Skip to content

Autopano - Stirching multiple images to form a seamless panorama

Notifications You must be signed in to change notification settings

thabsheerjm/My_AutoPano

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

My AutoPano - Image Stitching

In this project we are learning to stitch two or more images in order to create a seamless panorama by finding the homohraphy between the two images.

Phase 1 : Classical Computer Vision approach

The claasical method involves several steps, it starts with corner detection. Then Adaptive Non-maximal Suppression (ANMS) is applied to ensure an even distribution of conrners. Feature descriptors are created by encoding information at each feature point into a vector. The next step is feature matching, where feature points between images are matched. RANSAC is used for outlier rejection and to estimate a robust homography. Finally, the images are blended to produce the panorama.The deatiled steps are explained here.

How to Run:

Place the image pairs in the dir:"../Data/Train/Set1/" and run the following command:

 python3 Wrapper.py

The classical pipeline is given below:
Undistorted

Corner Detection

Undistorted

Non-Maximal Suppression

Undistorted

Feature Matching

Undistorted

Outlier Rejection

Undistorted

Parorama: Warping, Blending and Stitching

Undistorted

Phase 2: Deep Learning

The deep learning model effectively combines corner detection, ANMS, feature extraction, feature matching, RANSAC and estimate homography all into one robust generalizable network. The complete methodolgy is given here.

How to Run:

Run the following to generate the dataset

python3 Datagen.py

Divide train and validation. and run:

python3 network.py
python3 Train.py
python3 Test.py

The network pipeline is given below:
Undistorted

We followed a supervised model and obtained the follwing results:

The following picture shows the input and output patch from the trained network. Undistorted

The training loss over number of epochs is given below:

Undistorted

Acknowledgement

This project was part of RBE549- Computer Vision (Fall 22) at Worcester Polytechic Institute[1].
Team members :Thabsheer Machingal and Krishna Madhurkar

References

[1] RBE549-Project1

About

Autopano - Stirching multiple images to form a seamless panorama

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages