- Email: [email protected]
Research on safety evaluation model for in-vehicle secondary task driving
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS Accident Analysis and Prevention xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Research on safety evaluation model for in-vehicle secondary task driving Lisheng Jin a,1 , Huacai Xian b,∗ , Qingning Niu a,2 , Jing Bie c,3 a b c
Transportation College, Jilin University, ChangChun 130024, JiLin, China Transportation and Logistics Engineering College, Shandong Jiaotong University, JiNan 250023, Shandong, China Grontmij Nederland B.V., De Holle Bilt 22, 3732 HM De Bilt, The Netherlands
a r t i c l e
i n f o
Article history: Received 30 September 2013 Received in revised form 13 July 2014 Accepted 3 August 2014 Available online xxx Keywords: Driving safety Multi-task driving Driver eye movements ANP FCE
a b s t r a c t This paper presents a new method for evaluating in-vehicle secondary task driving safety. There are five in-vehicle distracter tasks: tuning the radio to a local station, touching the touch-screen telephone menu to a certain song, talking with laboratory assistant, answering a telephone via Bluetooth headset, and finding the navigation system from Ipad4 computer. Forty young drivers completed the driving experiment on a driving simulator. Measures of fixations, saccades, and blinks are collected and analyzed. Based on the measures of driver eye movements which have significant difference between the baseline and secondary task driving conditions, the evaluation index system is built. The Analytic Network Process (ANP) theory is applied for determining the importance weight of the evaluation index in a fuzzy environment. On the basis of the importance weight of the evaluation index, Fuzzy Comprehensive Evaluation (FCE) method is utilized to evaluate the secondary task driving safety. Results show that driving with secondary tasks greatly distracts the driver’s attention from road and the evaluation model built in this study could estimate driving safety effectively under different driving conditions. Crown Copyright © 2014 Published by Elsevier Ltd. All rights reserved.
1. Introduction* The growing popularity of information technology has led to an increase in the number and complexity of In-Vehicle Information Systems (IVIS). Concerns have been raised over the likeliness of driver distraction due to these technologies. Driver distraction can be defined as the diversion of attention from the driving task, as compelled by an activity or event inside or outside the vehicle, which competes for the attention of the driver. It usually results in a delayed recognition and processing of the information needed for safely accomplishing the driving task (Khushaba et al., 2013; Jin et al., 2012a,b; Wang et al., 2013). According to the report of Nevile and Haddington (2010), drivers commonly engage in distracting activities and this accounts for 14–21% of crashes caused by distraction. Estimates from the 100-Car Naturalistic Driving Study are similar, placing secondary task engagement as an important contributing factor for more than 22% of crashes and near-crashes. As
∗ Corresponding author. Tel.: +86 18943637802. E-mail addresses: [email protected] (L. Jin), [email protected] (H. Xian), [email protected] (Q. Niu), [email protected] (J. Bie). 1 Tel.: +86 0431 85095268. 2 Tel.: +86 1 514 317 3523. 3 Tel.: +03 188 811 4472.
complexities of in-vehicle and modernized technologies grow, this figure is expected to increase (Crundall, 2009; Hou et al., 2011). According to the report of World Health Organization (2011), many countries have attached great importance to such activities and have taken effective measures to prohibit multi-task driving. Countries such as United States, China, and Canada, have enacted laws banning the use of hand-held cell phone, text messaging, hands-held GPS or mapping services while driving. Although operating a motor vehicle while using a computer, making calls using a Bluetooth headset and interacting with other wireless devices are often legal in these countries (Safety Laws in California, 2012; PRC Road Traffic Law on State Security, 2004), these actions can still severely distract the driver. The issue of driver distraction still calls for further attention. On April 14, 2009, the National Highway Traffic Safety Administration of China listed on its website that 41% of traffic accidents were caused by driver distraction. On June 17, 2012, the website Distraction.Gov listed that over 3000 people were killed in distracted driving crashes in 2011 alone. One of the important reasons of these accidents is multi-task driving, a main factor of driver distraction. Therefore, it is very valuable and significant to develop an evaluation system for in-vehicle secondary task driving. Existing researches have focused on portable devices such as hands-held mobile phones, music player, and radio, while relatively few studies have systematically examined the effects of more recent technologies on driving performance, especially in-vehicle
http://dx.doi.org/10.1016/j.aap.2014.08.013 0001-4575/Crown Copyright © 2014 Published by Elsevier Ltd. All rights reserved.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
2
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
computer with a big touch-screen and answering telephone via a Bluetooth headset, which is the subject of this study. In addition, although researchers have made great strides in analyzing effects of multi-task driving on driving safety, their works have almost exclusively focused on the driving safety evaluation system to prevent traffic accidents. There are also only a few studies systematically examining the influence of in-vehicle task on measures of driver eye movements including fixations, saccades, and blinks. Aiming at that, the overall goal of this paper is to develop a driving safety evaluation system based on driver eye movements, and thus provide a new method for driving safety researching. 2. Driving experiment and data analysis Due to the deficiencies of high risk of real vehicle test and inconvenient of instrument assembling, the method of simulation experiment is adopted in this study. 2.1. Participants Forty participants (29 females, 11 males, mean age 32.5 years) between 20 and 52 years of age are recruited from JiLin University and surrounding communities. All participants hold a valid class C1 driver’s license for more than 2 years and have an annual mileage of more than 5000 km. The participants are also required to be in good physical and mental health, and not taking any medication or drugs that would affect their driving performance. 2.2. Apparatus and materials 2.2.1. The Smarteye6.0 eyes tracking system Driver eye movements while driving are captured by SmartEye6.0 (Smart Eye AB, Första Långgatan 28, Göteborg, Sweden). It uses four cameras mounted in front of windscreen to capture the driver’s eye movements and three scenes cameras to record the simulated traffic environment, consequently obtaining visual behavior measures such as the visual direction, fixations and saccades. Video analysis is performed using Smarteye Analysis, an offline data review and reduction analysis program. As compared to head-mounted eye movement tracking systems, Smarteye6.0 has the advantage of being separately mounted (no physical contact with the driver) and therefore has less influence on driving performance. 2.2.2. Driving simulator The driving simulator is a fixed-base, static driving simulator equipped with a Besturn B50 car. The display system projects onto a screen with a 115◦ wide horizontal field view in front of the vehicle. In addition, 60 Hz is adopted as sample frequency both in SmartEye6.0 and driving simulator. The experiment apparatus is shown in Fig. 1. 2.3. Experiment setup 2.3.1. Secondary tasks According to the report of Yang et al. (2010), auditory and visual tasks have different influence on drivers’ eye movements. So three visual secondary tasks (radio tuning, telephone touching, and computer (iPad4) touching) and two auditory task (telephone conversation, talking with laboratory assistant) are adopted as the in-vehicle devices operated by drivers while driving. As is illustrated in the Fig. 1, all in-vehicle devices except telephone are fixed roughly in the same place. The radio task is a press-button task and need ten operation steps, taking about 7–10 s to complete when test alone, with turning to a local radio station in the end. Task of
searching a certain song from the touch-screen telephone is completed by seven touch-menu steps. The touch-screen telephone is a hand-held device. Computer task is to touch the computer menu to a navigation system and need seven completion steps. Tasks of conducting a phone conversation via Bluetooth headset and talking with laboratory assistant include fifteen simple questions, such as how long have you got the driving license? The number of questions is random. 2.3.2. Road environment The highway from ChangChun to SiPing is selected as the driving road environment, for which a real-time 3D model is built using the visual simulation software Multigen Creator and Vega. This highway section has two lanes in each direction, separated by a grassy median. A “heavy” traffic scenario where there are 100 vehicles evenly distributed is adopted as the road environment and all vehicles maintain the same spacing about 115 m. If the spacing between vehicles is too long, the experimental data collected by our driving simulator will deviate the actual. 2.3.3. Experimental sessions The experiment is conducted in two sessions: (1) Training session: Participants practice with operating three secondary tasks and driving in the simulator first separately and then together. The practice session ends when the participants feel they can operate the secondary tasks safely while driving. (2) Test session: On the premise of security, drivers should complete six drives, a drive without secondary task (baseline) and five drives with five different secondary tasks. Participants are told to complete the secondary tasks at any time they feel safe while driving. 2.4. Data analysis As is known to us all, multi-task driving has great effect on driver eye movements, while different measures have different sensitivity to driving safety. So measures which have high sensitivity for driving safety are chosen to construct the evaluation index system. For this purpose, method of ANOVA is applied to analyze the significance difference between secondary tasks driving and the baseline conditions. In this study, according to the difference of secondary tasks, we divided the collected data into three groups. The first data group is collected from drivers driving without secondary tasks. Data collected from drivers tuning the radio to a local station, searching a song from touch-screen telephone, and talking with laboratory assistant while driving belong to the second data group. The remaining data belong to the third date group. The second and the third data groups all include a visual and an auditory secondary task at least. Among these data groups, the first and the second data group are used to extract secondary task driving safety evaluation indexes and the whole data are used to verify the secondary task driving safety model’s availability. 2.4.1. Fixations Measures of driver fixation which could express driver attention are adopted in the study, including percentage of time with eyes off-road, mean number of fixations on-road, maximum off-road fixations time, mean on-road fixations duration, pupil diameter, standard deviation of horizontal and vertical search angles. (1) Percentage of time with eyes off-road indicates the visual resources occupied degrees caused by the secondary task.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
3
Fig. 1. Experiment apparatus.
