Real-Time Vehicle Classification and Directional Counting at Urban Roads Using YOLOv11 Under Heterogeneous Traffic Conditions
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Urban roads in developing countries face highly heterogeneous traffic, making manual data collection labour-intensive and error-prone. This study presents a deep learning-based approach for real-time vehicle classification and directional counting using the YOLOv11 framework. A custom dataset of 5,200 locally collected images, encompassing 11 vehicle classes defined by the Road Development Authority of Sri Lanka, was used to train two model variants: YOLOv11 medium and YOLOv11 small. To enable directional flow estimation, the system implements a line-pass-based tracking algorithm that distinguishes between up- and down-traffic streams. The system demonstrated high accuracy for common classes, such as cars (Upstream 98.8% and Downstream 96.8%), motorcycles (Upstream 88.8% and Downstream 97.6%), and three-wheelers (Upstream 85.8% and Downstream 98.4%). However, performance variations in other categories highlight the challenges of occlusion and visual ambiguity in mixed-traffic environments. The findings demonstrate that YOLOv11 models trained on locally representative datasets provide a reliable, scalable, and cost-effective alternative to manual survey methods. This approach provides a disaggregated data source to support traffic signal optimisation, intersection analysis, and urban transport planning under complex heterogeneous traffic conditions.
Fields of Study & Subjects
Advanced Neural Network Applications
Video Surveillance and Tracking Methods
Traffic control and management
Publication details
- Journal / source
- Journal of South Asian Logistics and Transport
- Volume
- 6
- Issue
- 1
- Pages
- 23–40
- Publication date
- 31 March 2026
- Language
- EN