Hi all,
Above is a link to a viewable 3D model I made on Sketchfab.
I’m looking for some advice from some people with more experience than me please.
I’m very happy in general with the model. However, the side of the building facing the field has a veranda and it’s causing it to create a hole in the model. Any advice to solve this would be gratefully received.
I used dronedeploy, and have been thinking of getting Dronelink to be able to add some lower altitude orbits, as I suspect the drone is either too high, or to near the building making the angle unfavourable.
Each side has the same number of images. Its because the roof of the veranda overhangs the side of the building making the angle from the drone camera unable to see the true side of the building.
I tried taking some manual photos of the side, keeping in mind to angle the camera so as not to get any sky in the picture, but they didn’t merge properly. the result was two 3D models one above the over, and the lower one very strange and twisted.
If anyone knows the format needed to merge manual photos and set automated ones, that would help I think. Otherwise I was thinking of trying Dronelink to add more flight paths, but that costs money and don’t want to spend not knowing if it’ll fix my problem.
Just fly the drone manually (it doesn’t cost anything! ) for the extra photos, but make sure there are enough common areas with photos taken from higher up so that there are sufficient features in the overlaps.
Hi @MIles, I’m very glad you joined us and are asking interesting questions. We had a similar question recently here which might be informative:
I believe Gordon was trying to be more helpful and did read your question. We should assume that folks are showing up with best intentions, and respond appropriately.
Hello @Miles,
as @smathermather mentioned in the quoted post, Pix4Dcapture is a fantastic flight planner that keeps on getting better and better and is totally free.
When you download the app, you’ll have to register and log in, but once logged in, it did not ever bother me with anything again.
The app offers Double Grid Missions and Circular Missions while also having an option to let the camera face the center.
And instead of uploading to a Pix4D server I just take the images from the SSD and feed them to WebODM.
For a good 3D model I use several missions. First 2 double grid missions 45° shifted towards each other, another double grid at a higher altitude and another 2 or 3 circular missions at different altitudes and increasing camera angles. For all missions the camera faces center, except the double grid at higher altitude. That way also the surrounding can be reconstructed nicely.
That’s a lot of images, several flights and sometimes I have to restart the app on my mobile (Samsung phone with 6Gb RAM) since the app slows down with too many images.
Once I also did the flights on 2 different days since it took me some time to compute the model to find out I was missing some difficult faces. Which also works fine.
But difficult faces are usually underneath overhangs. That’s where low flying circular missions come in handy, just watch them trees
Thank you very much for your reply @shiva.
I tired Pix4Dcapture but had some permissions problems with the plugin ctrl+DJI that is required to run the app. I’ll definitely take another look and try to spend some time solving that. Perhaps I gave up on it too easily.
This is great information and very helpful to anyone trying to make 3D models. Can I ask a couple of clarifying questions pls?
When you say 45degrees shifted towards each other, do you mean if one is angled towards the west then the other will be towards the east?
The higher altitude double grid I’m assuming is using a straight down oblique birds eye view, is that correct?
Can I also ask about the circular missions, what is the lowest altitude you’ll fly? And, do you try to not include any sky in the picture as I heard this makes it hard for the program to match the photos on processing?
Lastly, I have also heard that sometimes too many photos can make the model worse, is that true or do you think the more the merrier?
yes there are some pitfalls when using Pix4Dcapture. I usually close all apps, then connect my phone to the controller and my phone will ask me which app to grant permissions.
Since I put that on manual, I can always choose and when wanting to fly a Pix4Dcapture mission, I choose Ctrl+DJI and then will have to reconfirm once or twice a permission pop-up to really allow Ctrl+DJI access to the remote connected via USB.
Once Ctrl+DJI has started and reports “connected”, only then I start Pix4Dcapture.
Yes for example. Here a screenshot of how that could look like:
When there are many trees near the building or the facade that I want to model, then manual flight is sometimes the only option. My drone is too valuable for me to have it fly into a tree.
Looking at the model you posted, I would even try to just manually fly a couple of times left and right in front of the veranda while pointing the camera at it and add those images to the process. The rest of your model looks great, only the veranda needs some more data and in front of the veranda there are also no trees.
But again, there are practical and physical limits to it. You will probably still have holes in the model, for sure the ceiling of the overhang will have no data.
Except you add pictures you take walking your drone
Or using another camera to take pictures, but that can be tricky to feed properly into the process.
Which brings us to the last question:
Yes, there is definitely a sweet spot of how much images you want to feed into the process. As much as I would like to explain it, I guess you’ll just have to try out
There are many helpful functions in WebODM where you can see the camera positions for example which can help you to figure out why a certain place maybe looks bad or which images ODM has skipped altogether.
That is why I am often very curious about the report PDF, it’s often a treasure trove for learning.
Though as mentioned, just try it out, play with some functions and a couple of hundred hours into the process I am sure you will find which knob you have to turn to tune your result
It can also go faster, depending on your perfectionism and dedication to the process.
Hope this helps and you are having fun playing with your drone and WebODM!
@shiva Again you’ve been super helpful, I really appreciate it.
Regarding to quality reports; what program do you use to read them as I don’t have anything installed on my PC that wants to open them in a readable way?
Hola @Miles,
with the quality report I mean the PDF report you can find the “odm_report” folder when downloading all assets from WebODM.
If there is no report.pdf in the “odm_report” folder, the shots.geojson should always be in there. I use QGIS or GPSPrune to view that. Shows locations of all images.
But the report.pdf contains much more info. Btw I am using linux. I think QGIS works also on windows and GPSPrune is written in Java, so should run on any system with a JRE.
Another help with 3D models is the “View 3D Model” function in WebODM.
In the left menu you’ll find “Cameras” where you can activate “Show Cameras”.
Looks like this:
Not sure if one can make it out in the screenshot, but you can see 2 double grid missions at slightly different altitudes and some orbital missions below.
The blue squares have little orange circles indicating the direction of the camera.
I had a few computations where a bunch of images was completely out of place.
Just click on a square and you can see the actual image.
When you find images out of place, you can either try tweaking your settings like GPS-accuracy or just remove them from the computation.
And, just checking on myself what I said in the last post: I clearly have some sky in the pictures and No, it did not mess up the process
) for the extra photos, but make sure there are enough common areas with photos taken from higher up so that there are sufficient features in the overlaps.