Under same conditions, the longer time it takes, the more visual resources it occupies (Xian et al., 2013a,b). Mean number of fixations on-road is defined as the number of glances at a target during the task where each glance is separated by at least one glance to a different target. Number of glances is related to information amount acquired by the driver. In general, the more important areas are, the more glance number is required. Maximum off-road fixations time. Driving safety will be gravely threatened when the driver’s sightline keeps away from road more than 1.6 s, which is called visual distraction (‘long glance’ for short). Mean on-road fixations duration represents the time it takes to deal with information related with driving safety, exposing the level of difficulty to extract information as a metric standard of the processing way to subjective information. Pupil diameter mainly expresses driver intense, as a critical parameter of visual information. Changes of the pupil diameter indicate that attention is aroused by different stimulus. The variance and standard deviation of horizontal and vertical visual search angles are used to evaluate the search breadth. The study chooses the latter as a physical quantity to represent deviation angles from the average fixation point, with degree as the unit. Obviously, the larger the standard deviation of search angles is, the wider the driver visual search ranges.
p1 and p2 are results of significance difference test among radio and talking with laboratory assistant tasks compared to the baseline condition. Significant difference exists when p is less than 0.05, which are described in bold in Table 1. From Table 1 we can see that parameters of percentage of time with eyes off-road and maximum off-road fixations time under visual secondary tasks driving increase significantly compared with the baseline. Maximum off-road fixations time decrease significantly while drivers driving with visual secondary tasks. However, these three measures are not changed more obviously while driving with auditory secondary task than the baseline. This shows that visual secondary tasks need more driver attention and have a larger impact on driving safety. In addition, visual and auditory secondary tasks all have significant effect on standard deviation of horizontal and vertical visual search angles. The results of data analysis indicate that secondary tasks significantly make the drivers’ visual range narrow.
Results of significance difference test between the baseline (the first data group) and the second data group are showed in Table 1.
(1) Mean saccade speed is the proportion of every saccade distance (angle) to the saccade duration, with ‘degrees/sec’ as its unit.
(2)
(3)
(4)
(5)
(6)
2.4.2. Saccades Saccade is one of the most important aspects of driver visual behavior. It is the process of substantial eye movements (Zhang et al., 2013). Measures of saccades summarized in Table 2 include mean saccade speed, mean saccade amplitude, and maximum saccade speed, which have also been similarly analyzed in the study by using ANOVA.
Table 1 Results of significance test of fixations difference between normal and multi-task driving. Measures
Normal driving
Radio task
p1
Telephone touching task
p2
Talking task
p3
Percentage of time with eyes off-road (%) Mean number of fixations on-road (times/s) Maximum off-road fixations time (s) Mean on-road fixations duration (s) Pupil diameter (mm) Standard deviation of horizontal search angles (◦ ) Standard deviation of vertical search angles (◦ )
7.99 3.24 0.69 0.58 3.27 6.77 9.84
18.77 2.88 0.97 0.59 3.36 5.83 8.79
0.000 0.022 0.001 0.454 0.152 0.040 0.000
16.45 2.94 0.78 0.56 3.44 5.92 8.79
0.000 0.047 0.034 0.377 0.152 0.045 0.001
8.31 3.56 0.70 0.630 3.12 5.61 8.78
0.313 0.071 0.097 0.845 0.243 0.011 0.014
Significant difference exists when p is less than 0.05, which are described in bold.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
4
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
Table 2 Results of significance test of saccades difference between normal and multi-task driving. Measures
Normal driving
Radio task
Mean saccade speed (◦ /s) Mean saccade amplitude (◦ /s) Average number of saccades (times/s) Maximum saccade speed (◦ /s)
122.1 2.77 3.70 366.8
99.7 2.94 3.59 377.4
Mean saccade speed can explain the information processing speed in the previous fixation and the searching speed to find the next target (Xian et al., 2013a,b). (2) Mean saccade amplitude is the range of a glance from the beginning to the end, usually with a visual angle as its unit. Average number of saccade means the searching frequency for information of the driver, decreasing with the rise of cognitive task’s complexity. (3) Average number of saccades indicates the number of objects that should be concerned by drivers. (4) Maximum saccade speed signifies the corresponding saccade speed with the sample of maximum speed from all samples in a continuous saccade. It is evident from the analysis in Table 2 that measures of mean saccade speed, mean saccade amplitude and average number of saccades are affected obviously by auditory secondary task driving, all p < 0.05. In addition, mean saccade amplitude is also influenced significantly by visual secondary task driving, p = 0.004 < 0.05. 2.4.3. Blinks Three measures of blinks: blink frequency, blink duration, and blink percentage are collected in the study. The blink frequency presents the measurements of averagely how many times blinks occur per minute. Blink duration is the average over duration of each blink in secondary task conducting. Blink percentage shows the percent in terms of time when blink happened in secondary tasks (Yang et al., 2010). A repeated measure ANOVA finds that the radio and telephone touching tasks have no significant effect on driver blink behavior. However, significantly longer blink duration and larger blink percentage are found during the talking secondary task (blink duration: M = 0.164 s, S.E. = 0.06 s; blink percentage: M = 0.139%, S.E. = 0.03%) compared to the baseline (blink duration: M = 0.152 s, S.E. = 0.027 s; blink percentage: M = 0.12%, S.E. = 0.02%), all p < 0.05. Significant differences are also found in blink frequency between talking task and the baseline, p < 0.05. 3. Safety evaluation model for multi-task driving 3.1. Evaluation index system For getting better estimate precision, measures which have high sensitivity for driving safety are chosen to evaluate secondary task driving safety and construct the evaluation index system. Through the analysis above we can see that measures of driver eye movements are significantly different with different secondary tasks except mean on-road fixations duration, pupil diameter, and maximum saccade speed during the multi-task driving compared to the baseline condition. Therefore, twelve measures of driver visual behavior are selected to constitute the evaluation index system. These measures are: percentage of time with eyes off-road e11 , mean number of fixations on-road e12 , maximum off-road fixations time e13 , standard deviation of horizontal search angles e14 , standard deviation of vertical search angles e15 , mean saccade speed e21 , mean saccade amplitude e22 , Average number of saccades e23 , maximum saccade speed e24 , blink frequency e31 , blink duration e32 , and blink percentage e33 .
p1 0.004 0.131 0.102 0.040
Telephone touching task
p2
Talking task
p3
102.7 2.57 3.88 251.8
0.024 0.095 0.835 0.617
101.2 3.08 4.38 361.7
0.001 0.002 0.001 0.916
The evaluation system hierarchical level divides into three layers: target stratum (secondary task driving safety evaluation index system), controller layer, and network layer. Index of C1 (fixation index), C2 (saccades index), and C3 (blinks index) belong to controller layer while measures from e11 to e33 are in network layer. 3.2. Obtain the global priorities based on ANP method The analytic network process (ANP) is a more general form of the analytic hierarchy process (AHP) used in multi-criteria decision analysis (Aghdaie et al., 2013; Gumus and Yilmaz, 2010). AHP structures a decision problem into a hierarchy with a goal, decision criteria, and alternatives, while the ANP structures it as a network. Both then use a system of pairwise comparisons to measure the weights of the components of the structure, and finally to rank the alternatives in the decision (Liang et al., 2013; Jiang et al., 2012). Due to the interaction between driving safety evaluation indexes, method of ANP is selected. The ANP method is comprised of the following four steps: 3.2.1. Step 1: Form the network structure In the first step, the criteria, the sub-criteria and the alternatives are identified. Then, the clusters of the elements are determined and a network is formed based on the relationship among the clusters and within the elements in each cluster (Das et al., 2012). Several different relationships could be found in a network. Direct relationship is a regular dependency in a standard hierarchy. Indirect relationship is a relationship that flows through another criteria or alternative. The direct relationship between a criterion and itself is characterized by “self-interacting” criteria (Ginevicius and Podvezko, 2008). Finally, interdependencies are relationships among criteria which form a mutual effect. 3.2.2. Step 2: Pairwise comparison matrices In the second step, pairwise comparisons are performed on the elements within the clusters as they influence each cluster and on those that it influences, with respect to that criterion. The pairwise comparisons are made with respect to a criterion or sub-criterion of the controller layer and network layer. Thus, the importance weights of the factors are determined (Datta et al., 2009). In pairwise comparison, decision makers (eight experts on driving safety) compare two elements. Then, they determine the contribution of the factors to the result (Medineckiene and Björk, 2011). In ANP, similar to AHP, pairwise comparison matrices are formed using the 1–9 scale of relative importance proposed by Saaty (2005, 2001). The Fundamental Scale used for judgment is given in Table 3. Table 3 Fundamental scale. Value
Relationship between these two measures
1 3 5 7 9 2,4,6,8 Use reciprocals for inverse comparisons
Equal important Moderate important of one over another Strong or essential importance Very strong or demonstrated importance Extreme importance Intermediate values
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model
ARTICLE IN PRESS
AAP-3569; No. of Pages 8
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
The comparison matrix of controller layer and network layer in our research are shown in Eqs. (1) and (2) respectively. In order to describe the matrix better, we take matrix of controller layer as an example. In Acontroller layer , 1/3 means C2 is three times more important than C3 .
