Application of Machine Learning in Soccer Broadcast: A Systematic Review

Abstract

Automatic sports video analysis has gained a significant increase in scale, due to the higher availability of video streams. This paper presents a review of modern techniques for video analysis of sports events, with a major focus on soccer.  Most scientific work in this field is centred on the last two decades, when higher computing power became common.  In the beginning, raw image processing algorithms were the standard, but recently the focus is the application of deep learning networks. This paper addresses several themes that are crucial in the development of content-aware video analysis systems for soccer games, namely object detection, classification and tracking, field segmentation and event detection.  A review of available datasets, related to soccer video data and annotations is also presented.  We believe that our findings can be useful for future video analysis of sports, in concrete soccer games.

Keywords: Computer Vision in soccer, Shot Classification, Field Segmentation, Object detection and tracking, Parsing, Event Detection.