Project 5: Panorama Stitching Part A

Lynn Zhou cs194-26-aet

Project Overview

This project involved shooting pictures and computing homographies to find the transformation so that we can either warp single images in specific directions and perspectives or create a mosaic out of multiple overlapping images, essentially stitching together a panorama.

Shooting the pictures

In shooting and choosing the pictures, I tried to select images that overlapped a lot so they would likely look better and tried to not vary the angle too much. The pictures I used for the mosaics are taken in a cafe on campus, of VLSB, and taken from a previous trip to Cancun

Recover Homographies

To solve for the homographies, we can define point correspondences for each of the pictures and use this equation

We set the scaling factor i to 1 and solve for the rest of the variables, a-h, using a least squares equation. In this case we need 8 equations, which when solving out the equation above, maps to 4 point correspondences at least.

Warp Images

After computing the homographies, I can warp an image using the transformation and inverse warping with the new shape. I warped two pictures, one of my TV screen (watching Frasier :D) and one of the empire state building

Original Empire State

Empire State Warped

Original TV

Warped TV

Blend images into mosaic

To blend images into a mosaic, I warped one of the images into the homography of the other and blended them together

Summary

I learned a lot about how the transformation for the homographies worked and it was surprisingly simple to mosaic multiple pictures

Bells and Whistles

TBD