⎡ C1 C2 C3 ⎤ 1
AController layer = ⎣ 1/2 1/5
2 1
1/3
5
C1
3 ⎦ C2 1
(1)
C3
e11 e12 e13 1/1.2 1/7 1.4
5
3.3.1. Steps 1: Determine the factors set of safety evaluation for secondary task driving In this study, three factors (fixations, saccades, blinks) are selected as the safety evaluation factors and expressed as U: U = {u1 , u2 , u3 } =
fixations, saccades, blinks
3.3.2. Identify the comments set of safety evaluation for secondary task driving As the evaluation factors for driving safety are vague, the comments set V of the evaluation of factors are divided into five levels: e11
e14 e15 e21 e22 e23 e24 e31 e32 e33 1/2 1/2 1/5.8 1/7.6 1/5.2 1/3 1/6.6
⎡ 1 ⎤ e12 1/4 1 1/6.6 1.7 1/2 1/2.2 1/4.7 1/7.2 1/4 1/2.8 1/6 1/3.4 ⎥ e13 ⎢ 1.2 ⎢ ⎥ ⎢ 7 6.6 1 5 3 7 3.8 1/2 4.2 5 2.2 3.5 ⎥ e ⎢ ⎥ 14 ⎢ 1/1.4 1/1.7 1/5 1 2 1/2 1/4.4 1/6 1/5 1/2.5 1/4 1/3 ⎥ ⎢ ⎥ e15 ⎢ 2 2 1/3 1/2 1 1/5 1/6 1/5 1/5 1/3 1/4.3 1/3.5 ⎥ ⎢ ⎥ ⎢ 2 2.2 1/7 2 5 1 1/5 1/8 1/4.2 1/2 1/6.5 1/3 ⎥ e21 ⎥ ANetwork layer = ⎢ ⎢ 5.8 4.7 1/3.8 4.4 6 5 1 1/5 2.8 3 1/3 2.4 ⎥ ⎢ ⎥ e22 ⎢ 7.6 ⎥ 7.2 2 6 5 8 5 1 5 6.2 3 4 ⎢ ⎥ e23 ⎢ 5.2 4 1/4.2 5 5 4.2 1/2.8 1/5 1 2.8 1/3.4 2 ⎥ ⎢ ⎥ ⎢ 3 2.8 1/5 2.5 3 2 1/3 1/6.2 1/2.8 1 1/4 1/2.2 ⎥ ⎢ ⎥ e24 ⎣ 6.6 6 1/2.2 4 4.3 6.5 3 1/3 3.4 4 1 2.9 ⎦ e31 4
3.4
1/3.5
3
3.5
3
1/2.4
1/4
1/2
2.2
1/2.9
1
(2)
e32 e33
3.2.3. Step 3: Establish the weight coefficient For establishing the weight coefficient of e11 to e33 , twelve numbers in a row of the network layer matrix are multiplied and then calculate the product’s 12th root. On the basis of the calculation, the initialization weight vector Wi (1, 2, . . ., 12) is obtained. After normalizing the Wi , the weighting coefficient values Wi (1, 2, . . ., 12) of the evaluation indexes are calculated by the Eq. (3). Wi =
Wi 12
(3)
Wj
V = {v1 , v2 , v3 , v4 , v5 } =
3.3. Realize safety evaluation based on FCE method Fuzzy comprehensive evaluation is a method, which adopts the principle of fuzzy relationship synthesis, comprehensively judges the membership grade status of the things to be judged from many factors (Tuzkaya and Önüt, 2008). The basic steps of fuzzy comprehensive evaluation method in this study are as following (Janic and Reggiani, 2002; Lee et al., 2009):
very high, high, average, low, very low
Constructing the membership functions and classifying the secondary task driving safety. The membership function of a fuzzy set is a generalization of the indicator function in classical sets (Zolfani et al., 2012). In this study, the relationship between the membership degree and the driving safety score is not in a monotony trend. So it is suitable to solve this problem using middle type membership function patterns. Therefore, sine curve is selected as the membership function. The fuzzy analytical expression is as follows.
j=1
Through the above calculation, the obtained weight vector of controller layer is (0.58, 0.31, 0.11). The local weight vector of network layer is (0.02, 0.02, 0.19, 0.02, 0.02, 0.02, 0.11, 0.26, 0.08, 0.05, 0.15, 0.07). According to the proportional allocation principle, the global weight of each evaluation index is G = (0.035, 0.035, 0.332, 0.035, 0.035, 0.019, 0.103, 0.243, 0.075, 0.017, 0.050, 0.023). After performing a consistency check on judgment matrixes, satisfactory agreements of the judgment matrixes are obtained. So the judgment matrixes are acceptable.
r1 (u) =
r2 (u) =
r3 (u) =
r4 (u) =
⎧ ⎪ ⎪ ⎨1 2 ⎪ ⎪ ⎩
⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩ ⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩
1
sin
u − 0.1 0.2
1 + 2
+1
0
u − 0.3 0.2
1 + 2
+1
0
⎧ ⎪ ⎪ ⎨1 2 ⎪ ⎪ ⎩
u − 0.5 0.2
1 + 2
+1
0
sin
0.2 0
u ≤ 0.1 0.1 < u < 0.5
(5)
u ≤ 0.3 0.3 < u < 0.7
(6)
u ≥ 0.7
0
u − 0.7
(4)
u ≥ 0.5
0
1 sin 2
0.1 < u < 0.3 u ≥ 0.3
0
1 sin 2
0 ≤ u ≤ 0.1
1 + 2
+1
u ≤ 0.5 0.5 < u < 0.9
(7)
u ≥ 0.9
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
6
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
r5 (u) =
⎧ ⎪ ⎪ ⎨1 2 ⎪ ⎪ ⎩
0
sin
u − 0.9 0.2
1 + 2
+1
u ≤ 0.7 0.7 < u < 0.9
driving safety levels are divided. So, driver1’s driving safety judgment set under normal driving condition is B1 , as follows. (8)
1 ≥ u ≥ 0.9
1
In these formulas: r1 to r5 represent the degree of secondary task driving safety belongs to ‘very low’, ‘low’, ‘average’, ‘high’, and ‘very high’ respectively. u is the value of criterion. Due to the maximum membership grade principle, the levels of secondary task driving safety could be determined.
3.3.3. Constructing the single factor assessment matrix for each driver In order to describe how the matrix constructed, we will take the driver 1 as an example. First of all, the data collected from driving simulator should be processed using Eq. (9). u11 =
ui − min(ui ) max(ui ) − min(ui )
(9)
In Eq. (9): max(ui )—the maximum of criteria value; min(ui )—the minimum of criteria value; ui —the criteria value of the present driver. Second, substituting the u11 into Eqs. (4)–(8), the single factor assessment matrix for each driver is determined. In accordance with this method, driver1’s single factor assessment matrix under normal driving condition is obtained and shown in the matrix R1 .
⎡ 0 ⎤ 0 0.13 0.87 0 0 0 0.77 0.23 ⎥ ⎢ 0 ⎢ ⎥ ⎢ 0 0 0.11 0.89 0 ⎥ ⎢ ⎥ ⎢ 0 0 0.94 0.06 0 ⎥ ⎢ ⎥ ⎢ 0 0 0.54 0.46 0 ⎥ ⎢ ⎥ ⎢ 0 0.22 0.78 0 0 ⎥ ⎢ ⎥ R1 = ⎢ 0 0 ⎥ ⎢ 0.31 0.69 0 ⎥ ⎢ 0 0.77 0.23 0 0 ⎥ ⎢ ⎥ ⎢ 0 0 0 0.88 0.12 ⎥ ⎢ ⎥ ⎢ 0 0 0 0.74 0.26 ⎥ ⎢ ⎥ ⎣ 0 0 0 0.69 0.31 ⎦ 0
0
0.27
0.73
⎡ 0.035 ⎤T ⎡ 0 ⎤ 0 0.13 0.87 0 0 0 0.77 0.23 ⎥ ⎢ 0.035 ⎥ ⎢ 0 ⎢ ⎥ ⎢ ⎥ ⎢ 0.332 ⎥ ⎢ 0 0 0.11 0.89 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.035 ⎥ ⎢ 0 0 0.94 0.06 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.035 ⎥ ⎢ 0 0 0.54 0.46 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.019 ⎥ ⎢ 0 0.22 0.78 0 0 ⎥ ⎥ ⎢ ⎥ B1 = G • R1 = ⎢ ⎢ 0.103 ⎥ ⎢ 0.31 0.69 0 0 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.243 ⎥ ⎢ 0 0.77 0.23 0 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.075 ⎥ ⎢ 0 0 0 0.88 0.12 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.017 ⎥ ⎢ 0 0 0 0.74 0.26 ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ 0.050 ⎦ ⎣ 0 0 0 0.69 0.31 ⎦ 0.023 0 0 = (0.032, 0.262, 0.170, 0.511, 0.037)
0.27
0.73
0
The maximum value of B1 is 0.511. According to the membership function, the score 0.511 belongs to ‘high’ level of driving safety. Similarly, the driving safety levels of other drivers under different driving conditions could be calculated. 4. Results and discussion Based on the evaluation results, we classify and analyze statistically according to driving conditions, driver’s driving experience, and types of secondary task. Concerning the driving experience division, we learn from the study of Yingshi Guo (Yingshi Guo, 2009) and classify driving experience into three groups: experienced driver (mileage is equal or greater than 5 million kilometers), general experienced driver (mileage is between 2 and 5 million kilometers) and inexperienced driver (mileage is equal or less than 2 million kilometers). Tables 4–7 are statistical data of the influence of driving conditions, types of secondary task and driver’s driving experience on driving safety. 4.1. Secondary task driving has significant influence on driving safety
0
In this way, the single factor assessment matrixes of drivers from 1 to 40 under different driving conditions are obtained.
3.3.4. Determining the driving safety levels The driving safety judgment sets are determined by multiplying the global weight of each criterion G and single factor assessment matrix R. With the maximum membership grade principle, the
From Tables 4 and 5, we can draw the conclusion that secondary task driving has significant influence on driving safety. ‘Average’ and above of driving safety levels under normal driving condition (93.0%) are higher than driving with bluetooth headset conversation (59.5%), laboratory assistant talking (70.2%), radio tuning (27.0%), and touch-screen touching (10.0%). Overall, during the secondary task driving, ‘Average’ and above of driving safety levels account for only 41.6%, while ‘Low’ and ‘Very low’ of driving safety levels are responsible for 58.4%. Through analysis for test data, it can be seen that secondary task driving, regardless of secondary task type, has significant influence on driving safety.
