Pulsed Millimeter Wave Radar for Hand Gesture Sensing and Classification
Pulsed Millimeter Wave Radar for Hand Gesture Sensing and Classification
复制标题
DOI：
10.1109/lsens.2019.2953022
复制
发表时间：
2019-12-01
影响因子：
2.8
通讯作者：
Wernersson, Lars-Erik
中科院分区：
文献类型：
--
作者：
Fhager, Lars Ohlsson;Heunisch, Sebastian;Wernersson, Lars-Erik
利用以144 Hz的帧速率操作的脉冲毫米波雷达来记录12个通用手势的2160个散射特征。手势识别是通过机器学习来实现的，利用预先训练的卷积神经网络上的迁移学习。这产生了出色的分类结果，验证准确率为99.5%，基于60%的训练与40%的验证分割。相应的混淆矩阵也被提出，示出了测试手势之间的高水平的分类正交性。这是第一次将脉冲毫米波雷达的数据用于机器学习的手势识别。证明了脉冲雷达高帧频数据的距离-时间包络表示适合于手势识别。对于更复杂的检测方案和定制的神经网络，预计会有进一步的改进。
A pulsed millimeter wave radar operating at a frame rate of 144 Hz is utilized to record 2160 scattering signatures of 12 generic hand gestures. Gesture recognition is achieved by machine learning, utilizing transfer learning on a pretrained convolutional neural network. This yields excellent classification results with a validation accuracy of 99.5%, based on a 60% training versus 40% validation split. The corresponding confusion matrix is also presented, showing a high level of classification orthogonality between the tested gestures. This is the first demonstration where data from a pulsed millimeter wave radar is used for gesture recognition by machine learning. It proves that the range-time envelope representation of high frame-rate data from a pulsed radar is suitable for hand gesture recognition. Further improvements are expected for more complex detection schemes and tailored neural networks.