基于渐进高斯滤波融合的多视角人体姿态估计

杨旭升; 吴江宇; 胡佛; 张文安

doi:10.16383/j.aas.c230316

基于渐进高斯滤波融合的多视角人体姿态估计

doi: 10.16383/j.aas.c230316

杨旭升^{1, 2,},
吴江宇^{1, 2,},
胡佛^{1, 2,},
张文安^{1, 2,}

1.
浙江工业大学信息工程学院杭州 310023
2.
浙江省嵌入式系统联合重点实验室杭州 310023

基金项目: 浙江省“尖兵”“领雁”研发攻关计划(2022C03114), 浙江省自然科学基金(LY23F030006)资助

详细信息

作者简介:
杨旭升：浙江工业大学信息工程学院副教授. 主要研究方向为信息融合估计, 人体姿态估计和目标定位. 本文通信作者. E-mail: xsyang@zjut.edu.cn

吴江宇：浙江工业大学信息工程学院硕士研究生. 主要研究方向为人体姿态估计和信息融合估计. E-mail: wujiangyu@zjut.edu.cn

胡佛：浙江工业大学信息工程学院助理研究员. 主要研究方向为人机交互, 情感计算和人工智能. E-mail: fohu@zjut.edu.cn

张文安：浙江工业大学信息工程学院教授. 主要研究方向为多源信息融合估计及应用. E-mail: wazhang@zjut.edu.cn

计量
- 文章访问数: 198
- HTML全文浏览量: 65
- PDF下载量: 85
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-29
- 录用日期: 2023-11-03
- 网络出版日期: 2024-02-21
- 刊出日期: 2024-03-29

Multi-view Human Pose Estimation Based on Progressive Gaussian Filtering Fusion

YANG Xu-Sheng^{1, 2
,},
WU Jiang-Yu^{1, 2
,},
HU Fo^{1, 2
,},
ZHANG Wen-An^{1, 2
,}

1.
College of Information Engineering, Zhejiang University of Technology, Hangzhou 310023
2.
Zhejiang Provincial United Key Laboratory of Embedded Systems, Hangzhou 310023

Funds: Supported by Zhejiang Province “Pioneer” and “Leading Goose” Research and Development Project (2022C03114) and Natural Science Foundation of Zhejiang Province (LY23F030006)

More Information

Author Bio:
YANG Xu-Sheng　Associate professor at the College of Information Engineering, Zhejiang University of Technology. His research interest covers information fusion estimation, human pose estimation, and target positioning. Corresponding author of this paper

WU Jiang-Yu　Master student at the College of Information Engineering, Zhejiang University of Technology. His research interest covers human pose estimation and information fusion estimation

HU Fo　Assistant researcher at the College of Information Engineering, Zhejiang University of Technology. His research interest covers human machine interaction, emotional computing, and artificial intelligence

ZHANG Wen-An　Professor at the College of Information Engineering, Zhejiang University of Technology. His research interest covers multi-sensor information fusion estimation and its applications

