ARCHITECTURE OF FOUR-ARM INTERSECTION PERFORMANCE EVALUATION SYSTEM BASED ON TURN MOVEMENT COUNT AND OMNIDIRECTIONAL CAMERA

• Main Author: Agil Alunjati, Figo
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87631
• Institution: Institut Teknologi Bandung

Traffic management involves the systematic regulation of the movement of people and vehicles to enhance safety, efficiency, and mobility. Area Traffic Control Systems (ATCS), as a component of traffic management, are designed to regulate traffic flow at intersections by coordinating vehicle movements across different arms and reducing potential conflicts. Intersections represent critical points of convergence within the road network. Previous research has developed many methods for controlling ATCS. However, there are still few studies that focus on using omnidirectional cameras and evaluating their accuracy and processing speed. Hence, this research focuses on developing a robust vehicle detection methodology that balances high accuracy and high frame rate using an omnidirectional ATCS camera. This research uses a case study approach at the Sedayu intersection in Yogyakarta, collecting data with omnidirectional CCTV cameras. The data, including vehicle turning movements, was analyzed using video analytics over two days: a weekday and a weekend. This study evaluates the robustness of various single-stage model architectures, focusing on both accuracy metrics (confidence level, mAP, precision, F1-Score, Recall) and processing speed (FPS, processing time, GPU memory, GPU load, GPU temperature). The objective is to identify a model that balances high accuracy with real-time processing capabilities. The results indicate that YOLOv11, with a batch size of 32, demonstrates robust real-time performance while maintaining reasonable accuracy. However, the findings also reveal consistent suboptimal performance, with Level of Service ratings of E and F, underscoring the need for strategic interventions in traffic signal control and lane configuration.