Fall-Detection-Using-CV

This project uses a combination of computer vision techniques, including the YOLOv5 object detection model and Mediapipe pose estimation, to detect falls in real-time. The system analyzes video feeds from a webcam or video file to identify people, estimate their posture, and detect potential falls. Upon detecting a fall, it triggers an alert sound to notify caregivers or monitoring personnel.

Table of Contents

• Features
• Installation
• Usage
• How It Works
• Configuration
• Future Improvements
• License

Features

• Real-time fall detection using a webcam or video file.
• Detects falls by monitoring a person's posture and dimensions.
• Triggers an alert sound when a fall is detected.
• Customizable frame skipping to optimize performance.
• Visual output with bounding boxes around detected individuals.

Installation

Requirements

• Python 3.7 or higher
• OpenCV
• PyTorch
• Mediapipe
• Cvzone (for bounding box and text display)
• Winsound (for Windows-based sound alerts)

Steps

1. Clone the repository:

   git clone https://github.com/Johnlorance/Fall-Detection-Using-CV.git

2. Install dependencies:

   pip install opencv-python torch numpy mediapipe cvzone

3. Configure YOLOv5:
   The pre-trained YOLOv5 model will be downloaded automatically when you first run the script.

4. Add alert sound file (optional for sound alerts):
   Ensure you have a valid .wav file for alert sounds.
   Update the alert_sound_path in the script with the path to your .wav file.

Usage

1. Run the script:

   • To start fall detection using your webcam:

     python fall_detection.py

   • To use a video file as input, modify the code in monitor_falls():

     monitor_falls("path_to_video.mp4")

   • Replace "path_to_video.mp4" with the path to your video file.

2. Quit the program:
   Press q while the video window is active to exit the program.

How It Works

1. Person Detection:
   YOLOv5 is used to detect people in the frame. Only the "person" class (class_id = 0) is processed.

2. Pose Estimation:
   Once a person is detected, Mediapipe's pose estimation is used to estimate the posture and identify if the person has fallen.

3. Fall Detection Logic:
   A fall is determined based on the ratio of height to width of the detected person's bounding box.
   If the person's height becomes smaller than their width (indicating a horizontal position), a fall is flagged.

4. Alert System:
   If a fall is detected, a separate thread is used to play an alert sound until the person recovers or the program is stopped.

Configuration

• Video Source:
  The monitor_falls() function can take a webcam (default 0) or a video file as input:

  monitor_falls(video_source=0, skip_frames=5)