A deep learning approach to fingerprinting indoor localization solutions

Fingerprinting Localization Solutions (FPSs) enjoy huge popularity due to their good performance and minimal environment information requirement. Considered as a data-driven approach, many modern data analytics can be used to improve its performance. In this paper, we propose tow learning algorithms, namely a deep learning architecture for regression and Support Vector Machine (SVM) for classification, to output the estimated location directly from the measured fingerprints. The design issues of the proposed neural network is discussed including the training algorithm, regularization and hyperparameter selection. It is discussed how data augmentation methods can be utilized to extend the measurements. The deep learning approach can be used to save the data collection time significantly using a pre-trained model. Moreover the run-time complexity is significantly reduced. The numerical analysis show that in some case, only 10 percent of original training database is enough to get acceptable performance on a pre-trained model.