Image stiching with openCV in python

We want a (python & java) algorithme using OpenCV to stitch images of a store shelf. - up to 20 images could be stitched to a unique image - The images are not token from a single point. the camera is moved for each photo (see attached file) - The order of raw images is not always respected. - Images are not always facing head there are some distorsions - Images could be taken in the order specified in the attached file Pay attention it's a store shelf image so there are so many repetitions of the same article. The final image must not change the number of exact article. (this image will be used for object detection) There must be no distortion nor perspective for the final image. The output: - One planogramme image - parallel shelves with no perspective distorsions - no seam no artifacts - all images stitched (not partly) Here is a link of an exemple of images to be stitch [login to view URL] It's recommanded to use this flow as a start: 0- detect images order 1- Correction of Perspective Distortion (for each photo) 2- Overlay Detection 3-Region-based feature detection (to avoid matching mistakes influenced by a high repetition of products on supermarket shelves.) 4- Matching 5- Optimization of homography angle 6- Compose and Blend NB: auto stitching dont work so please dont try it y're just loosing your time. This project is meant to those who have already tried this type of algo not to spend too much time.