Dear All
Hi, I am trying to get an object pose from 3D-2D point correspondences using
solvePnPRansac , but i don't know exactly how i should use this function in c++.
is there any example code that could help me do my work. I couldn't find any
useful code on the net.
thank you very much.
Best Regards
Navid

Hello,

   I have Delphi 7 with the interface to OpenCV libaries.
(I have the newest libraries 241 version)

Here is the code I wrote:

procedure Test(var img: PIplImage);
var
   imgDistorted: PIplImage;
begin
   //With: imgDisorted.. AV is raised on cvSaveImage, without is not ...
   imgDistorted := cvCreateImage(cvSize(img^.width and -2, img^.height and -2),
8, 1);

   cvSetImageROI(img, cvRect(0, 0, img^.width and -2, img^.height and -2));
   cvSaveImage(PChar('testOut.jpg'), img);//AV is raised here ...
end;

procedure TForm1.btn1Click(Sender: TObject);
Var
   img: PIplImage;
begin
   img := nil;
   try
     img := cvLoadImage(PChar('test.jpg'), 0);
     if not Assigned(img) then
       raise Exception.Create('B³±d podczas otwierania pliku skanu.');
     Test(img);
   finally
     if Assigned(img) then
       cvReleaseImage(img);
   end;
end;

On cvSaveImage line I am getting AV. Does anyone of you have an idea why there
is such problem?

Thanks

Hi!
I am using as pattern images taken from the webcam of some textured planar
object (more than 100x100 pixels).
I tried a sort of calibration of the tracking system doing this:
  - I put the camera on a know position (for example 500 mm from the template)
  - Let the program compute the distance
  - Read the data and get the measured error (E = Zcomputed - 500 )
After doing this for many times (actually i do this for 10 times)
I get a mean-error and I correct all the future data with an
empirical-coefficient.
It works but I think it's not the right method.
Some suggestion?
Where can I find some papers on this topic?
I know that pose estimation from feature-based tracking system are
not a trivial subject, especially if with a monocular camera...




--- In OpenCV@yahoogroups.com, Bence Magyar <mw.mzperx@...> wrote:
>
> Hello,
> what you have is called the detection error. It affects the Z axis the most
> since that's the axis which you are trying to guess without any direct
> sensor information of depth. Most monocular 3D pose estimation techniques
> suffer from this problem. There are plenty of possible workarounds but it's
> hard to get it right.
> What you should try in the first place:
>
>    - Don't use too small patterns. Usually 5x5 or 6x6 is robust for most
>    cameras and situations.
>    - Try to calibrate your camera more precisely.
>
> These steps should improve your results a bit, but not guaranteed that it's
> going to be flawless.
>
> --Bence
>
> 2012/7/6 alelr7 <alessandrolarosa7@...>
>
> > **
> >
> >
> > Hi everybody!
> >
> > I am working on Planar Object detection and Tracking, using SURF for a
> > Real-time application.
> >
> > I compute the camera pose from a planar pattern image, intrinsic
> > parameters and homography.
> >
> > I found some errors in the coordinates.
> > For example, the Z-axis coordinate is always bigger than the real one.
> > It's ratio (z-computed/z-measured) changes as the marker changes.
> >
> > What am I missing?
> >
> >
> >
>

Hi,


I am try to run the opencvforandroid tutorial-2-opencvcamera. It seems that I
can open the VideoCapture object, but the grab() method hangs.



I first get the VideoCapture object as below:

private VideoCapture mCamera;

mCamera = new VideoCapture(Highgui.CV_CAP_ANDROID);

Next I set the frame width and height. This is shortened code after I find an
acceptable widht and height.

  mCamera.set(Highgui.CV_CAP_PROP_FRAME_WIDTH, mFrameWidth);
   mCamera.set(Highgui.CV_CAP_PROP_FRAME_HEIGHT, mFrameHeight);


  Next I open the camera and start a thread to grab frames.

  if(mCamera.isOpened())
  {

                 (new Thread(this)).start();
   }



   In the thread, I try to grab a frame, but the code never goes past the grab
statement

  if (!mCamera.grab())

   {
            Log.i(TAG, "mCamera.grab() failed");
                      break;
    }



I am running on a atrix II. Maybe the Atrix II does not supprt the VideoCapture
class???