Table 4 Statistics of drivers’ driving safety levels under different driving conditions. Driving condition
Normal Bluetooth headset conversation Talking with laboratory assistant Radio tuning Touch-screen touching (telephone and Ipad4)
Percent of driving safety level Very high
High
Average
Low
Very low
42.0% 2.5% 6.2% 0% 0%
25.5% 17.7% 19.0% 4.0% 3.2%
25.5% 39.3% 45.0% 22.5% 6.9%
7.0% 27.5% 19.8% 61.2% 50.5%
0% 13.0% 10.0% 12.3% 39.4%
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model
ARTICLE IN PRESS
AAP-3569; No. of Pages 8
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
7
Table 5 Comparison results of driving safety levels under normal driving and secondary task driving conditions. Driving condition
Normal Secondary task driving
Percent of driving safety level Very high
High
Average
Low
Very low
42.0% 2.2%
25.5% 11.0%
25.5% 28.4%
7.0% 39.8%
0% 18.7%
Table 6 The influences of secondary task type on driving safety levels. Driving condition
Normal Cognitive tasks Visual tasks
Percent of driving safety level Very high
High
Average
Low
Very low
43.0% 4.4% 0%
22.0% 18.4% 3.6%
27.5% 42.2% 14.7%
7.5% 23.7% 55.9%
0% 11.5% 25.9%
Table 7 The influence of driver’s driving experience on driving safety levels. Driving condition
Driving experience
Percent of driving safety level Very high
High
Average
Normal driving
Rich General Bad Rich General Bad
61.5% 40.6% 20.8% 4.5% 0% 0%
19.0% 24.4% 23.2% 19.5% 3.5% 3.1%
19.5% 30% 33.8% 45.5% 25.5% 14.3%
Secondary task driving
4.2. Visual tasks have a greater impact on driving safety than cognitive tasks It is evident from the results in Table 6 that driving with visual tasks (‘Low’ and ‘Very low’ account for more than 81.0%) are dangerous than with cognitive tasks (‘Low’ and ‘Very low’ account for about 35.2%). Additionally, the security of driving with cognitive tasks is also lower than normal driving. 4.3. Driver’s driving experience has significant influence on driving safety
Low
Very low
0% 5% 22.2% 29.0% 55.0% 46.0%
0% 0% 0% 1.5% 16.0% 36.6%
driving, secondary task types and driver’s driving experience have influence on the driving safety obviously. Secondary task driving can decrease the driving safety significantly. In conclusion, the proposed approach is useful and effective to classify the safety level of the secondary task driving. Secondary task driving has bad effects on driving safety and drivers should enhance their consciousness away from interacting with secondary tasks while driving. The government should pay more attention to this issue and take action to prevent traffic accident caused by these secondary tasks. Acknowledgements
The statistic analysis indicates that, in any driving conditions, the more experienced the driver has, the stronger the ability keeps the driving safety. Under normal driving condition, driving safety level of experienced drivers is the highest (‘Average’ and above account for about 100%). The same situation applies to secondary task driving. Experienced drivers driving with secondary task can still keep good driving safety (‘Average’ and above account for about 69.5%) while inexperienced drivers cannot (‘Average’ and above account for only 17.4%). 5. Conclusions This study aimed to determine the driving safety level of secondary task driving by using approach of Analytic Network Process. The proposed approach takes into account the drivers’ fixations, saccades, and blinks safety with their controller layer. The weights of the evaluation indexes are determined by eight driving safety experts. After determining the weights of the evaluation indexes and constructing the single factor assessment matrix for each driver, the maximum membership grade principle is applied to divide the safety level of secondary task driving. Finally, the classification is implemented. Data of forty drivers engaging in driving simulating are used for verifying the applicability of the secondary task driving safety evaluation system. The analysis results showed that secondary task
This research was partially supported by Doctoral Study Special Research Foundation of Higher Education (20110061110036), Major Projects of JiLin Science and Technology Department (20116017), New Century Excellent Talent Foundation Program (NCET-10-0435) and Talent Development Foundation Projects of JiLin (801121100417). References Aghdaie, M.H., Zolfani, S.H., Zavadskas, E.K., 2013. Market segment evaluation and selection based on application of fuzzy AHP and COPRAS-G methods. J. Bus. Econ. Manage. 14 (1), 213–233. Crundall, D., 2009. The deceleration detection flicker test: a measure of experience? Ergonomics 6 (52), 184–674. Das, M.C., Sarkar, B., Ray, S., 2012. A framework to measure relative performance of Indian technical institutions using integrated fuzzy AHP and COPRAS methodology. Socio Econ. Plan. Sci. 46 (3), 230–241. Datta, S., Beriha, G.S., Patnaik, B., Mahapatra, S.S., 2009. Use of compromise ranking method for supervisor selection: a multi criteria decision making (MCDM) approach. IJVTE 1 (1), 7–13. Ginevicius, R., Podvezko, V., 2008. Multi-criteria evaluation of Lithuanian banks from the perspective of their reliability for clients. J. Bus. Econ. Manage. 9 (4), 257–267. Gumus, A.T., Yilmaz, G., 2010. Sea vessel type selection via an integrated VAHP-ANP methodology for high-speed public transportation in Bosphorus. Expert Syst. Appl. 37 (6), 4182–4189. Janic, M., Reggiani, A., 2002. An application of the multiple criteria decision making (MCDM) analysis to the selection of a new hub airport. EJTIR 2 (2), 113–141. Jiang, Y.Y., Wu, Z., Jin, L.S., Hou, H.J., 2012. Driver inattention monitoring system based on machine vision. Information 15 (12A), 5491–5498.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
8
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
Jin, L.S., Xian, H.C., Hou, H.J., Niu, Q.N., Lv, H.H., 2012a. Analyzing effects of touchscreen on driving performance. J. Beijing Inst. Technol. Suppl 2 (21), 6–10. Jin, L.S., Niu, Q.N., Hou, H.J., Xian, H.C., Wang, Y.L., Shi, D.D., 2012b. Driver cognitive distraction detection using driving performance measures. Discrete Dyn. Nat. Soc. 1, 1–12. Hou, H.J., Jin, L.S.H., Niu, Q.N., Sun, Y.Q., 2011. Driver intention recognition method using continuous hidden markov model. IJCIS 4 (3), 386–393. Xian, H.C., Jin, L.S., Hou, H.J., et al., 2013a. Analyzing effects of pressing radio button on driver’s visual cognition. CSIP2013 215, 69–78. Xian, H.C., Jin, L.S., Hou, H.J., et al., 2013b. Research on in-vehicle secondary task analysis model based on preview theory. In: The 2nd International Conference on Transportation Information and Safety (ICTIS 2013), pp. 1282–1287. Khushaba, R.N., Kodagoda, S., Liu, D., Dissanayake, G., 2013. Muscle computer interfaces for driver distraction reduction. Comput. Meth. Prog. Bio. 2 (110), 137–149. Liang, X.Y., Sun, X.X., Shu, G.Q., Sun, K., Wang, X., Wang, X.L., 2013. Using the analytic network process (ANP) to determine method of waste energy recovery from engine. Energ Convers. Manage. 66, 304–311. Lee, H., Kim, C., Cho, H., Park, Y., 2009. An ANP-based technology network for identification of core technologies: a case of telecommunication technologies. Expert Syst. Appl. 36 (1), 894–908. Medineckiene, M., Björk, F., 2011. Owner preferences regarding renovation measures - the demonstration of using multi-criteria decision making. J. Civil Eng. Manage. 17 (2), 284–295. Nevile, M., Haddington, P., 2010. In-car distractions and their impact on driving activities. Department of Infrastructure and Transport, Road Safety Grant, Report RSGR-2010-001. PRC Road Traffic Law on State Security, 2004. Chapter IV provisions on road passage. Article 62: Driver may do any of the following. http://baike.baidu.com/ view/20045.htm#5
Saaty, T.L., 2001. Decision Making with dependence and Feedback, 2nd Ed. RWS Publication, Pittsburg. Saaty, T.L., 2005. Theory and Applications of the Analytic Network Process: Decision Making with Benefits, Opportunities, Costs, and Risks. RWS Publications, Pittsburg. Safety Laws in California, 2012. Cell phone laws–Text messaging–GPS and other map apps. http://www.dmv.org/ca-california/safety-laws.php#Cell-Phone-Laws Tuzkaya, U.R., Önüt, S., 2008. A fuzzy analytic network process based approach to transportation-mode selection between Turkey and Germany: a case study. Inform. Sci. 178 (15), 3133–3146. Wang, J.Q., Zhang, L., Zhang, D., Zh Li, K.Q., 2013. An adaptive longitudinal driving assistance system based on driver characteristics. IEEE T INTELL TRANSP 14 (1), 1–12. World Health Organization, 2011. Mobile phone use: a growing problem of driver http://www.who.int/violence injury prevention/publications/ distraction. road traffic/en/index.html Yang, Y., McDonald, M., Zheng, P.J., 2010. The Effects of increasing workload on driver’s eye movement. The 12th WCTRS. World Conference on Transport Research Society, 11–15. Yingshi Guo, 2009. Study on effects of traffic environment and driving experience on driver’s eye movement and workload, A dissertation for degree of doctor, China, Xi’an Chang’an University. Zhang, Y., Harris, E., Rogers, M., Kaber, D., Hummer, J., Rasdorf, W., Hu, J., 2013. Driver distraction and performance effects of highway logo sign design. Appl. Ergon. 3 (44), 472–479. Zolfani, S.H., Chen, I.S., Rezaeiniya, N., Tamosaitiene, J., 2012. A hybrid MCDM model encompassing AHP and COPRAS-G methods for selecting company supplier in Iran. Technol. Econ. Dev. Econ. 18 (3), 529–543.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
ARTICLE IN PRESS Accident Analysis and Prevention xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Research on safety evaluation model for in-vehicle secondary task driving Lisheng Jin a,1 , Huacai Xian b,∗ , Qingning Niu a,2 , Jing Bie c,3 a b c
Transportation College, Jilin University, ChangChun 130024, JiLin, China Transportation and Logistics Engineering College, Shandong Jiaotong University, JiNan 250023, Shandong, China Grontmij Nederland B.V., De Holle Bilt 22, 3732 HM De Bilt, The Netherlands
a r t i c l e
i n f o
Article history: Received 30 September 2013 Received in revised form 13 July 2014 Accepted 3 August 2014 Available online xxx Keywords: Driving safety Multi-task driving Driver eye movements ANP FCE
a b s t r a c t This paper presents a new method for evaluating in-vehicle secondary task driving safety. There are five in-vehicle distracter tasks: tuning the radio to a local station, touching the touch-screen telephone menu to a certain song, talking with laboratory assistant, answering a telephone via Bluetooth headset, and finding the navigation system from Ipad4 computer. Forty young drivers completed the driving experiment on a driving simulator. Measures of fixations, saccades, and blinks are collected and analyzed. Based on the measures of driver eye movements which have significant difference between the baseline and secondary task driving conditions, the evaluation index system is built. The Analytic Network Process (ANP) theory is applied for determining the importance weight of the evaluation index in a fuzzy environment. On the basis of the importance weight of the evaluation index, Fuzzy Comprehensive Evaluation (FCE) method is utilized to evaluate the secondary task driving safety. Results show that driving with secondary tasks greatly distracts the driver’s attention from road and the evaluation model built in this study could estimate driving safety effectively under different driving conditions. Crown Copyright © 2014 Published by Elsevier Ltd. All rights reserved.
