Adaptive Regions of Interest Based on HSV Histograms for Lane Marks Detection

The lane detection is a vital component of autonomous vehicle systems. Although many different approaches have been proposed in the literature it is still a challenge to correctly identify road lane marks under abrupt light variations. In this work a vision-based ego-lane detection system is proposed with the capability of automatically adapting to abrupt lighting changes. The proposed method automatically adjusts the feature extraction and salient point tracking cues introduced by the GOLDIE (Geometric Overture for Lane Detection by Intersections Entirety) algorithm. The variance of the lighting conditions is measured using hue-saturation histogram and abrupt light changes on the road are detected based on the difference between histograms. Experimental comparison with previously proposed algorithms demonstrated that this method achieved efficient lane detection in the presence of shadows and headlights. In particular, the accuracy of the algorithm applied on the footage with highest light variation increased 12.5% on average. The overall detection rate increased 4%, which illustrated the applicability of the method.