| 摘要: |
| 【目的】研究基于声波的非接触式手势控制方法，实现对圈舍智能设备的非接触式手势控制，为减少畜牧业传染病的接触式传播风险提供技术支持。【方法】设计并实现了基于声波的手势识别系统，该系统由2套商用扬声器和麦克风组成声学雷达，采用功率谱密度对手势移动产生的多普勒效应进行提取，然后建立手势、位置与多普勒效应的映射关系，并提出一种融合规则和机器学习的手势识别方法，通过手势移动时产生的多普勒效应实现感知范围内任意位置对4种常见手势（前推、后移、左移、右移）的准确识别。【结果】采用奇异值分解算法进行特征提取，并对支持向量机模型、BP神经网络模型、K最邻近算法3种机器学习算法进行比较，可以得到支持向量机模型整体优于BP神经网络和K最邻近算法，且线性支持向量机的识别准确率最高，该系统对15位测试者在不同位置手势的平均识别准确率可以达到91.50%，且成功应用于照明设备的开关和亮度调节以及换气扇的开关和转速调节。【结论】综合考虑手势执行位置、手势移动速度和幅度等因素，采用奇异值分解算法进行特征提取，线性支持向量机算法进行分类，可达到较高的手势识别准确率，有望应用于圈舍智能设备的非接触式手势控制。 |
| 关键词: 声波感知 手势识别 多普勒效应 机器学习 |
| DOI： |
| 分类号: |
| 基金项目:中国博士后科学基金面上项目（2020M673504）;国家重点研发计划项目（2020YFD1100602） |
|
| Contactless gesture control of enclosure intelligent equipment using acoustic signal |
|
CHEN Ziyi,WANG Tianben,LIU Xiantao,et al
|
| Abstract: |
| 【Objective】This study designed an acoustic based contactless gesture control method to control enclosure intelligent equipment to provide support for reducing the spread risk of infectious disease caused by contact equipment control.【Method】The designed acoustic based gesture recognition system was composed of two pairs of off-the-shelf speakers and microphones working as an acoustic radar to recognize gestures.The Doppler shift caused by gestures was extracted using power spectrum density (PSD) of echo.Then,the relationship between gesture and Doppler shift was obtained and a fusion framework combining rule-based method and machine learning method was established.By fully leveraging the Doppler shifts caused by gesture movement,four common gestures including pushing forward,pulling backward,moving left and moving right at any location in perception range were recognized.【Result】The singular value decomposition (SVD) method was used for feature extraction and three methods of support vector machine (SVM),BP neural network and K nearest neighbor were compared.The SVM model with linear kernel function performed the best and achieved the average gesture recognition accuracy of 91.50% at different locations.The system was also successfully applied to control light equipment and ventilator.【Conclusion】Considering location,gesture moving speed and moving range,the established system achieved high gesture recognition accuracy with SVD for feature extraction and SVM for classification.It is expected to be applied to the contactless gesture control of enclosure intelligent equipment. |
| Key words: acoustic sensing gesture recognition Doppler effect machine learning |
