Automatic Projector Calibration with Embedded Light Sensors Johnny

Automatic Projector Calibration with Embedded Light Sensors Johnny C. Lee 1, 2 Paul H. Dietz 2 Dan Maynes-Aminzade 2, 3 Ramesh Raskar 2 Scott E. Hudson 1 1 Carnegie Mellon University 2 Mitsubishi Electric Research Labs 3 Stanford University Santa Fe, NM UIST 2004

Introduction to Projection

Projector Calibration

Our Approach - Embed light sensors into the target surface - optical fibers channel light energy from each corner to sensors - USB connection to the PC - White front surface hides fibers and acts as a light diffuser

Calibration Demonstration of calibration process

Gray Code Patterns - Binary sequence where only 1 -bit changes from one entry to the next. - Robust spatial encoding property • Frequently used in Range-Finding systems

Binary Gray 0000 0001 0010 0011 0010 0100 0110 0101 0111 0100 1000 1100 1001 1101 1010 1111 1011 1110 1100 1010 1101 1011 1110 1001 1111 1000

Binary Gray

Scalability and Robustness - Pattern count = log 2(pixels) - Constant time with respect to # of sensors - Decoding location requires only one XOR operation per location bit (cheap & fast) - Robust against inter-pixel sensor positioning - Robust against super-pixel size sensors - Accurate to the nearest pixel when in focus - Degrades gracefully in under defocusing - Strong angular robustness

Angular Robustness & Mirrors Demonstration Video

Optical Path Optical path between the projector and the sensor does not need to be known. Pixel location of a sensor can be found so long as there exists a path. Additional sensors in the target surface can increase robustness to partial occlusion.

Application Demonstrations Demonstration Video

Research Applications Everywhere Displays, IBM Digital Merchandising, MERL Shader. Lamps, projector AR, UNC/MERL

Other Applications Cheap, light-weight displays Projector array stitching - data walls - planetariums Redundant projector alignment - shadow reduction - stereoscopic displays - increasing brightness - high-dynamic range display

Trade Offs l Digital correction inherently sacrifices pixels and resamples the image. – – Image filtering Higher resolution projectors Pan-Tilt-Zoom projectors (preserve pixel density) Optical correction l Requires instrumented surface – Not a problem for some high Qo. S applications – Removable/reusable wireless calibration tags

Future Work l Interactive Rates - Movable Screens – High speed projection (DLP) – n-ary and RGB Gray Codes – Adaptive Patterns l Imperceptible calibration – High speed steganography – Infrared l Multiple projectors – Smart rooms – 3 D positioning

Concluding remarks • • • Robust Fast Accurate Low-Cost Scalable Applicable in HCI and out

Thanks! Contact Info Johnny Chung Lee johnny@cs. cmu. edu Haptic Pen: A Tactile Feedback Stylus for Touch Screens Wednesday 3 pm session

Homography Four sensor coordinates are used to compute a homography – (loosely) a transformation between two coordinate spaces. • Automatically flips image in the presence of mirrors. • Works with Open. GL and Direct. X matrix stacks for real-time warping on low-cost commodity hardware. • Warping extends beyond the bounds of the sensors (internal feature registration, characterization) • If more than 4 sensors are use, subpixel accuracy can be achieved through best-fit solutions

vs. Camera Based Approach l Standard computer vision problems – – Background separation Variable lighting conditions Material reflectance properties Non-planar/Non-continuous surfaces can be difficult l Accurate registration to world features requires high resolution cameras – Expensive (and high-speed is even more expensive) – High-computational overhead (Pentium vs. PIC) l Rigid camera-projector geometry – Requires calibration – Zooming may be problematic l Not as flexible – Projector stitching/Redundancy – Shader. Lamps/Non-planar surfaces