Shape and Texture Features based Human Action Recognition Using Collaborative Representation Classification
Shape and Texture Features based Human Action Recognition Using Collaborative Representation Classification
复制标题
使用协作表示分类的基于形状和纹理特征的人体动作识别
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
N. Parveen
中科院分区：
文献类型：
--
作者：
Lasker Ershad Ali;M. Islam;B. Madhu;M. Bulbul;N. Parveen
本文提出了一种基于形状和纹理的DMM-Haar特征的人体动作识别方法，并采用协同表示分类器对动作进行分类。在这项研究中，我们引入了有效的特征提取技术的基础上深度运动图（DMMs）和Haar小波变换，不同的动作可以表示与一系列的功能。首先，我们计算了三个DMM，如DMM前视图，顶视图和侧视图从三维动作视频序列的形状特征。之后，我们利用Haar小波对DMMs产生的图像提取纹理信息，并连接所有功能作为一个特征矩阵。我们利用主成分分析减少特征矩阵的特征维数。最后，采用基于2l赋范的协同代表性分类技术对不同行为进行分类.在这项研究中，我们已经分析了DMM-Haar功能的影响，在实验的基础上与DMM功能为基础的结果。所提出的方法的性能研究是与国家的最先进的方法来识别人类的行动公开可用的Microsoft Research Action 3D数据集。
This paper presents human action recognition by using shape and texture based DMM-Haar features where collaborative representation classifier is adopted for action classification. In this study, we have introduced effective feature extraction technique based on Depth Motion Maps (DMMs) and Haar wavelet transformation, where different actions can be represented with a range of features. Firstly, we have calculated three DMMs such as DMM front view, top view and side view from 3D action video sequences as the shape features. After that, we have utilized Haar wavelet on the DMMs generated images to extract texture information and concatenated all features as a feature matrix. We have utilized principal component analysis for reducing the feature dimensions of the feature matrix. Finally, 2 l normed based collaborative representative classification technique is adopted to classify different actions. For this research, we have analyzed the effects of the DMM-Haar features on experimental basis with DMM features based results. The performance study of the proposed method is comparable with the state-of-the-art methods to recognize human action on the publicly available Microsoft Research Action 3D dataset.