| Pedestrian video surveillance has been ubiquitous. Many public places including government unit, large public spaces, private apartment and grocery store have deployed monitoring systems. Although monitoring equipment has been popular, most still largely rely on human. It is often the case that one or a few operators sitting in front of the monitor to check if there are abnormal situation, the same as polices. Intelligent surveillance is not extensively applied. On account of multiple camera systems expand rapidly and visual datasets become complex and huge, depend on human is not workable. Intelligent pedestrian recognition technology must keep up with the pace. Person re-identification face multi-camera surveillance systems problem. Person re-identification refers to associate people which ender different camera views. Camera views change, scene changes and body posture can result in problems.This paper presents histograms of oriented gradients for human detection based on proposals algorithm, which can detect pedestrian rapidly, in view of the pedestrian detection problem. This paper also improved the algorithm of image segmentation based on Superpixel, which can apply to pedestrian segmentation. The accuracy of the pedestrian extract directly influence the outcome of people re-identification. At the same time the paper research pedestrian feature space conversion and matching algorithms. At last, the paper use CMC evaluation method to evaluate the overall effectiveness of people re-identification.In the course of the experiment the paper capture video from lab building. Experiments proved that pedestrian detection based on proposals algorithm faster than HOG pedestrian detection algorithm. And modified global and local search segmentation algorithm based on Superpixel could extract people from image. Finally, extract people feature and apply match algorithm so that acquire CMC result. Through the experiments improve the overall effect of people re-identification. |
PDF Full Text Request