H-008
Occlusion Handling of Multiple Objects in Indoor Depth-based Surveillance System

○Houari Sabirin・Sei Naito（KDDI R&D Laboratories, Inc.）

This paper proposes a method of handling rather complex occlusion of moving objects in a depth-based surveillance system using more than one static depth camera with no RGB signals. Individual tracking of occluded objects is made possible by separating the objects at the first stage of occlusion. The separation is performed by observing the depth data to find a distinctive statistical feature of the depth data of each of the occluded objects. After separation, each object is tracked individually by minimizing the spatio-temporal similarity feature of the objects. Experimental results show reliable identification of each object during and after occlusion..