Skip to content

MetriPlane v0.1.0 - Initial Public Release

Choose a tag to compare

View all tags
@Miko997 Miko997 released this 14 May 11:42
· 33 commits to main since this release
v0.1.0
dfd9e37

MetriPlane v0.1.0 - Initial Public Release

MetriPlane is a camera-first planar digital-twin toolkit for reproducible indoor object-state tracking, metric world-coordinate mapping, multi-camera fusion, zone analytics, WebSocket streaming, JSONL recording, deterministic replay, and benchmark-backed evaluation.

This is the first public release of the MetriPlane source and evidence package.

Highlights

  • Camera-first planar tracking using ArUco fiducial markers.
  • Metric XY world-coordinate mapping with Z=0.
  • One- and two-camera tracking pipeline.
  • Multi-camera fusion backend.
  • Zone dwell and transition analytics.
  • WebSocket state streaming.
  • JSONL recording and deterministic replay.
  • Operator dashboard and release demo workflow.
  • Docker/demo proof and operator UI smoke evidence.
  • Benchmark evidence for latency, mapping error, ID continuity, replay determinism, overload behavior, fusion jitter, and CPU/GPU fusion tradeoffs.
  • Public demo videos linked from the README.

Benchmark evidence

Primary evidence is available under:

Key reported results include:

  • Mapping error: 0.63 cm mean, 1.07 cm max over nine grid points.
  • Timing run: 4,384 frames over 15.056 s, mean 291.2 FPS.
  • Detection p95 latency: 1.50 ms for cam0 and 1.72 ms for cam1.
  • Fusion p95 latency: 0.19 ms.
  • Static fiducial continuity: 100% for primary markers.
  • Motion fiducial continuity: 97.42–99.11% for primary markers.
  • Replay determinism: 301-frame fixed-clock replay, 0.0 cm maximum positional difference, 0 event mismatches.
  • Backpressure: 120 Hz synthetic overload, bounded queue depth of 5, 2,600 stale frames dropped.
  • CPU/GPU equivalence: 0.0 cm RMSE difference over 13,161 samples.
  • CPU/GPU fusion benchmark: GPU backend correct but slower than CPU for tested N=1–1000 workloads.

Demo videos

Public demos are linked from the README and the project YouTube channel:

  • Public release demo.
  • Operator UI and live fusion.
  • Deterministic replay and provenance.
  • Backpressure and health behavior.
  • Two-camera fusion.
  • ArUco-driven digital-twin tracking.
  • ROS 2 / Omniverse-facing concept demos.

Known limitations

  • Planar XY tracking only; Z=0.
  • Fiducial markers required.
  • Not a marker-free object-recognition benchmark.
  • Not a safety-certified industrial-control system.
  • GPU benchmark covers only the fusion-compute block, not camera capture or marker detection.
  • Large raw JSONL sessions may be archived separately from the GitHub repository.

Provenance

MetriPlane was previously developed under the internal name VisionTwin. Historical evidence artifacts may preserve the old name where editing would break checksums or provenance.

Assets 2
Loading