Design Of Single Target Tracking System Based On FPGA

Object tracking has been an active research area in the computer vision community.It has a variety of uses in video surveillance and pedestrian tracking. Since the existing tracking system is difficult to meet the high-precision and real time requirements when the target is under rotation and scaling,we propose tracking algorithm based on feature points and Kalman filter and implement a single target tracking system on Xilinx Spartan-6 FPGA board in this paper.The tracking algorithm extract the feature points by using Difference of Gaussian(DoG) and Canny operator,and match the keypoints by using the simplified SIFT descriptor. We introduce direction constraints and offset distance mode to solve the mismatch of featured point,and then predict the object position by using Kalman filter to achieve real-time accurate tracking.In the system design,the feasibility of the tracking algorithm is verified by simulation,and then a tracking system is built by using CCD,TVP5158,TFP410,monitor and other devices.On this basis,several modules like DoG feature point extraction,Canny edge detection,feature points description and matching,Kalman filter are logically implemented and optimized on FPGA.Finally,the tracking system is used to track the human in the real scene.The experimental result shows that the tracking system can perform the tracking at a rate of 25fps for 720x576 sized video.In addition, the system enables the robust tracking of target with rotation and scaling in real time,which can meet the design requirements of the system.