Assessment model for perceived visual complexity of automotive instrument cluster

Applied Ergonomics 46 (2015) 76e83

Contents lists available at ScienceDirect

Applied Ergonomics journal homepage: www.elsevier.com/locate/apergo

Assessment model for perceived visual complexity of automotive instrument cluster Sol Hee Yoon a, Jihyoun Lim b, Yong Gu Ji a, * a b

Dept. of Information and Industrial Engineering, Yonsei University, Seoul, Republic of Korea Dept. of Industrial Engineering, Hongik University, Seoul, Republic of Korea

a r t i c l e i n f o

a b s t r a c t

Article history: Received 9 July 2013 Accepted 7 July 2014 Available online 15 August 2014

This research proposes an assessment model for quantifying the perceived visual complexity (PVC) of an in-vehicle instrument cluster. An initial study was conducted to investigate the possibility of evaluating the PVC of an in-vehicle instrument cluster by estimating and analyzing the complexity of its individual components. However, this approach was only partially successful, because it did not take into account the combination of the different components with random levels of complexity to form one visual display. Therefore, a second study was conducted focusing on the effect of combining the different components. The results from the overall research enabled us to suggest a basis for quantifying the PVC of an in-vehicle instrument cluster based both on the PVCs of its components and on the integration effect. © 2014 Elsevier Ltd and The Ergonomics Society. All rights reserved.

Keywords: Perceived visual complexity Quantifiable measurement variables Assessment model

1. Introduction Advancements in automotive information technology have led to the introduction of new interactive devices for drivers, such as the in-vehicle information system (IVIS; Horrey et al., 2006). An IVIS has the ability to display more information than a conventional instrument cluster, including information on the current driving status, as well as information indirectly related to driving, like maps, entertainment information, and multimedia. To address this opportunity, full LCD displays are replacing conventional analog instrument clusters, which consist of a single display that shows information related to speed, fuel level, navigation, driving assistance, and more (Bellotti et al., 2004; Huang, 2007). The addition of information in visual displays is closely linked with visual perception and complexity. The principles of visual perception are commonly explained by a top-down interpretation of what humans see based on sensory information from the physical world (Gregory, 1970), and bottom-up information driven by human knowledge (Gibson, 1966). In this research we defined PVC as the degree of detail or intricacy that the user perceives in a visual stimulus by a combination of bottom-up and top-down

* Corresponding author. Dept. of Information and Industrial Engineering, Yonsei University, 262 Seongsanno, Seodaemun-gu, Seoul 120-749, Republic of Korea. Tel.: þ82 10 9058 9810. E-mail addresses: [email protected], [email protected] (Y.G. Ji). http://dx.doi.org/10.1016/j.apergo.2014.07.005 0003-6870/© 2014 Elsevier Ltd and The Ergonomics Society. All rights reserved.

processing (Gregory, 1970; Gibson, 1966; Lim and Liu, 2009). Therefore, increased complexity in visual perception can negatively affect how drivers process visual information (Noy et al., 2004; Van der Horst, 2004). Tsimhoni and Green (2001) investigated the influence of visual demands on driving performance in an experiment that measured visual demands and mental workloads when participants read an electronic map displayed in a vehicle both while driving and while parked. Their results highlighted the payoff that exists between visual demands and driving performance. High demand for visual attention to a specific visual display decreases drivers' visual attention to the road (Lee et al., 2007; Liang and Lee, 2010; Xia et al., 2010). This is likely why increasing the amount of information in invehicle displays could disturb driving performance and lead to safety concerns. In solving the potential problem of visual distraction arising from an excess of information provided by in-vehicle visual displays, the design and amount of information presented to drivers must be controlled and optimized. The visual complexity of in-vehicle visual displays has long been a topic of concern. Several studies have focused on the possibility of measuring complexity by objective metrics (Lavie et al., 2011; Huang, 2007; Cummings and Tsonis, 2006). Previous studies applied an image compression technique whereby an image's degree of visual complexity is calculated based on its compressed size (Tuch et al., 2009). Other researchers have proposed an approach whereby an image's visual complexity is calculated based on the length of edges within the image (Lim et al., 2010; Schmieder and