1. Introduction* The growing popularity of information technology has led to an increase in the number and complexity of In-Vehicle Information Systems (IVIS). Concerns have been raised over the likeliness of driver distraction due to these technologies. Driver distraction can be defined as the diversion of attention from the driving task, as compelled by an activity or event inside or outside the vehicle, which competes for the attention of the driver. It usually results in a delayed recognition and processing of the information needed for safely accomplishing the driving task (Khushaba et al., 2013; Jin et al., 2012a,b; Wang et al., 2013). According to the report of Nevile and Haddington (2010), drivers commonly engage in distracting activities and this accounts for 14–21% of crashes caused by distraction. Estimates from the 100-Car Naturalistic Driving Study are similar, placing secondary task engagement as an important contributing factor for more than 22% of crashes and near-crashes. As
∗ Corresponding author. Tel.: +86 18943637802. E-mail addresses: [email protected] (L. Jin), [email protected] (H. Xian), [email protected] (Q. Niu), [email protected] (J. Bie). 1 Tel.: +86 0431 85095268. 2 Tel.: +86 1 514 317 3523. 3 Tel.: +03 188 811 4472.
complexities of in-vehicle and modernized technologies grow, this figure is expected to increase (Crundall, 2009; Hou et al., 2011). According to the report of World Health Organization (2011), many countries have attached great importance to such activities and have taken effective measures to prohibit multi-task driving. Countries such as United States, China, and Canada, have enacted laws banning the use of hand-held cell phone, text messaging, hands-held GPS or mapping services while driving. Although operating a motor vehicle while using a computer, making calls using a Bluetooth headset and interacting with other wireless devices are often legal in these countries (Safety Laws in California, 2012; PRC Road Traffic Law on State Security, 2004), these actions can still severely distract the driver. The issue of driver distraction still calls for further attention. On April 14, 2009, the National Highway Traffic Safety Administration of China listed on its website that 41% of traffic accidents were caused by driver distraction. On June 17, 2012, the website Distraction.Gov listed that over 3000 people were killed in distracted driving crashes in 2011 alone. One of the important reasons of these accidents is multi-task driving, a main factor of driver distraction. Therefore, it is very valuable and significant to develop an evaluation system for in-vehicle secondary task driving. Existing researches have focused on portable devices such as hands-held mobile phones, music player, and radio, while relatively few studies have systematically examined the effects of more recent technologies on driving performance, especially in-vehicle
http://dx.doi.org/10.1016/j.aap.2014.08.013 0001-4575/Crown Copyright © 2014 Published by Elsevier Ltd. All rights reserved.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
2
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
computer with a big touch-screen and answering telephone via a Bluetooth headset, which is the subject of this study. In addition, although researchers have made great strides in analyzing effects of multi-task driving on driving safety, their works have almost exclusively focused on the driving safety evaluation system to prevent traffic accidents. There are also only a few studies systematically examining the influence of in-vehicle task on measures of driver eye movements including fixations, saccades, and blinks. Aiming at that, the overall goal of this paper is to develop a driving safety evaluation system based on driver eye movements, and thus provide a new method for driving safety researching. 2. Driving experiment and data analysis Due to the deficiencies of high risk of real vehicle test and inconvenient of instrument assembling, the method of simulation experiment is adopted in this study. 2.1. Participants Forty participants (29 females, 11 males, mean age 32.5 years) between 20 and 52 years of age are recruited from JiLin University and surrounding communities. All participants hold a valid class C1 driver’s license for more than 2 years and have an annual mileage of more than 5000 km. The participants are also required to be in good physical and mental health, and not taking any medication or drugs that would affect their driving performance. 2.2. Apparatus and materials 2.2.1. The Smarteye6.0 eyes tracking system Driver eye movements while driving are captured by SmartEye6.0 (Smart Eye AB, Första Långgatan 28, Göteborg, Sweden). It uses four cameras mounted in front of windscreen to capture the driver’s eye movements and three scenes cameras to record the simulated traffic environment, consequently obtaining visual behavior measures such as the visual direction, fixations and saccades. Video analysis is performed using Smarteye Analysis, an offline data review and reduction analysis program. As compared to head-mounted eye movement tracking systems, Smarteye6.0 has the advantage of being separately mounted (no physical contact with the driver) and therefore has less influence on driving performance. 2.2.2. Driving simulator The driving simulator is a fixed-base, static driving simulator equipped with a Besturn B50 car. The display system projects onto a screen with a 115◦ wide horizontal field view in front of the vehicle. In addition, 60 Hz is adopted as sample frequency both in SmartEye6.0 and driving simulator. The experiment apparatus is shown in Fig. 1. 2.3. Experiment setup 2.3.1. Secondary tasks According to the report of Yang et al. (2010), auditory and visual tasks have different influence on drivers’ eye movements. So three visual secondary tasks (radio tuning, telephone touching, and computer (iPad4) touching) and two auditory task (telephone conversation, talking with laboratory assistant) are adopted as the in-vehicle devices operated by drivers while driving. As is illustrated in the Fig. 1, all in-vehicle devices except telephone are fixed roughly in the same place. The radio task is a press-button task and need ten operation steps, taking about 7–10 s to complete when test alone, with turning to a local radio station in the end. Task of
searching a certain song from the touch-screen telephone is completed by seven touch-menu steps. The touch-screen telephone is a hand-held device. Computer task is to touch the computer menu to a navigation system and need seven completion steps. Tasks of conducting a phone conversation via Bluetooth headset and talking with laboratory assistant include fifteen simple questions, such as how long have you got the driving license? The number of questions is random. 2.3.2. Road environment The highway from ChangChun to SiPing is selected as the driving road environment, for which a real-time 3D model is built using the visual simulation software Multigen Creator and Vega. This highway section has two lanes in each direction, separated by a grassy median. A “heavy” traffic scenario where there are 100 vehicles evenly distributed is adopted as the road environment and all vehicles maintain the same spacing about 115 m. If the spacing between vehicles is too long, the experimental data collected by our driving simulator will deviate the actual. 2.3.3. Experimental sessions The experiment is conducted in two sessions: (1) Training session: Participants practice with operating three secondary tasks and driving in the simulator first separately and then together. The practice session ends when the participants feel they can operate the secondary tasks safely while driving. (2) Test session: On the premise of security, drivers should complete six drives, a drive without secondary task (baseline) and five drives with five different secondary tasks. Participants are told to complete the secondary tasks at any time they feel safe while driving. 2.4. Data analysis As is known to us all, multi-task driving has great effect on driver eye movements, while different measures have different sensitivity to driving safety. So measures which have high sensitivity for driving safety are chosen to construct the evaluation index system. For this purpose, method of ANOVA is applied to analyze the significance difference between secondary tasks driving and the baseline conditions. In this study, according to the difference of secondary tasks, we divided the collected data into three groups. The first data group is collected from drivers driving without secondary tasks. Data collected from drivers tuning the radio to a local station, searching a song from touch-screen telephone, and talking with laboratory assistant while driving belong to the second data group. The remaining data belong to the third date group. The second and the third data groups all include a visual and an auditory secondary task at least. Among these data groups, the first and the second data group are used to extract secondary task driving safety evaluation indexes and the whole data are used to verify the secondary task driving safety model’s availability. 2.4.1. Fixations Measures of driver fixation which could express driver attention are adopted in the study, including percentage of time with eyes off-road, mean number of fixations on-road, maximum off-road fixations time, mean on-road fixations duration, pupil diameter, standard deviation of horizontal and vertical search angles. (1) Percentage of time with eyes off-road indicates the visual resources occupied degrees caused by the secondary task.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
3
Fig. 1. Experiment apparatus.
