Etude comparative sur les algorithmes de détection et de reconnaissance de véhicules dans une séquence d’images

Abstract

Road and highway safety has become a priority issue for public authorities due to the massive increase in road accidents every year. To this effect, the detection of vehicles perceived in a video has simplified the issue of traffic and the identification of suspect cars or vehicles. Deep learning techniques have emerged as a powerful strategy offering an alternative solution to human vision imitation. These techniques are based on deep learning. This thesis falls within the domain of computer vision. Our goal is to detect vehicles in a stream of images and classify them according to their type 'bus, car, motorcycle, truck, bicycle'. For this purpose, we adopted the two models YOLOv5 and Faster RCNN for vehicle detection. After a comparative study between these models, the best results were obtained by the Faster RCNN method with an accuracy value of 89%.