Human Action Recognition Based on 2D Poses and Skeleton Joints
Autor(a) principal: | |
---|---|
Data de Publicação: | 2022 |
Outros Autores: | |
Tipo de documento: | Artigo de conferência |
Idioma: | eng |
Título da fonte: | Repositório Institucional da UNESP |
Texto Completo: | http://dx.doi.org/10.1007/978-3-031-21689-3_6 http://hdl.handle.net/11449/249520 |
Resumo: | With the growing capacity of current technologies to store and process data at extremely fast speed, video processing and its diverse applications have been studied and several researches have emerged using videos as objects of study. One of them is human action recognition, which seeks to identify in a given video what actions the people present in it are performing, whether for recreational purposes or for monitoring public places for people’s safety. Being able to extract the necessary characteristics to perform the action classification is a complex task, because, unlike traditional image classification problems, the recognition of human actions requires that the solution works with spatio-temporal characteristics, which represent a pattern of movements performed by the person in both the spatial and temporal aspects along the frames that make up the action. One way to generate information that describes human movement is to identify the skeleton joints and limbs and use them to perform the classification, and the extraction of this data can be done by using 2D pose estimation algorithms, capable of tracking parts of the human body and returning their coordinates. This paper proposes a method for human action recognition that uses skeleton joints and limb information as features and a CNN together with an LSTM network for classification. The method was assessed on a public dataset (Weizmann dataset) and obtained superior recognition rates when compared to other action recognition methods of literature. |
id |
UNSP_3f95434764d29863e5607d79ac7aaba6 |
---|---|
oai_identifier_str |
oai:repositorio.unesp.br:11449/249520 |
network_acronym_str |
UNSP |
network_name_str |
Repositório Institucional da UNESP |
repository_id_str |
2946 |
spelling |
Human Action Recognition Based on 2D Poses and Skeleton Joints2D poseCNNHuman action recognitionHuman pose estimationLSTMWith the growing capacity of current technologies to store and process data at extremely fast speed, video processing and its diverse applications have been studied and several researches have emerged using videos as objects of study. One of them is human action recognition, which seeks to identify in a given video what actions the people present in it are performing, whether for recreational purposes or for monitoring public places for people’s safety. Being able to extract the necessary characteristics to perform the action classification is a complex task, because, unlike traditional image classification problems, the recognition of human actions requires that the solution works with spatio-temporal characteristics, which represent a pattern of movements performed by the person in both the spatial and temporal aspects along the frames that make up the action. One way to generate information that describes human movement is to identify the skeleton joints and limbs and use them to perform the classification, and the extraction of this data can be done by using 2D pose estimation algorithms, capable of tracking parts of the human body and returning their coordinates. This paper proposes a method for human action recognition that uses skeleton joints and limb information as features and a CNN together with an LSTM network for classification. The method was assessed on a public dataset (Weizmann dataset) and obtained superior recognition rates when compared to other action recognition methods of literature.Faculty of Sciences UNESP - São Paulo State University, SPFaculty of Sciences UNESP - São Paulo State University, SPUniversidade Estadual Paulista (UNESP)Belluzzo, Bruno [UNESP]Marana, Aparecido Nilceu [UNESP]2023-07-29T16:01:59Z2023-07-29T16:01:59Z2022-01-01info:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/conferenceObject71-83http://dx.doi.org/10.1007/978-3-031-21689-3_6Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), v. 13654 LNAI, p. 71-83.1611-33490302-9743http://hdl.handle.net/11449/24952010.1007/978-3-031-21689-3_62-s2.0-85145251348Scopusreponame:Repositório Institucional da UNESPinstname:Universidade Estadual Paulista (UNESP)instacron:UNESPengLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)info:eu-repo/semantics/openAccess2024-04-23T16:11:26Zoai:repositorio.unesp.br:11449/249520Repositório InstitucionalPUBhttp://repositorio.unesp.br/oai/requestopendoar:29462024-08-05T19:26:38.217490Repositório Institucional da UNESP - Universidade Estadual Paulista (UNESP)false |