Thanks,

Hello everyone! I have a project and want OpenCV to detect the heartbeat of the
user. I made an eye detected by "haarcascade_mcs_eyepair_big.xml." This function
detects the two eyes together. That done, divide the image into two parts, one
with the right eye and another with the left. I performed a filter to move to
grayscale and I'm trying to perform a binarization image to detect when the eyes
are open or closed. I'm trying to detect the white of the eye through the
binarization but even with the eye closed binarization is shown in a white part.
I believe this has to do with the threshold used. I'm trying this way because I
have researched is the most reliable way to do this, but I'm not succeeding.

Does anyone have any suggestions on how I can do to detect if the eye is open or
closed accurately?

Thanks for any help.

Hi all.
Does somebody know, how i can get extrinsic parameters using the 3d points
corresponds.

I found the following function: FindExtrinsicCameraParams2, but it is finds
extrinsic parameters between 3d object and its projection points(2d points).

Thanks!
I will try it.




--- In OpenCV@yahoogroups.com, "Sanjiv K. Bhatia" <sanjiv@...> wrote:
>
> To add to Bence's comment, OpenCV documentation talks about normalizing the
> results from match (using normalize).  The minMaxLoc function can be used
> to determine the maximum (or minimum) value which will give you the extent
> of match based on the last parameter.  Then, you can see the one that gave
> you the maximum match.
>
> Sanjiv
>
> On Fri, Jul 6, 2012 at 1:01 PM, Bence Magyar <mw.mzperx@...> wrote:
>
> >
> >
> > Hello,
> >
> > the problem is that "knowing which templates were matched" depends on
> > _your_ definition of successful template matching. As you can see by the
> > last parameter of cv::matchTemplate() it doesn't do anything else but runs
> > a cross-correlation operator (a normalized one in this case). To get the
> > result of the matching you have to search for the maximum or minimum value
> > in the resulting image (depending on the method you used in the last
> > parameter). To define a template matched you could set a threshold for the
> > maximum/minimum value in the result image.
> > After you have this, you can distinguish which templates are matching or
> > not matching.
> >
> > Have fun!
> >
> > --Bence
> >
> > 2012/7/5 paulal7489 <paulalopatniuk@...>
> >
> >> **
> >>
> >>
> >> Hello everybody,
> >>
> >> I am using matchTemplate with a high number of templates, but just two
> >> templates are always shown.
> >> So my idea was: Aplly the function matchTemplate for all templates and
> >> afterwords use some conditions (loops!?) to get an output which templates
> >> were found.
> >>
> >> Something like that:
> >>
> >> matchTemplate(src, templ6, result6, CV_TM_CCORR_NORMED);
> >>
> >> if( ... templ6)
> >> {
> >> fprintf( stderr, "Value = 6" );
> >> }
> >>
> >> I need to know, which templates was matched, because the templates are
> >> different values. When I know which one matched, I know the value, that is
> >> shown.
> >>
> >> Thanks a lot for help!!!
> >>
> >>
> >
> >
> >
>
>
>
>
> --
> Sanjiv K. Bhatia
> http://member.acm.org/~sanjiv
>

Hello Guys! Sorry about my dumb question, but I tried to find in internet
solution for my problem, but I have no sucess. I install the OpenCV 2.4.2 and
when I will type my program thats ok. The problem is when I will compile, the
windows show me the follow error:

"The application was unable to start correctly (0x0000013). Click OK to close
the application."

I tried everthing, I dont know what to do.

Somebody can I help me?

Tks in advance.

Hi Shervin! Thanks for your help. If I remove the highgui how can I do things
like this?

IplImage img = opencv_highgui.cvLoadImage(...);




--- In OpenCV@yahoogroups.com, Shervin Emami <shervin.emami@...> wrote:
>
> It sounds like you are using JavaCV so you should skip highgui anyway, and
> just try with the other modules. Use Java for all your GUI needs.
>
> Cheers,
> Shervin Emami.
> http://www.shervinemami.info/openCV.html
>
>
> On Thu, Jul 5, 2012 at 7:52 AM, Gabriela B <gabybosetti@...> wrote:
>
> > **
> >
> >
> > Hi! I'm trying to develop an application (JavaCV - JAR) that runs without
> > the user has previously installed OpenCV. I did it attaching all the .so of
> > OpenCV in the JAR, unpacking in runtime and then loading it with
> > System.load(). It works fine.
> >
> > Now, I can run the application if the user has installed all OpenCV
> > dependencies, but if not the app crashes. So my new challenge is attaching
> > these dependencies.
> >
> > For example, highgui needs libavcodec-dev but if I attach libavcodec.so It
> > can't be loaded, it throws an UnsatisfiedLinkError with no information
> > about the error. So, what I have to load before??
> >
> > Thanks for your help.
> > Gaby
> >
> >
> >
>

Hi! Thanks for your help! I don understand so much wath are you talking about.
Could you give me an example that swows how to do this?


--- In OpenCV@yahoogroups.com, "Rochester Beckham" <beckham.rochester@...>
wrote:
>
> --- In OpenCV@yahoogroups.com, "Gabriela B" <gabybosetti@> wrote:
> > For example, highgui needs libavcodec-dev but if I attach libavcodec.so It
can't be loaded, it throws an UnsatisfiedLinkError with no information about the
error. So, what I have to load before??
> > Gaby
> Unfortunately you need to have the libraries relinked if they are shared
objects.
>  if you have the static .a files you can always get binutils for the object
format; coff,elf and there opcode set for there architecture. if you were to use
an emulator or host translator you could compile in any format you like, but you
need to relink with -Bsymbolic or -Bdynamic on linux,unix,mac; problem is is
that the stack size grows by the api function object set. good luck.
>

If you think it is a bug, then it is better to create a ticket:
http://code.opencv.org/projects/opencv/issues

If it is not a bug, then the developers will close it anyway.

I have another question:
Is it possible to say, that the rectangle should be drawn, just if the template
was matched?

Cheers






--- In OpenCV@yahoogroups.com, "paulal7489" <paulalopatniuk@...> wrote:
>
> Hello everybody,
>
> I am using matchTemplate with a high number of templates, but just two
templates are always shown.
> So my idea was: Aplly the function matchTemplate for all templates and
afterwords use some conditions (loops!?) to get an output which templates were
found.
>
> Something like that:
>
> matchTemplate(src, templ6, result6, CV_TM_CCORR_NORMED);
>
> if( ... templ6)
>  {
> 	 fprintf( stderr, "Value = 6" );
>  }
>
> I need to know, which templates was matched, because the templates are
different values. When I know which one matched, I know the value, that is
shown.
>
> Thanks a lot for help!!!
>

Hi all,

I have developed a testing application for my company (a provider of Internet TV
systems). It works by blasting pre-defined IR commands at a set top box and then
taking a screen grab. Currently,the only way I have of verifying success or
failure of the test is by examining the screen grab myself and clicking on a
PASS or FAIL button.
I want to automate the verification step by comparing the screen grab against a
previous screen grab that I know is correct.
The screen grabs will typically be of a random TV program with graphics
overlaid. It will be the graphics that I am testing but the fact that the
background will be totally different in each pair of images puts the task way
beyond my knowledge and experience.
OpenCV seems to me to be a viable solution.
Can anybody offer any constructive advice/help?
I have no intention of plaguing any contributors with constant questions, my
company is willing to pay for serious professional help.
The test application runs on a MAC and uses the Cocoa framework so solutions can
be written in C. I am an experienced C Developer who knows very little about
OpenCV

I'm using OpenCV with python bindings , few days ago i was opening my webcam and
it works like charm , after that I hooked into a tutorial that using cv2 module
instead of cv , so I had to build the code and follow the tutorial , after that
my webcam is unable to open and results blank frames . what should I do ?

