基于IMU传感器与深度度量学习的人体行为识别算法

Human Activity Recognition Algorithm Based on Inertia Measurement Unit Sensors and Deep Metric Learning

人体行为识别可以定义为通过一系列观察和周围环境来确定一个人的各种姿势和日常活动。很多研究尝试将深度学习技术用于HAR中,然而,现有的基于DL的HAR方法存在复杂度较高、算力需求大和泛化性与鲁棒性不足的问题。为了解决上述问题,围绕基于智能手机内置IMU传感器的HAR方法,提出了一种名为RMDML的HAR方法,该方法结合了轻量化神经网络Res-MLP和深度度量学习的特征嵌入技术,旨在提取具有可分离性与可判别性的泛化特征,从而提高模型识别性能和泛化性能。RMDML模型在公开数据集UCI HAR上取得了97.26%的准确率,高于几种常见的HAR算法,证明了所提出方法的有效性。

Human activity recognition(HAR)is the process of determining a person’s various postures and daily activities through a series of observations and the surrounding environment.Many studies have attempted to use deep learning(DL)techniques for HAR.However,existing DL-based HAR methods suffer from issues such as high complexity,large computational requirements,and insufficient generalization and robustness.To address these issues,a new HAR method called RMDML is proposed that focuses on inertia measurement unit(IMU)sensors embedded in smartphones.RMDML combines a lightweight neural network called Residual Multi-Layer Perceptron(Res-MLP)with deep metric learning feature embedding technology to extract generalizable features with separability and discriminability,thereby improving the model recognition performance and generalization ability.RMDML achieves an accuracy of 97.26%on the publicly available UCI HAR dataset,which is higher than several common HAR algorithms,demonstrating the effectiveness of the proposed method.