摘要

摘要: 针对视觉遮挡引起的人体姿态估计(Human pose estimation, HPE)性能下降问题, 提出基于渐进高斯滤波(Progressive Gaussian filtering, PGF)融合的人体姿态估计方法. 首先, 设计分层性能评估方法对多视觉量测进行分类处理, 以适应视觉遮挡引起的量测不确定性问题. 其次, 构建分布式渐进贝叶斯滤波融合框架, 以及设计一种分层分类融合估计方法来提升复杂量测融合的鲁棒性和准确性. 特别地, 针对量测统计特性变化问题, 利用局部估计间的交互信息来引导渐进量测更新, 从而隐式地补偿量测不确定性. 最后, 仿真与实验结果表明, 相比于现有的方法, 所提的人体姿态估计方法具有更高的准确性和鲁棒性.
- 渐进高斯滤波 /
- 自适应滤波 /
- 分布式融合 /
- 人体姿态估计
Abstract: A human pose estimation (HPE) method based on progressive Gaussian filtering (PGF) fusion is proposed to address the performance degradation issue caused by visual occlusion. Firstly, a hierarchical performance evaluation method is designed to classify and handle multiple visual measurements, in order to adapt to the uncertainty problem caused by visual occlusion. Secondly, a distributed progressive Bayesian filtering fusion framework is constructed, and a hierarchical classification fusion estimation method is designed to improve the robustness and accuracy of complex measurement fusion. Specifically, to address the issue of measurement statistical characteristic variation, the interactive information among local estimators is utilized to guide the progressive measurement update, thereby implicitly compensating for measurement uncertainty. Finally, from simulation and experimental results, it demonstrates that compared with existing methods, the proposed human pose estimation method achieves higher accuracy and robustness.
- Progressive Gaussian filtering (PGF) /
- adaptive filtering /
- distributed fusion /
- human pose estimation (HPE)

HTML全文

图 1 多视觉人体姿态估计示意图

Fig. 1 Schematic diagram of multi-vision human pose estimation

下载: 全尺寸图片幻灯片

图 2 量测相容性分析

Fig. 2 Measurement compatibility analysis

下载: 全尺寸图片幻灯片

图 3 方法框图

Fig. 3 Method block diagram

下载: 全尺寸图片幻灯片

图 4 不同滤波融合方法下的位置误差

Fig. 4 Position error under different filtering fusion methods

下载: 全尺寸图片幻灯片

图 5 人体姿态估计实验平台

Fig. 5 Human pose estimation experimental platform

下载: 全尺寸图片幻灯片

图 6 不同滤波融合方法下的累积位置误差

Fig. 6 Cumulative position error under different filtering fusion methods

下载: 全尺寸图片幻灯片

表 1 累积误差均值统计(mm)

Table 1 Cumulative error mean statistics (mm)

实验方法腕关节肘关节肩关节

观测融合 166.44 124.44 96.56
CF 157.55 118.00 95.00
AMFKF 147.81 113.85 93.08
CI 127.63 117.85 99.62
IWCF 153.12 113.21 92.53
PGFFwoC 151.77 114.12 92.83
PGFFwC 119.47 108.98 84.11

下载: 导出CSV

参考文献(28)