Hi! I am working on a same project:
Markerless tracking with a monocular camera for tracking objects.

I use Planar textured object and SURF for detection.
Then I calibrate th system computing the camera pose from a known distance from
the target and using this ratio to correct the pose:

computed distance/measured distance.

Is it correct? Are there better solution?



--- In OpenCV@yahoogroups.com, Eder Souza <ederwander@...> wrote:
>
> I belive that the easy way is:
>
>
> use :
>
>
> distancex = point2.x - point1.x;
> distancey = point2.y - point1.y;
> cvSqrt( (float)(distancex*distancex + distancey*distancey));
>
>
> Before applying this formula you can use scale adjust to compensate the
> distance from the camera ...
>
> Eder de Souza
>
>
>
> On Thu, Apr 14, 2011 at 2:16 AM, Behzad Lotfi <behzad.lotfi@...>wrote:
>
> >
> >
> > Hi
> > i dont thing about
> > we want use the DPI information for object in the frame and map it on real
> > object then we want use the sin formula calculate the distance in 2D
> >
> >
> > ------------------------------
> > *From:* Shervin Emami <shervin.emami@...>
> > *To:* OpenCV@yahoogroups.com
> > *Cc:* amit.benade <amit.benade@...>
> > *Sent:* Thu, April 14, 2011 7:39:39 AM
> > *Subject:* Re: [OpenCV] calculation of distance of an object from camera
> > using opencv
> >
> >
> >
> > Hi,
> >
> > You cant just call a function and get the distance & direction of an object
> > from a camera image because this is impossible to do without more
> > information, for example if you saw a photo of a red square on a white
> > background then you don't know if the square is 1 inch wide or 100
> > kilometers wide, because there is not enough information. You can use things
> > like stereo cameras and depth cameras to help, or if you already know the
> > exact size & shape of the object you are tracking then it can be used with a
> > single camera, but it is not always easy. One method is with POSIT (that
> > OpenCV has), but this might not be easy to use for robotics since it gives
> > you a rotation & translation matrix. But another method is the one I used
> > for my Masters Thesis, which you can find at "
> > http://www.shervinemami.info/" or "http://www.shervinemami.co.cc/"
> >
> > Cheers,
> > Shervin Emami.
> >
> >
> > On Mon, Apr 11, 2011 at 12:10 AM, amit.benade <amit.benade@...>wrote:
> >
> >>
> >>
> >> i'm new to opencv i have to do a project for robotics in which there is a
> >> need for distance measurement and direction detection of motion of a object
> >> so which function should i use to find it ..............pls suggest me how
> >> to do it
> >>
> >>
> >
> >
> >
> >
>

Please help guys.
Where do you think i start working?

and if you have better thesis titles please advise me.

Hi,

I'm currently finishing my master course in computer vision and I work exactly
with traffic analysis using OpenCV.

You will need to study about:
* Background subtraction algorithms
* Optical flow algorithms
* Blobs tracking

Recently, I developed a free opensource library with more than 20 background
subtraction algorithms, see:
https://code.google.com/p/bgslibrary/

For Optical Flow tracking, you can use the opencv Lucas-Kanade default
implementation.

For Blob Tracking you can use cvBlob library available at:
http://code.google.com/p/cvblob/

After all, one of the simplest ways to count vehicles is to make a ROI (region
of interest) and trace a simple line. For each detected blob with cvblob you
check if his centroid is left or right of the line and store this state. When a
blob change his state (i.e. left to right) is because it has crossed the line
(counter = counter + 1)...

Any questions, please contact me...

