AutoPano

Part of coursework for Computer Vision at Worcester Polytechnic Institute (WPI) by Prof. Nitin J. Sanket. Course website: https://pear.wpi.edu/teaching/rbe549/spring2024.html

The purpose of this project is to stitch two or more images in order to create one seamless panorama image. Each image have few repeated local features (∼30-50% or more).

Phase1: Traditional Approach

Input Images

Feature extraction and ANMS

Feature matching

Feature matching after RANSAC

Stitched Panaroma

Phase2: Deep Learning Approach

Supervised Approach to estimate the homography

Supervised model

Results (Green: Ground truth, Blue: predicted)

Unsupervised Approach to estimate the homography

Unsupervised model

Results (Green: Ground truth, Blue: predicted)

Detailed explanations can be found in the project report.

Collaborators

• Ashwin Disa - amdisa@wpi.edu
• Mihir Deshmukh - mpdeshmukh@wpi.edu