Real-Time Scoliosis Posture Screening Chair

ISEF Category: Robotics and Intelligent Machines

Subcategory: Biomechanics  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: Full Year

The Hook

A spine can look almost straight from one angle and curved from another. That makes scoliosis screening tricky, even for trained eyes. Your project turns a chair, two cameras, and pose estimation into a fast screening tool. You can test whether software can track posture well enough to estimate spinal asymmetry in real time.

What Is It?

This project asks a simple question, can a camera system measure body alignment well enough to flag uneven posture? MediaPipe pose is software that finds body landmarks, like shoulders, hips, and knees, in video. You can use those landmarks to build a Cobb-angle proxy, which is not the same as the medical Cobb angle, but a repeatable estimate of trunk tilt or asymmetry.

Think of it like using a ruler to estimate the slope of a roof from a photo. The photo does not give you the full roof shape, but it can still tell you if one side is higher than the other. Your chair setup gives you a fixed viewing angle, two webcams give you depth clues, and marked landmarks on a phantom spine or volunteer help you check how close your estimate comes to a known reference.

Why This Is a Good Topic

This makes a strong science fair topic because you can test a real measurement problem, not just build a gadget. The system has clear variables you can change, like camera angle, landmark placement, posture type, or lighting. It also connects to a real screening need, since early scoliosis checks depend on finding asymmetry before it gets worse. You can learn computer vision, calibration, error metrics, and how to validate a model against a physical reference.

Research Questions

• How does camera placement affect the accuracy of Cobb-angle proxy estimates?
• What is the effect of landmark choice on the repeatability of posture asymmetry measurements?
• Does adding a second webcam improve agreement with phantom spine reference angles?
• To what extent does lighting variation change MediaPipe landmark stability?
• Which body posture features best predict the largest Cobb-angle proxy error?
• How does marked anatomical landmark validation compare with phantom spine validation?
• What is the effect of seat height or chair angle on measured trunk asymmetry?

Basic Materials

• Two webcams with tripod mounts or fixed stands.
• Computer with a USB port and enough processing power for live video.
• Chair with a stable backrest and fixed seat height.
• Measuring tape or ruler.
• Adhesive skin-safe markers for anatomical landmarks.
• Printed posture target sheet or grid background.
• 3D-printed phantom spine or a simple rigid curved model.
• Calibration board or printed checkerboard pattern.
• Notebook or spreadsheet for recording measurements.

Advanced Materials

• Two synchronized webcams with manual exposure control.
• Computer with a GPU for faster pose tracking.
• 3D-printed phantom spine with known reference geometry.
• Motion capture markers or reflective markers for comparison.
• Calibration wand or checkerboard calibration target.
• Adjustable chair frame or mounting rig.
• Open-source biomechanics reference dataset for validation.
• Force plate or pressure mat for posture context, if available.
• Digital angle gauge for bench reference checks.

Software & Tools

• MediaPipe: Detects body landmarks from webcam video and gives pose coordinates for analysis.
• Python: Runs your pose pipeline, data processing, and statistics.
• OpenCV: Captures video, calibrates cameras, and processes frames.
• ImageJ: Helps inspect images, measure alignment, and compare annotated frames.
• Jupyter Notebook: Lets you organize analysis, plots, and error calculations in one place.

Experiment Steps

1. Define the exact posture signal you want to estimate, then decide how your Cobb-angle proxy will translate body landmarks into one score.
2. Choose a validation target, such as a 3D-printed phantom spine or marked volunteer landmarks, and decide how you will treat it as a reference standard.
3. Design a camera layout that keeps the viewpoint fixed, then plan how you will calibrate both webcams into the same coordinate system.
4. Build a data sheet for recording landmark positions, frame quality, and reference angles so you can compare trials consistently.
5. Plan the control tests that separate software error from posture variation, such as repeated trials, lighting checks, and camera-position changes.
6. Decide how you will judge success, such as mean absolute error, correlation with the reference, and repeatability across trials.

Common Pitfalls

• Using loose camera mounts, which shifts the viewpoint and makes the proxy angle drift between trials.
• Letting the subject rotate their torso slightly, which changes the measured asymmetry more than the spine itself.
• Trusting raw MediaPipe landmarks without smoothing, which makes frame-to-frame jitter look like posture change.
• Comparing volunteer posture to a phantom spine without matching the body reference points, which makes validation misleading.
• Changing lighting or background texture during data collection, which reduces landmark detection quality.

What Makes This Competitive

A class-level version just shows that the system works. A stronger version tests how and when it fails. You could compare two camera geometries, compare volunteer data against a phantom spine, and report error bars, repeatability, and agreement metrics. A competitive project also explains which design choice matters most, and why.

Project Variations

• Use a smartphone stereo rig instead of webcams to test whether cheaper cameras still track posture well.
• Compare seated posture screening with standing posture screening to see which setup gives cleaner landmark data.
• Swap the phantom spine for ultrasound or marker-based volunteer validation if your lab has access to that equipment.

Learn More

• MediaPipe documentation: Read the pose tracking overview and landmark model notes on the official Google Developers site.
• OpenCV documentation: Find camera calibration, video capture, and image processing guides on the official OpenCV site.
• NIH PubMed: Search for review articles on scoliosis screening, posture assessment, and biomechanics validation.
• NCBI Bookshelf: Look for free textbook chapters on human movement, posture, and musculoskeletal anatomy.
• MIT OpenCourseWare: Search for free computer vision and image processing course materials that cover calibration and feature detection.
• Journal of Biomechanics: Search recent papers on posture measurement, spinal alignment, and validation methods through journal pages or PubMed.