The Research Of Intelligent Video Tracking Algorithm Based On Multiple Cameras

Intelligent video tracking is a new research goal in the field of computer vision in recent years.Intelligent vision detection and tracking areas include: pattern recognition,image processing,computer vision,artificial intelligence,etc.A large number of researchers and institutions in the world have launched the visual inspection and tracking research.The main work of this paper is as follows:(1)Starting from the current development and the need to solve the problem of multi camera tracking,we conducted in-depth analysis from the single camera tracking,overlapping multi-camera tracking and non-overlapping multi-camera tracking.At the same time,we summarized the difficulties and research emphases of intelligent distributed tracking system and found the importance of how to construct robust tracking features for multi camera video tracking.(2)We construct a robust video tracking framework and extend it to multi-camera video tracking.The tracking framework is coordinated with four mpdules: detection module,tracking module,prediction module and learning module.First of all,the detection module can make good use of detection information of each frame;secondly,the tracking module using well tracking information between frames;thirdly,the prediction module is used to estimate the target position of the current frame to reduce the detection time.Finally,the learning module can continuously update and learn the information of the detection module and the tracking module.(3)We propose a new texture feature,called x-bitBP,for multi camera video tracking.And artificial immune random forest is used in texture feature for rapid calculation and statistics;then,x-bit BP texture features and color features are fused to form a more robust feature;brightness transfer function is added to reduce the light change on video tracking results.(4)Aiming at the learning module and updating problem in multi camera video tracking algorithm,a multi-state self-learning template library updating algorithm is proposed,called RS-TLU algorithm.According to the similarity of tracking target,history template and instantaneous template,RS-TLU algorithm divides the tracking results of each frame into three states: steady state,gradual change state and sudden change state.Motion information and occlusion information are used to determine the tracking results for each state.(5)Based on the tracking algorithm and template updating algorithm proposed in this paper,we design an intelligent distributed tracking system with multiple cameras.In this system,the camera agent is applied to the video tracking process.The camera agent can learn the surrounding environment and cooperate with the communication mechanism between cameras to solve the problems of occlusion,illumination change and the target disappearance.At the end of the thesis,we summarize the research results and look forward to the future work direction.