This paper describes a new method for detecting and tracking multiple moving objects in a moving world. Motion detection from a moving observer has been a very important technique for autonomous driving and driver supporting systems. Because of the continuous background changing, detecting the real moving objects has become very difficult and always employs optical flow method to measure the flow vectors difference between background and the moving objects. Unfortunately, it has to pay a huge calculation cost to obtain the accurate optical flow vectors. A new motion detecting method based on camera 3D-motion analysis is proposed in this paper. With the special motion feature of on-board camera and FOE (focus of expansion), this method can detect the real moving objects by simply using the matching between two adjacent frames. The camera 3D-motion can theoretically be determined in the method with only three matching pairs, which make it fast and more efficient for real-time applications. The output moving object candidates are tracked over time in consideration of tracking cycle and estimation of their reliability and vitality degree. Experiments on plenty of real outdoor road scenes show the accuracy and efficiency of this approach. (A*)
Abstract