Features and matches on buildings

Rakefetmoor
21 ●1 ●1 ●3

Hi, I am using SURF with the default parameter to detect features in two images of urban area. Those images were taken with the same camera in different view angles. I plot the images with the key points and I see that the keypoints are on the ground as well as on the building tops. When I match the features between the two images with brute force matcher that picks the two near neighboars and use the ratio test of 0.7 threshold as well as the symmetry test, all the matches that left are on the ground. I loosed the constrain and I am getting more outliers but still all the matches points are on the ground. Do you have any suggestions why this is happening? I am interested to have matching points on the roof tops of the buildings and I am wondering if the parameter choice prompt the matches to be on a flat area. I'll appretiate any feedback.

Comments

Some example images might be helpful.

Ben (Jan 22 '13)edit

add a comment

answered Jan 24 '13

Geppertm
275 ●3 ●11

Surf is looking for eye-catching points. This occurs inter alia by the contrast between the umligenden pixels. I think your buildings have very few of these points.

Here are some Tips:

Use Grayscale Images (ore cast them to Grayscale)
Sharp them (Sharp Image = Original Image - Gausian(Original Image))

Maybe you can try a Sobel and add it to the sharpened Images. I think that would give you more matching Keypoints at the roof area and on the buildings.

You should also use a dynamic threshold for matching!. Like:

if((matches2[i][0].distance < 0.56*(matches2[i][1].distance)) && ((int) matches2[i].size()<=2 && (int) matches2[i].size()>0))

You could also try a KNN-Matcher instead of a BF-Matcher.

add a comment

Features and matches on buildings

Comments

1 answer

Links

Question Tools

Stats

Related questions

Features and matches on buildings edit savecancel

Comments

1 answer

Links

Question Tools

Stats

Related questions

Features and matches on buildings