A sample of some of the great reference on eye movements available at the BBC Motion Gallery
The retina of the human eye is divided into a large outer ring of highly light-sensitive but color-insensitive rods, and a comparatively small central region of color-sensitive cones. This is to allow for diurnal (day and night) vision. That tiny, cone-rich center region is the fovea, and provides our high acuity vision. The outer ring only provides peripheral vision. All detailed observations are made with the fovea, which must constantly be oriented to different parts of the viewed scene by successive fixations.
Frequent, precise eye movements are required because the high acuity vision is restricted to a very small (2º) region surrounding the fixation point. Visual acuity drops off precipitously from that point. The human visual system takes advantage of this high-acuity region by rapidly reorienting the eyes via very fast eye movements called saccades.
For reference, a visual angle of 2° is slightly less than the width of one’s thumb held out at arm’s length, or approximately the width of an average word held at normal reading distance. That’s ALL we normally see with clarity! Detailed observations of a reasonable part of the surrounding world therefore requires moving the eye (and head and body) to successively focus different parts of the ambient light array on the fovea, thus foveating various regions of the observed scene. This is what we must portray through our animation if our characters are to look alive and appear to be observing and interacting with their world.
The human visual system takes advantage of the high resolving power of the fovea by reorienting the fixation point around the viewed scene an average of three times each second via saccadic eye movements, but saccades aren’t the only type of eye movements. Eye movements can be classified into seven different types:
- is a motion of both eyes relative to each other that ensures that an object is still foveated by both eyes when its distance from the observer is changed. The closer the object is, the more the eyes point towards each other. This movement can be voluntarily controlled, but is normally the result of a moving stimulus.
- are the principal method for moving the eyes to a different part of the visual scene, and are sudden, rapid movements of the eyes. Saccades can be initiated voluntarily, but are ballistic: that is, once they are initiated, their path of motion and destination cannot be changed. Visual input is suppressed during a saccade.
- Pursuit motion
- is a much smoother, slower movement than a saccade; it acts to keep a moving object foveated. It cannot be induced voluntarily, but requires a moving object in the visual field. One frequent failing of thoughtless animation is having the eyes demonstrate pursuit motion when there is no object being followed by the character’s eyes.
- is a saw-toothed pattern of eye movements that occurs as a response to the turning of the head (acceleration detected by the inner ear) or the viewing of a moving, repetitive pattern (the train window phenomenon). It consists of smooth `pursuit’ motion in one direction to follow a position in the scene, followed by a fast motion in the opposite direction to select a new position. This is an eye movement that has probably never been animated, and if it was, it would probably get rejected by the supervisor or director because it would look so odd.
- Drift and microsaccades
- occur during fixations, and consist of slow drifts followed by very small saccades (microsaccades) that apparently have a drift-correcting function. These movements are involuntary, and their function is in question.
- Physiological nystagmus
- is a high-frequency oscillation of the eye (tremor) that serves to continuously shift the image on the retina, thus calling fresh retinal receptors into operation. Physiological nystagmus actually occurs during a fixation period, is involuntary, and generally moves the eye less than 1°. As with microsaccades, you’d need to be in an extreme close-up for this to register, but it’s another reason the human eye looks ‘alive’ in live-action extreme close-ups.
- of the eyes is a rotational motion around an axis passing through the fovea and pupil. It is involuntary, and is influenced by among other things the angle of the neck. Although this is also something we can safely ignore as animators, some rigs will automatically provide this when using the ‘eye-target’ controller.
Of the above, it’s really the first three we’re concerned with. Because of the precision that CG animation allows, we can (and should) pay much more attention to these types of movements than is typically done in hand-drawn animation.
Convergence is fairly straightforward. In a properly rigged character, convergence should happen naturally as the eye target is moved close to the character’s face. If a rig lacks this function, or the eye-target control isn’t being used, it should be a simple matter to manually add in some convergence when a character is looking at something very close. You don’t need to go too far with this to portray convergence.
Pursuit is also straightforward. I like to use the eye target to match the motion of the object being tracked, and I like to make sure the head motion (if the head is rotating in the direction of the object being tracked) is somewhat out of phase with the eye movement, otherwise the sense of the eyes actually moving in pursuit motion is lost.
