Research Target Detection And Tracking Of Moving Image Sequence

With the development of hardware and software, the computer vision has been applied widely and plays an important role in video surveillance, robot control, automatic driving, aerospace and so on. Moving object dectection and tracking is one of the most important technologies. Although a lot of researches on the detection and tracking have been employed by the scholars, there are still some problems in this area.Some algorithms of detection and tracking are improved and the ideal results by experiments are represented in this paper.(1) On the object detection, taking the update speed and calculation of backgrounds extracted by Gauss mixture model method into account, an improved method combining statistical histogram with pixel difference is proposed. Firstly statistical histogram method is used to obtain the candidate backgrounds and parameters for updating the backgrounds. Then iteration method is used to calculate the learning parameters by pixel differences. Finally the result proves the accuracy of the modified method and the real-time performance of system is improved.(2) On the object tracking, two parts which include single object tracking and multiple objects tracking are focused on to conduct the research. On the single object tracking, the form center and speed of search window are used to build the Kalman model to improve the Camshift algorithm for solving the problem caused by the similar colors of backgrounds. On the multiple objects tracking, the improved Gauss mixture model method is used to obtain and update the backgrounds at first. Secondly the expectation maximization method is used to build and update the object models. The data association and occlusion reasoning are used to solve the problem of occlusion. At last the result of the experiment proves the validity and accuracy of the improved method.