MultiPoint: Cross-spectral Optical–Thermal Aerial Imagery

Year: 2020
DOI: 10.3929/ethz-c-000788340

Aligned optical–thermal image pair dataset from fixed-wing UAV flights for cross-spectral registration. Captured with a UI-5261SE RGB camera (16 mm) and FLIR TAU2 19 mm (7.5–13.5 μm).

Dataset Description

• 10 flights at varying times of day; split into training and test sets
• HDF5 samples with three aligned images: optical, thermal, thermal_raw
  • 640×512 resolution, undistorted and aligned
  • Optical: grayscale normalized [0,1]
  • Thermal: normalized per-image (1st–99th percentile)
  • Thermal raw normalized by 16383; temperature °C = pixel_val * 16383 * 0.04 - 273.15
• Loader interface available in multipoint

Downloads

• Training set (24 GB, 9340 pairs): https://doi.org/10.3929/ethz-c-000788340
• Test set (12 GB, 4391 pairs): https://doi.org/10.3929/ethz-c-000788340
• Training keypoint labels (10 MB): https://doi.org/10.3929/ethz-c-000788340

Citation

F. Achermann, A. Kolobov, D. Dey, T. Hinzmann, J. J. Chung, R. Siegwart, N. Lawrance,
“MultiPoint: Cross-spectral registration of thermal and optical aerial imagery”, Conference on Robot Learning (CoRL), 2020.

Tags: aerial, multispectral, thermal, optical, registration