How make elements (e.g. car) more accurate

What settings do you recommend to make the body of the car more accurate?
Notice that it expands more towards the ground.

This made by default settings.

Thx !

2 Likes

What does the pointcloud look like? Let’s start there :slight_smile:

3 Likes

Hi,
thx rightly. Okay, I’m showing you a cloud of points. What next step do you suggest?



1 Like

Can you tell us more about your processing settings and your data (flight plan overlap/sidelap, gimbal angle, resolution in MP)?

It looks like we need to increase the quality of the pointcloud first.

I had to make a new render, the previous one is on a damaged computer in the service.

What you see is generated from the default settings

image


image
image

1 Like

What should I make to get better result?

1 Like

more lower altitude side angles so you can actually get images UNDER the car. Unless the angle is small enough, from above, it can’t really tell that the car isn’t connected to the ground.

2 Likes

Raise pc-quality, feature-quality, min-num-features to start. Go up by one quality level and raise numeric parameters by about 25% to start.

1 Like

This have you mean ?

image
image
image

1 Like

Ok. I Try so next misssion.

1 Like

Ok. I did as you said. It’s better. That’s true. Is there anything else that could be improved on this project?


image

1 Like

You can increase things again like last time, maybe add pc-filter 0.

Do you have a good idea of the mean GPS accuracy? Sometimes setting that or 2x that as GPS-accuracy can help a bit.

I’m following you in the right direction. The effects are getting better. I’m beginning to understand. these settings.
Here I set PC-feature =ULTRA (previously it was High) …and set PC-filter =0

The accuracy of details has increased noticeably. For example, already visible mirrors appeared. The shape of the main body of the vehicle is also correct.

However, there are these points that deviate from the norm. Do you want to go back and change the PC-filter setting? …or another setting?


3 Likes

That’s almost certainly a result of us disabling pc-filter.

You might find the default 2.5 appropriate, or you may want to relax it to 5 or 10, though the difference will be minimal.

1 Like

Thank you very much for all these valuable comments.
I want to know more about how and when GPS accuracy can be used?
I have two drones that I do raid:

  • DJI Mavic 2 Zoom, which saves GPS coordinates in EXIF ​​information to photos, but is without RTK

  • DJI Mavic 3 Enterprise, which also saves GPS coordinates to EXIF, but works with the RTK module enabled

Can you explain how manipulating the GPS accuracy settings in WebODM will work in both UAS cases?

I don’t remember which project I did. I have to check it out.

1 Like

I’m envious of your RTK ability. If I had that, that’s all I’d use. I fly the same mission over a long term construction site about once a month for documentation and comparison, as well as planning. The orthophotos never perfectly line up in QGIS because of my Air 2S’s crappy GPS (all consumer GPS is crappy, not particular to Air 2S). They’re all within 5-8 feet though, but still annoying when streets don’t line up, etc. I wish I had GCP’s at least, but gotta work with what I’ve got, which is just an Air 2S.

4 Likes

If you are interested in something, maybe I can help … if you want me to do some tests, I will gladly do it and I will also learn something on this occasion. I can leverage M3E RTK and WebODM and open a new thread in the community. Feel free to speak up if you have an idea.

Is the GPS in the Air 2s special compared to other consumer drones? Did I understand your statement ‘not particular to Air 2S’ correctly?

2 Likes

I was meaning the GPS in Air 2S is on par with other typical consumer drones, which is not very accurate for precise mapping. If I had an RTK capable drone, that’s all I would use for mapping just due to way better accuracy. RTK was designed for accuracy/mapping.

2 Likes

We normally recommend setting the GPS-accuracy to 2x your mean GPS error for the collection. This helps constrain the camera locations during reconstruction, which can help improve the accuracy of the reconstruction and its geolocation. Sometimes it can help improve completeness of reconstruction, but this is more a secondary benefit and consideration.

If you run with defaults and the camera locations in the report make sense and the reported GPS errors don’t look too bad you don’t have to tune it, but I believe it to be worth the effort to get a sense of what it is for your platforms and to use a value closer to that than our 10m.

1 Like

You might look at the align flag, which will align each of your datasets to a reference point cloud or elevation model.

I haven’t checked whether it is exposed in WebODM.

2 Likes