测绘通报 ›› 2017, Vol. 0 ›› Issue (10): 100-105.doi: 10.13474/j.cnki.11-2246.2017.0324

• 学术研究 • 上一篇    下一篇

利用单相机和神经网络估计的室内定位导航

张丽娜1, 彭力2   

  1. 1. 浙江安防职业技术学院信息工程学院, 浙江 温州 325000;
    2. 江南大学信息学院, 江苏 无锡 214000
  • 收稿日期:2017-02-21 出版日期:2017-10-25 发布日期:2017-11-07
  • 作者简介:张丽娜(1980-),女,硕士,副教授,主要研究方向为数据挖掘、大数据应用及定位等。E-mail:linazhang1980@163.com
  • 基金资助:
    浙江省教育厅一般科研项目(Y201635414);温州市科技局公益性科技计划(S20160021);浙江安防职业技术学院重点科研项目(AF2016Z02)

Indoor Positioning Navigation Based on Single Camera and Neural Network Estimation

ZHANG Lina1, PENG Li2   

  1. 1. Department of Information Engineering, Zhejiang Security Career Technical College, Wenzhou 325000, China;
    2. Department of Information Science, Jiangnan University, Wuxi 214000, China
  • Received:2017-02-21 Online:2017-10-25 Published:2017-11-07

摘要: 由于三维场景与二维图像之间存在着非线性和高度复杂的关系,使用相机对用户的位置进行估计需要建立复杂的数学模型。针对该问题,本文提出了使用神经网络估计的单相机进行室内定位的方法。室内定位系统的主要优点是LED能够使用可见光通信发送其位置信息。首先,该方法充分利用LED光线的投影不变性,借助图像传感器通信(ISC)完成虚拟直线的构建;然后,运用神经网络估计从该虚拟直线中提取出相机的方向信息;最后,使用一个简单数学方程估计用户位置。仿真试验考虑了4种情形,结果表明,本文提出的方法性能优于同类方法,对于一个房间内的大部分地方,定位误差在35 mm以内。

关键词: 室内定位, 单相机, 神经网络估计, 投影不变性, 定位误差

Abstract: Estimating the user's location with a camera requires the establishment of a complex mathematical model because the relationship between 3D scene and a 2D image is non-linear and highly complex. In order to solve this problem,a single camera based on neural network is proposed.The main advantage of the indoor positioning system is that LED can use the visible light communication to send its location information.The proposed method makes full use of LED light projection invariance,and completes the construction of virtual line by means of imaging sensor communication(ISC).Then,neural network estimation is used to extract information from the virtual camera direction lines.Finally,a simple mathematical equation is adopted to estimate the position of the user indoor.Four situations have been considered in the simulation experiments and the results show that the proposed method outperformed state-of-art techniques and its location error is less than 35 mm for most of the space within a room.

Key words: indoor positioning, single camera, neural network estimation, projection invariance, positioning error

中图分类号: