Research On Multi-cameras Cooperative Tracking Method Based On Adaboost

With the constant growth of surveillance area and the extension of surveillance requirements in the video surveillance, more and more various monitoring devices are applied to monitoring the environment, the intelligentization of video surveillance gets more and more important. As an important branch of intelligent video surveillance technology,multi-cameras coordination and control strategy based on target tracking is becoming one of the most great research direction. Multi-cameras cooperative target tracking can reduce resource redundancy of the system, enhance target tracking stability, and cooperative fusion of information can also enhance the richness and robustness of the target information in the video surveillance. This paper focuses on further study of multi-cameras coordination and control methods in the video surveillance environment with overlapping fields and a single target.Firstly, this paper illustrates the key technologies of target detection and tracking in different kinds of video surveillance environments, and analyzes overall architecture of the visual tracking systems and relevant theories of multi-cameras coordination and control.Secondly, factors that influence the dynamic tracking for a single target based a PTZ network camera are analyzed. When video stream captures from the network camera are pre-processed, the tracking algorithm with Kernelized Correlation Filters can predict the target position in the sequence frames, and proportional control algorithm is also used to control the motion of the camera based the PTZ camera control model and the deviation of target position and center point of picture, the PTZ camera can dynamicly track the target through the P/T motion.Thirdly, the target detection methods based on cooperative learning are analyzed.According to the detection classifier can be obtained by off-time training of samples, this paper focus on the classification of Cart and SVM. And for target detection and feature dimensionality reduction,an improved SVM-adaboost target detection method is proposed in this paper based on adaboost.Finally, this paper focus on method of multi-cameras coordination and control based on the integration strategy of target detection and dynamic tracking system of PTZ camera. It take the similarity between the target feature information and the corresponding fusion filter template as measuring parameter to select a series of cameras to track target dynamicly., and a strong classifier is formed by the integration of weak classifiers generated by the tracking algorithm with Kernelized Correlation Filters, which can re-detect target after missing the target.Based on several standard visual database and the indoor surveillance experimental platform, the dynamic tracking performance of single PTZ network camera, the effectiveness of ensemble learning strategies for target detection, and the multi-cameras scheduling performance of method of multi-cameras coordination and control are verified in this paper.