The function of adaptive front-lighting system is to improve the lighting condition of the road ahead and driving safety at night. The current system seldom considers characteristics of the driver’s preview behavior and eye movement. To solve this problem, an AFS algorithm modeling a driver’s preview behavior was proposed. According to the vehicle’s state, the driver’s manipulating input, and the vehicle’s future state change which resulted from the driver’s input, a dynamic predictive algorithm of the vehicle’s future track was established based on an optimal preview acceleration model. Then, an experiment on the change rule of the driver’s preview distance with different speeds and different road curvatures was implemented with the eye tracker and the calibration method of the driver’s preview time was established. On the basis of these above theories and experiments, the preview time was introduced to help predict the vehicle’s future track and an AFS algorithm modeling the driver’s preview behavior was built. Finally, a simulation analysis of the AFS algorithm was carried out. By analyzing the change process of the headlamp’s lighting region while bend turning which was controlled by the algorithm, its control effect was verified to be precise.
AFS (adaptive front-lighting system) is a front-lighting system that can change the light pattern and illumination area according to the vehicle’s state such as the velocity, the steering wheel angle, and road environment to light the road ahead effectively so as to reduce accidents at night. According to AFS productions of international companies and research institutes’ research findings and directions currently, the classic algorithm of AFS can be divided into the following four categories.
On the basis of these experiments, Young et al. used the regular pattern of heads’ turning angle at daytime and night to guide the design of AFS [
Considering the regulation and safety, the algorithm of AFS currently estimated the vehicles’ future driving track according to vehicles’ moving states, traffic environment, and drivers’ wheel input. Then the AFS algorithm controlled the headlamp’s deflection angle to illuminate the estimated driving track in advance. But most of these algorithms assumed that the speed and the wheel angle would not change and assumed the driving track as a circle. According to Zhang et al.’s study on vehicles’ lateral moving characteristics [
Meanwhile, as the most important active safety device, the headlamp’s main effect is to provide illumination for the road ahead. Hence, the AFS should satisfy the safety demand, ensure that the driver’s concerned region is fully illuminated, and must not influence the driver’s fixation behavior. Its illumination effect also has a direct relationship with the driver’s eye comfort. Compared to a good environmental illumination condition such as in the daytime, it should not additionally increase the driver’s visual fatigue. In this aspect, although some researches have used the statistic rule of drivers’ visual field to control the headlamp’s deflection, in the practical application, this method which was totally based on the statistic rule cannot ensure the safety.
Based on the considerations above, an AFS algorithm considering the driver’s preview behavior was proposed. The vehicle’s kinematics and dynamics characteristics were used comprehensively to calculate the vehicle’s future track which was more reliable than the method purely based on the vehicle’s state information the headlamp’s deflection lag and direction error was avoided. Under the premise of ensuring safety, parameters of the driver’s fixation behaviors were introduced to increase the measuring precision of the driver’s visual statistical rule by using the eye tracker and the headlamp’s illumination is more in compliance with the driver’s fixation behavior. This paper proposes an original technology route; the work was first carried out in early 2010, and it has obtained national invention patents [
By modeling the driver’s preview behavior at a bend, the vehicle’s future track in a period of time was predicted according to the vehicle’s current state and the vehicle’s steady state response characteristics to the driver’s wheel, gas pedal, or brake pedal input. Simultaneously, the driver’s fixation location on the future track was determined according to drivers’ preview behavior rule (the rule of fixation location or preview spot) at real bends; then the headlamp’s deflection angle was controlled and the location was illuminated effectively. The technical route of the algorithm was shown in Figure
The specific technical route.
The algorithm simulated the driver’s preview behavior at a large curvature bend and an integral algorithm to predict the vehicle’s track was proposed. The algorithm was based on the hypothesis of steady preview and dynamic correction [
The coordinate system.
Because every time piece is quite short, the mutual effect of the vehicle’s longitudinal and lateral movement can be neglected. On the basis of the vehicle’s state (location, velocity, and acceleration) at the initial moment of the time piece, the vehicle’s state at the end of the time piece was calculated using the rigid body kinematics principle [
Then the vehicle’s mass centre coordinate (
Adaptive front-lighting system deflection angle calculation mathematical model.
Based on the above theory, the measurement of the preview distance for real drivers was conducted. By the advanced eye tracker, the driver’s fixation behavior was recorded when the driver drove through the bend. The driver’s gazing direction was analyzed and the average value of the preview distance was calculated. Then an empirical equation of the relationship between preview distance, preview time and velocity, and road curvature was proposed and the preview time for the AFS algorithm was modified.
SmartEye Pro eye tracker (sample frequency: 60 Hz), A JETTA car made by FAW-Volkswagon, a 12 V spare battery, meter ruler, the adhesive tape to make the lane, and others.