dc.title.none.fl_str_mv |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
title |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
spellingShingle |
Human Action Recognition Based on 2D Poses and Skeleton Joints Belluzzo, Bruno [UNESP] 2D pose CNN Human action recognition Human pose estimation LSTM |
title_short |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
title_full |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
title_fullStr |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
title_full_unstemmed |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
title_sort |
Human Action Recognition Based on 2D Poses and Skeleton Joints |
author |
Belluzzo, Bruno [UNESP] |
author_facet |
Belluzzo, Bruno [UNESP] Marana, Aparecido Nilceu [UNESP] |
author_role |
author |
author2 |
Marana, Aparecido Nilceu [UNESP] |
author2_role |
author |
dc.contributor.none.fl_str_mv |
Universidade Estadual Paulista (UNESP) |
dc.contributor.author.fl_str_mv |
Belluzzo, Bruno [UNESP] Marana, Aparecido Nilceu [UNESP] |
dc.subject.por.fl_str_mv |
2D pose CNN Human action recognition Human pose estimation LSTM |
topic |
2D pose CNN Human action recognition Human pose estimation LSTM |
description |
With the growing capacity of current technologies to store and process data at extremely fast speed, video processing and its diverse applications have been studied and several researches have emerged using videos as objects of study. One of them is human action recognition, which seeks to identify in a given video what actions the people present in it are performing, whether for recreational purposes or for monitoring public places for people’s safety. Being able to extract the necessary characteristics to perform the action classification is a complex task, because, unlike traditional image classification problems, the recognition of human actions requires that the solution works with spatio-temporal characteristics, which represent a pattern of movements performed by the person in both the spatial and temporal aspects along the frames that make up the action. One way to generate information that describes human movement is to identify the skeleton joints and limbs and use them to perform the classification, and the extraction of this data can be done by using 2D pose estimation algorithms, capable of tracking parts of the human body and returning their coordinates. This paper proposes a method for human action recognition that uses skeleton joints and limb information as features and a CNN together with an LSTM network for classification. The method was assessed on a public dataset (Weizmann dataset) and obtained superior recognition rates when compared to other action recognition methods of literature. |
publishDate |
2022 |
dc.date.none.fl_str_mv |
2022-01-01 2023-07-29T16:01:59Z 2023-07-29T16:01:59Z |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/conferenceObject |
format |
conferenceObject |
status_str |
publishedVersion |
dc.identifier.uri.fl_str_mv |
http://dx.doi.org/10.1007/978-3-031-21689-3_6 Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), v. 13654 LNAI, p. 71-83. 1611-3349 0302-9743 http://hdl.handle.net/11449/249520 10.1007/978-3-031-21689-3_6 2-s2.0-85145251348 |
url |
http://dx.doi.org/10.1007/978-3-031-21689-3_6 http://hdl.handle.net/11449/249520 |
identifier_str_mv |
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), v. 13654 LNAI, p. 71-83. 1611-3349 0302-9743 10.1007/978-3-031-21689-3_6 2-s2.0-85145251348 |
dc.language.iso.fl_str_mv |
eng |
language |
eng |
dc.relation.none.fl_str_mv |
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/openAccess |
eu_rights_str_mv |
openAccess |
dc.format.none.fl_str_mv |
71-83 |
dc.source.none.fl_str_mv |
Scopus reponame:Repositório Institucional da UNESP instname:Universidade Estadual Paulista (UNESP) instacron:UNESP |
instname_str |
Universidade Estadual Paulista (UNESP) |
instacron_str |
UNESP |
institution |
UNESP |
reponame_str |
Repositório Institucional da UNESP |
collection |
Repositório Institucional da UNESP |
repository.name.fl_str_mv |
Repositório Institucional da UNESP - Universidade Estadual Paulista (UNESP) |
repository.mail.fl_str_mv |
|
_version_ |
1808129069859995648 |