Table tennis and computer vision: A monocular event classifier

Abstract

Detecting events in table tennis using monocular video sequences for match-play officiating is challenging. Here a low-cost monocular video installation generates image sequences and, using the Horn-Schunck Optical Flow algorithm, ball detection and location processing captures sudden changes in the ball’s motion. It is demonstrated that each abrupt change corresponds to a distinct event pattern described by its combined velocity, acceleration and bearing. Component motion threshold values are determined from the analysis of a range of table tennis event video sequences. The novel event classifier reviews change in motion data against these thresholds, for use in a rules based officiating decision support system. Experimental results using this method demonstrate an event classification success rate of 95.9%.