Mission-analysis tools for the underwater robot (ROV) built by Triton Robotics, a student team from Seattle that competes in the MATE ROV Competition (Ranger class). The team used these tools at the 2026 MATE ROV World Championship, where it placed 2nd overall. They turn the ROV's camera footage into the measurements and models that competition tasks are scored on: stereo depth and length measurements, 3D reconstructions, and corrected underwater images.
The vehicle's software is split across three repositories:
| Repository | Runs on | Role |
|---|---|---|
| TritonOS | Raspberry Pi on the ROV | Control loop, thruster output, sensor telemetry, camera streams |
| TritonPilot | Pilot computer | Controller input, live video and telemetry, logging |
| TritonAnalysis (this repo) | Separate analysis laptop | Stereo measurement, 3D reconstruction, mission tools |
Written and maintained by Thomas Gust, Triton's software and electrical lead.
TritonAnalysis contains standalone mission-analysis tools for Triton's competition workflow. These applets run on an analysis laptop against saved images, saved videos, or manually entered task data.
This repository is intentionally separate from TritonPilot and TritonOS. TritonAnalysis does not talk to the ROV, publish pilot commands, subscribe to live telemetry, start camera streams, or depend on the topside control UI. Its job is to help the team interpret mission data without adding complexity to the live piloting station.
The repository is organized as a unified competition app, packaged backup applets, and task-focused analysis modules:
- Unified tabbed competition-day analysis window
- Iceberg tracking and threat assessment
- Iceberg PVC segment measurement
- Planar height measurement
- Multi-rectangle planar length measurement
- Coral garden manual CAD model generation
- eDNA frequency analysis
- Underwater color correction and frame export
- Stereo rectification, disparity, depth, and length checks
- Stereo segment measurement presets for generic, iceberg, and coral lengths
- RealityScan stereo photogrammetry reconstruction wrapper
- Three.js OBJ viewport and distance measurement for reconstructed models
- OpenAI-assisted European green crab counter for saved board images
- Embedded SSH console for analysis-link shell access during field setup
Each applet can be launched directly with python -m ... from the repository
root.
main_triton_analysis.py Top-level unified GUI launcher
triton_analysis/apps/main_triton_analysis.py Unified tabbed competition-day GUI
triton_analysis/apps/main_iceberg_tracking.py Iceberg threat-assessment GUI
triton_analysis/apps/main_iceberg_measurement.py Iceberg variable-segment GUI
triton_analysis/apps/main_planar_height_measurement.py Planar height measurement GUI
triton_analysis/apps/main_multi_rect_length_measurement.py Multi-rectangle length GUI
triton_analysis/apps/main_coral_garden_model.py Coral garden CAD-model GUI
triton_analysis/apps/main_edna_analysis.py eDNA frequency GUI
triton_analysis/apps/main_stereo_depth.py Stereo depth/length-check GUI
triton_analysis/apps/main_stereo_segment_measurement.py Stereo segment measurement GUI
triton_analysis/apps/main_stereo_iceberg_measurement.py Iceberg preset shortcut
triton_analysis/apps/main_realityscan_reconstruction.py Stereo RealityScan reconstruction GUI
triton_analysis/apps/main_realityscan_model_viewer.py Three.js OBJ measurement viewer
triton_analysis/apps/main_crab_counter.py European green crab counting GUI
triton_analysis/apps/color_corr.py Underwater correction/frame-export GUI
triton_analysis/sync/pilot_transfer.py Pull-only TritonPilot media sync helper
triton_analysis/iceberg/tracking.py Coordinate/threat-assessment logic
triton_analysis/iceberg/measurement.py 2D/3D measurement algorithms
triton_analysis/measurement/planar.py Planar homography measurement algorithms
triton_analysis/stereo/calibration.py Stereo calibration artifact generation
triton_analysis/stereo/depth.py Stereo rectification/disparity helpers
triton_analysis/stereo/segment_measurement.py Stereo segment endpoint measurement helpers
triton_analysis/stereo/iceberg_measurement.py Iceberg measurement compatibility wrappers
triton_analysis/coral/garden_model.py Prism model and OBJ export
triton_analysis/edna/analysis.py eDNA frequency calculations and reports
triton_analysis/gui/ PyQt windows and responsive helpers
tools/ Batch/CLI helper tools
data/ Bundled app/test assets
tests/ Hardware-free tests and optional vision checks
docs/ Maintained user and maintainer docs
- Documentation Index
- Setup Guide
- Network And Data Handoff Guide
- Operations Guide
- Architecture Overview
- Applet Reference
- Subsystem Reference
- Data And Inputs
- Testing And Troubleshooting
On Windows:
powershell -NoProfile -ExecutionPolicy Bypass -File .\setup_windows.ps1 -IncludeDev
.\.venv\Scripts\Activate.ps1
python -m pytestThe Windows setup script installs requirements-windows.txt and verifies that
the unified app's SSH tab can import paramiko and load the Triton Pi preset.
