Behaviour, Fatigue & Fall Detection in Patients

Overview

This project aims to monitor patients’ behavior, fatigue, and fall events in real-time using computer vision techniques.
By leveraging MediaPipe, a framework for building multimodal applied ML pipelines, the system tracks human body movements, analyzes motion and facial patterns, and identifies anomalies such as falls or fatigue.

It can be integrated into healthcare monitoring systems for elderly care, hospital surveillance, and rehabilitation centers.

---

Features

• Real-time Pose Estimation: Uses MediaPipe Pose to extract body keypoints.
• Fall Detection: Detects sudden posture or vertical displacement changes.
• Fatigue Detection: Monitors facial landmarks (eyes/mouth) for drowsiness or fatigue.
• Behavioral Analysis: Identifies activity types or abnormal motion patterns.
• Alert System: Prints or triggers alerts upon detection.
• Modular Design: Each module (behavior, fatigue, fall detection) can be used independently.

---

Requirements

Python: 3.7 or higher

Dependencies:
Install all required libraries:

pip install -r requirements.txt

Run the Main Script

python main.py

What to Expect

• The webcam feed will open and start tracking movement in real time.
• The system continuously estimates pose and facial landmarks, monitoring for abnormal patterns.
• When a fall or fatigue event is detected, an alert is printed to the console (you can configure external notifications).

---

Visual Output

• A pose skeleton drawn on the webcam feed.
• Frames with detected falls or fatigue events are highlighted (e.g., in red).
• Real-time visualization of tracking and detection overlays.

---

Implementation Notes (High-Level)

• Uses MediaPipe Pose for full-body keypoints and MediaPipe Face Mesh for facial landmarks.
• Computes heuristics based on landmark positions, velocities, and vertical displacement for fall detection.
• Analyzes blink rate or eye aspect ratio (EAR) for fatigue detection.
• Optionally smooths landmark data and uses temporal windows to reduce false positives.
• Modular structure allows easy extension for additional behavior models.

---

Future Improvements

• Integration with SMS/email notifications for caregivers.
• Incorporation of ML-based classifiers for behavioral anomaly detection.
• Multi-person tracking in a single frame.
• Dashboard for analytics and visualization of logged data.
• IoT integration for real-time emergency alerts.

---

References

• MediaPipe Pose documentation
• OpenCV documentation
• NumPy documentation