Vision-Based Vehicle Classification for Smart City

Abstrak

Vehicle detection systems are essential for improving traffic management, enhancing safety, supporting law enforcement, facilitating toll collection, and contributing to smart city initiatives through real-time monitoring and data analysis. With the rapid growth of smart city technologies, the need for efficient, scalable, and high-accuracy vehicle detection models has become increasingly critical. This study aims to propose an advanced vehicle detection system using Convolutional Neural Networks (CNNs) in combination with the YOLOv5 model, which is known for its high-speed performance and superior accuracy in image recognition tasks. The proposed model is evaluated using a custom-trained YOLOv5s model, tested on a dataset comprising 1460 images of vehicles. These images are divided into five classes which are cars, motorcycles, trucks, ambulances, and buses. Performance evaluation metrics such as precision, recall, and mean Average Precision (mAP50-95) are used to assess the model's effectiveness. The results indicate that the YOLOv5-based model achieved impressive detection accuracy, with precision, recall, and mAP values exceeding 87%. The proposed system demonstrates its robustness in detecting and classifying various vehicle types across different conditions, including small, partially visible, and distant vehicles. The findings suggest that this model holds significant potential for real-world applications in urban traffic management and smart city infrastructure.

Kata kunci

• Vehicle Classification
• Image Recognition
• Vision-Based
• Smart City