A computer vision system for detecting motorcycle violations in pedestrian zones
This paper presents a system that relies on computer vision to identify instances of motorcycle violations in crosswalks utilizing CNNs. The system was trained and evaluated on a novel public dataset published by the authors, which contains traffic images classified into four categories: motorcycles...
- Autores:
-
Hernández-Díaz, Nicolás
Peñaloza, Yersica C.
Rios, Y. Yuliana
Martínez-Santos, Juan Carlos
Puertas, Edwin
- Tipo de recurso:
- Fecha de publicación:
- 2024
- Institución:
- Universidad Tecnológica de Bolívar
- Repositorio:
- Repositorio Institucional UTB
- Idioma:
- eng
- OAI Identifier:
- oai:repositorio.utb.edu.co:20.500.12585/12684
- Acceso en línea:
- https://hdl.handle.net/20.500.12585/12684
https://doi.org/10.1007/s11042-024-19356-9
- Palabra clave:
- IA model
Computer vision
Machine learning
Pedestrian areas
Autonomous traffic control system
LEMB
- Rights
- openAccess
- License
- http://creativecommons.org/licenses/by-nc-nd/4.0/
Summary: | This paper presents a system that relies on computer vision to identify instances of motorcycle violations in crosswalks utilizing CNNs. The system was trained and evaluated on a novel public dataset published by the authors, which contains traffic images classified into four categories: motorcycles in crosswalks, motorcycles outside crosswalks, pedestrians in cross walks, and only motorbike outside. We demonstrate the viability of leveraging deep learning models such as YOLOv8 for this purpose and provide details on the training and performance of the model. This system has the potential to enable intelligent traffic enforcement to mit igate accidents in pedestrian zones; to develop the system, a dataset comprising over 6,000 images was amassed from publicly available traffic cameras and subsequently annotated. Several models, including YOLOv8, SSD, and MobileNet, were trained on this dataset. The YOLOv8 model attained the highest performance with a mean average precision of 84.6% across classes. The study presents the system architecture and training process. Results illus trate the potential of utilizing deep learning to detect traffic violations in pedestrian zones, which can promote intelligent traffic enforcement and improved safety. |
---|