Sensor Fusion of 2D-LiDAR and 360-Degree Camera Data for Room Layout Reconstruction

High-resolution maritime charts (HighRes Charts) are essential for ensuring safety, accessibility, and operational efficiency within harbor environments. As a proof-of-concept, we reduce the complexity of large-scale outdoor environments and explore controlled indoor spaces as a surrogate problem. For this example, we introduce a method that seamlessly merges depth measurements from a 2D LiDAR with the RGB-image interpretation capabilities of artificial neural networks. First, we use geometric analysis and semantic segmentation to filter LiDAR points and identify candidate walls with a systematic Hough transform. The detected walls are then reviewed and selected using a confidence metric. The wall estimation is supplemented by the RGB data by determining azimuth position of ambiguous room corners, even when obscured by furniture. By fusing the final layout, our method can capture fully furnished rooms with remarkable spatial accuracy and reliability from a single scan. While relying on less assumptions than previous techniques, we report an average 2D IoU of $\mathbf{9 0. 4 7 \%}$, suggesting the feasibility of extending this approach to more complex environments. Finally, we discuss how the lessons learned can be transferred to mapping of maritime environments producing HighRes Charts.