Joint angle based learning to refine kinematic human pose estimation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Peng, Chang, Zhou, Yifei, Ren, Haoqiang, Huang, Shiqing, Chen, Chuangye, Yang, Jianming, Yang, Bao, Xi, Huifeng, Jiang, Zhenyu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914615027302400
author Peng, Chang
Zhou, Yifei
Ren, Haoqiang
Huang, Shiqing
Chen, Chuangye
Yang, Jianming
Yang, Bao
Xi, Huifeng
Jiang, Zhenyu
author_facet Peng, Chang
Zhou, Yifei
Ren, Haoqiang
Huang, Shiqing
Chen, Chuangye
Yang, Jianming
Yang, Bao
Xi, Huifeng
Jiang, Zhenyu
contents Marker-free human pose estimation (HPE) has found increasing applications in various fields. Current HPE suffers from occasional errors in keypoint recognition and random fluctuation in keypoint trajectories when analyzing kinematic human poses. The performance of existing deep learning-based models for HPE refinement is considerably limited by inaccurate training datasets in which the keypoints are manually annotated. This paper proposed a novel method to overcome the difficulty, in which the key techniques include: (i) A robust joint angle-based description of kinematic human poses; (ii) Approximating temporal variation of joint angles using high order Fourier series to get reliable "ground truth"; (iii) A bidirectional recurrent network is designed as a post-processing module to refine the estimation of single image-based HPE models. Trained with the high-quality dataset constructed using our method, the network demonstrates outstanding performance to correct wrongly recognized joints and smooth their spatiotemporal trajectories. Tests show that joint angle-based refinement (JAR) outperforms the state-of-the-art HPE refinement network in challenging cases like figure skating and breaking. JAR also demonstrates great potential to rectify existing datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2507_11075
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Joint angle based learning to refine kinematic human pose estimation
Peng, Chang
Zhou, Yifei
Ren, Haoqiang
Huang, Shiqing
Chen, Chuangye
Yang, Jianming
Yang, Bao
Xi, Huifeng
Jiang, Zhenyu
Computer Vision and Pattern Recognition
Artificial Intelligence
I.4.9; I.5.4; J.3
Marker-free human pose estimation (HPE) has found increasing applications in various fields. Current HPE suffers from occasional errors in keypoint recognition and random fluctuation in keypoint trajectories when analyzing kinematic human poses. The performance of existing deep learning-based models for HPE refinement is considerably limited by inaccurate training datasets in which the keypoints are manually annotated. This paper proposed a novel method to overcome the difficulty, in which the key techniques include: (i) A robust joint angle-based description of kinematic human poses; (ii) Approximating temporal variation of joint angles using high order Fourier series to get reliable "ground truth"; (iii) A bidirectional recurrent network is designed as a post-processing module to refine the estimation of single image-based HPE models. Trained with the high-quality dataset constructed using our method, the network demonstrates outstanding performance to correct wrongly recognized joints and smooth their spatiotemporal trajectories. Tests show that joint angle-based refinement (JAR) outperforms the state-of-the-art HPE refinement network in challenging cases like figure skating and breaking. JAR also demonstrates great potential to rectify existing datasets.
title Joint angle based learning to refine kinematic human pose estimation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
I.4.9; I.5.4; J.3
url https://arxiv.org/abs/2507.11075