Vision-Based Method for Forward Vehicle Brake Lights Recognition

This paper presents a recognition algorithm combining the vehicle detection and the color difference of RGB color space to recognize the brake-lights state of moving vehicles in order to achieve the intelligent-car’s rear-end collision warning about the vehicle in front of it. Firstly, we train and build AdaBoost cascade classifier by haar features samples and scaling sub-windows are used to detect the target vehicle from the region of interest of the resized image. Then we compare the adjacent frames to recognize brake light status, which including using color, shape, structural features to identify the third brake light; comparing the center of gravity coordinates and the color difference threshold to rear brake lights when vehicles are not red or yellow; according to subtraction of each RGB corresponding channel, binarization, and the color difference threshold of RGB color space to identify the red or yellow vehicles’ brake lights. Finally, experiments show that the algorithm can detect the front vehicle’s braking quickly and accurately.