Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (1): 150-160.DOI: 10.3778/j.issn.1002-8331.1909-0383

Previous Articles     Next Articles

Research on Improved Mask R-CNN Network Model for Human Keypoint Detection

SONG Ling, XIA Zhimin   

  1. College of Computer and Electronical Information, Guangxi University, Nanning 530004, China
  • Online:2021-01-01 Published:2020-12-31

人体关键点检测的Mask R-CNN网络模型改进研究


  1. 广西大学 计算机与电子信息学院,南宁 530004


To solve the problem that the computational cost of Mask R-CNN is too high and the training time is too long due to the large number of iterations, which caused by the parameter quantity in the deep learning solution of the human keypoint detection problem. An improvement of Mask R-CNN network model is proposed, which based on ShuffleNet network. The reassembly channel network of ShuffleNet is taken by this network model, through using group point-to-point convolution and channel shuffle operations and combining the calculation results of the bounding regression with the mask representation, The results show that the improved network model can speed up the operation and reduce the detection time in solving the problem of human key point detection in multi-person situations with preserving accuracy.

Key words: deep learning, Convolution Neural Network(CNN), Mask R-CNN, ShuffleNet, human keypoint detection


由于在现有的人体关键点检测问题中,深度学习解决方案采用的掩膜区域卷积神经网络Mask R-CNN存在参数量大导致计算成本过高、迭代次数多导致训练时间过长等问题,提出了一种基于重组通道网络ShuffleNet改进 Mask R-CNN网络模型。通过引入ShuffleNet的网络结构,使用分组逐点卷积与通道重排的操作与联合边框回归和掩膜分割的计算结果对Mask R-CNN进行轻量化改进。使用该方法改进网络模型在进行单人或多人情况下的人体关键点检测中,在保留精度的前提下,可以加快运行速度,减少检测时间。

关键词: 深度学习, 卷积神经网络(CNN), 掩膜区域卷积神经网络(Mask R-CNN), 重组通道网络, 人体关键点检测