How do all the technical components of VR headsets, e.g., screens, lenses, tracking, etc., actually come together to create realistic-looking virtual environments? Specifically, why do virtual environment in VR look more “real” compared to when viewed via other media, for example panoramic video?
The reason I’m bringing this up again is that the question keeps getting asked, and that it’s really kinda hard to answer. Most attempts to answer it fall back on technical aspects, such as stereoscopy, head tracking, etc., but I find that this approach somewhat misses the point by focusing on individual components, or at least gets mired in technical details that don’t make much sense to those who have to ask the question in the first place.
I prefer to approach the question from the opposite end: not through what VR hardware produces, but instead through how the viewer perceives 3D objects and/or environments, and how either the real world on the one hand, or virtual reality displays on the other, create the appropriate visual input to support that perception.
The downside with that approach is that it doesn’t lend itself to short answers. In fact, last summer, I gave a 25 minute talk about this exact topic at the 2016 VRLA Summer Expo. It may not be news, but I haven’t linked this video from here before, and it’s probably still timely:
Why do virtual objects close to my face appear blurry when wearing a VR headset? My vision is fine!
And why does the real world look strange immediately after a long VR session?
These are another two (relates ones) of those frequently-asked questions about VR and head-mounted displays (HMDs) that I promised to address a while back.
Here’s the short answer: In all currently-available HMDs, the screens creating the virtual imagery are at a fixed optical distance from the user. But our eyes have evolved to automatically adjust their optical focus based on the perceived distance to objects, virtual or real, that they are looking at. So when a virtual object appears to be mere inches in front of the user’s face, but the screens showing images of that object are — optically — several meters away, the user’s eyes will focus on the wrong distance, and as a result, the virtual object will appear blurry (the same happens, albeit less pronounced, when a virtual object appears to be very far away). This effect is called accommodation-vergence conflict, and besides being a nuisance, it can also cause eye strain or headaches during prolonged VR sessions, and can cause vision problems for a short while after such sessions.
“It can’t be comfortable or healthy to stare at a screen a few inches in front of your eyes.”
The popularity of Google Cardboard, and the upcoming commercial releases of the Oculus Rift, HTC Vive, and other modern head-mounted displays (HMDs) have raised interest in virtual reality and VR devices in parts of the population who have never been exposed to, or had reason to care about, VR before. Together with the fact that VR, as a medium, is fundamentally different from other media with which it often gets lumped in, such as 3D cinema or 3D TV, this leads to a number of common misunderstandings and frequently-asked questions. Therefore, I am planning to write a series of articles addressing these questions one at a time.
First up: How is it possible to see anything on a screen that is only a few inches in front of one’s face?
Short answer: In HMDs, there are lenses between the screens (or screen halves) and the viewer’s eyes to solve exactly this problem. These lenses project the screens out to a distance where they can be viewed comfortably (for example, in the Oculus Rift CV1, the screens are rumored to be projected to a distance of two meters). This also means that, if you need glasses or contact lenses to clearly see objects several meters away, you will need to wear your glasses or lenses in VR.
I have talkedmanytimes about the importance of eye tracking for head-mounted displays, but so far, eye tracking has been limited to the very high end of the HMD spectrum. Not anymore. SensoMotoric Instruments, a company with around 20 years of experience in vision-based eye tracking hardware and software, unveiled a prototype integrating the camera-based eye tracker from their existing eye tracking glasses with an off-the-shelf Oculus Rift DK1 HMD (see Figure 1). Fortunately for me, SMI were showing their eye-tracked Rift at the 2014 Augmented World Expo, and offered to bring it up to my lab to let me have a look at it.
Figure 1: SMI’s after-market modified Oculus Rift with one 3D eye tracking camera per eye. The current tracking cameras need square cut-outs at the bottom edge of each lens to provide an unobstructed view of the user’s eyes; future versions will not require such extensive modifications.
This is a tricky question, and we have to be precise. So let’s first define some terms.
When talking about “VR movies,” people are generally referring to live-action movies, i.e., the kind that is captured with physical cameras and shows real people (well, actors, anyway) and environments. But for the sake of this discussion, live-action and pre-rendered computer-generated movies are identical.
We’ll also have to define what we mean by “work.” There are several things that people might expect from “VR movies,” but not everybody might expect the same things. The first big component, probably expected by all, is panoramic view, meaning that a VR movie does not only show a small section of the viewer’s field of view, but the entire sphere surrounding the viewer — primarily so that viewers wearing a head-mounted display can freely look around. Most people refer to this as “360° movies,” but since we’re all thinking 3D now instead of 2D, let’s use the proper 3D term and call them “4π sr movies” (sr: steradian), or “full solid angle movies” if that’s easier.
