CS 428: Week 4

sources include: James Helman's Siggraph '93 Applied VR course notes, Dennis Proffitts '94 Developing Advanced VR Applications course notes, Lou Harrison's Siggraph '97 Stereo Computer Graphics for Virtual Reality notes, and the IRIS Performer guide.

The eye has a bandwidth of approx 1 gigabyte per second - much greater than the other senses. Any AR or VR systems we create should try to match or exceed what the human body can perceive.

The real world doesn't flicker. Classic theatrical films flickered and CRT displays flickered because the image is constantly being refreshed. Most modern displays also need to be refreshed regularly to keep energy flowing to the various elements. If the image isn't refreshed fast enough, we perceive flickering. Some people can perceive flickering even at 60Hz (the image being refreshed 60 times per second) for a bright display with a large field of view but most people stop perceiving the flicker between 15Hz (for dark images) and 50Hz (for bright images). LCD panels tend to have 60hz or 120hz refresh rates these days (though manufacturers often claim to have higher rates. Direct view LED tiles have similar rates.

Note that this is the rate that the hardware refreshes, which is (mostly) independent from the rate the graphics are being refreshed. We are at the point now that HMDs, like TVs, and phones generally have sufficient hardware refresh rate, though more exotic technologies for AR are still catching up. Making sure the graphics refresh fast enough will always remain an issue.

The eye can detect an amazing range of 10^14 (10^-6 to 10⁸ cd/m²) in luminance, but we can only see a range of 100:1 at the same time. The size of our pupil automatically adjusts to the amount of light available, allowing us to shift our window of much narrower range of luminance detection across a much larger range. If we spend 30 minutes dark adapting we can dramatically increase our vision at the low end.

The human eye is sensitive to ratios of intensities not absolute magnitude. Brightness = Luminance^0.33. To make something appear n times brighter the luminance must be increased by n^3.

Most perceptual processes are driven by intensity not colour. The human motion system is colour blind, depth perception is colour blind, object recognition is colour blind.

Today's displays still can not reproduce the full colour spectrum that human beings can see, but the range increases every year, and with the new focus on OLED displays, high contrast displays, and direct view LED displays the coverage is getting much better.

Each eye has approximately 145 degrees horizontal (45 degrees towards the nose and 100 degrees to the side) and 100 degrees vertically (30 degrees up and 70 degrees down)

Current VR headsets like the VIVE and Quest 3 have 110 degree FoV, which are getting close to matching human field of view. Higher end displas like the Valve Index are pushing 120 degrees.

AR headsets currently have much smaller fields of view which limits their ability to combine real and augmented content as you can only see the augmented content within a window within your field of view.

Google Glass display had FoV of 15 degrees monoscopic in the corner of your vision
Microsoft HoloLens has a 30 degree horizontal and 18 degree vertical FoV with HoloLens 2 improving to 43 degree horizontal and 29 degree vertical.
Magic Leap One has a 40 degree horizontal and 30 degree vertical FoV

image from https://venturebeat.com/2018/07/31/magic-leap-ones-field-of-view-leak-signals-another-ar-disappointment/

When you are holding your phone or a tablet the FoV depends on how close you are holding the screen to your face - the closer to your face the more the screen fills your vision.

With human scale systems like the CAVE and CAVE2, or even a planetarium dome, the field of view depends on the size and location of the physical displays and the type of 3D glasses used - 6 sided CAVEs surrounded the users with screens, but the glasses typically don't cover the complete field of view (roughly 120 horizontal and 90 vertical).

We often encounter visual acuity as measured with a Snellan Chart at the doctor's office where correct vision is described as being 20/20 (20/X where this viewer sees at 20 feet detail that the average person can see at x feet, 20/200 is legally blind). If you are metricly minded then the fraction is 6/6. https://en.wikipedia.org/wiki/Snellen_chart.

Back in the 90s our monitors at normal distance were 20/40, HMDs were 20/425, BOOMs were 20/85, and the original CAVE was 20/110.

Our current consumer HMDs are around 20/60 with a wide variety of lens distortion effects to give us a wider field of view, while most of our laptops and phones have so many pixels that they give us better than 20/20 across a small range of our vision.

What happens when we walk from a bright area into a dark area, say into a movie theatre? When we are outside the rods are saturated from the brightness. The cones which operate better at high illumination levels provide all the stimulus. When we walk into the darkened theatre the cones don't have enough illumination to do much good, and the rods take time to de-saturate before they can be useful in the new lower illumination environment so we cant see very well. It takes about 20 minutes for the rods to become very sensitive, so dark adjust for about 20 minutes before going stargazing.

