Defense of the thesis "Deep learning for continuous prediction of parking occupancy in an urban environment".

Jury

Mrs Florence Sèdes, Professor, Univ. Paul Sabatier, Toulouse 3, France, Rapporteur
M. Nicolas Saunier, Professeur, CIRRELT, Polytechnique Montréal, Canada, Rapporteur
Mr. Marian Scuturici, Professor, INSA Lyon, France, Examiner
Mr. Karine Zeitouni, Professor, CIRRELT, Polytechnique Montreal, Canada, Examiner Mrs. Karine Zeitouni, Professor, Université de Versailles Saint-Quentin-en-Yvelines, France, Examiner
Mr. Thierry Delot, Professor, INSA Lyon, France Mr. Thierry Delot, Professor, LAMIH, UPHF, France, Thesis co-supervisor
Mr. Martin Trépanier, Professor, LAMIH, UPHF, France, Thesis co-supervisor Mr Martin Trépanier, Professor, CIRRELT, Polytechnique Montréal, Canada, Thesis co-director
M. Abdessamad Ait EL Cadi, Professeur, LAMIH, UPHF/INSA, Co-Encadrant

Abstract:

Deep learning has been widely adopted in various fields due to its ability to extract complex features from large amounts of data. In this thesis, we propose a deep learning-based approach for the continuous prediction of parking lot occupancy. To this end, we have collected a large dataset on parking lot occupancy (for both covered and street-side parking lots) from different cities in France and Canada . Our experiments show that the proposed approach outperforms conventional and machine learning-based models in terms of forecast accuracy and real-time performance. What's more, our approach can also be easily integrated into existing intelligent parking systems to improve their efficiency and convenience. For city-wide deployment, we also propose a framework for sharing models between different parking lots by analyzing their spatial and temporal similarity. By identifying the relevant spatial and temporal characteristics of each parking lot (parking profile) and grouping them accordingly, our approach enables the development of accurate occupancy prediction models for a set of parking lots, reducing deployment costs and improving model transferability. Our experiments demonstrate the effectiveness of the proposed strategy in terms of reducing model deployment costs while maintaining good forecast quality. In conclusion, this work demonstrates the effectiveness of deep learning in solving the problem of continuous parking lot occupancy forecasting and highlights its potential for future smart parking applications.

Key words:

Parking occupancy prediction, parking profile, Spatiotemporal clustering, deep learning, machine learning, model sharing