Robust multi-object tracking to acquire object oriented videos in indoor sports

In this paper, we propose a robust multi-object tracking algorithm for acquiring object oriented multi-angle videos, which takes advantages of two different tracking techniques represented by subdivided color histogram based tracking and labeling based tracking. Object models based on color histograms are further subdivided to differentiate similar color regions. Another tracking technique utilizes automatic detection by background subtraction and matches labels by comparing previous tracking results and detected object locations. The developed tracking algorithm more robustly recovers from tracking failure cases such as missing objects or overlapping objects than histogram based tracking. To cover large areas of interest, multiple camera images are integrated using homography based transformation. Based on tracking results in the integrated coordinate system, panning-tilting-zooming (PTZ) cameras are incorporated to acquire multi-angle videos of the tracking object. The feasibility of the proposed tracking based multi-angle video acquisition is demonstrated through real indoor sport game scenarios.