A. Naftel and S. Khalid (UK)
object tracking, shadow detection, motion trajectory,occlusion handling
In this paper, we present a hybrid approach for tracking multiple objects through occlusion observed by a stationary camera. This tracking data can then be used to generate accurate object motion trajectories that provide an index key into a database of motion sequences. The system is intended for tracking people and objects in crowded environments such as supermarkets or shopping malls. It is shown how both static and dynamic occlusions are handled using a first-order Kalman Filter combined with a simple colour model incorporating histogram intersection and back-projection for each tracked object. In the next step we use a RANSAC-type approach to generate smooth motion trajectories for each object modelled with m-degree polynomials. Preliminary results are presented to show how our method produces robust trajectory paths insensitive to outliers containing high numbers of mis detected points. A novel nearest neighbour metric is then defined so that a user-defined motion query can be constructed to search a database of video motion clips and retrieve similar results.
Important Links:
Go Back
