design of a bright polarized head-mounted projection display · 2015. 12. 31. · design of a...

Design of a bright polarized head-mounted projectiondisplay

Hong Hua and Chunyu Gao

In optical see-through head-mounted displays, it has been a common challenge that the displayed imagelacks brightness and contrast compared with the direct view of a real-world scene. Consequently, suchdisplays are usually used in dimmed lighting conditions, which limits the feasibility of applying suchinformation displays outdoors or in scenarios where well-lit environments, such as in operation rooms,are required. The lack of image brightness is aggravated in the design of a see-through head-mountedprojection display (HMPD). For instance, the overall flux transfer efficiency of existing HMPD designs isless than 10%. The design of a polarized head-mounted projection display (p-HMPD) is presented. Theimages of a p-HMPD system can potentially be three times brighter than those in existing HMPD designs.It is further demonstrated that the p-HMPD design is able to dramatically improve image brightness,contrast, and color vividness with experimental results. Finally, the design of a compact optical systemand helmet prototype is described. © 2007 Optical Society of America

OCIS codes: 120.2040, 120.2820, 120.4570, 120.4820.

1. Introduction

Mixed- and augmented-reality (MR–AR) technology isa paradigm where computer-generated digital infor-mation is selectively superimposed upon a real-worldscene to supplement a user’s sensory perception of thephysical environment. It has been explored for a widerange of applications for 3D scientific visualization,medical training, and engineering processes.1–3 One ofthe key enabling technologies in MR–AR systems is a3D display that is able to seamlessly combine virtualand real information, which is called creating see-through capability.

Optical see-through head-mounted displays (OST-HMDs) have been one of the basic approaches to com-bining computer-generated virtual objects with theviews of real-world scenes required for MR–AR sys-tems.4 In an OST-HMD, the direct view of the phys-ical world is maintained, and computer-generatedvirtual images are optically superimposed onto thereal scene via an optical combiner. This optical ap-

proach allows a user to see the real world in fullresolution and introduces less intrusion into the viewof the real world than video see-through displayswhere real-world views are captured through videocameras. Therefore an OST-HMD system is preferredfor tasks where eye–hand coordination or a non-blocked real-world view is critical.5

Designing a wide field-of-view (FOV), compact, andnonintrusive OST-HMD, however, has been a chal-lenge. A head-mounted projection display (HMPD),pioneered by Fisher6 and Fergason,7 deviates from theconventional approaches to HMD designs. A sche-matic of a monocular HMPD configuration is illus-trated in Fig. 1. Two major aspects distinguish theHMPD technology from conventional HMDs and pro-jection systems: Projection optics replaces an eyepiece-or microscope-type lens system in a conventional HMDdesign, and a retroreflective screen substitutes for atypical diffusing screen in a conventional projectionsystem. The projected light is thus directly retrore-flected back to the exit pupil of the projection systemwhere the eye is positioned to view the projected im-age. The unique combination of projection and retro-reflection not only enables stereoscopic capabilitybut also provides intrinsically correct occlusion ofcomputer-generated objects by real ones and offersthe capability of designing wide FOV, low-distortionoptical see-through displays. More detailed discus-sions on the pros and cons of the HMPD technologycan be found in Hua et al.8 Recent technology ad-

H. Hua ([email protected]) is with the College of OpticalSciences, University of Arizona, 1630 East University Boulevard,Tucson, Arizona 85721, USA. C. Gao is with the Beckman Insti-tute, University of Illinois at Urbana-Champaign, Urbana, Illinois61801, USA.

Received 13 September 2006; revised 8 January 2007; accepted11 January 2007; posted 18 January 2007 (Doc. ID 75043); pub-lished 23 April 2007.

0003-6935/07/142600-11$15.00/0© 2007 Optical Society of America

2600 APPLIED OPTICS � Vol. 46, No. 14 � 10 May 2007

vancements and applications will be reviewed in Sec-tion 2.

Besides the challenge of designing wide FOV, low-distortion optics, the images of optical see-through dis-plays are commonly lack of brightness and contrastcompared with the direct view of a real-world scene.While the luminance level of an immersive HMD isusually required to be equal to or greater than approx-imately 17 cd�m2 for optimal visual acuity,9 the im-age brightness of an OST-HMD must match theaverage luminance level of its working environ-ments. The average luminance of outdoor scenesis typically approximately 5000 to 6000 cd�m2, anda well-lit indoor environment approximately aver-ages 400–500 cd�m2. The state-of-the-art microdis-plays suitable for HMDs, however, yield on average100 cd�m2 of luminance for backlit active-matrixliquid-crystal displays (AMLCD), 300 to 1000 cd�m2

for liquid crystal on silicon (LCOS) displays, and 50 to600 cd�m2 for organic light-emitting displays (OLED).

The problem of low image brightness and contrastis worsened due to the light attenuation through anoptical combiner interface required in see-throughdisplays, resulting in low luminance transfer efficiencyof the optical system. In conventional OST-HMDs,using a 50�50 beam splitter will attenuate the light,from both a displayed image and the real scene, by50%. Consequently, such displays are usually used indimmed indoor environments, which limits the fea-sibility of applying such information displays outdooror in scenarios where well-lit environments such as inan operation room are necessary.

