newbe

[castellum]

Hi There,

I am a new user of Zephyr and have been experimenting a lot with simple object's so far.
This weekend I have been trying something more complex. At least I think it is :-)

I am trying to scan a part of a motorbike while it is mounted on the bike.
So the photographs will not only contain the mounted part but everything around the bike as well.

I added an example image.
Is it possible to create q model out of 50 pictures like this, circling the Dashboard?
Should I prepare the images before processing in Zephyr? If so...How?

Kind regards

[Andrea Alessi]

Hi castellum,

yes, it is possible to reconstruct a bike, but beware of some things:

- uniform surfaces (like plastic) are hard for photogrammetry. You should consider taking your bike for a ride, have it dirty, then scan it

- make sure that your pictures are very well in focus. In your first image for example, the left handlebar is out of focus. Make sure your whole subject is well in focus. Here is a brief tutorial that goes a little bit deeper on photography for photogrammetry https://www.youtube.com/watch?v=E06kgYBftak

- never do pre-processing or post-processing of your images. I know it's counter-intuitive, but editing your images in photoshop or other software will actually remove information and make stuff harder for zephyr. The only pre-processing we suggest is using 3DF Masquerade to mask the subject, when needed. In you case, i assume the subject is still so no masking is required. If your shooting in that backyard however, there is the chance that zephyr will assume you want to reconstruct the building as well, so make sure to resize the bounding box (if needed) after the sparse point cloud has been generated.

Lastly, i'm not sure the full motorbike can be done with only 50 images, at least not every detail consider using 3DF Zephyr Lite after you've practiced a bit with the free version: you can download zephyr lite with a 14 days free trial, no registration required

[castellum]

Thanks Andrea,

Thanks for your advice's.

I will certainly do my best to get the bike dirty soon :-)

I think the focus is on the right place though. In this Case I would only like to create the cockpit, including the GPS mount in the picture.
Maybe the masking is what I was looking for. I now processed about 50 images. The green bush next to the bike came out reasonably good but that was not exactly what i was looking for.
And I did not find the bounding box so far either. SO I will look into that. next.
Does that provide the possibility to reduce the process to a pre-selected cube that surrounds only the part I want to create.

And yes I used the free version.
First wanted to get an idea what the tool can do and if it is not tot complex for my limited knowledge :-)

[Andrea Alessi]

If you want to do just the cockpit then it's probably fine, although i prefer to have everything crystal clear.

You can find the bounding box tutorial here: https://www.3dflow.net/technology/do...-bounding-box/

No worries, if you have any question feel free to ask, every question is welcome

[castellum]

I found the boundign box and did some experimenting with several configurations but with the current 50 images I don't seem to be able to create anything that looks like the cockpit what so ever.
After the Sparse Point cloud the output does not give me anything that looks like the model. No idea what part of the cloud is the cockpit so I can limit the bounding Box to that part.
It must be en issue with the Images.

I will post them in Drive so if anyone could give me a clue how to improve that would be great

https://drive.google.com/drive/folders/1TorR6OZiHcmAWoXmXFbstbuGjzUMXZtI?usp=sharing

[Andrea Alessi]

I had a quick look at your dataset, here are my thoughts:

- try to take sharper images. In many images many parts of the cockpit are not in focus. Sometimes you can work even with partially blurred images but that takes a bit of experience. In this case, the subject is hard already and having blurred photos is not going to make it an easy process.

- make sure the orientation is correct. If you ran in default/close range, you'll notice that your camera positions are correct - there are some cameras that have been incorrecty placed. This may happen for various reasons, i.e. reflections, repeated patterns, out of focus images that throw off the keypoints matching.

Overlaying to the camera helps understand this a lot. When cameras are not oriented correclty, either remove them or re-run the dataset with different settings. Here is what i mean by "you'll notice that your camera positions are incorrect"





- you started with a difficult subject many parts are uniform which are hard for this technology. Either dirty them or use a projected pattern on top of it - but it's an advanced process that i reccommend you investigate only when you have more experience in shooting and with zephyr. Moreover, there are parts with reflections in the cockpit, which are hard to minimize. Moreover, you also have a shiny and reflective part of the subject (The metal handle). It's possible to do, but I suggest you start with easier subjects to learn the ropes

To be honest, the input dataset is very challenging. I'm sure you could get the camera position correctly oriented, but then i'm fairly sure the dense point cloud would be pretty much unusable, so i would take another new photoset of the subject.

I suggest taking a look at our videotutorial series: They are short and to the point, i promise https://www.youtube.com/watch?v=17UT...eG4Jzu1CM5atRH

I hope this helps let me know if you have any question!

[castellum]

This helps a lot.

I did some easier objects first by the way. And since they went rather smooth I tried this project.
Obviously I will need some time understanding the process better and how to asses the image quality better.

Kind regards

[Andrea Alessi]

glad to help!

feel free to ask any questions you may have anytime

[castellum]

Hi Andrea,

I think I have done at least a dozen shoots of this object and processed them.
Read all I can find and watched all the Tutorials but none of the attempts gave me an output that even remotely looked like the pictures :-(
Obviously the SmartPhone pictures don't do the job.
I am in the process of selecting a new DSLR. I will give it another go after I get it.

Thanks for the support anyway.

[Andrea Alessi]

Hi castellum,

having a better camera will help, but please keep in mind that textureless, shiny/glossy objects are always going to be very difficult. Have you tried dirtying the surface?

best,

Andrea