Accurate Image Geo-Localization using Smartphones

To improve traveller services in train stations, SNCF and Keolis Canada seek to develop a mobile application for real-time geo-localisation based on still images captured by the user on its smartphone. This would allow guidance through the station and provide key information. This project will develop a Deep Learning (DL) model for accurate and real-time image-based geo-localisation, relying on novel computer vision and trained by image datasets previously collected and anonymized by SNCF. The DL model will allow for context-aware recognition, and for progressive pruning and knowledge distillation of deep networks. A mobile application compatible with Android and iOS operating systems, providing 2D classical visualisation, will also be developed with the constraints of limited processing resources (memory complexity and energy consumption). This cross-disciplinary project will train highly qualified personnel to face challenges in areas of strategic interest. It will intensify the exchange of ideas around new DL models in visual recognition applications, using large and unlabeled datasets. Findings will be disseminated in high caliber scientific journals and exploited in SNCF and Keolis applications.

Faculty Supervisor:

Éric Granger

Student:

Partner:

Keolis Canada;SNCF

Discipline:

Computer science

Sector:

Transportation and warehousing

University:

École de technologie supérieure

Program: