Automated Offside Detection by Spatio-Temporal Analysis of Football Videos

ACM MMSports 2021
Ikuma Uchida, Atom Scott, Hidehiko Shishido, Yoshinari Kameda
Full Paper
October 20, 2021

Automated Offside Detection by Spatio-Temporal Analysis of Football Videos


Photo by Prapoth Panchuea on Unsplash

What is it about?

We propose a new automated method to detect offsides from football match videos. Our pipeline includes several methods to detect, track and identify the players and ball on the pitch in order to automatically extract the necessary information. In contrast to previous attempts that focus on detecting offsides in still images, our proposed method aims to automatically determine offsides from video.

Why is it important?

The past decade has seen the proliferation of technology-enhanced refereeing. However, video assistant referee (VAR) systems are ultimately performed by a visual inspection of video footage. Therefore it is nonetheless a subjective decision made by the referee. In comparison, the advantage of our method is that it can strictly follow the official offside rules in which the dynamics of play actions are considered. The contributions of this study can be summarized as follows: 1) To the best of our knowledge, this is the first automatic offside detection approach that takes the dynamics of play into account to implement the International Football Association Board (IFAB) offside rules precisely. 2) We propose a novel pipeline to track players via kinematic Kalman Filters and assign team IDs by solving a constrained general assignment problem based on dominant colors extract from the detected bounding boxes. 3) We demonstrate a method of detecting offsides from video record using inexpensive equipment. Therefore, this method can benefit amateur teams that have tight budgets as well as well-funded professionals.


Atom Scott University of Tsukuba

Writing this article was a great pleasure as it resulted in the first full-paper accepted at an international conference co-authored with Ikuma. I worked on coding the tracking pipeline while Ikuma did the coding for offside detection. He also wrote the majority of the paper and designed the beautifully illustrated graphics. The bottleneck for offside detection accuracy is undoubtedly the tracking module. Thus, I hope to improve on this in the future. We are currently working on releasing the code for this part, a pre-release version of the code should be available soon on Github. Check it out!