calib3d: findChessboardCornersSB() minor updates#12668
Conversation
- avoid updating of input image during equalizeHist() call - avoid for() with double variable (use 'int' instead) - more CV_Check*() macros - use Mat_<T>, Matx - static for local variables
|
/cc @D-Alex BTW, I'm observing strange results if
|
|
I can reproduce the behaviour. The reason is that the input image has a patten which is basically like a forest for the detector after normalization detecting valid points everywhere. Response Map This should not be a problem as long as the pattern is big enough. However, because the board size is close to the initial size of 3x3 the checks are currently to weak to filter them out. The correct fix would be to enforce more consistency checks for the initial boards or to change the heuristic how initial boards are generated from a single seed point. I will have a look if I can improve the heuristic. Are there any time constraints for getting the fix done? |
|
OpenCV 4.0 gold release is planned in the end of October. This is non-default mode, looks like Do you have some objectives about this patch? |
| for(double angle=0;angle <= CV_PI;angle+=CV_PI*0.1) | ||
| for (int angle_i = 0; angle_i < 10; angle_i++) | ||
| { | ||
| double angle = angle_i * (CV_PI * 0.1); |
There was a problem hiding this comment.
@D-Alex Is this loop still correct? (<= condition has been replaced to < to get exact 10 "tests")
There was a problem hiding this comment.
Looks ok. PI was already tested during the first iteration and should not be tested twice like in the previous implementation.
|
👍 |
Uh oh!
There was an error while loading. Please reload this page.