Table tennis robot with stereo vision and humanoid manipulator II: Visual measurement of motion-blurred ball

In this paper a stereo-vision based framework is presented to track the motion of a table-tennis ball in motion-blurred images. In the proposed approach, first the motion direction is identified by minimizing L2 norm of directional derivatives of the blurred image and the motion width is estimated using the bounding rectangle of the foreground object. Then more accurate motion parameters of the ball are obtained based on circle fitting. Finally 3D location and velocity of the ball are reconstructed from a single image pair and weighted least-square method is utilized to estimate the flying trajectory of the ball. The proposed method is implemented in an actual robot table tennis system. GPU-based image processing and multi-thread technique are used to reduce the latency of the vision system. The obtained results verifies the effectiveness of the proposed method.