摘要
针对室内旅游场景复杂环境造成定位精度低和稳定性差等问题,以及为避免传统的人工设定、提取图像的特征复杂过程,提出了一种基于卷积神经网络的图像识别室内旅游场景的定位方法。该方法通过移动设备自带摄像头拍摄室内旅游场景图像,建立包括位置和方向信息的图像指纹数据库,运用深度卷积神经网络(CNN)对图像指纹数据进行训练。已训练的CNN对移动设备拍摄当前位置的图像进行图像匹配,从而实现位置的精准定位。实验测试用户用手机拍摄图像进行定位,结果:CNN模型训练,准确率高达99.5%;定位准确率大部分在90%以上,定位精度在直径1.5 m范围,证明了算法具有精度高、很好的鲁棒性、泛化能力较强。
To overcome the problems of low accuracy and poor stability brought by the complexity of scenarios,an indoor localization via image matching based on Deep Learning is proposed.The method includes taking images of indoor surroundings with cameras of mobile devices,setting up a dataset of images containing information on position and direction,and training a Convolutional Neural Network(CNN)with the image data.Then use the trained CNN to match the current images taken by the cameras of mobile devices to estimate precise location.The results of experiments show that the accuracy rate of CNN can reach up to 99.5%,positioning accuracy rate is up to 90%,and positioning precision is within 1.5 metres of diameter.This algorithm can achieve sound robustness,and fairly excellent generalization capabilities.
作者
梁宇
Liang Yu(Guilin Tourism University,Guilin,China)
出处
《科学技术创新》
2023年第12期105-109,共5页
Scientific and Technological Innovation
基金
广西高校中青年教师(科研)基础能力提升项目《基于智慧旅游的室内旅游场景的定位技术研究》(项目编号2019KY0828)(批准文件编号桂教科研(2019)1号)。
关键词
室内定位
深度学习
定位算法
图像匹配
indoor positioning
deep learning
location algorithm
image matching
