Improving motorcycle conspicuity through innovative headlight configurations

Accident Analysis and Prevention 94 (2016) 119–126

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Improving motorcycle conspicuity through innovative headlight configurations Maud Ranchet a,b,∗ , Viola Cavallo a , Nguyen-Thong Dang a , Fabrice Vienne c a b c

IFSTTAR, LEPSIS, Versailles, France Department of Physical Therapy, Georgia Regents University, Augusta, GA, USA IFSTTAR, LEPSIS, Marne la Vallée, France

a r t i c l e

i n f o

Article history: Received 29 September 2015 Received in revised form 4 May 2016 Accepted 12 May 2016 Keywords: Motorcycle Conspicuity Headlight ergonomics

a b s t r a c t Most motorcycle crashes involve another vehicle that violated the motorcycle’s right-of-way at an intersection. Two kinds of perceptual failures of other road users are often the cause of such accidents: motorcycle-detection failures and motion-perception errors. The aim of this study is to investigate the effect of different headlight configurations on motorcycle detectability when the motorcycle is in visual competition with cars. Three innovative headlight configurations were tested: (1) standard yellow (central yellow headlight), (2) vertical white (one white light on the motorcyclist’s helmet and two white lights on the fork in addition to the central white headlight), and (3) vertical yellow (same configuration as (2) with yellow lights instead of white). These three headlight configurations were evaluated in comparison to the standard configuration (central white headlight) in three environments containing visual distractors formed by car lights: (1) daytime running lights (DRLs), (2) low beams, or (3) DRLs and low beams. Video clips of computer-generated traffic situations were displayed briefly (250 ms) to 57 drivers. The results revealed a beneficial effect of standard yellow configuration and the vertical yellow configuration on motorcycle detectability. However, this effect was modulated by the car-DRL environment. Findings and practical recommendations are discussed with regard to possible applications for motorcycles. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction Despite the general improvement in road safety in Europe over the past few years, motorcycle1 safety is still a major concern. Accident statistics have shown that motorcyclists are among the most vulnerable road users (NHTSA, 2012). In France, fatalities of motorized two-wheeler drivers represent 24% of all motorcyclists whereas they only represent 1.9% of all road users (ONISR, 2014). In the US, the risk of a fatal accident per kilometer travelled is 22 times higher for motorcyclists than for automobilists (NHTSA, 2012). Typically, motorcycle crashes occur at intersections during the day and are often caused by right-of-way violations by another vehicle approaching head-on (MAIDS, 2009; Pai, 2011; Shaheed et al., 2013). Two kinds of perceptual errors contribute to the risk of such crashes: motorcycle-detection failures and/or motion-

∗ Corresponding author at: Augusta University, College of Allied Health Sciences, Department of Physical Therapy, 1120 15th Street, Augusta GA 30912, USA. E-mail addresses: [email protected], [email protected] (M. Ranchet). 1 By motorcycles, we mean all powered two-and three-wheel vehicles. http://dx.doi.org/10.1016/j.aap.2016.05.011 0001-4575/© 2016 Elsevier Ltd. All rights reserved.

perception errors. In-depth studies of motorcycle accidents have noted a high frequency of detection errors (Hurt et al., 1981; Van Elslande and Jaffard, 2010; Pai, 2011). These errors are often referred to as “looked but failed to see” errors, where the car driver reports having looked in the right direction but having not seen the approaching motorcycle (MAIDS, 2004, 2009). This failure may be partly explained by the poor visual conspicuity of motorcycles, which results from their small frontal size, their irregular contour, and their often dark color. Visual conspicuity can be defined as the ability of an object to attract attention by means of its physical characteristics (Connors, 1975), such as size, brightness, color, outline, and motion. Environmental factors also come into play, as shown by several studies (Olson et al., 1981; Cavallo and Pinto, 2012; Gershon et al., 2012): the visual context determines the figure-background relationship and thus the motorcycle’s brightness and color contrast levels. It may also contain competing visual elements that act as distractors. In order to improve motorcycle safety, European motorcycle manufacturers will be using “digital conspicuity”, i.e., intelligent transportation systems (ITSs) based on vehicle-to-vehicle communication. These technological solutions will solve problems related

