基于星型模型与多条件随机场的人体行为识别  

Human Action Recognition Based on Star Model and Multiple CRF Ensemble Model

作　　者：赵亚欣 赵怀勋 

机构地区：[1]武警工程大学研究生管理大队,西安710086 [2]武警工程大学信息工程系,西安710086

出　　处：《武警工程大学学报》2016年第4期12-16,共5页Journal of Engineering University of the Chinese People＇s Armed Police Force

摘　　要：为提高人体行为识别的实时性与准确性，提出一种基于星型模型与多条件随机场（Multiple Conditional RandomFields，MCRF）相结合的行为识别方法。首先，利用混合高斯模型进行背景建模，提取人体目标，建立人体星型模型。其次，通过建立人体坐标系，确定星型模型中各关键点的坐标提取距离、速度、角度、轨迹特征，并对每一特征集进行条件随机场建模，联合4类条件随机场模型建立多条件随机场模型，进行人体行为的识别。通过在KTH人体行为数据库上进行测试，结果证明，此方法能实时、准确的进行人体行为识别。In order to improve instantaneity and accuracy of action recognition, a combined method based on star model and multiple CRF ensemble model is proposed. After extracting human target by Gaussian Mixture Model（GMM） ,Which establish a background and a star model of human body. Then, establish a coordinate system of human body and extract fea- tures of distance, speed, angle and traiectory by coordinates of key points. CRF model is used for each feature subset and all the CRF models are combined to produce MCRF model,which is utilized to recognize human action. The experimental results in database of human behavior KTH indicate that human action can be recognized instantly and accurately by this method.

关 键 词：星型模型 特征提取 多条件随机场 行为识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]