Tracking multiple moving objects in video based on multi-channel adaptive mixture background model

This paper investigates tracking of multiple unknown objects in a video stream. Objects for tracking are added to the tracking list dynamically with the video going on. The task is to track all the moving objects in the video and obtain the trajectory of each object. In this paper, we propose a novel tracking strategy which can effectively locate a new moving object in the scene and its trajectory. Our strategy combines foreground extraction and optical flow points tracker. We extract the foreground using an improved Gaussian mixture model which can filter out the background. In order to determine the position of an object in latest frame, we calculate the displacement of objects’ feature points in previous frame using pyramidal Lucas-Kanade tracker. The reliability of the tracker can be degraded by a variety of factors, including illumination change, poor image contrast, occlusion of feature. To track the object more accurately, we implement a forward-backward error filter to evaluate the quality of results. Besides the already existing objects, our method also can detect new intruding object and add it into the track list, we implement a foreground search procedure to realize it. We develop an improved mix background model which can extract the video stream foreground real-time for moving objects detection. At the end of this paper, we illustrate the efficacy of our method by showing the performance of an example tracking system with this strategy.