Eight drivers with normal vision. Three quarter-circle lanes with curvature radius of 20 m, 30 m, and 40 m were used as the experiment road. With every lane, the experiment was done for three times repeatedly. Before the formal experiment, every driver did 10-minute exercise. The experiment velocity and the curvature radius were shown in Table
The experiment velocity and the curvature radius.
Radius of curvature (m) | Velocity (km/h) |
---|---|
20 |
|
30 |
|
40 |
|
The SmartEye Pro eye tracker recorded the driver’s raw eye movement data. According to the thresholds of eye movement parameters set up by the users, the fixation and saccade behaviors were confirmed. Meanwhile, by the corresponding software, the video of the driving scene was replayed and the fixation location (the green circle in Figure
Driver’s gaze point position comparison chart when “being close to the bend.”
The threshold settings of eye movement were as follows: the threshold of visual angle deviation: 2 deg.; the eyeball movement velocity threshold during fixation: 15°/s, which was the highest velocity of eyeball movement during a fixation; the eyeball movement velocity threshold during saccade: 35°/s, which was the lowest velocity of eyeball movement during a saccade; the fixation duration threshold: 200 ms.
According to the road environment, the state change of the vehicle, the fixation location (visual angle), and its change in the driving scene video, the entire driving process was divided into four sections: “straight lane,” “being close to the bend,” “entering the bend,” and “being out of the bend.”
“Straight lane”: in this section, the entire bend was shown in the driver’s visual field. Repeated experiments showed that the driver would sweep the bend and form a rough impression of the available track; then the fixation location would be a few meters away from the vehicle as shown in Figure
In the section of “being close to the bend,” because the vehicle’s movement state had not changed, the nonpredictive AFS algorithm could not predict the bend ahead and activate the AFS bend lighting function.
“Entering the bend”: as shown in Figure
Driver’s gaze point position comparison chart before and after into the corner: (a) is before into the corner, and (b) is after.
In the “entering the bend” section, the driver needed to steer and the AFS needed to activate bend lighting function and control the headlamp’s deflection according to some rules. So according to the purpose of this experiment and our algorithm’s parameters’ need, this section was the emphasis of our study.
“Being out of the bend”: the driver would brake when he were going to be out of the bend. This section would not be analyzed in detail.
The preview distance was calculated in the “entering the entrance” section during the driving process. The fixation point in the front near the centre line of the road was defined as the preview point. The SmartEye Pro system was installed as shown in Figure
The eye tracking system’s installation location.
The world coordinate system and gaze direction angle.
As shown above, the angle
The gaze direction’s horizontal angle is
The gaze direction’s vertical angle is
GazeDirection.
According to the gaze direction and the gaze origin in the world coordinate system, the preview point’s coordinate could be calculated in the world coordinate system as shown in Figure
Preview distance data under 40 m bend radius.
Data operation | Bend radius (20 m) | ||||
---|---|---|---|---|---|
10 km/h | 15 km/h | 20 km/h | 25 km/h | 30 km/h | |
Mean | 9.6253 | 9.7248 | 9.8515 | 10.0431 | 10.1059 |
Variance | 1.0292 | 0.9282 | 0.8827 | 0.8319 | 1.2074 |
Maximum | 7.5492 | 7.7634 | 8.0599 | 8.0124 | 8.2775 |
Minimum | 12.3558 | 12.0545 | 12.1912 | 12.4505 | 12.1334 |
Preview distance data under 20 km/h velocity.
Data operation | Velocity (20 km/h) | |
---|---|---|
30 m | 40 m | |
Mean | 9.4637 | 11.3121 |
Variance | 1.8896 | 1.4886 |
Maximum | 6.6242 | 7.6275 |
Minimum | 12.7901 | 13.9346 |
Preview distance calculation schematic.
When the road curvature radius was 20 m, the preview distance increased slowly from 9,6253 m to 10.1059 m with the adding of the velocity. When the velocity was constantly 20 km/h, the driver’s preview distance increased too while the road curvature radius increased from 30 m to 40 m. Noh et al. had used the preview time and the response of neuromuscular system as main human factors to study the preview distance at different velocities and different road curvatures whose conclusion was similar to ours [
The experiment’s result was consistent with the previous test’s conclusion. With the consideration of the characteristics of the preview distance’s distribution and change rule, the preview distance-velocity regression analysis was done to get the preview distance-velocity empirical equation (
At the speed of 20 Km/h, the empirical equation of the preview distance-curvature radius was as follows:
According to the relationship between the preview time and the preview distance,
According to (
The optical analysis software LucidShape was used to simulate the headlamp’s light distribution and acquire the headlamp’s isolux curve on the ground. In the following simulation process, the envelope area of the isolux curve at a certain illumination was regarded as the headlamp’s irradiation area. According to the relative relationship between the area and the road, the control effect of the algorithm was verified by the headlamp’s irradiation effect. With the reference of ECE regulation, a dipped headlight in accordance with the ECE regulation was chosen in our simulation and its optical structure was not changed. The dipped headlight’s location settings were as follows. The headlamps were installed in front of the vehicle and the headlamps were 1.2 m apart and were symmetrically on both sides of the vehicle’s longitudinal axis. The installation height of headlamps was 0.65 m.
