Suggestions for points far above ground in point cloud

Hi All,

I am running webODM on some drone images from a Sequoia camera.

Trying to solve problems one at a time. Since the dataset is 700 images (80% overlap over about 5 hectares @ 70 m altitude)

Any help in solving this kind of issue will help, if someone wants to help solve issues on this effort (RGB images from a sequoia camera) I can upload the dataset.


Thanks for any inputs!


1 Like

What drone is it mounted on?


Sequoia is mounted on a custom (small production run) quad rotor with pixhawk ardupilot flight controller. Camera is mounted on a 2 axis gimbal with a 5 degree forward-looking angle.

I have multiple data sets as I flew for RGB across the whole property for 70m and high resolution in the vines at 50m. (So I actually have two datasets for the vines at different altitudes if that is interesting)

I work specifically with vineyards and develop analytics from the outputs. I normally process with Pix4D but due to lots of issues, I am looking for better ways to build the data flow.

Thanks for any advice/comments.


1 Like

Nice! So you are probably getting better than 5m accuracy in your GPS (which was the underlying reason for the question). I would recommend updating:

./ down && ./ update && ./ restart 

And then use the new flag gps-accuracy and set it to 5 or even 3 if you are flying slowly or hovering at each point, you might even be able to set that to 1 meter.

1 Like

I should explain what this does: when we have difficult data, whether a low resolution sensor, lots of repeating features, or other issues, matching in our Structure from Motion model can be tricky. By specifying the accuracy of the data, we are able to constrain that process to ensure that matching between the images doesn’t move the position of the data by more than the expected error.

We have to be careful not to set this value lower than the real accuracy, but it can be really useful in cases like this.

1 Like

Is this analogous to this part of the workflow in Le produit Suisse?


(Edited for completeness)

1 Like

Thanks very much for your input.
I will give this a try.

Note though that I generally fly at about 8-9 m/s and get very good data at this speed.
I know that there may be some issues that derive out of the rolling shutter on the camera, but that should be individual warping of the image and blur.

The GPS accuracy for the UBLOX tends to be extremely good in general, and relative to it’s own reference it stays very very stable. (sub 0.3 m error over the course of a 20 minute flight) (drone will land within 0.3 of where it took off over 90% of the time, barring heavy wind)

I will try bringing the gps-accuracy down to 5 and see what it does.

I am curious though, do you think this is the issue why it is detecting point cloud values above the surface? It may be that at this point the drone caught a gust of wind and had a rapid movement / shift.

It is very likely that the camera maintained it’s 5 deg off nadir attitude, but it is possible that the drone jumped in altitude / orientation very quickly.

The images are snapped at 1 second intervals. There is also the case that this occurred close to the edge of the field in the over-fly area, where there are more images that average due to slowing, turning and accelerating.

I will follow your steps and see what I get.

Thanks again.

1 Like


That helped!!


I have a lot less big excursions from the field surface. There are still a couple still happening.

I am going to try 3m and re-run tonight.

New question:
I am also seeing a lot of field-mottling which is clearly coming because the camera choices are coming more from the off-boresight images. Note that I am flying with a 80% overlap and There should be a good nadir image of everything… but I am getting a lot of this square-patterned imagery.

In some ways this is actually higher quality than some ortho-imagery… but it definitely does not make for a good picture!

Any help on this kind of problem?

Note that normally I fly fields for agricultural purposes, but this farm has asked me for images to make a poster as well… and I am trying to get something nice from the RGB!

Thanks for your support!



1 Like