摘要
针对目前动态手势识别计算复杂度较高以及对实验器材有相应要求的问题,提出基于多特征融合的动态手势识别。使用OpenPose得到手部关键点信息,建立手势模型,将坐标信息利用手部的结构关系进行处理,得到手部的角度和长度特征。将角度特征序列和长度特征序列进行融合,利用阈值设定过滤序列中的奇异点,使用FastDTW算法计算待测动态手势与手势模板库中的序列距离,得到预测手势动作类别。实验表明,该方法计算复杂度较低,识别速度快,选取的四种手势动作的识别准确率均在90%以上,有较好的识别效果。
In view of the high computational complexity of dynamic gesture recognition and the high requirements for experimental equipment,this paper proposes dynamic gesture recognition based on multi-feature fusion.It used OpenPose to obtain key information of hand,established gesture model,and used coordinate information.The structural relationship of the part was processed to obtain the angle and length characteristics of the hand.The angle feature sequence and the length feature sequence were fused,the threshold was used to set the singular point in the filter sequence,and the FastDTW algorithm was used to calculate the sequence distance between the dynamic gesture to be tested and the gesture template library,and the predicted gesture action category was obtained.Experiments show that the computational complexity of this method is low and the speed is fast.The recognition accuracy of the four gestures selected in this paper is above 90%,which has a good recognition effect.
作者
李东东
张立民
邓向阳
姜杰
Li Dongdong;Zhang Limin;Deng Xiangyang;Jiang Jie(Naval Aviation University,Yantai 264000,Shandong,China)
出处
《计算机应用与软件》
北大核心
2021年第8期214-219,共6页
Computer Applications and Software
关键词
动态手势识别
动态时间规整
手势模板库
Dynamic gesture recognition
Dynamic time warping
Gesture template library