120

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

to the limitations of visual perception and attention, but such systems are not likely to be operational for at least 15 years. Until reliable systems are available and all vehicles are equipped with them, other solutions are needed to improve motorcycle safety. Within the past few decades, the use of daytime running lights (DRLs) by motorcyclists has been shown to improve the conspicuity of motorcycles (Olson et al., 1981; Smither and Torrez, 2010) and reduce motorcycle accidents (e.g., Muller, 1984; Zador, 1985). However, the now widespread use of DRLs on cars reduces the effectiveness of DRLs on motorcycles (Brendicke et al., 1994; Cavallo and Pinto, 2012). While car DRLs in the past consisted of turning on low beams during the day, dedicated DRLs with defined technical characteristics have become compulsory in many countries (in Europe since February 2011). Automobile manufacturers and suppliers presently favor the use of LED (light emitting diode) technology in designing car DRLs, not only to improve visibility but also to display their own visual signature. Having diversified shapes for car headlights might create “visual noise” for motorcycles and hamper their detection. In some countries, car DRLs are even turned on at the same time as low beams, thereby further increasing the number of potentially competing light sources. In this context, Rößger and Lenné (2015) synthesized recent results on motorcycle conspicuity and discussed the various methods employed to assess the conspicuity of powered two-wheelers. Several studies conducted within the last decade have attempted to find new means of improving the visual conspicuity of motorcycles using innovative headlight configurations that increase the apparent size of motorcycles and/or provide them with recognizable visual features (Maruyama et al., 2009; Rößger et al., 2012; Gershon and Shinar, 2013; Pinto et al., 2014). Different methods such as photographs, computer graphics simulation, and video clips, as well as different visibility conditions (nighttime, dusk or daytime) have been used, which could explain why the results are sometimes discrepant. Rößger et al. (2012) used photographs of daytime traffic scenes at intersections and found that subjects recognized a motorcycle with a T-shaped headlight configuration (additional lights on the fork and handlebars, forming a T) more quickly than with a standard headlight. Gershon and Shinar (2013) examined the effect of an alternating blinking-light system (ABLS) using video clips shot at daytime and at dusk. The ABLS was a 24-cm high helmet-mounted device with two vertically positioned lights that blinked alternately and created apparent motion. The findings showed that the ABLS increased motorcycle detection rates, especially at dusk. Maruyama et al. (2009) tested a much simpler headlight arrangement with a triangular design (one standard light and two additional lights on the rear view mirrors) called “face design”. The effectiveness of this configuration on motorcycle detection was observed in nighttime conditions using computer graphics simulation. However, Pinto et al. (2014) found no benefit of this configuration under more realistic conditions where photographs of complex urban traffic scenes (presence of cars with DRLs) and daytime visual conditions were used, probably because the additional lights on the motorcycle were confused with headlights of cars. Rather than with the triangular configuration, Pinto et al. (2014) showed significant benefits when the motorcycles had yellow headlights, and also when the motorcyclists had an additional light on their helmet. It can be assumed that color coding and/or a high-mounted (helmet) light represent quite simple, ergonomically realistic solutions likely to be accepted more easily by motorcycle riders than the T-shape light configuration (Rößger et al., 2012) or the ABLS (Gershon and Shinar, 2013). Other studies have also addressed the impact of innovative headlight configurations on the perception of an approaching motorcycle’s speed and time-to-arrival (Tsutsumi et al., 2008; Gould et al., 2012a,b). A recent study (Cavallo et al., 2015) pointed

out that increasing the vertical dimension of the motorcycle had a beneficial effect on the perception of the approaching motorcycle’s motion: longer time gaps were accepted for motorcycles equipped with this vertical configuration, especially in low-visibility conditions where the approach speed was high. Based on previous research into motorcycle detectability (Pinto et al., 2014) and perception of motorcycle motion (Cavallo et al., 2015), we assume that innovative motorcycle headlight configurations can be designed that benefit both kinds of visual mechanisms and thus globally improve motorcycle perceptibility. While the present experiment is devoted specifically to motorcycle conspicuity and detectability, the choice of configurations to be evaluated also took into account their potential for improving the motorcycle’s speed and time-to-arrival. Our experimental approach consisted of testing three innovative motorcycle headlight configurations in visually complex car-headlight environments in daytime conditions. The configurations, which involve color coding and/or accentuation of the motorcycle’s vertical dimension, were: (1) the “standard yellow” configuration, consisting of a central yellow headlight instead of a conventional white one; (2) the “vertical white” configuration, combining a central white headlight, an additional white light on the helmet, and two additional white lights on the fork; (3) the “vertical yellow” configuration, combining a central yellow headlight, an additional yellow light on the helmet, and two additional yellow lights on the fork. These innovative configurations were compared to a “standard white” central headlight. The four motorcycle headlight configurations were evaluated in three car headlight environments: cars with DRLs only on, cars with low beams only turned on, and cars with both DRLs and low beams turned on. We used a target detection task under time constraints. Short computer-generated video clips were presented to the participants. Compared to static images, these dynamic sequences added visual motion cues likely to affect conspicuity (Itti and Baldi, 2005). The visual targets were vulnerable road users (cyclists, pedestrians, and motorcyclists). The visual targets’ distance and eccentricity in the visual scene were also varied because it has been shown that these variables affect motorcycle detection performance (Engel, 1971; Rogé and Pébayle, 2009; Cavallo and Pinto, 2012; Pinto et al., 2014). We expected the innovative headlight configurations to improve motorcycle detection, and the visually complex car-light environments to be detrimental to motorcycle conspicuity.

