融合多元传播模型和指纹模型的免标定室内定位方法研究

项目名称： 融合多元传播模型和指纹模型的免标定室内定位方法研究

项目编号： No.61472399

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 王双全

作者单位： 中国科学院计算技术研究所

项目金额： 75万元

中文摘要： 室内定位在健康监护、公共安全、位置服务等领域具有重要价值，但常用的指纹模型需要人工采集大量标定数据，成为室内定位技术难以广泛应用的核心问题。本项目拟融合多元信号传播模型和指纹模型，利用非标定的用户轨迹数据，结合多重时空约束条件，研究免标定的室内定位方法，实现模型的高精度和通用性。研究内容包括：1）基于环境因素的多元信号传播模型。综合考虑距离衰减、墙壁遮挡、多径效应和人体遮挡等因素，构建基于环境因素的多元信号传播模型，提升模型精度和通用性；2）融合时空约束的位置估计模型。根据上述传播模型，结合贝叶斯层次图模型和最大似然估计算法求解粗略位置，利用空间约束的粒子滤波算法和核PCA降维算法对估计位置进行重映射，提高位置估计的精度；3）融合时空约束的增量式半监督指纹定位模型。根据位置估计，结合空间、轨迹、行为和标记点等约束条件，构建基于用户轨迹的HMM指纹定位模型，利用EM算法实现模型的增量更新。

中文关键词： 室内定位；免标定；多元传播模型；指纹模型；时空约束

英文摘要： Indoor localization is very important for the application development in the fields of health care, public security, location based services (LBS), etc. However, the popular fingerprint model requires much human labeling cost, which is the key problem for the wide use of indoor localization systems. This project fuses multivariate radio propagation model and fingerprint model to develop a calibration-free indoor localization method which has high localization accuracy and validity in various indoor environments, using the unlabeled user trace data and spatio-temporal constraints. The main research topics include: 1) environmental parameters based multivariate signal propagation model. Investigate the influences of path loss, wall attenuation, multi-path effects and human body attenuation, and build up new environmental parameters based multivariate signal propagation model, which can obtain higher localization accuracy and validity in different environments; 2) spatio-temporal constraints based location estimation model. The user's locations can be estimated by combining Bayesian hierarchical graphical model and maximum likelihood (ML) estimation method. Then, the location estimation results will be filtered and re-mapped using particle filter and kernel principal component analysis (KPCA) algorithm, respectively; 3) spatio-temporal constraints based semi-supervised and incremental fingerprint localization model. Using the location estimates and multiple constraints (spatial, trace, motion and landmark constraints), hidden Markov model (HMM) based semi-supervised fingerprint model is developed, which can be online updated using expectation maximization (EM) algorithm to further improve its localization accuracy.

英文关键词： indoor localization;calibration-free;multivariate radio propagation model;fingerprint model;spatio-temporal constraints