An improved occlusion handling for appearance-based tracking
Gwo-Cheng Chao, Shyh‐Kang Jeng, Shung-Shing Lee · 2011
The object occlusion is one serious issue in object tracking, especially when objects merge or split. The tracker may fail if it has no adaptability to such variations. In this paper, we present an improved solution of the appearance-based tracking for occlusion handling. The proposed method first recognizes the motion situations (merging and splitting of the moving objects), and then applies different template finding approach on each motion to get an association between detected blobs and targets. By this way, the accuracy of the data association can be improved when object occlusion occurs. Experiments were conducted to compare our results with those by other popular tracking algorithms. The proposed method is found robust for short time complete occlusion and partial occlusion.