Welcome to StreamPoseML Documentation¶
Turning human movement into machine learning insights¶
StreamPoseML is an open-source toolkit for creating real-time, video-based movement classification applications. Whether you’re a researcher studying movement patterns, a developer building interactive applications, or an artist exploring interactive technology, StreamPoseML helps you transform video of human movement into actions.
Choose Your Path¶
I want to get started quickly
Jump right in with a quick-start guide to see results in minutes.
Quick Start Guide →I want to understand concepts
Learn core concepts behind pose detection and feature engineering.
Core Concepts →StreamPoseML gives you two powerful components:
Python Package: A flexible toolkit for pose extraction, dataset creation, and model training that you can use in your Python projects
Web Application: A ready-to-use application for real-time pose classification from webcams or video files
Common Tasks¶
Process videos to extract pose keypoints Learn how →
Create labeled datasets for machine learning Guide →
Train models to classify movements Example →
Deploy for real-time classification Web App Guide →
Key Features¶
Powerful Pose Detection: Extract accurate body keypoints from videos using MediaPipe’s BlazePose
Smart Feature Engineering: Automatically calculate angles, distances, and other features from raw keypoints
Flexible Dataset Creation: Various tools for creating and transforming machine learning datasets
Streamlined Model Building: Train, evaluate, and deploy classification models with minimal code
Real-time Classification: Process live video streams for immediate feedback
Web Integration: Deploy models in browser-based applications
Documentation Structure¶
Getting Started
- Getting Started Guide
- Quick Start Guide
- Core Concepts
- Understanding how StreamPoseML works
- Pose Detection and Keypoint Extraction
- From Frames to Movement: Sequence Processing
- Making Movement Measurable: Feature Engineering
- Building Your Movement Library: Dataset Creation
- Teaching Computers to Recognize Movements: Model Training
- Putting It All Together: Real-time Classification
- Where to Go From Here
- Installation
Tutorials & Workflows
- Workflows
- Video Processing Workflow
- Complete Example Walkthrough
- Step 1: Import StreamPoseML
- Step 2: Set Input and Output Directories
- Step 3: Generate Keypoints and Sequence Data
- Step 4: Merge Video Sequence Data into a Dataset
- Step 5: Prepare Training Data
- Step 6: Model Training Approaches
- Step 7: Train a Random Forest Model (Alternative)
- Step 8: Save Model for Web Application Deployment
- Alternative Dataset Formats
- Using Models in the Web Application
Reference
Web Application
Development