locus-core

locus-core is the high-performance engine powering the locus-tag fiducial marker detection system.

Designed for robotics and autonomous systems, it provides a low-latency pipeline for detecting and decoding fiducial markers (tags) in real-time video streams.

Key Features

High Performance: Optimized with SIMD (SSE/AVX/NEON) for rapid image processing.
Architecture Support: Verified for x86_64 and aarch64.
Modular Design: Clean separation between thresholding, segmentation, and quad extraction.

Usage

Add this to your Cargo.toml:

[dependencies]
locus-core = "0.2.5"

Cargo features

Feature	Default	Description
`non_rectified`	off	Enables the `BrownConradyModel` and `KannalaBrandtModel` camera types and the corresponding `DistortionCoeffs` variants. Required for detecting markers in imagery with significant lens distortion.
`tracy`	off	Enables the Tracy profiler backend via the `tracing-tracy` bridge.
`rerun`	off	Enables Rerun SDK logging of intermediate pipeline artifacts for visual debugging.
`bench-internals`	off	Exposes internal modules (`bench_api`) for benchmarking and integration testing. Do not enable in production.

Opt in with:

[dependencies]
locus-core = { version = "0.3", features = ["non_rectified"] }

Basic Example

use locus_core::{Detector, Config, Image};

fn main() {
    let config = Config::default();
    let detector = Detector::new(config);
    
    // Load your grayscale image data
    let image = Image::from_raw(width, height, pixels);
    
    let detections = detector.detect(&image);
    for detection in detections {
        println!("Detected tag {} at {:?}", detection.id, detection.corners);
    }
}

Related Projects

locus-tag: The main repository containing Python bindings and CLI tools.

License

Licensed under either of:

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

at your option.