一种基于关节点信息的人体行为识别新方法  

A Novel Human Activity Recognition Method Using Joint Points Information

作　　者：田国会[1] 尹建芹[1,2] 韩旭[1] 于静[1] 

机构地区：[1]山东大学控制科学与工程学院,山东济南250061 [2]济南大学信息科学与工程学院山东省网络环境智能计算技术重点实验室,山东济南250022

出　　处：《机器人》2014年第3期285-292,共8页Robot

基　　金：国家自然科学基金资助项目(61075092;61203341);山东省自然科学基金资助项目(ZR2011FM011);山东大学自主创新基金资助项目(2011JC017)

摘　　要：为了提高机器人服务的主动性与智能性,使用Kinect体感设备获取人体的关节点数据解决人体行为识别问题.首先,利用Kinect采集人体关节点坐标,构造用于表示人体结构的3维空间向量,然后计算结构向量之间的角度和向量模的比值,进行人体姿态描述,同时以一段时间内连续的姿态序列作为行为表示特征量,最后选用动态时间规整(DTW)算法计算测试行为模板与参考行为模板之间的相似度以实现行为识别.实验结果表明,选用的行为表示特征量具有旋转与平移不变性.另外,对人在日常生活中的6种行为进行了识别实验,结果表明本文的行为识别算法可以取得较好的识别效果.In order to increase the activeness and intelligence of robot services, Kinect is used to collect the joint point information for the recognition of human activities. Firstly, coordinates of human joint points are acquired by Kinect, and the three-dimensional structure vector is constructed to represent the human body structure. Then, the angles and the ratios of structure vector lengths between the space vectors are computed to represent the human postures. Meanwhile, the behavior is represented by continuous posture sequences within a limited timespan. At last, dynamic time warping （DTW） algorithm is applied to implementing the activity recognition by calculating the similarity between test behavior sequence and reference activity template. The experimental results show that the behavior representation scheme is invariant to rotation and transla-tion. Moreover, experiments are conducted to recognize 6 kinds of human activities in daily living, and the results show that better recognition results can be acquired by the proposed algorithm.

关 键 词：人体行为识别 人体关节点 行为表示 动态时间规整 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]