Week 9 Depth Perception PDF

Preview

Summary

Week 9 lecture notes...

Description

[1] Properties of “depth” cues: Binocular vs monocular: - Could you get this info with only one eye? Pictorial vs non-pictorial: - Could the info be available in a photograph/picture/painting? - Pictorial depth cues tend to be Monocular Visual vs Oculomotor: - Visual: Information comes purely from the retinal image - Oculomotor: Information comes from eye muscle signals Ordinal vs Metric: - Does the information tell us which object is closer or by how much closer is it (i.e metric)

Advantages of 2 eyes: - Most animals have 2 eyes Some are placed laterally: An eye on each side - Generally tend to be prey animals (e.g duck, rabbit) - Visual fields overlap minimally - All round coverage of the world to detect predators Others are frontal: - Generally tend to be predator animals (e.g cat, owls) - Visual fields overlap considerably - Some areas of space not covered at all

Pros and Cons of frontal eyes: - Disadvantage of having frontally placed eyes is that it reduces the total visual field Note: But for objects that appear in the binocular field it gives us... Binocular summation: - Two “chanes” to see the object - Lower thresholds for detection/discrimination Depth perception: - Two seperate viewpoints - Allow us to extract depth information

Vergence: - Generally, fixated objects fall on our fovea in each eye - Each eye has to be pointing in the correct direction Vergence: The angular position of each eye can be used to calculate the depth of a fixated object - For long distances vergence angle will be small (i.e top image), for short distances vergence angle will be large (i.e bottom image) - This can give us weak depth information Note: Binocular, non-pictorial, oculomotor, metric

Stereopsis: Stereopsis = Binocular vision cues to depth - Each eye has a unique viewpoint, hence sees subtly different things Euclid & Galen (300-200BC): - It is impossible for both eyes to see the same parts of a sphere or column Leonardo da vinci (1500): - In real viewing, a foreground object blocks the view of different parts of the background in each eye. Paintings CAN’T simulate this Wheatstone (1838): - Difference in retinal position of objects in the two eyes signals depth Note: Although the far finger is to the left of the near one in your left eye, it is right to the near finger the in the right eye

Binocular disparity: Also known as retinal disparity/binocular parallax - For any two objects at different depths, the relative position of their image is different in the two eyes - First formally described by Wheatsone - Small depth difference = small disparity - Large depth difference = large disparity Note: Binocular, non-pictorial, visual, metric

Corresponding points: Imagine that both eyes are fixating at a cross a couple of metres away.. - Cross falls on the fovea of each eye - L & R half-images of the cross fall on corresponding points of each retina i.e This just means they are in the same place - A circle is at the same distance as the cross and just to the left - Half-images fall just to the right on each retina also corresponding points Note: For the triangle behind the cross… - Half-images fall just to the left of the cross on the left retina but a long way to the left on the right retina - Non-corresponding points - Object has binocular disparity

The Horopter: - A line of all possible locations where an objects half images fall on corresponding points - Defines locations at which objects have 0 disparity - Objects nearer than the horopter have a crossed disparity (i.e think of cross eyed) - Objects further than the horopter have an uncrossed disparity

Fusion & Diplopia: Q: If we have 2 different images, why do we see 1 object? A: Because of fusion

- Panum’s fusional area zone around the horopter where single vision occurs (i.e grey area in diagram) - Combine two locations of two objects equates to good metric depth - Outside this area, diplopia occurs which is where you see two unfused images Note: Diplopic images usually don’t give good metric depth

Fixation, fusion & focus: Fixation: The act of directing your gaze towards something so that its image falls in the middle of your fovea (i.e if you fixate an object with both eyes you “converge” on it ) Fusion: The phenomenon that some objects with small disparities are not seen as “double” instead their images are “fused” into one Focus: If the current state of the lens causes light rays spreading from a point at a given distance to meet at a single point on the retina, objects at this depth will be “in focus”

[2] How to stimulate stereoscopic 3D: - Need 2 images taken approximately 6.5cm apart - Display one to each eye Note: Many methods of presenting images and ensuring that each eye receives appropriate image - Mirror stereoscopes (e.g wheatstone stereoscope) - Prism/lens stereoscope (e.g viewmaster/brewster stereoscope) - Anaglyphs (i.e red/blue glasses) - Polarisation (i.e 3D cinema) - Shutters (e.g 3D TVs) - Vertically interlaced displays (i.e lenticullar/parallax barrier) - Free fusion (e.g RDS & autostereograms)

Mirror stereoscopes: - Mirrors angled at 45 degrees in front of observer - Half-images (E, E’) and mirrors (A, A’) should be arranged to simulate the appropriate vergence angle - Images are reversed & only for one observer

