Sensor fusion-based visual target tracking for autonomous vehicles

In this ariticle, a data fusion based algorithm is proposed to identify and track moving objects for autonomous vehicle navigation. It is a challenging problem because both the object and the cameras are moving. Here, the optical flow vector field, color features, and stereo pair disparities are used as visual features, while the vehicle's motion-sensor data are used to determine the cameras' motion. We propose a data fusion algorithm which integrates information obtained from different visual cues and the vehicle's motion-sensor data for target-tracking. The fusion algorithm determines the velocity and position of the target in the 3D world coordinates. Next, we present 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 natural image sequences. © 2008 International Symposium on Artificial Life and Robotics (ISAROB).