The low-efficiency problem is aggravated in a see-through HMPD in which the projected light is splittwice through a beam splitter as illustrated in Fig. 1.Using a 50�50 beam splitter leads to the loss of at

least 75% of the light from a displayed image and 50%of the light from the real scene. The light from thedisplayed image is further attenuated by as high as80% through an imperfect retroreflective screen.10

The actual luminance returned back to the exit pupilis approximately 4%–10% or less of the display lumi-nance. For instance, providing the usage of AMLCDs,the observed peak luminance is approximately 4 to10 cd�m2 or lower, which imposes significant restric-tions on the lighting conditions of working environ-ments and limits applications demanding well-litenvironments. Fergason suggested a conceptual im-provement to the technology by adding a secondaryretroreflective material located at 90° from the pri-mary retroreflective surface.7 This approach can in-crease the luminous efficiency of the display, but itcompromises a significant property of the originalHMPD concept, that is, the capability of natural oc-clusion of virtual objects by real objects placed be-tween the user and the retroreflective screen.

In this paper we present the design of a polarizedhead-mounted projection display (p-HMPD) to addressthe image brightness challenge.11 The new designscheme significantly improves the luminance effi-ciency of the display by applying polarization tech-niques. The observed image through the polarizedsystem can potentially be three times brighter thanexisting nonpolarizing HMPD designs. The rest of thepaper is organized as follows. In Section 2 recent ad-vancements in HMPD research and application arereviewed. The conceptual design of a p-HMPD designis described in Section 3. The optical design for a com-pact prototype is described in Section 4, the design of acompact helmet system is described in Section 5, andthe experimental results are presented in Section 6.

2. Related Work

The system described by Fisher was a biocular con-figuration with one microdisplay, and the same im-age was viewed by both eyes.6 Fergason extendedthe biocular concept to binocular displays.7 Sincethese pioneering efforts, the HMPD concept has beenrecently explored extensively by several groups ofresearchers, and the technology has evolved signifi-cantly.8,10–22

In the early work, head-mounted prototypes im-plemented from off-the-shelf components were dem-onstrated by Kijima and Ojika12 and by Parsonsand Rolland13 for medical visualization. Shortly after,Tachi et al.14 and his group demonstrated a non-head-mounted configuration that utilized two video pro-jectors together with a retroreflective screen. Theyfurther extended this configuration to a head-mountedsystem.15

Custom-designed optics for HMPDs and a studyof retroreflective material properties were initiallyexplored by Poizat and Rolland.16 Later, Hua andRolland investigated the engineering challenges indeveloping a fully custom-designed HMPD systemand studied the imaging properties of retroreflec-tive materials and their effects on image quality.8,10

Fig. 1. (Color online) Schematic of a monocular head-mountedprojection display.

10 May 2007 � Vol. 46, No. 14 � APPLIED OPTICS 2601

They made further efforts to design wide FOV,low-distortion, and lightweight miniature opticsfor HMPD systems, using advanced optical designtechnology such as diffractive optical elements, as-pheric surfaces, and plastic materials.17 The opticsfor each eye weighs only 6 g with a 52° FOV and lessthan 2.5% distortion. These efforts have led to thesuccess of custom-designed compact prototypes.18

Several efforts were made thereafter to exploremore compact or wider FOV display designs. Forinstance, Ha et al. designed a 70° wide FOV projec-tion lens,19 and Rolland et al.20 developed an OLED-based 42° prototype and a teleportal HMPD system.Martins et al. explored a design that integrated aretroreflective screen within the helmet for an out-door environment,21 and Curatu et al. reported thedesign of a HMPD integrated with eye-trackingcapability.22

A wide range of single-user applications, such asobject-oriented and visual-haptic displays,14,15 distrib-uted display environments,23 medical visualization,13

and optical camouflage,24 have been explored anddemonstrated using the HMPD concept. The HMPDtechnology has also been explored in multiuser collab-orative environments. For instance, Rolland et al.20

and Davis et al.25 have demonstrated a deployableroom display environment for medical training. Byexploring several unique properties inherent inHMPD technology, Hua et al. have developed a mul-tiscale collaborative infrastructure, referred to asSCAPE, to support versatile modes of user interac-tion and collaborative operation in augmented virtualenvironments.26–29

3. Conceptual Design of a Polarized HMPD

A. Illumination Challenge

As briefly discussed in Section 1, existing HMPD de-signs confront the challenge of low image brightnessand contrast. Consider a pixel on the microdisplay.Let us denote the viewing angle subtended by thechief ray of the given pixel in the eye space as �, whichcharacterizes the FOV of an HMPD system. The lu-minous flux of the pixel collected by the exit pupil ofa HMPD system can be modeled by

�v�� rB-trans��rretro��rB-refl��eff��l��, (1)

where �l is the luminous flux collected by the projec-tion system from the given pixel on the microdisplay,�eff is the transmittance of the projection system, rB-refl

and rB-trans are the reflectance and transmittance ofthe beam splitter, respectively, and rretro is the retro-reflectance of the retroreflective screen. rB-refl, rB-trans,and rretro depend on the viewing angle or the incidenceangle upon the associated optical surfaces.

The transmittance of a well-designed projectionsystem can typically be approximately 80% or higher.However, a theoretical 50�50 beam splitter will leadto a 75% loss of the light owing to the dual pass

through the beam splitter (Fig. 1), without takinginto account the factors of absorption and reflectionloss. A further light loss, varying from 50% to 80%, isobserved from the currently available retroreflectivematerials.9 Therefore the accumulative light effi-ciency in existing HMPD designs is approximately4%–10%. The variation depends mainly on the retro-reflectance of the screen. The low efficiencies of thebeam splitter and the retroreflective screen accountfor the major luminance attenuation. Minimizing theloss from the beam splitter is critical for improvingthe efficiency of luminance transfer in a HMPD sys-tem.

