Research On Moving Target Tracking Algorithm Based On The Cross-coupled Control Of Complex Scenes

Target detection and tracking is an important part of artificial intelligence. It hasbeen broadly researched and has made a wide range of applications in all aspects of life,such as intelligent video surveillance technology and precision guided weapon. All ofthem are based on static camera, motion detection in dynamic video scenes is inherentlydifficult, as the moving camera induces2D motion for each pixel, but the cameras are inthe state of moving in many applications, so the research of motion detection indynamic video scenes is of great importance.This paper focus on the technology of target detection and tracking, especially forthe target in moving background and the target detection algorithms which are base onfeature points and image matching and coupling are in-depth studied. And then thetarget detection algorithms combine with tracking filter, the optimal will be furtherverified. This paper focus on the following works:(1)Target detection is based on feature points. First, the common feature point isstudied, then the optimal feature point is used to extract the feature of the video indynamic background. And the features are classified by K-Means clustering, from thatwe can obtain the target detection.(2)Through matching the feature points between two frames we can get severalpairs of corresponding points, the motion vectors can be obtained through subbing thelocation of the corresponding points, at last by counting up the motion vectors we canget the motion vector histogram, in turn we can get the result of the target detection.(3)The matching point between two frames are used to solve the parameters ofcamera, then the parameters are used to crop the sub-region image which is used tomatching search in follow-up frames. At the same time, in order to improve theperformance of detection algorithm, we integrate the thinking of coupling with objectdetection, at last we can achieve object detection by image matching and coupling.(4)In order to verify several studied detection algorithms, we combine particlefilter with them to achieve the tracking result. At the same time, we can improve the performance of the tracking algorithm by integrate the feature points with state vector.In order to evaluate the validity of our algorithms, them are tested in datasetswhich are downloaded from internet and the landing dataset of the UCSD laboratory.Among the algorithms, the algorithm of image matching and coupling can get accurateregions of target. And also the promising results on datasets demonstrate theeffectiveness and robustness of our methods.