DSP-based Moving Target Recognition And Tracking System

Moving target identification and tracking is an important part of intelligent video processing and has very important value in the security surveillance, industrial production, military applications and other areas. Camera front side embedded video processing system with its advantage of low cost and high performance is being increasingly widely used, for identification and tracking of moving targets.This dissertation designed the overall embedded system plan, and completed the corresponding hardware design and software programming. After having implemented target recognition and tracking algorithm named Cam-shift in DM642 chip successfully, an improved Hu-Cam-shift tracking algorithm is proposed. Once the search window increases in Cam-shift iterative process, the search window for next calculating in Cam-shift algorithm is modified by comparing the Hu invariant moment between template and captured frames, then the Cam-shift algorithm is enhanced to keep tracing window from expanding to disturbing area in background.The improved algorithm performs well with low computational cost and makes the tracing robustly.The main contributions of this dissertation are summarized as follows: Firstly, a DM642-based target recognition and tracking system has been designed after a number of moving target detection and tracking algorithm are introduced, and the systems PCB board is already accomplished. In CCS3.3 software platform, the control of DM642's EMIF interface, interruption, EDMA, I2C module and programming of VP is realized. Based on in-depth analysis of the BT.656 digital video format, the video signal stream's sampling, transmission, storage and analysis modules are completed;Secondly, the Cam-shift tracking algorithm and the structure of DM642 chip is in-depth studied, then a target recognition and tracking system in DM642 chip is achieved successfully using the Cam-shift tracking algorithm, with which the low speed single target can be recognized and traced effectively. The main programming work includes probality look-up table establishing, anti-projection picture implementing and mean-shift iteration;Lastly, the development of video program utilizing the open source OpenCV library is discussed in detail. Considering the localization of traditional Cam-shift algorithm which is easily interfered by disturbing area in background of color similar to object, an improved Hu-Cam-shift algorithm is proposed. The search window for next calculating in Cam-shift algorithm is selected by comparing the Hu invariant moment of the image between template and captured frames.