Title: A semi-automatic system of web videos annotation and retrieval: application to events detection in soccer domain

Authors: Lamia Fatiha Kazi Tani; Abdelghani Ghomari; Mohammed Yassine Kazi Tani

Addresses: RIIR Laboratory, Computer Science Department, Faculty of Exact and Applied Sciences, University Oran1 Ahmed Ben Bella, Oran, Algeria ' RIIR Laboratory, Computer Science Department, Faculty of Exact and Applied Sciences, University Oran1 Ahmed Ben Bella, Oran, Algeria ' LabRI – SBA Laboratory, Ecole Supérieure en Informatique, Sidi Bel Abbes, Algeria

Abstract: Annotations and retrieval of soccer videos on the web is a challenging task that concerns human lives including sports. In this paper, we propose a novel approach based on deep learning and ontology formalism to detect objects and to recognise events in soccer videos. To overcome the semantic gap between low and high level semantic annotation of videos, we use a deep neural network to extract low-level features through a complete method called mask R-CNN based ResNet-101 architecture as a backbone. Then, we create and populate soccer ontology in accordance to the output predictions of the mask R-CNN. We then create a smart system able to learn how to detect objects and to infer events in soccer videos. To validate our approach, we experimented on 40 soccer videos of FIFA World Cup 2018 downloaded from YouTube and we compare the obtained results with those of the state of the art.

Keywords: soccer video annotation; event recognition; convolutional neural network; CNN; mask R-CNN; ontology; SWRL.

DOI: 10.1504/IJCAET.2022.10026118

International Journal of Computer Aided Engineering and Technology, 2022 Vol.16 No.4, pp.512 - 533

Received: 08 Nov 2019
Accepted: 29 Nov 2019

Published online: 06 Jul 2022 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article