7. Perspective transformation

Perspective transformation is also called projection transformation. The affine transformation we often call is a special case of perspective transformation. The purpose of perspective transformation is to convert an object that is a straight line in reality into a straight line through perspective transformation, which may appear as a diagonal line in the picture. Perspective transformation can map a rectangle to an arbitrary quadrilateral. This technology will be used later when our robot drives autonomously.

Perspective transformation via function:

dst = cv2. warpPerspective(src, M, dsize[,flag, [,borderMode[,borderValue]]])

l dst: Output image after perspective transformation, dsize determines the actual size of the output.

l src: source image

l M: 3X3 transformation matrix

l dsize: Output image size.

l flags: Interpolation method, the default is INTER_LINEAR (bilinear interpolation). When it is WARP_INVERSE_MAP, it means that M is an inverse transformation, which can realize the inverse transformation from the target dst to src.

l borderMode: edge type. Default is BORDER_CONSTANT. When the value is BORDER_TRANSPARENT, the values in the target image are not changed. These values correspond to the outliers in the original image.

l borderValue: border value, default is 0.

Like affine transformation, OpenCV will still provide a function cv2.getPerspectiveTransform() to provide the transformation matrix above.

The function is as follows:

matAffine = cv2.getPerspectiveTransform(matSrc, matDst)

l matSrc: input the four vertex coordinates of the image.

l matDst: output the four vertex coordinates of the image.

The code was run on jupyterlab

​x
import cv2

import numpy as np

import matplotlib.pyplot as plt

 

img = cv2.imread('yahboom.jpg',1)

 

imgInfo = img.shape

height = imgInfo[0]

width = imgInfo[1]

#src 4->dst 4 (upper left corner lower left corner upper right corner lower right corner)

matSrc = np.float32([[200,100],[200,400],[600,100],[width-1,height-1]])

matDst = np.float32([[200,200],[200,300],[500,200],[500,400]])

#combination

matAffine = cv2.getPerspectiveTransform(matSrc,matDst)# mat 1 src 2 dst

dst = cv2.warpPerspective(img,matAffine,(width,height))

img_bgr2rgb = cv2.cvtColor(dst, cv2.COLOR_BGR2RGB)

plt.imshow(img_bgr2rgb)

     plt.show()