drVit's profile - activity

If your video streams is well time synchronized (frame to frame), then the simplest (may be not the best) way is to process frames just like static images. Showing your video example could improve answers.

May be bool cv::ocl::haveOpenCL() and cv::ocl::setUseOpenCL(bool) is what you want.

be careful with OpenCL Canny, it is buggy at least ~one year

Since video is not space only but time too you should start from time-synchronization.

If you will create precise mask as described before you will be able to scan rows (and then columns) for mask background value presence. So the first mask top line without 'black' is your top crop rectangle side etc. Countnonzero will help you to find is it right slice or not.

Try to use fitellipse function.

Since your image contains black pixels there is no simple and safe way to distinguish where are black pixels from background and where -- from you image. You'd better create rectangle mask for every source image and then transform every mask the same way you transform every source images for stitching. Then you will have a very precise and correct mask.

If you want just to crop it use Rect to create submatrix from matrix:

UMat crop = src(Rect(left, up, width, height));

try estimaterigidtransform with appropriate fullAffine-flag

Is there any way in openCV to grab from videofile keyframes only? I googled ( http://code.opencv.org/issues/1419 ) before video.set(CV_CAP_PROP_POS_FRAMES) worked this way: position was "nearest keyframe", but now it is "fixed" and I can not find the way how to process keyframes for perfomance purpose only even in one way:

• 1st way : read() next keyframe
• 2nd way: do fast-working grab() for each frame, check it's propertie (which one?) and if it is keyframe then do slow-working retrieve() .

P.S. password reminder does not work, at least today for my old profile -- so I had to create new account