2. Method 2.1. Participants A total of 57 adults (12 women and 45 men) participated in the experiment. They were divided into three groups of 19 participants. Each group of participants had to detect vulnerable road users in a different car-light environment. The groups were matched on age (Group 1: 30.78 ± 7.42; Group 2: 30.79 ± 6.41; Group 3: 29.71 ± 6.01, p = 0.853), gender (15 men and 4 women) and driving experience (Group 1: 10.79 ± 6.45; Group 2: 11.21 ± 6.29; Group 3: 9.01 ± 6.31, p = 0.530). All participants were licensed drivers and had normal or corrected-to-normal vision (at least 6/10 binocular acuity). All participants underwent the useful field of view (UFOV) test (Ball et al., 1993) and all exhibited normal visuoattentional performance. Each of the three groups included two or three motorcyclists. Written informed consent was obtained before participation in the study.

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

121

Fig. 1. The four motorcycle headlight configurations: a) standard white, b) standard yellow, c) vertical white, and d) vertical yellow. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3. Mean and standard error for the motorcycle detection rate as a function of motorcycle headlight configuration (*** p< < 0.001).

Fig. 2. Screen views of the three car-DRL environments: a) DRLs only, b) low beams only, c) both DRLs and low beans.

2.2. Task Participants had to detect the presence or absence of vulnerable road users (VRUs) in an urban environment presented for 250 ms in a sequence of generated video images. The VRUs could be motorcyclists, bicyclists, or pedestrians. If one was detected, the participant had to identify it.

Fig. 4. Mean and standard error for the motorcycle detection rate as a function of car-headlight environment (*p < 0.05).

making a total of 72 experimental trials per block. Eighteen distractors containing a VRU (located at different distances from those used in the experimental trials) were added. In order to obtain a balanced design, 90 more trials without a target (50% of the trials) were added, so as to end up with 180 trials per block.

2.3. Experimental design 2.4. Apparatus and stimuli A mixed experimental design was employed, with four withinsubject variables and one between-subject variable. The four within-subject variables were as follows: - The motorcycle headlight configuration (standard white, standard yellow, vertical white, and vertical yellow) (Fig. 1). - The VRU (motorcyclist, bicyclist, or pedestrian) as the visual target; three types of VRUs were chosen to vary the visual situation. - The distance of the VRU in the scene (far vs. near). - The VRU’s eccentricity (central vs. off-centered); for the offcentered condition, the VRU was located on the right side of the visual scene in 50% of the trials and on the left side in the other 50%. The between-subject variable was the car-headlight environment: DRLs only, low beams only, both DRLs and low beams (Fig. 2). For each of the car-headlight environments, four blocks of 180 trials were presented, making a total of 720 trials per participant. Each block comprised one motorcycle headlight configuration and included 12 experimental conditions resulting from the combination of the distance, eccentricity, and vulnerable road-user variables. The 12 experimental conditions were repeated six times,

The video clips were generated by in-house software developed at IFSTTAR. Image generation (60 Hz) included high dynamic range (HDR) rendering. The videos were presented on a 47 flatpanel LCD display, which offered high-level light rendering features (luminance level: 4000 cd/m2; contrasts: >20,000:1; 1920 × 1080 resolution). Participants sat in a small-scale interactive driving simulator at a distance of 160 cm from the screen. The visual scenes showed urban traffic at intersections in daytime conditions. They depicted a number of cars in four car lanes approaching the intersection at a constant speed of 50 km/h, and a single vulnerable road user moving in the driving scene. Target distance was controlled by the VRU’s height in the image at the beginning of the video clip: 6 cm (angular size 2.15◦ ) and 4 cm (angular size 1.43◦ ) for the near and far conditions, respectively. For the eccentricity variable, a target was considered central if it was located in the central third of the visual scene and off-centered if it was located on the left or right side of the screen. Computergraphics simulation was used to create the car headlights (DRLs and low beams) and the four motorcycle headlight configurations. The DRLs on the cars simulated LEDs with different shapes, similar to what we find on the recent cars (Fig. 2a).

