Robust Multiple Targets Tracking Using Object Segmentation and Trajectory Estimation in Video

In this paper, a novel robust unsupervised video object tracking algorithm is proposed. The proposed algorithm combines several techniques: mathematical morphology, region growing, region merging, and trajectory estimation, for tracking several predetermined video objects, simultaneously. A modified mathematical morphological edge detector was employed to sketch the contour of the video frame; and an edge-based object segmentation algorithm was applied to the contour for partitioning the predetermined objects; moreover, according to the motion of the objects, the proposed algorithm can estimate and partition the objects in following video frames, automatically. The proposed algorithm is also robustness against mobile cameras. The experimental results show that the proposed algorithm can precisely partition and track multiple video objects