All based on nalgebra with f64 precision:
Type Definition Description
Realf64Scalar type Pt2Point2<f64>2D point Pt3Point3<f64>3D point Vec2Vector2<f64>2D vector Vec3Vector3<f64>3D vector Mat3Matrix3<f64>3×3 matrix Mat4Matrix4<f64>4×4 matrix Iso3Isometry3<f64>SE(3) rigid transform
Type Parameters Description
FxFyCxCy<S>fx, fy, cx, cyIntrinsics matrix BrownConrady5<S>k1, k2, k3, p1, p2, itersDistortion model Pinhole(none) Projection model IdentitySensor(none) No sensor tilt ScheimpflugParamstilt_x, tilt_yScheimpflug tilt Camera<S,P,D,Sm,K>proj, dist, sensor, kComposable camera CameraParamsprojection, distortion, sensor, intrinsicsSerializable camera description (see Serialization )
Type Variants Used by
DistortionKindNone, BrownConrady5, Rational8, ThinPrism9, Division1camera.model.distortion in every config that calibrates a camera (per camera through camera_overrides in a rig)SensorKindIdentity, Scheimpflugcamera.model.sensor in the same configsCameraModelSpec{ projection, distortion, sensor }camera.model in the planar, single-camera hand-eye, laserline-device, rig extrinsics and rig hand-eye configsCameraFix{ intrinsics, projection, distortion, sensor }, each Option<Vec<String>> of parameter namescamera.fix in the same configs, and joint_optimize.fix (one per camera) of the joint rig laser solve; None keeps the model's defaultCameraOverride{ camera, model, fix }camera_overrides in RigExtrinsicsConfig and RigHandeyeConfig: replaces one camera's model and fix lists (see Scheimpflug and Laser Rigs )RobustLossNone, Huber { scale }, Cauchy { scale }, Arctan { scale }solver.robust_loss and the laser calib_loss / laser_lossHandeyeModeEyeInHand, EyeToHandhandeye_init.handeye_mode
Type Fields Description
CorrespondenceViewpoints_3d, points_2d, optional cov_px2D-3D correspondences (build with CorrespondenceView::new(..)?, or new_with_covariances(..)? with a PointCovariance per point) View<Meta>obs, metaObservation + metadata PlanarDatasetviews: Vec<View<NoMeta>>Planar calibration input RigViewObscameras: Vec<Option<CorrespondenceView>>Per-camera observations of one rig frame RigView<Meta>obs: RigViewObs, metaMulti-camera view RigDataset<Meta>num_cameras, viewsMulti-camera input ReprojectionStatsmean, rms, max, countError statistics
Every problem's *Config embeds the same small structs by stage:
Type Fields Purpose
IntrinsicsInitConfigiterations, fix_k3Per-camera linear initialization; fits only the distortion terms the camera's solves estimate SolverConfigmax_iters, verbosity, robust_lossNon-linear solve RobotPoseConfigrefine, rot_sigma, trans_sigma, estimate_sigmaRobot-pose refinement in hand-eye bundle adjustment HandeyeInitConfighandeye_mode, min_motion_angle_degHand-eye linear initialization RigConfigreference_camera_idx, refine_intrinsics, refine_sensorRig frame options
Type Description
CalibrationSession<P>Generic session container SessionMetadataProblem name, version, timestamps (read via session.metadata()) LogEntryAudit log entry (timestamp, operation, success, notes; read via session.log()) ExportRecord<E>Timestamped export InvalidationPolicyWhat to clear on input/config change
vision-calibration-optim keeps the IR (ProblemIR, ParamBlock, ResidualBlock, FactorKind, ManifoldKind, ...), the backend dispatch and the optimize_* entry points internal; see Backend-Agnostic IR Architecture . Each estimate carries a SolveReport (final_cost, num_iters); a config's SolverConfig (backend, max_iters, verbosity, robust_loss) sets how each stage solves.
Type Description
HandeyeModeEyeInHand or EyeToHand RobotPoseMetabase_se3_gripper: Iso3, a view's robot pose (every hand-eye workflow)