Pose to Labanotation for Folk Dance Preservation

ISEF Category: Technology Enhances the Arts

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A dance can disappear even when a video survives. If no one can read the movement closely, the style fades frame by frame. Your project can turn motion into a written score, then test how much of the dance comes back when you read it again.

What Is It?

This project sits at the meeting point of computer vision and dance preservation. You use pose estimation, which means software finds body joint positions in video, then map those motions into Labanotation, a symbol system that records human movement. Think of it like turning a song into sheet music, except the notes describe steps, turns, levels, and direction.

The key question is not only whether the system can make symbols. You also need to ask whether those symbols keep enough detail to rebuild the dance. A forward renderer helps here. It takes the notation and generates motion again, so you can compare the recreated dance with the original video. That gives you a direct way to measure how much meaning survived the conversion.

Because folk dances often carry regional identity, this topic has real cultural value. You are not just building a classifier. You are testing whether digital notation can act as a long-term archive for movement.

Why This Is a Good Topic

This is a strong science fair topic because you can measure it in a clear way. You can change the dance type, the camera angle, the notation rules, or the reconstruction method, then compare accuracy with motion metrics. That makes the project testable, and the results are easy to graph. It also connects to cultural preservation, digital archiving, and motion analysis, so your work has a real-world purpose. You can learn pose estimation, data labeling, evaluation design, and basic model comparison, all in one project.

Research Questions

• How does the choice of pose-estimation model affect Labanotation reconstruction accuracy?
• What is the effect of camera angle on the quality of generated dance notation?
• Does adding temporal smoothing improve the match between original and reconstructed motion?
• To what extent does notation detail affect reconstruction accuracy for fast footwork versus arm gestures?
• Which folk-dance movements are easiest to preserve in Labanotation and which are hardest?
• How does the forward renderer's output compare to the original video when evaluated with joint-angle similarity?

Basic Materials

• Laptop with a webcam or phone camera
• Sample folk dance videos with clear permission for research use
• MediaPipe Pose or similar pose-estimation software
• Spreadsheet software for data logging
• Open-source video player with frame-by-frame control
• Basic annotations template for movement symbols
• External storage for video files.

Advanced Materials

• Workstation with a GPU, if available
• High-resolution motion-capture or multi-camera video dataset, if permitted
• Motion analysis software such as Blender or OpenPose-based tools
• Python environment with NumPy, pandas, OpenCV, and plotting libraries
• Labanotation reference materials from a university library or archive
• Annotation software for time-synced movement labeling
• Statistical analysis tools for sequence comparison.

Software & Tools

• MediaPipe: Detects body landmarks in dance video and gives you joint coordinates for notation mapping.
• Python: Helps you clean pose data, apply notation rules, and compare original and reconstructed motion.
• OpenCV: Supports video frame handling, tracking checks, and basic image analysis.
• ImageJ: Measures visual differences between original frames and rendered motion when you need simple image comparisons.
• Blender: Lets you build or inspect a forward renderer for motion reconstruction.

Experiment Steps

1. Define one dance style, one camera setup, and one notation standard before you collect data.
2. Decide which body landmarks and motion features your notation system will encode.
3. Build a small annotated sample set so you can test whether the mapping rules make sense.
4. Plan a forward renderer that turns notation back into motion, then choose a comparison metric for accuracy.
5. Set controls that separate pose-estimation error from notation-design error.
6. Compare multiple dances or movement sections to find where the system preserves motion well and where it fails.

Common Pitfalls

• Using videos with occluded limbs, which breaks pose tracking and creates missing symbols in the notation.
• Mixing different camera zoom levels, which changes apparent joint motion and weakens comparisons.
• Copying notation rules without testing whether they can represent fast turns, foot crossings, or hand flourishes.
• Comparing rendered motion to the original without aligning timing, which makes accurate reconstructions look wrong.
• Treating one short dance clip as enough evidence, which hides how the system fails on other movement patterns.

What Makes This Competitive

A stronger version of this project goes beyond a simple prototype. You can compare multiple pose models, multiple notation rules, or multiple dance families, then test which choices preserve meaning best. A competitive project also uses careful metrics, like joint-angle similarity, sequence alignment, or frame-level reconstruction error. If you can show where the system succeeds, where it fails, and why, your project looks much more like research than a demo.

Project Variations

• Test the system on one regional dance style and compare solo versus group performance capture.
• Focus on hand and arm gestures only, then measure whether upper-body notation is more accurate than full-body notation.
• Compare two reconstruction methods, one rule-based and one machine-learning-based, to see which preserves movement better.

Learn More

• PubMed: Search for review articles on motion analysis, pose estimation, and dance movement science.
• NASA ADS: Search for papers on computer vision methods if you want broader sequence reconstruction ideas.
• MIT OpenCourseWare: Look for computer vision and machine learning lecture materials and assignments.
• Journal of Dance Education: Search for articles on dance notation, preservation, and analysis in peer-reviewed form.
• Computer Vision and Image Understanding: Search for papers on pose estimation, motion capture, and reconstruction metrics.