3DAeroRelief: The first 3D Benchmark UAV Dataset for Post-Disaster Assessment

Area of Science:

• Computer Vision
• Geospatial Analysis
• Disaster Management

Background:

• 2D imagery limitations in disaster analysis include lack of depth, occlusions, and limited spatial context.
• Existing 3D benchmarks predominantly focus on urban/indoor scenes, neglecting disaster-affected areas.
• Accurate structural damage assessment is crucial for effective disaster response and recovery.

Purpose of the Study:

• To introduce 3DAeroRelief, the first 3D benchmark dataset tailored for post-disaster structural damage assessment.
• To provide a valuable resource for advancing 3D vision systems in real-world disaster scenarios.
• To facilitate research on 3D semantic segmentation in large-scale, outdoor, disaster-affected environments.

Main Methods:

• Dataset collection using low-cost unmanned aerial vehicles (UAVs) in hurricane-damaged regions.
• Generation of dense 3D point clouds via Structure-from-Motion and Multi-View Stereo techniques.
• Semantic annotation through manual 2D labeling projected into 3D space.

Main Results:

• 3DAeroRelief captures large-scale outdoor environments with fine-grained structural damage in real-world disaster contexts.
• Unmanned aerial vehicles (UAVs) offer affordable, flexible, and safe data collection in hazardous post-disaster areas.
• Evaluation of state-of-the-art 3D segmentation models on the dataset highlights challenges and opportunities for disaster response.

Conclusions:

• 3DAeroRelief is a significant resource for developing robust 3D vision systems for post-disaster assessment.
• The dataset enables research into 3D semantic segmentation for improved disaster response and recovery.
• This work addresses the gap in 3D benchmarks for analyzing disaster-affected environments.