Real-Time Person Detection with Landmark Detection and Depth Estimation

This project is aimed at detecting persons in real-time video streams, drawing landmarks on their bodies, and estimating the depth of each detected person. It utilizes computer vision techniques and deep learning models to achieve these tasks.

Features

• Real-Time Person Detection: Utilizes YOLOv5, a state-of-the-art object detection model, to detect persons in real-time video streams.
• Landmark Detection: Employs MediaPipe Pose, a powerful pose estimation library, to detect landmarks on the bodies of detected persons.
• Depth Estimation: Utilizes MiDaS, a depth estimation model, to estimate the depth of each detected person.
• Visualization: Draws bounding boxes around detected persons, landmarks on their bodies, and overlays depth maps on the video frames for visualization.

Prerequisites

Before running the project, ensure you have the following dependencies installed:

• Python 3.x
• OpenCV
• Torch
• NumPy
• Mediapipe
• Timm

You can install the dependencies by running:

pip install opencv-python opencv-python-headless torch torchvision matplotlib timm mediapipe

Usage

To run the project, execute the following command:

py app.py

Output

The output will be displayed in a new window showing the real-time video stream. Detected persons will be outlined with bounding boxes, landmarks will be drawn on their bodies, and depth maps will be overlayed on the frames.

** With landmarks drawing

** Without landmarks drawing

Acknowledgments

• YOLOv5: Ultralytics/YOLOv5
• MiDaS: Intel-isl/MiDaS
• MediaPipe: Google/MediaPipe

License

This project is licensed under the MIT License.