Development of an Anomaly Detection System for Rugby Match Based on Majority Rule Using Deep Neural Networks

Abstract

This paper describes the development of an anomaly detection system for players in a rugby match to provide appropriate aid and avoiding severe injury. In the developed system, deep neural networks with fine tuning and its combination of different types of construction are key techniques for anomaly detection in rugby match footage. Certain appearance-based characteristics are captured from movie streams, and the system uses deep neural networks to automatically classify concerned scenes, especially, lying down scenes. The proposed system performs very well by recognizing many types of appearances of players within occlusion from a free viewpoint. Experimental results show that the system can effectively detect concerned players with high accuracy.