ARU-DART

Automatic Recognition Unit for Darts — an autoscoring system using 3 USB cameras mounted on an LED ring around a steel-tip dartboard.

Development supported by Amazon Kiro.

Hardware Setup

• Winmau Blade 6 Triple Core board (standard 17" diameter)
• 3 OV9732 USB cameras at 120° intervals (800×600, MJPG)
• LED ring for stable illumination
• Raspberry Pi 4 (or macOS for development)
• Colored dart flights recommended (black flights reduce detection accuracy)

Quick Start

# Install dependencies
pip install -r requirements.txt

# Run 3-dart game mode (recommended)
python main.py --state-machine --dev-mode

# Run single-dart throw mode
python main.py --single-dart-test --dev-mode

# Run manual placement mode
python main.py --manual-dart-test --dev-mode

Game Modes

State Machine Mode (3-dart rounds)

python main.py --state-machine --dev-mode

Full game cycle: throw 3 darts → scores displayed → pull out darts → 2s cooldown → next round. Detects dart impacts via motion, scores each throw, counts to 3, then waits for pull-out before resetting.

Single Dart Test (thrown darts)

python main.py --single-dart-test --dev-mode

Throw a dart, system detects impact and scores it. Remove dart, system resets background, ready for next throw. One dart at a time.

Manual Dart Test (place by hand)

python main.py --manual-dart-test --dev-mode

Timed cycle: stabilize → "PUT IN NOW" (7s countdown) → detect → show result → "REMOVE DART" → repeat. For testing without throwing.

Accuracy Testing

Known Position Test

# Original 14-position set (DB, SB, T/D/BS/SS for sectors 20, 1, 5)
python main.py --accuracy-test --dev-mode

# Per-ring test across all 20 sectors
python main.py --accuracy-test --ring T --dev-mode   # Triples
python main.py --accuracy-test --ring D --dev-mode   # Doubles
python main.py --accuracy-test --ring BS --dev-mode  # Big singles
python main.py --accuracy-test --ring SS --dev-mode  # Small singles

Guides you through placing darts at specific board positions, compares detected vs expected scores, generates accuracy report.

Feedback Mode (score confirmation)

python main.py --single-dart-test --feedback-mode --dev-mode
python main.py --manual-dart-test --feedback-mode --dev-mode

After each detection, shows score on screen with "Correct? (y)es / (n)o". Press y/n in the camera window. Feedback stored in data/feedback/correct/ and data/feedback/incorrect/.

Feedback Analysis

PYTHONPATH=. python scripts/analyze_feedback.py        # Accuracy report
PYTHONPATH=. python scripts/generate_heatmaps.py       # Board heatmap images
PYTHONPATH=. python scripts/export_dataset.py           # ML training dataset CSV

Calibration

python main.py --calibrate --dev-mode              # Manual calibration
python main.py --calibrate-intrinsic --dev-mode    # Chessboard intrinsic calibration
python main.py --verify-calibration --dev-mode     # Verify calibration accuracy

Keyboard shortcuts in dev mode:

• c — trigger manual calibration
• v — toggle spiderweb overlay (projected board grid)

All CLI Flags

Flag	Description
--config PATH	Config file (default: config.toml)
--dev-mode	Enable camera preview windows
--state-machine	3-dart round mode with pull-out detection
--single-dart-test	Single dart throw mode (motion-detected)
--manual-dart-test	Manual placement mode (countdown-based)
--accuracy-test	Test against known board positions
--ring T|D|BS|SS	Ring filter for accuracy test
--feedback-mode	Enable score confirmation UI
--diagnostics	Enable per-throw JSON logging
--calibrate	Run manual calibration at startup
--calibrate-intrinsic	Run chessboard calibration
--verify-calibration	Run calibration verification
--manual-test	Legacy manual test (pause/place/detect)
--record-mode	Capture images for regression tests
--single-camera N	Test single camera (0, 1, or 2)
--show-histogram	Show RGB histogram overlay

Accuracy Results

Ring	Manual Placement	Live Throws (colored flights)
Triples	100% (20/20)	~94%
Big Singles	100% (20/20)	~94%
Small Singles	100% (20/20)	~94%
Doubles	80% (16/20)	~80%
Bulls	100%	100%

Doubles are the hardest ring — at the 170mm board edge where sub-mm calibration precision matters most.

Project Structure

src/
├── calibration/          # Coordinate mapping, board geometry
├── camera/               # Camera management
├── processing/           # Motion detection, dart detection, background model
├── fusion/               # Multi-camera fusion, scoring, dart hit events
├── state_machine/        # Throw lifecycle (3-dart rounds, pull-out)
├── feedback/             # Score parser, feedback collection, storage
├── analysis/             # Accuracy analyzer, heatmaps, dataset export
├── diagnostics/          # Diagnostic logging, accuracy test runner
└── util/                 # Logging, metrics

scripts/
├── analyze_feedback.py   # Generate accuracy report from feedback data
├── generate_heatmaps.py  # Generate board accuracy heatmaps
└── export_dataset.py     # Export verified dataset for ML training

tests/                    # Unit tests + property-based tests (hypothesis)
data/
├── feedback/             # Feedback storage (correct/incorrect)
├── diagnostics/          # Per-session diagnostic data
├── throws/               # Per-session throw images
└── recordings/           # Regression test recordings

Documentation

• .kiro/steering/project-context.md — Project overview and current status
• .kiro/steering/development-knowledge.md — Lessons learned, tuned parameters, CLI reference
• .kiro/specs/ — Spec-driven development documents (requirements, design, tasks per step)