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

Weathersby, 1983). However, these methods for quantifying the complexity of visual stimulus are difficult to apply in the process of designing an IVIS. In-vehicle information has the specific characteristic that different types of information are combined in a single display. Hence, it is important to consider the complexity of the display's individual components. Accordingly, in this research we analyzed the PVC of an in-vehicle instrument cluster based on the PVC of each of its components. To discuss issues related to the combination of different components to form a single visual stimulus, we revisit Gestalt theory of perception, which states that humans tend to focus on a whole stimulus rather than on its individual parts (Graham, 2008). Gestalt theory introduced the idea that there are differences between the perception of the whole and the perception of its components (Rock and Palmer, 1990). The theory states that different perceptions of the whole are created when different components interact (Rock and Palmer, 1990). This impacts the concept of organization, and provides insights into differences between the perception of individual components and the whole (Rock and Palmer, 1990). In the context of an in-vehicle instrument cluster, individual components can affect a system in different ways when they are combined. In this study, we investigated individual components' influences on the overall PVC and also evaluate their interrelationships. In this way, this research provides insights into the possibility of evaluating the PVC of an in-vehicle instrument cluster by estimating the complexity of its components. However, as in previous studies, we determined that the combination of different components impacts the perception of complexity (Rock and Palmer, 1990). We conducted a study to propose an assessment model for quantifying the PVC by including the PVC of the estimated components, and then added a factor that considered the effect caused by the combination of the components; these two approaches were combined to form a complete assessment model to quantify the visual complexity perceived by drivers. 2. Research framework 2.1. Research hypotheses In the present study, we suggest an assessment model to quantify the PVC of an in-vehicle instrument cluster. We first propose the possibility of quantifying the PVC of the cluster by estimating the perceived complexity of its components. That is to say, our first hypothesis was that the estimated PVC of the components that form the in-vehicle instrument cluster can be applied to explain the PVC of the whole instrument cluster. However, the estimated PVC of each component did not significantly influence the PVC of the whole cluster. Thus, to evaluate the combination of components, we conducted a second study based on Gestalt theory, which states that the whole is different from the sum of its parts, and thus suggests why we were unsuccessful in explaining overall PVC in terms of the components. Our second research hypothesis was that an integration effect exists when different components are combined into the one stimulus of an in-vehicle instrument cluster. Therefore, we added this integration effect as a new factor in the assessment model. Herein the integration effect is defined as the effect of combining different components such as the speedometer, navigation display, and menu into a single display; this definition is based on previous studies of Gestalt theory (Rock and Palmer, 1990; Palmer and Rock, 1994; Stickel et al., 2010; Xiang et al., 2007). Thus, we suggest an assessment model to quantify the PVC of the whole in-vehicle instrument cluster by considering its components and the effect of their integration.

77

In this study, we conducted a survey and a controlled experiment to evaluate our hypothesis. We developed a statistical model to estimate perceived complexity based on subjective and objective measurements of each of the eight components. We then evaluated the relationship between the estimated PVCs of the components and the subjective perception of the visual complexity of the invehicle instrument cluster as a whole. To assess the model to quantify the PVC of the whole display in relation to the PVCs of the components, a multiple linear regression model was used in addition to a factor quantifying the integration effect. 2.2. Measurement variables Eleven objective, quantifiable measurement variables were selected from previous studies on visual complexity. These measurement variables were applied to objectively quantify factors of the visual stimulus. The objective measurement variables used in this study were stimulus size, icon size, component quantity, number of divisions, color variety, font variety, icon quantity, blank space percentage, text percentage, graphic percentage, and text-tographic ratio (Cummings and Tsonis, 2006; Forsythe, 2009; Kemps, 1999; Olivia et al., 2004; Harper et al., 2009; Michailidou, 2008; Stickel et al., 2010; Xing, 2007; Rosenholtz et al., 2007). Table 1 lists the definitions of these variables. Moreover, for each component we determined which subset of objective quantifiable measurement variables were relevant to the characteristics of each component, as summarized in Table 1. 2.3. Components Components refer to the types of information that are included in an in-vehicle instrument cluster and presented to the driver. Eight components were identified in the instrument cluster based on the types of information they presented to drivers. The components were classified as providing either conventional information or additional information. Conventional information includes the standard information traditionally provided in a vehicle instrument cluster, such as speedometer, tachometer, other gauges, and the gear position indicator. The component labeled other gauges refers to other related components that present information in gauge format in the instrument cluster, such as fuel level and engine temperature readings. Finally, the gear position indicator refers to the display showing the gear rotation information as the driver selects or changes gears. Additional information refers to information that has been integrated into vehicle instrument clusters more recently, with technological advancements. In this research, we categorized the additional information into four groups: navigation, assistive information, entertainment, and menu. Navigation refers to information related to vehicle location. Assistive information refers to information to help drivers optimize their driving performance. Entertainment refers to information related to enjoyment while driving that is unrelated to the main task of driving. Finally, menu refers to a graphic user interface that allows drivers to select which information is displayed. 3. Study 1: in-vehicle instrument cluster and PVC of the components 3.1. Method The aim of this first study was to verify the influence of the estimated PVC of each component upon that of the overall invehicle instrument cluster. The study proposes a model to estimate the visual complexity of a component based on subjective responses and objective performance measurements. Subjective

78

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

responses were based on questionnaires regarding the PVC of various design components; each questionnaire included eight sections, one for each component. The objective performance measure was based on a laboratory experiment in which subjects' performance and ocular motor behavior were objectively measured during a visual search task. Multiple linear regression models were used to obtain the model used to estimate the PVC of components. Statistical analysis was then performed to investigate the relationship between the PVC of the components and that of the whole. In this way, the study enabled us to propose an estimation model to quantify the PVC of individual components, and to investigate the possibility of estimating the PVC of the whole in-vehicle instrument cluster. 3.1.1. Stimuli A total of 101 visual stimuli of individual components were used (15 for the speedometer, 10 for the tachometer, 14 for the gauge, 10 for the gear position indicator, 13 for the navigation display, 13 for assistive information, 13 for entertainment, and 13 for the menu). Each stimulus was saved as a video document (.SWF) and a flash document (.FLA) for the questionnaire, and the size of the image presented in the questionnaire was kept constant regardless of the size of the screen used. For the visual search experiments, a small star-shape figure (4  4 pixels) search target was embedded in each of the 101 visual stimuli. The locations of the target were randomized to make them unpredictable. A yellow star was used to avoid confusion with the background, and to facilitate target identification, the target was not placed in any regions of the visual stimulus that were of similar color. Moreover, 25 images were selected to evaluate the PVC of the whole in-vehicle instrument cluster. They were also saved as video and flash documents, and their evaluation was included in the questionnaire. 3.1.2. Equipment/apparatus A remote eye tracking system (SMI RED-m, SensoMotoric Instruments; SMI) was used to obtain data on the response time and number of fixations; it was attached to a 24-inch monitor where the visual stimuli were presented. An extra laptop was used to

