Skip to Main content Skip to Navigation
Journal articles

Génération de cartes tactiles photoréalistes pour personnes déficientes visuelles par apprentissage profond

Abstract : Photo-realistic tactile maps are one of the tools used by visually impaired people to understand their immediate urban environment, particularly in the context of mobility, for crossing crossroads for example. These maps are nowadays mainly hand-made. In this article, we propose an approach to produce a semantic segmentation of precision aerial imagery, a central step in this manufacturing process. The different elements of interest such as sidewalks, pedestrian crossings, or central islands are thus located and traced in the urban space. We present in particular how the augmentation of this imagery by vector data from OpenStreetMap leads to significant results using a deep learning technique (conditional generative adversarial network). After presenting the stakes of this work and a state of the art of existing techniques, we detail the proposed approach, and we study the results obtained, in particular by comparing the segmentations obtained without and with enrichment by vector data. The results are very promising.
Document type :
Journal articles
Complete list of metadata

https://hal.uca.fr/hal-03208233
Contributor : Jean-Marie Favreau Connect in order to contact the contributor
Submitted on : Monday, April 26, 2021 - 1:55:20 PM
Last modification on : Sunday, June 26, 2022 - 3:07:41 AM
Long-term archiving on: : Tuesday, July 27, 2021 - 7:03:44 PM

File

Post_print_gauthier_RIG.pdf
Files produced by the author(s)

Identifiers

Citation

Gauthier Fillières-Riveau, Jean-Marie Favreau, Vincent Barra, Guillaume Touya. Génération de cartes tactiles photoréalistes pour personnes déficientes visuelles par apprentissage profond. Revue Internationale de Géomatique, Lavoisier, 2020, 30 (1-2), pp.105-126. ⟨10.3166/rig.2020.00104⟩. ⟨hal-03208233⟩

Share

Metrics

Record views

133

Files downloads

142