Research on recognition method of vehicle lane-change behavior based on video image

The research statistics show that about 60%-70% traffic accidents are caused by vehicle collisions. And lane change as the most common behavior when driving a vehicle is also the main reason for vehicle collisions. This paper based on the theories of video image processing technology, adopts the improved method of HOUGH to extract and fit the lane line. Through multi-threshold setting we determine the underneath shadow and further position the region of interest of the vehicle (ROI). In the case of offline, take features of large number of positive and negative HOG samples, and use the SVM to do the classified training. After extracting the HOG features, combine them with SVM classifier to achieve recognition of the vehicle. Then position the center of the vehicle, and use the binocular ranging method to get the distance between the vehicle and the lane lines. Finally, calculate the dispersion of the distance and set appropriate thresholds and determine whether the vehicle have the lane change behavior.