feat: Add ByteTrack tracker with IoU matching and Kalman motion

[omkar-334]

Summary

This pr adds the ByteTrack tracking algorithm and the matching and motion primitives it builds on.

Changes

• matching/: IoU distance, score fusion, and threshold-gated Hungarian assignment (scipy.optimize.linear_sum_assignment).
• motion/: a constant-velocity Kalman filter in xyah state, with xyxy conversions.
• algorithms/bytetrack.py: ByteTrackTracker (two-stage high/low-score association, lost-track recovery, class-aware duplicate suppression) and ByteTrackConfig, registered under bytetrack.
• Config loading dispatches on the algorithm name, so TrackerConfig.from_yaml and BaseTracker.from_config resolve to ByteTrackConfig and construct ByteTrackTracker.
• scipy added to dependencies;

Behavior

• Two-stage association matches confirmed tracks against high-score detections, then unmatched active tracks against low-score detections.
• Output contains ACTIVE tracks only; lost tracks coast internally until they recover or age out.
• Thresholds follow the reference: high/low split at high_score_threshold, new tracks require a score 0.1 above it.

Checklist

• The PR title and description are clear and descriptive
• I have manually tested the changes
• All changes are covered by automated tests
• All related issues are linked to this PR (if applicable)
• Documentation has been updated (if applicable)

[Copilot]

Pull request overview

This PR introduces the ByteTrack multi-object tracking algorithm into libraries/getitrack, adding the matching (IoU + Hungarian assignment) and motion (constant-velocity Kalman filter in xyah) primitives it depends on, along with configuration defaults and unit tests.

Changes:

• Added ByteTrack tracker implementation + config variant and algorithm registration (bytetrack).
• Added matching utilities (IoU matrix / score fusion / threshold-gated Hungarian assignment) and a Kalman motion model with xyxy↔xyah conversions.
• Added SciPy as a dependency (for linear_sum_assignment and scipy.linalg) and introduced test coverage for matching, Kalman, config defaults, and ByteTrack behavior.

Reviewed changes

Copilot reviewed 14 out of 15 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
libraries/getitrack/uv.lock	Locks SciPy (marker-split by Python version) and updates resolution markers.
libraries/getitrack/pyproject.toml	Adds scipy>=1.13 runtime dependency.
libraries/getitrack/configs/default.yaml	Updates defaults for ByteTrack split threshold and motion velocity decay.
libraries/getitrack/src/getitrack/init.py	Ensures bundled algorithms are imported/registered on package import; keeps default logging disabled.
libraries/getitrack/src/getitrack/algorithms/init.py	Exposes bundled algorithms package and imports ByteTrack to register it.
libraries/getitrack/src/getitrack/algorithms/bytetrack.py	Implements ByteTrackTracker + ByteTrackConfig, including association, lifecycle handling, and duplicate suppression.
libraries/getitrack/src/getitrack/matching/init.py	Exports matching helpers (IoU + assignment).
libraries/getitrack/src/getitrack/matching/iou.py	Implements IoU computation, score fusion, and threshold-gated Hungarian assignment.
libraries/getitrack/src/getitrack/motion/init.py	Exports Kalman filter + bbox conversion helpers.
libraries/getitrack/src/getitrack/motion/kalman.py	Implements constant-velocity Kalman filter and xyxy↔xyah conversions.
libraries/getitrack/src/getitrack/config.py	Updates MotionConfig.velocity_decay default to 1.0 (no damping) and adjusts docstring.
libraries/getitrack/tests/unit/test_matching.py	Adds unit tests for IoU, score fusion, and gated assignment behavior.
libraries/getitrack/tests/unit/test_kalman.py	Adds unit tests for Kalman filter behavior and bbox conversions.
libraries/getitrack/tests/unit/test_config.py	Updates unit test expectation for motion default velocity decay.
libraries/getitrack/tests/unit/test_bytetrack.py	Adds ByteTrack behavior tests (registry dispatch, lifecycle, low-score recovery, duplicate suppression, det_indices).

[kprokofi]

What is the reference?

[omkar-334]

reference is the bytetrack code - https://github.com/FoundationVision/ByteTrack

[kprokofi]

is there a reason to keep all methods and variables private except for "reset"?

[omkar-334]

If you look at BaseTracker - three methods are public - reset, update and from_config, so this same convention is followed in downstream trackers. Only these 3 are prublic and rest are private since they are not user-facing

[leoll2-alias]

For the the Kalman filter, have you considered off-the-shelf implementations like pykalman or similar?

[omkar-334]

I did consider pykalman specifically, but i still had to write code for batch-prediction, conversions and other things. it also seems to have reduced maintenance over the last year. so i went with writing our own kalman implementation, since other algorithms can also reuse this