There really aren’t any timing considerations to convergence and pursuit, but there are for saccades. I’ve heard various rules of thumb regarding how many frames* a saccade should take, and how long the eyes should fixate between saccades. Rather than regurgitate those, I’ll lay out some data from physiology studies. I encourage readers to do their own research — look closely at reference like I’ve posted above, and see what the eyes really do.
It takes about 100-300 milliseconds (ms) to initiate a saccade. That is, from the time a stimulus is presented until the eye starts moving takes 0.1 to 0.3 seconds, or between 2 to 7 frames (a common mistake is to have the animated character react immediately — on the same frame — to a visual input).
A saccade takes another 30-120 ms to complete, depending on, among other things, the visual angle traversed. That means the typical saccade takes from 1 to 3 frames. This is where I see a lot of mistakes, usually with saccades that last 5, 6, even 8 frames, or sometimes very large saccades that occur in a single frame. Basically, you can reserve 1-frame saccades for very small movements, 2-3 frame saccades the rest of the time, and very rarely use a 4-frame saccade for a large, deliberate eye movement.
Again, saccades are ballistic, and they do not have any anticipatory movement (unless you’re going for an overtly comic or bizarre effect), nor do they have any overshoot. The fact that the saccades cannot be adjusted on the fly means that our brain makes a best guess for the fixation target of the saccade. The result is that often a relatively large eye dart is followed almost immediately by a smaller ‘adjustment’ saccade. This can be seen clearly in some of the BBC Motion reference footage.
Interestingly, not only is visual processing suppressed during saccades, but this suppression actually begins about 50ms (about 1 frame) before the saccade is initiated. I’ve heard animators speculate that one reason we blink during eye movements is that the closing of the eyes keeps us from getting confused by a ‘blurry’ visual image during an eye dart, or that the blinking ‘resets’ the eye’s focus. Since visual input is suppressed just before, and during, the saccade, we can see this logic is reversed. Our visual system takes advantage of the brief moment of visual ‘down time’ during the saccade to get some lubrication to the surface of the eye with a quick blink. The blink during the head movement also provides some physical protection to the surface of the eye from potential external trauma.
So processing of the retinal image takes place mainly between the saccades, during the fixations. These fixations last for about 200-600 ms (5-15 frames), with 300 ms (7-8 frames) being average, during scanning behavior. Obviously we can fixate on a target for much longer periods, depending on the situation.
Because of the above physiology, I like to use the eye-target control as often as possible. Most of the time this is how our eyes actually work: fixating on a point even as our head and body move. When it comes to the eyes, by far the most common mistake I see in beginning animator’s work is letting the eyes drift around with the head and body movement. In these kinds of scenes, the eyes tend to fixate only when the head and body settle. This will undermine even the most polished and nuanced acting scene.
Of course, if you’re using the eye-target, you need to build in appropriate fixations and saccades. This can take a lot of time, and once done, the lids need to be adjusted properly (more on this in a later post). One problem is that the eyes can look overly busy when they’re animated realistically in this manner. Maybe this is because animated characters tend to have abnormally large eyes (so we notice the eye darts more), or perhaps because we tend to accept ‘under-animated’ eyes in the animation we’ve grown up with. If you really pay attention to what people do with their eyes in everyday situations, you quickly see that we edit out a lot when we’re animating. Frankly, I think we edit out too much, and that there’s a tremendous amount of good acting that could be done with very subtle movements of the eyes and lids.
One trick I’ve found useful to avoid having too many saccades and fixations is to cheat the eye-target during a fixation. Imagine the character’s head is rotating from right to left. The eyes (and the eye-target controller) make a saccadic motion to the left (with the head motion), then fixate. If you ever-so-slightly drift the eye-target further to the left during the fixation, the eyes will appear to remain fixed on the same spot, and can be held for a longer fixation without the appearance of nystagmus. This way during a long close-up scene the number of saccades can be kept down.
What I haven’t discussed here is the relationship to these kinds of eye movements to acting, how to use these eye movements to show both intent and to show a character’s internal process, or how the eye lids and brows respond during various eye movements. Hopefully that’s coming up.
*As always, I’m using the feature film standard of 24 frames per second.