Since the cones do not operate well at low light intensities we can not see colour in dim light as only the rods are capable of giving us information. The rods are also more sensitive to the blue end of the spectrum so it is especially hard to see red in the dark (it appears black).

In VR where the programmer has total control of the visuals this means taking into account how long it will take a person to be able to see something in a dark area after being in a bright one, and how bright a scene can be after being in a dark one without blinding the user.

In AR this means we need to be careful how bright the computer graphics are, and we need to take into account the current conditions (is it a bright sunny day, is it overcast, is it dark) so the generated graphics are visible and comfortable, not hidden and not blinding.

There is also the optic nerve which is 10-20 degrees away from the fovea which connects your eye to your brain. This is the blind spot where there are no cones and no rods. We can not see anything at this point and our brain compensates by filling in that part of our vision with surrounding colours and simple patterns.

Chromatic Vision Simulator is a nice phone based tool for simulating what colour blind people see. Its free and available at http://asada.tukusi.ne.jp/cvsimulator/e/

Just as in the real world where we make sure that traffic lights work for colour blind people, we need to make sure that our VR and AR worlds work as well. You should make sure that the worlds you create work for color blind people.

In general it is a good idea to give the user control over colors, text size, contrast and brightness etc. in VR and AR just like it is a good idea to do this on a smart phone or a laptop to enable more people to use your hardware and software.

One of the major takeaways from this is that as our display technology gets better and better we will increasingly be running into the limitations of the human eye rather than the limitations of the display technology - just as we are on our smartphone and laptop screens.

Another major takeaway is that as we are creating worlds, in the case of VR, where we can take total control of various senses , this allows us to get a better understanding of how those senses operate, alone and together, and the more we understand, the less likely we are to make people sick, and the more likely we are to give them richer experiences that take advantage of all the relevant senses.

One of the most important things in VR, and increasingly in AR, is creating an artificial world with a sense of depth.

Interactive computer graphics, and monoscopic AR (hand held smartphone or tablet) adds 5

VR gives and stereoscopic AR adds on 5,6,7

some specialized multi-plane displays and light field displays can give 8.

In VR, as in stereo movies or stereo photographs, the brain is getting two different cues about the virtual world. Some of these cues indicate this world is 3D (convergence and stereopsis). Some of these cues indicate that the world is flat (accommodation). The eyes are converging depending on the position of the virtual objects floating in front of or behind the screen, but the eyes are focusing on the flat screen. This can lead to headaches, or more serious forms of simulator sickness, but in general its not a serious problem, especially if you are doing it for only a couple hours at a time.

For most people VR 'works' and virtual objects appear to be the correct size because they 'trust' convergence more than they trust accomodation (focus on the screen), but some people trust accomodation more, and for them the virtual objects are always on the projection plane and generally the wrong size.

If we start thinking about wearing AR glasses (which could potentially become VR glasses whenever we wanted) all the time then we may run into more issues.

Note there are some potential issues of having young children seeing this kind of mediated 3D and research is ongoing there in terms of whether it can alter the normal development of their vision. Most of the current headsets are for 12 or 13 year olds or older to be safe.

We have stereo vision because each eye sees a slightly different image. Well, almost all of us (90-95%) do.

Presenting stereo imagery through technology dates back to the mid 1800ds. Here is a stereoscope from the early 1900ds.

As a kid you may have had a View-Master(tm) and saw the cool 3D pictures. Those worked exactly the same way as the stereoscope. Your left eye was shown one image on the disc, while your right eye was shown a different image. These images were typically taken with a pair of identical cameras with identical settings sitting horizontally next to each other, or two images taken by the same camera that is quickly moved horizontally. Often cameras, especially film cameras, were too large requiring a series of mirrors to get their lenses close enough to mirror human vision.

As a kid you might have read some 3D comics which came with red/ blue (anaglyphic) glasses, or you might have even seen a movie with the red/blue glasses (maybe down at U of Chicago's Documentary Film Group with films like 'House of Wax' or 'Creature from the Black Lagoon' or 'It Came from Outer Space'). You can either use 2 synchronized projectors each with a black and white print of the film for one eye, one projector with a red filter and one with a blue filter, or, less expensively, you can have one projector with the red and blue on a single color print of the film , which is why this would become popular at second run theatres.

If you have a pair of red/cyan glasses and a correctly calibrated display then this image will become 3D.) The coloured lenses make one image more visible to one of your eyes and less visible to your other eye, even though both eyes are viewing both images.