B. Polarized HMPD

To address the image brightness challenge describedabove, we present a simple but elegant design of ap-HMPD.11 The main departure from a conventionalHMPD design is that the polarization states of thedisplay system are manipulated to maximize the ef-ficiency of reflection and transmission. A schematic ofa monocular p-HMPD configuration is shown in Fig.2. One of the modifications is to replace a nonpolar-izing beam splitter in Fig. 1 with a polarizing beamsplitter (PBS). To gain maximum reflection and min-imize the transmission loss of light incident upon thePBS from the projection optics, we further ensurethat the projected light is linearly polarized, and itspolarization direction is matched with the high-reflection polarization of the PBS, which is usuallyreferred to as the S polarization.

The microdisplays used in HMD designs are usu-ally liquid-crystal (LC) flat panels, and thus the lightemergent from these microdisplays is often linearlypolarized. In most of the LC-based microdisplays, thepolarization direction is usually parallel with thewidth or height side of the panel. Simply aligningthe polarization direction of the panel with theS-polarization axis of the PBS can ensure high reflec-

Fig. 2. (Color online) Schematic of a monocular polarized head-mounted projection display.


tion of light. In HMD systems, however, it is usuallydesirable to align the width of the panel with thehorizontal FOV for a preferred aspect ratio of thevisual field. Therefore when the polarization axis ofthe microdisplay is perpendicular to the S polariza-tion of the PBS, a half-wave plate retarder can beinserted in front of the PBS, with its fast axis orientedat a 45° angle with the S polarization, to rotate thepolarization axis of the projected light by 90°. In mi-crodisplays where their polarization direction is at anarbitrary angle with the S polarization of the PBS, apolarization rotator can be carefully designed to ro-tate the polarization axis of the projected light ac-cordingly. Except for a low-percentage absorptionloss through a retarder or polarization rotator, therequirement for a linearly polarized projected lightusually does not lead to significant loss of light usingLC-based microdisplays. Owing to the high reflec-tance of a PBS for the S-polarization light, the pro-jected light is reflected with very high efficiency (e.g.,approximately 93% for the wire-grid PBS we used inSubsection 4.C), opposed to approximately 50% ormore loss through a non-PBS interface.

After the projected light is reflected by the PBS, itis retroreflected back to the same PBS by a retrore-flective screen. The potential depolarization effects bythe retroreflective screen were recently tested withan Axiometric polarimeter. Results show that depo-larization is less than 10% for incidence angles within�20° and is less than 20% for angles up to �30°. Thetest further showed that the retroreflected light re-mains dominantly the same type of polarization as itsincidence light. For instance, when the incident lightis S polarized, the retroreflected light remains S po-larized, perpendicular to the high-transmission po-larization of the PBS. The depolarization artifactswill cause a decrease of luminance transfer effi-ciency varying with incidence angles. Such angulardependence of luminous efficiency visually createsvignettinglike artifacts and reduces image unifor-mity.

To minimize transmission loss, it is required tomatch the polarization axis of the retroreflected lightwith the P-polarization axis of the PBS. As shown inFig. 2, a quarter-wave retarder is inserted between thePBS and the retroreflective screen, with its fast axis ata 45° angle with the polarization direction of the lightemergent from the PBS. With this modification, theprojected light is manipulated through a consecutivesequence of polarization states, from its initial state ofS polarization to right circular polarization (RCP) bythe first pass of the quarter-wave retarder, from RCPto left circular polarization (LCP) at the interface of theretroreflective screen, and from LCP to P polarizationafter the second pass of the retarder.

Because of the high transmittance of P-polarizationlight by a PBS (e.g., approximately 87% for the wire-grid PBS in Subsection 4.C), the retroreflected light istransmitted efficiently and is collected at the exitpupil, opposed to approximately 50% of transmissionloss through a non-PBS interface. With the abovemodifications, Eq. (1) for a p-HMPD can be rewrit-

ten as

�v��wp2��rP-trans��rC-retro��rS-refl��eff��l��,

(2)

where �wp is the transmittance of the wave-plateretarder, rS-refl and rP-trans are the reflection and trans-mission efficiencies of the PBS for S and P polariza-tions, respectively, and rC-retro is the retroreflectanceof the retroreflective screen for circularly polarizedlight. The luminance transfer efficiency in a p-HMPDcan be up to four times the efficiency in nonpo-larizing HMPD designs using a theoretical 50�50beam-splitting method. Practically, we expect the ob-served image luminance to be approximately 15%–40% of the microdisplay luminance. In addition to FOVand wavelength dependence, the efficiency variationdepends mainly on the retroreflectance of screenchoices.

4. Design of a Compact Optical System

Based on the schematic design described in Section 3,a monocular prototype was designed and built on anoptical bench from off-the-shelf optics to validate theimprovement on imaging quality.11 Further effortswere made recently to design a compact, low-costoptical system and to develop a head-mounted proto-type.

A. System Specifications

A pair of existing 1.3 in. (1 in. � 2.54 cm) backlit colorAMLCDs, with a resolution of �640 � 3� � 480 pixelsand a 42 �m pixel size, was selected as the miniaturedisplays. The display, tested with a polarization an-alyzer, is linearly polarized, and its polarization axisis aligned with the width of the panel.

