A multiple object tracking method based on object chain

摘要

In this paper, we propose a multi-object tracking method. It uses the object chain from different cameras. Firstly, according to the Kalman filter, the object region in the next frame could be predicted, and if the overlap area between the predicted region and the detected region is larger than the threshold, setting the value to 1 in the matching matrix, otherwise setting the value to 0. Secondly, it will separately update the state of the previous and current frame by line and column scanning the matching matrix. The same object state should be put into a chain, at the same time, the relevant frame number and position is noted. Finally, it will realize multi-target tracking by the matched feature in the object chain from different videos. The experimental results reveal that the proposed method could track the occluded targets better and reduce greatly the tracking time.