--- In OpenCV@yahoogroups.com, "Narayan" <petermax4@...> wrote:
>
> Hi everybody, I'm completely new to OpenCV library. However, I've gone through
plenty of books and articles written about this library. I'm doing a project
about detecting the number of vehicles i.e. vehicle traffic using image
processing or should I say using OpenCV but so far I can't get anywhere. All I
want is to determine the number of vehicles from the given video frame or for
now an image. Can anybody suggest me how or from where should I start doing what
I want?
> I really need to complete this project within 1 month or so as it affects my
semester marks.
>
>
> P.S.I'm familiar enough with C and C++ programming language but have no idea
about other languages so do suggest me directions on the basis of these
languages...!
>

I've been working on my video processor as a multithreaded
application. One thread decodes video frames, another analyzes them,
another compresses the results and finally the last thread displays a
number of the frames as they're processed so you can keep an eye on
how the program is doing.

The docs for VideoCapture::Read say:

OpenCV 1.x functions cvRetrieveFrame and cv.RetrieveFrame return image
stored inside the video capturing structure. It is not allowed to
modify or release the image! You can copy the frame using
cvCloneImage() and then do whatever you want with the copy.

To me this implies that with OpenCV 2.4 you can deal with the returned
frames directly. A lot of weirdness in my program seems to demonstrate
otherwise.

I've always found the Mat object's sharing of underlying data a bit
confusing, but I think it's somehow biting me. The video decoding
thread has been re-written a number of times to try and separate the
read data so it is only read once and not copied around 100 times.
From there it goes into a queue and another thread picks it up for
processing.

			 Mat *frame = new Mat();
		   (*video) >> (*frame);
			 if (!frame->empty()) {
				 // create the video frame and give up ownership of the Mat object
(frame data)
				 TVideoFrame *image=new TVideoFrame(frame);
				 paramAccess.Lock();
				 image->setFrameNum(fileFramesProcessed++);
				 image->setBatchFrameNum(totalFramesProcessed++);
				 image->setFilename(currentFile);
				 paramAccess.Unlock();

         frameAccess.Lock();
				 frameBuffer.push(image);
				 frameAccess.Unlock();
			 }

Something is wrong with that. I have a feeling that the Mat object
doesn't seem to fully deal with when to release the shared data (copy
on write if shared anyone???). Any ideas on how my frame object can
take ownership of the underlying image without having to do an extra
clone operation? It would be nice for the VideoCapture object to do
its decoding into a Mat and then on next call decode the next frame
into a different Mat object.

-Michael

Thanks for the hint, you are right. But I tested my sample code again with the
new version 2.4.2 and it seems to work now. I'll do some more testing at first!

--- In OpenCV@yahoogroups.com, "r1w3n" <r1w3n@...> wrote:
>
>
>
> If you think it is a bug, then it is better to create a ticket:
> http://code.opencv.org/projects/opencv/issues
>
> If it is not a bug, then the developers will close it anyway.
>

I'm using OpenCV's stitching module to stitch together a hundred or so 4288x2848
images. I'm stitching a few images a time; I plan on stitching these smaller
stitches into one final panorama.

The stitching module works perfectly most of the time. However, I get an
exception when I accidentally include duplicates, or images that have no
features in common. No big deal, that's my fault. But sometimes, after the
stitching module begins to composite the images onto a panorama, I get this
error in the command line:

OpenCV Error: Assertion failed (y == 0 || data && dims >= 1 && (unsigned)y <
(unsigned)size.p[0])) in unknown function, file
C:\slave\WinInstallerMegaPack\src\opencv\modules\core\include\opencv2/core/mat.h\
pp, line 425

This boggles my mind. Does it have something to do with memory? I don't know.
Any help would be appreciated.

is it possible to make a mobile object recognition system using opencv and
android.

please advise me on how to do it.

i am making a thesis and my title would be

Object recognition system for visually impaired people using android and opencv

please help me