row1
Abstract
A deep Gaussian process (DGP) model is trained to create indoor radio maps with confidence intervals, which are referred to as uncertainty maps. Utilizing the uncertainty maps, an LSTM-based location prediction model is pretrained with artificial trajectory data sampled from the uncertainty maps, and then fine-tuned with the signal measurements collected in the field. In the training process, auxiliary outputs are implemented to overcome overfitting and improve the robustness of the system. Our extensive experiments demonstrate the outstanding performance of the proposed MapLoc system.
Summary
This article presents MapLoc, a system which maps out indoor environments. More specifically, it’s an indoor fingerprinting system that utilizes a Deep Gaussian Process (DGP) to regress uncertainty maps and incorporates a long short-term memory (LSTM)-based method for location estimation.
The process is as follows: A phone which can collect magnetic field intensity data and Wi-Fi RSS data simultaneously (using a proprietary app) is set at multiple random places within a location to capture data. The system first feeds this data into the DGP. This allows it to create 2D overhead maps of the environment in which it’s tested, called Uncertainty Maps.
During testing, these 2D maps are compared against the location’s floorplan to estimate accuracy. The location-prediction model is trained from the sequences generated by the DGP and fine-tuned by signal measurements from the field. The LSTM network is then fed the fine-tuned data from the DGP to create an accurate estimated map of the environment.
