U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-04-137
Date: December 2005
Enhanced Night Visibility Series, Volume VI: Phase II—Study 4: Visual Performance During Nighttime Driving in Fog
PDF Version (1.27 MB)
PDF files can be viewed with the Acrobat® Reader®
Thirty individuals participated in this study. They had not participated in any other studies for the ENV project (i.e., clear, rain, snow, glare, or pavement marking studies). Participants were divided into three age categories: 10 drivers were between the ages of 18 and 25 (younger category), 10 were between the ages of 40 and 50 (middle-aged category), and 10 were over 65 (older category). There were five males and five females in each age category. Candidates for the study had to meet the conditions of a screening questionnaire (appendix A). Participants also had to sign an informed consent form (appendix B), present a valid driver’s license, pass the visual acuity test (appendix C) with a score of 20/40 or better (as required by Virginia State law), and have no health conditions that made operating the research vehicles a risk.
Participants were told they could withdraw freely from the research program at any time without penalty and that no one would try to make them participate if they did not want to continue. If they chose at any time not to participate further, they were instructed that they would be paid for the amount of actual participation time. Participants received 20 dollars per hour for their participation. All data gathered as part of this experiment were treated with complete anonymity.
A mixed-factor design was used to collect data during the onroad portion of the study (i.e., detection and recognition tasks). There were two independent variables:
The between-subjects variable of the experiment was age. The within-subject variable was VES configuration. Unlike the three previous studies (i.e., clear, rain, and snow), this study used just one object type, a pedestrian in white clothing. The reason for limiting the number of objects was the variability of the fog-making conditions. (See Smart Road section for more details.) Table 1 shows a representation of the experimental design including the object presented to participants. A detailed explanation of each of the independent variables follows the table.
The one object selected for this study was a pedestrian (table 2) mainly because of the high crash-fatality rates pedestrians represent.(1,2) This study used real pedestrians. Although pedestrian mockups have been used in previous research of this type, using mockups in this study would have restricted the performance capabilities of the infrared thermal imaging system (IR–TIS) and limited the external validity of the study.(3)
In the experiment, the pedestrian was presented at the following contrast level: white clothing against the fog background at night. The pedestrian walked perpendicular to the vehicle path, representing a pedestrian crossing the road (figure 1). Because of the variability of the fog-making conditions, each participant saw a single object multiple times with each VES. This provided a more precise assessment of object visibility independent of slight fluctuations in fog density. A pedestrian dressed in white was selected because it represents the type of object that would benefit the most from the ultraviolet A (UV–A) headlamps based on the results from the other weather condition studies and the expected fluorescence of the clothing material. The assumption was that the detection and recognition levels with the UV–A configurations for a pedestrian dressed in white would be better than for a pedestrian dressed in black because the white clothing would fluoresce more. In addition, a pedestrian crossing in front of vehicles has the greatest potential to affect safety. Detailed object characterization information is provided in ENV Volume IX.
The age factor had three levels: younger participants (18 to 25), middle-aged participants (40 to 50), and older participants (65 or older). These age groups were created based on literature review findings (refer to ENV Volume II) that suggest changes in vision during certain ages. (See references 4, 5, 6, 7, and 8.) Each age group comprised five males and five females. Gender was used as a control but not as a factor of interest.
Following is a list of the VES configurations:
For an indepth look at the technical specifications of each headlamp, refer to ENV Volume XVII, Characterization of Experimental Vision Enhancement Systems.
The presentation orders for each VES configuration were counterbalanced. Table 3 provides an example of the VES configuration order for a pair of participants. The first column, “Order,” indicates the order in which the VESs were presented. The second column, “VES,” presents the VES configuration that was used. The third column, “Vehicle,” indicates whether a sport utility vehicle (SUV) or a sedan was used for the VESs.
The six VES configurations tested were selected based on several considerations. The HLB and the HID headlamps are currently available on the market, and they reflect the most commonly used headlamps (HLB) and the headlamp type with a growing section of the market (HID). They were included as two of the configurations to allow comparison of new VES alternatives to what is readily available.
There was also some concern about possible changes in the detection and recognition distances resulting from the use of high-profile headlamps, such as halogens on an SUV, versus lower-profile headlamps, such as halogens on a sedan. This is important to consider because of the growing number of higher-profile vehicles on the Nation’s roadways; therefore, two halogen-based VESs of different heights were included, one at a low profile (i.e., HLB–LP) and one at a high profile (i.e., HLB) on a sedan and SUV, respectively.
