Pose Estimation

Real-time pose detection from camera using RTMPose and ONNX Runtime, built with Flutter.

Live demo on GitHub Pages

This project is also a practical example of cross-platform camera access in Flutter using flutter_webrtc — a single implementation that works on all six platforms without platform-specific code.

Platforms

Platform	Status
Android	✅
iOS	✅
Web	✅
Windows	✅
Linux	✅
macOS	✅

Features

• Real-time 17-keypoint body pose detection
• Runs fully on-device via ONNX Runtime (no server required)
• Camera input via WebRTC — works on mobile, desktop, and browser
• RTMPose-t model — lightweight and fast

Screenshot

Getting Started

flutter pub get
flutter run -d chrome          # web
flutter run -d windows         # Windows desktop
flutter run -d macos           # macOS desktop
flutter run -d linux           # Linux desktop
flutter run                    # Android or iOS (device/emulator)

Prerequisites

The ONNX model file must be placed at assets/models/rtmpose.onnx.
See assets/models/README.md for download instructions.

Architecture

lib/
  main.dart              — app entry point
  screens/
    pose_screen.dart     — camera + inference UI
  services/
    pose_detector.dart   — ONNX Runtime inference
    camera_service.dart  — WebRTC camera capture
  painters/
    skeleton_painter.dart — keypoint overlay rendering

Model

RTMPose-t exported to ONNX format.
Input: 1 × 3 × 256 × 192 (RGB, normalized).
Output: 1 × 17 × 3 (x, y, score per keypoint, COCO format).

Tech Stack

Component	Library
UI	Flutter 3.x
Inference	flutter_onnxruntime
Camera	flutter_webrtc
Image processing	image

License

MIT