mrzl's profile - activity

Hey everybody,

I'm currently in the process of clustering a set of points into different sections. Though, the k means algorithm requires me to explicitly set the number of clusters to divide the set of points in. Is there an alternative algorithm, where I can pass in a set of points and a maximum distance, which divides point clusters from each other? What I would like to have returned is a list of clusters of yet unknown size, containing a sub-set of points.

I would be very thankful for any directions.

Thanks, Marcel

I've got the same issue on my machine. I'm using the Metaio framework, though. This is bollocks.

Hey everybody,

I'm looking for a GPU accelerated version of the cv::medianBlur function. It doesn't seem to be documented( http://docs.opencv.org/modules/gpu/doc/image_filtering.html ), and I can't seem to find it in the library itself either.

I'm just wondering, since it's not going to be the biggest problem to implement it with a custom kernel.

Thanks in advance, Marcel

Why don't you give us a little more details about what you've got, otherwise I've got to simply tell you that you gotta implement it yourself.

Hey everybody,

I'm trying to use cv::gpu::FarnebackOpticalFlow, but I'm getting an error I don't know how to interpret:

When I'm creating an object like

    cv::gpu::FarnebackOpticalFlow farn;

I'm getting an error stating that I don't have a specific processor feature present. It's happening in the constructor of the FarnebackOpticalFlow class:

    isDeviceArch11_ = !DeviceInfo().supports(FEATURE_SET_COMPUTE_12);

Even more down the stack trace I'm ending up here:

    #if defined (_M_IX86) || defined (_M_X64)
    if (IsProcessorFeaturePresent(PF_FASTFAIL_AVAILABLE))
    #endif  /* defined (_M_IX86) || defined (_M_X64) */
    __fastfail(FAST_FAIL_STACK_COOKIE_CHECK_FAILURE);

and this is where I fail to come up with a solution. Maybe somebody of you can give me a hint to solve this, since the CPU version of the Farneback flow is a little too slow for my needs.

Thanks in advance. Yours, Marcel

I'm using Win8 64bit with the standard cv::VideoCapture implementation ( not sure what you mean ) with opencv version 2.4.7

Hey everyody,

I've come to the point that I've got a scene and the objects I would like to track are separated from its background and I'm calculating the optical flow of that image. Furthermore I would like to detect and track this object/blob, but I don't know what approach to choose to detect objects on top of the calculated flow.

This is a small illustration of my problem:

I'm not looking for a solution to my problem, just an approach to how I should continue. Thanks a lot for the help in advance.

Yours, Marcel

I'm doing some motion detection on a video and while doing the processing, it sometimes is handy to simply skip a couple of frames in order to have a bigger impact on the e.g. optical flow detection. Currently I'm simply loading frames in a loop like that:

for( int i = 0; i < playbackSpeed; i++ )
{
    originalImage = videoReader.getNextImage();
}

Though I'd much rather just skip a couple of frames instead of just loading more, because doing this ends up in a little performance problem. Cheers for the help.

mrzl

I'd be interested, too.

Yeah eventually i figured it out- somehow. I decided to compile the 2.4.7. version from the github repo and compiled with the WITH_QT option DISABLED. I'm using Qt within VS12 via the Qt plugin now and it's going great. The version I sucessfully compiled is the one tagged 2.4.7. in Itseez's github repo ( not the very latest one ). I hope this helps. let me know if you need any more info.

Hey!

I've got a videofile loaded via the VideoCapture class and everything is working fine. I can jump to specific frames via the .set( CV_CAP_PROP_POS_FRAMES, frameNr ) function. Though now I want to accomplish the same thing with the cv::gpu::VideoReader_GPU class. If anybody did this before I'd love to hear how. :-)

I'm running Windows 8 with OpenCV 2.4.7.

Thanks a lot in advance! Yours, mrzl

Hey, I've got exactly the same problem. Everything compiles perfectly except from the highgui module. I've read that enabling WITH_QT in cmake might resolve the problem, but I'm getting tons of configuration errors I haven't resolved, yet, when I'm enabling QT.

I'd love to hear from your progress if you managed to fix this somehow. Thanks!

mrzl