Advice on shooting a 14ft Bridge model

[WillP]

Hi All

I’m shooting a model of a bridge but struggling to get any image alignment.  This initial test shoot has a few issues but I wanted to get some advice on best practice before re-shooting.  For the test I shot 986 images with this spec - 50MP RAW, 35mm Lens at F14 using a Ring Flash. I edited the images down to the ones that should have good alignment but still can’t get any alignment in Metashape.  I’ve uploaded the gallery of images here https://londonrevolution.net/Bridge/

Clearly I’ve over shot (!) but wanted to have too many images to use rather than too few.  I’m guessing that the biggest issue is the checkered table cloth underneath the model which is creating too many false alignments.  The model is housed in a church crypt and due to its fragility its not currently possible to move it into a bigger space.

Has anyone got any advice on how to approach this?  I’m eager to create the best possible results for the Church!

Thanks loads

Will

[Alexey Pasumansky]

Hello Will,

Can you send the project with the alignment results only (original images are not required) to support@agisoft.com, so that we could check, the distribution of the tie points on the images and any sample original image, so that we could check the shooting settings and image quality?

Background (table cloth) and reflections may cause some problems, but I would also suggest to improve the shooting, by taking more detailed images (from closer distance, while the lens allows to get sharp image) and pointing the camera on the object directly, avoiding taking the images in the following way, where the scene depth changes greatly:
https://londonrevolution.net/Bridge/images/thumbnails/441A0404.jpg
https://londonrevolution.net/Bridge/images/thumbnails/441A0536.jpg
For such shots it is not possible to get sharp details for every part of the image and some areas will be blurred considerably.

The examples of good shooting, according to my understanding, are the following:
https://londonrevolution.net/Bridge/images/thumbnails/441A0584.jpg
https://londonrevolution.net/Bridge/images/thumbnails/441A0682.jpg
https://londonrevolution.net/Bridge/images/thumbnails/441A0792.jpg

Providing that you don't have any limitation to the number of images, I suggest to take more detailed images with better overlap, even though only a small part of the object is captured, instead of making a kind of overview images, which wouldn't give detailed and accurate information for processing.

I would recommend to take several rows of vertical (nadir) shooting lines - similar to aerial survey. For these lines also change the camera orientation by 180 degrees on the neighboring lines, then add the images from the sides (taken from 2 or 3 heights, pointing cameras normal to the object surface), if necessary, add some oblique shots for complex areas with many small elements, but prefer detailed (close) shots instead of overview (distant) shooting.


You may need to use a tripod with adjustable height at least for the side shots, thus you can avoid using flash and provide longer shutter speed (especially, if remote control is available).
The diffuse indirect light is preferred, as the flash is causing glares and hard shadows.

[WillP]

Thanks Alexey!  Your advice was really helpful.  I did a re-shoot of the model and replaced the tablecloth with a random high contrast print.  It worked perfectly