Video from a fixed position and rotating plate

[Selbi]

Hello there, I've just started out with Photogrammetry and so I'm really new to the whole technology as well as 3DF Zephyr itself.

I want to recreate small office objects such as tackers, scissors, and pencils (really just to get started). To minimize blur and the tedious work of having to move around the small objects, I've bought a rotating plate and a smartphone tripod.

I've posted what the setup looks like in the attachments below.

However, the results are ATROCIOUS. It doesn't seem to recognize more than half of all pictures, and during the recreation it was only able to capture the plate (from its one perspective). You can also find that in the attachments.

I've received significantly better results by shaky hands and awkwardly moving around the object manually.

Are there any settings I need to give to 3DF Zephyr to tell it "all photos were shot from the exact same location"? It doesn't make sense to me that more stable shots result in a much worse model.

I've used the default settings for everything and set the preset to "Close Range".

[Yogesh Sonar]

Hello friend,

I am also new in photogrammetry, there is facilities in software to handle this task, but I use simple way at least for my application, I simply delate object background by using external photo adding soft, and than normally reconstruct, the result was successful.

Thanks

[Andrea Alessi]

Hello Selbi,

it looks like you did not mask the background. Unfortunately, zephyr can not know if you wish to reconstruct the object or the table. And by having the camera fixed, zephyr is exactly doing its work in the first phase: computing the camera position, which is, well, fixed

So you have to trick zephyr into thinking that there is only the subject, and not the table. You can do it with masking: as yogesh suggest, you need to delete the background, but i suggest you do not use third party software - if you use, say, photoshop, you'll strip away exif data and probably end up cropping the images (which is bad) so you should use the bundled 3DF Masquerade application.

Masquerade also have a turntable mode you can enable, which speeds up the process even more (as it looks for a similar background from the previous processed image).

here is a tutorial https://www.3dflow.net/technology/do...-masking-tool/ and here is a videotutorial https://www.youtube.com/watch?v=dGRw8LbXknU

[Selbi]

I've now done the masking for every single image on multiple datasets (some extracted from a video, the most recent one from actual pictures), but I still struggle to get proper results. The biggest issue seems to be that the tool discards more than half of all images because it doesn't "recognize" them. But the images don't seem to be so bad for that.

Please help me identify what I'm doing wrong. Do my images have a too low resolution?

Here is my dataset of a stapler with my masking: https://nofile.io/f/c3e9fPCe6pT/stapler.zip

[Andrea Alessi]

Hi Selbi,

the photos are okay, considering that you are using a cellphone - however, try to change height when doing a new orbit around the subject. See the tutorial 1 example photos to see what i mean https://www.3dflow.net/technology/do...ls-3df-zephyr/

However, the dataset is already difficult per se - you are trying to reconstruct a subject that is mostly plain surface with no texture, moreover, with some shiny surfaces too.

I suggest you try starting with easier subjects (small objects that are not composed by a uniform surface: for example try on some fruits such as apples, bananas, etc or small statues, etc) so that you can learn how the process works a little bit more in detail.

This is going to be hard even with proper photography equipment, however, if you really want to try with this subject, i suggest you try putting under the object something textured (such as a newspaper). It will be easier for you to move around the object rather than moving and masking.