Video Transformation for the Facilitation of Skill Learning in Sport

Abstract

This study was conducted on an video image transformation method that can facilitate skill learning in sport using recorded competition videos. The proposed method for the skill learning targeting karate competition are described in this paper. The method consists of two stages. The first stage compensates for partially filmed karate competition field in the input video image by generating a complete field image via a deep neural network (DNN) based image-generator. The generator was developed based on a pix2pix framework to perform this image transformation. The second stage utilizes the output image of the generator and the players’ stick models to convert the output image into a front view image. The transformed video is useful for observing not only the players’ motions, but also the entire flow of the competition. The evaluation experiment showed that the transformed video promotes ease of observation. By the proposed method, it becomes possible to simultaneously observe players' movement skills and the overall tactics on the field such as formations, thereby opening up new possibilities for video-based sports learning.