3dv_tutorial/examples/image_stitching.py at master · mint-lab/3dv_tutorial · GitHub

Repository navigation

35 lines (29 loc) · 1.34 KB

import numpy as np

import cv2 as cv

# Load two images

img1 = cv.imread('../data/hill01.jpg')

img2 = cv.imread('../data/hill02.jpg')

assert (img1 is not None) and (img2 is not None), 'Cannot read the given images'

# Retrieve matching points

fdetector = cv.BRISK_create()

keypoints1, descriptors1 = fdetector.detectAndCompute(img1, None)

keypoints2, descriptors2 = fdetector.detectAndCompute(img2, None)

fmatcher = cv.DescriptorMatcher_create('BruteForce-Hamming')

match = fmatcher.match(descriptors1, descriptors2)

# Calculate planar homography and merge two images

pts1, pts2 = [], []

for i in range(len(match)):

pts1.append(keypoints1[match[i].queryIdx].pt)

pts2.append(keypoints2[match[i].trainIdx].pt)

pts1 = np.array(pts1, dtype=np.float32)

pts2 = np.array(pts2, dtype=np.float32)

H, inlier_mask = cv.findHomography(pts2, pts1, cv.RANSAC)

img_merged = cv.warpPerspective(img2, H, (img1.shape[1]*2, img1.shape[0]))

img_merged[:,:img1.shape[1]] = img1 # Copy

# Show the merged image

img_matched = cv.drawMatches(img1, keypoints1, img2, keypoints2, match, None, None, None,

matchesMask=inlier_mask.ravel().tolist()) # Remove `matchesMask` if you want to show all putative matches

merge = np.vstack((np.hstack((img1, img2)), img_matched, img_merged))

cv.imshow('Planar Image Stitching', merge)

cv.waitKey(0)

cv.destroyAllWindows()