Application Of The Improved Kalman Filter In Target Tracking

Target tracking is the process of studying, anglicizing and understanding the moving objects in a sequence of video images, for finding the location of the interested target. As an important branch in a field of computer vision, it is widely used in security, intelligent video surveillance, human motion analysis, intelligent traffic management and other fields, so the study of the target tracking algorithm has a high practical value and broad development prospects.The moving target detection is the basis of the target tracking. This paper firstly introduces the basic method of visual target detection. Based on the characteristics of each algorithm, this paper chooses to use the background subtraction method to achieve the detection target, and uses the digital morphology to reduce interference noise.For the simple moving target model, this paper proposes the moving target tracking algorithm based on Kalman filter. Firstly, it calculates the nucleus histogram target model for using the background subtraction to detect moving targets in the second chapter, thereby establishing the state space model of the moving target. Then it gets the parameters of position and speed using the historical frame, completing the update process of the Kalman filter, and then to obtain the step prediction covariance matrix. Finally, the use of measurement information (position or speed information) from the current frame is obtained by measuring the motion estimation calculated update position of the target to achieve tracking moving targets.In order to further improve the tracking accuracy, this paper proposes the method of combining the Kalman filter algorithm and Mean Shift algorithm for tracking moving targets.The method uses the Kalman filter algorithm to predict the position of the target, to give the area of the next frame target that may arise, and makes the Kalman filter estimated forecast as the starting point of the Mean Shift iterative algorithm to find the true position of the target, in order to achieve the goal of effective tracking. However, this method is only applicable to the linear model, which means the complex motion will no longer apply.For the nonlinear model moving target, and the problem of the error covariance matrix calculation overflow when the target is blocked, this paper presents the moving targets tracking based on the Cabuture Information filter algorithm, for using the inverse form of the Cabuture Kalman filter, namely the Cabuture Information filter, to estimates the target state,thereby to avoid the inverse covariance matrix, reducing the amount of computation target prediction process, to achieve the optimal filter design, combined with the Mean Shift algorithm for nonlinear visual target tracking.Finally,the thesis simulates and experiment on the visual target tracking algorithm based on the Cabuture Information filter algorithm by using simple motion model,complex motion model when the target is blocked.Also,the thesis analysesthe results of the simulation experiment,verifying the correctness of thethesis of the visual target tracking algorithm based on the Cabuture Information filter algorithm.