ringgrid

Pure-Rust detector for dense ring calibration targets on hex or rectangular lattices. Detects markers with subpixel edge precision, decodes 16-sector binary IDs from a shipped baseline 893-codeword profile (with an opt-in extended profile available for larger ID spaces), fits ellipses via Fitzgibbon's direct method with RANSAC, corrects projective center bias, and estimates a board-to-image homography. No OpenCV dependency.

Key Features

• Subpixel edge detection — gradient-based radial sampling produces edge points fed to a direct ellipse fit, yielding subpixel-accurate marker localization
• Projective center correction — recovers the true projected center from inner/outer conic pencil geometry, correcting the systematic bias of ellipse-fit centers
• Consistency-first ID correction — verifies decoded IDs against local hex-lattice structure, clears contradictory IDs, and recovers safe missing IDs before global filtering
• Stable baseline IDs plus opt-in extension — shipped base profile keeps 893 stable IDs at minimum cyclic Hamming distance 2; opt-in extended grows capacity to 2180 IDs with a weaker minimum distance of 1 without introducing new polarity ambiguity beyond the shipped baseline
• Distortion-aware — supports external camera models (Brown-Conrady) via the PixelMapper trait, or blind single-parameter self-undistort estimation
• Compositional target model — TargetLayout composes hex/rect lattices, coded (16-sector) or plain rings, and optional origin-dot fiducials; legacy v4 board_spec.json files still load via TargetLayout::from_json_* auto-migration (see migration notes)
• Pure Rust — no C/C++ dependencies, no OpenCV bindings

Pipeline Stages

Named stage order: proposal -> local fit/decode -> dedup -> projective center -> id_correction -> optional global filter -> optional completion -> final homography refit.

Installation

[dependencies]
ringgrid = "0.10"

Rust Target Generation

The library can generate canonical target JSON plus printable SVG/PNG directly:

use ringgrid::{PngTargetOptions, SvgTargetOptions, TargetLayout};
use std::path::Path;

// `coded_hex` uses a deterministic geometry-derived name; use `TargetLayout::new`
// for full control over lattice, ring geometry, coding, and origin fiducials.
let target = TargetLayout::coded_hex(8.0, 15, 14, 4.8, 3.2, 1.152).unwrap();

target.write_json_file(Path::new("target.json")).unwrap();
target
    .write_target_svg(Path::new("target.svg"), &SvgTargetOptions::default())
    .unwrap();
target
    .write_target_png(
        Path::new("target.png"),
        &PngTargetOptions {
            dpi: 300.0,
            ..PngTargetOptions::default()
        },
    )
    .unwrap();

render_target_svg returns the SVG as a string, and render_target_png returns an in-memory grayscale image::GrayImage when you want to avoid file I/O. write_target_png embeds the requested DPI as PNG print metadata.

Equivalent Command-Line Workflow

The Rust API above is equivalent to the published ringgrid CLI when you want the same artifact set from the terminal instead of from application code:

cargo install ringgrid --features cli
ringgrid example --name hex_coded --out hex_coded.toml   # start from a recipe
ringgrid gen hex_coded.toml --out ./out/target           # SVG + PNG + DXF + JSON

Both paths write a v5 target_spec.json. Legacy v4 board_spec.json files still load in detection — TargetLayout::from_json_* auto-migrates the v4 schema. See the CLI Guide for the recipe format and every flag.

• tools/out/target_faststart/target_spec.json
• tools/out/target_faststart/target_print.svg
• tools/out/target_faststart/target_print.png

Use the generated JSON in detection:

use ringgrid::{Detector, TargetLayout};
use std::path::Path;

let target = TargetLayout::from_json_file(Path::new("tools/out/target_faststart/target_spec.json")).unwrap();
let detector = Detector::new(target);

Complete step-by-step target generation docs (Rust API, Rust CLI, Python script, and helper tools):

• https://vitalyvorobyev.github.io/ringgrid/book/target-generation.html

Simple Detection

use ringgrid::{Detector, TargetLayout};
use std::path::Path;

let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
let image = image::open("photo.png").unwrap().to_luma8();

let detector = Detector::new(target);
let result = detector.detect(&image).unwrap();

for marker in &result.detected_markers {
    if let Some(id) = marker.id {
        println!("Marker {id} at ({:.1}, {:.1})", marker.center[0], marker.center[1]);
    }
}

result is a slim DetectionResult. It contains the final marker list plus image size, frame metadata, optional homography, and optional self_undistort output. Per-marker fit/decode metrics, edge points, and homography RANSAC stats are an opt-in diagnostics channel — call detector.detect_with_diagnostics(&image) to also get a DetectionDiagnostics. For the serialized JSON shape and field meanings, see:

• https://vitalyvorobyev.github.io/ringgrid/book/output-format.html

With a marker diameter hint for better scale tuning:

# use ringgrid::{Detector, TargetLayout};
# use std::path::Path;
# let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
let detector = Detector::with_marker_diameter_hint(target, 32.0);

