Need Suggestions: Reducing Error in GCP for Sloped Terrain

[lupionpe]

Hi everyone,

I’m struggling with a challenging issue and could really use your expertise. I’ve been working on a project with a DJI Mavic Pro and the ODM Engine 3.3.3 (GPU). Despite trying various processing parameters, I keep getting errors above 2m in my Ground Control Points (GCPs). I’m looking for suggestions on how to improve this.

Exemple on GCP deviation

Quality Report

Flight Details:

• Equipment: DJI Mavic Pro
• ODM Engine: 3.3.3 (GPU)

Flight Plan:

I did a crosshatch flight, each direction in a different set-up:

• Set-Up 1:

  • Altitude: 75m (following terrain)
  • Front Overlap: 85%
  • Side Slap: 75%
  • Gimbal Angle: 85°
  • Perimeter Flight at 45°
  • Speed: 4m/s

• Set-Up 2:

  • Altitude: 80m (following terrain)
  • Front Overlap: 85%
  • Side Slap: 75%
  • Gimbal Angle: 85°
  • Speed: 4m/s

Processing Attempts:

I’ve tried processing with these parameters:

1. auto-boundary: true, crop: 10, dem-resolution: 3, dsm: true, dtm: true, feature-quality: ultra, gps-accuracy: 0.05, matcher-neighbors: 32, orthophoto-resolution: 2, pc-classify: true, pc-quality: high, pc-rectify: true, pc-sample: 0.01, rolling-shutter: true, use-3dmesh: true
   

2. auto-boundary: true, crop: 10, dem-resolution: 3, dsm: true, dtm: true, feature-quality: ultra, gps-accuracy: 0.05, orthophoto-resolution: 2, pc-classify: true, pc-quality: high, pc-rectify: true, pc-sample: 0.01, rolling-shutter: true, use-3dmesh: true
   

Additional Resources:

• Link to the images
• Link to the GCP file (Datum: SIRGAS 2000)
• Quality report

GCP Distribution

Any advice or suggestions on how to reduce the error in GCP for sloped terrain would be incredibly appreciated. Thanks in advance for your help!

Cheers,
Lupion

[Saijin_Naib]

Can you try on Defaults without GPU to compare?

[luc23]

How were the GCP’s coordinates taken?

[Jason_Owsley]

I found that using orb instead of sift improved my reported accuracy to centimeter level.

[lupionpe]

I collected with an RKT (Polaris S100), using NTRP.

In the pictures I used arucos to find it.

[lupionpe]

I tried, did not improve the results.

[lupionpe]

I will be trying this tonight, I let you know if there is an improvement.

[lupionpe]

I tried, it actually reduce the error, but the final products are worst.

Parameters:

auto-boundary: true, crop: 10, dem-resolution: 3, dsm: true, dtm: true, feature-quality: ultra, feature-type: orb, gps-accuracy: 0.05, orthophoto-resolution: 2, pc-classify: true, pc-quality: high, pc-rectify: true, pc-sample: 0.01, rerun-from: dataset, rolling-shutter: true, use-3dmesh: true

Any suggestions?
Maybe use matching-neigbors: 32?

[luc23]

@lupionpe

Polaris S100

Could this not be an issue with your GCP’s rather than your ODM settings?

Are you sure your RTK survey coordinate system settings are configured correctly and you are including the correct EPSG or Proj string into your GCP file?

[lupionpe]

I am doing some further investigation here to make sure this is not the issue.
Regarding the data collection and tolerances I am positive it’s not the problem, but I will test today converting the coordinate system to WG84 and make sure everything is in the correct datum.