Lecture 6 - Depth Perception Flashcards
what is depth perception
- the ability to perceive visual info in 3D - which is important for interacting with the word and recognising objects
the inverse problem:
and the cues we use
- any retinal image is consistent with infinitely many possible configurations of the world
- we resolve these using cues:
1. binocular cues
2. motion cues
3. pictorial cues
4. oculomotor cues
binocular cues:
- stereo vision with binocular disparity
- arises as we have two eyes with different locations so images received vary and therefore we can match up images to work out the differences
motion cues:
- motion parallax (self-motion)
- Kinetic depth (KDE) - object motion - different parts of surface move at different speeds.
oculomotor cues:
- convergence: muscles in the ye that tell them how much to converge
- accommodation: if looking at something quite close to you lens ned to be fat - thinner if far away - another source of info being the ciliary muscles.
ambiguity in depth perception: PERSPECTIVE
- to overcome this - we can make the assumption that LINES IN THE WORLD TEND TO BE PARALLEL.
- if lines converging this informs us of depth of that scene
ambiguity in depth perception: SHADING
- retinal image can be created by:
- a convex object that is lit from above
- concave lit from below - we can overcome this ambiguity by using prior knowledge, or prior assumptions to interpret the image
- assumptions gained through knowledge and experience of the physical properties of the world
- top-down processing (Constructivist view on vision)
- we can use the assumption THAT LIGHT COMES FROM ABOVE
Ambiguity in Texture
to interpret these retinal images that could be 2D or 3D (e.g. square with different dot sizes or sphere with uniform pattern) we assume two things:
- SURFACES TEXTURS ARE ISOTROPIC (unbiased orientation)
- and HOMOGENEOUS (uniform density)
Main assumptions in depth perception
A: we assume objects rest on the ground plane
B: we assume lines are parallel or at right angles
C: we assume light comes from above
D: we assume surface texture is isotropic and homogeneous
when these assumptions are violated we experience illusions.
Multi-cue perception, why do we need integration
to overcome issues with:
- reliability
- ambiguity
and conflict
types of multi-cue integration:
- compromise (take the average, but take into account reliability of the source)
- dominance (trust one ignore the other)
- interaction (info incomplete or ambiguous, need one source to disambiguate the other)
Young et al (1993)
testing COMPROMISE
- Ps viewed computer-generated cylinder defined by 2 cues:
- texture
- motion
- motion: cylinder rocked back and forward meaning the texture had to move to give this perception
- Ps had to decide if the cylinder looked flattened, circular or stretched
- Found that they could use motion to simulate a v flattened shape or stretched shape by moving different parts of textures at different rates.
Follow up of Young et al 1993
- two cues put into conflict
- each defines a different shape:
- motion: stretched
- texture: flattened
- FOUND: Ps perceived shape as a compromise between the two cues.
Dominance:
- cue with valid assumptions will dominate
- e.g. Ames room illusion - invalid cue is dominating as perspective is the dominant cue versus relative size
Dominance:
- cue with valid assumptions will dominate
- e.g. Ames room illusion - invalid cue is dominating as perspective is the dominant cue versus relative size
