Multi-view tracking using Kalman filter and graph cut

In this paper, we propose a multi-view approach to detect and track based on graph-cut and Kalman filter algorithms to solve this problem. The first, object appears in the scene be detected as foreground in each view using a background model and background difference. Next, for related between cameras used homographic constraint. Any pixel inside the foreground object in every view will be related by homographies inducted by the reference view. reference view Images converted to binary images by a graph-cut segmentation. This step separated the position of the intersection points from other parts inside reference images. This added step significantly reduce false positives and missed detections due to points noise or when it cannot be guaranteed that a single reference view image will consistently by scene objects. To track, We measurement the average position of the points. The kakman filter provides an optimal estimate of its position at each time step. The filter kalman, the first one is the prediction of the next state estimate using the previous one; the second is the correction of that estimate using the measurement. Experimental results with detailed qualitative analysis are demonstrated in challenging multiview crowded scenes.