Research Of Target Detection And Tracking Algorithm In Image Sequences

Target detection and tracking technology based on the visual image sequences is ahotspot in research field of computer vision, it has a wide range of applications inaspects of weapons guidance, security monitoring, and intelligent robots. The study ofthis paper focuses on the algorithms of automatic detection, mark positioning, trajectoryprediction and template matching tracking about moving target in a static scene. Underthe premise of meeting real-time requirements of the algorithm, the improvedalgorithms and related strategies are proposed in this paper, which suppress ambientnoise and scenes interference. With them the detection and location of multiple movingtargets and continuous and stable tracking of the selected single target have beenimplemented.The stage of target detection and location mainly realizes automatic detection andmark positioning of the target can be tracked in the image sequences. However, thebackground illumination changes, low contrast and small movement of objects in thescene will interfere with the correct detection of effective targets in the process of targetdetection. This paper makes a contrastive analysis of commonly used detectionalgorithms, and selects the background difference algorithm which has the better resaultof target detection. Combined with morphological filter the corresponding optimizedalgorithm scheme is proposed. Lack of background difference algorithm is too sensitiveto suppress scene interference. The experimental results confirmed that new algorithmscheme can overcome this problem and extract the effective moving targets in the fullfield of view. To distinguish and locate between multiple effective moving targets, thispaper studies recursive and non-recursive labeling algorithm, and summarizes theprinciple of selecting optimal labeling algorithm. Combined with centroid localizationstrategy the positions of multiple moving targets can be determined. To complete thetarget detection, the single target used in continuous tracking is determined by artificialselection.The process of target continuous tracking determines the same object’s positions indifferent frames. The basic idea of this paper is carried out target position matching orcalculation in each frame with the correlation of the image sequences and trajectoryprediction, which is a "prediction"-"matching"-"correction" algorithm framework.The "prediction" mainly determines the search and matching range of tracking target to ensure that the algorithm does not perform blind position matching operation. Thispaper studies prediction algorithms based on the motion trajectory fitting and Kalmanfilter, and puts forward the optimal of selecting optimal prediction algorithm. In theprocess of position "matching", this paper makes an in-depth study of the templatematching tracking algorithm, and proposes an improved correlation matching algorithmremoving the mean which has high matching precision and strong anti-jammingcapability. To overcome the defects of template matching algorithm, this paper studiesand proposes the corresponding accelerated algorithm scheme. To ensure the continuousand stable target tracking, this paper studies the "correction" processing strategies in thecontinuous tracking stage, such as similarity judgment strategies based on two-levelthresholds, adaptive template update strategy based on improved dynamic weightingmethod and target re-capture detection algorithm based on improved image differenceaccumulation algorithm.Through the results of simulation and actual system experiments about groundclose-up and slow-moving people and vehicles in the static scene, they confirm that thealgorithm scheme in this paper has accuracy, stability and real-time performance.