摘要
行为识别是图像处理的一个热点问题。一些研究表明在监督学习的框架下,通过提取时空兴趣点(Spatial-temporal interest points)能较好地识别人体的行为。由于兴趣点中包含与人体行为无关的噪声点,为了改进兴趣点的提取,提出了一种基于人体骨架的改进方法。该方法通过粒子滤波(Particle filter)算法改进人体骨架的精度,而改进后的人体骨架,能得到更有效的兴趣点。通过在"Weizmann","KTH"数据集的测试,实验结果表明,该算法不仅能够提高人体行为的识别,而且能够改进人体骨架的精确度。
Action recognition is a hot research topic in image processing area. Some studies have shown that based on supervised learning, spatial-temporal interest points which are extracted from images can recognize human action. Since interest points contain some noises which are not related to human action, a method which is based on human skeleton is presented to refine interest points. This method can improve the precision of human skeleton by particle filter. The refined human skeleton is used to get better interest points. Based on “Weizmann”,“KTH” dataset, experiment results show that the method can improve the precision of human action recognition and human skeleton.
出处
《重庆师范大学学报（自然科学版）》
CAS
CSCD
北大核心
2014年第5期105-109,共5页
Journal of Chongqing Normal University:Natural Science
关键词
行为识别
时空兴趣点
粒子滤波
姿态估计
支持向量机
human action recognition
spatial-temporal interest points
particle filter
pose estimation
support vector machine
