Hello, can you help me with my research

[wislemmm]

Hello guys, I'm new to photogrammetry. I'm getting familiar with the technique by watching several videos and reading some articles on the internet, but I'm having a hard time doing my job, despite having several tips from the internet. So I decided to use this forum, I hope you can help me.





So I tried to complement the photographic protocol with other photographs. I tried to use the zoom or to position my smartphone more freely to focus on the regions of interest. In the end, the model had not had many changes or the result was getting worse. Sometimes the software also did not recognize some photos.
I read that the more photos the better to capture more details, but, contrary to theory, I am getting better results when I take fewer photos. I don't know if I don't take the photos correctly.

I will share a link with the photos and models from my last test so that you can better analyze my situation.

Anyway, I would like a help for my specific case. I've done several tests and got better results over time, but I still can't fully understand what I'm missing in order to not be able to generate a model with all well-defined regions. I would like to thank you in advance and all your tips and opinions to help me with this work are welcome.

Taking advantage of the text, some other doubts. 1) I am carrying out the construction on an average notebook. I would like to know if this interferes a lot in the quality of the generated model or the quality of the photos have a greater weight in this. 2) I have already tested photographic protocols with a greater number of photos and others with fewer. What determines the size of the .obj file that I export? There have been cases where I rebuilt a model with 100 photos and the file weighed less than 10MB, while a model generated with 50 photos weighed 300MB. In the workflow I use in the software, all presets are default and defaut multi-texture in the last step and the category is general.

[cam3d]

Hi wislemmm - Welcome to the forum!

Thanks for sharing the photographic dataset and the model created from that set.

A quick look at the meta-data in the image files and it looks like you are using a Moto G (5S) Plus phone camera. Using a phone camera will typically yield worst results than a DSLR or mirrorless camera as the sensor size (and subsequently the photosites on the sensor) are much smaller and receive less light.

Your ISO is quite high, sitting between ISO 793-1000. Higher ISO settings result in higher levels of image noise, so you want to keep this as low as possible (Ideally sub ISO400).

Your F/stop is quite low (F/2) - This is generally fixed on mobile cameras and can't be adjusted. Using a DSLR will open up F/stop options and using this calculator you can establish an approximate DOF which encompasses the subject and keeps everything sharp: https://www.photopills.com/calculators/dof

Your shutter-speed is quite slow (around 1/24) - If you are shooting handheld then this speed may result in motion blur which you want to avoid. Using a tripod with a delayed trigger will allow you to reach slower shutter-speeds with less concern about motion blur. Having a slower shutter speed will mean you can reduce your ISO to less noisy levels and improve image quality.

From the looks of things you also have automatic settings enabled - Auto-settings are to be avoided as much as possible to keep your images consistent across the set.

Here's an example of cast teeth which I captured using a Canon 600D with 50 photos: https://skfb.ly/6AEWL

-------------


1. Your computer hardware will either speed up or slow down the processing, all things equal - The model quality will remain the same regardless of hardware. Photographic input is a major factor when it comes to reconstruction quality.

2. In most scenarios image count is not strictly tied to polygon count. The OBJ file size is dependent on the number of polygons generated - Polygons generated is a result of many factors, including framing of the subject, overlap of images, resolution of images, processing settings etc... Polygon count can be reduced using decimation, or increased using densification tools. (Found under Tools>MeshFilters)

[wislemmm]

Hi cam3d, thanks for the tips. I left all settings in manual mode and moved my camera closer to the plaster model to keep the focus on the teeth. The result got better. The new file is in Google Drive, in case you want to take a look.

In this first stage of my work I have to use the smartphone to take the photos. Really with the camera aperture fixed, my depth field is limited.

Is it a good idea to take new photos to try to better capture the details of some regions that are not well defined? I did this in some previous tests, but it seemed that the resolution got worse in regions that were previously good.

Any tips on how I can complement my photographic protocol? Or do I already achieve the best result I could using a smartphone?

[cam3d]

Great to hear you made improvements by switching to manual settings. Ideally you will have every point on the surface you wish to reconstruct viewed by 3 distinct camera positions. If you take more photos and they result in a worse reconstruction then it is likely that the additional photos are poorly orientated for some reason. This could be because something in frame has moved relative to the subject and is taking the matching procedure off-course.

Lighting is a major factor when it comes to creating high quality images. I would prefer a lower quality camera and higher quality lighting, rather than low quality lighting and a high quality camera (having both high quality is obviously the ideal) - If you can take your subject outside into daylight (overcast and still) then you will have plenty of nice high quality light to work with. The alternative is studio lighting which offers much more control however this can be quite a complex subject.

Here is more info on turntable scanning: https://www.3dflow.net/technology/do...for-turntable/