Orthomosaics from panoramic photos for Hawaiian roadways

Natural hazards pose a significant risk to transport infrastructure and can cause annual direct damage of 3.1 to 22 billion US dollars globally, with 84% of it being flooding-related. Cost-effective approaches to assessing road damage and conditions are vital for repairing and reconstructing the transportation infrastructure after hazards. We conducted a study that presents a novel methodology developed for generating highly detailed orthomosaics of road surfaces, achieving millimeter-level spatial resolution. The approach utilizes panoramic photos obtained from a mobile camera system, coupled with Structure-from-Motion (SfM) technology. A key aspect of the methodology is the accurate masking of the ego-vehicle, sky, and moving objects (such as vehicles, bicycles, and pedestrians) present in the street scenes captured by the photos. This masking process involves a combination of deep learning algorithms, image processing techniques, and manual editing. The study demonstrates that removing these objects from the images significantly improves photo alignment precision and enhances the overall quality of the orthomosaics. The resulting orthomosaics are found to be highly applicable for GIS analysis and the assessment of road conditions and damages.

自然灾害对交通基础设施构成严重威胁，全球每年由此造成的直接经济损失达31亿至220亿美元，其中84%与洪涝灾害相关。灾后交通基础设施的修复与重建工作，亟需具备成本效益的道路损伤与状况评估方法。本研究提出一种全新方法，可生成具备毫米级空间分辨率的道路表面高细节正射影像镶嵌图（orthomosaics）。该方法依托移动摄像系统采集的全景照片，并结合运动恢复结构（Structure-from-Motion, SfM）技术。该方法的核心环节之一，是对拍摄场景中的自车（ego-vehicle）、天空及移动物体（如机动车、非机动车与行人）进行精准掩膜处理。该掩膜流程融合深度学习算法、图像处理技术与人工编辑手段。研究表明，从图像中移除上述干扰物体，可显著提升图像配准精度，并优化正射影像镶嵌图的整体质量。最终生成的正射影像镶嵌图可广泛应用于地理信息系统（GIS）分析，以及道路状况与损伤评估工作。