Under same conditions, the longer time it takes, the more visual resources it occupies (Xian et al., 2013a,b). Mean number of fixations on-road is defined as the number of glances at a target during the task where each glance is separated by at least one glance to a different target. Number of glances is related to information amount acquired by the driver. In general, the more important areas are, the more glance number is required. Maximum off-road fixations time. Driving safety will be gravely threatened when the driver’s sightline keeps away from road more than 1.6 s, which is called visual distraction (‘long glance’ for short). Mean on-road fixations duration represents the time it takes to deal with information related with driving safety, exposing the level of difficulty to extract information as a metric standard of the processing way to subjective information. Pupil diameter mainly expresses driver intense, as a critical parameter of visual information. Changes of the pupil diameter indicate that attention is aroused by different stimulus. The variance and standard deviation of horizontal and vertical visual search angles are used to evaluate the search breadth. The study chooses the latter as a physical quantity to represent deviation angles from the average fixation point, with degree as the unit. Obviously, the larger the standard deviation of search angles is, the wider the driver visual search ranges.
p1 and p2 are results of significance difference test among radio and talking with laboratory assistant tasks compared to the baseline condition. Significant difference exists when p is less than 0.05, which are described in bold in Table 1. From Table 1 we can see that parameters of percentage of time with eyes off-road and maximum off-road fixations time under visual secondary tasks driving increase significantly compared with the baseline. Maximum off-road fixations time decrease significantly while drivers driving with visual secondary tasks. However, these three measures are not changed more obviously while driving with auditory secondary task than the baseline. This shows that visual secondary tasks need more driver attention and have a larger impact on driving safety. In addition, visual and auditory secondary tasks all have significant effect on standard deviation of horizontal and vertical visual search angles. The results of data analysis indicate that secondary tasks significantly make the drivers’ visual range narrow.
Results of significance difference test between the baseline (the first data group) and the second data group are showed in Table 1.
(1) Mean saccade speed is the proportion of every saccade distance (angle) to the saccade duration, with ‘degrees/sec’ as its unit.
(2)
(3)
(4)
(5)
(6)
2.4.2. Saccades Saccade is one of the most important aspects of driver visual behavior. It is the process of substantial eye movements (Zhang et al., 2013). Measures of saccades summarized in Table 2 include mean saccade speed, mean saccade amplitude, and maximum saccade speed, which have also been similarly analyzed in the study by using ANOVA.
Table 1 Results of significance test of fixations difference between normal and multi-task driving. Measures
Normal driving
Radio task
p1
Telephone touching task
p2
Talking task
p3
Percentage of time with eyes off-road (%) Mean number of fixations on-road (times/s) Maximum off-road fixations time (s) Mean on-road fixations duration (s) Pupil diameter (mm) Standard deviation of horizontal search angles (◦ ) Standard deviation of vertical search angles (◦ )
7.99 3.24 0.69 0.58 3.27 6.77 9.84
18.77 2.88 0.97 0.59 3.36 5.83 8.79
0.000 0.022 0.001 0.454 0.152 0.040 0.000
16.45 2.94 0.78 0.56 3.44 5.92 8.79
0.000 0.047 0.034 0.377 0.152 0.045 0.001
8.31 3.56 0.70 0.630 3.12 5.61 8.78
0.313 0.071 0.097 0.845 0.243 0.011 0.014
Significant difference exists when p is less than 0.05, which are described in bold.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
4
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
Table 2 Results of significance test of saccades difference between normal and multi-task driving. Measures
Normal driving
Radio task
Mean saccade speed (◦ /s) Mean saccade amplitude (◦ /s) Average number of saccades (times/s) Maximum saccade speed (◦ /s)
122.1 2.77 3.70 366.8
99.7 2.94 3.59 377.4
Mean saccade speed can explain the information processing speed in the previous fixation and the searching speed to find the next target (Xian et al., 2013a,b). (2) Mean saccade amplitude is the range of a glance from the beginning to the end, usually with a visual angle as its unit. Average number of saccade means the searching frequency for information of the driver, decreasing with the rise of cognitive task’s complexity. (3) Average number of saccades indicates the number of objects that should be concerned by drivers. (4) Maximum saccade speed signifies the corresponding saccade speed with the sample of maximum speed from all samples in a continuous saccade. It is evident from the analysis in Table 2 that measures of mean saccade speed, mean saccade amplitude and average number of saccades are affected obviously by auditory secondary task driving, all p < 0.05. In addition, mean saccade amplitude is also influenced significantly by visual secondary task driving, p = 0.004 < 0.05. 2.4.3. Blinks Three measures of blinks: blink frequency, blink duration, and blink percentage are collected in the study. The blink frequency presents the measurements of averagely how many times blinks occur per minute. Blink duration is the average over duration of each blink in secondary task conducting. Blink percentage shows the percent in terms of time when blink happened in secondary tasks (Yang et al., 2010). A repeated measure ANOVA finds that the radio and telephone touching tasks have no significant effect on driver blink behavior. However, significantly longer blink duration and larger blink percentage are found during the talking secondary task (blink duration: M = 0.164 s, S.E. = 0.06 s; blink percentage: M = 0.139%, S.E. = 0.03%) compared to the baseline (blink duration: M = 0.152 s, S.E. = 0.027 s; blink percentage: M = 0.12%, S.E. = 0.02%), all p < 0.05. Significant differences are also found in blink frequency between talking task and the baseline, p < 0.05. 3. Safety evaluation model for multi-task driving 3.1. Evaluation index system For getting better estimate precision, measures which have high sensitivity for driving safety are chosen to evaluate secondary task driving safety and construct the evaluation index system. Through the analysis above we can see that measures of driver eye movements are significantly different with different secondary tasks except mean on-road fixations duration, pupil diameter, and maximum saccade speed during the multi-task driving compared to the baseline condition. Therefore, twelve measures of driver visual behavior are selected to constitute the evaluation index system. These measures are: percentage of time with eyes off-road e11 , mean number of fixations on-road e12 , maximum off-road fixations time e13 , standard deviation of horizontal search angles e14 , standard deviation of vertical search angles e15 , mean saccade speed e21 , mean saccade amplitude e22 , Average number of saccades e23 , maximum saccade speed e24 , blink frequency e31 , blink duration e32 , and blink percentage e33 .
p1 0.004 0.131 0.102 0.040
Telephone touching task
p2
Talking task
p3
102.7 2.57 3.88 251.8
0.024 0.095 0.835 0.617
101.2 3.08 4.38 361.7
0.001 0.002 0.001 0.916
The evaluation system hierarchical level divides into three layers: target stratum (secondary task driving safety evaluation index system), controller layer, and network layer. Index of C1 (fixation index), C2 (saccades index), and C3 (blinks index) belong to controller layer while measures from e11 to e33 are in network layer. 3.2. Obtain the global priorities based on ANP method The analytic network process (ANP) is a more general form of the analytic hierarchy process (AHP) used in multi-criteria decision analysis (Aghdaie et al., 2013; Gumus and Yilmaz, 2010). AHP structures a decision problem into a hierarchy with a goal, decision criteria, and alternatives, while the ANP structures it as a network. Both then use a system of pairwise comparisons to measure the weights of the components of the structure, and finally to rank the alternatives in the decision (Liang et al., 2013; Jiang et al., 2012). Due to the interaction between driving safety evaluation indexes, method of ANP is selected. The ANP method is comprised of the following four steps: 3.2.1. Step 1: Form the network structure In the first step, the criteria, the sub-criteria and the alternatives are identified. Then, the clusters of the elements are determined and a network is formed based on the relationship among the clusters and within the elements in each cluster (Das et al., 2012). Several different relationships could be found in a network. Direct relationship is a regular dependency in a standard hierarchy. Indirect relationship is a relationship that flows through another criteria or alternative. The direct relationship between a criterion and itself is characterized by “self-interacting” criteria (Ginevicius and Podvezko, 2008). Finally, interdependencies are relationships among criteria which form a mutual effect. 3.2.2. Step 2: Pairwise comparison matrices In the second step, pairwise comparisons are performed on the elements within the clusters as they influence each cluster and on those that it influences, with respect to that criterion. The pairwise comparisons are made with respect to a criterion or sub-criterion of the controller layer and network layer. Thus, the importance weights of the factors are determined (Datta et al., 2009). In pairwise comparison, decision makers (eight experts on driving safety) compare two elements. Then, they determine the contribution of the factors to the result (Medineckiene and Björk, 2011). In ANP, similar to AHP, pairwise comparison matrices are formed using the 1–9 scale of relative importance proposed by Saaty (2005, 2001). The Fundamental Scale used for judgment is given in Table 3. Table 3 Fundamental scale. Value
Relationship between these two measures
1 3 5 7 9 2,4,6,8 Use reciprocals for inverse comparisons
Equal important Moderate important of one over another Strong or essential importance Very strong or demonstrated importance Extreme importance Intermediate values
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model
ARTICLE IN PRESS
AAP-3569; No. of Pages 8
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
The comparison matrix of controller layer and network layer in our research are shown in Eqs. (1) and (2) respectively. In order to describe the matrix better, we take matrix of controller layer as an example. In Acontroller layer , 1/3 means C2 is three times more important than C3 .