122

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

Table 1 Response rate (in %) for the motorcycle headlight configurations in each car-headlight environmement. GROUP 1: DRLs on Motorcycle stimuli Standard white Detections

Non-Detections

No-target distractors Standard yellow

Correct detections 88.16 62.94 Identification errors 0.44 3.07 Misses 33.99

11.40

Vertical white

Vertical yellow

Standard white

Standard yellow

Vertical white

Vertical yellow

63.82

93.20

False alarms 1.70

1.11

2.81

1.46

3.07

0.88

33.11

5.92

Correct rejections 98.30

98.89

97.19

98.54

GROUP 2: Low beams on Motorcycle stimuli Standard white Detections

Non-Detections

No-target distractors Standard yellow

Correct detections 77.41 60.31 Identification errors 2.41 3.29 Misses 36.40

20.18

Vertical white

Vertical yellow

Standard white

Standard yellow

Vertical white

Vertical yellow

63.60

91.23

False alarms 3.10

2.40

2.92

2.28

2.41

0.66

33.99

8.11

Correct rejections 96.90

97.60

97.08

97.72

GROUP 3: DRLs and low beams on Motorcycle stimuli Standard white Detections

Non-Detections

No-target distractors Standard yellow

Correct detections 70.39 55.26 Identification errors 4.39 3.95 Misses 40.79

25.22

Vertical white

Vertical yellow

Standard white

Standard yellow

Vertical white

Vertical yellow

58.77

82.02

False alarms 3.39

2.28

3.51

2.75

4.82

2.41

36.40

15.57

Correct rejections 96.61

97.72

96.49

97.25

2.5. Procedure After giving their informed consent, all participants performed the visual acuity test (Ergovision) and the UFOV test (Ball et al., 1993) and then read the task instructions. A series of 12 practice trials per block were given to familiarize the participants with the task. Then they performed four blocks of 180 trials each. The order of the blocks was counterbalanced across participants. Each trial included a 1500-ms long cross displayed in the middle of a gray screen, which the participants had to fixate, followed by the video clip. The participants had to give an oral response and press the button to go on to the next trial. No feedback on response accuracy was given. Participants were allowed breaks during and between the blocks. The entire experimental session lasted approximately 1 h and 30 min.

Table 2 GEE analysis results. Variables

Chi-squared

p-value

Motorcycle headlight configuration Car-DRL environment Distance Eccentricity Motorcycle headlight × Car-DRL environment Motorcycle headlight × Excentricity Motorcycle headlight × Distance Motorcycle headlight × Distance × Excentricity

51.067 12.189 0.530 98.035 13.656 10.654 9.005 2.522

0.000 0.002 0.466 0.000 0.034 0.014 0.029 0.641

to the analysis of pairwise comparisons. The ␣-level was set at .05 for all statistical analyses.

3. Results 2.6. Data analysis 3.1. Descriptive results Correct detections, identification errors, false alarms, and misses were determined for the motorcycle stimuli, and false alarms and correct rejections were counted for the no-target distractor trials. Correct detections were further analyzed using the generalized estimating equations (GEE) method which takes the dependency of the repeated measurements into account. Because of the binary structure of the dependent variable, a logit link function and a binomial distribution were chosen. The structure of the correlation matrix was set at exchangeable. The results were obtained using SPSS 21.0 software. Main effects and interactions given by the GEE procedure are detailed below. Bonferroni corrections were applied

The response rate for the motorcycle headlight configurations in each car-headlight environment are described in Table 1. The d values for correct detections was between 1.96 and 3.47, indicating that the participants’ responses in each group were likely to differ significantly from chance.

3.2. GEE analysis Table 2 shows the GEE analysis results.

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

123

Fig. 5. Mean and standard error for the motorcycle detection rate as a function of motorcycle headlight configuration and car-headlight environment (***p < 0.001; **p < 0.01; *p < 0.05).

