calib-targets-core

Shared types and geometric utilities for the calib-targets workspace. Small, dependency-light, and purely geometric — it does not depend on any concrete corner detector or image crate. Every detector in the workspace (chessboard, ChArUco, marker board, PuzzleBoard) emits its result in terms of these types.

Most users install calib-targets (the facade) rather than this crate directly. Install calib-targets-core only if you are writing a new detector or consuming detection results without the facade.

Install

[dependencies]
calib-targets-core = "0.13"
nalgebra = "0.34"

Types you will see

Type	Role
[`AxisEstimate`]	One local grid-axis direction: an angle in `[0, π)` plus a 1σ uncertainty. Corners carry two.
[`LabeledCorner`]	A detected corner with grid label: `position`, `grid: Option<(i, j)>`, `id`, `target_position`, `score`. The common detector output.
[`TargetDetection`]	`{ kind: TargetKind, corners: Vec<LabeledCorner> }`. Uniform wrapper across all detector types.
[`TargetKind`]	`Chessboard`, `ChArUco`, `PuzzleBoard`, `CheckerboardMarker`. Non-exhaustive.
[`Coord`]	Integer lattice index `{ u, v }` (`u` right = grid column `i`, `v` down = grid row `j`). Labels are always rebased so that the bounding-box minimum sits at `(0, 0)`.
[`GridTransform`]	Canonical affine integer-grid mapping owned by `projective-grid`: lattice kind + 2×2 matrix + translation.
[`GridAlignment`]	Semantic alias for `GridTransform` used by detector result fields; not a second structure.

Utilities

[homography_from_4pt] + [Homography] — 4-point DLT solver with Hartley normalisation.
[warp_perspective_gray] — grayscale perspective warp.
[sample_bilinear] / [sample_bilinear_fast] / [sample_bilinear_u8] — subpixel sampling helpers on a [GrayImageView].
[cluster_orientations] + [OrientationClusteringParams] — axis-angle histogram clustering used by the chessboard detector.
[ChessConfig] + friends — shared ChESS corner-detector configuration struct, consumed by every higher-level detector through ChessboardParams::chess.

Coordinate conventions

Image pixels. Origin at top-left; x right, y down. Pixel centre sampling uses (x + 0.5, y + 0.5).
Grid indices. (i, j) with i right, j down. Grid labels are non-negative — every detector rebases the bounding-box minimum to (0, 0).
Quad / homography corner order. TL, TR, BR, BL (clockwise). Never self-crossing.
Corner orientation. A corner's axes: [AxisEstimate; 2] holds two ordered axis angles with axes[1] − axes[0] ≈ π/2. The CCW sweep from axes[0] to axes[1] crosses a dark sector. See the workspace conventions chapter. (The raw per-corner input type lives in the detector crates — e.g. calib_targets_chessboard::ChessCorner — not here; calib-targets-core carries only the shared output types.)

Quickstart

use calib_targets_core::{Coord, LabeledCorner, TargetDetection, TargetKind};
use nalgebra::Point2;

// Public types are `#[non_exhaustive]`: build them with the named
// constructor plus `with_*` setters, not a struct literal.
let corner = LabeledCorner::new(Point2::new(10.0, 20.0), 1.0)
    .with_grid(Coord::new(0, 0));

let detection = TargetDetection::new(TargetKind::Chessboard, vec![corner]);

println!(
    "{:?} {}",
    detection.corners[0].position,
    detection.corners.len(),
);