manipulate and design the experiment. SMI Experiment Center software was employed to design the search task experiment, and SMI BeGaze software was used to analyze the experimental data. 3.1.3. Participants Sixty-two males and forty females between the ages of 20 and 60 years (mean ¼ 36.44 years, SD ¼ 12.53) with driving experience participated in the questionnaire for Study 1. Demographic data on the participants were gathered regarding their average time spent driving per day and the total distance they had driven (driving experience). Among the participants, 57.84% drove an average of less than one hour per day, 30.39% drove between one and two hours, 6.86% drove between two and three hours, and 1.96% drove between three and four hours; 30.39% of the participants had driven less than 2000 km, 19.61% had driven between 2000 km and 20,000 km, and 48.04% had driven more than 20,000 km. Forty-five participants were recruited for the visual search experiment (23 males and 22 females). Data from five participants (2 males and 3 females) were excluded due to eye movement tracking ratios below 95% (participant ocular limitation). The remaining 40 participants (21 males and 19 females) each had a driver's license and driving experience. Their ages ranged between 22 and 56 years (mean ¼ 36.98 years, SD ¼ 9.89). Demographic data on the participants were gathered regarding their average time spent driving per day and driving experience. 3.1.4. Questionnaire design The design of the questionnaire was based on previous research regarding an information questionnaire for an air traffic control visual display (Xing, 2008) and an information complexity questionnaire to validate visual interfaces (Ling et al., 2011). Our redesigned questionnaire consisted of three questions related to the perception of each of the visual stimuli; participants scored each image on a 7-point Likert scale (1: strongly disagree, 7: strongly agree). All participants were Korean; therefore, the entire questionnaire was translated into Korean. The questionnaire was divided into eight sections to cover each of the eight components of the in-vehicle instrument cluster; the components were presented

Table 1 Quantifiable measurement variables for components. Quantifiable measurement

Definition

Stimulus size

Entire region (area) covered by a visual stimulus, measured in mm2 Icon size Average size of a symbol that is used to express a well-known meaning. Component quantity Number of components in the visual stimulus. Number of divisions Number of areas on the visual stimulus that can be visually distributed by lines or different background colors. Color variety Number of different colors perceived. (Gradation of color was counted as two colors, for the initial color and final color). Font variety Number of different fonts used in the visual stimulus. Icon quantity Number of different symbols used in the visual stimulus. Blank space Percentage of area in the visual stimulus that does percentage not provide any kind of information. Text percentage Percentage of area in the visual stimulus that provides information in text form. Graphic percentage Percentage of area in the visual stimulus that provides information in graphic form. Text-to-graphic ratio Ratio between the total area of text information to the total area of graphic information in the visual stimulus. Y ¼ quantifiable; N ¼ not quantifiable.

Conventional information components

Additional information components

Speedometer Tachometer Other Gear Navigation Assistive Entertainment Menu gauges position Y

Y

Y

Y

Y

Y

Y

Y

N

N

Y

Y

Y

Y

Y

Y

N N

N N

N N

N N

Y Y

Y Y

Y Y

N Y

Y

Y

Y

Y

Y

Y

Y

Y

Y N

Y N

Y Y

N N

Y Y

Y Y

Y Y

Y Y

Y

Y

Y

Y

Y

Y

Y

N

Y

Y

Y

N

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

Y

N

Y

Y

Y

Y

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

in random order to prevent ordering effects. The same questionnaire was also used for the whole in-vehicle instrument cluster stimuli. The questionnaire was administered online to provide the same visual stimulus size to all participants regardless of the monitor used. In the questionnaire, the demographic questions were asked first, followed by those related to the eight in-vehicle components. Participants had to answer each question about PVC for all 101 visual stimuli. The questionnaire on the 25 whole in-vehicle instrument cluster stimuli was administered one month after the questionnaire on the components. 3.1.5. Visual search experiment design To observe the relationship between users' behavior and the PVC, an experiment was conducted in which participants performed a visual search task while their searching behaviors were encoded and analyzed by an eye tracking system. The experiment was designed based on research on visual clutter by Rosenholtz et al. (2007). The data collected were the response time and number of fixations, and the visual stimuli used were the same 101 components presented in the questionnaire. Demographic data was gathered before the experiment, and a brief introduction to the experiment was given, during which the task was explained. Participants were asked to find the target hidden within the visual stimulus. Once the target was found, the participants had to click on it in order to continue to the next visual stimulus. The experiment consisted of three sets, with 50 or 51 visual stimuli in each set. The order of stimuli within the sets was randomized to prevent any ordering effect. Between each set, each participant was given a 10-min break. Response time was measured as the time it took for participants to click on the hidden target. The number of fixations was counted by analyzing a participant's scan path while searching for each hidden target. Fig. 1 illustrates how the data were obtained using the eye tracking equipment. 3.1.6. Statistical analysis Three types of statistical analysis were conducted: 1) analysis of the relationship between quantifiable measurements for each of the components to determine which measurement variables were significantly correlated and thus were suitable for inclusion in the estimation model; 2) regression model analysis to suggest an estimation model for the PVC of each component; and 3) analysis of

