Offside Detection System in Football Matches using Human Pose Estimation

Abstract

Offside decisions are important in context of any football game. In recent times the decision making in sports have been heavily dependent on technology. Football is no exception. Recently the offside decisions and many other similar decisions on the football field have been made by a Video Assistant Referee.These decisions have been broadly inconsistent with the referees.Also these VAR decisions can sometimes take a lot of time causing delays. We can use machine learning techniques to tackle the problem of the offside rule in football. We make use of Keypoint R-CNN so that we can perform human pose estimation of players which information can be used for offside detection and image processing techniques to detect offside in a given frame. This dissertation will tackle all the problems we encounter in the process of offside detection in an image. The dissertation presents an improved offside decision system for football match images. We have also presented various challenges that the current method faces so as to facilitate further research in this area.