Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model

Yajuan Wei; Chuan Dai; Zhijie Xu; Minsi Chen; Ying Liu

doi:10.1007/978-3-031-49421-5_4

Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model

Yajuan Wei, Chuan Dai, Zhijie Xu, Minsi Chen, Ying Liu

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Human pose estimation (HPE) is the basis of a wide variety of computer vision tasks. However, existing approaches are designed mainly for static 2D images, ignoring the temporal continuity and geometric consistency between video frames. With the aim of resolving the aforementioned issues, Pose-DWT-Former, an enhanced transformer-based approach, was proposed in this paper. The principle is to employ the 2D skeleton-based pose sequences extracted from the video frames and the discrete wavelet domain (DWT) information of those sequences as the network input for the purposes of 3D HPE. The evaluations were performed on two datasets, with good results in speed, accuracy, and robustness against noise, laying a solid foundation for the subsequent use of 3D human posture information for action recognition.

Original language	English
Title of host publication	Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2
Editors	Andrew D. Ball, Huajiang Ouyang, Jyoti K. Sinha, Zuolu Wang
Publisher	Springer, Cham
Pages	41-49
Number of pages	9
Volume	152
ISBN (Electronic)	9783031494215
ISBN (Print)	9783031494208, 9783031494239
DOIs	https://doi.org/10.1007/978-3-031-49421-5_4
Publication status	Published - 29 May 2024
Event	The UNIfied Conference of DAMAS, InCoME and TEPEN Conferences - Huddersfield, United Kingdom, Huddersfield, United Kingdom Duration: 29 Aug 2023 → 1 Sep 2023 https://unified2023.org/

Publication series

Name	Mechanisms and Machine Science
Publisher	Springer
Volume	152 MMS
ISSN (Print)	2211-0984
ISSN (Electronic)	2211-0992

Conference

Conference	The UNIfied Conference of DAMAS, InCoME and TEPEN Conferences
Abbreviated title	UNIfied 2023
Country/Territory	United Kingdom
City	Huddersfield
Period	29/08/23 → 1/09/23
Internet address	https://unified2023.org/

Access to Document

10.1007/978-3-031-49421-5_4Licence: Unspecified

Cite this

Wei, Y., Dai, C., Xu, Z., Chen, M., & Liu, Y. (2024). Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model. In A. D. Ball, H. Ouyang, J. K. Sinha, & Z. Wang (Eds.), Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2 (Vol. 152, pp. 41-49). (Mechanisms and Machine Science; Vol. 152 MMS). Springer, Cham. https://doi.org/10.1007/978-3-031-49421-5_4

Wei, Yajuan ; Dai, Chuan ; Xu, Zhijie et al. / Pose-DWT-Former : An Improved Transformer-Based 3D Human Pose Estimation Model. Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2. editor / Andrew D. Ball ; Huajiang Ouyang ; Jyoti K. Sinha ; Zuolu Wang. Vol. 152 Springer, Cham, 2024. pp. 41-49 (Mechanisms and Machine Science).

@inproceedings{0f1e7aa56db44b8386835a27615fe8ca,

title = "Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model",

abstract = "Human pose estimation (HPE) is the basis of a wide variety of computer vision tasks. However, existing approaches are designed mainly for static 2D images, ignoring the temporal continuity and geometric consistency between video frames. With the aim of resolving the aforementioned issues, Pose-DWT-Former, an enhanced transformer-based approach, was proposed in this paper. The principle is to employ the 2D skeleton-based pose sequences extracted from the video frames and the discrete wavelet domain (DWT) information of those sequences as the network input for the purposes of 3D HPE. The evaluations were performed on two datasets, with good results in speed, accuracy, and robustness against noise, laying a solid foundation for the subsequent use of 3D human posture information for action recognition.",

keywords = "DWT, HPE, Skeleton pose sequences, Transformer",

author = "Yajuan Wei and Chuan Dai and Zhijie Xu and Minsi Chen and Ying Liu",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.; The UNIfied Conference of DAMAS, InCoME and TEPEN Conferences, UNIfied 2023 ; Conference date: 29-08-2023 Through 01-09-2023",

year = "2024",

month = may,

day = "29",

doi = "10.1007/978-3-031-49421-5_4",

language = "English",

isbn = "9783031494208",

volume = "152",

series = "Mechanisms and Machine Science",

publisher = "Springer, Cham",

pages = "41--49",

editor = "Ball, {Andrew D.} and Huajiang Ouyang and Sinha, {Jyoti K.} and Zuolu Wang",

