Calibrated and automated recognition of barcodes in complex, highly variable scenes

The main objective of this project is to investigate, develop and evaluate state-of-the-art artificial-intelligence (AI) algorithms for the automatic recognition of extremely diversified barcode instances, in complex, highly variable and non-controlled scenes. The application area is of high interest to the industrial partner, and the ultimate goal is to build a state-of-the-art, automated and well-calibrated barcode analysis system, which meets challenging industrial expectations of high accuracy and computational efficiency, while providing measures of confidence of the predictions (to mitigate the risks associated with potential errors). The recent years have witnessed substantial technical advances in deep learning (e.g., convolutional neural networks), which promises to address such complex image analysis problems. The project will involve one intern (a PhD student), whose objective is to adapt, extend and develop state-of-the-art convolutional neural networks for training automatic recognition models, while accounting for specific industrial challenges: high-accuracy requirements, the need for effective uncertainty estimation to mitigate the high cost of errors, and the need for active learning and exploring the diversity of large-scale unlabeled data. These model-calibration, semi-supervision and active-learning problems are of great interest to the industrial partner and to the computer vision community in general.

Faculty Supervisor:

Ismail Ben Ayed

Student:

Partner:

Systemes Electroniques Matrox Ltee

Discipline:

Engineering

Sector:

Manufacturing

University:

École de technologie supérieure

Program: