Location Estimation Through Inexact Machine Learning Approach

[摘要] Location estimation has become a field of increasing interest in recent years. The main reason is the multiple applications that can be enabled based on this technology. Fields such as entertainment, health care, tourism and advertisement are some of the areas where a plethora of applications can be implemented. In outdoors this problem is solved, for most of the cases, with Global Navigation Systems (GNSS). However, in indoors is a current topic of interest that has been addressed from different perspectives with different technologies. Nonetheless, there is no technology that is as established as GNSS is for outdoors. One promising approach is Inertial Measurement Units (IMU) which are low cost and widely accessible in multiple SmartDevices such SmartPhones, SmartWatches, WristBands, among others. Two of the main difficulties that hinder the wide adoption of this technology are the error accumulation between estimations and the scarce availability of the Ground Truth data to train and test the models. In this work both challenges are addressed by two methods, one which corrects the error by using the structure of the map where the user is located and the other method improves the Ground Truth data provided by GNSS measurements. Energy consumption is reduced by a factor 27x when compared with GPS and the accuracy of the labels is improved by 26% on average.

[发布机构] Rice University

[效力级别] estimation