[1]	Desmarais Y, Mottet D, Slangen P, Montesinos P. A review of 3D human pose estimation algorithms for markerless motion capture. Computer Vision and Image Understanding, 2021, 212: Article No. 103275 doi: 10.1016/j.cviu.2021.103275
[2]	Yang X, Yin S, Zhang W A, Hu F, Yu L, Asynchronous Gaussian filtering fusion for human motion estimation based on RGB-D cameras. IEEE Sensors Journal, 2023, 23(22): 28044-28054 doi: 10.1109/JSEN.2023.3323869
[3]	杜惠斌, 赵忆文, 韩建达, 赵新刚, 王争, 宋国立. 基于集员滤波的双Kinect人体关节点数据融合. 自动化学报, 2016, 42(12): 1886-1898 Du Hui-bin, Zhao Yi-wen, Han Jian-da, Zhao Xin-Gang, Wang Zheng, Song Guo-Li. Data fusion of dual Kinect human body joints based on ensemble filtering. Acta Automatica Sinica, 2022, 42(9): 2830-2837
[4]	Wang J, Tan S, Zhen X, Xu S, Zheng F, He Z, et al. Deep 3D human pose estimation: a review. Computer Vision and Image Understanding, 2021, 210: Article No. 103225 doi: 10.1016/j.cviu.2021.103225
[5]	蔡兴泉, 霍宇晴, 李发建, 孙海燕. 面向太极拳学习的人体姿态估计及相似度计算. 图学学报, 2022, 43(4): 695-706 Cai Xing-Quan, Huo Yu-Qing, Li Fa-Jian, Sun Hai-Yan. Human posture estimation and similarity calculation for Taijiquan learning. Journal of Graphics, 2022, 43(4): 695-706
[6]	张鋆豪, 何百岳, 杨旭升, 张文安. 基于可穿戴式惯性传感器的人体运动跟踪方法综述. 自动化学报, 2019, 45(8): 1439-1454 Zhang Jun-Hao, He Bai-Yue, Yang Xu-Sheng, Zhang Wen-An. A review of human motion tracking methods based on wearable inertial sensors. Acta Automatica Sinica, 2019, 45(8): 1439-1454
[7]	Casalino A, Guzman S, Maria Zanchettin A, Rocco P. Human pose estimation in presence of occlusion using depth camera sensors, in human-robot coexistence scenarios. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems. Madrid, Spain: IEEE, 2018. 1−7
[8]	Moon S, Park Y, Ko D W, Suh I H. Multiple Kinect sensor fusion for human skeleton tracking using Kalman filtering. International Journal of Advanced Robotic Systems, 2016, 13(2): 315-323
[9]	Liu G, Tian G, Li J, Zhu X, Wang Z. Human action recognition using a distributed RGB-depth camera network. IEEE Sensors Journal, 2018, 18(18): 7570-7576 doi: 10.1109/JSEN.2018.2859268
[10]	He H, Liu G, Zhu X, He L, Tian G. Interacting multiple model-based human pose estimation using a distributed 3D camera network. IEEE Sensors Journal, 2019, 19(22): 10584-10590 doi: 10.1109/JSEN.2019.2931603
[11]	Ahmed F, Hossain Bari A S M, Sieu B, Sadeghi J, Scholten J, Gavrilova M L. Kalman filter-based noise reduction framework for posture estimation using depth sensor. In: Proceedings of the 18th IEEE International Conference on Cognitive Informatics and Cognitive Computing. Milan, Italy: IEEE, 2019. 150−158
[12]	Yeung K Y, Kwok T H, Wang C C L. Improved skeleton tracking by duplex Kinects: a practical approach for real-time applications. Journal of Computing and Information Science in Engineering, 2013, 13(4): Article No. 041007
[13]	Olfati-Saber R, Fax J A, Murray R M. Consensus and cooperation in networked multi-agent systems. Proceedings of the IEEE, 2007, 95(1): 215-233 doi: 10.1109/JPROC.2006.887293
[14]	Kalman R E, Bucy R S. New results in linear filtering and prediction theory. Journal of Basic Engineering, 1961, 83(1): 95-108 doi: 10.1115/1.3658902
[15]	杨旭升, 张文安, 俞立. 适用于事件触发的分布式随机目标跟踪方法. 自动化学报, 2017, 43(8): 1393-1401 Yang Xu-Sheng, Zhang Wen-An, Yu Li. Distributed random target tracking method suitable for event triggering. Acta Automatica Sinica, 2017, 43(8): 1393-1401
[16]	Yang X, Zhang W A, Chen M Z Q, Yu L. Hybrid sequential fusion estimation for asynchronous sensor network-based target tracking. IEEE Transactions on Control Systems Technology, 2017, 25(2): 669-676 doi: 10.1109/TCST.2016.2558632
