Homography and image alignment in Computer Vision - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to compute the homography matrix using OpenCV.

Computer Vision

import cv2
import numpy as np

# pts_src and pts_dst are arrays of corresponding points
H, status = cv2.findHomography(pts_src, pts_dst, [1])

Drag options to blanks, or click blank then click option'

Acv2.LMEDS

Bcv2.RANSAC

Ccv2.USAC_DEFAULT

Dcv2.USAC_FAST

Attempts:

3 left

💡 Hint

Common Mistakes

Using a method other than RANSAC may cause poor homography estimation.

Not specifying a method leads to default behavior which might not handle outliers well.

The cv2.findHomography function uses cv2.RANSAC as the robust method to estimate the homography matrix while ignoring outliers.

2fill in blank

medium

Complete the code to warp the source image to align with the destination image using the homography matrix.

Computer Vision

height, width = img_dst.shape[:2]
img_aligned = cv2.warpPerspective(img_src, [1], (width, height))

Drag options to blanks, or click blank then click option'

AH

BH.T

Cnp.linalg.inv(H)

Dnp.eye(3)

Attempts:

3 left

💡 Hint

Common Mistakes

Using the inverse of H instead of H causes incorrect warping.

Passing identity matrix results in no transformation.

The warpPerspective function uses the homography matrix H to transform the source image to align with the destination image.

3fill in blank

hard

Fix the error in the code that extracts matching keypoints from two images for homography estimation.

Computer Vision

matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
pts_src = np.float32([kp1[[1]].pt for m in matches]).reshape(-1,1,2)

Drag options to blanks, or click blank then click option'

Am.queryIdx

Bm.trainIdx

Cm.distance

Dm.imgIdx

Attempts:

3 left

💡 Hint

Common Mistakes

Using trainIdx instead of queryIdx swaps source and destination points.

Using distance or imgIdx causes attribute errors.

For the source image keypoints, we use m.queryIdx to get the index of the keypoint in kp1 corresponding to each match.

4fill in blank

hard

Fill both blanks to create a dictionary of aligned images and their homography matrices.

Computer Vision

aligned_data = {
    'image': img_src, 
    'homography': [1], 
    'status': [2]
}

Drag options to blanks, or click blank then click option'

AH

Bstatus

Cmatches

Dimg_aligned

Attempts:

3 left

💡 Hint

Common Mistakes

Confusing matches with status causes incorrect data storage.

Using img_aligned instead of H for homography is incorrect.

The dictionary stores the homography matrix H and the status mask returned by findHomography to indicate inliers.

5fill in blank

hard

Fill all three blanks to filter good matches and extract corresponding points for homography.

Computer Vision

good_matches = [m for m in matches if m.distance < [1]]
pts_src = np.float32([kp1[[2]].pt for m in good_matches]).reshape(-1,1,2)
pts_dst = np.float32([kp2[[3]].pt for m in good_matches]).reshape(-1,1,2)

Drag options to blanks, or click blank then click option'

A30

Bm.queryIdx

Cm.trainIdx

D50

Attempts:

3 left

💡 Hint

Common Mistakes

Using too low a threshold filters out good matches.

Swapping queryIdx and trainIdx reverses source and destination points.

We filter matches with distance less than 50, then extract source points using m.queryIdx and destination points using m.trainIdx.