MultiPose Real-Time Human Pose Estimation

A real-time multi-person human pose estimation system built using TensorFlow MoveNet Multipose (Lightning) and OpenCV. This project detects and visualizes human keypoints and skeletal connections from both video files and live webcam streams with high efficiency and low latency.

---

📌 Project Overview

This project implements a real-time multi-person pose estimation pipeline capable of detecting multiple humans simultaneously and rendering 17 anatomical keypoints per person.
The system is optimized for speed and accuracy, making it suitable for real-world computer vision applications.

The architecture leverages a MobileNetV2 image feature extractor, combined with a Feature Pyramid Network (FPN) decoder and CenterNet-style prediction heads, enabling fast inference while maintaining robustness.

---

🧠 Model Architecture

https://www.kaggle.com/models/google/movenet/tensorFlow2/multipose-lightning/1?tfhub-redirect=true

• Backbone: MobileNetV2 (Depth Multiplier: 1.75)
• Decoder: Feature Pyramid Network (FPN) with stride 4
• Prediction Head: CenterNet-style keypoint detection
• Model Variant: MoveNet Multipose Lightning
• Output Format:
  • 17 keypoints per person
  • Each keypoint: (x, y, confidence score)
  • Supports detection of up to 6 people per frame

---

⚙️ Key Features

• ✅ Real-time multi-person pose estimation
• ✅ Supports both video files and live webcam input
• ✅ Confidence-based keypoint filtering
• ✅ Skeletal connection rendering
• ✅ GPU acceleration support (optional)
• ✅ Modular OpenCV-based visualization pipeline
• ✅ Efficient inference suitable for real-time applications

---

🛠️ Technology Stack

• Programming Language: Python
• Deep Learning Framework: TensorFlow, TensorFlow Hub
• Computer Vision: OpenCV
• Numerical Computing: NumPy
• Visualization: OpenCV drawing utilities
• Hardware Acceleration: GPU (optional)

---

🎥 Input Sources

• Pre-recorded video files (MP4)

• Live webcam stream for real-time pose estimation

---

📈 Use Cases

• Human activity recognition
• Sports performance analysis
• Gesture recognition systems
• Surveillance and crowd analysis
• Human–Computer Interaction (HCI)
• AI-assisted fitness and posture monitoring

---

🚀 Learning Outcomes

• Hands-on experience with state-of-the-art pose estimation models
• Real-time deep learning inference pipeline design
• Integration of deep learning models with OpenCV
• Confidence-based post-processing and visualization
• Practical exposure to multi-person detection challenges

📌 Notes

• The project includes both video-based and webcam-based inference pipelines.
• GPU usage is optional and automatically configured when available.
• Confidence thresholds can be adjusted to tune detection accuracy.

🧑‍💻 From Author

Developed as part of an AI/ML-focused computer vision project to demonstrate real-time deep learning inference, multi-person pose estimation, and applied computer vision engineering.