Detection and tracking of objects using a multi-camera system (Point Grey's Bumblebee XB3)

Abstract

Detection and Tracking of targets is essential in robotics world due to its promising applications in many areas. This final year report focuses on detection of moving objects in the vicinity and then tracking the detected objects as long as they stay in the view. Stereo vision provides human like vision to assist robots in it basic movements. It helps the robots to find the pose of an object in 3D, reconstructing 3D environment for path planning, retrieving the geometry of a 3D object. By taking images from three cameras with a known positional relationship, I can match positional points to determine disparity, and therefore depth. It uses the background as height map that is employed to easily extract foreground objects to determine the distance and position away from the camera. Tracking based solely on position information is unreliable when targets establish close position with the camera thus having the knowledge of color of the object will allow the system to increase the tracking robustness. Once an object has been detected in the view of Bumblebee XB3, the system is capable of tracking it while it still continues to detect for more object. A simple GUI interface was also developed.