Moving Target Collaborative Tracking With Multi-Camera

This paper has mainly studied the target tracking algorithm of multi-cameracoordination．Firstly, the target tracking algorithm of single camera in the limitedcomplex scene is further studied．studied．And the research results can realize themoving target tracking in the complex scene． Finally studied the multi-cameracoordination target tracking, on the related theories and algorithms are studied anddiscussed．The main researches are as follows：①Huge computation and template drift are the defeats of traditional movingobject tracking method based on template matching, which may cause tracking failure.To solve these problems, template matching tracking algorithm based on motioningregion detection is presented. This algorithm use optical flow to estimate the movingregion, find the centroid of optical flow field motion, reduce the range of trackingregion, then template matching is used to track the object. The results of theexperiment show that the algorithm can effectively solve the problems of template drift,improve real-time of tracking and track the object in the video images.②Mean shift tracker is an efficient technique for target tracking. However, thetraditional Mean-shift tracker has difficulty to tracking fast moving target and suffers inlocal optimal problem. To overcome the limitation of Mean-shift tracking algorithm,SIFT feature is employed to detect the rough position of target. When Mean Shifttracking failure, use SIFT algorithm to extract the target feature. Then compute thecentroid as Mean Shift iteration initial point the next frame. Based on the rough positionMean Shift algorithm is employed to achieve precise tracking of the target.③Research on target handoff by using camera field of view line. When camerasinstalled the field of view line will never changed, not restrict by the environment andcamera parameters. Use SIFT algorithm to select4pairs of matching pointsautomatically, then determine the field of view line by projective invariants method.When the value of the target visibility discrimination function changes from negative topositive，it means the target will soon present in the scene of the adjacent camera，thendetects all the moving objects in the relative area and the targets which distance fromfield of view line are less than a certain threshold are considered to be identified target,take those target’s color histogram matching the real target’s, the best matchingendowed identification, then achieve the target handover. Finally, it summarizes the whole research and points out the direction of the futurework.