一种基于深度学习的高精度行人步长估算方法  

A high-precision pedestrian step length estimation method based on deep learning

作　　者：朱平 余学祥[1,2,3] 韩雨辰 张灿 肖星星 ZHU Ping;YU Xuexiang;HAN Yuchen;ZHAN Can;XIAO Xingxing(School of Spatial Information and Surveying and Mapping Engineering.Anhui University of Science and Technology,Huainan,Anhui 232001,China;Key Laboratory of Aerospace ground Cooperative Monitoring and Early Warning for Mining Disasters,Anhui University of Science and Technology,Huainan,Anbui 232001,China;Engineering Research Center of Coal Industry for Collaborative Monitoring of Environment and Disaster in Mining Area,Anhui University of Science and Technology,Huainan,Anhui 232001,China)

机构地区：[1]安徽理工大学空间信息与测绘工程学院,安徽淮南232001 [2]安徽理工大学矿山采动灾害空天地协同监测与预警安徽普通高校重点实验室,安徽淮南232001 [3]安徽理工大学矿区环境与灾害协同监测煤炭行业工程研究中心,安徽淮南232001

出　　处：《测绘科学》2023年第6期19-26,130,共9页Science of Surveying and Mapping

基　　金：2021年度安徽省科技重大科技专项(202103a05020026);2021年度安徽省重点研究与开发计划项目(202104a07020014);安徽省自然科学基金项目(2008085MD114);安徽理工大学2022年研究生创新基金项目(2022CX2168)。

摘　　要：针对行人惯性导航定位(PDR)算法中传统步长估计方法精度不高,现有的长短时记忆(LSTM)网络计算复杂度高的问题,该文提出一种基于堆栈式降噪自编码器、灰狼优化算法与门控循环单元(SDAE-GWO-GRU)的步长估计模型。采用深度学习网络SDAE对加速度数据进行降噪重构,然后将降噪后的数据输入经GWO优化后的GRU网络中进行训练。实验表明,该文的步长估计模型的单步步长误差率均值为1.09%,标准差为0.94%,精度优于传统的步长估计方法。Aiming at the problem that the accuracy of the traditional step length estimation method in the PDR algorithm is not high and the computational complexity of the existing LSTM network is high,a step length estimation model based on SDAE-GWO-GRU was proposed in this paper.The deep learning network SDAE was used to denoise and reconstruct the acceleration data,and then the denoised data was input into the GRU network optimized by GWO for training.Experimental results showed that the average error rate of the step length estimation model in this paper was 1.09%,and the standard deviation was 0.94%,which was better than the traditional step length estimation method.

关 键 词：步长估计 堆栈式降噪自编码器 灰狼算法 GRU网络 行人定位 

分 类 号：TN929.53[电子电信—通信与信息系统] TP212[电子电信—信息与通信工程]