We target for a display with a diagonal full FOVbetween 50° and 60°, which corresponds to 35.3–28.6 mm of focal length for the projection lens. Thetargeted FOV stems from several considerations andtrade-offs. First, our previous investigation on retro-reflective materials shows that the retroreflectance ofcurrently available materials drops off significantlyfor light incident at angles beyond �35°.10 A FOVbeyond 70° will inevitably cause vignettinglike arti-facts and compromise image uniformity. Second, in-creasing the FOV will degrade the angular resolutionof the display. The given miniature display allowsonly a narrow (i.e., 13°) FOV to meet the requirementfor 1 arc min visual acuity at the fovea, while a 60°diagonal FOV offers a balanced angular resolution of4.5 arc min per pixel. Finally, a wide FOV requiresbeam splitters and retarders with large dimensions,which consequently compromises the compactnessand light weight of the display system.

Several high-quality projection optics have beendesigned for HMPDs, using the combination of dif-fractive optical elements, aspheric surfaces, andplastic materials.17,19 The fabrication costs of theselenses, however, are intimidating. Rather than de-signing such a costly system, we aim to design com-pact projection optics composed of 3–4 glass elements


with low-cost spherical surfaces. The system, to-gether with the polarizing components, should beoptimized to minimize all primary aberrations andparticularly achieve low distortion and chromatic ab-errations. The back focal length of the projection lensis required to be at least 25 mm owing to the pack-aging considerations of the helmet design discussedin Section 5.

The targeted diameter of the exit pupil in the pro-jection system is 10–12 mm, which leads to a projec-tion system with a f�# of 2.5–3.5. The large pupil sizeallows a swivel of approximately �21 up to 26.5°within the eye sockets without causing vignetting orloss of image with a typical 3 mm eye pupil in thelighting conditions provided by HMPDs. Further-more it tolerates approximately 7 to 9 mm differencesof the interpupilary distances (IPD) among differentusers without the need to mechanically adjust theIPD of the binocular optics.

In the design of a head-mounted prototype, an ef-fective eye clearance of 23 mm is necessary to accom-modate users wearing eyeglasses. Though it is lesschallenging to achieve large eye clearance in HMPDdesigns than in eyepiece-based HMDs, it can be prob-lematic when the FOV is large. In a HMPD design,the exit pupil of the system is the mirror image of theentrance pupil of the projection optics. On one hand,if the separation between the entrance pupil of theprojection optics and the beam splitter is not suffi-cient, the reflected light by the beam splitter at largefield angles may be blocked by the projection optics.On the other hand, the required dimensions of thebeam splitters and the retarders scale with the FOVand the eye clearance distance. To ensure compact-ness, the entrance pupil distance of the projectionoptics is optimized to be approximately 5–10 mm in-side the lens. The specification of the system is sum-marized in Table 1.

B. Design of Projection Optics

After initial trials of optimization starting fromseveral patent lenses, we selected U.S. patent lens2799207 as a starting point. It is a five-element f�2.0system with a full FOV of 70°, which offered us theflexibility to explore a range of FOV and f�# combi-nations. We initially scaled the lens to a range of focallengths between 25 and 35 mm, and roughly opti-mized the scaled lenses to meet the first-order re-quirements specified in Table 1. For each of theselenses, we then explored the main considerations ofpackaging and helmet design factors discussed inSection 5. This exercise allowed us to specify validconstraints for the design of a projection lens in thecontext of helmet design and to visualize how thesubsequent helmet design would be affected. Finally,we decided to aim for a f�3.2 projection system with56° FOV, which offers a good balance between displayquality and helmet compactness. The projection lenshas a 31.5 mm focal length and a 10 mm entrancepupil.

The initial five-element system was then optimizedwith CODE V by Optical Research Associates.30 Dur-

ing the initial steps of optimization, all the curva-tures of the refractive surfaces and the thicknessbetween the surfaces were set as variables. The sur-face materials were also set as variables; however,they were constrained to eliminate high-refraction-index materials. Five visual fields, 0, 0.28, 0.55, 0.76,and 1 (i.e., on axis, 8°, 16°, 22°, and 28°, respectively)were optimized. The weights for the five fields wereadjusted accordingly during the optimization process.A few iterations led to an optimized five-element sys-tem shown in Fig. 3(a) offering a circular FOV of 58°for a full unvignetted 10 mm pupil. The optimizedsystem demonstrates 30% of modulation contrast at aspatial frequency of 40 lps�mm. It offers a spot size ofless than 1

4 of the pixel size for the on-axis field andapproximately 1

2 of the pixel size for the maximumfield. Both the modulation transfer function (MTF)and spot size performances are significantly betterthan the minimal requirements listed in Table 1.Overall aberrations are well corrected and balanced.However, the system has approximately 13% distor-tion and a fairly large field curvature, which is notdesirable in display designs. The overall length of theoptics itself is approximately 42 mm, with a backfocal length of 41 mm. The diameter of the largestlens element is over 35 mm.

