基于L_(2,1)范数稀疏特征选择和超法向量的深度图像序列行为识别

Activity Recognition from Depth Image Sequences Based on L_(2,1)-norm Sparse Feature Selection and Super Normal Vector

结合L_(2,1)范数稀疏特征选择和超法向量提出了一种新的深度图像序列行为识别方法。首先从深度图像序列中提取超法向量特征;然后利用L_(2,1)范数稀疏特征选择方法从超法向量特征中选择出最具判别性的稀疏特征子集作为特征表示;最后利用线性分类器Liblinear进行分类。在MSR Action3D数据库上的实验结果表明,所提方法使用2%的超法向量特征获得的识别率为94.55%,并且具有比其他方法更高的识别精度。

This paper presented a novel method of activity recognition from depth image sequences based on L_（2,1）-norm sparse feature selection and super normal vector.First,the super normal vector feature is extracted from depth image sequences.Then the most discriminative feature subset is selected from the whole super normal vector feature set based on the method of L_（2,1）-norm sparse feature selection.Finally,the classification is based on Liblinear classifier.Experimental results on MSR Action3 Ddataset show that the proposed method achieves 94.55% of recognition accuracy using only 2% of the whole super normal vector feature,and is superior to the state-of-art methods.