Part-based multi-target tracking with structured learning

Multi-target tracking in video sequences is a difficult problem when most objects have very similar appearance and objects are close to each other in the image. The paper proposes a new approach to handling these problems by incorporating the target structure information. The structure information of a target is represents by a part-based model which contains appearance measurement and spatial constraints. To better estimate the parts of target, the spatial constraints are learned using an online structured learning algorithm. Based on this model, the adaptive tracker can search a region similar to the target while avoiding nearby targets. The experiments validate the feasibility of the proposed approach under the condition of occlusions and pose changes.