Design And Implementation Of Target Detection And Tracking Framework For Efficient Video Processing System

Today target detection and tracking of video processing systems framework constantly adds features to improve the real-time requirements, and the existing system framework to handle a variety of different shapes and sizes of the target still has deficiencies, thus paper puts forward an improved target detection and tracking processing algorithms, which can be applied to many different types of treatment goals and achieve higher accuracy and computing speed, to build a more closely integrated than the original system on the basis of software and hardware in an efficient framework.The main work is to first introduce current development target detection and tracking framework for video processing systems and problems, due to the wide variety of ever-improving gradually raised target detection and tracking algorithms, software and hardware for the implementation and platforms of the algorithm have a higher demand. Then paper based on the analysis of the basic theory of wavelet function, in order to solve the handling of non-local (NL)-Mean unable to meet the real-time nature of the problem and the lack of edge features, proposes adaptive threshold Curvelet fast weighted mean coefficient NL-noising algorithm, using weights in the upper and lower thresholds to limit the search window pixel, with smooth partition merges two pairs of integral sub-band, the proposed algorithm has better denoising effect. Then for the camera motion, different size and shape target, similar goals and mutual occlusion and undetectable clear background model and other issues, paper raises Harris3D histogram weighted density optical flow estimation block credit target goal mark and constraints associated with PN detection and tracking algorithm to search using estimation algorithm with time as the center of the non-static background for moving object detection, search algorithms and PN learning combined with structural constraints and target tracking algorithm, there is an effective approach to solving occlusion target detection and tracking in video stream under division target detection problem and to improve the detection results by density integration algorithm. The proposed algorithm successfully splits similar goals and objectives is blocked, gets higher accuracy in detection, tracking higher recall rate and the average overlap rate.In order to meet the system framework for dealing with different shapes of unknown size and reach goals or shelter with needs of real-time processing, the paper points out the detailed design work on the kind of hardware and software,which is complementary target detection and tracking of video processing systems framework principles and implementation results, using DSP and FPGA hardware framework combining experiments, paper proposed framework to more accurately and more efficiently complete multiple tasks in parallel, effectively dividing obscured targets, getting adaptive target frame size, and meetting real-time requirement.