Novel automated detection of sports activities using shadow videos

Barua, PD; Tuncer, T; Dogan, S; Ooi, CP; Acharya, RU

Novel automated detection of sports activities using shadow videos

Barua, PD Tuncer, T Dogan, S Ooi, CP Acharya, RU

Permalink

Publisher:: Springer
Publication Type:: Journal Article
Citation:: Multimedia Tools and Applications, 2024, 83, (15), pp. 44933-44954
Issue Date:: 2024-01-01

Closed Access

	Filename	Description	Size
	s11042-023-17407-1.pdf	Published version	1.61 MB	Adobe PDF	View/Open

Copyright Clearance Process

Recently Added
In Progress
Closed Access

This item is closed access and not available.

Full metadata record

Field	Value	Language
dc.contributor.author	Barua, PD
dc.contributor.author	Tuncer, T
dc.contributor.author	Dogan, S
dc.contributor.author	Ooi, CP
dc.contributor.author	Acharya, RU
dc.date.accessioned	2025-04-06T23:43:20Z
dc.date.available	2025-04-06T23:43:20Z
dc.date.issued	2024-01-01
dc.identifier.citation	Multimedia Tools and Applications, 2024, 83, (15), pp. 44933-44954
dc.identifier.issn	1380-7501
dc.identifier.issn	1573-7721
dc.identifier.uri	http://hdl.handle.net/10453/186680
dc.description.abstract	The aim of this work is to investigate the use of shadow videos for daily sports activity detection and to contribute to the emerging field of shadow-based classification in machine learning. A novel deep feature engineering model is proposed, and a new shadow video dataset is collected to validate the proposed model. Furthermore, the use of shadow videos ensures privacy protection for individuals. To evaluate the proposed system, a dataset comprising five sports activities (i.e., squat, steady run, standing butterfly, overhead side bend, and knee lift) recorded from 33 participants is used. The proposed model works in the following way: (i) videos are divided into frames and aggregated into non-overlapping blocks of six frames to create images, (ii) deep features are extracted from three fully connected layers of the pre-trained AlexNet, resulting in 4096, 4096, and 1000 features from fc6, fc7, and fc8 layers, respectively. These three feature vectors are then merged to generate a final feature vector with a length of 9192 (= 4096 + 4096 + 1000). (iii) The Chi2 selector is employed to select the most distinctive 1000 features in the feature selection phase, and (iv) the support vector machine (SVM) with leave-one-subject-out (LOSO) validation is used to classify the five sports activities. The proposed deep features coupled with Chi2 based model achieved a classification accuracy of 88.49% using the SVM classifier with LOSO cross-validation (CV) on our collected dataset.
dc.language	en
dc.publisher	Springer
dc.relation.ispartof	Multimedia Tools and Applications
dc.relation.isbasedon	10.1007/s11042-023-17407-1
dc.rights	info:eu-repo/semantics/closedAccess
dc.subject	0801 Artificial Intelligence and Image Processing, 0803 Computer Software, 0805 Distributed Computing, 0806 Information Systems
dc.subject.classification	Artificial Intelligence & Image Processing
dc.subject.classification	Software Engineering
dc.subject.classification	4009 Electronics, sensors and digital hardware
dc.subject.classification	4603 Computer vision and multimedia computation
dc.subject.classification	4605 Data management and data science
dc.subject.classification	4606 Distributed computing and systems software
dc.title	Novel automated detection of sports activities using shadow videos
dc.type	Journal Article
utslib.citation.volume	83
utslib.for	0801 Artificial Intelligence and Image Processing
utslib.for	0803 Computer Software
utslib.for	0805 Distributed Computing
utslib.for	0806 Information Systems
pubs.organisational-group	University of Technology Sydney
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology/School of Biomedical Engineering
utslib.copyright.status	closed_access	*
pubs.consider-herdc	false
dc.date.updated	2025-04-06T23:43:19Z
pubs.issue	15
pubs.publication-status	Published
pubs.volume	83
utslib.citation.issue	15

Abstract:

The aim of this work is to investigate the use of shadow videos for daily sports activity detection and to contribute to the emerging field of shadow-based classification in machine learning. A novel deep feature engineering model is proposed, and a new shadow video dataset is collected to validate the proposed model. Furthermore, the use of shadow videos ensures privacy protection for individuals. To evaluate the proposed system, a dataset comprising five sports activities (i.e., squat, steady run, standing butterfly, overhead side bend, and knee lift) recorded from 33 participants is used. The proposed model works in the following way: (i) videos are divided into frames and aggregated into non-overlapping blocks of six frames to create images, (ii) deep features are extracted from three fully connected layers of the pre-trained AlexNet, resulting in 4096, 4096, and 1000 features from fc6, fc7, and fc8 layers, respectively. These three feature vectors are then merged to generate a final feature vector with a length of 9192 (= 4096 + 4096 + 1000). (iii) The Chi2 selector is employed to select the most distinctive 1000 features in the feature selection phase, and (iv) the support vector machine (SVM) with leave-one-subject-out (LOSO) validation is used to classify the five sports activities. The proposed deep features coupled with Chi2 based model achieved a classification accuracy of 88.49% using the SVM classifier with LOSO cross-validation (CV) on our collected dataset.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/186680