Research On Multiple People Tracking Algorithm In Video Stream

Research of Multiple People Tracking Algorithm in Video StreamThe moving object tracking is a very important issue in computer vision. The moving objects in the frames will be found out, and then their path and other information could be obtained. This information helps to analysis and understands the behavior of the moving object, and then more intelligent process can be executed. This article is mainly about the multiple people tracking algorithm in the video stream.Moving object detection should be executed before object tracking. This article analyzes and improves the motion detection algorithms. This article combined the background subtraction and frame differential method. Firstly, background subtraction is used to find the possible moving object, and the frame differential method is used to identify the scope of the moving object, then the moving targets will be segmented in their scopes. According to the features of the noises in the binary images, this article adopts improved threshold value area expunction algorithm to remove noises. Experimental results show that the improved threshold value area expunction algorithm has high efficiency.The features of human body are analyzed and improved in this paper, including color histogram, position, speed and gesture. Color histogram is often used to represent the feature of the color. In this article we set the color of the background black to avoid the influence of it. The speed is calculated in a very short time. Then the speed could reflect the moving state of the target at a certain time. The target is segmented from the binary image. Then the difference value of the partitioning partial between two targets decides the difference of their posture.In this paper, a tracking algorithm based on moving object detection is proposed. If the target enters this area firstly, he will be regarded as a new target and his features will be recorded in the database. If he is not a new target, the algorithm will find the best match record in the database. The target could be tracked successfully if the best match record is found.Occlusion often occurs when there are two or more targets in the surveillance area, and most tracking algorithm could solve the problem only when a target is sheltered partly in a short time. So a more prefect algorithm used to solve the occlusion problem is in need, because long-time or wholesale occlusion is unavoidable. In this article, the targets which are sheltered in the same area will merge into a new target, and then the new target will be tracked in the follow frames and its features renew in every frame until the occlusion problem is solved. When the targets which are sheltered separate, all of them could be tracked again. The algorithm we proposed could solve more complex occlusion problem. We associate the algorithm proposed in this article with the particle filter algorithm to track moving targets. The targets are tracked by the algorithm proposed in this article when occlusion does not happen. When the occlusion happens, particle filter algorithm will be used to solve the problem when a target is sheltered partly in a short time. So when the two algorithms associate, it is able to solve the occlusion problem.