Automatic Training and Radiomap Collection for Indoor Location Estimation

Mapsted是一家加拿大技术公司，专门从事室内定位服务。他们开发了一种核心定位技术，可以提供实时的室内定位。该项目的目标是开发 ** 进一步增强其核心本地化技术的特定模块。更具体地说，该项目将 ** 开发用于室内定位的系统校准自动构建的新算法。我们将使用图论来构建无线电地图，其中包含WiFi接收信号强度（RSS）和磁场（MF）的无监督测量。当下载应用程序的用户访问感兴趣的区域时，他/她的手机从可用的接入点捕获WiFi**RSS以及从手机磁力计获得的MF，并传输到 ** 中央服务器。这些数据没有任何附加到测量的RSS的位置。然而，RSS值，** 由用户捕获，也自然地受到环境的几何形状和结构限制的约束。在实践中，不同的RSS测量之间存在一些相关性。例如，在彼此附近收集的 **RSS值可能会与 ** 类似的RSS值共享大量可见的接入点。在这个项目中，我们将使用RSS测量对之间的相关性来移植痕迹 ** 以构建数据图，这实际上是RSS值的排列，其结构化以创建RSS测量的新 ** 图形表示。然后，我们将提出算法来将数据图 ** 与环境的平面图重叠。

Mapsted is a Canadian Technology company which specializes in indoor positioning services. They have developed**a core localization technology which provides real-time indoor positioning. The goal of this project is to develop**further enhancements to a specific module of their core localization technology. More specifically, this project will**develop new algorithms for automatic construction of system calibration for indoor localization. We will use graph**theory to build a radio map with unsupervised measurements of WiFi received signal strength (RSS) and magnetic**field (MF). When a user with the downloaded application visits an area of interest, his/her phone captures WiFi**RSS from available access points as well as the MF obtained from the phones magnetometer, and transmits to a**central server. Such data do not have any location attached to the measured RSS. However, the RSS values,**captured by the user, are also naturally constrained by the geometry of the environment and structural**limitations. In practice, there exist some correlation among the different RSS measurements. For example, the**RSS values collected in proximity of each other will likely share a large number of visible access points with**similar RSS values. In this project, we will use the correlation between pairs of RSS measurements to graft traces**to build a data graph, which is indeed an arrangement of RSS values that are structured to create a new**graphical representation of the RSS measurements. We will then propose algorithms to overlap the data graph**with the floor plan of the environment.