Occlusion prediction algorithms for multi-camera network

The mode of object tracking has evolved from single camera tracking to multi-camera tracking over the last few years. Even though multi-camera model overcomes limitations present in single camera system, it introduces complexities of handling network of multiple cameras. In this paper, we propose a novel real time occlusion prediction method for multi-camera network thus reducing complexities and cost of tracking in multi-camera model without losing track of the subject. The proposed method constitutes sequential phases as (a) estimation of direction of relative motion in real plane based on change in bounding pixel positions of tracked subject, (b) constructing algorithms for occlusion prediction, and (c) mitigation of occlusion by awaking a minimal subset of cameras in network. The method uses the pattern of change in the dimensions of bounding box with respect to frame number, obtained by applying background subtraction to the video of mutual motion of the subjects. On the basis of estimated direction of motion it uses proposed algorithms to decide the possibility and proximity of occlusion and thus awaking the minimal set of camera in the network that does not encounter occlusion. The proposed approach has been verified using various video samples and is observed that the proposed method successfully predicts the occurrence of occlusion.