Constellation, le dépôt institutionnel de l'Université du Québec à Chicoutimi

Résumé

A fully autonomous car does not yet exist. But the vehicles have continued to gain in range in recent years. The main reason? The dazzling progress made in artificial intelligence, in particular by specific algorithms, known as machine learning. These example-based machine learning methods are used in particular for recognizing objects in photos. The algorithms developed for the detection and identification must respond robustly to the various disturbances observed and take into account the variability in the signs’ appearance. Variations in illumination generate changes in apparent color, shadows, reflections, or backlighting. Besides, geometric distortions or rotations may appear depending on the viewing angle and the panels’ scale. Their appearance may also vary depending on their state of wear and possible dirt, damage. In this work, to improve the accuracy of detection and classification of sign road partially covered by snow, we use the Fast Region-based Convolutional Network method (Fast R-CNN) model. To train the detection model, we collect an image dataset composed of multi-class of road signs. Our model can simultaneously multi-class of a road sign in nearly real-time.