Moving object detection by a mounted moving camera

Accurate moving object detection is one of the most important topic for surveillance systems. Background subtraction works well for videos acquired by stationary mounted camera. However, it does not work for mounted moving cameras since the background objects also move in two consecutive frames. In this paper, we propose an optical flow based moving object detection algorithm for video sequences obtained by a mounted moving camera. Our algorithm first finds the interest points in the consecutive video frames and then tracks them with pyramidal Lucas-Kanade method. After matching interest points in two frames, it generates optical flow vectors. We assume that the majority of the optical flow vectors are based on camera motion. Our algorithm determines the camera motion vector by averaging these vectors. It then crops frames for removing camera motion to determine overlapping areas of two frames. Finally, we calculate the frame differences to detect moving objects. We tested our algorithm on several video frames and observed that our algorithm determines moving objects with a very high accuracy.