講演情報

[NpVBMPAo-07]Multimodal AI–Based Pavement Inspection with an Action Camera: Wide-Area Deployment Impact Assessment in Hokkaido, Japan

*Yuri Konno1, Takumi Asada1, Hiroyuki Oikawa3, Shuichi Kameyama2 (1. Muroran Institute of Technology (Japan), 2. Hokkaido University of Science (Japan), 3. Docon Co., Ltd. (Japan))

キーワード:

Deep learning、Machine learning、Cost-effectiveness、Network-level、work saving

We have developed a low-cost, easy-to-implement pavement inspection technology that combines an action camera with multimodal AI, and it is now being used in real operations on many roads in Japan. This presentation focuses on a wide-area deployment impact assessment in Hokkaido, Japan, and discusses how the resulting inspection data can support Big-Data–Driven Pavement Management.
First, we outline the proposed inspection workflow (from data acquisition to automated damage detection and digital mapping) using video, acceleration, and location information collected with a single vehicle-mounted action camera. By integrating these heterogeneous data streams with multimodal AI, we estimate key pavement condition indicators such as surface cracking, rutting, and the International Roughness Index (IRI). This approach eliminates the need for dedicated survey vehicles and enables consistent evaluation with minimal additional fieldwork.
Second, we present results from a wide-area operational deployment in Hokkaido that evaluates practical impacts for road agencies. Over an approximately 1,400 km network, the proposed approach reduced inspection costs by about 70% compared with conventional methods and decreased total labor input by about 80%, demonstrating substantial savings in both budget and workforce resources while maintaining operational usability.
Third, we present a method for converting continuously accumulated onboard video data and inspection results into actionable management information. High-frequency, spatially dense condition records enable network-level analysis, such as identifying areas of concentrated damage or prioritizing sections affecting bicycle comfort, safety, and the reliability of emergency transport. Finally, we discuss how this inspection big data complements traditional periodic surveys and accelerates the transition from reactive repairs to data-driven, preventive pavement asset management across extensive networks.