DAPM: UAV Monocular Depth Estimation from Any Height, Pitch, Roll and FOV

Researchers introduced DAPM, a monocular UAV model for depth and camera pose estimation across changing flight heights, pitch, roll, and field of view.

The framework uses geometric analysis of viewing angles, an Ideal Ground Depth supervision module, and coarse-to-fine quantization to better handle viewpoint variation.

They also released the UAPD dataset and reported state-of-the-art performance on depth and pose benchmarks.

The work addresses a major limitation in drone vision, with potential impact on aerial mapping, 3D reconstruction, and autonomous navigation.