Development Of Multi-camera Tracking System Based On DM6446

In recent years, target tracking has become a hot research topic in research areas such as machine vision, artificial intelligence, and video processing. However, due to the limited field of view of single camera, it is challenging to track target continuously in large-scale surveillance scenes. Therefore, target tracking by multiple cameras has become a dominant paradigm in modern industrial surveillance.In order to improve the robustness of the target tracking and achieve reliable large-scale tracking, a novel distributed target tracking system including target detection, target tracking, target handoff, cameras communication and cooperation is proposed. Firstly, image registration is done in the system initialization stage to calculate the image plane projection parameters among the cameras. After that, multiple subjects’ detection and tracking is carried on in each single camera utilizing Gaussian Background Modeling and Kalman Filter. Once there is target entering the overlapping areas among the cameras, target handoff is performed according to cameras projection parameters obtained. Moreover, all algorithms have been ported and real-time optimized to DSP platform. In system general framework, master and slaver configuration is employed. Specifically, master client is implemented using Microsoft Visual C++2008and slave client is realized using C language. Then TCP/UDP protocol is adopted for communication between cameras. Finally, plenty of tests have been done in different circumstances to validate the system’s robustness and efficiency. Experimental results show the robustness and efficiency of the presented system in real-time target tracking with multiple cameras in actual environment.