The goal for this part of the project was to be able to stitch together two or more images that have overlapping regions. These images are taken by rotating a camera, so they are therefore taken from different angles. By taking advantage of the linear algebra concept of homographic transforms, we are able to change one of the images to look like it was taken from the angle of the other image.
I was able to view an image as if it was taken from a different angle by taking advantage of homography matrices. I first used a set of correspondence points that provided information for how to construct a homography matrix from one of the angle to the other. From there I was able to use the matrix to shift the angle.
| Original | Rectified |
|---|---|
|
|
|
|
|
|
To construct the mosaic, one of the images has to be warped into the angle of the other image. This is done by defining correspondence points from one image to another. These provide information to be able to construct the homagraphy matrix. The matrix is then used to warp one of the images into the plane of the other image. Once that happens, it is important to blend the images together. We do this by linearly blending the images from the center using an alpha term that is based off the x coordinate.
| Image 1 | Image 2 | Final Mosaic |
|---|---|---|
|
|
|
|
|
|
|
|
|
In this project I learned how to construct a mosaic using multiple pictures. The concept that really interested me is the homography matrix. I think it is really cool that such a transform that preserves lines exists.