How transformation matrix is calculated for removing the perspective using cv2.getPerspectiveTransform in opencv

Question

I am trying to understand cv2.getPerspectiveTransform? In the official document (https://docs.opencv.org/4.x/d2/de8/group__core__array.html#ga12b43690dbd31fed96f213eefead2373), cv2.getPerspectiveTransform needs two sets of coordinates (e.g. one:tilted images, another: frontal images).

And, the homography relationship of these two set of coordinates are calculated using SVD (Singular Value Decomposition).

So my exact question is that how SVD can solve this problem? could you please kindly explain how SVD solves this indeed??

Thank you