Estimation of Racket Grip Vibration from Tennis Video by Neural Network

Abstract

In this paper, vibrotactile perception of a person in a video is automatically estimated from the visual and audio information. We limit the video scene to the back view of a tennis player rallying, but other factors such as locations, a player’s clothes, sound environments are arbitrary. We use tennis videos taken in three locations for neural network learning of the relation between the video and measured acceleration of the racket grip. Then we show the grip sensation can be successfully estimated from an unknown video. The quality of the produced vibrotactile sensation is evaluated by a subject experiment. The system is based on a similar concept to VibVid proposed by the authors. In this paper, we examine more general case than the previous research..