Auto-stereoscopic Light-Field Display By: Jesus Caban George Landon

Stereoscopic Displays Stereoscopic displays: a 3-D display type that presents a left view of the imaged scene to the left eye and a right view to the right eye. Stereoscopic displays: a 3-D display type that presents a left view of the imaged scene to the left eye and a right view to the right eye. Examples: Examples: head-mounted displays head-mounted displays active stereo displays active stereo displays polarized displays polarized displays Note: Note: User have to use glasses or devices in order to see the stereo and depth perception User have to use glasses or devices in order to see the stereo and depth perception Most of them are limited to a single viewing perspective Most of them are limited to a single viewing perspective

Auto-Stereoscopic Displays Auto-stereoscopic displays: a 3-D display type that presents concurrent independent views of the imaged scene without special viewing aids. Auto-stereoscopic displays: a 3-D display type that presents concurrent independent views of the imaged scene without special viewing aids. Examples: Examples: Stackable (Volumetric) Displays Stackable (Volumetric) Displays Lenticular Displays Lenticular Displays Holography Displays Holography Displays

Stackable (Volumetric) Displays One or more surfaces that can generate or scatter visible radiation into a volume. One or more surfaces that can generate or scatter visible radiation into a volume. Limitations Limitations Resolution Resolution Cost Cost

Holography Optical Displays Holography Optical Elements are recordings of interference patterns when two or more laser beams intersect. Holography Optical Elements are recordings of interference patterns when two or more laser beams intersect. These make it easy to fabricate a display which puts light from one image into one of the viewer's eyes and light from a second image into the other eye These make it easy to fabricate a display which puts light from one image into one of the viewer's eyes and light from a second image into the other eye From: MIT’s Mark II Holographic Video Display

Lenticular Displays Consist of an array of cylindrical or circular lenses placed in front of interleaved pictures that emits light differently in distinct viewing angles. Consist of an array of cylindrical or circular lenses placed in front of interleaved pictures that emits light differently in distinct viewing angles. At a given viewing angle, only one of the set of interleaved pixels can be seen. At a given viewing angle, only one of the set of interleaved pixels can be seen. As the viewer moves, the perceived image rotates to provide the proper perspective. As the viewer moves, the perceived image rotates to provide the proper perspective. From:

Lenticular Systems From: Acquisition Acquisition Acquire image through calibrated lens array Acquire image through calibrated lens array Display Display Display image using equivalent lens array Display image using equivalent lens array

Lenticular Systems Reference: Epigem Ltd and other partners were De Montfort University, National Physical Laboratory, DRS Hadland Ltd and ImageQuest Ltd The LAIRD Project

Auto-stereoscopic Light-Field Display Goals: Goals: Flexible high-resolution system capable of providing 3D parallax effect and stereoscopic view. Flexible high-resolution system capable of providing 3D parallax effect and stereoscopic view. Multiple Viewers Multiple Viewers Cost Effective Cost Effective Display filter/lens array can be calibrated by regular users in any display configuration Display filter/lens array can be calibrated by regular users in any display configuration Scalable Scalable

Dynamically Reparameterized Light Fields SIGGRAPH 2000 Paper: SIGGRAPH 2000 Paper: Aaron Isaksen, Leonard McMillan, Steven J. Gortler, Dynamically Reparameterized Light Fields Aaron Isaksen, Leonard McMillan, Steven J. Gortler, Dynamically Reparameterized Light Fields Based on: Based on: Extended work on light field and lumigraph Extended work on light field and lumigraph What’s new in this paper? What’s new in this paper? Added ability to vary the apparent focus within a light field using intuitive camera-like controls (i.e. variable aperture and focus ring) Added ability to vary the apparent focus within a light field using intuitive camera-like controls (i.e. variable aperture and focus ring) Parameterization works independently of scene geometry Parameterization works independently of scene geometry

Lens Array By using a microlens (array of tiny lenslets) we can create an array of view dependent pixels. By using a microlens (array of tiny lenslets) we can create an array of view dependent pixels.

View Dependent Pixels We need higher spatial resolution We need higher spatial resolution Only a fraction of the native pixels are visible from any direction Only a fraction of the native pixels are visible from any direction How many views? How many views? The maximum independent number of view for each lenslet is determined by the number of pixels covered. The maximum independent number of view for each lenslet is determined by the number of pixels covered.

Color Pixelization In LCD panels a pixel consists of RGB triad In LCD panels a pixel consists of RGB triad If from a lenslet only one element of the triad is visible, only one color can be displayed in that viewing direction. If from a lenslet only one element of the triad is visible, only one color can be displayed in that viewing direction.

Calibration We need to know the correspondence between framebuffer and 3D rays We need to know the correspondence between framebuffer and 3D rays Each lenslet can be modeled as a pinhole camera Each lenslet can be modeled as a pinhole camera Focal length (provided by the manufacturer) Focal length (provided by the manufacturer) Principal point of a lenslet Principal point of a lenslet A simplified calibration matrix can be define for each A simplified calibration matrix can be define for each defined for camera translation defined for camera translation defined as rotation defined as rotation By using a single calibrated camera and their relative position, determine the framebuffer mapping. By using a single calibrated camera and their relative position, determine the framebuffer mapping.

Calibration Setup

Calibration Results Green Pixels Visible by Left Camera Blue Pixels Visible by Right Camera Red Pixels Visible by Both Cameras

Calibrated lens array Left viewRight view

Left viewRight view

Open Issues Calibration is SLOW! Calibration is SLOW! 512*512 pixels window 512*512 pixels window 262,144 (right/left images)  524, ,144 (right/left images)  524,288 Is there is a way to speed up this process? Is there is a way to speed up this process? Can we do this in parallel? Can we do this in parallel? How can we identify which pixel is which pixel? How can we identify which pixel is which pixel? Can we use this information to develop a calibration model? Can we use this information to develop a calibration model? Pixelization problem Pixelization problem Can we calibrate for R, G and B? Can we calibrate for R, G and B? 524,288 * 3  1,572, ,288 * 3  1,572,864

References Aaron Isaksen, Leonard McMillan, and Steven J. Gortler. Dynamically Reparameterized Light Fields. In Proceedings of SIGGRAPH 2000, pages 297–306, August Michael Halle. Autostereoscopic Displays and Computer Graphics. Computer Graphics, 31(2):58–62, Steven J. Gortler, Radek Grzeszczuk, Richard Szeliski, and Michael F. Cohen. The Luminograph. SIGGRAPH 96, pages 43-54, 1996