The Pedestrian Detection And Tracking Technology Research Based On Depth Map

Pedestrian detection and tracking technology is a hot research in the past decades years in the field of computer vision. It has a very wide application in intelligent transportation, machine vision, intelligent monitoring, computer auxiliary driving, and other fields. But conventional moving targets detection algorithms using visible image was often affected by the change of illumination conditions, interference of complex backgrounds, shadow of moving objects and moving objects of self-occlusion or mutual-occlusion phenomenon.We present a pedestrian detection method using the HOG of head and shoulder features based on depth map and detecting moving objects in particular scene in this paper. We calculate the integral images of depth images before extracting pedestrian’s features information of head and shoulder (HOG) to improve the efficiency of the algorithm. Using the threshold segmentation method based on distance information to Locate the moving objects areas (region of interest), which can remove the background information in the depth map effectively, also can improve the detection efficiency of the algorithm. Our algorithm detection efficiency can reach more than95%, each image processing time ranges from15milliseconds to100milliseconds, which have reached the requirement of real-time. Our algorithm is the good solution to the poor real-time and low efficiency of detection and identification in traditional pedestrian detection algorithm.On the basis of based pedestrian detection on depth map algorithm in this paper, through making a comparative study of several kinds of movement target tracking algorithm, selecting meanshift tracking method as pedestrian tracking method, the meanshift algorithm is the most commonly used in target tracking algorithm, firstly analysis the target, establishes a target model, then in the subsequent frames establish test model. By comparing the Bhattacharyya coefficient of similarity function between target model and measurement model, then after many times of iteration, eventually reaching an ideal result, to complete the target tracking. This algorithm can solve the problem in the traditional tracking algorithm, that is when the pedestrian moving fast, it is easy to lose tracked target. It also solved the problems of traditional pedestrian detection system low efficiency and poor real-time performance.