Research On Pedestrian Detection And Multiple Pedestrian Tracking And Application On Video Surveillance Platform

In this paper,we propose one method of pedestrian detection algorithm and multiple pedestrian tracking algorithm based on human's head and shoulder,and we also develop a video surveillance platform to validate these two algorithms.In the section of human head and shoulder's detection,we firstly introduce the concept of tracklets feature.Based on this concept,we train one head and shoulder's tracklets' traversing energy map.Furthermore,we propose the foreground' tracklets extraction,energy analysis strategy and local main gradients strategy to filter out the features which do not belong to head and shoulder.At last,we use the real head and shoulder' tracklets' feature and tracklets' energy map to calculate the final results.After getting the detection results,we propose one multiple pedestrian tracking algorithm based on Markov Chain Monte Carlo(MCMC).The special points of our tracking algorithm include bidirectional information node,hierarchical data driven MCMC and post-process algorithm.These points can help to get good performance under the condition of insufficient detection results.At last,we develop one video surveillance platform based on server and client.We use our detection and tracking algorithm to do early warning work to validate these algorithm.