Posts Tagged ‘panorama stitching’

Making of: step 2, the stitching of Paris

Thursday, March 11th, 2010

The first step after the shooting is to cope with RAW files. Let Arnaud Frich explain his method and goal :

The stitching itself remains a bit complicated with a project of this size. Even if using a motorized panoramic head helps a lot during the shooting, the stitching still doesn’t give perfect results because of certain points : first, robustness of the motorized panoramic head (we had a break after each row to manually correct the focus) and second, repeating patterns in some of the buildings.

Let me remind you of the shooting layout in the Paris image : 138 columns by 17 rows, that is 2346 images. The first stitching gave the following image :

We can notice a shift in the rows which comes from a shooting issue: we missed 3 images during the shooting at the edge of one row and the beginning of the next. It’s the kind of stuff you don’t notice when out in the field but when you come back to the office (BTW: it was really a nightmare figuring out which images were missing as they were not named after the actual shooting order. Luckily, the missing images were at the edges of the panorama so the consequences were really low. We cropped the zone on the left side of the panorama. Yes! We could have 2 more columns in the panorama because we used only 136 out of 138 shot columns => this would have been 27 Gigapixels.  Anyway…).

After spending an entire day finding and renumbering the missing images, we were able to use this new well-numbered set of images with the Clauss plugin of Autopano Giga in combination with the shooting log file.

The first optimization of the project with all images was quite good but not perfect because of some bad links remaining in the editor.  It’s easy to guess why: repeating patterns in an urban landscape are really common; for example, there is nothing more similar than two windows next to each other. This can lead to alignment issues in detected control points between neighboring images. That’s what we can see in the following screen shot which raises this issue :

This screen shot shows a standard detection without using the Clauss plugin of Autopano Giga. You can see a lot of red zones where all false links have been found. If you use a motorized head, the issue is not as big as this one. It will be really low but can still exist. Nevertheless, isn’t that screen shot beautiful ?

With practice, coping with these false links is quite quick. What takes much longer is coping with out-of-focus zones (for example with the top of the roof of St-Sulpice tower). No miracles here. You need to manually move around the images with the move mode in the editor. A full day of work was needed to do that because even if each operation was really easy to do, it took time for each move operation to happen on a panorama of that size. Regular saving of the project is highly advised.

Once we arrived at that step, we had a perfectly geometrical panorama. But we still need to cope with color correction. The lighting of the scene changed a lot during such a long shooting so we needed to be careful with that. BTW : yes, if you look accurately at the image, you can see that shadows moved in the picture. Not easy to guess, but it’s findable. So we put a lot of color anchors in the panorama :

With the help of multiple selection, it’s a fairly quick operation. Optimization was done right after so we have a nearly finished project, as illustrated below :

We still need to crop, but we’ve got the final size : 354159 x 75570 = 26,763,795,630 pixels. Wow. Not bad ;)

We can now proceed to the rendering.

Alexandre Jenny