The simulation result of the headlamp’s illumination on the road was shown in Figure
The simulation result of the headlamp’s illumination on the road.
Meanwhile, because the road’s curvature radius was relatively small in our simulation, to clearly verify our algorithm’s irradiation effect, the envelope area of 321x isolux curve was drawn in LabVIEW as the headlamp’s irradiation area which was shown in Figure
The envelope area of 321x isolux curve.
The vehicle model, driver model, and road model of CarSim were used in the simulation. The AFS algorithm and the real-time display of the headlamp’s deflection angle, beam location, and irradiation effect were realized in LabVIEW. By the cosimulation of CarSim and LabVIEW, the bend driving simulating real environment was achieved. According to the calculated headlamp’s deflection angle, irradiation area, and the vehicle’s future track, the algorithm’s control effect was verified.
The vehicle’s speed was 30 km/h, the curvature radius of the bend was 20 m, and the preview time
The headlamp’s beam deflection angle of our algorithm was shown in Figure
The headlamp’s deflection angle and steering wheel angle.
The front-lighting illumination area simulation schematic.
It was shown in Figure
The contrast algorithm was a nonpredictive algorithm proposed by Ishiguro and Yamada which was relatively mature [
It was shown in Figure
The headlamp’s deflection angle and steering wheel angle (contrast algorithm and our algorithm diagram 1).
The front-lighting illumination area simulation schematic diagram 1.
The headlamp’s deflection angle and steering wheel angle (contrast algorithm and our algorithm diagram 2).
It was shown in Figure
At the moment of
At the moment of
At the moment of
Both algorithms’ deflection angle could follow the input of the driver’s steering angle. Relative to our algorithm, the contrast algorithm’s deflection angle was about 20° bigger when the steering wheel began to deflect.
The specified control effect was shown in Figure
The front-lighting illumination area simulation schematic diagram 2.
By the simulation result analysis of these two conditions, the conclusions were as follows. The future track’s prediction of our algorithm was more precise (the purple line in the image during the entire simulation was more close to the lane’s center line), so the algorithm would control the headlamp’s deflection more precisely to ensure the front road’s illumination effect well. Relative to the contrast algorithm, when the vehicle was in the bend and was going to be out of the bend, the headlamp’s deflection angle of our algorithm was smaller; the headlamp’s beam could cover the front road more well and make sure that the region the driver was interested in was in the central zone of the headlamp’s beam which meant better illumination effect. When the vehicle was just in the bend and was out of the bend after a while, the headlamp’s deflection angle of both algorithms was the same and their illumination effects were basically the same too.
In summary, the proposed algorithm could predict the vehicle’s future track more precisely; on the basis of it, the headlamp’s deflection angle was controlled to improve the front road’s illumination condition effectively. Meanwhile, the proposed algorithm could make sure that the central zone of the headlamp’s beam covered more regions which the driver was interested in.
The previous road experiment has found that the preview distance increased linearly with the velocity and the road curvature radius’s increasing. Based on this rule, the precise preview distance was acquired by the eye tracker. The average preview time under the speed of 10–30 km/h is 1.70 s. Under the curvature radius of (20–40) m, the average preview time is 1.44 s. An empirical model of the preview distance was proposed which accorded with the driver’s visual characteristics.
Based on the dynamic predict algorithm of the vehicle’s future track using the driver’s best preview acceleration model, the algorithm’s fixed preview time was modified by road experiments and a full AFS algorithm was developed based on the algorithm. The future track’s prediction of the proposed algorithm was precise.
Finally, by the cosimulation of CarSim and LabVIEW, the headlamp’s illumination area and the vehicle’s future track of different time were analyzed in detail in the bend driving process. The illumination effect was analyzed in two aspects and was compared with a mature nonpredictive algorithm. The result showed that the control effect was obvious and the algorithm had a good application prospect.
The authors declare that there is no conflict of interests regarding the publication of this paper.
This paper is supported by the National Natural Science Foundation of China (50975120), Specialized Research Fund for the Doctoral Program of Higher Education (20120061110028), Jilin Provincial Research Foundation for Technology Guide (20130413058GH), and Program for Chang Jiang Scholars and Innovative Research Team in University (no. IRT1017), China. Thanks are due to them.