计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (7): 29-35.DOI: 10.3778/j.issn.1002-8331.1712-0405

• 热点与综述 • 上一篇    下一篇

基于稀疏编码局部时空描述子的动作识别方法

赵晓丽,田丽华,李  晨   

  1. 西安交通大学 软件学院,西安 710049
  • 出版日期:2018-04-01 发布日期:2018-04-16

Action recognition method based on sparse coding local spatio-temporal descriptors

ZHAO Xiaoli, TIAN Lihua, LI Chen   

  1. School of Software Engineering, Xi’an Jiaotong University, Xi’an 710049, China
  • Online:2018-04-01 Published:2018-04-16

摘要: 针对已有动作识别算法训练速度慢且识别精度不高等问题,提出了基于稀疏编码局部时空描述子的动作识别方法。该方法首先对深度图像进行法线提取,同时应用基于运动能量的自适应时空金字塔对动作帧分块;然后局部聚集法线,得到显著性局部时空描述子;对局部时空描述子进行稀疏编码得到一组字典向量来重构样本数据;最后利用简化粒子群(sPSO)优化SVM分类器找到最适合样本数据的分类模型。实验在MSRAction3D和MSRGesture3D公开数据集上达到了93.80%和95.83%的识别率,且训练速度较传统方法有明显提升,证明了该方法的有效性和鲁棒性。

关键词: 动作识别, 稀疏编码, 简化粒子群, 深度序列, 局部时空描述子

Abstract: To overcome the slow training speed and low recognition rate of the existing action recognition algorithm, an action recognition method based on the sparse coding local spatio-temporal descriptor is proposed. The method firstly extracts the normal of the depth image, and uses the adaptive spatio-temporal pyramid based on the action energy to block the action frames. Then the local spatio-temporal descriptors are obtained by the local aggregation normal. The local spatio-temporal descriptors are encoded by sparse coding to get a set of dictionary vectors to reconstruct the sample data. Finally, the simplified Particle Swarm Optimization(sPSO) is used to optimize the SVM classifier to find the most suitable sample data classification model. The experiment achieves a recognition rate of 93.80% and 95.83% on MSRAction3D and MSRGesture3D datasets, and the training speed is significantly improved compared with the previous methods, which proves the effectiveness and robustness of the method.

Key words: action recognition, sparse coding, simplified Particle Swarm Optimization(sPSO), depth sequence, local spatio-temporal descriptors