Research On Target Tracking Algorithm Based On KCF And The Implementation Of Embedded System

With the rapid development of computer technology,machine vision has attracted wide attention and plays an important role in traffic monitoring,navigation guidance,weaponry,national defense operation and so on.Target tracking as a key technology in machine vision is becoming a hot topic in this field.In this paper,we mainly study the KCF algorithm and design an embedded system based on the architecture of DSP and FPGA.The specific research contents are as follows:(1)Three typical discriminant target tracking algorithms are studied which include TLD algorithm,Struck algorithm and KCF algorithm.Considering the robustness,real-time and portability,we use Benchmark video library to compare the tracking performance and real-time performance of these algorithms.The experimental results show that KCF has the best real-time performance,but the worst occlusion performance.(2)To overcome the shortcoming of poor occlusion performance of KCF,an improved KCF algorithm is proposed.The improved algorithm uses the peak response of the classifier to determine the target occlusion.If it is detected that the target is seriously occluded,the tracking result is determined by using the Bhattacharyya distance of continuous tracking box's histogram.When the target tracking failure is detected,the template matching method is used to search and relocate the target,then initializing the tracker.In addition,we use central position error,tracking success rate and real-time performance to evaluate it.The experimental results show that the improved algorithm has better robustness for target occlusion.(3)In order to transplant tracking algorithm into embedded system,we design an image processing platform based on FPGA+DSP architecture which consists of three parts: image acquisition,image transmission and processing,image display.Image acquisition part: In order to facilitate the installation of image sensor,a small industrial camera is designed;Image transmission and processing part: the FPGA is responsible for receiving and preprocessing image,and the DSP is responsible for the implementation of tracking algorithm;Image display part: successfully writing network port driver and using the computer interface to realize the observation of the camera image.