Distant object recognition with Grabcut for an active-zooming camera

In this paper a method for distant object recognition is proposed. We propose an iterative structure for distant object recognition using an active-zooming camera like PTZ camera. We adopt Graphcut-based segmentation algorithm, Grabcut, to our structure in order to select candidate regions for zooming and gazing. Grabcut is utilized through the modified strategy for our purpose. And we also propose two failure detection methods based on contour points and homography for our object recognition system. We show that our candidate region selection method is available for distance change from the camera, and also validate our recognition method through the real experiment.