On macOS or Linux:
python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt
python -m pip install pytest
python -m pytestUnified competition window:
python main_triton_analysis.py
python -m triton_analysis.apps.main_triton_analysis
python -m triton_analysis.apps.main_triton_analysis --stereo-manifest path\to\stereo_session --calibration path\to\stereo_calibration.json
python -m triton_analysis.apps.main_triton_analysis --pilot-transfer-url http://10.77.0.1:8765
python -m triton_analysis.apps.main_triton_analysis --workspace D:\TritonAnalysisWorkspaceThe unified window opens with Coral Reconstruction, Stereo Iceberg Length,
Iceberg Tracking, eDNA Analysis, Crab Counter, Crab Dataset, Stereo Calibration,
Backup Coral Measurement, Backup Iceberg Measurement, and SSH. The standalone applets below remain
available as backups.
The status bar includes automatic TritonPilot media sync status and the active
destination folder.
Long-running analysis jobs (Coral Reconstruction, Crab Counter, Crab Dataset,
and Stereo Calibration) report into a shared activity tracker so simultaneous
runs are never lost across tabs: each tab shows a status dot (running, done, or
failed), a finished job on a background tab raises a click-to-open toast, and the
Activity button (top-right) lists every running and recently finished job. The
Notifications menu flashes the taskbar when a job finishes while the window is
unfocused and can play an optional finish sound.
The SSH tab includes a Triton Pi preset for triton@tritonpi.local with the
shared field password already filled.
By default, synced media and generated outputs live under .\Workspace inside
this checkout. The Workspace menu can move that root on each computer while
keeping subfolders like incoming\pilot, results, reports, and
calibrations consistent.
Iceberg tracking threat applet:
python -m triton_analysis.apps.main_iceberg_trackingCoral garden CAD model applet:
python -m triton_analysis.apps.main_coral_garden_modeleDNA frequency analysis applet:
python -m triton_analysis.apps.main_edna_analysis
python -m triton_analysis.apps.main_edna_analysis --sampleEuropean green crab counter:
$env:OPENAI_API_KEY="sk-..."
python -m triton_analysis.apps.main_crab_counter path\to\arm_camera_snapshot.png
python -m triton_analysis.apps.main_triton_analysis --tab crabThe crab counter auto-fills its repo-bundled references from
data\crab\templates, then writes JSON and an annotated image under
Workspace\results\crab_counter. Downloads and old Workspace\data template
folders are fallback-only compatibility locations.
Measurement applets:
python -m triton_analysis.apps.main_iceberg_measurement [image-or-video ...]
python -m triton_analysis.apps.main_planar_height_measurement [image-or-video ...]
python -m triton_analysis.apps.main_multi_rect_length_measurement [image-or-video ...]Stereo calibration from TritonPilot capture sessions:
python -m triton_analysis.apps.main_stereo_calibration_gui path\to\manifest.json
python -m triton_analysis.apps.main_stereo_calibration path\to\manifest.json [more-manifests-or-folders ...] --charucoStereo depth and 3D length checks:
python -m triton_analysis.apps.main_stereo_depth path\to\manifest.json
python -m triton_analysis.apps.main_stereo_depth path\to\manifest.json --calibration path\to\stereo_calibration.json
python -m triton_analysis.apps.main_stereo_segment_measurement path\to\manifest.json
python -m triton_analysis.apps.main_stereo_segment_measurement path\to\manifest.json --preset coral
python -m triton_analysis.apps.main_stereo_iceberg_measurement path\to\manifest.json
python -m triton_analysis.apps.main_stereo_iceberg_measurement path\to\manifest.json --calibration path\to\stereo_calibration.jsonFor low-texture PVC structures, use the stereo depth applet's rectified left/right endpoint clicks for direct triangulation; dense disparity is still best treated as a diagnostic unless the object has visible texture. For task-focused straight-line measurements, use the stereo segment applet and choose the Generic Segment, Iceberg Keel, or Coral Rig Length preset. The older stereo iceberg command still opens the same applet in Iceberg Keel mode.
RealityScan stereo reconstruction and model viewing:
python -m triton_analysis.apps.main_realityscan_reconstruction path\to\stereo_session --calibration path\to\stereo_calibration.json
python -m triton_analysis.apps.main_realityscan_model_viewer path\to\underwater_model_metric.objThe reconstruction GUI includes a Model Viewer tab that embeds the Three.js
viewport when PyQt6-WebEngine is installed, with a browser fallback.
Underwater color correction and frame export:
python -m triton_analysis.apps.color_corrPilot media transfer helper:
python -m tools.pilot_transfer_sync http://10.77.0.1:8765 --output ".\Workspace\incoming\pilot"Use TritonPilot to operate the ROV and capture clean media. Move saved files or manual task values to the analysis computer. Use TritonAnalysis to run the appropriate applet, export results, and prepare judge-facing numbers or artifacts.
Keeping analysis off the pilot station protects the live-control computer from long-running CV work, accidental UI clutter, and mission-specific changes.