3.2.1. Effect of motorcycle headlight configuration The GEE analysis revealed a significant effect of motorcycle headlight configuration on the correct detection rate (Fig. 3). Pairwise comparisons showed that both the yellow vertical and the yellow standard configurations were detected significantly better than were the white standard or white vertical configurations (89%, 79%, 60%, and 62%, respectively, p < 0.001). No significant correct detection differences were observed between the white vertical configuration and the white standard configuration, nor between the yellow vertical configuration and the yellow standard configuration. 3.2.2. Effect of car-headlight environment The GEE results indicated a significant effect of car-headlight environment on the correct detection rate (Fig. 4). Pairwise comparisons showed that participants who saw the car-headlight environment with DRLs only detected significantly more motorcyclists than did those who saw the car-headlight environment with both low beams and DRLs (77% vs. 67%, p < 0.05). No significant differences were found between the car-headlight environment with low beams only and the other two environments. 3.2.3. Effects of distance and eccentricity No significant effect of distance was observed: the detection rate for participants was 72% when the motorcyclists were located in far-away positions and 72% when the motorcyclists were located in near positions. The findings indicated a significant main effect of eccentricity: motorcycles were detected more often when they were located in center (95%) of the visual scene than when they were off-centered (49%). No significant interaction between distance and eccentricity was observed. 3.2.4. Effect of motorcycle headlight configuration as a function of car-Headlight environment The significant interaction between motorcycle headlight configuration and car-headlight environment indicates that the effect of motorcycle headlight configuration was significant when cars had only their DRLs or only their low beams on (Fig. 5). When the cars had their DRLs on, both the vertical yellow and standard yellow configurations were detected significantly better than

standard white configuration (93% vs.. 63% p < 0.01; 88% vs. 63%, p < 0.01; respectively) or the vertical white configuration (93% vs. 64%, p < 0.01; 88 vs. 64%; p < 0.001; respectively). When the cars had their low beams on, however, only the vertical yellow configuration was detected significantly better than the standard white configuration (91% vs. 60%, p < 0.001). When the cars had both their DRLs and their low dipped beams on, no significant differences between the standard yellow configuration (70%) or the vertical yellow configuration (82%) and the standard white configuration (55%) were found. The standard yellow configuration and the vertical yellow configuration differed significantly from the vertical white configuration only (70% vs. 59%, p < 0.05; 82% vs. 59%, p < 0.01, respectively). 3.2.5. Effect of motorcycle headlight configuration as a function of eccentricity The significant interaction between motorcycle headlight configuration and eccentricity showed that the effect of motorcycle headlight configuration was significant only when the motorcycle was located in an off-centered position in the visual scene (p < 0.001) (Fig. 6). Both the vertical yellow configuration (80%) and the standard yellow configuration (59%) were detected significantly more often than the standard white configuration (29%) when the motorcycle was off-centered. No significant differences between the four headlight configurations were found when the motorcycle was located in the center of the visual scene. 3.2.6. Effect of motorcycle headlight configuration as a function of distance The significant interaction between motorcycle headlight configuration and distance revealed that the vertical yellow and the standard yellow configurations were detected significantly better than the standard white configuration in both near (87%, 79%, and 60%, respectively, p < 0.001) and far (91%, 78%, and 58%, respectively, p < 0.001) locations (Fig. 7). Both yellow configurations were also detected significantly more often than was the vertical white configuration, in both near (64%, p < 0.001) and far locations (60%, p < 0.001). No significant differences were observed between the near and far positions for any of the four motorcycle headlight configurations.

124

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

Fig. 6. Mean and standard error for the motorcycle detection rate as a function of eccentricity (***p < 0.001; *p < 0.05).

Fig. 7. Mean and standard error for the motorcycle detection rate as a function of distance (***p < 0.001).

4. Discussion The purpose of this study was to investigate the effect of three innovative headlight configurations on motorcycle detectability in visually complex, daytime environments. The results brought out the beneficial effect of both the standard yellow and the vertical yellow configurations on the detectability of motorcycles as compared to the standard white or the vertical white configurations. This finding suggests that the use of color coding (in the present study, yellow) specifically improved motorcycle identification by clearly distinguishing motorcycles from other light sources. These findings are in accordance with our previous research showing better detection of motorcycles when they are equipped with a yellow headlight rather than a conventional white headlight (Pinto et al., 2014). They are also in line with studies showing that color coding reduces search time and increases the number of correct answers (Christ, 1975; Nowell, 1997). The conspicuity advantage of the two yellow configurations was also shown to depend on the visual complexity of the car-headlight environment. Both yellow configurations improved motorcycle detection in situations where cars were equipped with DRLs whose shapes clearly differed from the round motorcycle headlights. When cars had their low beams on, the vertical yellow configuration was more effective than the standard yellow configuration. The vertical yellow configuration presented two additional conspicuity elements. Firstly, the additional light on the motorcyclist’s helmet introduced an element of dissimilarity in terms of the height in the visual field of the motorcycle and the other road users, which

