Neural Network Evaluation for Magnetic Fingerprinting Positioning

Fingerprinting indoor positioning enables providing location information in GPS non-coverage areas. This problem is typically tackled through machine learning models due to their pattern-matching capabilities. Magnetic field is a promising alternative to commonly used radio-frequency signals for fingerprinting indoor positioning. In this work, we study different neural network architectures to evaluate magnetic fingerprinting indoor positioning. We use a dataset with different phones and varied regions to study the influence of these factors in the positioning results. Amongst the tested architectures the shallow multilayer perceptron resulted in the lowest positioning error across all regions and phones, closely followed by the convolutional neural network, and phone’s antennas resulted in noticeable differences in the positioning error.