booktitle = "Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2",

address = "Switzerland",

url = "https://unified2023.org/",

}

Wei, Y, Dai, C, Xu, Z, Chen, M & Liu, Y 2024, Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model. in AD Ball, H Ouyang, JK Sinha & Z Wang (eds), Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2. vol. 152, Mechanisms and Machine Science, vol. 152 MMS, Springer, Cham, pp. 41-49, The UNIfied Conference of DAMAS, InCoME and TEPEN Conferences, Huddersfield, United Kingdom, 29/08/23. https://doi.org/10.1007/978-3-031-49421-5_4

Pose-DWT-Former : An Improved Transformer-Based 3D Human Pose Estimation Model. / Wei, Yajuan; Dai, Chuan; Xu, Zhijie et al.

Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2. ed. / Andrew D. Ball; Huajiang Ouyang; Jyoti K. Sinha; Zuolu Wang. Vol. 152 Springer, Cham, 2024. p. 41-49 (Mechanisms and Machine Science; Vol. 152 MMS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Pose-DWT-Former

T2 - The UNIfied Conference of DAMAS, InCoME and TEPEN Conferences

AU - Wei, Yajuan

AU - Dai, Chuan

AU - Xu, Zhijie

AU - Chen, Minsi

AU - Liu, Ying

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

PY - 2024/5/29

Y1 - 2024/5/29

N2 - Human pose estimation (HPE) is the basis of a wide variety of computer vision tasks. However, existing approaches are designed mainly for static 2D images, ignoring the temporal continuity and geometric consistency between video frames. With the aim of resolving the aforementioned issues, Pose-DWT-Former, an enhanced transformer-based approach, was proposed in this paper. The principle is to employ the 2D skeleton-based pose sequences extracted from the video frames and the discrete wavelet domain (DWT) information of those sequences as the network input for the purposes of 3D HPE. The evaluations were performed on two datasets, with good results in speed, accuracy, and robustness against noise, laying a solid foundation for the subsequent use of 3D human posture information for action recognition.

AB - Human pose estimation (HPE) is the basis of a wide variety of computer vision tasks. However, existing approaches are designed mainly for static 2D images, ignoring the temporal continuity and geometric consistency between video frames. With the aim of resolving the aforementioned issues, Pose-DWT-Former, an enhanced transformer-based approach, was proposed in this paper. The principle is to employ the 2D skeleton-based pose sequences extracted from the video frames and the discrete wavelet domain (DWT) information of those sequences as the network input for the purposes of 3D HPE. The evaluations were performed on two datasets, with good results in speed, accuracy, and robustness against noise, laying a solid foundation for the subsequent use of 3D human posture information for action recognition.

KW - DWT

KW - HPE

KW - Skeleton pose sequences

KW - Transformer

UR - http://www.scopus.com/inward/record.url?scp=85195558791&partnerID=8YFLogxK

UR - https://doi.org/10.1007/978-3-031-49421-5

U2 - 10.1007/978-3-031-49421-5_4

DO - 10.1007/978-3-031-49421-5_4

M3 - Conference contribution

AN - SCOPUS:85195558791

SN - 9783031494208

SN - 9783031494239

VL - 152

T3 - Mechanisms and Machine Science

SP - 41

EP - 49

BT - Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2

A2 - Ball, Andrew D.

A2 - Ouyang, Huajiang

A2 - Sinha, Jyoti K.

A2 - Wang, Zuolu

PB - Springer, Cham

Y2 - 29 August 2023 through 1 September 2023

ER -

Wei Y, Dai C, Xu Z, Chen M, Liu Y. Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model. In Ball AD, Ouyang H, Sinha JK, Wang Z, editors, Proceedings of the UNIfied Conference of DAMAS, IncoME and TEPEN Conferences (UNIfied 2023) - Volume 2. Vol. 152. Springer, Cham. 2024. p. 41-49. (Mechanisms and Machine Science). Epub 2024 May 28. doi: 10.1007/978-3-031-49421-5_4

Pose-DWT-Former: An Improved Transformer-Based 3D Human Pose Estimation Model

Abstract

Publication series

Conference

Access to Document

Other files and links

Cite this