Detection And Tracking Of Moving Object Static In Static Scene

Motion target detection and tracking is an important research direction in computer vision research, moving target detection is the most-derived and the basis of movement target tracking, accurately detect moving objects from video scene has important significance the moving target tracking. Moving target tracking accuracy not only related to the accuracy of the motion target detection, but also to the surrounding environment changes, the stability, accuracy and real-time property of target tracking algorithm has become an important research subject.This thesis mainly studies motion target detection and tracking method on the static scenarios.For moving target detection problem, this paper describes the moving target of the commonly used static scene detection method, then study a detection method to solve the background of local area quickly changed circumstances, which uses the original into a foreground object of the background detection method, by fusing the background model frame difference method, gaussian mixture model and the code of the model algorithm, and inter-frame difference method using the background model object region can be divided into the background area and sports area, reintegration of Gaussian mixture model and the model code, can be more precise without leaving smear in the case of complete target detection. Experiments show that the algorithm can extract accurate real-time moving target. For moving target tracking problem, based on the analysis method of existing static scene moving target tracking, the usual tracking method of meanshift algorithm and adaptive camshift algorithms are combined with Kalman filter algorithm to track the moving target.Due to the moving object can be approximated as linear model, establish moving targets linear state model and observations model, detect the resulting center of the target position gives the Kalman filter observation, and to determine whether the target under occlusion factor is blocked. If not obscured,then the kalman filter as the target of prediction using the current frame position meanshift algorithm or camshift algorithm is the optimal value iteration to the center of the moving target, if the target is blocked, then the Kalman filter predicted position as the center of the optimal value moving target. UKF is a non-linear distribution approximated using sampling methods to approximate suboptimal nonlinear filtering problem solving, with stable performance, high precision, adaptability characteristics. But in the practical application, moving targets is usually nonlinear, in order to improve the tracking accuracy, this paper will UKF filter respectively with meanshift algorithm and camshift algorithm in order to improve the tracking accuracy. The target detection to get the center of the moving target gives UKF filter observations, and according to the barrier factors to determine whether a target obscured. If it is blocked,the UKF predictive value as the current frame position using the algorithm or camshift meanshift iterative algorithm to target the center of the optimal value. If the target is blocked, then the UKF filtering predict the optimal target location as the target location. UKF algorithm which fusions meanshift algorithms and camshift algorithm, compared to the KF which fusions meanshift and camshift algorithm has high precision. The algorithm can improve the real-time target tracking, accuracy and anti-blocking properties. And proved by experiments that the method is effective.The method of this paper in the case of moving target partial background change rapidly,moving targets can be accurate and effective real-time detection, and in the moving target with cover case, be able to quickly and accurately track the moving target.