CASA: Cricket Action Similarity Assessment in Video Footage Using Deep Metric Learning

Tevin Moodley, Dustin van der Haar

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Cricket batters will often measure their performance through comparisons against successful batters or feedback provided by experts. Action Similarity Assessment is the task of comparing the similarity or dissimilarity of an action between two actors to determine how similar the actions they perform are to one another. This research paper proposes the use of a Siamese Convolution Neural Network to compute the similarity distances between different batters using video footage. Due to the limited research surrounding action similarity, a new dataset is proposed to help foster future works pertaining to action similarity. Three architectures are proposed to determine which architecture is best suited for the domain: a custom CNN, Inception Resnet V2, and Xception. From the results obtained, it can be concluded that the best solution for the action similarity assessment task within cricket video footage is a Siamese Xception architecture, achieving a model accuracy of 98%.

Original languageEnglish
Title of host publicationArtificial Intelligence Research - Third Southern African Conference, SACAIR 2022, Proceedings
EditorsAnban Pillay, Edgar Jembere, Aurona Gerber
PublisherSpringer Science and Business Media Deutschland GmbH
Pages139-153
Number of pages15
ISBN (Print)9783031223204
DOIs
Publication statusPublished - 2022
Event3rd Southern African Conference on Artificial Intelligence Research, SACAIR 2022 - Stellenbosch, South Africa
Duration: 5 Dec 20229 Dec 2022

Publication series

NameCommunications in Computer and Information Science
Volume1734 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference3rd Southern African Conference on Artificial Intelligence Research, SACAIR 2022
Country/TerritorySouth Africa
CityStellenbosch
Period5/12/229/12/22

Keywords

  • Cricket action similarity
  • Inception Resnet V2
  • Siamese network
  • Xception

ASJC Scopus subject areas

  • General Computer Science
  • General Mathematics

Fingerprint

Dive into the research topics of 'CASA: Cricket Action Similarity Assessment in Video Footage Using Deep Metric Learning'. Together they form a unique fingerprint.

Cite this