Obstacle Detection Using a Motion-Corrected Plan-View Transformation

Abstract This paper describes an obstacle detection method for a vision-based autonomous land vehicle (ALV) based on frame differencing. The simple frame-differencing approach has several drawbacks. and an improved method is presented, in wich the difference is taken of perspective projections of the scene. Since the ALV's motion is measured directly using sensors, a motion-corrected version of this difference can be produced. Parallax due to motion is then used to identify objects which project up from the ground plane. This process can be performed at near frame rate