How popla69/hand-sign-detection-final works
This repository is a webcam-based ASL alphabet recognizer. Its principal workflow captures video, detects a hand, extracts landmarks and engineered features, evaluates an ensemble of three models, filters and stabilizes predictions, then displays the recognized letter. The README documents that workflow; sampled main.py code confirms component setup and ensemble use. The repository also contains a separate pipeline orchestrator and temporal-sign handling. Source evidence is partial, so relationships not documented or visible in the excerpts are omitted.
Main entry points: ensemble_predictor.py Ensemble prediction · data_structures.py Shared data types · feature_engineering.py Feature engineering · camera_capture.py Camera capture
Components 19
Capture and Features
- Camera capture
src/camera_capture.py - Hand detection
src/hand_detection.py - Landmark extraction
src/landmark_extraction.py - Feature engineering
src/feature_engineering.py - Temporal modeling
src/temporal_modeling.py - Dynamic signs
src/dynamic_sign_handler.py
Recognition Models
- Landmark MLP
src/landmark_mlp.py - Temporal CNN
src/temporal_cnn.py - MobileNet vision
src/mobilenet_vision.py - Ensemble prediction
src/ensemble_predictor.py - Confidence calibration
src/temperature_scaling.py
Prediction Output
- Confidence filtering
src/confidence_gate.py - Temporal stabilization
src/temporal_stabilizer.py - Display output
src/text_output_handler.py
Shared Runtime
- Pipeline orchestrator
src/pipeline.py - System configuration
src/config.py - Shared data types
src/data_structures.py - Logging
src/logger.py
Other components
- ASL signer
Connections 23
- ASL signer → Camera capture: shows signs
README.md - Camera capture → Hand detection: provides frames
README.md - Hand detection → Landmark extraction: provides hand region
README.md - Landmark extraction → Feature engineering: provides landmarks
README.md - Feature engineering → Temporal modeling: buffers features
README.md - Temporal modeling → Ensemble prediction: provides sequence
README.md - Feature engineering → Landmark MLP: provides features
README.md - Ensemble prediction → Landmark MLP: evaluates model
src/ensemble_predictor.py - Ensemble prediction → Temporal CNN: evaluates model
src/ensemble_predictor.py - Ensemble prediction → MobileNet vision: evaluates model
src/ensemble_predictor.py - Ensemble prediction → Confidence calibration: calibrates confidence
src/ensemble_predictor.py - Ensemble prediction → Confidence filtering: provides prediction
README.md - Confidence filtering → Temporal stabilization: filters result
README.md - Temporal stabilization → Display output: provides stable letter
README.md - Camera capture → Pipeline orchestrator: initialized by
src/pipeline.py - Hand detection → Pipeline orchestrator: initialized by
src/pipeline.py - Temporal modeling → Pipeline orchestrator: configured by
src/pipeline.py - Feature engineering → Shared data types: uses data types
src/feature_engineering.py - Landmark extraction → Shared data types: uses data types
src/landmark_extraction.py - Landmark MLP → Shared data types: uses data types
src/landmark_mlp.py - Temporal CNN → Shared data types: uses data types
src/temporal_cnn.py - Ensemble prediction → Shared data types: uses data types
src/ensemble_predictor.py - Display output → Shared data types: uses data types
src/text_output_handler.py