The configurations that used the UV–A headlamps had to be paired with HLB headlamps because UV–A headlamps provide minimal visible light. These UV–A headlamps stimulate the fluorescent properties of objects irradiated by the UV radiation, producing visible light. Their purpose is to supplement, not eliminate, regular headlamps. Two different UV–A headlamp configurations were used in this study: hybrid UV–A and five UV–A. The hybrid UV–A headlamp is an experimental prototype that produces a significant amount of visible light, although not enough light to allow driving without standard headlamps. The UV–A headlamps used in the five UV–A configuration produce far less visible light. The UV–A and HLB pairings resulted in two different VES configurations: hybrid UV–A + HLB and five UV–A + HLB.
The IR–TIS was included because of its ability to present the driver with images of the environment based on the temperature differential of objects. This approach has the potential to allow very early detection of pedestrians, cyclists, and animals (i.e., objects generating heat) as well as roadway infrastructure objects that shed heat (e.g., guardrails, light posts).
There are several reasons for the decrease in the number of VES configurations from the clear and rain conditions (ENV Volumes III and IV). The need to limit participation to one night became apparent during pilot testing. The fog-making environment changed from night to night; therefore, all the VESs that needed to be tested were run in the same night to avoid the potential for confounding environmental changes and the within-subject variable, VES. To account for potential changes within the same night, fog density measurements were taken. (See Covariate section under Objective Dependent Variables.) Based on the results from the clear and rain conditions, the tested high output halogen (HOH) configuration was either no different or worse than the HLB, so it was decided that no further testing for this configuration was necessary. Similarly, the UV–A and HID pairings also were found to provide little improvement in clear and rain conditions, and therefore, were not included in this study. Halogen high beam was not included because this configuration is known for its potential to hinder driver’s visibility during fog at night because of increased backscatter.
Detection and recognition distances were obtained to analyze the degree to which the different VES configurations enhanced nighttime visibility while driving. These two variables were selected because of their common use and acceptance in the human factors transportation literature. (See references 9, 10, 11, 12, and 13.) Both terms, detection and recognition, were explained to the participants during the training session. Detection was explained as follows: “Detection is when you can just tell that something is on the road in front of you. You cannot tell what the object is, but you know something is there.” Recognition was explained as follows: “Recognition is when you not only know something is there, but you also know what it is.”
During training and practice, the participants were instructed on the use of a hand-held pushbutton wand used to mark the moments when they detected and recognized objects. The participants were instructed to press a button on the wand when they detected an object on the road, then press it again when they recognized the object. The front-seat experimenter flagged the data collection the moment the participant drove past the object. Detection and recognition distances were calculated from distance data collected at these three points in time.
The light reflected back to the driver from the fog, or backscatter, was used as a potential covariate for the detection and recognition distances. The fog movement behavior varied from night to night. Confounds resulting from this movement were mitigated by running all VES conditions in the same night; however, variations in fog density within the same night were still possible, which could have changed the amount of backscatter. For example, a less dense fog will result in a lower value of illuminance at the eye of the driver than will a denser fog because less light is reflected back from the fog (i.e., less backscatter). To account for this variation, the backscatter from the fog was recorded at the point of detection. An illuminance meter was positioned between the rearview mirror and the windshield. The meter was connected to a laptop that continuously read the illuminance measurements from the meter. The back-seat experimenter recorded the illuminance measurement obtained with the meter when the participant pressed the button for detection (appendix G). Backscatter was evaluated to determine if variations in fog density were enough to have a significant effect in the detection and recognition distances. If the illuminance due to backscatter affected the results, the values recorded at the point of detection (lux) could be used as a covariate to adjust the mean detection and recognition distances before evaluating the main effects and interactions for statistical significance.
Subjective ratings also were collected as dependent variables. Participants were asked to evaluate seven statements for each VES using a seven-point Likert-type scale. The two anchor points of the scale were one (indicating “Strongly Agree”) and seven (indicating “Strongly Disagree”). The statements addressed each participant’s perception of improved vision, safety, and comfort after experiencing a particular VES. Participants were asked to compare each experimental VES to their regular headlamps (i.e., the headlamps on their own vehicle). Researchers assumed that participants’ own vehicles represented what they knew best, and therefore, were most comfortable using. Following is the list of statements on the questionnaire:
Safety procedures were implemented as part of the experiment. These procedures were used to minimize possible risks to participants during the experiment. The safety measures included the following requirements:
The onroad pedestrians were trained when to clear the road, based on a preset safety-envelope mark. In addition, they were provided with radios in case the front-seat experimenter needed to communicate with them.
Onroad driving was conducted using four vehicles. The experimental vehicles included three SUVs and a sedan. All vehicles were equipped with laptops for data collection. Software was developed to link the data collection system to the vehicle to obtain distances traveled (i.e., electronic odometer) and speed. The distances from the points of detection and recognition to the pedestrian were obtained from the electronic odometer data. The software program on the laptop allowed the front-seat experimenter to change between VES configurations and pedestrian orders (figure 2). After all the pedestrians had been presented for a VES condition, the program went directly into the subjective questions asked by the experimenter. In addition, the software also gathered information such as the participant’s age, gender, and assigned identification number.
The software used by the back-seat experimenter was developed to link the data collection system to the illuminance meter to obtain the backscatter reading at the point where the participant detected the pedestrian (appendix G).
The VESs were distributed among the four vehicles (figure 3 through figure 6). The three SUVs had light bars installed, allowing experimenters to aim the test headlamps by adjusting its horizontal and vertical position. The HLB–LP and IR–TIS were the only exceptions; these were factory installed on the sedan. Note that either of the two SUVs equipped with the UV–A headlamps could also be used for the HLB-only configuration.
The all-weather testing facility on the Smart Road was used in this study (figure 7 and appendix H). The pedestrian was presented at four different locations on the Smart Road (figure 8). Each participant changed vehicles on the second turnaround of the Smart Road. One onroad experimenter was assigned to each participant; this experimenter was responsible for escorting the participant to the next vehicle, showing him or her where the controls were, and verifying that the correct VES configuration was being tested. Two other onroad experimenters were positioned at the various locations. One onroad experimenter was assigned to locations 1 and 5, and another onroad experimenter was assigned to locations 2 and 4. See appendix J for details on the protocol for the onroad experimenters. A total of four onroad and four in-vehicle experimenters (one in the front seat to monitor distance data collection and one in the back seat to monitor the illuminance meter software) were part of the study each night.
The all-weather testing facility on the Smart Road generated the fog (figure 9 and figure 10; overhead lighting was turned off for the study). Data were not collected during heavy wind conditions. Fog generation was the weather condition that varied the most from night to night. Operation of the fog generation system required continuous monitoring during each night of data collection to ensure the fog density was constant. The system used water and compressed air mixed at nozzles mounted over the center of the road to produce fog. The pressure ratio of the air and water was controlled to provide a consistent fog density. A visibility meter was positioned on the roadside to assist in this system control.
Figure 8. Diagram. Locations where pedestrians were presented for the adverse weather condition (note the area where fog was generated).
Figure 10. Photo. Smart Road overhead lighting system and fog towers starting to make fog.
The headlamps used for the HLB, HID, and UV–A configurations were located on external light bars. Each light assembly required a re-aiming process, which took place before the experimental session began. At the beginning of the Phase II studies, a headlamp aimer was not available to the contractor, so an aiming protocol was developed with the help of experts in the field. (See references 14, 15, 16, and 17.) The details of the aiming protocol used for this specific study are described in appendix K. During the photometric characterization of the headlamps, it was discovered that the position of the maximum intensity location of the HLB configuration was aimed higher and more toward the left than typically specified. The effect of this aiming deviation on detection and recognition distances is undetermined. The aiming could have resulted in more illumination on pedestrians leading to increased detection and recognition distances. The aiming could also have led to increased backscatter from the fog, resulting in decreased detection and recognition distances. Details about the aiming procedure and the maximum intensity location are discussed in ENV Volume XVII, Characterization of Experimental Vision Enhancement Systems.
Two participants performed the experiment simultaneously. The experiment, which consisted of two sessions, was performed in one night for each pair of participants. The first session included vision screening, laboratory training, and IR–TIS training. The second session involved the experiment on the Smart Road. The entire experiment lasted approximately 3.5 h. During the onroad session, participants were familiarized with the Smart Road and the experimental object before starting the experiment. Six VES configurations were presented to the participants; the order of VES presentation was counterbalanced. A discussion of the details follows.
