Accurate Motion Detection in Dynamic Scenes Based on Ego-Motion Estimation and Optical Flow Segmentation Combined Method

This paper presents a novel method for accurate motion detection in dynamic scenes without any prior information about moving object or dynamic scenes. Moving object detection is mainly performed by segmentation of estimated optical flow field, which is calculated by classical Horn Schunck algorithm. Robust ego-motion estimation is performed prior to the optical flow segmentation, which largely decreases the computational complexity in that a compensated background shows very small optical flow vectors and more distinguishable than the optical vector from moving object. Experiments on real video sequences from moving cameras demonstrate the effectiveness of the proposed method.