| Etablissement | Université 8 mai 1945 de Guelma | | Affiliation | Département d'Electrotechnique et automatique | | Auteur | Zentar, Mohamed Dhia El Hak | | Directeur de thèse | Bencheriet Chemesse Ennehar (Maitre de conférence) | | Filière | Automatique | | Diplôme | Doctorat LMD | | Titre | Vehicle and Pedestrian Detection Based on Deep VIT | | Mots clés | Vehicle detection, pedestrian detection, Deep learning, Convolutional Neural Network, Vision Transformer | | Résumé | Road accidents cause thousands of victims each year. The alarming increase in this number has prompted researchers to develop applications to help the driver detect obstacles, road signs, or vehicles in suspicious conditions that can cause road accidents.
The main causes of these accidents are not only excessive speed by vehicles but also animals or
pedestrians crossing the road. In 2021, on the east-west highway of Ain Defla, a terrifying accident caused by a broken-down motorist hit by a car while trying to cross the road caused a pile-up of 34
vehicles, killing at least five and injuring 12.
Driving assistance systems seem to be a good solution for limiting road accidents. We seek to develop an intelligent driving assistance system whose main objective is detecting vehicles and pedestrians in road areas to alert the driver in time to take safety precautions (slowing down, braking, etc.). The application is based on deep learning of Visual Transformers combined with
Convolutional Networks (CNN). Usually used in NLP (Natural Language Processing), visual processors have demonstrated their robustness in artificial vision with performances exceeding CNN's | | Statut | Validé |
|