Crate edgefirst_tracker 

Multi-object tracking for EdgeFirst inference pipelines.

This crate provides the Tracker trait and a concrete implementation, ByteTrack, which associates per-frame detections into persistent tracks using two-pass IoU matching and Kalman-smoothed trajectory prediction.

Quick start

use edgefirst_tracker::{bytetrack::ByteTrackBuilder, Tracker, MockDetection};

let mut tracker = ByteTrackBuilder::new().build();

// Timestamps are nanoseconds; here we use small integers for clarity.
let frame1 = vec![MockDetection::new([0.1, 0.1, 0.5, 0.5], 0.9, 0)];
let infos = tracker.update(&frame1, 1_000_000);

// Each element corresponds to the detection at the same index; `Some` means
// the detection was either matched to an existing track or started a new one.
if let Some(info) = infos[0] {
    println!("track {} created at t={}", info.uuid, info.created);
}

Terminology

• Tracklet — an internal state machine (Kalman filter + metadata) that follows one object across frames.
• Detection — a single-frame bounding box with confidence score and class label, produced by a model decoder.
• Track — the public view of a tracklet: a TrackInfo value returned by Tracker::update or Tracker::get_active_tracks.

Architecture

See crates/tracker/ARCHITECTURE.md for the full design description, Mermaid class diagram, tracking flow, and Kalman filter state documentation.

Re-exports

pub use bytetrack::ByteTrack;

pub use bytetrack::ByteTrackBuilder;

Modules

bytetrack

kalman

Structs

ActiveTrackInfo

A TrackInfo paired with the most recent raw detection box for that track.

TrackInfo

Per-track metadata returned by Tracker::update and Tracker::get_active_tracks.

Uuid

A Universally Unique Identifier (UUID).

Traits

DetectionBox

A bounding box produced by a detection model, suitable for use with a Tracker.

Tracker

Multi-object tracker interface.