Robust Camera Pan and Zoom Change Detection Using Optical Flow

Detection of camera motion is essential for automated video analysis. Changes in camera pan and zoom are two important events to be identified. This paper proposes optical flow based techniques to identify these events. The scheme detects pan change by finding the number of flow vectors that correspond to the dominant orientation. An orientation histogram has been proposed to quantify the direction of the flow vectors. Zoom-out and zoom-in are identified using the convergence and divergence of flow vectors respectively. A distance based method has been presented to identify the presence and nature of zoom. Numerical results presented for various data sets show that the proposed scheme is very robust.