safe-fall-detection

Affordable Fall Detection for Safer Living.

Comment

Fall Detection and Emergency Alert System

Project Overview

This project leverages YOLO-based computer vision to detect falls from video input and trigger emergency alerts if a person remains on the ground for more than 10 seconds.

It is implemented as a Streamlit web app, allowing users to upload videos or (in future versions) use live webcam feeds for real-time monitoring. The system notifies family members and nearby hospitals, providing location, contact info, and fall snapshots for faster response.

An affordable, open-source solution designed to save lives for elderly individuals, patients, and workers in high-risk environments.

Current Features & Functionality

  • Detects human activities: Walking, Sitting, Fall Detected
  • Starts a fall timer automatically
  • Alerts triggered if the person remains fallen for ≥10 seconds
  • Outputs processed video with annotated bounding boxes:
    • Green → Walking / Sitting
    • Yellow → Fall detected, countdown running
    • Red → Fall alert triggered
  • Family Dashboard: Notified instantly with snapshot, location, and fall details
  • Hospital Emergency Center: Nearby hospitals (3 by default) notified with patient info, location, and fall snapshot
  • The ‘Get Directions & Dispatch Ambulance’ button opens Google Maps with directions from the hospital to the patient’s location, showing how the system can speed up emergency response.
  • Optimized for performance by analyzing every 2nd frame without delaying alerts
  • Base64 snapshot storage allows quick display in dashboards

Demo & Current Limitations

  • Users can upload a video for fall detection analysis
  • Processed video can be downloaded with annotations
  • Alerts are simulated for demo purposes:
    • 3 nearby hospitals notified in demo
    • Family dashboard displays fall snapshots and alert info
  • Live webcam monitoring is possible but currently limited:
    • Streamlit’s live webcam support is not optimized for real-time high FPS
    • Video upload is recommended for smooth demo

Who It Helps

  • Elderly people living alone → reduces risk of serious injuries
  • Patients in hospitals or care centers → enables timely intervention
  • Industrial and construction workers → improves workplace safety
  • Caregivers and healthcare staff → provides real-time awareness of dangerous falls
  • Communities without access to expensive hardware → offers a low-cost, open-source solution

Why It Matters

Falls are a leading cause of injury worldwide. Early detection can save lives and reduce recovery time:

  • 37.3 million severe falls annually require medical attention (WHO)
  • Rapid alerts improve response times for caregivers
  • Provides accessible, low-cost monitoring for individuals and small facilities

Measurable Impact

  • Accuracy: YOLO model trained on Walking, Sitting, and Falling datasets
  • Threshold: Alerts triggered for falls lasting ≥10 seconds (customizable)
  • Demo: Streamlit app allows video upload and annotated download
  • Real-time detection ensures no missed alerts, even with frame-skipping optimization

Tech Stack

  • Python
  • OpenCV
  • Ultralytics YOLO
  • Streamlit
  • NumPy
  • Base64 for snapshot storage

Current Workflow

  1. User uploads a video or uses a live demo feed
  2. YOLO detects human activity in frames
  3. If a fall is detected and persists >10 seconds:
    • Snapshot is captured
    • Emergency alert sent to family dashboard
    • Nearby hospitals notified (demo: 3 hospitals)
  4. Dashboards display:
    • Fall snapshot
    • Patient location and contact
    • Fall duration
    • List of hospitals notified
  5. Processed video with bounding boxes is available for download

Future Roadmap

  • Multi-Person Detection – monitor multiple individuals simultaneously
  • Real-Time Notifications – integrate with SMS, WhatsApp, or email for instant alerts
  • Live Webcam Integration – support real-time monitoring in care centers or homes
  • Wearable & IoT Integration – combine with smart devices for health monitoring
  • Cloud & Edge Deployment – run on Raspberry Pi, Jetson Nano, or cloud for live monitoring
  • Dataset Expansion & Accuracy – improve detection under diverse lighting, angles, and body types
  • User-Friendly Dashboards – mobile/web dashboards for caregivers to view live alerts and history
  • Accessibility & Inclusivity – multilingual alerts, voice notifications for visually impaired users
  • Real-Time Family & Hospital Alerts – integrate IoT sensors, wearable devices, and live cameras so that falls are detected immediately and family + nearby hospitals are notified automatically with location and patient details.

Impact & Social Value

  • Open-source, low-cost, and scalable solution
  • Empowers families and caregivers to monitor safety without expensive hardware
  • Deployable in homes, hospitals, or workplaces
  • Designed for real-time emergency response: as soon as a fall is detected, family members and nearby hospitals receive instant alerts with location and fall details, enabling faster medical assistance. Integration with IoT devices, cameras, and wearables is planned for continuous, automated monitoring.
  • Ready-to-use demo for education, research, and community safety programs

Contributing

We welcome contributions from the community:

  • Fork the repository
  • Submit issues or pull requests for improvements
  • Add new features, datasets, or optimizations

How to Run Locally

# Clone the repo
https://github.com/tumblr-byte/safe-fall-detection.git
cd safe-fall-detection

# Install dependencies
pip install -r requirements.txt

# Run the Streamlit app
streamlit run app.py

Built With

Share this project:

Updates