The second component, at least as important, is “3D,” which is of course a very fuzzy term itself. What “normal” people mean by 3D is that there is some depth to the movie, in other words, that different objects in the movie appear at different distances from the viewer, just like in reality. And here is where expectations will vary widely. Today’s “3D” movies (let’s call them “stereo movies” to be precise) treat depth as an independent dimension from width and height, due to the realities of stereo filming and projection. To present filmed objects at true depth and with undistorted proportions, every single viewer would have to have the same interpupillary distance, all movie screens would have to be the exact same size, and all viewers would have to sit in the same position relative the the screen. This previous post and video talks in great detail about what happens when that’s not the case (it is about head-mounted displays, but the principle and effects are the same). As a result, most viewers today would probably not complain about the depth in a VR movie being off and objects being distorted, but — and it’s a big but — as VR becomes mainstream, and more people experience proper VR, where objects are at 1:1 scale and undistorted, expectations will rise. Let me posit that in the long term, audiences will not accept VR movies with distorted depth.
Figure 1: The zSpace holographic display and how it would really look like when seen from this point of view.
So I drove around the bay to get a close look at the zSpace, to determine its viability for my purpose. Bottom line, it will work (with some issues, more on that below). My primary concerns were threefold: head tracking precision and latency, stylus tracking precision and latency, and stereo quality (i.e., amount of crosstalk between the eyes).
My friend Serban got his Oculus Rift dev kit in the mail today, and he called me over to check it out. I will hold back a thorough evaluation until I get the Rift supported natively in my own VR software, so that I can run a direct head-to-head comparison with my other HMDs, and also my screen-based holographic display systems (the head-tracked 3D TVs, and of course the CAVE), using the same applications. Specifically, I will use the Quake ||| Arena viewer to test the level of “presence” provided by the Rift; as I mentioned in my previous post, there are some very specific physiological effects brought out by that old chestnut, and my other HMDs are severely lacking in that department, and I hope that the Rift will push it close to the level of the CAVE. But here are some early impressions.
Figure 1: What it would look like to unbox an Oculus VR dev kit, if one were to have such a thing.
I received an email about a week ago that reminded me that, even though stereoscopic movies and 3D graphics have been around for at least six decades, there are still some wide-spread misconceptions out there. Those need to be addressed urgently, especially given stereo’s hard push into the mainstream over the last few years. While, this time around, the approaches to stereo are generally better than the last time “3D” hit the local multiplex (just compare Avatar and Friday the 13th 3D), and the wide availability of commodity stereoscopic display hardware is a major boon to people like me, we are already beginning to see a backlash. And if there’s a way to do things better, to avoid that backlash, then I think it’s important to do it.
So here’s the gist of this particular issue: there are primarily two ways of setting up a movie camera, or a virtual movie camera in 3D computer graphics, to capture stereoscopic images — one is used by the majority of existing 3D graphics software, and seemingly also by the “3D” movie industry, and the other one is correct.
Toe-in vs skewed frustum
So, how do you set up a stereo camera? The basic truth is that stereoscopy works by capturing two slightly different views of the same 3D scene, and presenting these views separately to the viewers’ left and right eyes. The devil, as always, lies in the details.
Say you have two regular video cameras, and want to film a “3D” movie (OK, I’m going to stop putting “3D” in quotes now. My pedantic point is that 3D movies are not actually 3D, they’re stereoscopic. Carry on). What do you do? If you put them next to each other, with their viewing directions exactly parallel, you’ll see that it doesn’t quite give the desired effect. When viewing the resulting footage, you’ll notice that everything in the scene, up to infinity, appears to float in front of your viewing screen. This is because the two cameras, being parallel, are stereo-focused on the infinity plane. What you want, instead, is that near objects float in front of the screen, and that far objects float behind the screen. Let’s call the virtual plane separating “in-front” and “behind” objects the stereo-focus plane.
So how do you control the position of the stereo-focus plane? When using two normal cameras, the only solution is to rotate both slightly inwards, so that their viewing direction lines intersect exactly in the desired stereo-focus plane. This approach is often called toe-in stereo, and it sort-of works — under a very lenient definition of the words “sort-of” and “works.”
The fundamental problem with toe-in stereo is that it makes sense intuitively — after all, don’t our eyes rotate inwards when we focus on nearby objects? — but that our intuition does not correspond to how 3D movies are shown. 3D (or any other kind of) movies are not projected directly onto our retinas, they are projected onto screens, and those screens are in turn viewed by us, i.e., they project onto our retinas.
