Automatic extraction of relevant road infrastructure using connected vehicle data and deep learning model
Adu-Gyamfi, Kojo Konadu
Is Version Of
Civil, Construction, and Environmental Engineering
This thesis presents a novel approach for extracting road infrastructure information from connected vehicle trajectory data, employing geohashing and image classification techniques. The methodology involves segmenting trajectories using geohash boxes and generating image representations of road segments. These images are then processed using YOLOv5 to accurately classify straight roads and intersections. Experimental results demonstrate a high level of accuracy, with an overall classification accuracy of 95%. Straight roads achieve a 97% F1 score, while intersections achieve a F1 score of 90%. These results validate the effectiveness of the proposed approach in accurately identifying and classifying road segments. The integration of geohashing and image classification techniques offers numerous benefits for road network analysis, traffic management, and autonomous vehicle navigation systems. By extracting road infrastructure information from connected vehicle data, a comprehensive understanding of road networks is achieved, facilitating optimization of traffic flow and infrastructure maintenance. The scalability and adaptability of the approach make it well-suited for large-scale datasets and urban areas. The combination of geohashing and image classification provides a robust framework for extracting valuable insights from connected vehicle data, thereby contributing to the advancement of smart transportation systems. The results emphasize the potential of the proposed approach in enhancing road network analysis, traffic management, and autonomous vehicle navigation, thereby expanding the knowledge in this field and inspiring further research.