Automated Vehicle Classification (AVC) using machine learning implementation in Malaysia's toll system

Congestion at Malaysian toll plazas persists due to manual toll rate settings at multiclass lanes, leading to errors and inefficient traffic flow, causing significant economic losses during peak hours. Thus, this project aims to develop the best model detector for an automated vehicle classification system using computer vision and machine learning algorithms to enhance toll collection efficiency. The methodology involves understanding the business context, acquiring 1,735 images spanning seven vehicle classes, modeling, user evaluation, and deployment using Streamlit and MySQL. Model training utilizes YOLOv8, YOLO-NAS, and Faster R-CNN, with evaluation metrics such as Mean Average Precision (MAP), precision, and others. Key materials include OpenCV, Ultralytics, TensorFlow 2.0, and others. YOLOv8 exhibits superior performance with the highest MAP of 0.995 after fine-tuning compared to other models, demonstrating effectiveness in real-time object detection. The system employs a single detection process, ensuring only one vehicle is detected at a time, enhancing accuracy. The project contributes to the accomplishment of Sustainable Development Goals (SDG), including SDG 11, SDG 9, and SDG 15, supporting sustainable mobility practices. Future enhancements may involve multi- sensor fusion and axle detectors for improved accuracy.