Table 1. Specification of the p-HMPD Optical System

ParameterDescription Specification

Microdisplay Type Active Matrix LCD,backlighting

Active display area 26.4 mm (width) � 19.8 mm(height), 33 mm (diagonal)

Pixel resolution (640 � 3) � 480 pixelsPixel size �42 �m, square

Projectionoptics

Effective focallength

28.6–35.3 mm

Entrance pupildiameter

10–12 mm

Entrance pupildistance

5–10 mm

Back focal length �24 mmTotal number of

optical elements3–4

PBS Size 64 mm � 45 mmS-polarization

reflection�90%

P-polarizationtransmission

�85%

Retarder Retardance Quarter-wave (broadband)Size 60 mm � 45 mm

System Wavelength VisibleFOV 50°–60° (diagonal)Distortion �5% over the entire FOVModulation transfer

function (MTF)�30% @ 12.5 lps�mm (given

by the pixel size of themicrodisplay)

Spot size (rms) �42 �mEye clearance 20–25 mm


Further optimization was performed to balance be-tween performance and compactness. The cementeddoublet in Fig. 3(a) was replaced by a biconcave neg-ative singlet with the same equivalent optical power.Several iterations of optimization led to a four-element starting point, from which a global optimi-zation process was performed. During the globaloptimization process, the major constraints were theeffective focal length, overall length, as well as gen-eral constraints on the minimum and maximumthickness of the elements.

Among the many potential solutions synthesizedthrough the global optimization process, the format

shown in Fig. 3(b) was selected for further optimiza-tion, mainly because of the combination of its elegantlens shapes, compactness, and relatively low value oferror function. Further optimization was performedprogressively by adjusting the weights to the fivevisual fields to achieve approximately the same MTFperformance across the FOV, in addition to the con-straints on the overall length of the lens assemblyand the back focal distance. We further constrainedthe glass map so that high-refraction-index materialscould be avoided. After reaching a well-balanced op-tical performance and compactness, the fictitious ma-terials were replaced with closely matched low-costglasses.

During the final stage of optimization, a constraintwas added to ensure the back focal length to be atleast 25 mm and limit the third-order distortion,which was less corrected than other third-order ab-errations. After a few iterations, the system distor-tion was reduced to 3.8% at the 28° field. The layoutof the final design, with a total weight of 6 g, is shownin Fig. 3(c). It is worth noting that the telecentricrequirement was relaxed in the design to gain com-pactness. In the display space, the chief ray angle ofthe 1.0 field is approximately 32°. Such a steep in-cident angle at the marginal field can potentiallyreduce image uniformity for LC-based displays, yield-ing vignettinglike artifacts and compromising theimage contrast of the peripheral field. However, en-forcing telecentric constraint requires that the lensaperture be at least the same size as the microdisplaysource, which significantly compromises the compact-ness of a HMD design. The AMLCD microdisplaysused in this design offer considerably large viewingangles. The luminance and contrast attenuationwithin �30° of viewing angles appears to be accept-able. In desktop displays 20% nonuniformity acrossthe entire visual field is not unusual.

The spot diagrams across the five fields are shownin Fig. 4(a). The rms spot diameter is smaller thanthe pixel size of the LCD display across the entirevisual field. The rayfan plots and field curves areshown in Figs. 4(b) and 4(c), respectively. The resid-ual astigmatism reaches a maximum of 0.25 mm atthe 21° FOV, and there is some residual coma at the28° field. The distortion of the system is well cor-rected and less than 3.8% across the overall FOV. Thepolychromatic MTF for the full 10 mm pupil acrossthe five representative field angles is shown in Fig4(d). The modulation contrast of the design acrossthe entire visual field is approximately 30% at aspatial frequency of 20 cycles�mm and over 50% at12.5 cycles�mm, which is the spatial frequency of thetargeted LCDs (in Table 1). Therefore the optics per-formance slightly exceeds the requirements by themicrodisplays.

The performance of the design in Fig. 3(c) canhardly be improved without making significantchanges to the surfaces. We experimented addingan aspheric surface to the first surface of the secondelement. It turned out to be very useful to correctthe residual aberrations. However, the design with-

Fig. 3. (Color online) Design layout of the projection optics forp-HMPD prototype. (a) Five-element starting point optimized froma U.S. patent lens. (b) Four-element starting point obtained bysimplification from (a) and global synthesis. (c) Final four-elementdesign optimized from (b).


out aspherics met the requirements, and we decidedto leave the option of an aspheric surface out.

C. Polarizing Components

One of the key requirements for the polarizing com-ponents (i.e., PBS and retarders) are their spectralresponses to visible wavelengths to minimize thewavelength dependence of luminous efficiency, whichresults in color temperature displacements of the dis-play system. It is also required that the PBS and theretarder have a wide acceptance angle to match theFOV of the projection optics. Finally, the form factorand weight of these components are greatly con-cerned for the consideration of display compactnessand portability.

Our initial experiments showed that the widely usedPBS cubes are inadequate for the proposed system.While PBS cubes usually have high reflection andtransmission ratios (e.g., as high as 99%) and highextinction ratios (e.g., 1000:1), their acceptance angleis usually very narrow owing to the limited range ofthe Brewster window.31 A cube PBS with a larger ac-

ceptance angle might be designed at the cost of usinghigh-refractive-index prisms and poor reflection–transmission uniformity across the angular aperture.Furthermore, given the wide FOV of the projectionoptics, the required aperture for the PBS is larger than70 mm. The form factor and the unneglectable weightof two cemented prisms at such a large dimensioninevitably challenge the design of a light and compacthelmet.

