DM642-based Moving Target Recognition And Tracking System

Recognition and tracking of moving target is an important content in the field ofthe intelligent video analysis. By analyzing the correlation characteristics of the videosequence images, we could extract, locate, recognize and track the moving target aswell as obtain the behavior of the moving target. Furthermore, explanation of thisbehavior provides a scientific basis for the final behavioral decision.We designed an embedded system for moving target recognition and trackingbased on the DSP chip, utilizing the Camshift (Continuously the adaptive mean shift)algorithm combined with the frame difference method. The system uses the framedifference method to recognize the moving object when the PTZ (Pan, Tilt, and Zoom)camera is stationary and then uses the Camshift algorithm to track the moving object.Controls of the PTZ are used to track its moving path. The system demonstrates goodperformance in effectiveness, real-time response and strong anti-interference ability astracking a single moving object under daily background. In order to cope with thedynamically changed background of the camera, we proposed an object detectionmethod based on the SIFT features match.The main contributions of this dissertation are summarized as follows:Firstly, we design a TMS320DM642based system of moving target recognitionand tracking, based on the completion of the system circuit board production andcommissioning work. Depth study of AVR chip Atmega16, realize the Atmega16andDM642communication, completion of the transmission actuator control signalfunction. On the CCS4.0software platform, DM642implementation of interrupt,EDMA module, VP, I2C and external SDRAM, Flash and codec chip programmingcontrol;Secondly, the architecture of DM642chip based on the deep research, a detailedstudy of Camshift tracking algorithm, a corresponding.Asm and.Cmd files. At thesame time in the hardware programming from YUV space to HSV space of the fastFourier transform algorithm. Finally, in the DM642chip to achieve: the use ofadjacent frame difference method of moving target recognition and tracking; usingCamshift algorithm based on simple background on a low speed single moving objecttracking;Lastly, based on the image SIFT feature point matching remove backgroundinterference method, using the OpenCV open source video library. Considering themovement of a camera makes the moving object detection more difficult under thecomplex background. In order to copy with dynamically changed background, anobject detection method based on the SIFT features match is designed. The feature points are detected by the SIFT algorithm to compute the parameters of the affinetransform model, guided by RANSAC, to compensate the global motion between theimages.