Implementation of the paper Low-Constraint Ray Optimization for Photogrammetric Shallow Water Bathymetry to be published in 3D Underwater Mapping from Above and Below, Vienna, 2025.
you will need eigen3, opencv4.x, apriltag, nlohmann_json, fmt, and cxxopts to build the C++ optimizer. CCTag and ceres-solver come bundled as subprojects. For the python wrapper viswrap.py, you need vispy and numpy.
meson setup builddir --buildtype release
meson compile -C builddirpython3 viswrap.py path/to/data --lone=0.04 --solvewhere path/to/data contains each camera's images, saved as 0.png for camera 0, 1.png for camera 1 etc. with optional masks as 0_mask.png. Additionally, for each pair of cameras, (0,1), (0,2), (1,2), etc. there needs to be a file 0-to-1.json, 0-to-2.json, etc. that contains calibration matrices K{1,2}, distortion coefficients d{1,2} and the relative orientation R,T of camera 2. R,T are defined according to OpenCV's stereoCalibrate() (4.11.0).
optionally, --abcd or --point and --perpvec can be used to manually supply an initial guess for the water plane, with abcd as the Hesse normal form or point as a 3d point on the plane, and perpvec as its perpendicular vector. --solve will make the optimizer try to solve the problem. without it, only the initial guess is displayed. for robustness, --lone supplies a floating point value for SoftL1-loss robustification on the backrefractive disparity.
for feature detection, three methods are available: --cctag, --sift or apriltag (default). CCTag and apriltag will require fiducials to be present in the underwater scene.