Rapid urban expansion and rising vehicle numbers in metropolitan areas have intensified traffic congestion and increased the risk of road traffic accidents. However, traditional traffic data collection methods remain limited in spatial coverage and temporal resolution, reducing their effectiveness for traffic management and road safety planning. Closed-circuit television (CCTV) systems continuously capture traffic conditions, yet their analytical potential has not been fully utilized. Advances in deep learning, particularly the You Only Look Once (YOLO) algorithm, provide strong capabilities for real-time object detection and classification from image data. This study aimed to evaluate the performance of YOLOv5m, YOLOv8m, and YOLOv12m in detecting and classifying vehicle types from CCTV video data collected from three major bridges in Bangkok, and to analyze traffic volume by time period and travel direction. The results showed that YOLOv12m demonstrated competitive overall detection performance compared with YOLOv5m and YOLOv8m. Temporal analysis revealed higher inbound traffic in the morning and greater outbound traffic in the evening, reflecting commuting patterns. In addition, the model supported traffic volume analysis across selected bridge corridors. These findings highlight the potential of integrating CCTV and YOLO for automated traffic monitoring, traffic volume analysis, road safety support, and smart transportation planning in Bangkok.
Worachairungreung et al. (Tue,) studied this question.