基于DeepPose和Faster RCNN的多目标人体骨骼节点检测算法

doi:10.7523/j.issn.2095-6134.2020.06.015

中国科学院大学学报 ›› 2020, Vol. 37 ›› Issue (6): 828-834.DOI: 10.7523/j.issn.2095-6134.2020.06.015

基于DeepPose和Faster RCNN的多目标人体骨骼节点检测算法

余保玲¹, 虞松坤¹, 孙耀然², 杨振³, 傅旭波¹

1. 浙江大学公共体育与艺术部, 杭州 310058;
2. 浙江大学光电科学与工程学院, 杭州 310058;
3. 中国科学院自动化研究所, 北京 100190

收稿日期:2019-05-07 修回日期:2019-10-30 发布日期:2020-11-15
通讯作者: 傅旭波
基金资助:
国家重点研发计划项目（2018AAA0100405）和胶州人工智能产业技术研究院开放课题资助

Human body joint nodes detection based on DeepPose and Faster RCNN

YU Baoling¹, YU Songkun¹, SUN Yaoran², YANG Zhen³, FU Xubo¹

1. Department of Public Physical and Art Education, Zhejiang University, Hangzhou 310058, China;
2. College of Optical Science and Engineering, Zhejiang University, Hangzhou 310058, China;
3. Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China

Received:2019-05-07 Revised:2019-10-30 Published:2020-11-15

摘要/Abstract

摘要： 近年来，随着计算机视觉技术的不断发展，深度学习技术在人体关节节点检测中得到了很好的应用。但是由于人体关节结构复杂，关节之间存在相互依赖的关系和互相遮挡的问题，因此人体骨骼节点检测依然是一个极具挑战的任务。传统的模型难以预测多个目标的骨骼节点，为了解决这个问题，提出一种基于Faster RCNN和DeepPose相结合的方法，首先通过Faster RCNN检测出包含人体的感兴趣区域，并将该区域作为改进的DeepPose算法的输入，使其能够处理多目标关节节点检测的问题。实验表明，该算法在MPII数据集的手腕、膝盖两种关键节点检测上均取得最好结果，比之前的最好结果各提升1.2%和0.3%，在全部的关键节点检测上PCKh为87.6%。

关键词: Faster RCNN, DeepPose, 人体关节节点检测

Abstract: Human body joint nodes detection is a considerably challenging task which has drawn enormous attention in the field of computer vision recently. The challenges of this task include: coping with the complex structure of human body joints, denoting the interdependence between joint nodes, and dealing with the sheltered and overlapped body joint nodes. Among the common solutions to this task, the models based on deep learning are widely applied and provide useful results. However, the existing models have following drawbacks: 1) comparatively low accuracy in prediction; 2) poor performance in multi-objective tasks. In our work, we proposed a novel method aiming at more satisfactory results. We firstly detect the relevant regions of human body with Faster RCNN, and then input the regions into a modified DeepPose algorithm. We achieve the state-of-the-art results in the detection of the wrist and knee on MPII dataset, improving 1.2% and 0.3% in PCKh, respectively. The total PCKh is 87.6% on MPII dataset.

Key words: Faster RCNN, DeepPose, human body joint nodes detection

中图分类号:

TP212.9

余保玲, 虞松坤, 孙耀然, 杨振, 傅旭波. 基于DeepPose和Faster RCNN的多目标人体骨骼节点检测算法[J]. 中国科学院大学学报, 2020, 37(6): 828-834.

YU Baoling, YU Songkun, SUN Yaoran, YANG Zhen, FU Xubo. Human body joint nodes detection based on DeepPose and Faster RCNN[J]. , 2020, 37(6): 828-834.

参考文献

[1] Tompson J J, Jain A, LeCun Y, et al. Joint training of a convolutional network and a graphical model for human pose estimation[C]//Advances in Neural Information Processing Systems. 2014:1799-1807.
[2] Toshev A, Szegedy C, DeepPose G. Human pose estimation via deep neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Columbus, OH, USA. 2014:24-27.
[3] Sapp B, Taskar B. Modec:multimodal decomposable models for human pose estimation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2013:3674-3681.
[4] Gkioxari G, Toshev A, Jaitly N. Chained predictions using convolutional neural networks[C]//European Conference on Computer Vision. Springer, Cham, 2016:728-743.
[5] He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions On Pattern Analysis and Machine Intelligence, 2015, 37(9):1904-1916.
[6] Girshick R. Fast r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015:1440-1448.
[7] Gkioxari G, Girshick R, Malik J. Contextual action recognition with r* cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015:1080-1088.
[8] He K, Gkioxari G, Dollár P, et al. Mask r-cnn[C]//Proceedings of the IEEE International Conference on Computer Vision. 2017:2961-2969.
[9] Ren S, He K, Girshick R, et al. Faster r-cnn:Towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems. 2015:91-99.
[10] Everingham M, Van Gool L, Williams C K I, et al. The pascal visual object classes (voc) challenge[J]. International Journal of Computer Vision, 2010, 88(2):303-338.
[11] Johnson S, Everingham M. Clustered pose and nonlinear appearance models for human pose estimation[C]//BMVC 2010. doi:10.5244/c.24.12.
[12] MPI human pose dafaset[DB/OL].[2019-05-05]. http://human-pose.mpi-inf.mpg.de/#evaluation.
[13] Tompson J, Goroshin R, Jain A, et al. Efficient object localization using convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015:648-656.
[14] Pishchulin L, Insafutdinov E, Tang S, et al. Deepcut:joint subset partition and labeling for multi person pose estimation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016:4929-4937.
[15] Wei S E, Ramakrishna V, Kanade T, et al. Convolutional pose machines[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016:4724-4732.

基于DeepPose和Faster RCNN的多目标人体骨骼节点检测算法

Human body joint nodes detection based on DeepPose and Faster RCNN

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 3

编辑推荐

Metrics

本文评价

访问统计

联系我们

[1]	朱士坤, 李冬, 黄奎, 张宝贤. 一种高效的轻量级无线传感器网络文件系统设计与实现[J]. 中国科学院大学学报, 2014, 31(1): 130-134.
[2]	李丽娟, 赵彤. 无线传感器网络中跨层能耗优化模型及求解算法[J]. 中国科学院大学学报, 2011, 28(3): 375-381.
[3]	姜志鹏, 高随祥. 一个无需知道节点间距离的无线传感器网络定位方法[J]. 中国科学院大学学报, 2011, 28(3): 382-388.