79

the relationship between the PVC of each component and the PVC of whole in-vehicle instrument clusters comprising those components. In the first stage, we discovered the objective quantifiable measurements that were statistically significantly correlated with the subjective questionnaire data on the PVC of the components and the objective visual search task results (i.e., response time and number of fixations). We performed a Pearson correlation analysis, which provided a confidence interval of 95%. As a result, objective measurement variables were selected for the next stage. Secondly, a model to estimate PVC was presented for each of the components. This estimation model was developed using a stepwise forward regression model. Finally, in stage 3, we conducted a Pearson correlation analysis with a confidence interval of 5% to verify the ability of the estimated PVC of each component to predict the subjective PVC of the whole in-vehicle instrument cluster. PASW Statistics 18 software was employed for all statistical analyses. 3.2. Results 3.2.1. Estimation model for conventional information components Among the speedometers' quantifiable measurement variables, stimulus size, color variety, font variety, blank space percentage, text percentage, graphic percentage, and text-to-graphic ratio measurements were significantly correlated with the speedometers' PVC (Table 2). Stimulus size, blank space percentage, and text percentage were significantly correlated with response time; stimulus size, color variety, blank space percentage, text percentage, and graphic percentage were significantly correlated with the number of fixations (Table 2). The estimation model for the PVC of the speedometer (r ¼ 0.624; r2 ¼ 0.390) was developed using the blank space percentage (p < 0.001) and text percentage (p < 0.001) measurements (Table 4). Most of the tachometers' objective measurement variables except the text percentage were significantly correlated with their PVC; however, the user experiment indicated that the text percentage had the highest correlation with tachometer visual search performance and visual search behavior. More specifically, stimulus size, font variety, blank space percentage, and text percentage were significantly correlated with response time in the user experiment, while stimulus size, font variety, blank space percentage, and text percentage were significantly correlated with the number of fixations (Table 2). The estimation model for the PVC of the tachometer

Fig. 1. Example of data obtained from the eye-tracking experiment.

80

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

Table 2 Quantifiable measurement variables for components of conventional type information. Quantifiable measurement

Conventional information components

PVC

RT

Fix. #

PVC

RT

Fix. #

PVC

RT

Fix. #

PVC

RT

Fix. #

Stimulus size Icon size Color variety Font variety Icon quantity Blank space percentage Text percentage Graphic percentage Text-to-graphic ratio

0.226c e 0.282c 0.224c e 0.599c 0.594c 0.258c 0.205c

0.251c e 0.027 0.011 e 0.104b 0.163c 0.057 0.057

0.257c e 0.073a 0.066 e 0.103b 0.193c 0.103b 0.065

0.123c e 0.340c 0.186c e 0.712c 0.026 0.638c 0.119c

0.301c e 0.007 0.123b e 0.097a 0.164c 0.014 0.056

0.425c e 0.022 0.156b e 0.095a 0.213c 0.013 0.080

0.205c 0.371c 0.257c 0.426c 0.404c 0.389c 0.619c 0.238c 0.500c

0.218c 0.191c 0.007 0.062 0.159c 0.008 0.107b 0.019 0.167c

0.348c 0.258c 0.068 0.146c 0.225c 0.021 0.147c 0.061 0.265c

0.731c 0.086a 0.016 e e 0.424c e 0.389c e

0.095a 0.048 0.032 e e 0.014 e 0.030 e

0.092a 0.027 0.041 e e 0.008 e 0.014 e

Speedometer

Tachometer

Other gauges

Gear position indicator

PVC ¼ perceived visual complexity; RT ¼ response time; Fix. # ¼ number of fixations. a Significant at p < 0.05. b Significant at p < 0.01. c Significant at p < 0.001.

(r ¼ 0.714; r2 ¼ 0.509) was developed using the stimulus size (p < 0.001) and blank space percentage (p < 0.05) measurements (Table 4). For the other gauges component, all of the quantifiable measurements selected from the expert evaluation were significantly correlated with the subjective results for PVC. However, our user experiment showed some differences. Stimulus size, icon size, icon quantity, text percentage, and text-to-graphic ratio were significantly correlated with response time, and stimulus size, icon size, font variety, icon quantity, text percentage, and text-to-graphic ratio were significantly correlated with the number of fixations (Table 2). The estimation model for the PVC of the other gauges component (r ¼ 0.633; r2 ¼ 0.401) was developed using the stimulus size (p < 0.001), icon quantity (p < 0.001), text percentage (p < 0.001), and text-to-graphic ratio (p < 0.01) measurements (Table 4). Finally, among the quantifiable measurements for gear position indicators, stimulus size, icon size, blank space percentage, and graphic percentage were significantly correlated with their PVC. In the user experiment, only stimulus size was significantly correlated to the response time and number of fixations (Table 2). The estimation model for the PVC of the gear position indicator (r ¼ 0.731; r2 ¼ 0.534) was developed using the stimulus size (p < 0.001) measure (Table 4).

3.2.2. Estimation model for additional information components All of the navigation displays' quantifiable measurement variables were significantly correlated with their PVC, except component quantity and number of divisions. However, all quantifiable measurement variables except icon size and font variety were significantly correlated with response time and the number of fixations (Table 3). The estimation model for the PVC of the navigation display (r ¼ 0.594; r2 ¼ 0.352) was developed using the text information percentage (p < 0.001) and graphic percentage (p < 0.001) measurements (Table 4). All of the quantifiable measurement variables of assistive information components were significantly correlated to their PVC, except stimulus size and component quantity. Contrastingly, the icon size, component quantity, number of divisions, color variety, text percentage, and text-to-graphic ratio were significantly correlated with response time, and icon size, component quantity, number of divisions, color variety, font variety, text percentage, and text-tographic ratio were significantly correlated with the number of fixations (Table 3). The estimation model for the PVC of the assistive information component (r ¼ 0.573; r2 ¼ 0.328) was developed using the blank space percentage (p < 0.001) measure (Table 4). All of the quantifiable measurement variables for the entertainment components were significantly correlated with their PVC,

