A Strong Geometric Baseline for Cross-View Matching of Multi-person 3D Pose Estimation from Multi-view Images

被引：3

作者：

Dehaeck, Sam ^{[1
]}

Domken, Corentin ^{[1
]}

Bey-Temsamani, Abdellatif ^{[1
]}

Abedrabbo, Gabriel ^{[1
]}

机构：

[1] Flanders Make, Decis Prod CodesignS, Oude Diestersebaan 133, B-3920 Lommel, Belgium

来源：

IMAGE ANALYSIS AND PROCESSING, ICIAP 2022, PT II | 2022年 / 13232卷

关键词：

Pose estimation; Multi-person 3d pose; Multi-view geometry; Cross-view matching;

D O I：

10.1007/978-3-031-06430-2_7

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Reconstructing the 3D pose from multiple people viewed from several cameras is a complex problem. After the 2D human pose estimation is performed in each image separately (e.g. by a deep learning algorithm), the next step is matching the individuals on the different images. In the present contribution, a novel non-deep learning and parameter-free algorithm is developed to this end, which is only based on geometrical cues (i.e. reprojection errors). This is shown to achieve perfect precision and recall figures on a challenging 102-person benchmark.

引用

页码：77 / 88

页数：12

共 13 条

[1] 3D Pictorial Structures Revisited: Multiple Human Pose Estimation [J].

Belagiannis, Vasileios ;

Amin, Sikandar ;

Andriluka, Mykhaylo ;

Schiele, Bernt ;

Navab, Nassir ;

Ilic, Slobodan .

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2016, 38 (10) :1929-1942

[2]

Bridgeman Lewis, 2019, 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Proceedings, P2487, DOI 10.1109/CVPRW.2019.00304

[3] 3D Pictorial Structures for Multiple View Articulated Pose Estimation [J].

Burenius, Magnus ;

Sullivan, Josephine ;

Carlsson, Stefan .

2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2013, :3618-3625

[4] Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields [J].

Cao, Zhe ;

Simon, Tomas ;

Wei, Shih-En ;

Sheikh, Yaser .

30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017), 2017, :1302-1310

[5]

Cheng BW, 2020, Arxiv, DOI [arXiv:1908.10357, 10.48550/ARXIV.1908.10357]

[6] Fast and Robust Multi-Person 3D Pose Estimation from Multiple Views [J].

Dong, Junting ;

Jiang, Wen ;

Huang, Qixing ;

Bao, Hujun ;

Zhou, Xiaowei .

2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2019), 2019, :7784-7793

[7]

Elmi A, 2020, Arxiv, DOI [arXiv:2004.02688, 10.1109/icpr48806.2021.9412652, DOI 10.1109/ICPR48806.2021.9412652]

[8]

Hartley R., 2004, Multiple view geometry in computer vision, DOI [10.1017/CBO9780511811685, DOI 10.1017/CBO9780511811685]

[9] TesseTrack: End-to-End Learnable Multi-Person Articulated 3D Pose Tracking [J].

Reddy, N. Dinesh ;

Guigues, Laurent ;

Pishchulin, Leonid ;

Eledath, Jayan ;

Narasimhan, Srinivasa G. .

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021, 2021, :15185-15195

[10]

Tanke J., 2019, GERMAN C PATTERN REC, DOI [10.1007/978-3-030-33676-938, DOI 10.1007/978-3-030-33676-938]

← 1 2 →