Lecture 10 Perception: The problem of depth perception Flashcards
The problem of depth perception
the world is 3D but the retinal image is 2D issue of the visual system extracting 3D depth from the 2D retinal image
Two-dimensional image in each eye
straight lines project onto curved retina
euclidean input
non-euclidean image
What changes the visual angle
both size and distance
Relation between the degree of visual angle on the retina and size of the object
directly proportionate
the bigger the object the larger the angle on the retina
Relation between visual angle and how far away the object is from the person
inversely proportional - the same size object at twice the distance from a person is going to generate a smaller angle on the retina.
Monocular depth cue: based on projective geometry: occlusion
occluded objects are farther in depth, unoccluded objects are nearer
Monocular depth cue: based on projective geometry: relative size
All else being equal:
- Retinal image size is smaller for objects farther in depth
- Retinal image size is larger for objects nearer in depth
Monocular depth cue: based on projective geometry: relative height
The height of the image from the base of the image/image plane.
If the object is lower in the image plane then it will be closer to you, if the object is higher it will be further away.
Objects farther away from us in the visual field lower in the retinal image
Monocular depth cue: based on projective geometry: texture gradients
Gradual change in the appearance of an object from coarse to fine - some objects appear closer because they are coarse and more distinct, but gradually become less and less distinct and and more fine, which makes the objects appear to get further and further away.
Monocular depth cue: based on projective geometry: familiar size
depth cue based on knowledge of the typical size of objects
Monocular depth cue: aerial perspective
light scatter by the atmosphere causes farther objects to appear hazier, less distinct
Monocular depth cue: linear perspective
parallel lines in a 3D world converge in the 2D image
Vanishing point: point at which lines converge
Anamorphic art
use of linear perspective to create depth in a 2D image from a single view point
only works if the person viewing the scene is standing at a particular point
motion parallax
to do with moving through a scene
depth cue bases on head movement, or retinal image at two different points in time
Objects in depth are displaced more in the image than objects further away
Extraretinal depth cues: accomodation
When we focus on an object close to us our lens is bulging - more convex
When we focus on an object far away from us our lens is becoming flatter’
Change in accommodative state - a source of info as to how far away the object is from us
When does accomodation become an ineffective cue to depth
objects further than 2-3 metres as there is a limit as to how flat the lens can get
Extrarential depth cues: convergence
Eyes rotate inward or outward focus on objects: Vergence angle is a depth cue.
Looking at something far away the vergence angle is smaller, looking at something close the vergence is smaller
The convergence state of the eye tells us how far an object is away from us.
Near: converge, Far: diverge
convergence
the way your eyes move together and point inward when you look at nearby objects
diverge
an object farther away, they rotate away from each other
human field of vision compared to rabbits
humans: 190 deg - predator eyes close in front of the head
rabbits: 360 deg - prey need to see all round
binocular disparity
110 deg seen by both eyes
but eyes separated in the head by 6 cm so each eye sees a slightly different image
disparity between retinal image is a depth cue
stereopsis
the ability to use disparity as a depth cue
where is disparity
centre of gaze
where is non-zero disparity
outside the centre of gaze
zero disparity meaning
means that the images received by both eyes are identical
non-zero disparity meaning
a difference in the image seen by each eye
corresponding retinal points
fovea and pairs of points equidistant from fovea
what is the Vieth-Müller circle:
points in space whose images fall on corresponding retinal points
theoretical geometric prediction
in reality isn’t exactly a circle
what happens to objects that arent in Vieth-Müller circle
will have some non-zero disparity
What is a horopter
true zero-disparity plane, flatter than a circle
deviated systematically from V-M circle
measured behaviourally using special apparatus
always included the objects of fixation (fovea)
Zero-disparity objects and horopter
lie on the horopter and are seen as fused
Non-zero disparity objects and horopter
lie off the horopter and produce diplopia (double vision)
Panum’s fusional area
Zone of non-zero disparity around the horopter where objects are perceived as fused.
Crossed disparity
objects closer than fixation (nearer to you than the horopter)Fixating at something far away and the object closer to you will appear double.
uncrossed disparity
objects farther from fixation (farther from you than the horopter)
Fixating at something near the object far away will appear double
how the disparity information is resolved
by disparity sensitive neurons in the brain
Stereoacuity
The smallest difference in depth can be detected by from disparity
How much objects can be displaced before realising they arent lined up
0.1 arc minute (i.e., 1/60 x 1/10 = 0.0016 degrees)
Stereoblindness
the inability to use binocular disparity as a depth cue
3-5% of population
Strabismus:
misalignment of the ocular axes
a condition where the eyes are not aligned properly, causing them to point in different directions, potentially leading to double vision or other vision problems
2-5% of the population
Strabismus: esotropia
eye deviates inward
Strabismus: exotropia
eye deviates outwards
How do disparity-sensitive neurons in the brain work
Binocular neurons tuned to sign and magnitude of disparity
* some neurons respond more for zero or low disparity
* others for higher disparity
* Similar orientation and spatial frequency preference
3D illusions and applications from stereopsis: Wheatstone stereoscope (Charles Wheatstone, 1830s)
an optical instrument that uses two slightly different images, one for each eye, to create the illusion of a three-dimensional image
3D illusions and applications from stereopsis: Holmes stereoscope, 1850s Inventors: David Brewster, Oliver Wendell Holmes
Creates a more realistic impression of what it would actually be like if you were at the scene
3D illusions and applications from stereopsis: Stereoscopic camera, 1950’s
- Sputnik: took 3D photos that could be viewed with a stereoscope
- Screens in the camera separated by same distance as eyes in the head are if it was much more cant fuse if it is too less doesn’t match out disparity perception
- Causes fusion to happen - producing a 3D image
3D illusions and applications from stereopsis: Free fusion (poor man’s stereoscope)
refers to the ability to view stereoscopic images in 3D without the aid of special glasses or a stereoscope, by simply converging or diverging your eyes to fuse the two images.
Random dot stereograms
refers to the ability to view stereoscopic images (like those in a stereoscope) in 3D without the aid of special glasses or a stereoscope, by simply converging or diverging your eyes to fuse the two images.
Dichoptic viewing
the input has been split to the two eyes using the filters, the key difference in the image is the disparity as if projected to each eye in the head.
Polarized glasses (modern dichoptic method)
- Light waves vibrate in different directions, but can be filtered to vibrate in a single direction
- Two cameras, spaced 6-7 cm apart, capture the scene with different polarizations (e.g., horizontal vs. vertical)
- The two images are projected on the screen, each with its own polarization
- Viewers wear glasses with polarizing filters that allow only the corresponding image to reach each eye, creating 3D effect
Other applications of stereopsis:
- 3D medical imaging and surgical planning
- 3D terrain mapping
- 3D ground scanning, surveillance and other defence applications
- Robotics, machine vision, self-driving cars
Size constancy
Perceiving objects as constant in size changes in viewing distance
Closer objects cast larger retinal images, further objects cast smaller retinal images
Yet perceived size doesn’t change that much
Perceived size is constant becuase we discount distance.
Ames room:
A person appears to grow in size when walking across the room
room appears regular but is actually distorted (trapezoidal)
* corners of the room are at different distances from the observer
* retinal image size differs (is smaller for farther corner)
* perceived distance to the two corners is the same (while actual distance differs)
* therefore, perceived size differs
Unable to use the distance cues correctly to judge their size correctly.
Ponzo illusion
Same physical size, upper line appears longer
* could be due to depth induced by linear perspective (upper line appears farther away)
* could be an artifact of how the visual system processes tilted lines when they are shown
with horizontal lines (i.e., not related to depth)
