融合角度特征的CNN可见光室内定位算法  

CNN visible light indoor positioning algorithm fused with angle features

作　　者：杨薇[1] 邵建华[1,2] 杜聪 YANG Wei;SHAO Jianhua;DU Cong(School of Computer and Electronic Information,Nanjing Normal University,Nanjing 210023,China;Key Laboratory of Optoelectronics of Jiangsu Province,Nanjing 210023,China)

机构地区：[1]南京师范大学计算机与电子信息学院,南京210023 [2]江苏省光电重点实验室,南京210023

出　　处：《激光杂志》2022年第1期113-118,共6页Laser Journal

基　　金：教育部博士点基金(No.2013102SBJ0265)。

摘　　要：针对可见光室内定位精度不高的问题,提出一种融合角度特征的卷积神经网络(Convolutional Neural Network, CNN)可见光室内定位算法。首先在基于接收信号强度(Received signal strength, RSS)的基础上,考虑到LED(Lighting Emitting Diode, LED)辐射角度对接收端PD(Photo Detector, PD)接收到的信号强度的影响,融合发送端和接收端之间的角度特征信息建立空间三角模型,构建基于角度特征值的位置指纹库,接着利用卷积神经网络模型,训练定位模型,预测待定位目标的位置,然后通过指纹库中的已知指纹点,计算定位误差,得到高精度的室内定位误差。最后在4 m×4 m×2.5 m实验空间中验证该算法,得到平均定位误差为4.16 cm,且定位误差累积分布在4.5 cm以内的概率为80%,在8 cm以内的概率为90%,定位误差稳定。Aiming at the problem of low indoor positioning accuracy in visible light, a convolutional neural network(CNN) visible light indoor positioning algorithm fused with angle features is proposed. First, based on the received signal strength(RSS), considering the influence of the LED radiation angle on the signal strength received by the receiving end photo detector(PD), integrate the angle feature information between the sender and the receiver to establish a spatial triangle model, and build a location fingerprint database based on the angle feature value, then the convolutional neural network model is used to train the positioning model to predict the location of the target to be located, and then through the known fingerprint points in the fingerprint database, the positioning error is calculated to obtain a high-precision indoor positioning error. Finally, the algorithm is verified in an experimental space of 4 m×4 m×2.5 m, and the average positioning error is 4.16 cm, and the probability of the cumulative distribution of positioning error within 4.5 cm is 80%, and the probability within 8 cm is 90%. The error is stable.

关 键 词：室内定位 位置指纹库 卷积神经网络 平均定位误差 

分 类 号：TN209[电子电信—物理电子学]