Detection of occluded multiple objects using occlusion activity detection and object association

This paper proposes the detection of occluded moving objects using occlusion activity detection and an object association algorithm. When multiple objects are occluded between them, a simultaneous feature based tracking of multiple objects using tracking filters fails. To estimate feature vectors such as location, color, velocity, and acceleration of a target is a critical factor affecting the tracking performance and reliability. To resolve this problem, the occlusion activity detection and object association algorithm are addressed. The occlusion activity detection method provides the occlusion status of next state using the Kalman prediction equation. By using this predicted information, the occlusion status is verified once again in its current state. If the occlusion status is enabled, an object association technique using a partial probability model is applied. Using these algorithms, we can obtain the reliable center points of occluded objects respectively. For an experimental evaluation, the image sequences for a scenario in which three rectangles are moving within the image frames are made and evaluated. Finally, the proposed algorithms are applied to real image sequences. Experimental results in a natural environment demonstrate the usefulness of the proposed method.