A Collision Warning Oriented Brake Lights Detection and Classification Algorithm Based on a Mono Camera Sensor

Abstract

In this paper, a brake lights detection and classification algorithm based on a monocular camera and oriented to collision warning is described. First, the current driving lane is identified through a lane detection algorithm. Shrinking the search area to the actual driving lane, the preceding vehicle is detected exploiting YOLO object detector. Then, preceding vehicle lateral and third brake lights are identified in the $L^*a^*b$ colorspace by harnessing brightness and color information, along with geometrical considerations. Finally, brake lights status is classified by means of SVMs classifiers, based on features computed both on light regions and the overall vehicle image, and the braking status of the preceding vehicle is determined. The algorithm is designed to work in different illumination conditions during day time: experimental results prove the robustness of the proposed strategy with respect to different illumination conditions and brake lights shape, with an overall algorithm precision of 96.3%