Object Location and Tracking in Binocular Vision System

A framework of detecting and tracking object is given by binocular vision system approach. Selecting the image in one frame of video stream from camera, instead of modeling object's features, the framework finds out the given object's position and pose, and then starts the tracking. There are three research points in the framework. First of all, online learners and detectors using object's image from different visual angle is built. Secondly, integrating with object's location calculation, a method of image matching is defined. Finally, based on the previous velocity, position and pose, object's tracking strategy is proposed. At the end of this paper, a vehicle tracking experiment is given and the result shows that the framework is robust and fast.