摘要
本文针对人体身份及动作识别的问题,提出一种基于低分辨率红外阵列传感器并使用卷积神经网络进行分类识别的方法,这种方法可以识别出人的身份和跌倒、坐下以及行走动作。本文使用的卷积神经网络是基于VGGNet搭建的,由输入层、5层卷积层、3层池化层、1层全连接层和输出层构成,自动提取红外热图像中的信息特征,对身份及动作进行分类,在良好的隐私保护下避免了繁琐的人工提取特征。经过实验测试,卷积神经网络算法识别动作平均准确率为93.3%,其中行走识别准确率达到100%,坐下识别准确率为90%,跌倒识别准确率为90%,身份识别准确率为96.7%。
Aiming at the problem of human identity and motion recognition,a method based on low resolution infrared array sensor and using convolutional neural network for classification and recognition is proposed,which can identify the identity of people and actions of falling,sitting and walking.The convolutional neural network used in this paper is based on VGGNet.It consists of input layer,five-layer convolutional layer,three-layer pooling layer,one layer of fully connected layer and output layer.It automatically extracts information features in infrared thermal images,and classifies actions,avoids the cumbersome manual extraction features under good privacy protection.After experimental testing,the average accuracy of convolutional neural network algorithm recognition is 93.3%,of which the walking recognition accuracy rate is 100%,the sitting recognition accuracy is 90%,the fall recognition accuracy is 90%,and the identity recognition accuracy is 96.7%.
作者
王召军
许志猛
Wang Zhaojun;Xu Zhimeng(College of Physics and Information Engineering,Fuzhou University,Fuzhou 350108)
出处
《电气技术》
2019年第11期6-10,26,共6页
Electrical Engineering
基金
国家自然科学基金资助项目（61401100）
福建省自然科学基金资助项目（2018J01805）
福州大学人才基金（GXRC-18083)
关键词
低分辨率红外阵列传感器
卷积神经网络
动作识别
身份识别
low resolution infrared array sensor
convolutional neural network
motion recognition
identification
