Research On The Pedestrian Recognition Algorithm In The Vehicle Collision Avoidance Safety System Under Urban Road Conditions

In recent years, the road traffic safety is becoming more and more serious. How to reduce the traffic accidents and to ensure the personal safety has become the focus of the modern society. Automobile anti-collision safety system technology is one of the effective ways to solve the problem of road traffic safety, and thus, it was paid more attention of each big automobile manufacturer of the world. As a leading-edge technology in the research of vehicle anti-collision safety system, the pedestrian recognition in front of the vehicle can effectively reduce the probability of the pedestrian crash, so it is getting more and more attention by vehicle technology researchers.In this paper, the vehicle CCD camera is used as a visual sensor to collect the road image, and the DSP embedded controller is used as the hardware platform to recognize and track the pedestrians in front of the vehicle. Firstly, the virtual road region which is established in this paper is segmented in the image. Namely, the search scope of pedestrian targets is limited in the region with the greatest probability of occurrence, which can improve the efficiency of pedestrian recognition. Then it is to preliminary position of pedestrian targets according to the vertical symmetry characteristic of pedestrian body contour. And it can solve the recognition problems successfully caused by clothes changing or body with partial occlusion. Then it is the feature extraction of the pedestrian target area which is initially located. On the basis of the traditional LBP and HOG feature, this paper proposed a new pedestrian feature which fully inherited the advantages the two feature. Next, it is to classify pedestrians by pedestrian classifier which is trained by support vector machine. Finally, it is to use Kalman filter for tracking and forecasting the position of the pedestrian target which can further narrow the area of searching pedestrian target and improve the speed and efficiency of pedestrian recognition.The experiment result of real-time pedestrian recognition in the real road shows that the pedestrian recognition algorithm proposed in this paper has a good recognition effect on the pedestrians with different clothes、different postures、different motion states、partial body occlusion within a certain range from the car. Besides that, the system has a good performance in real-time when the vehicle is in a certain speed.