What people saw in the theatre back in the 1950s, including that photo above from Life magazine, was closer to the image below, using 2 synchronized projectors each with a black and white print of the film for one eye, each projector with a different polarizing filter, and each person in the audience wearing polarized glasses. This is basically the same technology as when 3D movies came back in the 1980s and when they came back again in the early 2000s. Now we have digital projectors so the image is much more stable and we don't have film breaks so the final stereo effect is much cleaner but the core technology is identical.

Today we can also put polarizing material on the front of our flat panel displays in alternating horizontal lines of the left and right eye images, so a single video (in color or black and white) can contain imagery for both eyes, though in half resolution each. This is how evl's CAVE2 creates stereo.

here is another red/blue image from an Ocean Going Core Drilling ship. The red/blue or red/cyan trick works ok for greyscale images (like old movies or black and white comics) but has a hard time with color. Other color variants have been used over the years.

Another inexpensive way to show stereo imagery is to draw two slightly different images onto the screen (or onto a piece of paper), place them next to each other and tell the person to fuse the stereo pair into a single image without any additional hardware. This is easy for some people, very hard for other people, and impossible for a few people.

Some of these images require your left eye to look at the left image, others require your left eye to look at the right image (cross-eye stereo).

To see the pictures below as a single stereo image look at the left image with your right eye and the right image with your left eye. If you aren't used to doing this then try this: Hold a finger up in front of your eyes between the two images on the screen and look at the finger. You should notice that the two images on the screen are now 4. As you move the finger towards or away from your head the two innermost images will move towards or away from each other. When you get them to merge together (that is you only see 3 images on the screen) then all you have to do is re-focus your eyes on the screen rather than on your finger. Remove your finger, and you should be seeing a 3D image in the middle.

As mentioned above, in the 1950s when stereo films like 'Creature from the Black Lagoon' were shown theatrically, the theatres did not use red/cyan. They used a system pretty much identical to what 3D movies used in the 80s and in the 2010s - passive polarization. In this case one projector shows the left eye image with a polarizing filter. A second synced projector shows the right eye image with a different polarizing filter. The screen preserves polarization and the viewers wear lightweight polarized glasses that allow the correct image through and not the wrong one.

This same technique can be used on flat panel displays where a layer is added on top of the display with lines of alternating polarization matching up with the rows of pixels below. The same lightweight glasses are used but now one eye only sees the even lines of the display while the other eye only sees the odd lines. This has the disadvantage of cutting your vertical resolution in half.

In active stereo the projector or the panel shows the image for the left eye, then black, then the image for the right eye, then black, and then repeats. The viewer wears slightly heavier glasses with LCD shutters the turn either clear or opaque and the shutter for the appropriate eye is open when the display shows the image for that eye. This requires the display to also talk to the glasses (usually via IR) and for the glasses to be powered through a cable or batteries.

To make sure your visuals look good, its very important to keep in mind what kind of display you will be using, the space that it will be used in, who will be using it, and what kinds of things they will be doing - are these designers interacting with a new car model, or museum visitors exploring an ancient site?

Projection-based systems, even modern laser based 4K ones, can give much larger borderless fields of view, but typically need to be in dimly lit or dark rooms. If the system is front-projection then how close will the user be able to get to the display before they begin to cast shadows on the screen? If its rear-projection from behind the screen then how much space are you prepared to waste behind the screen. Modern short throw and ultra-short throw projectors allows the users to get much closer to the screens than earlier projectors.

Large flat panel displays have a different problem - they are designed for on-axis TV viewing and the further you are off-axis (horizontal or vertical) increases the chance that you will see degraded colour / contrast / stereo vision. They also tend to have borders which can be distracting when they are tiled. They are also thin so you cant stand on them without breaking them so its hard to make floors out of them.

Head mounted VR displays avoid issues of dark rooms and off-axis viewing but you still need enough space for the user to move around in, and you need to make sure the user doesn't hurt themselves while moving around without seeing the real world.

Head mounted AR displays allow the user to see the real world, making them safer than VR displays, but as these displays may be used in a dim room or in bright sunlight the graphics need to be visible across that wide range of use cases, similar to how you want to be able to use your phone or smart watch in a variety of settings, so the contrast and brightness ranges need to be greater than for displays that are only used indoors, and the brightness and contrast of the display need to match the real world behind the graphics.

Interocular Distance - the distance between the viewer's left and right eyes, usually about 2.5 inches - in order for virtual objects to appear to have the exact correct size the AR or VR system needs to know your specific interocular distance. https://en.wikipedia.org/wiki/Interpupillary_distance

note that there can be good reasons to change the interocular distance if you are looking at very small things or very big things, i.e. back in 2006 we helped out with the NASA STEREO project using two satellites ahead and behind the earth in its orbit to take stereo movies of the sun, and on the other end of scale we work with biologists that take stereo images of cells.

