Assessing semantic consistency of pavement markings and signs using street view images – a case study on lane-turning information

Road asset management (RAM) is crucial in road construction and maintenance. Previous efforts have focused on the digitization of the physical state of road facilities, such as location and condition. However, the semantic information conveyed by these facilities, such as instructions, controls, and warnings, and the consistency of semantic information across multiple facilities has been neglected. Inconsistent semantic information can confuse road users, disrupt traffic, and endanger lives. To address this critical problem, this study proposes the concept of ‘semantic space’ for road facilities and presents a comprehensive framework that combines street view images with deep learning techniques to detect, localize, and analyze semantic consistency with this space, specifically focusing on lane-turning information. To validate the effectiveness of our framework, we conducted experiments on 81 km of urban roads in Nanjing, Jiangsu, China. The experimental results show that our method has an overall precision of 77.6% and an overall recall of 94.2% for detecting defined semantic inconsistency errors. While this study focuses on lane-turning information, the proposed framework for semantic space detection and assessment shows promise in analyzing inconsistencies in other road semantic information conveyed by diverse and discrete road facilities, contributing to an enhanced RAM.