⎡ C1 C2 C3 ⎤ 1
AController layer = ⎣ 1/2 1/5
2 1
1/3
5
C1
3 ⎦ C2 1
(1)
C3
e11 e12 e13 1/1.2 1/7 1.4
5
3.3.1. Steps 1: Determine the factors set of safety evaluation for secondary task driving In this study, three factors (fixations, saccades, blinks) are selected as the safety evaluation factors and expressed as U: U = {u1 , u2 , u3 } =
fixations, saccades, blinks
3.3.2. Identify the comments set of safety evaluation for secondary task driving As the evaluation factors for driving safety are vague, the comments set V of the evaluation of factors are divided into five levels: e11
e14 e15 e21 e22 e23 e24 e31 e32 e33 1/2 1/2 1/5.8 1/7.6 1/5.2 1/3 1/6.6
⎡ 1 ⎤ e12 1/4 1 1/6.6 1.7 1/2 1/2.2 1/4.7 1/7.2 1/4 1/2.8 1/6 1/3.4 ⎥ e13 ⎢ 1.2 ⎢ ⎥ ⎢ 7 6.6 1 5 3 7 3.8 1/2 4.2 5 2.2 3.5 ⎥ e ⎢ ⎥ 14 ⎢ 1/1.4 1/1.7 1/5 1 2 1/2 1/4.4 1/6 1/5 1/2.5 1/4 1/3 ⎥ ⎢ ⎥ e15 ⎢ 2 2 1/3 1/2 1 1/5 1/6 1/5 1/5 1/3 1/4.3 1/3.5 ⎥ ⎢ ⎥ ⎢ 2 2.2 1/7 2 5 1 1/5 1/8 1/4.2 1/2 1/6.5 1/3 ⎥ e21 ⎥ ANetwork layer = ⎢ ⎢ 5.8 4.7 1/3.8 4.4 6 5 1 1/5 2.8 3 1/3 2.4 ⎥ ⎢ ⎥ e22 ⎢ 7.6 ⎥ 7.2 2 6 5 8 5 1 5 6.2 3 4 ⎢ ⎥ e23 ⎢ 5.2 4 1/4.2 5 5 4.2 1/2.8 1/5 1 2.8 1/3.4 2 ⎥ ⎢ ⎥ ⎢ 3 2.8 1/5 2.5 3 2 1/3 1/6.2 1/2.8 1 1/4 1/2.2 ⎥ ⎢ ⎥ e24 ⎣ 6.6 6 1/2.2 4 4.3 6.5 3 1/3 3.4 4 1 2.9 ⎦ e31 4
3.4
1/3.5
3
3.5
3
1/2.4
1/4
1/2
2.2
1/2.9
1
(2)
e32 e33
3.2.3. Step 3: Establish the weight coefficient For establishing the weight coefficient of e11 to e33 , twelve numbers in a row of the network layer matrix are multiplied and then calculate the product’s 12th root. On the basis of the calculation, the initialization weight vector Wi (1, 2, . . ., 12) is obtained. After normalizing the Wi , the weighting coefficient values Wi (1, 2, . . ., 12) of the evaluation indexes are calculated by the Eq. (3). Wi =
Wi 12
(3)
Wj
V = {v1 , v2 , v3 , v4 , v5 } =
3.3. Realize safety evaluation based on FCE method Fuzzy comprehensive evaluation is a method, which adopts the principle of fuzzy relationship synthesis, comprehensively judges the membership grade status of the things to be judged from many factors (Tuzkaya and Önüt, 2008). The basic steps of fuzzy comprehensive evaluation method in this study are as following (Janic and Reggiani, 2002; Lee et al., 2009):
very high, high, average, low, very low
Constructing the membership functions and classifying the secondary task driving safety. The membership function of a fuzzy set is a generalization of the indicator function in classical sets (Zolfani et al., 2012). In this study, the relationship between the membership degree and the driving safety score is not in a monotony trend. So it is suitable to solve this problem using middle type membership function patterns. Therefore, sine curve is selected as the membership function. The fuzzy analytical expression is as follows.
j=1
Through the above calculation, the obtained weight vector of controller layer is (0.58, 0.31, 0.11). The local weight vector of network layer is (0.02, 0.02, 0.19, 0.02, 0.02, 0.02, 0.11, 0.26, 0.08, 0.05, 0.15, 0.07). According to the proportional allocation principle, the global weight of each evaluation index is G = (0.035, 0.035, 0.332, 0.035, 0.035, 0.019, 0.103, 0.243, 0.075, 0.017, 0.050, 0.023). After performing a consistency check on judgment matrixes, satisfactory agreements of the judgment matrixes are obtained. So the judgment matrixes are acceptable.
r1 (u) =
r2 (u) =
r3 (u) =
r4 (u) =
⎧ ⎪ ⎪ ⎨1 2 ⎪ ⎪ ⎩
⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩ ⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩
1
sin
u − 0.1 0.2
1 + 2
+1
0
u − 0.3 0.2
1 + 2
+1
0
⎧ ⎪ ⎪ ⎨1 2 ⎪ ⎪ ⎩
u − 0.5 0.2
1 + 2
+1
0
sin
0.2 0
u ≤ 0.1 0.1 < u < 0.5
(5)
u ≤ 0.3 0.3 < u < 0.7
(6)
u ≥ 0.7
0
u − 0.7
(4)
u ≥ 0.5
0
1 sin 2
0.1 < u < 0.3 u ≥ 0.3
0
1 sin 2
0 ≤ u ≤ 0.1
1 + 2
+1
u ≤ 0.5 0.5 < u < 0.9
(7)
u ≥ 0.9
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
6
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
r5 (u) =
⎧ ⎪ ⎪ ⎨1 2 ⎪ ⎪ ⎩
0
sin
u − 0.9 0.2
1 + 2
+1
u ≤ 0.7 0.7 < u < 0.9
driving safety levels are divided. So, driver1’s driving safety judgment set under normal driving condition is B1 , as follows. (8)
1 ≥ u ≥ 0.9
1
In these formulas: r1 to r5 represent the degree of secondary task driving safety belongs to ‘very low’, ‘low’, ‘average’, ‘high’, and ‘very high’ respectively. u is the value of criterion. Due to the maximum membership grade principle, the levels of secondary task driving safety could be determined.
3.3.3. Constructing the single factor assessment matrix for each driver In order to describe how the matrix constructed, we will take the driver 1 as an example. First of all, the data collected from driving simulator should be processed using Eq. (9). u11 =
ui − min(ui ) max(ui ) − min(ui )
(9)
In Eq. (9): max(ui )—the maximum of criteria value; min(ui )—the minimum of criteria value; ui —the criteria value of the present driver. Second, substituting the u11 into Eqs. (4)–(8), the single factor assessment matrix for each driver is determined. In accordance with this method, driver1’s single factor assessment matrix under normal driving condition is obtained and shown in the matrix R1 .
⎡ 0 ⎤ 0 0.13 0.87 0 0 0 0.77 0.23 ⎥ ⎢ 0 ⎢ ⎥ ⎢ 0 0 0.11 0.89 0 ⎥ ⎢ ⎥ ⎢ 0 0 0.94 0.06 0 ⎥ ⎢ ⎥ ⎢ 0 0 0.54 0.46 0 ⎥ ⎢ ⎥ ⎢ 0 0.22 0.78 0 0 ⎥ ⎢ ⎥ R1 = ⎢ 0 0 ⎥ ⎢ 0.31 0.69 0 ⎥ ⎢ 0 0.77 0.23 0 0 ⎥ ⎢ ⎥ ⎢ 0 0 0 0.88 0.12 ⎥ ⎢ ⎥ ⎢ 0 0 0 0.74 0.26 ⎥ ⎢ ⎥ ⎣ 0 0 0 0.69 0.31 ⎦ 0
0
0.27
0.73
⎡ 0.035 ⎤T ⎡ 0 ⎤ 0 0.13 0.87 0 0 0 0.77 0.23 ⎥ ⎢ 0.035 ⎥ ⎢ 0 ⎢ ⎥ ⎢ ⎥ ⎢ 0.332 ⎥ ⎢ 0 0 0.11 0.89 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.035 ⎥ ⎢ 0 0 0.94 0.06 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.035 ⎥ ⎢ 0 0 0.54 0.46 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.019 ⎥ ⎢ 0 0.22 0.78 0 0 ⎥ ⎥ ⎢ ⎥ B1 = G • R1 = ⎢ ⎢ 0.103 ⎥ ⎢ 0.31 0.69 0 0 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.243 ⎥ ⎢ 0 0.77 0.23 0 0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.075 ⎥ ⎢ 0 0 0 0.88 0.12 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0.017 ⎥ ⎢ 0 0 0 0.74 0.26 ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ 0.050 ⎦ ⎣ 0 0 0 0.69 0.31 ⎦ 0.023 0 0 = (0.032, 0.262, 0.170, 0.511, 0.037)
0.27
0.73
0
The maximum value of B1 is 0.511. According to the membership function, the score 0.511 belongs to ‘high’ level of driving safety. Similarly, the driving safety levels of other drivers under different driving conditions could be calculated. 4. Results and discussion Based on the evaluation results, we classify and analyze statistically according to driving conditions, driver’s driving experience, and types of secondary task. Concerning the driving experience division, we learn from the study of Yingshi Guo (Yingshi Guo, 2009) and classify driving experience into three groups: experienced driver (mileage is equal or greater than 5 million kilometers), general experienced driver (mileage is between 2 and 5 million kilometers) and inexperienced driver (mileage is equal or less than 2 million kilometers). Tables 4–7 are statistical data of the influence of driving conditions, types of secondary task and driver’s driving experience on driving safety. 4.1. Secondary task driving has significant influence on driving safety
0
In this way, the single factor assessment matrixes of drivers from 1 to 40 under different driving conditions are obtained.
3.3.4. Determining the driving safety levels The driving safety judgment sets are determined by multiplying the global weight of each criterion G and single factor assessment matrix R. With the maximum membership grade principle, the
From Tables 4 and 5, we can draw the conclusion that secondary task driving has significant influence on driving safety. ‘Average’ and above of driving safety levels under normal driving condition (93.0%) are higher than driving with bluetooth headset conversation (59.5%), laboratory assistant talking (70.2%), radio tuning (27.0%), and touch-screen touching (10.0%). Overall, during the secondary task driving, ‘Average’ and above of driving safety levels account for only 41.6%, while ‘Low’ and ‘Very low’ of driving safety levels are responsible for 58.4%. Through analysis for test data, it can be seen that secondary task driving, regardless of secondary task type, has significant influence on driving safety.
