Looking for ideas on automating orthomosaic quality checks

Deine

Hi everyone,

I'm currently working on an automating my processing pipeline using WebODM and had a question regarding orthomosaic quality assessment.

At the moment, quality inspection is still a manual process where an operator visually checks the orthomosaic for issues such as:

  • Ghosting or duplicated objects
  • Distortions or warped features
  • Blurred regions
  • Visible seamline artifacts
  • Missing or incomplete areas

I'm wondering if anyone has explored automating this process.

Specifically, I'm considering two approaches:

  1. Using the ODM Quality Report (reprojection error, reconstruction statistics, feature counts, GPS/GCP errors, etc.) to infer the quality of the final orthomosaic.
  1. Using computer vision/image processing to analyze the generated orthomosaic directly and detect visual artifacts such as ghosting, blur, seamlines, and distortions.

My question is:

  • Which approach do you think would be more reliable for automatically identifying orthomosaic quality issues?
  • Have you seen any existing tools, research, or projects that perform automated visual quality assessment on WebODM/OpenDroneMap outputs?
  • Are there other intermediate products or metrics generated by ODM that would be useful for detecting localized distortions automatically?

Ultimately, I'd like to automatically flag orthomosaics (or specific regions within them) that require manual review instead of having someone inspect every orthomosaic manually.

I'd appreciate any suggestions or experiences you can share. Thanks!

2 Replies

pierotofy Best answer

A very challenging problem; I think the only robust method would involve some sort of vision model, or an ensemble of models, trained on thousands (or dozen of thousands) of annotated orthophoto tiles, but you can imagine the difficulty of annotating and generating such a dataset. Keep us posted with your findings?

Deine OP

Thanks for your insight sir! I completely agree, the biggest challenge would be building and annotating a large dataset with different types of orthomosaic artifacts.

I also think a vision-based approach is probably the best direction for automating quality assessment. While ODM's quality report provides useful reconstruction statistics, it doesn't always reflect the visual issues that operators look for in the final orthomosaic. I'll definitely keep the community posted if I make progress. Thanks!

Need more help?

This discussion happened in the WebODM community Discord. Join and ask your own question.

Join View Original Thread

Related questions

This page mirrors a public discussion from the WebODM Discord. How to request a removal.