论文部分内容阅读
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 improve the initiative and intelligence of robot service, Kinect somatosensory equipment is used to obtain the body’s joint point data to solve the problem of human behavior identification.Firstly, Kinect is used to collect the coordinates of the human joint point to construct a 3-dimensional space vector for representing human body structure, Calculate the ratio between the angle of the structure vector and the vector modulus, describe the human pose, at the same time take the continuous gesture sequence for a period of time as the behavioral representation feature quantity, and finally choose the dynamic time warping (DTW) algorithm to calculate the test behavior template and the reference behavior template The experimental results show that the chosen behavior indicates that the feature quantity has rotation and translational invariance.In addition, six kinds of behaviors of people in daily life are identified and the results show that the behavior recognition The algorithm can achieve better recognition effect.