Free Fusion: - Present images side by side - Adjust own vergence to align images i.e Convergence (i.e cross eyed) or Divergence (i.e wall eyed) - Vergence is quite unnatural; however, it is cheap and Easy

Autostereograms (Tyler, 1979): - Also “SIRDS” single image random dot stereograms - One image with pattern that repeats horizontally - Adjust vergence to align neighbouring regions Note: Horizontal distance between repeating items varies as it introduces disparity

RDS & Cyclopean vision: - With the RDS, Julesz introduced a new phenomenon: Cyclopian vision - Left eye sees a bunch of dots - Right eye sees a bunch of dots - When the two are presented together, we see two square regions (in depth) that were NOT present in either half of the image - These squares are described as cyclopian as they emerge only after binocular combination

[3]: Familiar size - Familiar objects that have a stereotypical size can appear further away when the retinal image is smaller - For objects of constant size (i.e almost all objects) it’s retinal image varies with distance - Simple relationship (e.g half the distance = double the retinal image size) Note: Monocular, pictorial, visual, metric

Perspective: - Uses essentially the same information as familiar cue size - Don’t have to recognise the object, BUT assumptions are still made Linear perspective (convergence): - Assumes lines are parallel Detail perspective (texture gradient): - Assumes texture elements are of familiar size Note: Monocular, pictorial, visual, metric - Perspective can be “forced” to make you mispercieve depth e.g Sports field advertising or anamorphic art

Motion parallax: - As the observer moves, stationary objects at different depths move at different velocities Note: Imagine staring off into the distance through a train/car - Closer things move more quickly - More distant things move more slowly Note: Monocular, visual, metric - Not strictly pictorial as it requires motion

Height in the visual field: - Assuming that things generally tend to rest on the ground plane, things higher up in the visual field are often more distant - Note also the apparent size of the 3 different men in this figure Note: Monocular, visual, pictorial, metric

Aerial/Atmospheric perspective: - Not connected to linear detail/perspective - More distant objects tend to be lower contrast, lighter, more blue

- Particles in the atmosphere (i.e gas, water) fade the image by scattering light - The larger the distance the larger the scattering of light by intermediate particles - Contrast-depth relationship depends on what local atmosphere is like Note: Monocular, visual, pictorial metric

Occlusion (Interposition): - Opaque foreground objects occlude more distant objects for areas lying along the same line of light Note: Monocular, pictorial, visual, ordinal - Some artists deliberately get occlusion “wrong” the create a specific effect - This can also create “impossible figures”

Cast Shadow: - Position of cast shadow can influence perceived depth - Depth is ambiguous as an object could lie anywhere along the line of sight Note: Assume light source is constant within a scene. Here it’s overhead as shadow lies on the ground plane (i.e whose depth is revealed through linear perspective) - Separation between object and shadow represents height of object in image - Given visual direction and height above ground plane, depth is disambiguated Note: Monocular, pictorial, visual, metric

Attached Shadow: - Light reflectance patterns reveal the angle of surfaces with respect to the light source and the observer e.g Circular regions appear concave or convex, depending on pattern of shading - Rotating 180 degrees reverses the perceived depth - Again, assume light source is uniform, and located above Note: Monocular, pictorial, visual, metric

Blur: - The object being viewed foveally is usually in focus - Blur varies with depth compared to viewed object - BUT both near and far objects are blurred Note: Look at the edge between regions - Occlusion means that the edge takes on the blur property of the near object - Blurred edge => Blurred object is near - Sharp edge => Sharp object is near Note: Monocular, pictorial, visual, metric

Accomodation: - Lens must change shape to maintain a sharply focused image as object distance varies Note: State of ciliary muscles controlling accommodation offers a cue to depth

- While fixating a distant object (i.e left) the ciliary muscles are relaxed and the lens is relatively thin - While fixating a near object (i.e right) the cilary muscles are tensed to allow the lens to take on a thicker shape Note: Monocular, non-pictorial, oculomotor, metric

Perceived depth & Perceived size: - Stare at the cross in the image on top right for 30 sec then stare at your hand. How large is the afterimage? - Then stare at a far surface. How large is the afterimage now Note: Afterimage looks small projected at a short distance but large for long distances Why? Emmerts law: - For a given retinal image size, perceived size of object is proportional to perceived distance - Could this explain visual illusions such as the ponzo illusion Misapplied constancy scaling: - Visual system assigns a depth to the object - Perceived size is calculated with respect to this depth percept - If the depth percept is inaccurate, an error in perceived size would be predicted Note: Ames room - Julian Beevers pavement art (i.e figures on left look small cause depth is underestimated)...