Research Of Object Tracking In Non-overlapping Multi-camera

Modern monitoring systems are mostly used in shopping malls, banks and other largeestablishments, it should be used a large number of cameras to cover all monitoring, but thecost is too high, and there is no need. Therefore the non-overlapping multi-camerasurveillance system came into being.Because of the targets in this system are discrete inspace and time, how to matching the detected target in different cameras and using theresult of the matching to obtain the trajectory of the target within a certain time is thecritical issues in the field of non-overlapping multi-camera surveillance. In order to solvethe above problems, the paper proposes a combination of topological relations betweencameras and multiple target representation models, and the use of intelligent optimizationalgorithm for solving a set of objective optimal path, in order to achieve the goal ofcontinuous tracking. Work and innovation of the paper are as follows:(1) Research the acquisition of camera topology. This paper use a online topologyobtain method. We obtain the import and export of a single camera by the objects detectionresults, and use the Gaussian model to describe the transfer time between different camerasand provide a reliable time-constraint relations of target matching and association. Its effectis good.(2) Research the sample matching problems between the different cameras. Multipleappearance features and their matching algorithm are proposed included color histogramcorrected, H chrome model and SURF feature points. BTF can eliminate differences amongmulti-camera and compensate for the color histogram disadvantage of the lack of spatialinformation; taking into account the H component is not sensitive to light changes, wecombined it with that LFDA method can eliminate the interference besides the light; theSURF feature point is good robustness of rotation and scale changes, the similarity betweenthe moving targets is obtained by comparing the distance between the feature points. Thethree matching models can work together to improve the accuracy of the model matching.(3) Research the objects association problems among cameras. We using D-S evidencetheory to fuse the above features and improve the accuracy of the target tracking. And weuse discrete particle swarm optimization (DPSO) to extract effective association,which is toget the target trajectory in a certain period of time by the similarity of the sample.