Alternatively, a 65 mm � 45 mm nanometer-scalewire-grid plate PBS was custom designed, with aplate thickness of 1.6 mm. The wire direction of thePBS is aligned with the width of the plate to matchwith the polarization axis of the microdisplay formaximum efficiency. The wire-grid PBS offers manyadvantages such as wide acceptance angles, low lightabsorption, light weight and plate form, broadband,and potential low cost owing to integrated-circuit-type fabrication processes.32 The wire-grid coatingcan be thought to function as a dielectric interface forthe P polarization but as a metal surface for the Spolarization. Consequently, a wire-grid PBS usually

Fig. 4. (Color online) Performance analysis of the design in Fig. 3(c) for a 10 mm unvignetted pupil. (a) Spot diagram across five fieldangles; (b) rayfan plots; (c) plots for spherical aberration, astigmatism, and distortion; (d) polychromatic MTF as a function of the spatialfrequency in line pairs per millimeter (cyles�mm).


has a high contrast ratio for S polarization and alower contrast for P polarization.

After comparing the various retarder techniques, a60 mm � 45 mm glass–mica–glass cementedquarter-wave retarder was custom made. The fast axis of theretarder is at a 45° angle with the width of the com-ponent, which matches with the polarization axis ofthe microdisplay. When it is used together with thePBS, the retarder manipulates an incident linear po-larization into a circular polarization and vice versa.The experimental testing of these polarization ele-ments will be discussed in Section 6.

5. Design of a Compact Prototype

Light weight and compactness are always highly de-sirable features for head-mounted devices. Besidesthe design of a compact optical system, one of themain challenges in developing an HMD prototype isto design an ergonomically conceived and lightweighthelmet, which accounts for a wide range of con-straints imposed by the optical design, the electronicsof the microdisplays, and various human factors.

Experiments with a first-generation prototype pro-vided us with extremely valuable experiences and

knowledge about the important human factors inparticular.18 For example, the weight of the first-generation helmet is not well balanced. The heavyfront adds significant stress to a user because it tendsto pull the helmet forward and forces the user toadjust the helmet frequently. Furthermore the en-closed top with microdisplay electronics inside causesdifficulty with heat dissipation. Finally, mountingthe optics in a vertical configuration leads to the over-lay of ghost views directly reflected from the groundby the beam splitters. In the first-generation proto-type, ground reflection is less of a concern mainlybecause the lighting in an environment has to beextremely dim owing to the low brightness of thevirtual images. Owing to the enhancement of imagebrightness in our proposed p-HMPD design, the dis-play is anticipated to function in normal room light-ing conditions.

Accounting for various ergonomic factors and de-sign constraints, we chose to mount the optics in ahorizontal configuration. To avoid a front-heavy de-sign, the optical path of the projection optics wasfolded 90° by inserting a mirror between the projec-tion optics and the microdisplay [Fig. 5(a)]. This

Fig. 5. (Color online) Design of a compact p-HMHD prototype. (a) Overall optical layout in prototype packaging, (b) complete CADassembly of the helmet, (c) front and (d) side views of the p-HMPD prototype.


folded design allows the installation of the microdis-play, the associated electronics, and the cables to thesides of the helmet. The folded design also contrib-utes to the reduction of the horizontal width of thehelmet and satisfies both ergonomic and aestheticconsiderations. As a result, the overall width of thehelmet is in proportion to the average size of an adulthead,33 and the overall weight is nicely balancedaround the head.

The total weight of the first-generation prototype isapproximately 750 g.18 A significant portion of theweight is attributed to complex metallic optome-chanical structures within the helmet, which housesthe optics and provides adequate adjustments of dis-play focusing, alignment, and IPD. To minimize theweight of the new prototype, we decided to eliminatethe use of metallic optomechanical structures in thenew design. Considering the free-form fabrication ca-pability of rapid prototyping (RP) techniques, alsoknown as layered manufacturing, we shaped the hel-met shell in such a way that the necessary structuressupporting the optics and electronics were integratedwith the shell as one single piece. The main shellprovides various shaped structures to mount themirror, microdisplay, electronics, projection optics,PBS, and wave-plate retarder. Considering the factthat these structures were designed as an integratedpiece, the positioning of these optical elements wasensured by the accuracy of fabrication, and fine po-sitioning was warranted by spacing the adjustmentduring the helmet assembly.

To allow the adjustment of the IPD, the housingsfor the left and right arms of the optics were designedas symmetric but separate parts. The parts were con-nected together by two rods. By pulling or pushingthe two parts, the IPD can be adjusted appropriatelywhen necessary. This simple method eliminated thecomplex mechanism used in the previous design. Acomplete computer-aided design (CAD) assembly ofthe helmet shell is shown in Fig. 5(b). The left andright parts are identical, each of which consists of themain shell with the supporting structures and a coverpiece. It is worth noting that the sides of the mainshells were shaped to hold and guide the thick andlong video cables that run from the microdisplays toa computer system.

The helmet shells were fabricated using the RPtechnique, in which physical models are fabricatedlayer by layer directly from a 3D CAD model. Thehelmet shells were assembled and attached to anoff-the-shelf headband that offers head-size adjust-ment. The front and side views of the prototype areshown in Figs. 5(c) and 5(d), respectively. The totalweight is approximately 450 g, significantly reducedfrom the previous prototype.

6. Experimental Results

The custom-made wire-grid polarizing beam splittersand the retarders were tested for FOV dependencewith an Axiometric polarimeter at a 550 nm wave-length. The reflectance for S-polarized light is ap-proximately 93% for 0° field angle, varying between

88% and 96% for field angles in the range of �30°.The transmittance for P-polarized light is approxi-mately 87% for 0° field angle, varying between 82%and 60% for field angles in the range of �30°. The 0°field corresponds to a 45° incidence angle on the PBS,and a positive field angle indicates that a ray im-pinges on the PBS at an angle less than 45°. At the 0°field, the ratio of the reflectance for S-polarized lightto P-polarized light is approximately 95, and theratio of the transmittance for P-polarized light toS-polarized light is approximately 480.