has been proven effective (Pinto et al., 2014). Secondly, the addition of the two lights on the fork and the helmet light accentuated the upright position of the motorcycle, and so its shape may have acted as a recognizable visual signature. A similar effect of a vertical configuration was found in the study by Rößger et al. (2010), which showed that a T-shaped light configuration allowed subjects to recognize motorcycles in intersection scenes faster, particularly when the motorcycle was competing with other motorized road users for the subject’s attention. Among the three innovative configurations tested here, the vertical yellow configuration thus seems to have the greatest potential for improving motorcycle conspicuity. In the car-headlight environment where both DRLs and low beams were on at the same time, motorcycles were detected less well, as a whole, than when they were surrounded by cars equipped with DRLs only. In this complex visual environment, no significant detection-rate differences between the innovative configurations and the standard white configuration were found. The two yellow configurations obtained significantly better motorcycle detection rates only when compared to the vertical white configuration. This result suggests that the greater number of car-headlight sources in fact produced visual noise that was detrimental to motorcycle detection, even when the motorcycles were equipped with yellow headlight configurations. The standard yellow configuration and the vertical yellow configuration were both found to be particularly effective when the motorcycle was off-centered, whereas no effect of motorcycle distance was observed. In our previous work (Pinto et al., 2014), a beneficial effect of the standard yellow configuration was found only when the motorcycle was in the middle of the visual scene and in a far-away location. The discrepancies between the two studies could be due to methodological differences. In the present study, we used dynamic image sequences that provided motion information, whereas our previous study used static images, i.e., photographs of road scenes (Pinto et al., 2014). When discussing different methodological approaches to motorcycle conspicuity research, Rößger et al. (2015) pointed out that findings from static and dynamic images are often closely related, and that experiments can justifiably use static stimuli to study attentional aspects of road-user behavior. However, the authors also encourage researchers to look carefully at the generalizability of the results and to take additional approaches into account. The present experiment demonstrates the merits of using dynamic image sequences, and suggests that the motion information available may have improved the detection of the motorcycles, specifically in off-centered positions where angular velocities were higher than in centered positions. Motion cues are known to enhance object conspicuity (Itti and Baldi, 2005), and have also been shown to improve pedestrian detection accuracy (Papagiorgiou, 1997). Motorcycles moving in real-world situations are usually perceived in a dynamically changing environment and have motion features of their own. The present study thus points out the merits of using video clips rather than photographs to investigate innovative motorcycle headlight configurations. One of the advantages of the present study is that we used a high-quality screen and HDR technology, which provided better rendering of contrasts and reproduced higher levels of luminance than ordinary LCD screens do (Petit and Bremond, 2010). To our knowledge, this is the first study in the field of motorcycle conspicuity to use this recent technology. The results obtained are promising. However, despite the HDR rendering, the light sources presented on the screen were still not as luminous as in real-life conditions, so further research in the real world is needed to validate the findings. Considering that high contrasts and luminance levels favor conspicuity, our study suggests that the conspicuity effect of the two yellow configurations would even be stronger with real light sources. Further studies in real-world conditions are needed to determine the effects of the contrast and luminance

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

levels of the tested headlight configurations so as to increase the ecological validity of the recommended solutions. The most suitable approach for increasing validity seems to be the experimental method used by Olson et al. (1981), where conspicuity-enhancing devices were tested in realistic driving situations and the gaps between naïve drivers and the motorcycles were measured. While the present experiment only addressed the visual-detection and identification components of the driving task, the Olson method encompasses the whole perception-action loop and analyzes the behavioral consequences – in terms of accepted gaps – of the perceptual activity. This method is challenging, but likely to ensure optimal external validity.

