Annotation Techniques for Judo Combat Phase Classification from Tournament Footage

TL;DR

This paper introduces a semi-supervised method for classifying judo combat phases from tournament footage, aiming to automate annotation and analysis with limited labeled data, showing promising initial results.

Contribution

The study develops a transfer learning-based semi-supervised approach to classify judo combat phases from video, addressing data scarcity issues in sports analytics.

Findings

Achieved F1 scores of 0.66, 0.78, and 0.87 for different classes.

Demonstrated the feasibility of semi-supervised learning in sports video analysis.

Provided a framework for automating judo match annotation.

Abstract

This paper presents a semi-supervised approach to extracting and analyzing combat phases in judo tournaments using live-streamed footage. The objective is to automate the annotation and summarization of live streamed judo matches. We train models that extract relevant entities and classify combat phases from fixed-perspective judo recordings. We employ semi-supervised methods to address limited labeled data in the domain. We build a model of combat phases via transfer learning from a fine-tuned object detector to classify the presence, activity, and standing state of the match. We evaluate our approach on a dataset of 19 thirty-second judo clips, achieving an F1 score on a $20\%$ test hold-out of 0.66, 0.78, and 0.87 for the three classes, respectively. Our results show initial promise for automating more complex information retrieval tasks using rigorous methods with limited labeled data.