Human Fall Detection

Introduction

This project is a human fall detection system based on the MoveNet model. The system is able to detect human fall in real-time using a video or webcam.

Detection Method

• The system uses the MoveNet model to detect the keypoints of the human body
• The keypoints are then used to calculate:
  • The x and y coordinates of the nose that represent the head position
  • The aspect ratio of the body that represents the body position
  • The angle of centroid of the body with respect to the x-axis that represents the body orientation
• The head position and body position are then used to determine the fall detection
• The body orientation is then used to confirm the fall detection

Usage

1. Clone the repository
2. Install the required packages
3. Change the input based on your needs

3.1 For video input, change the queda.mp4 in VideoCapture to the path of your video file

# Insert video file name here
cap = cv2.VideoCapture('queda.mp4')

3.2 For webcam input, use the following code

# If using webcam, uncomment the line below and comment the line above
cap = cv2.VideoCapture(1)

4. Run the code
5. Press esc to exit the program

References

• MoveNet
• Video Fall Detection
• TensorFlow Multipose Estimation by Nicholas Renotte
• Automatic Detection of Human Fall in Video by Vinay Vishwakarma, Chittaranjan Mandal, and Shamik Sural