Stereo Window (Plane) - the point at which there is no difference in parallax between the two eye views - usually at the same depth as the monitor screen or the projection surface. HMD optics create their own projection plane, commonly about 1-2 meters away from the user. Looking at things on the plane is most comfortable since convergence and accommodation are both correct, but there is no stereo effect there. Often the subject of a stereo photograph or movie is placed on the plane with other elements in front or behind to heighten the 3D effect.

Homologous Points - points which correspond to each other in the separate eye views (e.g. the tip of a person's finger, or a person's left pupil)

Horizontal Parallax - when the retinal images of an object fall on disparate points on the two retinas, these points differ only in their horizontal position (since our eyes are at the same vertical position). The value given by R - L. Note that this is obviously true for stereo movies, but in full VR you may be tilting your head (and your two eyes along with your head) so the 'horizontal' part refers to the line between your eyes defining horizontal and vertical.

Positive Parallax - the point lies behind the stereo window - in the Creature from the Black Lagoon image, the background is far from the camera and has blue to the right of red for homologous points

Zero Parallax - the point is at the same depth as the stereo window - in the Creature from the Black Lagoon image, the base of the fossilized hand is at zero parallax and there is no red / blue disparity.

Negative Parallax - the point lies in front of the stereo window - in the Creature from the Black Lagoon image, the fingers of the fossilized hand are near to the camera and have red to the right of blue for homologous points like the fingertips

Vertical Displacement - vertical parallax between homologous points relative to the line that the two eyes form - this should be zero in a correctly calibrated setup. Our eyes and brain can actual handle a certain amount of vertical parallax but it gets uncomfortable.

Interocular Crosstalk (Ghosting) - when one eye can see part of the other eye's view as well, reducing the effect of the stereo imagery - Crosstalk should also be zero in a correctly calibrated setup, but this can be hard to achieve, especially in scenes with very high contrast.

In an HMD or the boom, the screens in front of each user's eye move with the user - that is the screens are always perpendicular to the eye's line of site (assuming the eyes are looking straight ahead). This allows the traditional computer graphics 'camera paradigm' to be extended so that there are 2 cameras in use - one of each eye.

In the CAVE, Fish Tank VR, or any large screen VR, this is not the case. The projection planes stay at a fixed position as the user moves around. The user may not be looking perpendicular to the screen and certainly cant be looking perpendicular to all of the screens in the CAVE simultaneously - in this case 'off-axis projection' is used, and the math/geometry is a bit more complex to be more general, though still something that someone with a knowledge of computer graphics from CS 425 could implement.

With head mounted displays for AR or VR there is a single user looking at the display. With room scale or spatially immersive displays like CAVEs there are typically multiple users, but usually only one user being head tracked. This means everyone is seeing through this one person's eyes, meaning the farther you are standing form this person the less correct the visuals are. The visuals are also moving based on that tracked person's movements, so if that person is moving around a lot the visuals are moving around a lot.

Naively we would like to update the graphics every frame in order to use the most recent head (eye) positions.

Since there will be jitter in the tracker values and latencies in getting information from the tracking hardware to deal with, this may result in the computer generated image jittering.

One way to avoid this is to only update the image when the head has moved (or rotated) a certain amount so the software knows that the motion was intentional.

Another common option is to interpolate between the previous and current position and rotation values in order to smooth out this motion. This results in smoother transitions, and minimizes tracker error, but will also increase the lag slightly.

Another option is to extrapolate into the future by predicting how the user is going to move in the next couple seconds and proactively render for the position you believe the user will be in since people typically do not move their heads too abruptly.

In the 1990s Silicon Graphics (the folks who created GL, OpenGL, and would go on to found nVidia) optimized graphics through a pipelined approach (image from SGI Performer manual) and overlapping different parts of the pipeline.

and then it helps to have tools to help visualize how long each part of the process takes and where improvements can be made.

The following image show a-application, c-cull, d-draw in the same color for the same frame, e.g. when its drawing frame 5 its culling frame 4 and doing the application phase of frame 3. In this case the application is spending almost all its time drawing, so it has time to do more application work or culling which might improve the overall experience.

While in videogames it is good to maintain a particular frame rate, in VR and AR it is much more important to do that.

Current game engines are pretty good at optimizing what is drawn based on where the user is in the scene. If you create your own software from scratch you will need to take care of those things.

