Research And Improvement Strategy On Moving Target Tracking Algorithm In Video

With the development of computer technology, artificial intelligence has affected every aspect of people’s life. And one of the most important research areas in artificial intelligence how to make computers have visual ability, which is called machine vision. Target tracking is the most basic and important branch in the field of computer vision. It includes image processing, machine learning, pattern recognition and knowledge of other aspect. In many tracking algorithms, tracking algorithm based on filter theory is also in focus. Because of the characteristics such as good stability, high efficiency and easy to use Kalman filter algorithms widely used in the field of computer vision.Kalman filter was first used in the state space equation or linear dynamic system. Kalman filter provides a recursive method to solve the optimal estimation problem in linear system. Because it not only has the filtering ability but also has the ability to predict, it often used in visual tracking. However, Kalman filter is only applicable to linear systems and the computational accuracy will profoundly decrease when it is used in nonlinear system. Extended Kalman filter(EKF) applies Kalman filter theory to nonlinear system. However, when the system is strongly nonlinear EKF would be contrary to the local linear hypothesis, resulting in the decrease of filtering accuracy and even filter divergence. Moving objects in video sequences often exhibit strong nonlinearity such as target mutations. It is very easy to lost target and fail to track using the standard Kalman filter or EKF filter. Therefore, it is very necessary to researches on how to improve the Kalman filtering algorithm and EKF algorithm. To solve the problems above, this paper will focus on how to improve the KF algorithm and EKF algorithm which can handle the strongly nonlinear moving targets in target tracking. The main contents are as follows:(1)Thoroughly research the Kalman filter and apply it to video track. Further study the multi-innovation, propose Kalman filtering algorithm based on multi-innovation theory(MIKF) combining with multi-innovation theory and verify MI-KF in two aspects, experiment and theory.(2)Combining with multi-innovation theory, we propose Extended Kalman filtering algorithm based on multi-innovation theory(MI-EKF) and improve the filtering precision of the EKF in strongly nonlinear systems. Finally, verify the effectiveness of MI-EKF by simulation experiment and convergence.(3)Apply algorithm proposed by this paper to vehicle counting system. By using the object-oriented programming method, we analyze and design each module of the system in detail. The system consists of vehicle detection module and vehicle tracking module. The detection module uses the target detection theory mentioned in this paper and vehicle tracking module uses the MI-KF tracking algorithm proposed in this paper. Finally, by means of the actual project we prove that the improved algorithm proposed in this paper has reached the level of engineering application.