Proposal-Only Diagnostics

When you want to inspect candidate centers before fit/decode, use the proposal API directly:

use ringgrid::{Detector, ProposalConfig, TargetLayout};
use std::path::Path;

let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
let image = image::open("photo.png").unwrap().to_luma8();

let detector = Detector::with_marker_diameter_hint(target, 32.0);
let proposals = detector.propose(&image);
let diagnostics = detector.propose_with_heatmap(&image);

let result = ringgrid::find_ellipse_centers_with_heatmap(
    &image,
    &ProposalConfig {
        r_min: 4.0,
        r_max: 18.0,
        min_distance: 12.0,
        ..ProposalConfig::default()
    },
);

println!("{}", proposals.len());
println!("{:?}", diagnostics.image_size);
println!("{:?}", result.heatmap.len());

ProposalResult.heatmap is the post-Gaussian-smoothed vote accumulator used for thresholding and NMS. Proposal tutorial and Python plotting workflow:

• https://vitalyvorobyev.github.io/ringgrid/book/detection-modes/proposal-diagnostics.html

Adaptive Scale Detection

For scenes with large marker size variation, use adaptive multi-scale methods:

# use ringgrid::{Detector, ScaleTiers, TargetLayout};
# use std::path::Path;
# let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
# let detector = Detector::new(target);
# let image = image::open("photo.png").unwrap().to_luma8();
let result = detector.detect_adaptive(&image).unwrap();
let result = detector.detect_adaptive_with_hint(&image, Some(32.0)).unwrap();
let result = detector.detect_multiscale(&image, &ScaleTiers::four_tier_wide()).unwrap();

Which method to choose:

Inspect adaptive tiers before detecting:

# use ringgrid::{Detector, TargetLayout};
# use std::path::Path;
# let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
# let detector = Detector::new(target);
# let image = image::open("photo.png").unwrap().to_luma8();
let tiers = detector.adaptive_tiers(&image, Some(32.0));
let result = detector.detect_multiscale(&image, &tiers).unwrap();

Adaptive scale guide:

• https://vitalyvorobyev.github.io/ringgrid/book/detection-modes/adaptive-scale.html

Detection with Camera Model

When camera intrinsics and distortion coefficients are known, use detect_with_mapper for distortion-aware detection via a two-pass pipeline:

use ringgrid::{
    CameraIntrinsics, CameraModel, Detector, RadialTangentialDistortion, TargetLayout,
};
use std::path::Path;

let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
let image = image::open("photo.png").unwrap().to_luma8();
let (w, h) = image.dimensions();

let camera = CameraModel {
    intrinsics: CameraIntrinsics {
        fx: 900.0, fy: 900.0,
        cx: w as f64 * 0.5, cy: h as f64 * 0.5,
    },
    distortion: RadialTangentialDistortion {
        k1: -0.15, k2: 0.05, p1: 0.001, p2: -0.001, k3: 0.0,
    },
};

let detector = Detector::new(target);
let result = detector.detect_with_mapper(&image, &camera).unwrap();

for marker in &result.detected_markers {
    // center is always image-space
    println!("Image: ({:.1}, {:.1})", marker.center[0], marker.center[1]);
    // center_mapped is working-frame (undistorted)
    if let Some(mapped) = marker.center_mapped {
        println!("Working: ({:.1}, {:.1})", mapped[0], mapped[1]);
    }
}

Self-Undistort (No Calibration Required)

When camera calibration is unavailable, ringgrid can estimate a single-parameter division-model distortion correction from the detected markers:

use ringgrid::{DetectConfig, Detector, TargetLayout};
use std::path::Path;

let target = TargetLayout::from_json_file(Path::new("target.json")).unwrap();
let image = image::open("photo.png").unwrap().to_luma8();

let mut cfg = DetectConfig::from_target(target);
cfg.self_undistort.enable = true;

let detector = Detector::with_config(cfg);
let result = detector.detect(&image).unwrap();

if let Some(su) = &result.self_undistort {
    println!("Lambda: {:.3e}, applied: {}", su.model.lambda, su.applied);
}

Custom PixelMapper

Implement the PixelMapper trait to plug in any distortion model:

use ringgrid::PixelMapper;

struct Identity;

impl PixelMapper for Identity {
    fn image_to_working_pixel(&self, p: [f64; 2]) -> Option<[f64; 2]> {
        Some(p)
    }
    fn working_to_image_pixel(&self, p: [f64; 2]) -> Option<[f64; 2]> {
        Some(p)
    }
}

Then use it with detector.detect_with_mapper(&image, &mapper).

Coordinate Frames

• DetectedMarker.center — always raw image pixel coordinates
• DetectedMarker.center_mapped — working-frame (undistorted) coordinates when a mapper is active
• DetectedMarker.grid_coord — lattice cell ([q, r] for hex, [col, row] for rect); the only marker key on plain targets
• DetectedMarker.board_xy_mm — board-space marker coordinates in millimeters for valid decoded IDs
• DetectionResult.center_frame / homography_frame / board_frame — explicit frame metadata

Documentation

• User Guide — comprehensive mdbook covering marker design, detection pipeline, mathematical foundations, and configuration
• API Reference — rustdoc for all public types

License

Licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE)
• MIT license (LICENSE-MIT)

at your option.