The horizon can be moved in - moving in Z-far and perhaps covering this with fog

2 things are needed: a functioning vestibular system (canals in the inner ear) and a sense of motion

Symptoms: Nausea, eyestrain, blurred vision, difficulty concentrating, headache, drowsiness, fatigue

Causes: still unknown but one common hypothesis is a mismatch between visual motion (what your eyes tell you) and the vestibular system (what your ears tell you)

Why would this cause us to become sick? Possibly an inherited trait - a mismatch between the eyes and ears might be caused by ingesting a poisonous substance so vomiting would be helpful in that case.

Another hypothesis deals with the lack of a rest frame. When a user views images on a screen with an obvious border that border locates the user in the real world. Without that border the user loses their link to the real world and the affects of motion in the virtual world are more pronounced.

Current HMD environments default to allowing the user to walk around a physical space but then use teleporting rather than controller based 'flying' common in first person shooters to move larger distances. This is primarily to reduce the chances of people getting sick. Current HMD environments also often stop drawing graphics in the periphery when the user is using a joystick to move quickly through the space to reduce the chances of simulator sickness.

Fighter pilots have 20 to 40 percent sickness rates in flight simulators - but experienced pilots get sick more often than novice pilots.

In a rotating field when walking forward, people tilt their heads and feel like they are rotating in the opposite direction. Here is a physical version but you can do the same thing with just computer graphics
https://www.youtube.com/watch_popup?v=WP9Edx7--PU

If a person is walking on a treadmill holding onto a stationary bar and you change the rate the the visuals are passing by, it will feel to the person like the bar is pushing or pulling on their hands.

Open fields are less likely to cause problems than walking through tight tunnels; tunnels are very aggressive in terms of peripheral motion. This doesn't mean that you should not have any tunnels, but you should be careful how much time the users spend there and how fast they are allowed to move through them.

Its easy (and fun) to induce vertigo. Most people really seem to enjoy jumping off of high places and falling in VR.

With VR, you very likely have visuals covering the user's field of view, like a child up close to the TV, and you have a full array of special effects to chose from. Choose carefully.

Link to that part of the episode - the content is regularly removed and comes back under a different link - warning - watching this full screen can cause issue for some people.

- audio processing can also help with realism - the same sound will sound different in a closet vs a cathedral vs outside, and if its sounds wrong then its another cue that the virtual world is somehow 'wrong'.

- directional audio can be good to tell a user where something is occurring that might be out of the users current field of view

- audio can also be used to send speech between multiple participants, but we will talk more about that in a future lecture

In a virtual world you have full control of all audio sources. In an augmented reality world you need to blend your audio with the audio of the real world

Simple audio can be monaural with a single speaker but there are many advantages to having multiple speakers as in a surround sound system, or using headphones to give directional audio where sounds seem to come form a particular location. As you get close to these sounds they get louder and are more localized. This can tell you where things have happened, or lead you towards something.

Subwoofers can be good for rumbling. Connecting them to a vibrating floor can add additional feedback and increase the reality of a train passing or an dragon walking by.

The sounds themselves can be prerecorded clips that are played back or looped, or sounds can be synthesized. Recorded sounds tend to be used in realistic virtual worlds. Synthesized sounds can be useful in scientific environments to give feedback on the current state of the world based on data.

Its important to have high-quality prerecorded clips - background hiss can be very noticeable. As with models or textures it is important to credit the creators, and make sure you abide by their restrictions for any audio files that you use. Creating good audio is hard.

You will usually want to play back multiple sounds at the same time - some of these will be looping environmental sounds and others will be sounds played when certain events occur. There are several free audio libraries out there that can handle these things in a pretty straightforward way.

Its important to balance all of these sounds so that some sounds do not unintentionally hide others. You also need to be careful that you do not overload the speakers or the headphones with too many loud sounds.

You also need to be careful that you are not playing too many sounds at the same time, or playing the same sound too many times. For example when there are multiple people running and you want to hear the footsteps, or multiple water droplets hitting a pool, it is probably a bad idea to play a sound for each of those events or you will just hear noise. One method is to have m slightly different versions of the sound and only play that sound if it hasn't been played in the previous n seconds. Its pretty much the same as doing foley work in a film where you want to create the impression of a sound but that doesn't necessarily mean recreating the actual sound.

Its usually a good idea to load all of your sounds into memory at the beginning of the program and store them in memory. Any repeating sound can be set to repeat and play at zero volume, and then faded up when needed and faded down when not needed.

Week 4

Vision / Visuals and Audio

Coming Next Time