How cdtalley/fall-detection---computer-vision works
Project type and purpose
• This is a privacy-preserving, real-time fall-risk detection system built around a computer-vision/ML pipeline.
• It consists of two user-facing front ends—a Next.js marketing/demo site and a Streamlit dashboard—and a Python-based processing backend (pose estimation → feature extraction → ML classification → alerting).
File structure
• Top-level:
• frontend/ – static Next.js app (marketing site, demo launcher)
• src/ – core Python modules (data_processing.py, model_training.py, realtime_inference.py, performance_analysis.py)
• streamlit_app/ – live monitoring dashboard (realtime_fall_detection.py)
• notebooks/ – exploratory/model investigation notebooks
• quick_start.py – helper to kick off the pipeline end-to-end
• reports/ – MD reports on model performance
• Configuration/build: package.json, tsconfig.json, next.config.js in frontend; requirements.txt at root
Extract insights from README
• System pipeline:
Video Input → MediaPipe Pose Estimation → Keypoint Feature Extraction → LSTM/CNN Classification → Streamlit Alert Dashboard
• Tech stack: OpenCV, MediaPipe, TensorFlow, scikit-learn, Streamlit; data analysis via Pandas/NumPy/Matplotlib