Convolutional Neural Network as sensor fusion algorithm applied to IPIN2019 dataset

Abstract

This work-in-progress explores the use of Convolutional Neural Net-works (CNNs) in a sensor fusion approach for indoor localization, a crucial component in computer vision, robotics, and navigation. CNNs have emerged in the last few years as sensor fusion algorithms, using deep learning to process multi-sensor data. We apply CNN to the IPIN 2019 competition dataset. The method consists of processing sensor data and transforming it into images, and training a CNN model for position estimation. Preliminary results show promise in specific scenarios, but the CNN approach struggles with generalization on diverse tracks.