Now, when a normal camera records a movie, the assumption is that the movie will later be projected onto a screen that is orthogonal to the projector’s projection direction, which is implicitly the same as the camera’s viewing direction (the undesirable effect of non-orthogonal projection is called keystoning). In a toe-in stereo camera, on the other hand, there are two viewing directions, at a slight angle towards each other. But, in the theater, the cameras’ views are projected onto the same screen, meaning that at least one, but typically both, of the component images will exhibit keystoning (see Figures 1 and 2).
Figure 1: The implied viewing directions and screen orientations caused by a toe-in stereo camera based on two on-axis projection cameras. The discrepancy between the screen orientations implied by the cameras’ models and the real screen causes keystone distortion, which leads to 3D convergence issues and eye strain.
Figure 2: The left stereo image shows the keystoning effect caused by toe-in stereo. A viewer will not be able to merge these two views into a single 3D object. The right stereo image shows the correct result of using skewed-frustum stereo. You can try for yourself using a pair of red/blue anaglyphic glasses.
The bad news is that keystoning from toe-in stereo leads to problems in 3D vision. Because the left/right views of captured objects or scenes do not actually look like they would if viewed directly with the naked eye, our brains refuse to merge those views and perceive the 3D objects therein, causing a breakdown of the 3D illusion. When keystoning is less severe, our brains are flexible enough to adapt, but our eyes will dart around trying to make sense of the mismatching images, which leads to eye strain and potentially headaches. Because keystoning is more severe towards the left and right edges of the image, toe-in stereo generally works well enough for convergence around the center of the images, and generally breaks down towards the edges.
And this is why I think a good portion of current 3D movies are based on toe-in stereo (I haven’t watched enough 3D movies to tell for sure, and the ones I’ve seen were too murky to really tell): I have spoken with 3D movie experts (an IMAX 3D film crew, to be precise), and they told me the two basic rules of thumb for good stereo in movies: artificially reduce the amount of eye separation, and keep the action, and therefore the viewer’s eyes, in the center of the screen. Taken together, these two rules exactly address the issues caused by toe-in stereo, but of course they’re only treating the symptom, not the cause. As an aside: when we showed this camera crew how we are doing stereo in the CAVE, they immediately accused us of breaking the two rules. What they forgot is that stereo in the CAVE obviously works, including for them, and does not cause eye strain, meaning that those rules are only workarounds for a problem that doesn’t exist in the first place if stereo is done properly.
So what is the correct way of doing it? It can be derived by simple geometry. If a 3D movie or stereo 3D graphics are to be shown on a particular screen, and will be seen by a viewer positioned somewhere in front of that screen, then the two viewing volumes for the viewer’s eyes are exactly the two pyramids defined by each eye, and the four corners of the screen. In technical terms, this leads to skewed-frustum stereo. The following video explains this pretty well, better than I could here in words or a single diagram, even though it is primarily about head tracking and the screen/viewer camera model:
In a nutshell, skewed-frustum stereo works exactly as ordered. Even stereo pairs with very large disparity can be viewed without convergence problems or eye strain, and there are no problems when looking towards the edge of the image.
To allow for a real and direct comparison, I prepared two stereoscopic images (using red/blue anaglyphic stereo) of the same scene from the same viewpoint and with the same eye separation, one using toe-in stereo, one using skewed-frustum stereo. They need to be large and need to be seen at original size to appreciate the effect, which is why I’m only linking them here. Ideally, switch back-and-forth between the images several times and focus on the structure close to the upper-left corner. The effect is subtle, but noxious:
I generated these using the Nanotech Construction Kit and Vrui; as it turns out, Vrui is flexible enough to support bad stereo, but at least it was considerably harder setting it up than good stereo. So that’s a win, I guess.
There are only two issues to be aware of: for one, objects at infinity will have the exact separation of the viewer’s eyes, so if the programmed-in eye separation is larger than the viewer’s actual eye separation, convergence for very far away objects will fail (in reality, objects can’t be farther away than infinity, or at least our brains seem to think so). Fortunately, the distribution of eye separations in the general population is quite narrow; just stick close to the smaller end. But it’s a thing to keep in mind when producing stereoscopic images for a small screen, and then showing them on a large screen: eye separation scales with screen size when baked into a video. This is why, ideally, stereoscopic 3D graphics should be generated specifically for the size of the screen on which they will be shown, and for the expected position of the audience.
