perspective_remap.py: new module that inverts the terrain-reference warp
in TERRAIN-DISP-MERGED frames using ELEV_GND depth (64×80 tiles, tilted
plane, no NaN). Produces natural-parallax perspective images per scene,
valid only in the overlap with the reference camera's ground footprint.

elphel_to_colmap_perspective.py: new CLI script (parallel to the original
elphel_to_colmap.py, which is kept unchanged) that adds --elev_gnd and
applies the remap before writing COLMAP output.

Tested on 185-frame NC dataset: valid-pixel min=0.10, mean=0.54, max=1.00.
SciPy added to Dependencies in README.
Co-authored-by: Claude <claude@elphel.com>

Verifies Phase 0 output before any OpenMVS tools run:
- camera intrinsics (focal length, FoV)
- camera-centre trajectory plots (XY/XZ/YZ)
- inter-frame step statistics with outlier detection
- consecutive rotation change (warns if > 10°)
- sample image display with NaN/saturation diagnostics

Fixed a stride-2 parse bug: empty 2D-point lines must be kept
in the line list before the range(0,N,2) loop, otherwise only
every other image record is parsed.
Co-authored-by: Claude <claude@elphel.com>

Shell wrapper: elphel_to_colmap.py → InterfaceCOLMAP →
DensifyPointCloud → ReconstructMesh → RefineMesh → TextureMesh.
Configurable via OPENMVS_BIN, NUM_THREADS, HFOV env vars.
Includes installation note for building OpenMVS from source.
Co-authored-by: Claude <claude@elphel.com>

Reads 32-bit float scene slices from *-TERRAIN-DISP-MERGED.tiff and
per-scene poses from *-INTERFRAME.corr-xml, outputs COLMAP text-format
sparse model (cameras.txt, images.txt, empty points3D.txt) plus
16-bit PNG images for OpenMVS ingestion.

Camera: PINHOLE, fl derived from 32° HFoV (1116 px for 640 px width).
Rotation: mirrors GeometryCorrection.getCommonRotMatrix(); convention
verified in Phase 3 Blender smoke test.
Co-authored-by: Claude <claude@elphel.com>