用微型机器学习和传感器实现低功耗步态识别
Gait Recognition Based on Tiny ML and IMU Sensors
- 在微控制器上部署深度神经网络，实时分类四种运动状态
- 模型测试准确率超80%，支持异常检测提升系统鲁棒性
- 适合电池供电设备，适用于可穿戴健康监测场景
本项目基于微型机器学习（Tiny ML）与惯性测量单元（IMU）传感器开发了一种步态识别系统。系统采用XIAO-nRF52840 Sense微控制器与LSM6DS3 IMU传感器，采集行走、静止、上楼、下楼四类活动的加速度与角速度数据。数据经边缘人工智能平台Edge Impulse处理，通过滑动窗口与数据归一化等预处理，训练深度神经网络（DNN）分类器实现活动识别。模型在测试集上准确率超过80%，具备有效分类能力，并支持异常检测功能。系统依托Tiny ML实现低功耗运行，适用于电池供电或能量采集设备。
原文摘要 · Abstract (English)
This project presents the development of a gait recognition system using Tiny Machine Learning (Tiny ML) and Inertial Measurement Unit (IMU) sensors. The system leverages the XIAO-nRF52840 Sense microcontroller and the LSM6DS3 IMU sensor to capture motion data, including acceleration and angular velocity, from four distinct activities: walking, stationary, going upstairs, and going downstairs. The data collected is processed through Edge Impulse, an edge AI platform, which enables the training of machine learning models that can be deployed directly onto the microcontroller for real-time activity classification.The data preprocessing step involves extracting relevant features from the raw sensor data using techniques such as sliding windows and data normalization, followed by training a Deep Neural Network (DNN) classifier for activity recognition. The model achieves over 80% accuracy on a test dataset, demonstrating its ability to classify the four activities effectively. Additionally, the platform enables anomaly detection, further enhancing the robustness of the system. The integration of Tiny ML ensures low-power operation, making it suitable for battery-powered or energy-harvesting devices.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。