aledalgrande's profile - activity

PS: don't have access to ResearchGate, but by a quick look at the abstract I was wondering if it was something to spend my time on. You guys confirmed it's not.

Thanks for the comments. It seemed too easy to be truly usable in a real case...

The guys here detected the contour of people. They say they used OpenCV, but the link to the source code is broken.

I think it could be an intersection of HOG people detection and Canny. Do you have any other idea on how they could make this with OpenCV?

I downloaded the 3.0 alpha, but I had a lot of link errors (e.g. couldn't find cv::rectangle) and I also couldn't find how to use UPNP, which I can see on master in Github.

Is there any guide for the transition?

That is called the Kronecker product and it's not supported by OpenCV, so you have to write your own.

I have implemented a 2D->3D point triangulation method, which uses 2 poses P1 and P2 and 2 points back-projected to z=1 point1 and point2 to recover depth:

Matx14d A1 = point1(0) * P1.row(2) - P1.row(0),
        A2 = point1(1) * P1.row(2) - P1.row(1),
        A3 = point2(0) * P2.row(2) - P2.row(0),
        A4 = point2(1) * P2.row(2) - P2.row(1);
double normA1 = norm(A1), normA2 = norm(A2), normA3 = norm(A3), normA4 = norm(A4);
Matx44d A(A1(0)/normA1, A1(1)/normA1, A1(2)/normA1, A1(3)/normA1,
          A2(0)/normA2, A2(1)/normA2, A2(2)/normA2, A2(3)/normA2,
          A3(0)/normA3, A3(1)/normA3, A3(2)/normA3, A3(3)/normA3,
          A4(0)/normA4, A4(1)/normA4, A4(2)/normA4, A4(3)/normA4);

SVD svd(A, SVD::MODIFY_A);
Mat_<double> svdVT = svd.vt;
Mat smallest = svdVT.row(3);
double scale = smallest.at<double>(0,3);

if (scale == 0.0)
    scale = 0.00001;

Point3d point3D(smallest.at<double>(0,0)/scale, smallest.at<double>(0,1)/scale, smallest.at<double>(0,2)/scale);

This is based on the book Multiple View Geometry in Computer Vision. However, this doesn't seem to provide the right results visually and in fact, when I calculate the difference between the projected 3D point and the original image point coordinates, there is an error of >100px.

What is the problem? Even with synthetic values and transformations I can't get a good enough triangulation.

The error was in how I passed the value to position, which is a Point2f type. In this case it was a memory leak, but in XCode it would show as a Point2f variable with empty values for x and y, not as a null value. That's what was confusing.

Always check for a valid value, in this case when a Point2f variable is initialized, it is set to Point2f(0,0) by the constructor.

I solved it, had a stupid variable shadowing in an upper for loop that was passing a null reference for the position in this method, so it would "seem" valid to the compiler, but at runtime it would crash. No worries, everybody has a bad day, but keep contributing to the OpenCV community!

PS: I was using the hashtags because they come up first in the auto-complete

Added the beginning of the method as well.

Ah that is Objective C. It's just a check to see if position is inside the image.

Sometimes this code triggers a critical error. I tried to put every condition I could think of to prevent bad input, but it's still not working sometimes.

- (Vec3f)extractColorWithKeyframe:(Keyframe*)keyframe WithPosition:(Point2f)position
{
    Mat image = keyframe.originalImage;
    if (image.rows < 1 || image.cols < 1 || ![keyframe.camera isInFrame:position withBoundary:0] || image.channels() < 4)
        return Vec3f(0,0,0);

    Vec4b bgra;
    bgra = image.at<Vec4b>(position.x, position.y);

Any clue?

You cannot get the physical sensor size, you have to look it up on the manufacturer's website.

I would love to use some functionalities of 3.0.0 on iOS, but it's not out yet, so is there any compiled version of the iOS framework I can use? Last time I tried to build it I threw away days of work, unsuccessfully.

Thanks

You also have to use OpenCV >= 3.0.0

You can actually deblur if it comes from video, search for "blind deconvolution". There are some Matlab implementations around.

I'm trying to write some code to estimate the fundamental matrix:

img1 = cv2.imread('glassL.jpg')
img2 = cv2.imread('glassR.jpg')
K = calibration['cameraMatrix']
F = cv2.findFundamentalMat(img1, img2, cv2.FM_RANSAC, 0.1, 0.99)

Getting the error:

OpenCV Error: Assertion failed (npoints >= 0 && points2.checkVector(2) == npoints && points1.type() == points2.type()) in findFundamentalMat, file /tmp/opencv-XbIS/opencv-2.4.8.2/modules/calib3d/src/fundam.cpp, line 1103
Traceback (most recent call last):
  File "sfm.py", line 27, in <module>
    F = cv2.findFundamentalMat(img1, img2, cv2.FM_RANSAC, 0.1, 0.99)
cv2.error: /tmp/opencv-XbIS/opencv-2.4.8.2/modules/calib3d/src/fundam.cpp:1103: error: (-215) npoints >= 0 && points2.checkVector(2) == npoints && points1.type() == points2.type() in function findFundamentalMat

Any clue?

If I have cameraMatrix and distCoeffs from plain calibration (1 camera) how can I use stereoRectify to prepare some images for reprojectImageTo3D? It seems rotation and translation matrices are needed, but I don't have those in my calibration output...

Whenever I try to run OpenCV code in Ruby bindings I get this error:

/usr/local/var/rbenv/versions/2.1.0/lib/ruby/2.1.0/rubygems/core_ext/kernel_require.rb:55:in `require': dlopen(/usr/local/var/rbenv/versions/2.1.0/lib/ruby/gems/2.1.0/extensions/x86_64-darwin-13/2.1.0-static/ruby-opencv-0.0.12/opencv.bundle, 9): Library not loaded: /usr/local/opt/opencv/lib/libopencv_nonfree.2.4.8.dylib (LoadError)
  Referenced from: /usr/local/var/rbenv/versions/2.1.0/lib/ruby/gems/2.1.0/extensions/x86_64-darwin-13/2.1.0-static/ruby-opencv-0.0.12/opencv.bundle
  Reason: image not found - /usr/local/var/rbenv/versions/2.1.0/lib/ruby/gems/2.1.0/extensions/x86_64-darwin-13/2.1.0-static/ruby-opencv-0.0.12/opencv.bundle

I installed OpenCV today from Homebrew (2.4.8.2), and the gem is ruby-opencv "0.0.12".

Do you have any clue? My C++ skills are very rusty ;)

I'm looking into stereo matching to build a point cloud from two photos.

Will it work correctly if the images come from the same camera, that has just been moved/rotated? Will I pass the same calibration parameters to stereoRectify (will I need to use that function)?

Thanks