Panorama

Computer Vision

Overview

After taking multiple photos with the same center of projection, I aligned and blended these photographs together to form a new panorama view.

Initial Images used to create the panorama

Final Panorama

Steps

1. Point Correspondence

Select at least 4 corresponding points between each central and peripheral image.

Central Image

Peripheral Image

2. Homography Transformations

Find a homography (3x3 transformation matrix) using linear least squares that map the selected points in the peripheral images back to the corresponding points in the base (central) image.

3. Warping

Warp the peripheral images to align with the base (central) image using the homography found in step2

Warped Images

4. Blending

To remove bad artifacts in regions where images overlap, I created an alpha mask for each image using Gaussian blur and then normalize these alpha maps to make sure that the sum of the alpha weights is equal to 1 across all the images at that location.

5. Final Result

Put all the pieces together to generate the final panorama.