标题:
基于搜索算法K-Means动作关键特征序列的行为识别方法Action Recognition Based on Key Poses Sequences with Searching-Based K-Means Algorithm
作者:
殷鑫, 龚声蓉, 刘纯平
关键字:
行为识别, 集合论, 平均聚类算法, 动态时间规整算法Action Recognition, Set Theory, K-Means, Dynamic Time Warping
期刊名称:
《Journal of Image and Signal Processing》, Vol.4 No.1, 2015-02-16
摘要:
行为识别是近年来计算机视觉领域的一个研究热点。本文在当今已有的行为识别算法的基础之上进行优化改进。通过基于代数理论的背景减法提取轮廓并进行姿势表达、通过聚类算法提取动作关键特征，并基于DTW动态时间规整算法完成动作识别。由于原始K-means算法中聚类结果对于初值的依赖性，我们引入基于搜索算法的K-means聚类算法，避免了初值对聚类结果的影响。通过在国际主要数据库上的实验，达到了较高的准确率和稳定度，并能够实现在线实时识别。Vision-based human action recognition is currently one of the most active research fields. Action recognition is a cross-disciplinary field which contains theories of image processing, computer vi-sion and artificial intelligence. Firstly, we get contours and pose presentation through background subtraction algorithm based on algebra theory and then we get the key poses of action through improved searching-based K-means algorithm. Finally actions are recognized through dynamic time warping algorithm. Experimental results on the main datasets show suitability for online recognition and real-time scenarios.