The transmittance of the retarder is approximatelyconstant across the entire FOV (less than 1.5% ofvariation). The retardance magnitude remains ap-proximately constant up to approximately �16° FOVand increases gradually by 40 nm (approximately 7%of the testing wavelength) at �30° FOV. The FOVdependence of the retardance magnitude will cause areduction of the overall efficiency at a marginal visualfield of the display, creating vignettinglike artifacts.

To validate the predicted improvement on imagebrightness, we initially implemented two monocularprototypes on an optical bench for the convenienceof testing flux efficiency: one prototype based on thepolarizing design and the other without polarizationmanipulation using a regular 50�50 nonpolarizingbeam splitter.11 The projection lenses used in thebench prototypes, with an effective focal length of30 mm, were assembled from two off-the-shelf achro-mats.

Using a calibrated, collimated light source, we quan-tified the luminance efficiency of the two bench setupsover �30° of the FOV. The experimental results dem-onstrate approximately four times of consistent im-provement on luminance efficiency, across the entireFOV. Combining the efficiencies of the projection op-tics, the PBS, the retarder, and the retroreflectivefilm, the overall efficiency of the polarized setup isapproximately 17% at the center field and slowlydrops down to approximately 12% at �30° fields. Onthe contrary, the overall efficiency of the nonpolar-ized setup is approximately 4% at the center anddrops down to approximately 2% at �30°. Furthermeasurements will be made for other wavelengths. Acomprehensive analysis of the above polarimetricproperties and luminance efficiency quantification aswell as their impacts on image quality will be re-ported in a follow-up article.

To compare the image brightness and contrast ofthe two setups with and without polarization control,the collimated light source used in the previous test-ing was replaced with microdisplays, and an identicalimage was projected through the setups. Under iden-tical room lighting conditions and camera exposuresettings, a set of photographs was taken from the twobench prototypes by aligning the camera with the exitpupil of the optics. Two examples of the photographsunder identical room lighting from the polarized andthe nonpolarizing displays are shown in Figs. 6(a)and 6(b), respectively. It is worth noting that neitherpostprocessing nor brightness enhancement was per-formed on these photographs. The photographs from


the polarized setup demonstrated a considerable in-crease in intensity and significant improvement inimage contrast and color vividness over those fromthe nonpolarizing setup.

From the photograph sets, we analyzed the histo-grams of the region representing the display view,which is marked with dotted circular lines in thefigure. The histograms for the two examples in Figs.6(a) and 6(b) are shown in Figs. 6(c) and 6(d), respec-tively. The mean intensity values increase by approx-imately 50%, from 78 for the nonpolarizing HMPDdesign to 111 for the p-HMPD, and the standarddeviations increase by approximately 52%, from 21 to33. Such wider intensity distribution for the p-HMPDindicates an improvement in image contrast and dy-namic range.

7. Conclusion and Future Work

In this paper we presented the design of a p-HMPDwith considerably improved luminance efficiencycompared with existing HMPD designs. The lumi-nance efficiency of the proposed design is approxi-mately three times brighter than a nonpolarizingdesign. The paper also detailed the optical systemand helmet design for a compact head-mounted pro-

totype. In future work, experiments will be per-formed to characterize the depolarization effects ofthe retroreflective material, quantify the chromaticdependence of the polarization elements, and quan-titatively evaluate the luminance efficiency and thechromatic responses of the display. We will alsocustom design a high-resolution, brighter displayprototype using LCOS-based microdisplays for moredemanding applications.

This research was supported by National ScienceFoundation grant IIS 05-34777.

References1. R. Azuma, Y. Baillot, R. Behringer, S. Feiner, S. Julier, and

B. MacIntyre, “Recent advances in augmented reality,” IEEEComput. Graphics Appl. 21, 34–47 (2001).

2. M. Bajura, H. Fuchs, and R. Ohbuchi, “Merging virtual objectswith the real world: seeing ultrasound imagery within thepatient,” in Proceedings of ACM SIGGRAPH (ACM, Chicago,1992), pp. 203–210.

3. T. Caudell and D. Mizell, “Augmented reality: an application ofheads-up display technology to manual manufacturing pro-cesses,” in Proceedings of Hawaii International Conferences onSystems Sciences (Hawaii, 1992), pp. 659–669.

4. J. P. Rolland and H. Fuchs, “Optical versus video see-through

Fig. 6. (Color online) Sample images from bench prototypes. (a) Polarized HMPD, well-lit environment, (b) nonpolarized HMPD, well-litenvironment, (c) histogram of the display view region for photograph (a), (d) histogram of the display view region for photograph (b).


head-mounted displays in medical visualization,” Presence:Teleoperators and Virtual Environments 9, 287–309 (2000).

5. F. Biocca and J. P. Rolland, “Virtual eyes can rearrange yourbody: adapting to visual displacement in see-through, head-mounted displays,” Presence: Teleoperators and Virtual Envi-ronments 7, 262–277 (1998).

6. R. Fisher, “Head-mounted projection display system featuringbeam splitter and method of making same,” U.S. patent5,572,229 (5 November 1996).

7. J. Fergason, “Optical system for head mounted display usingretro-reflector and method of displaying an image,” U.S. patent5,621,572 (15 April 1997).

