Multi-sensor AI for Surface Fault Detection

This talk presents state-of-the-art research outcomes from the imec.ICON HAIROAD project. We developed computer vision pipelines grounded in key concepts such as domain generalization and data-efficient learning, aimed at automatically detecting surface faults on materials including asphalt, concrete, and block pavements. To identify faults with depth characteristics—such as rutting and subsidence—we fused data from camera and LiDAR sensors. For faults with rich texture, such as cracks and potholes, a camera-only solution was employed. Additionally, HAIROAD explored the refinement of acoustic data to assess road surface roughness. This visual and acoustic profiling contributes to the construction of BRRC indicators, which are essential for planning timely and cost-effective road maintenance.