5. Practical applications and research perspectives The findings of this research suggest that the vertical yellow configuration with two lights on the fork and one on the motorcyclist’s helmet is a promising solution for improving the detection and identification of motorcycles in daytime urban environments. This headlight arrangement has the advantage of being less complex than the T-configuration (Rößger et al., 2012) and more realistic than an alternating-blinking light on the helmet (Gershon and Shinar, 2013). It should also be preferred to the triangular configuration, whose effectiveness for motorcycle detection has been shown at night (Maruyama et al., 2009), but not during the day (Pinto et al., 2014). Another major advantage of the vertical yellow configuration is that it has also been shown to improve the perception of the approaching motorcycle’s movement and lead to accepted time gaps large enough to keep motorcyclists out of danger (Cavallo et al., 2015). Furthermore, the vertical yellow arrangement has proven to be more effective than the triangular configuration recommended by several researchers (Gould et al., 2012a) and motorcycle manufacturers (Maruyama et al., 2009): recent research suggests that the triangular configuration does not significantly enhance the gaps accepted for motorcycles (Cavallo et al., 2015). In the present state of our knowledge, the vertical yellow configuration appears to be the best ergonomic solution that simultaneously improves motorcycle detection and motion perception. In terms of applications, the introduction of innovative headlight configurations could have strong implications for motorcycle manufacturers in terms of producing safer products in the near future. However, with respect to the best theoretical solution found in the present study, i.e., the vertical yellow arrangement, it is probably not realistic to assume that motorcycles could be equipped with yellow headlights, since yellow lights are less efficient for lighting up the street and could also produce color distortions in road-sign perception. Instead, we recommend a motorcycle headlight configuration that combines a central white light and three additional yellow lights, one on the helmet and two on the fork. Using LED technology for these additional lights would be a good solution for limiting power demands. Future studies should test the effectiveness of such a mixed configuration on both motorcycle detection and motion perception. Until “electronic conspicuity” is available, innovative motorcycle headlight configurations could easily be implemented and could notably improve motorcycle safety in the coming years. Finally, our study attracts attention to the often neglected fact that motorcycle detectability also depends on environmental features and suffers from distracting light sources on cars, especially when cars simultaneously turn on their low beams and their DRLs. Better separation of these two functions, by way of an improvement in car lighting regulations could help make motorcycles easier to detect by other vehicle drivers.

125

Acknowledgment This work was financed by the French MAIF Foundation.

