Navigating indoor spaces using machine learning: Train stations in Paris

This chapter presents the project of navigating indoor spaces using machine learning: train stations in Paris. Legibility is particularly important in indoor spaces primarily used for commuting, such as train stations and airports. As the ability of artificial neural networks to analyse visual information is getting to a human-level performance, the chapter proposes a classification problem using a deep convolutional neural network (DCNN) to proxy the modelling of legibility of indoor spaces using photographic images as input. A DCNN is a popular architecture widely applied in interpreting images. The performance of a DCNN is achieved by a bank of filters whose weights are learnt during the training deployed to extract features from images. It relies on high-dimensional descriptors, considering the interplay between aspects of images, instead of taking only one or two aspects into consideration.