The other issue is that virtual objects very close to the viewer will appear blurry. This is because when the brain perceives an object to be at a certain distance, it will tell the eyes to focus their lenses to that distance (a process called accommodation). But in stereoscopic imaging, the light reaching the viewer’s eyes from close-by virtual objects will still come from the actual screen, which is much farther away, and so the eyes will focus on the wrong plane, and the entire image will appear blurry.
Unfortunately, there’s nothing we can do about that right now, but at least it’s a rather subtle effect. In our CAVE, users standing in the center can see virtual objects floating only a few inches in front of their eyes quite clearly, even though the walls, i.e., the actual screens, are four feet away. This focus miscue does have a noticeable after-effect: after having used the CAVE for an extended period of time, say a few hours, the real world will look somewhat “off,” in a way that’s hard to describe, for a few minutes after stepping out. But this appears to be only a temporary effect.
Taking it back to the real 3D movie world: the physical analogy to skewed-frustum stereo is lens shift. Instead of rotating the two cameras inwards, one has to shift their lenses inwards. The amount of shift is, again, determined by the distance to the desired stereo-focus plane. Technically, creating lens-shift stereo cameras should be feasible (after all, lens shift photography is all the rage these days), so everybody should be using them. And some 3D movie makers might very well already do that — I’m not a part of that crowd, but from what I hear, at least some don’t.
In the 3D graphics world, where cameras are entirely virtual, it should be even easier to do stereo right. However, many graphics applications use the standard camera model (focus point, viewing direction, up vector, field-of-view), and can only represent non-skewed frusta. The fact that this camera model, as commonly implemented, does not support proper stereo, is just another reason why it shouldn’t be used.
So here’s the bottom line: Toe-in stereo is only a rough approximation of correct stereo, and it should not be used. If you find yourself wondering how to specify the toe-in angle in your favorite graphics software, hold it right there, you’re doing it wrong. The fact that toe-in stereo is still used — and seemingly widely used — could explain the eye strain and discomfort large numbers of people report with 3D movies and stereoscopic 3D graphics. Real 3D movie cameras should use lens shift, and virtual stereoscopic cameras should use skewed frusta, aka off-axis projection. While the standard 3D graphics camera model can be generalized to support skewed frusta, why not just replace it with a model that can do it without additional thought, and is more flexible and more generally applicable to boot?
Update: With the Oculus Rift in developers’ hands now, I’m getting a lot of questions about whether this article applies to head-mounted displays in general, and the Rift specifically. Short answer: it does. There isn’t any fundamental difference between large screens far away from the viewer, and small screens right in front of the viewer’s eyes. The latter add a wrinkle because they necessarily need to involve lenses and their concomitant distortions so that viewers are able to focus on the screens, but the principle remains the same. One important difference is that small screens close to the viewer’s eyes are more sensitive to miscalibration, so doing stereo right is, if anything, even more important than on large-screen displays. And yes, the official Oculus Rift software does use off-axis projection, even though the SDK documentation flat-out denies it.
I don’t think there’s need to introduce the Oculus Rift HMD. Everyone’s heard of it, and everyone’s psyched – including me.
However, HMDs are prone to certain issues, and while that shouldn’t detract us from embracing them, we should be careful to do it right this time. The last thing the VR field needs right now is a viral YouTube video along the lines of “Oh, an Oculus Rift! Cool! Let me try it on… Wow, that’s awesoBLEEAAARRGHHH.”
To back up a little: when HMDs became a thing in the 80s, they tended to induce dizziness and nausea in viewers, after a relatively short time of using them. Interestingly, HMDs had generally worse effects than other types of immersive display environments such as CAVEs. The basic theory of simulation sickness is based on virtual motion, and does not account for this difference.
The commonly stated explanation for this difference is display lag. In an HMD, the screens move with the viewer’s head, and any delay will cause the virtual world to move along with the viewer until the display system catches up. Imagine wearing an HMD and quickly turning your head to the side. Say it takes 30ms total until this motion is noticed by the head tracking system, the application updates its internal state, renders the new state, and refreshes the HMD’s screens. During this interval, the world will turn with you, and it will snap back to its original orientation once the delay time has passed. The real world does not behave like that, and because HMD-based graphics tap deeply into our brain’s visual system, this is very disorienting and adds to the discomfort. In a CAVE, on the other hand, the screens do not move with the viewer. Delay will still cause a disturbance in the projection of the virtual world, as the actual viewer position will not match the virtual one, but because screens are large and relatively far away, this will be barely noticeable. So far, so good.
