Image segmentation and feature extraction for recognizing strokes in tennis game videos

摘要

This paper addresses the problem of recognizing human actions from video. Particularly, the case of recognizing events in tennis game videos is analyzed. Driven by our domain knowledge, a robust player segmentation algorithm is developed real video data. Further, we introduce a number of novel features to be extracted for our particular application. Different feature combinations are investigated in order to find the optimal one. Finally, recognition results for different classes of tennis strokes using automatic learning capability of Hidden Markov Models (HMMs) are presented. The experimental results demonstrate that our method is close to realizing statistics of tennis games automatically using ordinary TV broadcast videos.