Registration techniques for using imperfect and partially calibrated devices in planar multi-projector displays

Area of Science:

• Computer Vision
• Display Technology
• Image Processing

Background:

• Multi-projector displays require precise geometric and photometric registration for optimal performance.
• Current registration methods rely on pre-calibrated, perfect projectors and cameras, limiting accessibility.
• Real-world devices often exhibit imperfections like lens distortion and are uncalibrated, posing significant challenges.

Purpose of the Study:

• To develop accessible registration techniques for multi-projector displays using inexpensive, imperfect devices.
• To enable geometric alignment in the presence of severe projector lens distortion.
• To achieve photometric self-calibration with uncalibrated cameras.

Main Methods:

• A novel closed-form geometric registration technique using rational Bezier patches to model projector-camera relationships in planar displays.
• Utilizing a low-resolution camera (640x480 VGA) for calibrating a high-resolution display (3000x2500).
• A photometric self-calibration method for projector-camera pairs without prior camera calibration.

Main Results:

• Successful geometric registration of a 3x3 array of nine severely distorted projectors using a low-resolution camera.
• Demonstration of photometric self-calibration for a multi-projector display with an uncalibrated camera.
• The proposed methods overcome limitations of existing techniques by accommodating geometric and photometric imperfections.

Conclusions:

• The developed techniques significantly advance the accessibility of multi-projector display calibration for layman users.
• This work is the first to successfully calibrate multi-projector displays with geometrically imperfect projectors and photometrically uncalibrated cameras.
• The findings pave the way for more robust and user-friendly tiled projection systems.