Targets Re-identification And Matching Based On Intelligent Assistance For Blind-area-crossing

In video surveillance applications, the target tracking technology has been drawing the most attention in the research communities home and abroad. As the demand for public security is rapidly growing, analyzing the behaviors of pedestrians by utilizing the computer vision technology has made some progress in both application and theory area. However, in the surveillance camera network where some blind areas exist, re-identifying and matching the pedestrians crossing the blind spots still faces a lot of problems to be solved.In many public areas, the monitored field is composed of several sub-areas and each of these sub-areas is electronically watched by only a single camera. There normally exists an area that is not monitored by either of the two adjacent cameras and usually is called "blind area" or “nonoverlapping area”. Since the spatial and time information about the targets to be tracked are discontinuous due to the non-overlapping area, it brings greater challenges in solving the problems with re-identification and matching of the pedestrians, and adds more complexity on analyzing the activities that take place in the blind-area. In this paper, the following four issues are mainly dealt with:Firstly, in order to better detect the pedestrians in single camera, in this paper, based on the statistical learning methods, the author uses the well-trained differentiating part(WTDP) as the model to identify the pedestrians. A WTDP is composed of three parts: root filter, part filter, and spatial location model. The root filter covers the whole target and the part filters are responsible for the parts of pedestrians with high resolution. The spatial location represents the position of each part of the model.Secondly, for the case of target tracking with a single camera, a new target model is put forward based on the modified Heaviside function. It mainly uses the difference between the foreground and background characteristics of the distribution to establish the new target model. The experimental results show that the new algorithm gives better performance in accuracy and speed of tracking robustly than the traditional algorithm. This algorithm is also applied to the multipedestrian tracking.Thirdly, the target’s feature model and similarity measurement methods are analyzed. Based on fusion of target features, an algorithm is presented for matching the blind-area crossing pedestrians.Fourthly, summary is made and the weaknesses of the proposed method are analyzed. A matching method is proposed which is based on the optimal combination of target features for the blind-area crossing pedestrians, which finds the best feature combination from the comparison experiments. The experimental results indicate that the matching rate is further improved than that presented in the previous chapter. In addition, the matching speed is compared respectively with that found in the existing literature.