8. H. Hua, A. Girardot, C. Gao, and J. P. Rolland “Engineering ofhead-mounted projective displays,” Appl. Opt. 39, 3814–3824(2000).

9. R. N. Berry, L. A. Riggs, and C. P. Duncan, “The relation ofvernier and depth discriminations to field brightness,” J. Exp.Psychol. 40, 349–354 (1950).

10. H. Hua, C. Gao, and J. P. Rolland, “Imaging properties ofretroreflective materials used in head-mounted projective dis-plays (HMPDs),” Proc. SPIE 4711, 194–201 (2002).

11. H. Hua and C. Gao, “A polarized head-mounted projectivedisplay,” in Proceedings of IEEE and ACM International Sym-posium on Mixed and Augmented Reality 2005 (IEEE, 2005),pp. 32–35.

12. R. Kijima and T. Ojika, “Transition between virtual environ-ment and workstation environment with projective head-mounted display,” in Proceedings of IEEE VR 1997 (IEEE,1997), pp. 130–137.

13. J. Parsons and J. P. Rolland, “A non-intrusive display tech-nique for providing real-time data within a surgeons criticalarea of interest,” in Proceedings of Medicine Meets VirtualReality 1998 (1998), pp. 246–251.

14. N. Kawakami, M. Inami, D. Sekiguchi, Y. Yangagida, T.Maeda, and S. Tachi, “Object-oriented displays: a new type ofdisplay systems—from immersive display to object-orienteddisplays,” in Proceedings of IEEE SMC 1999, IEEE Interna-tional Conference on Systems, Man, and Cybernetics, Vol. 5, pp.1066–1069.

15. M. Inami, N. Kawakami, D. Sekiguchi, Y. Yanagida, T. Maeda,and S. Tachi, “Visuo-haptic display using head-mounted pro-jector,” in Proceedings of IEEE Virtual Reality 2000, pp. 233–240.

16. D. Poizat and J. P. Rolland, Use of Retro-reflective Sheets inOptical System Design, Technical report TR98-006 (Universityof Central Florida, 1998).

17. H. Hua, Y. Ha, and J. P. Rolland, “Design of an ultra-light andcompact projection lens,” Appl. Opt. 42, 1–12 (2003).

18. H. Hua, C. Gao, F. Biocca, and J. P. Rolland, “An ultra-light

and compact design and implementation of head-mountedprojective displays,” in Proceedings of IEEE VR 2001, pp.175–182.

19. Y. Ha, H. Hua, R. Martins, and J. P. Rolland, “Design of awearable wide-angle projection color display,” in Proceedingsof International Optical Design Conference 2002 (IODC, 2002),pp. 67–73.

20. J. P. Rolland, F. Biocca, F. Hamza-Lup, Y. Ha, and R. Martins,“Development of head-mounted projection displays for distrib-uted, collaborative, augmented reality applications,” Presence:Teleoperators and Virtual Environments 14, 528–549 (2005).

21. R. Martins, V. Shaoulov, Y. Ha, and J. P. Rolland, “Projection-based head-mounted displays for wearable computers,” Proc.SPIE 5442, 104–110 (2004).

22. C. Curatu, H. Hua, and J. P. Rolland, “Projection-based head-mounted display with eye-tracking capabilities,” Proc. SPIE5875, 587050J (2005).

23. R. Kijima, K. Haza, Y. Tada, and T. Ojika, “Distributed displayapproach using PHMD with infrared camera,” in Proceedingsof IEEE 2002 Virtual Reality Annual International Sympo-sium (IEEE, 2002), pp. 33–40.

24. M. Inami, N. Kawakami, and S. Tachi, “Optical camouflageusing retro-reflective projection technology,” in Proceedings ofISMAR 2003 (ISMAR, 2003), pp. 348–349.

25. L. Davis, J. P. Rolland, F. Hamza-Lup, Y. Ha, J. Norfleet, andC. Imielinska, “Enabling a continuum of virtual environmentexperiences,” IEEE Comput. Graphics Appl. 23, 10–12 (2003).

26. H. Hua, L. Brown, and C. Gao, “A new collaborative infrastruc-ture: SCAPE,” in Proceedings of IEEE VR 2003 (IEEE, 2003),pp. 171–179.

27. H. Hua, L. Brown, and C. Gao, “SCAPE: supporting stereo-scopic collaboration in augmented and projective environ-ments,” IEEE Comput. Graphics Appl. 24, 66–75 (2004).

28. H. Hua, L. Brown, and C. Gao, “System and interface frame-work for SCAPE as a collaborative infrastructure,” Presence:Teleoperators and Virtual Environments 13, 234–250 (2004).

29. L. Brown and H. Hua, “Magic lenses for augmented virtualenvironments,” IEEE Comput. Graphics Appl. 26, 64–73(2006).

30. Optical Research Associates, http://www.opticalres.com.31. J. L. Pezzaniti and R. A. Chipman, “Angular dependence of

polarizing beam-splitter cubes,” Appl. Opt. 33, 1916–1929(1994).

32. D. P. Hansen, R. T. Perkins, and E. Gardner, “Broadband wiregrid polarizing beamsplitter for use in the visible wavelengthregion,” U.S. patent 6,243,199 (5 June 2001).

33. L. G. Farkas, Anthropometry of the Head and Face, 2nd ed.(Raven Press, 1994).


design of a bright polarized head-mounted projection display · 2015. 12. 31. · design of a...

Documents