Panorama stitching

For more details - Project Report

Repository - Github

Stitching multiple images with 30 − 50% overlap to generate a panorama using classical approach. Steps involve corner detection, ANMS, feature extraction and matching, RANSAC and estimating homography.

Input data

Undistorted images


Corner detection

Harris corner & Adaptive Non-maximal Suppression (ANMS) for uniform distribution of features.


Feature matching and RANSAC for outlier rejection

Match keypoints (encoded as feature vectors) across pair of images. Refine the matches using RANSAC.


Stitch all images

Estimated homography using the refined matches, warp and stitch images with overlap.


More panoramas