Table 4 Statistics of drivers’ driving safety levels under different driving conditions. Driving condition
Normal Bluetooth headset conversation Talking with laboratory assistant Radio tuning Touch-screen touching (telephone and Ipad4)
Percent of driving safety level Very high
High
Average
Low
Very low
42.0% 2.5% 6.2% 0% 0%
25.5% 17.7% 19.0% 4.0% 3.2%
25.5% 39.3% 45.0% 22.5% 6.9%
7.0% 27.5% 19.8% 61.2% 50.5%
0% 13.0% 10.0% 12.3% 39.4%
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model
ARTICLE IN PRESS
AAP-3569; No. of Pages 8
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
7
Table 5 Comparison results of driving safety levels under normal driving and secondary task driving conditions. Driving condition
Normal Secondary task driving
Percent of driving safety level Very high
High
Average
Low
Very low
42.0% 2.2%
25.5% 11.0%
25.5% 28.4%
7.0% 39.8%
0% 18.7%
Table 6 The influences of secondary task type on driving safety levels. Driving condition
Normal Cognitive tasks Visual tasks
Percent of driving safety level Very high
High
Average
Low
Very low
43.0% 4.4% 0%
22.0% 18.4% 3.6%
27.5% 42.2% 14.7%
7.5% 23.7% 55.9%
0% 11.5% 25.9%
Table 7 The influence of driver’s driving experience on driving safety levels. Driving condition
Driving experience
Percent of driving safety level Very high
High
Average
Normal driving
Rich General Bad Rich General Bad
61.5% 40.6% 20.8% 4.5% 0% 0%
19.0% 24.4% 23.2% 19.5% 3.5% 3.1%
19.5% 30% 33.8% 45.5% 25.5% 14.3%
Secondary task driving
4.2. Visual tasks have a greater impact on driving safety than cognitive tasks It is evident from the results in Table 6 that driving with visual tasks (‘Low’ and ‘Very low’ account for more than 81.0%) are dangerous than with cognitive tasks (‘Low’ and ‘Very low’ account for about 35.2%). Additionally, the security of driving with cognitive tasks is also lower than normal driving. 4.3. Driver’s driving experience has significant influence on driving safety
Low
Very low
0% 5% 22.2% 29.0% 55.0% 46.0%
0% 0% 0% 1.5% 16.0% 36.6%
driving, secondary task types and driver’s driving experience have influence on the driving safety obviously. Secondary task driving can decrease the driving safety significantly. In conclusion, the proposed approach is useful and effective to classify the safety level of the secondary task driving. Secondary task driving has bad effects on driving safety and drivers should enhance their consciousness away from interacting with secondary tasks while driving. The government should pay more attention to this issue and take action to prevent traffic accident caused by these secondary tasks. Acknowledgements
The statistic analysis indicates that, in any driving conditions, the more experienced the driver has, the stronger the ability keeps the driving safety. Under normal driving condition, driving safety level of experienced drivers is the highest (‘Average’ and above account for about 100%). The same situation applies to secondary task driving. Experienced drivers driving with secondary task can still keep good driving safety (‘Average’ and above account for about 69.5%) while inexperienced drivers cannot (‘Average’ and above account for only 17.4%). 5. Conclusions This study aimed to determine the driving safety level of secondary task driving by using approach of Analytic Network Process. The proposed approach takes into account the drivers’ fixations, saccades, and blinks safety with their controller layer. The weights of the evaluation indexes are determined by eight driving safety experts. After determining the weights of the evaluation indexes and constructing the single factor assessment matrix for each driver, the maximum membership grade principle is applied to divide the safety level of secondary task driving. Finally, the classification is implemented. Data of forty drivers engaging in driving simulating are used for verifying the applicability of the secondary task driving safety evaluation system. The analysis results showed that secondary task
This research was partially supported by Doctoral Study Special Research Foundation of Higher Education (20110061110036), Major Projects of JiLin Science and Technology Department (20116017), New Century Excellent Talent Foundation Program (NCET-10-0435) and Talent Development Foundation Projects of JiLin (801121100417). References Aghdaie, M.H., Zolfani, S.H., Zavadskas, E.K., 2013. Market segment evaluation and selection based on application of fuzzy AHP and COPRAS-G methods. J. Bus. Econ. Manage. 14 (1), 213–233. Crundall, D., 2009. The deceleration detection flicker test: a measure of experience? Ergonomics 6 (52), 184–674. Das, M.C., Sarkar, B., Ray, S., 2012. A framework to measure relative performance of Indian technical institutions using integrated fuzzy AHP and COPRAS methodology. Socio Econ. Plan. Sci. 46 (3), 230–241. Datta, S., Beriha, G.S., Patnaik, B., Mahapatra, S.S., 2009. Use of compromise ranking method for supervisor selection: a multi criteria decision making (MCDM) approach. IJVTE 1 (1), 7–13. Ginevicius, R., Podvezko, V., 2008. Multi-criteria evaluation of Lithuanian banks from the perspective of their reliability for clients. J. Bus. Econ. Manage. 9 (4), 257–267. Gumus, A.T., Yilmaz, G., 2010. Sea vessel type selection via an integrated VAHP-ANP methodology for high-speed public transportation in Bosphorus. Expert Syst. Appl. 37 (6), 4182–4189. Janic, M., Reggiani, A., 2002. An application of the multiple criteria decision making (MCDM) analysis to the selection of a new hub airport. EJTIR 2 (2), 113–141. Jiang, Y.Y., Wu, Z., Jin, L.S., Hou, H.J., 2012. Driver inattention monitoring system based on machine vision. Information 15 (12A), 5491–5498.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013
G Model AAP-3569; No. of Pages 8
ARTICLE IN PRESS
8
L. Jin et al. / Accident Analysis and Prevention xxx (2014) xxx–xxx
Jin, L.S., Xian, H.C., Hou, H.J., Niu, Q.N., Lv, H.H., 2012a. Analyzing effects of touchscreen on driving performance. J. Beijing Inst. Technol. Suppl 2 (21), 6–10. Jin, L.S., Niu, Q.N., Hou, H.J., Xian, H.C., Wang, Y.L., Shi, D.D., 2012b. Driver cognitive distraction detection using driving performance measures. Discrete Dyn. Nat. Soc. 1, 1–12. Hou, H.J., Jin, L.S.H., Niu, Q.N., Sun, Y.Q., 2011. Driver intention recognition method using continuous hidden markov model. IJCIS 4 (3), 386–393. Xian, H.C., Jin, L.S., Hou, H.J., et al., 2013a. Analyzing effects of pressing radio button on driver’s visual cognition. CSIP2013 215, 69–78. Xian, H.C., Jin, L.S., Hou, H.J., et al., 2013b. Research on in-vehicle secondary task analysis model based on preview theory. In: The 2nd International Conference on Transportation Information and Safety (ICTIS 2013), pp. 1282–1287. Khushaba, R.N., Kodagoda, S., Liu, D., Dissanayake, G., 2013. Muscle computer interfaces for driver distraction reduction. Comput. Meth. Prog. Bio. 2 (110), 137–149. Liang, X.Y., Sun, X.X., Shu, G.Q., Sun, K., Wang, X., Wang, X.L., 2013. Using the analytic network process (ANP) to determine method of waste energy recovery from engine. Energ Convers. Manage. 66, 304–311. Lee, H., Kim, C., Cho, H., Park, Y., 2009. An ANP-based technology network for identification of core technologies: a case of telecommunication technologies. Expert Syst. Appl. 36 (1), 894–908. Medineckiene, M., Björk, F., 2011. Owner preferences regarding renovation measures - the demonstration of using multi-criteria decision making. J. Civil Eng. Manage. 17 (2), 284–295. Nevile, M., Haddington, P., 2010. In-car distractions and their impact on driving activities. Department of Infrastructure and Transport, Road Safety Grant, Report RSGR-2010-001. PRC Road Traffic Law on State Security, 2004. Chapter IV provisions on road passage. Article 62: Driver may do any of the following. http://baike.baidu.com/ view/20045.htm#5
Saaty, T.L., 2001. Decision Making with dependence and Feedback, 2nd Ed. RWS Publication, Pittsburg. Saaty, T.L., 2005. Theory and Applications of the Analytic Network Process: Decision Making with Benefits, Opportunities, Costs, and Risks. RWS Publications, Pittsburg. Safety Laws in California, 2012. Cell phone laws–Text messaging–GPS and other map apps. http://www.dmv.org/ca-california/safety-laws.php#Cell-Phone-Laws Tuzkaya, U.R., Önüt, S., 2008. A fuzzy analytic network process based approach to transportation-mode selection between Turkey and Germany: a case study. Inform. Sci. 178 (15), 3133–3146. Wang, J.Q., Zhang, L., Zhang, D., Zh Li, K.Q., 2013. An adaptive longitudinal driving assistance system based on driver characteristics. IEEE T INTELL TRANSP 14 (1), 1–12. World Health Organization, 2011. Mobile phone use: a growing problem of driver http://www.who.int/violence injury prevention/publications/ distraction. road traffic/en/index.html Yang, Y., McDonald, M., Zheng, P.J., 2010. The Effects of increasing workload on driver’s eye movement. The 12th WCTRS. World Conference on Transport Research Society, 11–15. Yingshi Guo, 2009. Study on effects of traffic environment and driving experience on driver’s eye movement and workload, A dissertation for degree of doctor, China, Xi’an Chang’an University. Zhang, Y., Harris, E., Rogers, M., Kaber, D., Hummer, J., Rasdorf, W., Hu, J., 2013. Driver distraction and performance effects of highway logo sign design. Appl. Ergon. 3 (44), 472–479. Zolfani, S.H., Chen, I.S., Rezaeiniya, N., Tamosaitiene, J., 2012. A hybrid MCDM model encompassing AHP and COPRAS-G methods for selecting company supplier in Iran. Technol. Econ. Dev. Econ. 18 (3), 529–543.
Please cite this article in press as: Jin, L., et al., Research on safety evaluation model for in-vehicle secondary task driving. Accid. Anal. Prev. (2014), http://dx.doi.org/10.1016/j.aap.2014.08.013