Visual 3D target tracking for autonomous vehicles

In this paper, an algorithm is proposed to identify and track moving objects for the autonomous vehicles target following application. It is a difficult problem because both the targets and the cameras are moving. Here, optical flow fields, color features, stereo pair disparity are used as visual features while vehicles´ motion sensors are used to determine the camera motion. Then this paper proposes a data fusion algorithm which integrates information obtained from different visual cues and the camera motion sensor data. The fusion algorithm determines the speed and relative position of the interested target in the 3D world coordinate for the vehicle to track. This paper presents a detailed description of the three-dimensional (3D) target tracking algorithm using an extended Kalman filter. Experimental results are presented to demonstrate the performance of the proposed scheme using different image sequences