Skip to content

galitau/CCCare

Repository files navigation

CareSystem AI — Vision-Based Rehabilitation & Monitoring

CareSystem AI is a comprehensive, real-time exercise tracking and patient monitoring platform developed for the Centre for Community, Clinical and Applied Research Excellence (CCCARE). By integrating advanced computer vision, biometric security, and live medical sensor data, this system automates the supervision of physical rehabilitation for patients with complex clinical needs.


Impact on CCCARE

Traditional rehabilitation monitoring is resource-intensive, often requiring one-on-one supervision to ensure patient safety and form accuracy. CareSystem AI empowers CCCARE by:

  • Scaling Clinical Expertise
    Automated form correction allows a single therapist to oversee multiple patients simultaneously without a drop in quality of care.

  • Objective Progress Tracking
    Replaces manual logs with high-precision, timestamped data on repetitions, joint angles, and physiological responses.

  • Frictionless Workflow
    Biometric identification automatically loads patient history, medications, and specific exercise protocols upon entry.


View It Here

Screenshot 2026-02-15 193630 Screenshot 2026-02-15 200312
CCCare.Video.mp4

Key Features

1. Privacy-First Biometric Identification

  • DeepFace & FaceNet512
    Identifies patients using one-way neural embeddings — mathematical fingerprints that cannot be reversed into photos.

  • Photo-less Database
    Stores numerical vectors rather than raw images, supporting privacy-by-design healthcare compliance.


2. Intelligent Exercise Detection

  • Skeletal Landmark Tracking
    Uses MediaPipe to track 33 body points and calculate joint angles to differentiate between complex movements.

  • State Machine Rep Engine
    Custom logic combined with Exponential Moving Averages ensures repetitions are only counted when full range of motion is achieved.


3. Real-Time Feedback & Coaching

  • Audio Coaching
    Multilingual voice feedback provides actionable guidance (e.g., “Go lower”, “Keep your back straighter”).

  • Visual Debouncing
    Temporal smoothing ensures stable, flicker-free feedback for patients and clinicians.


4. Live Clinical Dashboard

  • WebSocket Streaming
    Streams real-time heart rate and oxygen saturation data from sensors to a React dashboard.

  • Edge Processing
    Video analysis occurs locally while only anonymized analytics are synced to cloud storage.


Safety & Clinical Alerts

CareSystem AI acts as a second set of eyes for therapists:

  • Heart Rate Spike Detection
    Alerts triggered when patient heart rate exceeds configured safety thresholds.

  • Oxygen Level Monitoring
    Alert triggered when abnormal oxygen levels are detected.

  • Full-Body Presence Validation
    Rep counting pauses if critical skeletal landmarks are lost.


Getting Started

Prerequisites

  • Node.js
  • Python 3.9+
  • Webcam
  • Optional: medical sensor hardware

Frontend Setup

npm install
npm run dev

Backend Setup

npm install
npm run dev

Future Roadmap

We plan to evolve CareSystem AI from movement tracking toward predictive rehabilitation intelligence.

Gamification & Engagement

  • AR overlays with interactive virtual targets
  • Game-like progression systems to increase adherence
  • Personalized motivational feedback loops

Predictive Recovery Analytics

  • Machine learning models trained on historical patient data
  • Plateau detection and automated intervention suggestions
  • Adaptive exercise intensity recommendations

Automated Clinical Documentation

  • Session metric summarization
  • EHR-ready export pipelines

Platform Expansion

  • Multi-clinic deployment scaling
  • Role-based clinician dashboards
  • Secure interoperability with hospital systems

About

Vision-based rehabilitation monitoring platform using computer vision and biometric embeddings to detect exercises, count reps, and stream real-time vitals to a clinician dashboard. Built for CCCARE during the HealthTech Innovation Challenge.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors