Based The Ti Dm6437 Pedestrian Tracking Algorithm

Human tracking is an important branch in intelligent video analysis. Such algorithms can be used in people counting system, detecting people-show-up in forbidden region. Also, many higher level video information extraction algorithms such as anomaly behavior detection and human identification are base on human tracking algorithms, which makes them more useful. Meanwhile, the embedded processors'advantages are more and more obvious, so studying the implementation of human tracking algorithms on embedded processors is of great significance for real-world applications.Firstly, this dissertation briefly introduces the software and hardware development environment on DM6437. Then several already existed background subtraction methods are analyzed and compared, based on which and concerning the frequently occurring issue of camouflage and shadow in background subtraction results, a new method based on watersheds segmentation is introduced to compensate the background subtraction results. This method can relieve the problem of broken and fragmented objects caused by camouflage, and can get rid of some of those fake foreground regions that still exist after processing the background subtraction results using morphology methods. Primary experiment results show that, to some extent, this method can also reduce the false alarm rate of shadow detection algorithms. Then some related human detection and tracking algorithms are introduced, and the tracking algorithm's framework used by this dissertation is presented. Finally, concerning the algorithm implementation issue, software development and algorithm optimization methods based on DSP/BIOS real-time operating system are described, and the algorithm's performance on DM6437 is evaluated and analyzed.