Alas, there is an additional, often overlooked, factor — display calibration. Any immersive graphics system, HMD or CAVE or else, needs to exactly replicate how virtual objects are projected onto the system’s real screens, and then seen by the user (how exactly that works is a topic for another post). The bottom line is that the graphics software needs to know the absolute positions and orientations of all screens, and the absolute positions of the viewer’s eyes. Determining this is the job of head tracking and system calibration. But in an HMD, unlike in a CAVE, the tolerances for calibration are very low. The screens are very small and very close to the viewer’s eyes, which doesn’t leave much room for error (see Figure 1). Even worse, there is no way to precisely don an HMD short of putting screws into one’s skull; every time you put it on, it sits slightly differently. And that means any pre-configured projection parameters will not match reality.
Figure 1: Diagram of a hypothetical HMD for calibration purposes. The HMD consists of small real screen mounted directly in front of the viewer’s eyes, and uses optics to create larger virtual screens at a longer distance away to allow users to properly focus on those screens. For proper calibration, graphics software needs to know the precise positions of the viewer’s pupils and the exact positions and sizes of the virtual screens, in some coordinate system. Head tracking will provide the mapping from this viewer-attached coordinate system to the world coordinate system to allow users to look and walk around.
These mismatches have several effects. For one, imagine that a viewer wears an HMD slightly askew, so that the two screens have different vertical positions in front of their respective eyes. If the software does not account for that, the two stereo images will be vertically displaced, something that does not happen in real life. The viewer’s eyes will make up for it, up to a point, by moving up/down independently, but that is an unnatural motion and causes eye strain. It’s the same effect as watching a 3D movie in a theater while not holding one’s head level — it will hurt later.
Another, more subtle, effect is that in a miscalibrated display system the virtual world does not behave as the real world would. Do a simple experiment: fire up some first-person video game that allows view configuration, such as Doom3, and set a high field of view. Then rotate the view and observe. The virtual world will display a strong distortion effect, meaning that the sizes of objects, and their internal angles, change as the viewpoint changes. This is an extreme example, but even slight discrepancies are subconsciously unsettling, because our visual system is very good at detecting if something is not right with the world, and it tells us that by making us sick.
Even in non-immersive 3D graphics, a too large discrepancy between real field of view (how large the screen looms in our visual field) and programmatic field-of-view is known to cause motion sickness, and immersive 3D graphics with the same issue will be much worse. FOV discrepancy is only one symptom of miscalibration, but it’s the one that’s easiest to demonstrate; the others are more subtle (but that’s a topic for another post). In the end, miscalibration is a nasty problem because it is subtle, very hard to correct, and causes significant ill effects.
I noticed these things when I started experimenting with my own HMDs a while ago (I have an eMagin Z800 3DVisor and a Sony HMZ-T1). I experimented with rapid motions, but those didn’t really make me dizzy. I did notice, however, that the world didn’t seem solid, but as if it was made from jelly. I expected that, not having done proper calibration yet, so I used an interactive calibration utility to set up the system just so. After that, the world seemed stable, and interestingly I didn’t notice any more issues from lag. Not having done any further experiments, my hunch is that miscalibration is actually a bigger problem than lag. (Disclosure: while I was using a low-latency Intersense IS-900 tracking system, the computer running the show was fairly old, and the Quake3 renderer had no particular performance tweaking, so I estimated total system delay around 30ms).
So what’s the take-home message from this wall of text? If we want HMDs to succeed, we need to treat them properly in our graphics software. We need to use proper projection models instead of the standard camera model (but that’s a topic for another post), and not simply apply ad-hoc stereo models such as toe-in etc. (but that’s a topic for another post). It might work for a demo, but it won’t be pretty, and it will make our users sick. Instead, we need to know exactly how the HMD is laid out internally (screen placement and size, effects from the optical system in front of the screens, lens distortions, etc.), and, just as importantly, we need to know exactly where the viewer’s eyes are with respect to the screens (see Figure 1). This last one is the hard part. Maybe a future perfect HMD will contain one pair of stereo cameras per screen that will accurately track the viewer’s pupils and allow the graphics software to set up the projection parameters correctly, no matter how the HMD is worn and how the viewer moves. But until then, we need to come up with a practical approach, and we need to find simple methods to calibrate HMDs on the fly, and teach our users how to use those methods.
Well, and, of course, we mustn’t forget about minimizing lag, either. That would be too easy.
Oh, and by the way, want to get a quick glimpse of just how immersive the Oculus Rift will be (going by current specs)? If your monitor is X inches wide, put your eye X/2 inches in front of the monitor’s center — that’s about what it will look like. If you want to play a first-person game from that viewpoint and have it look right, set the horizontal field of view to 90 degrees.