Moving Target Detection And Tracking In Video Sequences,

Moving targets detection and tracking related to computer image processing, video image processing, pattern recognition, artificial intelligence and other areas, is widely applied in many aspects such as military, industrial, traffic, life and so on.This paper mainly focuses on moving targets under the static background, especially under the traffic background. The study is divided into three areas:the moving target detection and extraction, the target's shadow inhibition and removal, the target tracking.The innovation and the major works of this paper include:1. As for the extraction of moving targets from the static background, two algorithms are introduced in this paper, algorithm based on the difference between the consecutive frames and the algorithm based on the difference from the background. But the experiment proved that these two algorithms have their own shortcomings. In order to make up for their shortcomings, this paper combines the two algorithms. An improved method of moving targets detection is proposed. The algorithm to extract the initial background based on gray-change statistic, and update the background in real time are proposed. Then the targets detection has a certain self-adaptive. Finally, by the use of smallest external rectangular box, these targets are marked.2. In the shadow detection aspect, this paper firstly introduces the current commonly used in several different color space under the shadow of detection method, and then gives the experiment results and analysis. In order to improve the speed and accuracy of shadow detection, through studying the shadow of the formation mechanism, the paper firstly determines a rough area of the shadow, and then detects the shadow in the rough area based on normalized cross-correlation function. After experiments, this algorithm of moving objects with the shadow have higher detection capability and speed than the commonly used method.3. In the targets tracking aspect, this paper focuses on several tracking methods based on the template matching and the characteristics matching. Template matching algorithm locates the targets accurately, but has a larger operation. At the same time, the template image of the imprecise estimation will have a huge impact on the results of tracking. In order to improve the tracking speed, the paper proposes a number of improved algorithms. Firstly, we can ascertain the rough location of the target, then matching the template point-by-point in the region. This algorithm has greatly enhanced the speed of the track. Secondly, amend and update template dynamically. It will reduce the adverse effects of background disturbance and noise and ensure tracking accuracy. In the feature matching algorithm, the paper selects the target's location, size and corner to matching characteristics, using least squares and optical flow estimation to predict the characteristics. So that the amount of calculation of matching is smaller than template matching and greatly increasing the speed of matching. Finally, using the cost function to amend the predicted characteristics value. It will update the target's characteristics adaptively, improve the matching accuracy. Even if the targets to be tracked have the change in size or shape, the algorithm can accurately track the targets.