Quantitative model of the driver's reaction time during daytime fog – application to a head up display‐based advanced driver assistance system

Abstract

Road accidents because of fog are relatively rare but their severity is greater and the risk of pile-up is higher. However, processing the images grabbed by cameras embedded in the vehicles can restore some visibility. Tarel et al. (2012) proposed to implement head up displays (HUD) to help drivers anticipate potential collisions by displaying dehazed images of the road scene. In the present study, three experiments have been designed to quantify the expected gain of such a system in terms of the driver's reaction time (RT). The first experiment compares the RT with and without dehazing, giving quantitative evidence that such an advanced driving assistance system (ADAS) may improve road safety. Then, based on a modified Piéron's law, a quantitative model is proposed, linking the RT to the target visibility (V t ), which can be computed from onboard camera images. Two additional experiments have been conducted, giving evidence that the proposed RT model, computed from V t , is robust with respect to contextual cues, to contrast polarity and to population sample. The authors finally propose to use this predictive model to switch on/off the proposed HUD-based ADAS.

Vehicle Collision Prediction under Reduced Visibility Conditions

“…In 2015, Halmaoui et al proposed a quantitative model to describe the relationship between visibility and PRT during daytime fog. Then, the dehazing of images was performed using visibility enhancement algorithms in a Head Up Display [ 19 ].…”

“…Visibility has two definitions. One is the meteorological visibility: “the distance at which the contrast of a black object is attenuated by 95%”, and the other one is the target visibility: “it rates to what extent a target is detected in a scene, when an observer looks at it” [ 19 ]. Meteorological visibility can be estimated by detection of atmospheric light extinction from weather stations or visibility cameras, and the range is in kilometer level [ 20 ].…”

“…Especially in the visibility range between 40 m and 120 m, free-driving speeds are too high to allow for a successful stop for a stationary object [ 23 ]. Halmaoui et al also found the relationship between visibility and PRT during daytime fog [ 19 ]. Most work focus on how to detect and enhance vision under low visibility, such as in fog [ 24 ].…”

“…However, we do not speculate on how to convert foggy images into clear ones for some target visibility. This work focuses on “how to improve performance of rear-end CWS under poor visibility conditions.” Based on the research work mentioned above (mainly [ 19 ]), it is concluded that target visibility can affect driver speed, following distances, and the PRT.…”

“…Note that the prediction horizon is not fixed in this system, rather it is dynamically changed using the prediction horizon estimation model. For changing the prediction horizon, this model takes human factors into consideration by estimating PRT under different visibility conditions [ 19 ]. Under severe visibility conditions, PRT will be estimated to be longer than normal, and thus the prediction horizon will be increased so as to be able to predict future possible collisions and thus enhance driving safety.…”