Table 3 Quantifiable measurement variables for components of additional type information. Quantifiable measurement

Additional information components Navigation display PVC

Stimulus size Icon size Component quantity Number of divisions Color variety Font variety Icon quantity Blank space percentage Text percentage Graphic percentage Text-to-graphic ratio

Assistive information

RT

Fix. #

PVC c

RT

Entertainment Fix. #

PVC

RT

PVC c

RT

Fix. #

b

0.080 0.322c 0.018 0.005 0.322c 0.188c 0.185c 0.081b

b

0.105 0.026 0.085a 0.096a 0.100b 0.024 0.340c 0.317c

0.213 0.034 0.143c 0.132b 0.112b 0.043 0.375c 0.304c

0.003 0.087b 0.007 0.063a 0.116c 0.102b 0.105b 0.573c

0.045 0.043 0.196c 0.090a 0.200c 0.146c 0.068 0.146c

0.033 0.055 0.212c 0.045 0.189c 0.180c 0.060 0.145c

0.134 0.039 0.053 0.178c 0.098b 0.256c 0.658c 0.204c

0.021 0.029 0.033 0.143c 0.132b 0.077a 0.126b 0.036

0.139 0.016 0.150c 0.261c 0.174c 0.040 0.445c 0.044

0.045 0.290c e 0.216c 0.246c 0.045 0.041 e

0.256 0.146c e 0.256c 0.127b 0.122b 0.063 e

0.316c 0.161c e 0.329c 0.207c 0.148c 0.099b e

0.583c 0.151c 0.330c

0.195c 0.321c 0.185c

0.188c 0.308c 0.173c

0.141c 0.548c 0.234c

0.085a 0.090a 0.009

0.153c 0.063 0.067

0.097b 0.187c 0.067a

0.055 0.089a 0.033

0.194c 0.114b 0.069

0.602c 0.172c 0.291c

0.095a 0.108b 0.234c

0.158c 0.093a 0.317c

PVC ¼ perceived visual complexity; RT ¼ response time; Fix. # ¼ number of fixations. a Significant at p < 0.05. b Significant at p < 0.01. c Significant at p < 0.001.

c

Menu Fix. #

c

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

except icon size and quantity of items. The number of divisions, color variety, font variety, icon quantity, and graphic percentage were significantly correlated with response time, and stimulus size, component quantity, number of division, color variety, icon quantity, text percentage, and graphic percentage were significantly correlated with the number of fixations (Table 3). The estimation model for the PVC of the entertainment component (r ¼ 0.658; r2 ¼ 0.433) was developed using the icon quantity (p < 0.001) measure (Table 4). Finally, all of the quantifiable measurement variables for the menus were significantly correlated with their PVC, except font variety and icon quantity. All the quantifiable measurement variables except icon quantity were significantly correlated with response time, and all were significantly correlated with the number of fixations (Table 3). The estimation model for the PVC of the menu (r ¼ 0.607; r2 ¼ 0.368) was developed using the text percentage (p < 0.001) and text-to-graphic ratio (p < 0.001) measurements (Table 4). 3.2.3. Relationship between components and the overall in-vehicle instrument cluster PVC Once we obtained a quantifiable model for the PVC of each component (Table 4), we applied the model to estimate the PVC of each of the components' visual stimuli. We then conducted a correlation analysis between the PVC of each component and the PVC of the in-vehicle visual display (the whole visual stimulus). Our results show that only the speedometer (p < 0.01), other gauges (p < 0.05), and gear position indicator (p < 0.05) were significantly correlated with the PVC of the overall in-vehicle instrument cluster (Table 5). 4. Study 2: integration effect and PVC Results from Study 1 showed that not all components' PVCs were significantly correlated with the PVC of the in-vehicle instrument cluster: the speedometer, other gauges, and gear position indicator were the only components that showed significant correlations. These results agree with the Gestalt law of perception. Accordingly, it is impossible to suggest a model that quantifies the PVC of an in-vehicle instrument cluster by knowing only the PVC of each component. Therefore, we developed a quantitative model for the PVC of an instrument cluster, applying the findings from Study 1 and conducting a second study to develop a factor to account for Gestalt effects. 4.1. Method In Study 2, we investigated a factor capable of explaining the effects of combining different components, thus enabling us to

81

Table 5 Correlation analysis between the perceived visual complexity of components and the in-vehicle instrument cluster. Category

Component

Correlation of component PVC with whole-cluster PVC R

Significance level

Conventional type of information

Speedometer Tachometer Other gauges Gear position indicator

0.055b 0.011 0.071a 0.053a

0.005 0.658 0.015 0.012

Additional information

Navigation display Assistive information Entertainment Menu

0.041 0.028 0.054 0.019

0.227 0.496 0.097 0.533

a b

Significant at p < 0.05. Significant at p < 0.01.

