Skip to Main Content
This work constructs a real-time ball capture robot using the stereo computer vision technique and a novel biologically inspired 3D trajectory prediction method. The proposed trajectory prediction algorithm emulates the flight-to-light flight behaviour of the moth. The developed system can catch a flying ball according to the early prediction position generated from the proposed algorithm. The capture system is composed of four main components: camera calibration, object segmentation, ball location transformation from 2D to 3D coordinates, and 3D trajectory prediction. The experimental results demonstrate that the proposed moth flight-to-light tracking algorithm can predict the capture location more accurately than the Kalman filter does. The developed system can make early prediction in less than 69.582 ms and command the robot to catch the flying ball in time.