Candidates initially were screened over the telephone (appendix A). If a candidate qualified for the study, a time was scheduled for testing. After the candidate arrived at the contractor facility, he or she received an overview of the study and then was asked to complete the informed consent form (appendix B) and to take an informal vision test for acuity using a Snellen chart, a contrast sensitivity test, and a color vision test (appendix C). The vision tests were performed to ensure that all participants had at least 20/40 vision and identify any type of vision disparity that might have influenced the results. A detailed experimenter protocol for vision testing appears in appendix D. If no problems were identified, the participant was trained on the tasks in the experiment to be performed during the drive.
Each participant was taught how to perform the tasks associated with object detection and recognition and told how the questionnaires would be used. The study protocol and the picture of the objects were presented at this point (appendixes D and E). The participants for this study received training similar to the training participants from the previous ENV studies received (i.e., clear, rain, and snow conditions). Even though only one type of object, the perpendicular pedestrian dressed in white, was used, participants were shown all the objects used in the other studies to maintain their multiple-object visual scanning behavior. Participants also were instructed in the detection and recognition definitions, use of the pushbutton wand, and the Likert-type scales for the subjective questionnaire. The purpose of this lab training and practice was to allow all participants to begin the experiment with a standard knowledge base. After the lab training, each participant practiced detecting an example of the experimental object using the IR–TIS.
Because the participants drove multiple vehicles as part of the study, each participant was familiarized as soon as he or she reached the next experimental vehicle. While the vehicle was parked, the onroad experimenter reviewed general information concerning the vehicle’s operation (appendix L). The participant was asked to adjust the seat and steering wheel positions for his or her driving comfort. When the participant felt comfortable with the controls of the vehicle, the experiment was ready to begin.
The participant was instructed to place the vehicle in park when he or she reached turnarounds 2 and 3 (figure 8) while the onroad experimenters prepared for the next lap. The participant also was instructed to drive at 16 km/h (10 mi/h) during the experimental sessions and to follow instructions from the in-vehicle experimenter at all times.
Each participant conducted a practice drive down the road to become familiar with the road and the vehicle. No object was presented during the practice drive. At the turnaround, the front-seat experimenter gave the pushbutton wand to the participant and instructed him or her that this portion of the session was a practice run to familiarize him or her with the protocol. The participant then drove back up the road for a practice run of detection and recognition tasks, obtaining feedback from the front-seat experimenter as needed. After the practice tasks, the participant began the tasks of the experiment, driving with each of the six VESs in the assigned order.
Distance data were collected while the participant drove with each VES. The front-seat experimenter provided the participant with a pushbutton wand to flag the data collection program when the participant detected and recognized the object. The participant performed no tasks while driving other than detection, recognition, and maintaining a speed of 16 km/h (10 mi/h). The front-seat experimenter sat in the passenger seat and told the participant when he or she could begin driving and where to park. The front seat experimenter also administered the subjective questionnaires after each VES configuration and controlled the data collection program. Details on the front-seat experimenter protocol appear in appendix F. The back-seat experimenter recorded the following information for each run: windshield wiper speed (low, medium, or high), illuminance reading at the detection point, and any additional information about the fog. Details on the back-seat experimenter protocol appear in appendix G.
Every participant followed the same sequence of events when collecting the data for each of the VES configurations:
The study was performed once every night from 7:30 p.m. to 11 p.m. until all pairs of participants had completed the experiment. Each participant was paid for the total number of hours (training and experimental session) at the end of the experimental session.
Data for this research were contained in one data file per VES configuration per participant. A separate data file contained illuminance data per VES per participant. All the data collected for the 30 participants were merged into a single database that included objective, subjective, and illuminance data. The illuminance measures were evaluated to determine how the fog varied for each VES. An analysis of variance (ANOVA) was performed to evaluate drivers’ visual performance under each of the different treatments. PROC GLM was used in SAS® statistical software to compute the ANOVA. The full experimental design model was used in the data analysis (table 4).
A Bonferroni post hoc analysis was performed for the significant main effects (p < 0.05). For the significant interaction, the means and standard errors were graphed and discussed. Post hoc analyses assisted in identifying experimental levels that were responsible for the statistical significance of the main effects. Note that the significance of a main effect or interaction does not make all interior levels significantly different. For a detailed discussion of post hoc tests and ANOVA, see Winer, Brown, and Michels.(18)