Research On Moving Target Tracking And Identification

In recent years, the rapid development of electronic technology makes the digitalvideo processing gradually become a hot research topic in the field of imageprocessing.Tracking and identification of moving objects are important researchtopics in the field of video processing and also the basic technology of many visualapplications. Firstly it detects a moving target from the surveillance video, then it doessome real time tracking according to the corresponding algorithm, and finally itmakes the judgment and the identification of species and behavior of the target. Theperformance of target tracking system and identification system directly determinesthe accuracy and reliability of the subsequent application.The main work of this paper is to study the algorithm of trackingand identification moving target that based on surveillance video. It contains threeaspects which are the moving target detection, tracking and identification. In theprocess of object testing, this paper combines the symmetric difference andmixed Gauss background model algorithm, as a result it manages to do accuratedetection and extraction of the target.As to the tracking, through the study oftraditional Camshift method, this paper proposes two improved algorithms, whichmake the tracking have better robustness and reliability. In the identification ofmoving targets in this paper,using the principle of support vector machine, byextracting the feature vector,it do classification of the moving pedestrians and vehiclesin video.Detection of moving objects: first of all, the basic process of detection isintroduced, including image preprocessing, detection operation, difference imagebinarizing processing and mathematical morphology processing. Secondly, by theclassic detection methods such as inter frame difference method、 backgroundsubtraction、optical flow method, it proposes a algorithm which merges symmetrydifference and mixed Gauss model, and succeed to do accurate detection andextraction of the target by doing phase or operation of three frame difference methodand two value image produced by mixed Gauss model.Tracking of moving target: discussing the basic principles of Mean-shift trackingas well as the Camshift target algorithm. Because the traditional Camshift can onlyextract the color characteristics of the target, particularly sensitive to the light andsurroundings, so when the target moving too fast or the chrominance components oftarget and background are close, it can easily lost the target or get phenomenon offailure. According to the above two problems, this paper put forward respectively theimprovement of Camshift tracking algorithm based on Kalman filter and SURF tracking algorithm based on Camshift: when the speed of the target is high, addingKalman filter to Camshift tracking,and do estimation of centroid position to the target;when the target color and background color are similar, we introduce SURF algorithmfor feature point matching between the two frame, to relocation the target, in order toachieve accurate and continuous tracking, improving the robustness and effectivenessof the system. Through the comparative experiments on the traditional algorithm andthe improved algorithm, the improved Camshift algorithm can track the targetaccurately that failed in the traditional algorithm, and it can improve the robustnessand effectiveness of the whole tracking system.Study on identification of moving target: It discusses the principle andcharacteristics of support vector machine (SVM), and classified-identification methodbased on SVM. In this paper, through the extraction of four dimensional featurevector of moving objects, by using the support vector machine classifier,we canclassify the pedestrians and vehicles of video. The experiment shows, the accuracyrate of two kinds of samples classification reached95%, which shows that the supportvector machine is an effective classification method. It can find the support vectorbetween classes of many training samples, to determine the optimal hyper plane, andeventually separate the two types of samples.