🧠 Human Fall Detection System

🔗 Live Demo

Click here to try the app live

Overview

A computer vision system that automatically detects human fall events in video using deep learning. Built entirely as a self-initiated project to explore real-world applications of AI in safety and healthcare monitoring.

The system analyzes uploaded videos frame by frame, sends each frame to a Roboflow-hosted deep learning model, and classifies human posture as either "Fall Detected" or "Standing" in real time.

What I Built

• Integrated Roboflow's object detection API for pose-based fall classification
• Built a fully interactive web application using Streamlit
• Implemented frame extraction and real-time inference pipeline using OpenCV
• Added confidence threshold control so users can tune detection sensitivity
• Deployed the app live on Streamlit Cloud

Tech Stack

• Python
• OpenCV — frame extraction and image processing
• Roboflow API — deep learning model inference
• Streamlit — web app interface and deployment
• NumPy — array and image data handling

How It Works

1. Upload a video (MP4, MOV, AVI)
2. Frames are extracted using OpenCV
3. Each frame is sent to Roboflow model via REST API
4. Model returns fall or standing classification with confidence score
5. Result is displayed live in Streamlit UI

How to Run Locally

pip install -r requirements.txt

streamlit run fall_detection_streamlit/project_streamlit.py

Developer

Hetanjali Vaghela

B.E. Robotics & Automation Engineering | LDCE Ahmedabad