suggest an assessment model for quantifying the PVC of an invehicle instrument cluster based on the PVCs of its components. First, we selected the quantifiable measurements that were most suitable for determining the integration effect, choosing stimulus size, component quantity, number of divisions, color variety, font variety, icon quantity, blank space percentage, text percentage, graphic percentage, clutter, and text-to-graphic ratio. Component quantity refers to the number of components in the instrument cluster (Olivia et al., 2004; Xing, 2007). That is, if the visual stimulus of an instrument cluster is composed of a speedometer, tachometer, and gear position indicator, its component quantity is three. Clutter refers to a state in which there are an excessive number of items (based on representation, method, or organization) and a resulting degradation in performance (Rosenholtz et al., 2007; Forsythe, 2009). Thus, we measured clutter by determining the edge probability of the visual stimulus. In Study 2, we applied the same methodology that was used for Study 1: we analyzed questionnaire results for 25 visual stimuli of whole in-vehicle instrument clusters, and the results of user experiments on visual search performance and behavior. Then, correlation analysis was conducted to select significant variables which were later used for estimating the PVC of integration effect. An additional correlation analysis was then conducted between the estimated integration effect and the subjective PVC of the whole in-vehicle instrument cluster. The integration effect was considered to be an additional component, accounting for the perception of the combination of different components. Thus, once the results were obtained, we suggested a final model to enable the quantification of the PVC for an in-vehicle instrument cluster based on an estimation of the PVC of the components and the integration effect.

Table 4 Estimation model for perceived visual complexity of components. Component

Estimation model

r

r2

Speedometer Tachometer Other gauges

0.465 þ 0.032 * (blank space percentage)c  0.033 * (text percentage)c 10.583  0.087 * (stimulus size)c  0.0003 * (blank space percentage)a 2.471  0.142 * (stimulus size)c þ 0.133 * (icon quantity)c  0.004 * (text percentage)c þ 0.482 * (text-to-graphic ratio)b 2.365 þ 0.080 * (stimulus size)c 2.126 þ 0.090 * (text percentage)c þ 0.002 * (graphic percentage)c 5.357  0.036 * (blank space percentage)c 2.306 þ 0.231 * (icon quantity)c 2.426 þ 0.041 * (text percentage)c þ 0.044 * (text-to-graphic ratio)b

0.624 0.714 0.633

0.390 0.509 0.401

0.731 0.594 0.573 0.658 0.607

0.534 0.352 0.328 0.433 0.368

Gear position indicator Navigation display Assistive information Entertainment Menu a b c

Significant at p < 0.05. Significant at p < 0.01. Significant at p < 0.001.

82

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

following estimation model was used for the integration effect (r ¼ 0.453; r2 ¼ 0.205):

4.2. Statistical model analysis The statistical analysis was divided into four stages: 1) statistical analysis of the relationship between the quantifiable measurement variables and the integration effect for an in-vehicle instrument cluster; 2) stepwise forward regression model analysis to suggest an estimation model for the integration effect; 3) statistical analysis of the influence of the integration effect on the PVC of an in-vehicle instrument cluster; and 4) stepwise forward regression model analysis to suggest an estimation model for the PVC of an in-vehicle instrument cluster based on both the estimated PVC of the components and the integration effect. In the first stage, we analyzed the effect of computational objective measurement variables on the integration effect, visual search performance, and visual search behavior. A Pearson correlation analysis was performed with a confidence interval of 5%, and the measurement variables showing a significant correlation for all the three cases were selected as relevant quantifiable measurement variables for the estimation model. In the second stage, a model was developed to estimate the integration effect; this estimation model was based on a stepwise forward regression model that utilized the quantifiable measurement variables selected from stage 1. A stepwise forward regression model starts with no variables, and then uses predefined selection criteria to sequentially add variables that improve the outcome; accordingly, only the quantifiable measurement variables that showed a significant correlation (p < 0.05) were considered in the model. In the following stage, Pearson correlation analysis with a confidence interval of 5% was performed to test the correlation between the integration effect and the PVC of the whole in-vehicle instrument cluster. Finally, in the last stage, the significantly correlated components obtained from Study 1 and the estimated integration effect were applied in a regression model analysis with respect to the subjective PVC for the whole in-vehicle instrument cluster. Thus, we proposed a model to assess the PVC of the instrument cluster. PASW Statistics 18 software was employed for all statistical analyses. 4.3. Results The results from the correlation analysis showed that stimulus size (p < 0.01), icon size (p < 0.05), component quantity (p < 0.001), blank space percentage (p < 0.05), graphic percentage (p < 0.01), and clutter (p < 0.001) were significantly correlated with the results of the questionnaire and the user experiment. Therefore, we selected the following quantifiable measurement variables to propose an estimation model for the integration effect. Table 6 shows results from the forward regression model. For the estimation model, the component quantity, clutter, blank space percentage, graphic percentage, and stimulus size were selected as significant quantifiable measurements (p < 0.001). Thus, the

Table 6 Results of regression model analysis for the integration effect.

Intercept Component quantity Clutter Blank space percentage Graphic percentage Stimulus size a

Significant at p < 0.001.

