Heatmap Visualization and Badminton Player Detection using Convolutional Neural Network
Muhammad Abdul Haq, Shuhei Tarashima, Norio Tagawa
Abstract
Badminton coaches and analysts are more interested in how well athletes do in games by watching video matches. However, they still keep watching the whole video by hand, which is inconvenient and might cause them missing important information in the video. Most studies for sports video analysis have been done on soccer and volleyball, but badminton has not been fully focused on. Based on this observation, in this work we aim to build an automated system that can track the position of a player from an input badminton broadcast video, and visualize its position statistics on a heatmap. Convolutional neural network is used to track players and their position is projected on 2D court map using homography. In this paper we validate our approach using videos collected from the Badminton World Federation (BWF) channel on YouTube.