Method CameraModelCamRayFromImg

src/colmap/sensor/models.h:2853–2872

Unproject a pixel to a unit bearing vector in camera coordinates. Unlike CameraModelCamFromImg (which is limited to the forward hemisphere via the 2D normalized-coordinate representation), this function returns a valid 3D unit ray for any pixel on the camera model's image plane, including back-facing rays on omnidirectional cameras. Downstream geometry code (two-view geometry, absolute/relative

// Downstream geometry code (two-view geometry, absolute/relative pose,
// triangulation) should prefer this function when a 3D bearing is needed.
std::optional<Eigen::Vector3d> CameraModelCamRayFromImg(
    const CameraModelId model_id,
    const std::vector<double>& params,
    const Eigen::Vector2d& xy) {
  Eigen::Vector3d ray;
  switch (model_id) {
#define CAMERA_MODEL_CASE(CameraModel)                                      \
  case CameraModel::model_id:                                               \
    if (CameraModel::CamRayFromImg(                                         \
            params.data(), xy.x(), xy.y(), &ray.x(), &ray.y(), &ray.z())) { \
      return ray;                                                           \
    }                                                                       \
    break;

    CAMERA_MODEL_SWITCH_CASES

#undef CAMERA_MODEL_CASE
  }
  return std::nullopt;
}

double CameraModelCamFromImgThreshold(const CameraModelId model_id,
                                      const std::vector<double>& params,