Panoramic image stitcher based on 'Automatic Panoramic Image Stitching using Invariant Features' by Matthew Brown and David Lowe.

Using Python 3, install:

• numpy
• cv2
• argparse
• scipy
• ordered_set

python3 ./src/stitcher/main.py <path_to_input_images_directory>

1. Extract SIFT features from all images
2. Find similar SIFT features using KD tree
3. Verify matches using RANSAC, more inliers = good match
4. Iteratively add each image to the bundle adjuster (with the best matches being added first)

• Rotation matrix is converted to rotation vectors to allow the rotation to be parameterised with fewer variables in order to reduce the problem size
• Uses Symbolic differentiation to build the Jacobian matrix
• Optimises for the reprojection error between matches, this bundle adjuster does not take into account the 3D projections however this will be added as an option in the future

• The Adobe Panoramas Dataset has been used for testing: https://sourceforge.net/adobe/adobedatasets/panoramas/home/Home/

• Improve RANSAC speed
• Improve bundle adjustment speed
• Add spherical projection for output image
• Add gain compensation
• Add blending