摘要
针对传统的二维图片动作识别算法识别率相对不高、实时性不强的问题,提出一种三维的人体动作实时识别的方法。该方法首先通过Kinect获取人体三维骨骼数据,然后对骨骼数据信息采取归一化的方法进行数据对齐的预处理,使得与实时数据与标准数据的角度阈值和距离参考值统一。最后与标准动作采用多特征融合的识别算法对动作进行识别与匹配的方法,并在此基础上改进基于关节点角度的动作识别方法。实验结果表明,本文方法运行速度较快,可有效消除角度测量不稳定以及距离测量无法检测方向上的差异造成的动作匹配不准确。满足三维动作识别的实时性、鲁棒性要求。
Aiming at the problems of low recognition rate and low real-time performance of the traditional two-dimensional image motion recognition algorithm,a three-dimensional real-time human motion recognition method was proposed.Firstly,Kinect was used to obtain the three-dimensional bone data of human body.Then,the normalization method was used to preprocess the data alignment.Finally,the recognition algorithm of multi feature fusion was used to recognize and match the standard actions.The experimental results show that the method runs fast and can effectively eliminate the movement recognition inaccuracy caused by the instability of angle measurement and the inability of distance measurement to quantitatively describe the movement difference.It can meet the real-time and robust requirements of 3D motion recognition.
作者
胡新荣
王梦鸽
刘军平
彭涛
HU Xin-rong;WANG Meng-ge;LIU Jun-ping;PENG Tao(Engineering Research Center,Hubei Province for Clothing Information,Wuhan 430200,China;Mathematics and Computer School,Wuhan Textile University,Wuhan 430200,China)
出处
《科学技术与工程》
北大核心
2020年第34期14133-14137,共5页
Science Technology and Engineering
基金
湖北省高等学校优秀中青年科技创新团队计划(T201807)。
关键词
三维
动作识别
实时识别
骨骼数据
动作匹配
three-dimensional
posture recognition
real-time recognition
skeletal data
motion matching