Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (15): 169-176.DOI: 10.3778/j.issn.1002-8331.1805-0493

Previous Articles     Next Articles

Human Action Recognition Based on DBN-HMM

YANG Shiqiang, LUO Xiaoyu, LI Xiaoli, YANG Jiangtao, LI Dexin   

  1. Faculty of Mechanical and Precision Instrument Engineering, Xi’an University of Technology, Xi’an 710048, China
  • Online:2019-08-01 Published:2019-07-26

基于DBN-HMM的人体动作识别

杨世强,罗晓宇,李小莉,杨江涛,李德信   

  1. 西安理工大学 机械与精密仪器工程学院,西安 710048

Abstract: The action recognition enables the machine to discriminate and understand the intention of the human action. And then it realizes efficient human-machine interaction. A new limb angle model is proposed to express the human action in 3-D space. The model has certain invariability and low computational complexity. In view of the traditional action recognition method based on Gaussian Mixture Model and Hidden Markov Model(GMM-HMM), an action recognition model is proposed combining the Deep Belief Network(DBN) model and Hidden Markov Model(HMM). A nonlinear DBN model based on the condition restricted Boltzmann machine(CRBM) is constructed. It has strong capable to model because of its deep structure, and it combines with historical information. It is more suitable for action recognition. Experiments results show that the algorithm has higher recognition rate, and it is feasible.

Key words: action recognition, limb angle model, Hidden Markov Mode(HMM), Conditional Restricted Boltzman Machine(CRBM), deep belief network

摘要: 动作识别使得机器能够对人体动作的意图进行判别理解,进而实现高效的人机交互。提出一种肢体角度模型,实现在三维空间中对人体动作进行表示,该模型具有一定的不变性,计算复杂度低。针对传统的基于混合高斯的隐马尔可夫模型(GMM-HMM)的动作识别,提出深度置信网络模型(DBN)和隐马尔可夫模型相结合的动作识别模型,构建了一种非线性的基于条件限制玻尔兹曼机(CRBM)的DBN深度学习模型,深层次结构使其建模能力更强,且能够结合历史信息建模,更适用于动作识别。实验表明该算法具有较高的识别结果。

关键词: 动作识别, 肢体角度模型, 隐马尔可夫模型, 条件限制玻尔兹曼兹曼机, 深度置信网络