The Research Of Moving Object Detection And Tracking Algorithms For Single Target

Due to the complexity of scene,it is difficult to track a single-target persistently,and making it's hard to be popularized in surveillance video.In order to realize intelligent tracking for targets in surveillance video,an in depth research for Tracking-Learning-Detection target tracking algorithm was done,aiming at real-time tracking and detecting of targets.The algorithm can constantly tracking moving objects,however,because of the complexity of this algorithm,it is slow to process images.What's more,the algorithm has poor robustness when target rotates or strongly deforms,therefore,it cannot meet the requirements of real-time and stability in surveillance video system.To solve above problems,the main work includes following points:(1)Deeply research several classical single-target tracking algorithms,compares its experiment results with TLD tracking algorithm,analyze their benefits and drawbacks respectively.(2)For the real-time issue of algorithm,on the basis of TLD algorithm framework,reduce the number of samples to be inputted to detector by dynamically adjusting the size of images and using local scan instead of global scan,to meet the real-time requirement.(3)To solve the robustness of the algorithm,on one hand,use the SIFT algorithm instead of uniformly spread to obtain the feature points;on the other hand,use the SIFT matching algorithm to improve the tracking process of tracker.This two steps provide double assurance for TLD algorithm,and enhance the robustness when target rotates or strongly deforms.(4)Design and implement the software and hardware environment of monitoring system platform,and a set of video monitoring system was developed by using QT development tool.Meanwhile,embed improved TLD algorithm into this system and test the tracking performance of improved TLD algorithm in the system.