基于镀银玻璃微球/固体橡胶压阻传感器与机器学习的手势识别技术研究

Research on Gesture Recognition Technology Based on Silver-Coated Glass Microsphere/Solid Rubber Piezoresistive Sensors and Machine Learning

摘要: 针对聋哑人群与普通人群之间存在的沟通障碍问题，本文提出一种基于镀银玻璃微球/固体橡胶(Ag@GMs/SR)可拉伸压阻传感器(SPTS)与机器学习算法相结合的手势识别方案。该传感器以固体橡胶为柔性基体，采用浸渍干燥法将镀银玻璃微球作为导电填料附着于表面，具备高灵敏度、大拉伸范围与优异的环境稳定性，可缝入手套实现多通道手部运动捕获。采集信号经调理电路放大、滤波并通过蓝牙传输至上位机；利用滑动窗口平均与差分阈值法抑制环境干扰；采用主成分分析(PCA)降维并以径向基核支持向量机(SVM)实现手势分类。实验结果表明，系统对30类常用手语手势的平均识别率达到96.3%，响应时间约128 ms，在温湿度波动下识别率变化小于2%，为聋哑人与普通人之间实时高效沟通提供了可行的可穿戴解决方案。

Abstract: To address the communication barriers between deaf-mute individuals and the hearing population, this paper proposes a gesture recognition scheme based on a stretchable piezoresistive sensor (SPTS) composed of silver-coated glass microspheres and solid rubber (Ag@GMs/SR), integrated with machine learning algorithms. The sensor uses solid rubber as a flexible matrix, with silver-coated glass microspheres attached onto its surface as conductive fillers via an impregnation-drying method, providing high sensitivity, a wide stretchable range, and excellent environmental stability. The sensor can be sewn into a glove to achieve multi-channel hand motion capture. The collected signals are amplified, filtered, and transmitted to a host computer via Bluetooth. A sliding window averaging method combined with a differential thresholding technique is used to suppress environmental interference. Principal component analysis (PCA) is employed for dimensionality reduction, and a support vector machine (SVM) with a radial basis function kernel is adopted for gesture classification. Experimental results demonstrate that the proposed system achieves an average recognition rate of 96.3% for 30 common sign language gestures, with a response time of approximately 128 ms. Under fluctuating temperature and humidity conditions, the recognition rate varies by less than 2%. This provides a feasible wearable solution for real-time and effective communication between deaf-mute and hearing individuals.