Resolving occlusions with structural kalman filter

摘要

Moving object tracking is one of the most important techniques in motion analysis and understanding and it has many difficult problems to solve. Most of all occlusions during the process of tracking is very diffcult problem and it could make the tracking system fail to track moving objects reliably. With occlusions, a tracking systme can not continue to track targets because it can not match occluded regions in an image. So we need to study some effective methods to track moving objects in spite of potential occlusions. In this paper, we propose an advanced Kalman filter named structural Kalman filter which can predict moving objects reliably in case of partial occlusion by combining several Kalman filters corresponding to sub-regions of a moving object. By predicting occluded sub-regions, a trackign system can track such occluded regions continuously when the regions appear again. In other words, the predicted information is used to find the reappearing regions again.