Research Of The Object Matching Algorithm In Non-overlapping Multi-camera

With the social progress, target detection and tracking is hard to meet the practical needs for single camera. In order to expand the scope of monitoring, in recent years, multi-camera target detection and tracking is taken seriously. Considering the contradiction between the broadness of the surveillance areas and the limited view range of the signal camera, it is impossible for the cameras to cover the whole region to be watched. Single camera target detection and tracking technology is mature, and how to track and judgment objects through these blind area to another camera is hard, the no-overlap multi-camera target tracking is now a research hotspot.Due to the difference between the multi-camera, so the same targets should have very different characteristics in them, and the time and space of them is separated on them, so how to match the same target in non-overlapping vision is the key problem in target tracking, we propose a multi-feature data confusion algorithm which combines several object appearance features with the topology relationship among the cameras. The main jobs and contributions in this paper are list as follows:1) Because the weakness of real-time and robustness for the movement target detection in single camera, and have to study dozens of frames, this paper proposes a background extraction algorithm of VIBE to extract moving object. And VIBE use the first frame to build the background model, combined with the target pixels eight neighborhood to establish pixel model. Generally in the second frame we can extract the good moving objects. With the background will introduce noise and illumination change, this paper proposes a algorithm using pixel level and frame level based on the background to update the background model. Experimental show that the proposed updating algorithm have a good adaptability on the illumination change.2) For the matching problem of object in non-overlapping multi-camera system, multiply appearance features and their matching algorithm are proposed including color histogram corrected by bright transfer function, UV model and SURF feature points. This paper puts forward using SURF feature point matching Euclidean distance as an important condition, through the comparison of the feature point nearest and second close method, to work out the distance ratio to mark the similarity between the two moving target, the smaller of the ratio of two object matching degree is good enough.3) This article use the system topology structure as one additional features, due to the different time and space model between multi-camera, that is, between multi-camera is connected or not, and the average distance between connected camera, we generally use the average time to express the distance between them. We use the mixture of Gaussians model to describe the time between different camera.4) Considering the characteristics of target tracking about overlapping vision, using D-S evidence theory to fuse the above features, and realize tracking the goals which through non-overlapping region, avoid the conflict of the above characteristics, and improve the accuracy of the target tracking. At the same time use the method of graph theory to extract effective path, combine with the minimum cost flow model and use maximum cost to extract effective association.