tenta4's profile - activity

It's seems you beginner in programming, if you can't to implement a loop like for x {for y}

cv::Mat img(400, 400, CV_8UC3);

// fill by white

img = cv::Scalar(255, 255, 255);

// change the pixel

int x = 50; int y = 100;

img.data[(y * img.size().width + x) * 3] = 0; // r

img.data[(y * img.size().width + x) * 3+1] = 0; //g

img.data[(y * img.size().width + x) * 3+2] = 0; //b

// write

cv::imwrite("myimg.png", img);

// open

cv::Mat opened = cv::imread("myimg.png");

cv::imshow("opened", opened);

cv::waitKey(0);

distance - is not the distance between 2d points. It is Euclidean distance between descriptors of two points.

solvePnP returns book's translation and rotation in camera coordinate system. You can use it for model rendering. Look here to know how

http://answers.opencv.org/question/23089/opencv-opengl-proper-camera-pose-using-solvepnp/?answer=23123#post-id-23123

Hi. I working for augmented reality application. For pattern recognition i use next pipe:

• orb feature detecting
• orb descriptor extracting
• feature matching with Bruteforce matcher
• pose estimation (SolvePNPRansac)

But it's pipeline works fine only for small distances and not "hard" angles.

Is there the better approach for pose estimation?

Is there additional capabilities for pattern recognition, when it flat?

yes, it's normal. When set flag withDerivatives, output vector size = purCount * 2

Is this normal, that when i build pyramid with level = 4, function creates vector <mat> with size = 8?

It's sample code.

cv::buildOpticalFlowPyramid(queryImage, mQueryData->mPyr, cv::Size(15, 15), 3);

Here size of each mat

D/CMarkerlessRecognitionStrategy(32362): size: 432 240

D/CMarkerlessRecognitionStrategy(32362): size: 432 240

D/CMarkerlessRecognitionStrategy(32362): size: 216 120

D/CMarkerlessRecognitionStrategy(32362): size: 216 120

D/CMarkerlessRecognitionStrategy(32362): size: 108 60

D/CMarkerlessRecognitionStrategy(32362): size: 108 60

D/CMarkerlessRecognitionStrategy(32362): size: 54 30

D/CMarkerlessRecognitionStrategy(32362): size: 54 30

Hi, i use solvePnPRansac function in in Android application in native. With identical input parameters it returns different results (rvec, tvec and status)

Can this appears or i have memory corruptions?

maybe you have not enough memory?

Hi. Is there any reason for use OpenCV for Android instead of native library via JNI?

you also can apply matrix P to point F using perspectiveTransform() function

Try to convert images to HSV color model and subtract only H channel

Hi. I use calcOpticalFlowPyrLK function for pattern tracking for my Augmented Reality application.

It's very fast and stable, but i found, that Pyramid Optical Flow is not rotation invariant.

Is there alternative methods with the same speed? I tried calcOpticalFlowFarneback, but it is not fast enough

I solve this problem. It's cause i used OpenCV 2.4.0. This feature is available in version >= 2.4.5

I try to use built pyramids in the calcOpticalFlowPyrLK function, but it wrote "Assertion failed (0 <= i && i < (int)v.size())"

Why it doesn't want? Documentation says prevImg – first 8-bit input image or pyramid constructed by buildOpticalFlowPyramid().

I use cvFindExtrinsicCameraParams2 function for determinate rotation matrix and translation vector.
I have a strange situation.
When my pattern close to camera, everything is OK, but when it so far from camera, i have (look at picture) trouble.
It appears when i look at the pattern from angle 90* between optical axis and marker's plane

My question is why this appears and how can i solve it?

If you want to add opencv libraruies in your Qt project, you must do next

• add pathes to opencv in pro file
  for example
  INCLUDEPATH += C:/OpenCV/mingw/install/include
  LIBS += C:/OpenCV/mingw/install/lib/*.a

• make this http://answers.opencv.org/question/18081/opencv_core246ddll-is-not-in-your-computer/#18095

Your
Mat obj_corners = new Mat(4,1,CvType.CV_32FC1);
Mat scene_corners = new Mat(4,1,CvType.CV_32FC1);

must be CvType.CV_32FC2

Its a points 2d, that's why they have 2 (x,y) coordinates.

Thank you, i know this, but quaternion components also interdependent. I found this article (http://www.malcolmdshuster.com/pub_1993f_c_qnorm_mds.pdf), maybe it will be helpful for me but i hoped to find the easyest solution

I want to use Kalman filter for rotation. I have rotation matrix, but i can't use filtration for it, because all it's elements interdependent. That's why i try to convert matrix to euler angles and to filter each of compoments.

My questions:

• is this filtering of euler angles will be right? (is all components x, y, z independent? I think no) If yes, how to avoid gimbal lock during filtration?

• is there a better way to filter the rotation?

Take your opencv_core246d.dll and copy it to folder, where lies your executed file. For example, i take opencv_core246d.dll here C:\OpenCV\opencv\build\x64\vc10\bin and copy in VS project debug folder C:\Users\pc\Documents\Visual Studio 2010\Projects\opencv_test\Debug

It's easiest way

You have 2 variants:
1) Copy all opencv_*.dll to folder with your .exe file (i do so).
2) Add path [opencv]/bin in you PATH in environment variables.

The ansver is simple: Rotation vector - is not an Euler Angles.

see my answer here http://answers.opencv.org/question/17723/android-application-to-recognize-known-object-and/#17879

You should check if the keypoints were found
if (keypoints.size() == 0 || keypointsRef.size() == 0) return mRgba;

Your matchesList must be checked also

Are you sure that the images are opened?

I want to use simple artoolkit marker to calibrate my device. Whan i found marker, i have 4 image points and 4 object points (corners of a square). I use calibrateCamera function for my 4 points to try to get intrinsic parameters. But received focals have too big amplitude (if my real fx = 960, it can calculates as 500 and 3000)

Okay, maybe marker too small (hence the large error of calculation) or four points is too few.

I try to use CV_CALIB_USE_INTRINSIC_GUESS flag, and give the initial approximation of intrinsic parameters. But output matrix is absolutely dependent of initial approximation. It means, that if i give it fx = 500, it's returns fx = 500 +/- 20, if i give it 2000, it returns 2000 +/- 20 Here my questions

1. How i can improve my calibration results.
2. Why CV_CALIB_USE_INTRINSIC_GUESS flag set so large dependence from initial value of intrinsic parameters?

You can call mat.data attribute to get uchar* from cv::Mat

I don't know the best way, but maybe this source will be helpful for you https://code.ros.org/trac/opencv/changeset/564/trunk/opencv/src/highgui/utils.cpp if you will not find the better decision