The Third Dimension PDF

Seeing in Depth 3 categories of visual cues give us depth information: 1. oculomotor cues: moving our eyes and focus 2. monocular cues: cues about the image of one eye 3. binocular cues: cues requiring both eyes and disparity Depth cues Cues can also be divided into whether they give us absol...

Seeing in Depth 3 categories of visual cues give us depth information: 1. oculomotor cues: moving our eyes and focus 2. monocular cues: cues about the image of one eye 3. binocular cues: cues requiring both eyes and disparity Depth cues Cues can also be divided into whether they give us absolute or relative information about depth metrical cues provide quantitative information about distance in depth non-metrical cues only provide information about depth order (relative depth) OCULOMOTOR CUES Oculomotor cues: cues provided by how we control our eyes when we look at an object accommodation is how much we've contracted our ciliary muscles to focus on a visual object More contraction, thick lens Less contraction, thin lens vergence is how we've positioned our eyes to look at a part of the scene convergence is where we turn our eyes inward, to look at near objects divergence is where we turn our eyes outward, to look at far objects Monocular depth Monocular depth cues Position-based cues Partial occlusion: if one object occludes another, the occluding object must be in front provides relative (non-metrical) depth information Relative height: the vertical position (height) of an object within the field of view relative to eye level objects further away from eye level appear closer, at eye level further away Relative-Metrical: can be used to infer relative depth (ie object A is twice as far as object B); does not provide absolute depth information Size-based cues Relative size: a comparison of the size of objects, without knowing the true size of either for alike objects, smaller means farther Relative-Metrical cue Texture gradient: repeating patterns of relative size cues Familiar size: if we know what size the object is, we know what size it should be on the retina at different depths absolute metrical cue Linear perspective: parallel lines that get farther away appear to converge Relative cue Lighting-based cues Atmospheric perspective: the farther an object, the more light is scattered by the air between the object and observer  as a result, distant objects appear less distinct and bluer than near objects Relative cue Shading: light falling on a smooth surface leads to differences in shading, giving cues about depth in the absence of explicit information about the light's location, we assume it is from above Dynamic Cues Motion parallax: a dynamic depth cue caused by motion of the observer as we move, objects parallel to our direction of movement will move at different rates across our visual field depending on how far away they are Further objects move slower than objects closer to us this relative motion allows us to infer the distances of the objects Optic flow: a dynamic depth cue caused by motion of the observer as we move forwards, objects expand from a central point in the direction of motion Deletion and accretion: changes in occlusion over time As an object moves behind another, they become occluded, as they come back out, they are accreded BINOCULAR DEPTH Binocular disparity: the difference between the 2 viewpoints from each eye provide us with a very strong cue about the depth of an object, at least for near to moderately distant objects stereopsis: our sense of depth that arises from binocular disparity Because of the convergence of our eyes to a fixation point, an object at that point will appear at the fovea of each retina Horopter: imaginary curve or surface in space where objects appear at the same distance from both eyes, meaning that light from those objects falls on corresponding points in both retinas Zero disparity: occurs when an object is located on the horopter. Since it projects to corresponding points on both retinas, the brain interprets it as being at the same depth as the fixation point Other points will have a certain disparity (crossed/uncrossed) depending on their distance from the horopter NEAR OBJECTS crossed disparity: Objects that are closer to the observer than the horopter project to non-corresponding points on the retinas, but they are projected closer to the nose (crossed). This makes the object appear closer than the point of fixation their image will be displaced to the left in the right eye, and to the right in the left eye FAR OBJECTS uncrossed disparity: Objects that are farther away than the horopter also project to non-corresponding points, but they are projected away from the nose (uncrossed). These objects appear farther than the fixation point. their image will be displaced to the right in the right eye, and to the left in the left eye Binocular disparity: changing the point of fixation will change the horopter, and whether different objects appear with crossed or uncrossed disparity magnitude of the disparity changes increases with greater distance from the horopter Recreating binocular depth cues Cyclopean stimuli are stimuli are entirely defined by binocular disparity random dot stereograms are cyclopean stimuli created in random patterns of dots and shifting part of the random pattern for the left and right images free fusion: requires the uncoupling of vergence and focus Correspondence problem Correspondence problem: the challenge the brain faces in matching visual information from the two eyes to create a unified perception of depth and 3D structure each eye receives a slightly different image due to their different viewpoints The brain must figure out which points in the left eye's image correspond to the same points in the right eye's image Neurophysiological basis for a neuron to code for binocular disparity, it must receive input from both eyes the earliest place this can occur is in V1 binocular cells: tuned to zero disparity (the horopter) while others respond best when similar images occupy slightly different positions on the retina found in V1 and at later stages of both the dorsal and ventral pathways Requires the object to be at a particular depth in order for cells to fire COMBINING DEPTH CUES we use the available cues in combination with our knowledge of how things work, ie what is the most likely interpretation of the available information, given the physical constraints of the world The Bayesian approach P(Sc | I ) = P(Sc) x P (I | Sc) P(Sc | I) = the probability of the Scene given the observed input what we're trying to infer) P(Sc) = the probability of the scene occurring at all, based on our knowledge of how the world works, and our experience P(I | Sc) = the probability of the Image being produced by a particular Scene Example: If a coin partly occludes another, it is more probably that it is 2 coins with one occlude (different depth) vs 2 broken coins vs one coin larger than the other Depth Cues We use depth cues to interpret a visual scene depth can affect our perception of an object's size AMES ROOM All things being equal: smaller images = smaller things Making size estimates based off of knowledge and assumptions