Tracking People in Presence of Occlusion
In most tracking algorithms, person-to-person occlusion is not dealt with explicitly, and the tracker is unable to label persons during occlusion. We present a framework that deals well with partial and complete occlusion for multiple people. Person-to-person occlusion is frequent in many indoor and outdoor situations, especially when cameras are mounted with a small angle of depression.
In this outdoor sequence, no person-to-person overlap occurs but one person hands over an object to the second person. While they are segmented out as one motion blob, our algorithm differentiates between them during overlap, and spatial constraints switch the identity of the object from person 1 to person 2