Parameter settings for low-lying barrier island environments

Hi everyone,

We’ve been testing the use of ODM to process DSM from fixed wing drone imagery on a low-lying barrier island environment. Images were collected with a fixed wing ebee drone equipped with RGB camera. We are running ODM through docker on a linux box and everything seems smooth there.

The images are of an undeveloped island with elevations ranging from 0 (shoreline) to about 3-4m dune heights into a somewhat vegetated back barrier but it is fairly sparse. We do not have visual GCPs (no fixed features on the island) but we do have a few on-the-ground survey points taken with RTK-GPS a few days before/after the drone flight that we are using for verification. Only 8 points in two lines from shoreline to the dune so they aren’t well-distributed. We also have a previous DSM for two of the flights created with pix4D that we are comparing to for reference.

We first used ODM with defaults but then (one-by-one) changed the following based on looking through this community:

ignore-gsd: true
orthophoto-resolution: 3
dem-resolution: 3
camera-lens: brown

Those settings gave us our lowest error but our elevations in the DSM still have a mean offset of 16 cm from the sampled points and the pix4D DSM. My next thought was to test the sensitivity to the smrf parameters but I was curious if anyone has worked in these types of environments and has a recommendation for optimal settings. I searched around but may have missed previous posts.

I put a small example of the orthophoto here: Dropbox - masonboro_orthophoto.PNG - Simplify your life

and a difference map of odm-pix4D here: Dropbox - odm_minus_pix4d.PNG - Simplify your life

We aren’t worried about waterways behind the island, our focus is on just the main part of the island. I also don’t expect the differences with pix4D to be zero everywhere – each program has their advantages but we may not have access to pix4D moving forward and I want to be confident I am getting reasonable DSMs with ODM.

Sorry for the long-winded post, any suggestions appreciated!

2 Likes

Welcome!

Cool study, and very nice comparing the two.

Not sure what I’d recommend at the moment other than maybe constraining the gps-accuracy flag a bit if you’re confident it is better than 10m on average to see if that might tighten the reconstruction up a bit.

1 Like

Good thought! The drone is equipped with RTK-GPS which the manufacturer says is 3cm horizontal accuracy. We’ll lower the flag to 1m which is a big jump from default but it will give us an initial indication of how sensitive our reconstruction is to it. Thanks!

1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.

Just keep in mind with fixed wing drones that knowing the real position at velocity is tricky. Any accuracy claims are super tricky, because of glide speed vs. GPS update. If there aren’t good IMU corrections, most fixed wing sellers are over-stating their per image accuracy.

2 Likes