Abstract:
To efficiently resolve action classification problem, a classification algorithm based on Action Energy Image(AEI) is proposed. The high dimensional feature space is reduced to lower dimensional space with (2D)2PCA. The nearest-neighbor classifier is adopted to distinguish different actions. It need not extract the period of the video, which is indispensable in some other methods. Experimental results show that the algorithm achieves higher classification accuracy with less running time and less memory space.
Key words:
action recognition,
intelligent supervision,
Action Energy Image(AEI),
(2D)2PCA
摘要:
提出一种基于行为能量图像(AEI)和双向二维主成分分析((2D)2PCA)的行为分类算法解决行为分类问题。该算法利用AEI作为识别特征,无需运动周期的分割,运用(2D)2PCA对特征空间降维,用最近邻方法分类。实验结果表明,该算法能以较少的运行时间获得较高的分类准确率。
关键词:
行为识别,
智能监控,
行为能量图像,
双向二维主成分分析
CLC Number:
LIN Chun-Li, WANG Ke-Dun, WANG Ke-Cheng, JIA Tu- , CHENG Mo-Qing. Action Classification Algorithm Based on AEI and (2D)2PCA[J]. Computer Engineering, 2010, 36(24): 145-146.
林春丽, 王科俊, 王克成, 夏余 , 程万胜. 基于AEI与(2D)2PCA的行为分类算法[J]. 计算机工程, 2010, 36(24): 145-146.