马永涛, 刘康伟, 高鑫
AuthorsHTML:马永涛, 刘康伟, 高鑫
AuthorsListE:Ma Yongtao, Liu Kangwei, Gao Xin
AuthorsHTMLE:Ma Yongtao, Liu Kangwei, Gao Xin
Unit:天津大学微电子学院，天津 300072
Unit_EngLish:School of Microelectronics, Tianjin University, Tianjin 300072, China
Abstract_Chinese:针对指纹法定位中的指纹库构建耗费大量人力和时间的问题, 提出了一种基于群智感知和非监督式学习的室内定位指纹库构建算法．首先利用群智感知的方式采集无线接入点的接收信号强度获取原始指纹数据集．然后针对原始指纹数据的冗余杂乱和指纹注释问题, 提出基于学习向量量化(LVQ)和多维标度(MDS)结合的一种新颖算法FLM来解决．最终有效构建室内定位指纹库．最后基于射线跟踪模型(ray-tracing)建立仿真实验场景, 仿真结果表明其指纹库建立效率得到显著提高, 应用于基本定位算法的定位误差80% 在2.6 m以下, 而且单次定位的计算量下降63% ．
Abstract_English:Aiming at solving the problem of massive manpower and time consumption in fingerprint database construction for fingerprint localization，an algorithm for indoor positioning fingerprint database construction based on crowdsensing and unsupervised learning was proposed in this paper. Firstly，the received signal strength was acquired from access points by crowdsensing，which is the original fingerprint data set. Then，to deal with the problem of data redundancy and fingerprint annotation of the original fingerprint data，a FLM-algorithm based on the combination of learning vector quantization(LVQ)and multidimensional scaling(MDS)was proposed. Finally，the construction of indoor positioning fingerprint database was effectively achieved. Based on ray-tracing，a simulation scenarios of the proposed algorithm was established. Simulation results show that the efficiency of the fingerprint database construction is improved greatly. 80% of localization error is lower than 2.6 m by appling to the basic localization algorithm，and the amount of localization calculation of the single position is reduced by 63% .
Keyword_Chinese:群智感知; 指纹库; 原型指纹; 接收信号强度; 非监督式学习
Keywords_English:crowdsensing; fingerprint database; prototype fingerprint; received signal strength; unsupervised learning

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6096