0

我目前正在研究一个使用 ORB 的小算法。它必须在某个时候重新计算关键点和描述符,因为它们的位置和大小会发生变化。但是,在“useExistingKepoints”标志打开的情况下调用detectAndCompute,在断言“级别> = 0”处失败。我很困惑,因为在关键点或描述符中没有称为“级别”的属性。我的问题是究竟是什么导致断言失败以及如何避免这种情况?

仅供参考:由于跟踪器使用矩形边界框而不是圆形关键点,因此我必须将一个转换为另一个。

Keypoint2BoundingBox:

//Create Rect2d with data from the keypoint. 
//Orientation and octave don't need to be saved, since  the object gets reused
//kp: KeyPoint
Rect2d(round(kp.pt.x - kp.size / 2), round(kp.pt.y - kp.size / 2)
       round(kp.size), round(kp.size)));

BoundingBox2Keypoint(此方法的结果被传递给 ORB 导致问题)

//obtain previous kp object and update it accordingly
//kp: KeyPoint;   rect: Rect2d
kp->size = int(round(max(rect->height,rect->width)));
kp->octave = int(round(size2Octave(kp->size, patchsize, scale)));
kp->pt = Point2f(int(round(rect->x-(rect->width/2))),
int(round(rect->y-(rect->height/2))));

大小2八度

//Uses keypoint size and some ORB parameters to compute the octave 
//the keypoint would have
//size: Keypoint Size;   patchsize: size of keypoints at octave 0;
//scale: stepsize to next octave
//returns fraction if Keypoint size is not multiple of patchsize
log10(size/patchsize)/log(scale);
4

1 回答 1

1

正如您在源代码中看到的:

 level = keypoints[i].octave;
 CV_Assert(level >= 0);

你需要修正你计算的方式octave

于 2019-07-16T08:51:55.203 回答