How sanjana-ml/driver-safety-system works
This is a laptop-webcam driver-drowsiness monitoring system, with a separate model-training and evaluation workflow. In live use, the GUI or console entry point captures frames and passes them to the pipeline, which checks face landmarks and frame quality, calibrates and selects cues, combines CNN predictions with temporal and confidence checks, then reports status and can trigger an audio alert. The README documents reporting and training capabilities not fully sampled; those branches are shown where documented, while implementation details and some internal calls remain unverified.
Main entry points: pipeline.py Detection pipeline · app.py Monitoring GUI · predict.py Console monitor · cue_selector.py Adaptive cue selection
Components 24
Live monitoring
- Monitoring GUI
gui/app.py - GUI entry point
main.py - Console monitor
predict.py - Webcam capture
camera.py - Audio alert
utils/alert.py - Image overlays
utils/image_utils.py
Vision and inference
- Detection pipeline
utils/pipeline.py - Face landmarks
utils/landmarks.py - Frame quality gate
utils/quality.py - Adaptive cue selection
utils/cue_selector.py - Driver calibration
utils/calibration.py - Head pose
utils/head_pose.py - PERCLOS tracking
utils/perclos.py - Temporal voting
utils/sliding_window.py - CNN classifier
models/cnn_cbam.py - CBAM attention
cbam/cbam.py
Training and evaluation
- Dataset loading
training/data_loader.py - Model training
train.py - Model evaluation
testing/evaluate.py - Re-evaluation runner
testing/run_tests.py
Shared support
- Logs and reports
utils/logger.py - System configuration
config.py - Error handling
utils/exceptions.py
Other components
- Driver
Connections 32
- Driver → GUI entry point: starts monitoring
README.md - Driver → Console monitor: starts monitoring
README.md - GUI entry point → Monitoring GUI: launches
main.py - Monitoring GUI → Webcam capture: starts capture
gui/app.py - Console monitor → Webcam capture: reads frames
predict.py - Monitoring GUI → Detection pipeline: processes frames
gui/app.py - Console monitor → Detection pipeline: processes frames
predict.py - Detection pipeline → Face landmarks: detects face
utils/pipeline.py - Detection pipeline → Frame quality gate: checks quality
utils/pipeline.py - Detection pipeline → Driver calibration: calibrates cues
utils/pipeline.py - Detection pipeline → Adaptive cue selection: selects cues
utils/pipeline.py - Detection pipeline → CNN classifier: predicts drowsiness
utils/pipeline.py - Detection pipeline → PERCLOS tracking: tracks eye closure
utils/pipeline.py - Detection pipeline → Temporal voting: smooths predictions
utils/pipeline.py - Adaptive cue selection → Driver calibration: uses thresholds
utils/cue_selector.py - Adaptive cue selection → Face landmarks: reads landmarks
utils/cue_selector.py - Adaptive cue selection → Frame quality gate: checks reliability
utils/cue_selector.py - Detection pipeline → Head pose: estimates pose
utils/pipeline.py - CNN classifier → CBAM attention: applies attention
models/cnn_cbam.py - Monitoring GUI → Audio alert: controls alarm
gui/app.py - Monitoring GUI → Image overlays: draws overlays
gui/app.py - Model training → Dataset loading: loads samples
train.py - Model training → CNN classifier: builds classifier
train.py - Model training → Model evaluation: evaluates model
train.py - Re-evaluation runner → Dataset loading: loads test data
testing/run_tests.py - Re-evaluation runner → Model evaluation: generates reports
testing/run_tests.py - Model evaluation → Logs and reports: writes metrics
README.md - Monitoring GUI → Logs and reports: logs predictions
gui/app.py - Console monitor → Logs and reports: logs predictions
predict.py - Detection pipeline → System configuration: uses thresholds
utils/pipeline.py - Webcam capture → Error handling: raises camera errors
camera.py - Dataset loading → Error handling: raises dataset errors
training/data_loader.py