A Method Research Of Pedestrian Detection And Tracking

Pedestrian detection and tracking based on video is a hot reach research area in the field of computer vision. We study has a wide range of applications prospects in intelligent transportation, intelligent monitoring, intelligent robotics, video concentration and video transmission etc. However, the research is still facing many difficulties, because the complexity of the detected environment, the different posture of pedestrian, the shield, and the interference and so on. So that, the accuracy of detection and the stability of tracking are need to be improved in the near further.Our paper focus on the follwing issues based on the study of the existing pedestrian detection and tracking algorithms.1. Feature extraction is a key issue of pedestrian detection because the non-rigid feature of pedestrian. Firstly, the paper analyzed the pedestrian feature extractioa In order to improve the accuracy of the detection, we use multi-feature fusion with HOG feature and LUV feature to extract pedestrian feature.2. In the background subtraction method, stationary objects has a ghost and easy affect by illumination. Our paper analyzes the causes of the problem, and finally propose a sohition Besides, pedestrians mutual occlusion which often exist in a pedestrian tracking process affects the tracking accuracy. We builds a model by analyzing the specific process of mutual occlusion, in order to track accurately even if there is occlusion in the pedestrian tracking process.3. After detailed study of the reasons of the tracking instability in the complex scene, we use Online Boosting to update online to improved particle filter model, in order to track the target in complex scenes with occlusion. Experimental results show that the proposed Online Boosting pedestrian tracking algorithm based on particle filter improves pedestrian tracking accuracy and stability, and can achieve better results at the changing nature scenes.