Trinity Phase 2
Autonomous vehicle testing and validation suite using a 3-camera infrastructure setup
- Status
- Research
- Version
- 1.0.0
- Platform
- GPU workstation
- License
- Not specified
Validating AV behaviour needs ground truth that does not depend on the vehicle's own sensors; Trinity uses fixed external cameras to provide it.
How it is put together.
Trinity Phase 2 is an autonomous vehicle testing and validation suite built around three fixed infrastructure cameras that observe a test vehicle independently of its own sensors. It provides synchronised capture and recording, YOLO detection with DeepSORT tracking and 3D triangulation, ArUco-based AV identification, safety metrics such as TTC and near-miss detection, edge-case detection, perception validation and dataset export. Results feed a PyQt6 GUI, an analytics dashboard, scheduled PDF/HTML/Excel reports and a FastAPI REST/WebSocket API. The README reports the project as about 72% complete.
- 1
Core
Camera manager, H.265 recorder, calibration, config
trinity/core/ - 2
Detection and tracking
YOLO detector, DeepSORT tracker, 3D position estimation
trinity/detection/ - 3
AV monitoring
ArUco AV identification, trajectory, safety metrics, event detection
trinity/av_monitoring/ - 4
Advanced features
Edge cases, perception validation, dataset management
trinity/advanced/ - 5
Reporting
Report generation, templates, charts, scheduler
trinity/reporting/ - 6
REST API
FastAPI application with WebSocket endpoints
trinity/api/ - 7
GUI
PyQt6 main window and tabs
trinity/gui/
Specification.
- Cameras
- 3 synchronised USB cameras, ArUco marker calibration, H.265/H.264 event-triggered recording
- Detection and tracking
- YOLOv8/v10 (80 COCO classes), DeepSORT, 3D position via triangulation, BEV view
- AV monitoring
- ArUco vehicle ID, trajectory analysis, TTC / near-miss / safe distance, 7 event rules
- Advanced
- Edge-case detection (18 categories), MOTA/MOTP perception validation
- Dataset export
- COCO, KITTI and nuScenes formats with PII anonymisation
- Reporting
- PDF/HTML/Excel reports, daily/weekly/monthly scheduling
- Interfaces
- PyQt6 GUI, FastAPI REST API with WebSocket streaming, PostgreSQL/SQLite via SQLAlchemy
- Status
- README: ~72% complete; CARLA integration, test suite and Docker packages not done
- Runs on
- README minimum requirements include an NVIDIA GPU with 6 GB+ VRAM; pyproject.toml lists onnxruntime-gpu and cupy-cuda12x; used as a PyQt6 desktop app on a GPU host