stereo Flashcards
(27 cards)
What is stereo vision?
A technique for estimating 3D structure by comparing two images taken from different viewpoints.
What is disparity in stereo vision?
The horizontal pixel shift between corresponding points in the left and right images.
What does disparity allow us to compute?
Depth, using the formula Z = (focal length × baseline) / disparity.
What is sparse stereo matching?
Matching a small number of feature points to compute a sparse 3D reconstruction.
What is dense stereo matching?
Computing disparities for nearly every pixel to generate a full depth map.
Why is image rectification important for stereo?
It aligns image rows so that corresponding points lie on the same horizontal line, simplifying matching.
What is the goal of window-based matching?
To find the best matching pixel in the right image for each pixel in the left image by comparing patches.
What is the most common matching cost in stereo vision?
Sum of Absolute Differences (SAD).
What is the SAD formula in stereo?
SAD = ∑ |I_L(x,y) - I_R(x-d,y)| over a local window.
What are challenges for SAD matching?
Lighting variation, occlusion, textureless regions, and repetitive patterns.
What is scanline stereo?
An approach that aligns entire rows (scanlines) using dynamic programming to handle occlusions and improve consistency.
What is dynamic programming used for in stereo?
To find an optimal sequence of pixel matches across a row, treating the problem like path alignment.
What does a diagonal move in the scanline DP grid represent?
A pixel match between the left and right images.
What do horizontal or vertical moves in the DP grid represent?
Occlusions or unmatched pixels.
What is a major limitation of scanline stereo?
It processes each row independently, leading to horizontal streaking artifacts and lack of 2D smoothness.
What is Semi-Global Matching (SGM)?
A stereo matching method that aggregates matching costs over multiple directions to approximate global optimization.
Who developed SGM?
Heiko Hirschmüller.
What is the key idea behind SGM?
To balance data fidelity and smoothness by combining local matching costs along multiple paths through the image.
How does SGM handle occlusions and noise?
By penalizing disparity changes between neighboring pixels, promoting consistency.
What are typical window sizes used in SGM?
Small windows, such as 3×3 or 5×5, to preserve detail while maintaining robustness.
What kind of output does SGM produce?
A dense disparity map that can be converted into a depth map or point cloud.
What are the main advantages of SGM?
Accurate, robust to noise and occlusion, and efficient enough for real-time or near real-time use.
What is the main input assumption for dense stereo algorithms?
That images are rectified so that epipolar lines are horizontal.
What is the difference between sparse and dense stereo?
Sparse stereo matches a few keypoints; dense stereo attempts to match most or all pixels.