Beta

Standard error

t-Value

Significance level

5.388 0.133 1.542E7 0.040 0.040 0.001

0.457 0.008 0.000 0.005 0.005 0.000

11.788 17.039 9.190 8.671 7.772 4.851

0.000a 0.000a 0.000a 0.000a 0.000a 0.000a

Integration effect ¼ 5:388 þ 0:133*ðComponent quantityÞ  1:542E  7*ðClutterÞ  0:040*ðBlank space percentageÞ  0:040*ðGraphic percentageÞ þ 0:001*ðStimulus sizeÞ Once we obtained the estimation model for the integration effect, we measured the estimated integration effect for each of the visual stimuli on the in-vehicle instrument cluster. The Pearson correlation coefficient for the correlation between the integration effect and the overall PVC of the display was r ¼ 0.575 (p < 0.001). Therefore, the integration effect was significantly correlated with the PVC of the whole in-vehicle instrument cluster, thus solving the problem arose in the results of Study 1. A stepwise forward regression model was conducted using the PVCs of the speedometer, other gauges, and gear position indicator, as well as the integration effect. Table 7 lists the results of this statistical analysis and presents the estimation model for the PVC of the overall in-vehicle instrument cluster. Thus, we developed an estimation model for the PVC of an invehicle instrument cluster. The components used in this model were the integration effect (p < 0.001) and the estimated PVC of the speedometer (p < 0.001). The model is given as follows, and represents a method for quantitatively assessing the PVC of the invehicle instrument cluster by considering its components (r ¼ 0.580; r2 ¼ 0.337):

PVC of the in-vehicle instrument cluster ¼ 0:289 þ 1:162*ðIntegration effectÞ  0:083*ðEstimated PVC of speedometerÞ 5. Discussion and conclusions In this study, we examined the influence of the components of an in-vehicle instrument cluster on the PVC of the whole cluster. An estimation model was proposed to quantify the PVC of each of the components as a function of their quantifiable measurements, such as size, quantities, and ratios of design elements (font, icon, and text). The results of Study 1 showed that not all components had any measures studied that were significantly correlated with the PVC of the entire instrument cluster. The usability of each measurement as an estimator of PVC was tested by correlation analysis between each measurement and subjective and objective measures of PVC. Stimulus size, icon size, component quantity, blank space percentage, graphic percentage, and clutter were identified as factors that correlate with PVC. This reduced set of attributes was utilized to develop an estimation model for PVC of the whole in-

Table 7 Results of regression model analysis for the perceived visual complexity of the invehicle instrument cluster.

Intercept Integration effect Estimated PVC of the speedometer a b

Beta

Standard error

t-Value

Significance level

0.289 1.162 0.083

0.108 0.032 0.018

2.666 36.150 4.648

0.008a 0.000b 0.000b

Significant at p < 0.01. Significant at p < 0.001.

S.H. Yoon et al. / Applied Ergonomics 46 (2015) 76e83

vehicle instrument cluster. The results suggested that drivers respond to specific properties of the display design; therefore, these properties should be prioritized during the design process. Regarding conventional information displays, only the estimated PVCs of the speedometer, the other gauges, and the gear position indicator showed significant correlations with the subjective PVC of the complete in-vehicle instrument cluster; the estimated PVC of the tachometer was not significantly correlated with the overall display complexity. This suggests that the tachometer is less important to drivers than the speedometer, other gauges, and gear position indicator. The PVCs of the additional information displays were not statistically significantly correlated with the PVC of the whole instrument cluster; this might be due to participants' unfamiliarity with the additional information given in the in-vehicle instrument cluster. Accordingly, additional factors, such as user knowledge and preference, may play a role in the perception of visual complexity. Thus, further research is needed to investigate factors that influence perceptions of complexity. The results of this study agreed with the Gestalt law. However, the research objective was to propose a model to quantify the PVC of whole in-vehicle instrument clusters. Thus, we followed up with a study to investigate the possibility of identifying a new quantifiable component capable of representing the effect of integrating the eight components of the in-vehicle instrument cluster, since these components had only been considered separately in the initial study. In the case of an in-vehicle instrument cluster, there may be a perceptual grouping based on the Gestalt law. The cluster consists of several components, and the PVC of its whole visual stimulus depends on how the components are organized. Hence, grouping and organizing the components depending on their perceptual factors, such as size, common region, location, and color (Palmer, 1992), might provide different results. The model developed to estimate the integration effect was evaluated by analyzing the relationship between the integration effect, as calculated by using the estimation model, and the PVC of the overall in-vehicle visual display. As expected, there was a significant relationship between them. The results of this evaluation emphasized the strong relationship between the integration effect and the PVC of the overall in-vehicle display. The Pearson correlation coefficient for the integration effect was 0.575, which was the largest of all the components. This demonstrates the importance of harmoniously and meticulously integrating each component when attempting to decrease the perception of complexity in a visual display. The results of this study suggest that the integration effect can be used as a component to complement the estimation model for the PVC of an in-vehicle instrument cluster. This research aimed to develop an assessment model to quantify drivers' perceptions of the visual complexity of in-vehicle instrument clusters. The results revealed the differences in human perception when visual stimuli are seen together or separately. We focused herein on in-vehicle instrument clusters, but we expect that our findings apply more broadly to other visual displays with similar characteristics, namely, displays composed of different and independent components. This research was conducted using static visual stimuli; further studies are required to examine the perceived complexity of dynamic visual displays. Acknowledgment This work was supported by Mid-career Researcher Program through NRF grant funded by the MSIP (Ministry of Science, ICT and Future Planning) (Grant-# NRF-2013R1A2A2A03014150).

