Human Pose Estimation using Machine Learning

A real-time human pose estimation system that processes images, videos, and webcam feeds using OpenCV DNN and MediaPipe.

Table of Contents

• Overview
• Features
• Installation
• Usage
• Author

---

Overview

This project estimates human poses using a hybrid approach:

• OpenCV DNN for images and pre-recorded videos.
• MediaPipe for real-time webcam processing.
• Streamlit for an interactive user interface.

Ideal for applications in sports analysis, healthcare monitoring, and surveillance.

---

Features

• ✅ Multi-input support (Image/Video/Webcam).
• ✅ Adjustable confidence threshold and frame skip.
• ✅ Real-time processing with optimization for edge devices.
• ✅ Customizable background image with transparency.

---

Installation

Prerequisites

• Python 3.8 - 3.12
• pip

Steps

1. Clone the Repository

   git clone https://github.com/0monish/Human-Pose-Estimation-using-ML.git  
   cd Human-Pose-Estimation-using-ML  

2. Install Dependencies

   pip install opencv-python==4.5.5.64 streamlit==1.13.0 numpy==1.23.5 mediapipe==0.8.11 Pillow==9.4.0  

---

Usage

Launch the App

streamlit run human_pose_estimation.py  

Select Input Type

• Image: Upload a JPG/PNG/etc file.
• Video: Upload an MP4/MOV/etc file and adjust frame skip/resolution.
• Webcam: Start live pose estimation.

Adjust Parameters

• Confidence Threshold: Lower values detect more joints (may include noise).
• Frame Skip: Process every 2nd/3rd/4th frame for faster video analysis.
• Processing Width: Reduce for faster inference (e.g., 320px).

---

Author

Monish Khandelwal