基于时空兴趣点的人体行为识别  

Human Action Recognition Based on Space-time Interest Point

作　　者：付朝霞[1,2] 王黎明[1] 

机构地区：[1]中北大学电子测试技术重点实验室&仪器科学与动态测试教育部重点实验室,信息通信工程学院,山西太原030051 [2]山西省委党校信息网络教研部,山西太原030006

出　　处：《微电子学与计算机》2013年第8期28-30,35,共4页Microelectronics & Computer

基　　金：国家重点基础研究发展计划(2011CG311804);国家自然资金重点项目(61071193)

摘　　要：提出一种基于时空兴趣点的人体行为识别算法,该算法从行为视频中提取丰富的时空兴趣点,然后在数据空间采用基于高斯混合模型的聚类方法完成对特征点的分类.在动作识别过程中,采用平均Hausdorff距离来衡量序列间匹配的相似性,提高运算效率.KTH数据库上的实验证明了该算法的有效性和鲁棒性.Human action analysis and recognition are increasingly attracting much attention from computer vision and pattern recognition researchers.This paper presents a human action recognition based on space-time interest point algorithm.The algorithm detects dense space-time interest points from action videos and classifies these feature points using cluster method based on Gaussian mixture model in the data space.In the process of action recognition,a matching-based approach with the mean Hausdorff distance is measured the similarity between image sequences to improve the operation efficiency.The experiments on the KTH database prove the effectiveness and robustness of the algorithm.

关 键 词：行为识别 特征提取 时空兴趣点 隐马尔可夫模型 平均Hausdorff距离 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]