How besteaydogan/smart-vest-ai works
Smart Vest AI is a fall-detection prototype: SisFall recordings are transformed into window features, used to train and evaluate classifiers, then processed as streaming decisions and optionally exported for embedded inference. The graph preserves the documented preprocessing, LOSO evaluation, threshold/FSM analysis, real-time simulation, stress testing, and model-export branches. README statements support the high-level workflow; source excerpts and the SOURCE INDEX support the narrower implementation links. Several listed modules were not sampled, so their internal calls are not inferred.
Main entry points: models Saved models · pencere_motoru.py Window decision engine · train_model.py Model training · preprocess.py Window preprocessing
Components 22
Signal preparation
- Signal utilities
utils.py - Window preprocessing
preprocess.py - Feature extraction
utils.py
Modeling and evaluation
- Model training
train_model.py - Model comparison
guclu_model.py - Compact model training
kucuk_model_egit.py - Subject-independent evaluation
loso_testi.py - Confidence analysis
esik_analizi.py - FSM tuning
fsm_optimize.py - Saved models
models - Evaluation reports
reports
Fall decision runtime
- Window decision engine
pencere_motoru.py - Streaming classifier
real_time_classifier.py - Stream simulation
gercek_zamanli_sim.py - Runtime stress test
stres_testi.py - Fast stress test
stres_testi_hizli.py
Export and integration
- C++ model export
model_translator.py - Embedded model header
models/model.h
Other components
- SisFall recordings
- Sensor stream
- Alarm or protection
- ESP32-S3 firmware
Connections 27
- SisFall recordings → Window preprocessing: provides recordings
README.md - Window preprocessing → Signal utilities: uses signal helpers
preprocess.py - Window preprocessing → Feature extraction: extracts features
preprocess.py - Window preprocessing → Model training: prepares training data
README.md - Feature extraction → Model training: feeds training
README.md - Model training → Saved models: persists model
README.md - Model comparison → Saved models: exports model
guclu_model.py - Compact model training → Saved models: trains compact model
README.md - Model training → Subject-independent evaluation: evaluated by
README.md - Model training → Confidence analysis: supports analysis
README.md - Confidence analysis → FSM tuning: tunes confirmation
README.md - Subject-independent evaluation → Evaluation reports: records results
README.md - Sensor stream → Window decision engine: feeds samples
README.md - Window decision engine → Signal utilities: uses filter
pencere_motoru.py - Window decision engine → Saved models: loads classifier
pencere_motoru.py - Window decision engine → Alarm or protection: confirms fall
README.md - Sensor stream → Streaming classifier: feeds samples
README.md - Streaming classifier → Signal utilities: uses filtering
real_time_classifier.py - Streaming classifier → Saved models: loads classifier
real_time_classifier.py - Stream simulation → Window decision engine: feeds stream
gercek_zamanli_sim.py - Stream simulation → Alarm or protection: publishes alarm
gercek_zamanli_sim.py - Runtime stress test → Window decision engine: exercises engine
stres_testi.py - Fast stress test → Saved models: loads classifier
stres_testi_hizli.py - Fast stress test → Signal utilities: uses data helpers
stres_testi_hizli.py - Saved models → C++ model export: supplies trained model
README.md - C++ model export → Embedded model header: generates header
README.md - Embedded model header → ESP32-S3 firmware: included for inference
README.md