Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
15 Cards in this Set
- Front
- Back
- 3rd side (hint)
|
Edge detection, what is good edge? |
Minimize false positive ; detection Close to true edge ; localization One point for edge ; single response |
|
|
|
How to get edge from gradient image? |
Smooth, Enhance : contrast Edge local: maxima, thinning |
|
|
|
Issue of edge detection |
Scale change edge ( smooth) Thresh sensitivity |
|
|
|
Canny edge detector steps |
Ori img Gradient image (rough edges) Thresh (for level of detail) NMS (for one pixel) Hysteris thresh (connect ) |
|
|
|
How hysteris thresholding works? |
Choose 2 thresh high n low Start pixel (strong=above high thresh) connect if next pixel above LOW thresh High/low = 2 |
|
|
|
Line fiting what is? |
Fit a line to texture of line in image |
|
|
|
Problem of line fitting |
Occlusion extra edges Parts missing Noise (detect param) |
|
|
|
Line fitting using Hough trafo idea |
Vote pixel value belong to line or not In hough space |
Hough trafo |
|
|
Trafo of hough space n image math! |
Img,hough Y=m*x+b Line,point(bo,mo) Point(xo,yo),line(b=-xo*m+yo). Line = multiple points |
|
|
|
Hough trafo explain |
Points along one line, get translated into lines in houghspace, crossed at one point. Point of crossing = line equation in img space
Use sin cos cuz slope can be infinity Line: d = x cos0 + y sin 0 |
One line? Point? What infinite? |
|
|
Effect of noise on Hough trafo Sol? |
-peak spread not clear voting bin count lower -multiple peaks on wrong place
Sol: Use gradient orientation (as teta) and magnitude (dtronger more vote) Change discretization( resol) Use shapes (circle square etc) |
Line / edges from image? |
|
|
Procon hough trafo |
Pro - Independent points occlusion egal - Robust to noise - One pass, multiple instance Con - Complexity exponent w/ #param - Spurious peaks non targt shape - Bin size hard to choose |
Pro Occ why? Noise? Onepass? Con Curse dimmensional for? Spurious Discretization |
|
|
Ransac what idea |
Random sample consensus Remove outlier by looking at support of points combination |
Outlier |
|
|
Ransac loop |
Select seed, Compute trafo Find inlier If sufficient, calc least square of trafo Keep traco w/ most #inliers |
|
|
|
Ransac decide # samples needed? |
Probability single sample corr w^n Probabilitit all k samples fail (1-w^n)^k Choose k such that all below desired rate |
|
