期刊论文

Acceleration gesture recognition based on deep convolution long short-term memory network

中文

电子测量技术

1002-7300

2019

42

21

109-113

10.19651/j.cnki.emt.1903021

国家科技支撑计划课题(2015BAK33B00)资助。

本校为第一机构

手势加速度识别是基于惯性传感器手势交互的重要研究内容。针对已有的手势识别方法严重依赖于人工选取特征或不能有效融合手势动作的时空特征等局限性较大的问题,提出了基于卷积神经网络和长短时记忆网络结合的加速度手势识别算法。该算法通过构造三层卷积神经网络提取手势加速度数据的空间特征,并经过Droupout正则化操作,避免了特征过拟合的问题,再通过构造两层长短时记忆网络学习手势加速度数据的时序特征,融合手势动作的时空特征进行手势识别分类。实验表明,该方法能够有效识别加速度手势动作。

Gesture acceleration recognition is an important research based on inertial sensor gesture interaction.Aiming at the problem that the existing method relies on the artificial selection or can not effectively fuse the space-time feature,an acceleration gesture recognition algorithm based on Convolutional neural network and long short-term memory network is proposed.The algorithm extracted the spatial characteristics by constructing a three-layer Convolutional Neural Network,and avoided the problem of feature over-fitting by Droupout regularization operation.Then it learned the time-order charac...