摘要
研究一种基于MEMS(Micro-Electro-Mechanical System)传感器的人体运动识别系统,适用于乒乓球及羽毛球运动员在比赛或训练中的动作识别以及计数。该识别系统采集运动员持拍手臂的三轴加速度、三轴角速度及三轴姿态角信号。通过平滑滤波并寻找波峰波谷以及零点的方法对信号进行动作区间分割,提取出每一个单独动作数据,并对每段动作数据进行特征值提取。利用BP神经网络算法对收集的训练样本进行训练,通过BP神经网络输出动作识别结果。实现了乒乓球羽毛球运动中多达7种动作的识别及计数,具有较高准确性以及较好实时性。
In this paper, a motion recognition system based on MEMS（Micro-Electro-Mechanical System） sensor was studied, which was used in the competition and training of table tennis and badminton to recognize and count players’ motions. The recognition system collected the triaxial acceleration signal, triaxial angular velocity signal and triaxial attitude angle signal of the player’s arm which held the bag. By smoothing filtering and finding peaks and troughs and zeros, the signal was segmented by action intervals, and each individual action data was extracted. Feature values were extracted from each action data and trained by BP neural network algorithm. The recognition and counting of up to 7 kinds of action in table tennis and badminton realized through the output of BP neural network. The proposed recognition system has high accuracy and reliable real-time performance.
作者
李元良
史中权
李少辉
李嘉昕
陈富东
王瑞琪
丁汉祥
Li Yuanliang,Shi Zhongquan,Li Shaohui,Li Jiaxin,Chen Fudong,Wang Ruiqi,Ding Hanxiang(1.School of Mechanical Engineering, Hohai University, Changzhou 213022,Jiangsu,Chin)
出处
《计算机应用与软件》
北大核心
2018年第8期243-248,285,共7页
Computer Applications and Software
基金
中央高校基本科研业务费学生项目(2017B716X14)
关键词
平滑滤波
神经网络
运动识别
动作分割
Smooth filtering
Neural network
Motion recognition
Motion segmentation
