Motion estimation using the Radon transform in dynamic scenes

This article describes a Radon transform based method for estimating the displacement in a monocular image sequence from dynamic scenes which contain rigid moving objects. The Radon transform is used to take projections along parallel directions from two consecutive frames in a monocular image sequence while the camera is stationary. The method consists of three stages: (a) motion detection, (b) identification of moving object areas in rectangular regions, (c) motion estimation for each isolated moving object. Moving object detection and displacement measurement using this method is insensitive to sudden or gradual overall illumination changes in the sequence. The method has the capability to be implemented in real time by using pipeline architecture. Motion estimation operations using the Radon transform do not need much time. Results using synthetic and real scene images are included.