Research On Intelligent Target Tracking Algorithm And Its Implementation On Embedded System

With the development and progress of the information age,machine vision technology has been developed by leaps and bounds.As a key research topic in this field,target tracking technology has been more and more widely applied in military and civilian areas.In this thesis,the KCF target tracking algorithm is deeply explored and improved,and the design and implementation of embedded target tracking system based on the framework of FPGA+DSP is completed.The main work is as follows:(1)To overcome the shortcomings of poor anti-occlusion of KCF algorithm,target occlusion decision,model updating strategy and search and reposition strategy after target loss are added,and propose AKCF algorithm.AKCF algorithm certain degree of occlusion and occlusion of the determination by the response peak correlation filters,and controls updating the tracking algorithm according to the intensity level of the covered object.When it is determined that the target is seriously blocked and lost,the algorithm updates are stopped and the target is searched and relocated.After recapturing the target,the tracking algorithm is updated and the tracking continues,thus achieving a highly robust and effective tracking.(2)For scale transformation,this thesis adds an asynchronous multi-scale detection strategy based on response discrimination to the AKCF algorithm,and presents a SAKCF algorithm.In the SAKCF algorithm,the peak response of the filter is also used to evaluate the tracking quality.When a possible scale change is detected,an asynchronous multi-scale detection method is used to find the appropriate tracking scale to solve the scale transformation problem.At the same time,an improved segmentation algorithm is used to modify the selected target template to improve the accuracy and robustness of the tracking algorithm.The experimental results show that the SAKCF algorithm proposed in this paper improves the accuracy and success rate by 4.4%and 8.9%,respectively,compared with the original KCF algorithm,and has better robustness.(3)To improve the real-time performance of the algorithm on the embedded platform of DSP,this thesis makes further improvement and optimization of the algorithm.The optimization algorithm runs in a structure that reads the image data directly from the original image as much as possible,avoiding unnecessary time and space consumption;combines with some algorithm libraries improved by the DSP platform,realizes efficient two-dimensional FFT calculation,and performs some operations such as multiplication,division and matrix operation in the optimization algorithm.The experimental results prove that the optimized algorithm has high.real-time performance on the DSP platform,and the tracking frame rate is about 42FPS.(4)The software framework of embedded target tracking system based on FPGA+DSP is designed and implemented,and the overall structure of the system is introduced,including image transmission based on SRIO,host computer interaction based on RS422 serial port,etc.Finally.the SYS/BIOS framework is used to improve the software framework in a multithreaded way,making the system more stable,fast and easy to expand.