How yaseenupskills-ops/ml-project works
FallGuard is an on-device fall-detection system: camera/video input becomes pose keypoints and temporal features, then model predictions pass through voting, a grace period, and caregiver alerts. The Streamlit dashboard provides alert and live-monitor views; local metrics and alert history support operations. Evaluation and the CNN-LSTM model are separate documented capabilities. The live pipeline and several integrations are supported by indexed references; unsampled dashboard and server details limit some relationship claims.
Main entry points: live_detection.py Live detection · app.py Caregiver dashboard · alert.py Alert delivery · stream_server.py Local stream server
Components 18
Capture and pose
- Camera sources
camera.py - Pose extraction
pose_extraction.py - Temporal features
features.py
Detection pipeline
- Live detection
live_detection.py - Random Forest
model_rf.py - CNN-LSTM
model_cnn_lstm.py - Voting and tiers
decision_logic.py - Grace period
grace_period.py
Caregiver operations
- Alert delivery
alert.py - Caregiver dashboard
dashboard/app.py - Alert history
alert_store.py - Local stream server
stream_server.py
Model evaluation
- Held-out evaluation
evaluate.py
Runtime support
- Live metrics
metrics.py - Configuration
project_config.py
Other components
- Camera scene
- Caregiver
- Email service
Connections 19
- Camera scene → Camera sources: provides video
README.md - Camera sources → Pose extraction: supplies frames
- Live detection → Pose extraction: extracts keypoints
live_detection.py - Live detection → Temporal features: computes features
live_detection.py - Live detection → Random Forest: predicts falls
live_detection.py - Live detection → Voting and tiers: votes predictions
live_detection.py - Live detection → Grace period: confirms candidate
live_detection.py - Live detection → Alert delivery: sends alert
live_detection.py - Live detection → Live metrics: updates metrics
live_detection.py - Caregiver dashboard → Local stream server: uses live stream
dashboard/app.py - Local stream server → Camera sources: opens camera
stream_server.py - Local stream server → Live metrics: reads metrics
stream_server.py - Caregiver dashboard → Alert history: views alert history
dashboard/app.py - Alert delivery → Alert history: records alerts
alert.py - Caregiver dashboard → Caregiver: presents status
README.md - Alert delivery → Email service: sends email
alert.py - Held-out evaluation → Random Forest: evaluates model
evaluate.py - Configuration → Live detection: configures runtime
live_detection.py - CNN-LSTM → Held-out evaluation: supports evaluation