Deep Learning-Based Vehicle Speed Estimation in Bidirectional Traffic Lanes

Abstract

Accurate vehicle speed estimation is critical for efficient traffic management and safety, particularly in areas with complex traffic patterns such as bidirectional lanes. This study proposes a deep learning-based system utilizing the YOLOv8 model to estimate vehicle speeds in bidirectional traffic. By leveraging existing camera infrastructure and advanced image processing techniques, the proposed system focuses on regions of interest (ROI) for more accurate speed calculation. Three YOLOv8 model variants—Nano, Small, and Medium—are evaluated, with YOLOv8 Medium achieving a mean Average Precision (mAP) of 93.6%. The results demonstrate the potential of YOLOv8 for improving real-time object detection and speed estimation, with future integration of additional sensor modalities, such as lidar and radar, paving the way for more robust intelligent transportation systems.