Multi View Image Surveillance And Tracking

With the help of computer vision system and artificial intelligence ,machine can instead human brain to processing ,solving quite a lot problems in our life: unmanned traffic flow automatic statistics intelligent behavior partern recognition face recognition virtual reality and so on. These techniques is the core of the analysis and processing of video content, a single camera with its restrictions, however, cannot satisfied the demand of people's monitoring,people increasingly focus on multi-camera's collaboration .With the help of different view from the cameras , using computer vision ,deep learning ,provide convenience for people's work and life . And it is quite popular today. Object tracking as an important branch of computer vision, it is particularly important to research on its depth.For multiple cameras target tracking are mainly concentrated in multiple camera picture fusion and object tracking with computer vision algorithm those two aspects. Image fusion is used to expand the monitoring of the single camera, target tracking is used to reduce the work intensity of monitoring person. We know that the camera capture video frame sequence, is actually a 3d scene in the two-dimensional plane, the fusion of two planar image, involves the image coordinate system, imaging coordinate system, camera coordinate system and the transformation between them. Object tracking is to predict the current frame in interested in the position of target in the next frame, the traditional target tracking algorithm with Kalman Filter to track, the mean shift and particle filter. Modern tracking algorithm is tracking by detection, namely by training target characteristic,matching the characteristics of the target detection in the next frame.On the premise of deeply study the relevant theoretical basis, and studied the related content. The main contents are as follows:(1) The study of' image fusion algorithms: image fusion algorithm, the time cost of the various steps in analysis algorithm, select the computation of the optimal scheme for image fusion, including different feature points matching and the performance of different projection plane, Final, statistics 500 frames under complex scene image fusion time.(2) Object tracking algorithm research about the fusion image: research Kalman filter algorithm and the TLD tracking algorithm, modified TLD algorithm of random forest classifier in forest tree and the number of leaf nodes, to reduce the calculation time.Propose a new detection way in this paper,compared to original algorithm,it saves a lot of time.