[17]	Yang X, Zhang W A, Yu L, Yang F. Sequential Gaussian approximation filter for target tracking with nonsynchronous measurements. IEEE Transactions on Aerospace and Electronic Systems, 2019, 55(1): 407-418 doi: 10.1109/TAES.2018.2852398
[18]	Coskun H, Achilles F, DiPietro R, Navab N, Tombari F. Long short-term memory Kalman filters: Recurrent neural estimators for pose regularization. In: Proceedings of the IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017. 5525−5533
[19]	Pathirana P N, Li S, Trinh H M, Seneviratne A. Robust real-time bio-kinematic movement tracking using multiple Kinects for tele-rehabilitation. IEEE Transactions on Industrial Electronics, 2016, 63(3): 1822-1833 doi: 10.1109/TIE.2015.2497662
[20]	贾晓凌, 张文安, 杨旭升. 视觉遮挡下的人体姿态鲁棒估计. 高技术通讯, 2021, 31(11): 1210-1218 doi: 10.3772/j.issn.1002-0470.2021.11.011 Jia Xiao-Ling, Zhang Wen-An, Yang Xu-Sheng. Robust estimation of human posture under visual occlusion. High Technology Communications, 2021, 31(11): 1210-1218 doi: 10.3772/j.issn.1002-0470.2021.11.011
[21]	郑婷婷, 杨旭升, 张文安, 俞立. 一种高斯渐进滤波框架下的目标跟踪方法. 自动化学报, 2018, 44(12): 2250-2258 Zheng Ting-Ting, Yang Xu-Sheng, Zhang Wen-An, Yu Li. A target tracking method in Gaussian progressive filtering framework. Acta Automatica Sinica, 2018, 44(12): 2250-2258
[22]	Yang X, Zhao C, Chen B. Progressive Gaussian approximation filter with adaptive measurement update. Measurement, 2019, 148: Article No. 106898 doi: 10.1016/j.measurement.2019.106898
[23]	Zhang J, Yang X, Zhang W A. A progressive Bayesian filtering framework for nonlinear systems with heavy-tailed noises. IEEE Transactions on Automatic Control, 2023, 68(3): 1918-1925 doi: 10.1109/TAC.2022.3172165
[24]	Ivorra E, Ortega Pérez M, Alcaniz Raya M L. Azure Kinect body tracking under review for the specific case of upper limb exercises. MM Science Journal (Online), 2021, 2021: 4333-4341 doi: 10.17973/MMSJ.2021_6_2021012
[25]	Yang X, Zhang W A, Liu A, Yu L. Linear fusion estimation for range-only target tracking with nonlinear transformation. IEEE Transactions on Industrial Informatics, 2020, 16(10): 6403-6412 doi: 10.1109/TII.2019.2955931
[26]	Romeo L, Marani R, Malosio M, Perri A G, D＇Orazio T. Performance analysis of body tracking with the Microsoft Azure Kinect. In: Proceedings of the 29th Mediterranean Conference on Control and Automation. Puglia, Italy: IEEE, 2021. 572−577
[27]	Antico M, Balletti N, Laudato G, Lazich A, Notarantonio M, Oliveto R, et al. Postural control assessment via Microsoft Azure Kinect DK: an evaluation study. Computer Methods and Programs in Biomedicine, 2021, 209: Article No. 106324 doi: 10.1016/j.cmpb.2021.106324
[28]	Nagymáté G, M Kiss R. Application of OptiTrack motion capture systems in human movement analysis. Recent Innovations in Mechatronics, 2018, 5(1): 1-9

施引文献

资源附件(0)

访问统计

点击查看大图

图(6) / 表(1)

计量

文章访问数: 198
HTML全文浏览量: 65
PDF下载量: 85
被引次数: 0

姓名
邮箱
手机号码
标题
留言内容
验证码

留言板

基于渐进高斯滤波融合的多视角人体姿态估计

doi: 10.16383/j.aas.c230316

计量

Multi-view Human Pose Estimation Based on Progressive Gaussian Filtering Fusion

计量

目录

实验方法	腕关节	肘关节	肩关节
观测融合	166.44	124.44	96.56
CF	157.55	118.00	95.00
AMFKF	147.81	113.85	93.08
CI	127.63	117.85	99.62
IWCF	153.12	113.21	92.53
PGFFwoC	151.77	114.12	92.83
PGFFwC	119.47	108.98	84.11

留言板

基于渐进高斯滤波融合的多视角人体姿态估计

doi: 10.16383/j.aas.c230316

计量

出版历程

Multi-view Human Pose Estimation Based on Progressive Gaussian Filtering Fusion

计量

出版历程

目录