Development of Road Asset Mapping using Dashcam

Abstract

Road asset mapping significantly benefits transportation authorities, infrastructure management, and road users. Recent advancements in Geographic Information Systems (GIS) and digital mapping technologies have substantially improved the effectiveness of inventory and asset management. However, the current technology, including Light Detection and Ranging (LiDAR) and mobile mapping cameras, is relatively expensive compared to dashcams. Therefore, this study aims to identify various road assets using You Only Look Once version 8 (YOLOv8) and map them using GIS. The data for this study was collected from a car’s dashcam. Images were extracted from the video and displayed in multiple frames using VLC Media Player. Subsequently, bounding boxes and labeling were used in this dataset for image annotations with the help of Roboflow software. The dataset was then divided into train, validation, and test sets before it could be used for modeling and model evaluation. YOLOv8 was chosen as the model for this study due to its reputation for improving image recognition, segmentation, detection, and retrieval by comprehensively understanding the details within an image. The model’s precision, recall, and mean Average Precision (mAP) at 50% Intersection over Union (IoU) values are 0.895, 0.873, and 0.876, respectively. Hence, gathering more high-quality datasets to enhance the study and achieve higher accuracy for a road asset mapping system is recommended for future studies. This research demonstrates that the system can effectively identify and map various road assets, including road signs, streetlights, and traffic lights, using data collected by dashcams, thereby establishing a low-cost road asset mapping system.