3D localization of moving object by high-speed four-camera vision system

Take the flying ping-pong ball as an example, this paper focuses on 3D localization of moving object based on a high-speed and high-precision vision system. The vision system is composed of two stereo pairs, and each pair stares at half of the table. The advantage of the system is that: it can redouble the precision while covering the same field of vision. In order to obtain the 3D positions of the whole trajectory, each stereo pair is first off-line calibrated. And then the online process is to detect the 2D position and reconstruct the 3D position of the ping-pong ball in each image frame. The detection process using circularity, color as well as motion features. Multiple visual features fusion method is applied to improve robustness. After 3D reconstruction, the final trajectories are merged from positions obtained by the two stereo pairs. Experiments show the feasibility of the system.