How can I improve the textured mesh of a building’ façade?

[Constantinos]

My first attempt to create a 3D model of a part of a single house did not brought the desired outcome.

However, I would like to start by giving some information and then I will explain the above issue. All the photographs that I took were with a DJI Phantom Pro V2 and I used the 3DF Zephyr Lite version. Before to start to create a 3D model of a building, I created a 3D model of a field with some ground piles. With almost 100 photographs, which I took them through a flight mission of an application, I can say that the 3D model was pretty good. Nevertheless, I believe that there are some allowances of improvement.

Then I said to experiment with something different, like a façade of a house. I took manually 139 photographs through my drone from different heights and distances. I inserted them to the software and I created the textured mesh. The textured mesh looked very bad, as you can notice from the attached picture. There are some holes; the texture is not smooth or clear. The exterior walls are not flat but there are a lot of concavities. I noticed that the holes and the concavities were on the walls’surfaces, which were painted white or beige. On the other hand, the spots that were rougher, e.g. roof tiles or exterior wall cladding looked better than the others.

I decided to do some other trials by changing some settings through the process or adding a filter. I checked the ‘Image Quality Index’ (most of them were almost red, which is bad, and I do not know why) and the ‘Camera Stats’ and I started a new project by subtracting 20 photographs. I used the following mesh filters: laplacian smoother, bilateral smoother, fill holes and photoconsistency. I used the aerial or urban preset and the default or deep in details preset. By doing all these changes I did not manage to accomplish a better result from the first textured mesh with the 139 photographs. Frankly, they were worst.

So I would like to ask for your help by giving me advices and guidelines in order to improve my 3D model but at the same time to ameliorate my skills with this software because I really liked it. As a civil engineer I am interested with the potentials that this software provides.

Which parts I have to check in order to change some settings, which will give me the desired outcome?
You have to be more careful when you want to create a 3D model of a façade than a surface of a field?
Shall I use different techniques when I take photographs of a façade than a field? The answer of this question will play a significant role for the process of the 3D model’s creation?
Which are the characteristics of a good 3D model? What checks I have to do?
What do you advice me to do in order to learn better the software?

Thanks a lot.

[cam3d]

Hi Constantinos

Thank you for your detailed explanation of what you're trying to achieve, and what you have managed to achieve so far.

Fundamentally, photogrammetric processing relies on matching feature points across a set of images for camera orientation. I see that you have already found that surfaces that are homogenous and feature-poor (such as painted walls) resolve in a less detailed way than a feature-rich surface.

Rough/bare concrete is a good example of a feature-rich surface that will give you better reconstruction results. In the case outlined above (field with ground piles) I'd hazard a guess that the scene was feature-rich and therefore reconstructed in a satisfactory manner.

WIth feature-poor surfaces, your photographic acquisition becomes much more important. High ISO can result in image noise, motion blur or focus issues can obscure feature points which otherwise could be detected by Zephyr and used to aide the reconstruction, and capture patterns/flight paths can have a significant impact on the final result. There are many other things that could be at play here and without looking at the data itself, it can be quite tricky to determine the exact cause of the problem.

Regarding settings, it's best practice to aim to capture a dataset which will run on default settings. In my personal experience, it's better to shoot a dataset again than to try and make a tricky dataset work through adjusting settings.

The acquisition technique changes from scenario to scenario, but it's a good idea to take more photographs with greater overlap when you know the subject is feature poor.

Depending on your application a good 3D model will have simply the elements you need, and nothing you don't. If it's for 3D printing, you want to ensure there are no holes in the surface of the mesh which would otherwise be a 'watertight' volume. If the model is for visualization only then the requirements are much more forgiving.

Scaling a subject in Zephyr is easiest and most accurate with the full version as you can use control points to drive the 3D transform/rotate/scale.

Practice practice practice is the best way to improve your skill level using 3DF Zephyr, & we additionally have a wide range of documentation and video tutorials which can be found here: https://www.3dflow.net/technology/do...hyr-tutorials/ - Anything not covered by these resources can most likely be answered here on the forum, our Discord server or if you would rather, you can send questions to support@3dflow.net

If you are able to share the photographic dataset and your .ZEP project file we can have a look into other reasons why the results of your house scan are non-ideal.

I hope this was informative! Let me know if you have any further questions. Happy to help!