References Ball, K., Owsley, C., Sloane, M.E., Roenker, D.L., Bruni, J.R., 1993. Visual attention problems as a predictor of vehicle crashes in older drivers. Invest. Ophthalmol. Vis. Sci. 34, 3110–3123. Brendicke, R., Forke, E., Schäfer, D., 1994. Auswirkungen einer allgemeinen tageslichtpflicht auf die sicherheit motorisierter zweiräder. 6. Fachtagung‘motorrad’. VDI Berichte 1159, Düsseldorf, Germany, pp. 283–318. Cavallo, V., Pinto, M., 2012. Are car daytime running lights detrimental to motorcycle conspicuity? Accid. Anal Prev. 49 (0), 78–85. Cavallo, V., Ranchet, M., Pinto, M., Espié, S., Vienne, F., Dang, N.-T., 2015. Improving car drivers’ perception of motorcycle motion through innovative headlight configurations. Accid. Anal. Prev. 81, 187–193. Christ, R.E., 1975. Review and analysis of color coding research for visual displays. Hum. Factors 17 (6), 542–570. Connors, M.M., 1975. Conspicuity of Target Lights: The Influence of Color. NASA, Washington, DC. Engel, F.L., 1971. Visual conspicuity directed attention and the retinal locus. Vision Res. 17, 95–108. Gershon, P., Shinar, D., 2013. Increasing motorcycles attention and search conspicuity by using alternating-blinking lights system (abls). Accid. Anal. Prev. 50 (0), 801–810. Gershon, P., Ben-Asher, N., Shinar, D., 2012. Attention and search conspicuity of motorcycles as a function of their visual context. Accid. Anal Prev. 44 (1), 97–103. Gould, M., Poulter, D.R., Helman, S., Wann, J.P., 2012a. Errors in judging the approach rate of motorcycles in nighttime conditions and the effect of an improved lighting configuration. Accid. Anal. Prev. 45 (0), 432–437. Gould, M., Poulter, D.R., Helman, S., Wann, J.P., 2012b. Judgments of approach speed for motorcycles across different lighting levels and the effect of an improved tri-headlight configuration. Accid. Anal. Prev. 48 (0), 341–345. Hurt, H.H.J., Ouellet, J.V., Thom, D.R., 1981. Motorcycle Accident Cause Factors and Identification of Countermeasures, vol. 1. NHTSA, U.S. Department of Transportation, Washington DC. Itti, L., Baldi, P.F., 2005. A principled approach to detecting surprising events in video. In: (Cvpr), P.I.C.O.C.V.a.P.R. (Ed.), pp. 631–637. MAIDS, 2004. Motorcycle accident in-depth study maids: In-depth investigations of accidents involving powered two wheelers. Final report 1.2. Acem-association des constructeurs européens de motocycle (the motorcycle industry in europe). Brussels. MAIDS, 2009. Motorcycle accident in-depth study maids: In-depth investigations of accidents involving powered two wheelers. Final report 2.0. Acem-association des constructeurs européens de motocycle (the motorcycle industry in europe). Brussels. Maruyama, K., Tsutsumi, Y., Murata, Y., 2009. Study of face design, lighting system design for enhanced detection rate of motorcycles In: Proceedings of the 21st International Technical Conference On the Enhanced Safety of Vehicles, Stuttgart, Germany, pp. Paper No. 09-406. Muller, A., 1984. Daytime headlight operation and motorcyclist fatalities. Acc. Anal. Prev. 16 (1), 1–18. NHTSA, 2012. Traffic Safety Facts. NHTSA, Washington, DC. Nowell, L.T., 1997. Graphical Encoding for Information Visualization: Using Iconcolor, Shape, and Size to Convey Nominal and Quantitative Data. Virginia Polytechnic Institute and State University, Blacksburg, VA, USA. ONISR, 2014. Provisional road casualty results. Le bilan de la sécurité routière 2014. La documentation francaise, Paris, France. Olson, P.L., Halstead-Nussloch, R., Sivak, M., 1981. The effect of improvements in motorcycle/motorcyclist conspicuity on driver behavior. Hum. Factors 23 (2), 237–248. Pai, C.-W., 2011. Motorcycle right-of-way accidents—a literature review. Accid. Anal. Prev. 43 (3), 971–982. Papagiorgiou, C., 1997. Object and pattern detection in video sequences. In: Master’s Thesis. MIT. Petit, J., Bremond, R., 2010. A high dynamic range rendering pipeline for interactive applications. Vis. Comput. 26 (6), 533–542. Pinto, M., Cavallo, V., Saint-Pierre, G., 2014. Influence of front light configuration on the visual conspicuity of motorcycles. Accid. Anal. Prev. 62, 230–237. Rößger, L., Lenné, M.G., 2015. Increasing motorcycle conspicuity: design and assessment of interventions to enhance rider safety. Ashgate. Rößger, L., Hagen, K., Krzywinski, J., Schlag, B., 2010. Motorcycle’s conspicuity depending on signal pattern of the frontal light configuration. In: Proceedings of the International Conference on Safety and Mobility of Vulnerable Road Users: Pedestrians, Motorcyclists, and Bicyclists, Jerusalem, Israel. Rößger, L., Hagen, K., Krzywinski, J., Schlag, B., 2012. Recognisability of different configurations of front lights on motorcycles. Accid. Anal. Prev. 44 (1). Rößger, L., Lenné, M.G., Espié, S., 2015. Summarized assessment of the results on motorcycle conspicuity. In: Underwood, G. (Ed.), Increasing Motorcycle Conspicuity: Design and Assessment of Interventions to Enhance Rider Safety (human Factors of Simulation and Assessment). Ashgate.

126

M. Ranchet et al. / Accident Analysis and Prevention 94 (2016) 119–126

Rogé, J., Pébayle, T., 2009. Deterioration of the useful visual field with ageing during simulated driving in traffic and its possible consequences for road safety. Saf. Sci. 47 (9), 1271–1276. Shaheed, M.S.B., Gkritza, K., Zhang, W., Hans, Z., 2013. A mixed logit analysis of two-vehicle crash severities involving a motorcycle. Accid. Anal. Prev. 61 (0), 119–128. Smither, J.A.-A., Torrez, L.I., 2010. Motorcycle conspicuity: effects of age and daytime running lights. Hum. Factors 52 (3), 355–369. Tsutsumi, Y., Maruyama, K., Kazumitsu, K., 2008. Long lighting system for enhanced conspicuity of motorcycles. JSAE Trans. 39 (5), 119–124.

Van Elslande, P., Jaffard, M., 2010. Typical human errors in traffic accidents involving powered two-wheelers. In: Proceedings of the 27th International Congress of Applied Psychology, Melbourne, Australia, pp. 1363–1364.. Zador, P.L., 1985. Motorcycle headlight-use laws and fatal motorcycle crashes in the US, 1975–1983. Am. J. Public Health 75 (5), 543–546.