An appearance based human motion tracker

Human motion tracking is a common requirement for many real world applications such as video surveillance, games, cultural and medical applications. In this paper an appearance based tracking system is proposed which tracks human motion from a video scene. The system is based on appearance based color histogram and color correlogram model. Multiple people can be tracked consistently by this proposed system. This system works well in indoor and semi-indoor environment. The proposed system comprises of five major steps. Firstly all motion blobs in the video scene are extracted. Morphological image closing are applied then on extracted motion blobs for removing anomalies associated with blobs. After that all the motion blobs are counted and labeled by connected component finding. Correlogram and histogram model for each blob are built from the color information of moving blobs in the next step. Similarity measures between the blobs of the current image frame and previous image frame are calculated. Finally system performs tracking based on the similarity measures from frame to frame. The system is consistent during partial occlusion. During occlusion occluding entities are tracked as one entity and tracked separately when they split.