轻量级CNN实时跌倒预测及嵌入式系统实现  

Lightweight CNN real-time fall prediction and embedded system implementation

作　　者：杜群贵[1] 钟威 Du Qungui;Zhong Wei(School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学机械与汽车工程系,广州市510640

出　　处：《电子测量技术》2022年第11期10-15,共6页Electronic Measurement Technology

基　　金：广东省自然科学基金(2021A1515012258)项目资助。

摘　　要：为了实现实时而准确的跌倒预测,同时将深度学习模型移植到于可穿戴端设备中运行,提出了一种轻量级卷积神经网络模型。借鉴深度可分离网络的轻量级模型思想,设计了网络结构,并优化通道数和卷积核尺寸,在保证准确率基本不变的情况下大大减小了模型计算复杂度。为将算法部署于可穿戴跌倒保护设备,提出了模型在嵌入式端的实时运行框架,并将算法编写为C程序,移植到了STM32单片机中。此模型在Sisfall数据集中获得了97.5%的准确率,204.3 ms的裕量时间。移植的模型仅有11.65 KB大小,在STM32单片机中的算法延时仅为8.24 ms。实验结果表明,该模型具有较高的预测精度和很好的实时性,为跌倒预测算法和跌倒保护装置的开发提供了进一步的参考。In order to achieve real-time and accurate fall prediction, and transplant the deep learning model to run on wearable devices, a lightweight convolutional neural network model is proposed. Drawing on the lightweight model idea of DSC network, the network structure is designed, and the number of channels and the size of convolution kernel are optimized, which greatly reduces the computational complexity of the model while keeping the accuracy rate basically unchanged. In order to deploy the algorithm in the wearable fall protection device, a real-time running framework of the model on the embedded side is proposed, and the algorithm is written as a C program and transplanted to the STM32 microcontroller. This model achieves 97.5% accuracy with 204.3 ms lead time on the Sisfall dataset. The transplanted model is only 11.65 KB in size, and the algorithm delay in the STM32 microcontroller is only 8.24 ms. The experimental results show that the model has high prediction accuracy and good real-time performance, which provides a further reference for the development of fall prediction algorithms and fall protection devices.

关 键 词：跌倒检测 深度可分离网络 嵌入式 可穿戴设备 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]