视频序列中基于LBP特征的人体行为识别  

Human Action Recognition Based on Local Binary Pattern Feature in Video Sequences

作　　者：王宪[1] 慕鑫[1] 宋书林[1] 陈向阳[1] 

机构地区：[1]轻工过程先进控制教育部重点实验室(江南大学),江苏无锡214122

出　　处：《光电工程》2013年第3期108-114,共7页Opto-Electronic Engineering

基　　金：国家自然科学基金(60574051);江苏省产学研联合创新资金-前瞻性联合研究项目(BY2012067)

摘　　要：视频序列中的行为分析与识别已经成为当前计算机视觉领域的研究热点。为了更加有效地提取人体行为序列中的轮廓特征的信息,提出了一种基于局部二值模式(Local Binary Pattern,LBP)特征的人体行为识别的算法。通过背景差分法从视频中提取完整的人体运动序列,利用LBP算子计算运动序列的LBP特征谱,组成样本的LBP轮廓特征空间,接着将特征空间通过K-means聚类的方法生成行为特征。最后,采用隐马尔可夫模型(HMM)对特征进行识别。实验过程中,分别在两个公共行为数据库上进行了测试实验,平均识别率能达到85%以上,并且在两个数据库的交叉实验结果表明了本文算法具有一定的鲁棒性。Human action recognition in the video sequence have become a hot research topic in computer vision field. In order to extract the contour feature of the human＇s behavior sequence more effectively, a new algorithm for human action recognition based on Local Binary Pattern （LBP） is proposed. Firstly, background subtraction algorithm is used to extract the complete human motion sequence in the video, and the LBP operators are used to calculate the samples＇ LBP feature space which is composed of the motion sequences＇ LBP feature spectrum. Then, the behavior feature is generated by k-means clustering method. Finally, the Hidden Markov Model （HMM） is adopted for the classification. During the experiment, the test experiment is performed in the two public behavior databases respectively, and the average recognition rate can reach more than 85%. The intersection of the two databases experimental results shows that the proposed algorithm has certain robustness

关 键 词：行为识别 局部二值模式 K-MEANS聚类 隐马尔科夫模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]