Fall Detection AI

Inspiration

As people age, they become increasingly vulnerable to falls and serious injuries. Falls are the leading cause of injury-related hospitalisations for older adults, and when they occur, immediate medical attention is often critical. Traditional monitoring systems either require wearable devices that people forget to use or rely on periodic check-ins that may miss critical moments. We built this system to provide continuous, non-intrusive monitoring that can detect falls in real-time and immediately alert caregivers or emergency services.

Purpose & Use Cases

For Elderly Care:

Monitor aging parents or grandparents who live independently
Detect falls in assisted living facilities and nursing homes
Provide peace of mind for family members who can't be present 24/7
Reduce response time to medical emergencies

For People with Disabilities:

Monitor individuals with mobility impairments who may be prone to falls
Support people with balance disorders or seizure conditions
Assist caregivers in monitoring multiple people simultaneously

Other Applications:

Rehabilitation progress tracking
Post-surgery recovery monitoring
Child safety in high-risk areas
Workplace safety in hazardous environments

The beauty of this system is that it's completely non-intrusive - no one needs to wear anything or remember to activate it. It just works in the background, providing a safety net that can literally save lives.

What It Does

Our fall detection system uses live video monitoring to continuously watch for potential falls. When the AI algorithm detects a fall pattern - such as rapid downward movement, sudden impacts, or someone ending up in a horizontal position - it immediately triggers a warning notification on screen. The system also activates an AI chatbot interface, allowing the person to verbally confirm whether they're okay or need immediate assistance. This dual approach ensures both automated detection and human verification.

How We Built It

The system combines a Python backend with a JavaScript/HTML/CSS frontend, connected through Flask, allowing live communication between video processing and user interaction.

But the main highlight of our product is in how we detect collisions and falls. Video processing and pose estimation are handled by MediaPipe, which provides real-time body landmark detection. The core fall detection logic analyses forces and movements using our custom BodyData class in track.py, powered by NumPy for mathematical calculations.

Challenges We Ran Into

The biggest technical hurdle was developing an accurate fall detection algorithm. We needed to distinguish between normal activities like jumping, intentional lying down, and actual falls. Getting precise measurements of body forces and camera positioning was complex, requiring careful calibration of velocity thresholds, acceleration patterns, and body orientation analysis. The algorithm had to be sensitive enough to catch real falls while avoiding false positives from normal movement.

Accomplishments We're Proud Of

In such a short timespan, we successfully created a working fall detection algorithm that can identify falls in real-time. Each member of our 4-man team knew exactly what they were responsible for and in the end we were able to combine our different expertise and skills together for this project. Most importantly, we believe we built something that could genuinely help people stay safe and get help when they need it most.

What We Learned

This project taught us how to integrate MediaPipe for real-time pose estimation and design algorithms that can analyse complex body movements. We gained experience with force calculations, video processing, and building AI systems that need to work reliably in real-world conditions. The challenge of balancing detection sensitivity with accuracy gave us valuable insights into computer vision applications.

What's Next for Fall Detection AI

The fall detection algorithm has room for improvement in accuracy and false positive reduction. Future enhancements could include machine learning integration for better pattern recognition, mobile app development for remote monitoring, and integration with smart home systems for automated emergency responses or even a live chatbot that you can talk to live and assess your injury. We're also exploring ways to make the system more accessible for different living environments and user needs.