融合深度学习和特征点识别的室内定位研究  

An Indoor Positioning Technology Based on Deep Learning and Feature Point Recognition

作　　者：王陈哲 陈宇佳 陶诗量 李明[1] 胡博 贾军辉 陈浩男 WANG Chenzhe;CHEN Yujia;TAO Shiliang;LI Ming;HU Bo;JIA Junhui;CHEN Haonan(National Geomatics Center of China,Beijing 100830;Baidu(China)Company Limited,Beijing 100193;The Third Institute of Photogrammetry and Remote Sensing,Ministry of Natural Resources,Chengdu 610100;State Grid Location Based Service Co.,Ltd.,Beijing 102209,China)

机构地区：[1]国家基础地理信息中心,北京100830 [2]百度(中国)有限公司,北京100193 [3]自然资源部第三航测遥感院,四川成都610100 [4]国网思极位置服务有限公司,北京102209

出　　处：《地理与地理信息科学》2023年第6期23-27,共5页Geography and Geo-Information Science

基　　金：国家自然科学基金面上项目(62076027);国家档案局科技项目(2022-R-13)。

摘　　要：传统室内定位技术多需提前铺设硬件设备,成本较高且对受控环境有一定要求,为此,该文提出一种融合深度学习和特征点识别的室内定位方法。首先利用手机等移动设备终端获得实时视频流,通过基于卷积神经网络的标志物识别器对视频流中的每帧图像进行标志物识别,从而获得其粗略位置,完成概略定位;其次,对视频流中每帧图像进行特征点识别,并与构建的环境特征信息库中带有位置信息的特征点进行匹配,计算其精确位置,完成精细定位。该方法完全基于图像识别,无须依赖硬件设备,对受控环境普适性较高,在Andriod移动设备终端的试验表明,该方法具有较高的准确性与稳定性。Most of the traditional indoor positioning technologies need to lay hardware equipment in advance,which is expensive and difficult to apply to part of the space where additional equipment cannot be laid.In order to reduce equipment cost as much as possible and reduce preparation work,this paper proposes an indoor positioning technology that integrates deep learning and feature point recognition.This technology uses mobile devices such as mobile phones to obtain real-time video streams,and uses a convolutional neural network-based marker recognizer to identify markers in each frame of the video stream,so as to obtain its rough position and complete rough positioning.At the same time,the feature points are identified for each frame of image in the video stream,and matched with the feature points with location information in the constructed environment feature information database,and their positions are calculated to complete fine positioning.The technology proposed in this paper is based entirely on image identification,without relying on the laying of hardware equipment,and can be better applied to secret spaces.This paper has completed the program deployment of the proposed indoor positioning technology on the Andriod mobile device terminal.The experiments show that the proposed indoor positioning technology has high accuracy and stability.

关 键 词：室内定位 卷积神经网络 特征点识别 环境特征信息库 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]