w7 gemini Flashcards
What is the main goal of stereo vision?
To infer depth from two or more images taken from different viewpoints.
What are the two main sub-problems in stereo vision?
Correspondence and Reconstruction.
What is the correspondence problem in stereo vision?
The challenge of finding matching points in the left and right images that correspond to the same 3D point in the scene.
What are some of the stereo constraints that can be used to solve the correspondence problem?
Epipolar constraint
maximum disparity constraint
continuity constraint
uniqueness constraint
ordering constraint.
Explain the epipolar constraint for coplanar cameras.
For coplanar cameras, corresponding points will lie on the same row (the ‘epipolar’ line) in both images.
What is the maximum disparity constraint?
The length of the search region for a corresponding point is limited by the maximum expected disparity.
Explain the continuity constraint in the context of stereo vision.
Neighbouring points in an image are likely to correspond to points at similar depths.
What is the uniqueness constraint in stereo vision?
A location in one image should ideally match only a single location in the other image.
What is the ordering constraint in stereo vision?
Matching points along corresponding epipolar lines should appear in the same order in both images.
What are the basic requirements for solving the correspondence problem?
Most scene points must be visible in both images, and corresponding image regions should appear similar.
Under what conditions do the basic requirements for correspondence tend to hold?
When the distance of the 3D point from the cameras is much larger than the baseline (z»_space; B).
What are the two main approaches to solving the stereo correspondence problem?
Correlation-based methods and Feature-based methods.
What is the output of correlation-based methods for correspondence?
Dense disparity maps.
What is the output of feature-based methods for correspondence?
Sparse disparity maps.
What are the characteristics of a simple case of stereo vision with coplanar cameras?
Image planes of cameras are coplanar, focal lengths are equal, optical centers are at the same height, and optical axes are parallel.
How is disparity calculated for coplanar cameras?
d = x’L - x’R
What is the relationship between depth and disparity?
Depth is inversely proportional to disparity.
What is a disparity map?
A field of disparity vectors for a pair of stereo images.
What is the significance of the baseline in a stereo vision system?
It is the distance between the optical centers of the two cameras.
What are the characteristics of a complex case of stereo vision with non-coplanar cameras?
Cameras are not aligned such that their image planes are coplanar.
What is measured instead of distance for disparity in non-coplanar cameras?
Angles.
What is the formula for disparity in non-coplanar cameras using angles?
Disparity = αL - αR
What is the horopter?
A curve in 3D space where objects project to corresponding points on the retina with zero disparity.
What are some other cues to depth besides stereo vision?
Binocular, Oculomotor, Monocular, and Motion cues.
What are the oculomotor cues to depth?
Accommodation and Convergence.
Explain accommodation as a depth cue.
The shape of the lens in the eye adjusts to focus on objects at different depths.
Explain convergence as a depth cue.
The angle of rotation of the eyes needed to fixate on an object at a certain depth.
What are some monocular cues to depth?
Interposition, Size familiarity, Texture gradients, Linear perspective, Aerial perspective, and Shading.
Explain interposition as a depth cue.
When one object blocks the view of another, the occluding object is perceived as being closer.
Explain size familiarity as a depth cue.
Knowing the typical size of an object helps in judging its distance.
Explain texture gradients as a depth cue.
For uniformly textured surfaces, the texture elements appear smaller and more closely spaced with increasing distance.
Explain linear perspective as a depth cue.
Parallel lines appear to converge as they recede into the distance.
Explain aerial perspective as a depth cue.
Distant objects appear fuzzier and have lower luminance contrast and color saturation.
Explain shading as a depth cue.
The distribution of light and shadow on objects provides information about their shape and depth.
What are some motion-induced cues to depth?
Motion parallax, Optic Flow, Accretion and deletion, and Structure from motion.
Explain motion parallax as a depth cue.
The apparent speed and direction of motion of objects differ based on their distance.
Explain optic flow as a depth cue.
As an observer moves, the pattern of apparent motion of objects provides information about relative depth.
Explain accretion and deletion as motion-induced depth cues.
As an observer moves, parts of a background object may become covered or uncovered by a foreground object.
Explain structure from motion (kinetic depth) as a depth cue.
The perception of a 3D structure can be derived from the changing 2D projections of the object.
What is epipolar geometry?
The geometry of the situation involving two cameras looking at the same scene.
What is an epipole?
The projection of the optical center of one camera onto the image plane of the other camera.
What are epipolar lines?
Lines in one image that correspond to a point in the other image.
What are conjugated epipolar lines?
The epipolar lines generated by the same 3D point on the left and right image planes.
What is the Epipolar Constraint in the context of general (non-coplanar) cameras?
Corresponding points must lie on conjugated epipolar lines.
What is rectification in the context of stereo vision?
A transformation applied to the stereo image pair to make the epipolar lines parallel to the image rows.
What knowledge is required for triangulation and rectification in stereo vision?
Knowledge of intrinsic and extrinsic camera parameters.
Why is recovery of depth information important in computer vision?
For controlling movement, 3D reconstruction, and object recognition.
What is the difference between coplanar and non-coplanar cameras in terms of epipolar lines after rectification?
After rectification, epipolar lines in both cases become horizontal.
What is a limitation of stereo vision related to the field of view?
A stereo pair of cameras can only find depth for locations within the common field of view.
How does the length of the baseline affect the common field of view and depth error in coplanar stereo vision?
A short baseline leads to a larger common field of view but a larger depth error.
What is the convergence angle in non-coplanar stereo vision?
The angle between the optical axes of the two cameras when they are fixated on a point.
How does the location of the horopter depend on the vergence angle?
The location of the horopter changes based on the convergence angle.
What is the significance of retinal disparity in human vision?
Some cortical neurons are tuned to retinal disparity.
What are ‘crossed’ and ‘uncrossed’ disparities in human vision?
Points nearer than the horopter create ‘crossed’ disparities, while points more distant create ‘uncrossed’ disparities.
What are some limitations of the uniqueness constraint?
It doesn’t hold when surfaces lie along a line-of-sight for one camera.
What are some limitations of the ordering constraint?
It doesn’t hold when objects have different depths, causing occlusions.
Why is correspondence fundamentally ambiguous?
Because for a given point in one image, there can be multiple potential matching points in the other image.
What are two reasons why some points in each image may have no corresponding points in the other image?
Due to occlusion or because the cameras might have different fields of view.