83

References Bellotti, F., De Gloria, A., Poggi, A., Andreone, L., Damiani, S., Knoll, P., 2004. Designing configurable automotive dashboards on liquid crystal displays. Cogn. Technol. Work 6, 247e265. Cummings, M.L., Tsonis, C.G., 2006. Partitioning complexity in air traffic management tasks. Int. J. Aviat. Psychol. 16 (3), 277e295. Forsythe, A., 2009. Visual complexity: is that all there is?. Engineering psychology and cognitive ergonomics. In: HCII 2009. LNAI, vol. 5639, pp. 158e166. Gibson, J.J., 1966. The problem of temporal order in stimulation and perception. J. Psychol. Interdiscip. Appl. 62 (2), 141e149. Graham, L., 2008. Gestalt theory in interactive media design. J. Hum. Soc. Sci. 2 (1), 1e12. Gregory, R.L., 1970. On how little information controls so much behaviour. Ergonomics 13 (1), 25e35. Harper, S., Michailidou, E., Stevens, R., 2009. Toward a definition of visual complexity as an implicit measure of cognitive load. ACM Trans. Appl. Percept. 6 (2), 1e18. Horrey, W.J., Wickens, C.D., Consalus, K.P., 2006. Modeling driver's visual attention allocation while interacting with in-vehicle technologies. J. Exp. Psychol. Appl. 12 (2), 67e78. Huang, K.C., 2007. Effects of computer icons and figure/background area ratios and color combinations on visual search performance on an LCD monitor. Displays 29, 237e242. Kemps, E., 1999. Effects of complexity on visuo-spatial working memory. Eur. J. Cogn. Psychol. 11 (3), 335e356. Lavie, T., Oron-Gilad, T., Meyer, J., 2011. Aesthetics and usability of in-vehicle navigation displays. Int. J. Hum.eComp. Stud. 69 (1e2), 80e99. Lee, Y.C., Lee, J.D., Boyle, N.L., 2007. Visual attention in driving: the effects of cognitive load and visual disruption. Hum. Factors 49 (4), 721e733. Liang, X., Lee, J.D., 2010. Combining cognitive and visual distraction: less than the sum of its parts. Assess. Saf. Driv. Simul. 42 (3), 881e890. Lim, J.H., Liu, Y., 2009. Modeling the influence of cyclic top-down and bottom-up processes for reinforcement learning in eye movements. IEEE Trans. Syst. Man Cybernet. Part A: Syst. Hum. 39 (4), 706e714. Lim, J.H., Tsimhoni, O., Liu, Y., 2010. Investigation of driver performance with night vision and pedestrian detection systems e part I: empirical study on visual clutter ad glace behavior. IEEE Trans. Intell. Transp. Syst. 11 (3), 670e677. Ling, C., Lopez, M., Shehab, R., 2011. Complexity questionnaires of visual displays: a validation study of two information complexity questionnaires of visual displays. Hum. Factors Ergon. Manuf. Serv. Ind. 23 (5), 1e21. Michailidou, E., 2008. Determining Users' Perception of Web Page Visual Complexity and Aesthetic Characteristics (HCW Technical Report). University of Manchester, School of Computer Science, Manchester, United Kingdom. Noy, Y.I., Lemoine, T.L., Klachan, C., Burns, P.C., 2004. Task interruptability and duration as measure of visual distraction. Appl. Ergon. 35, 207e213. Olivia, A., Mack, M.L., Shrestha, M., Peeper, A., 2004. Identifying the perceptual dimensions of visual complexity of scenes. In: Paper Presented at the 26th Annual Meeting of the Cognitive Science Society, Chicago, United States. Palmer, S., 1992. Common region: a new principle of perceptual grouping. Cogn. Psychol. 24, 436e447. Palmer, S., Rock, I., 1994. Rethinking perceptual organization: the role of uniform connectedness. Psychonomic Bull. Rev. 1 (1), 29e55. Rock, I., Palmer, S., 1990. The legacy of gestalt psychology. Sci. Am. 263 (6), 84e90. Rosenholtz, R., Li, Y., Nakano, L., 2007. Measuring visual clutter. J. Vision 7 (2), 1e22. Schmieder, D.E., Weathersby, Marshall R., 1983. Detection performance in clutter with variable resolution. IEEE Trans. Aerosp. Electron. Syst. 19 (4), 622e630. Stickel, C., Ebner, M., Holzinger, A., 2010. The XAOS metric e understanding visual complexity as measure of usability. In: Leitner, G., Hitz, M., Holzinger, A. (Eds.), HCI in Work and Learning, Life and Leisure, vol. 6389. Springer, Berlin, Heidelberg, pp. 278e290. Tsimhoni, O., Green, P., 2001. Visual demand of driving and the execution of displayintensive in-vehicle tasks. In: Proceeding of the Human Factors and Ergonomics Society 45th Annual Meetings, pp. 1586e1590. Tuch, A.N., Bargas-Avila, J.A., Opwis, K., Wilhelm, F.H., 2009. Visual complexity of websites: effects on user's experience, physiology, performance, and memory. Int. J. Hum.eComp. Stud. 67, 703e715. Van der Horst, R., 2004. Occlusion as a measure for visual workload: an overview of TNO occlusion research in car driving. Appl. Ergon. 35, 189e196. Xia, R.T., Fan, J., Wang, Z., 2010. The spatial allocation of drivers' visual attention during driving in traffic environment. In: Paper Presented in 2010 International Conference of Intelligent Computation Technology and Automation, Chansha, Hunan, China. Xiang, P., Yang, X., Yuanchun, S., 2007. Web page segmentation based on Gestalt theory. In: IEEE International Conference on Multimedia and Expo, July, Beijing, pp. 2253e2256. Xing, J., 2007. Information complexity in air traffic control displays. In: HumaneComputer Interaction, Part IV, HCII 2007. LNCS, vol. 4553, pp. 797e806. Xing, J., 2008. Designing Questionnaires for Controlling and Managing Information Complexity in Visual Displays (Tech. Rep. to Federal Aviation Administration No. DOT/FAA/AM-08/18). Civil Aerospace Medical Institute, Office of Aerospace Medicine, Oklahoma, United States.