Detection of incomplete orthophoto

Hi and thanks for your support, which I’ve already received.

Is it possible to detect from ODM outputs that the orthophoto is incomplete? E.g. I’d like to run processing with default parameters and when I decide (by output) that the orthophoto doesn’t cover the whole area I’d like to run processing again with more resource-consuming parameters. I’m doing it manually (Processing forest data with similar photos) I’d like to make it automatic using bash script or so.


That’s a really interesting problem space!

How will you know that the reconstruction is incomplete? Will you have an expected Area Of Interest boundary you’ll compare the reconstructed area to, and then decide if it is less than 90% coverage you’ll need to re-run?

Interesting question indeed… Pix4d has this feature, I think it’s marking the areas with red colour. But you will only see this after it has been running for a while.

If you are using a flight planner, this is what I was seeing when you don’t have enough coverage. Please note the lack of photographs taken. So I went back right after and made some pictures where the holes were:

1 Like

Isn’t that for cameras that don’t get calibrated? You can still have coverage/reconstruction in those areas, and vice-versa, you can have every camera calibrate and be part of the reconstruction and still be missing areas in the reconstruction :grimacing:

Tbh it was 2 months ago I had the trial version, not sure anymore. I remember the red coloured-cameras and in the end the mesh was incomplete :speak_no_evil: Ofc that was not pix4d’s fault but mine.
If I remember correctly I had the option to manually tie the pictures (?) but it was too much of a hassle.

After that experience I shoot rather too many pictures instead. Lessons learned.

1 Like

It could be as you said.

On the other hand, is there some parameter from SFM process which will tell us that there was relatively small amount of matched images? Orthophoto shouldn’t be complete in this case.

1 Like

There is a parameter num_candidate_image_pairs from project/opensfm/reports/reconstruction.json. I see that in proper reconstruction of orthophoto the number is twice the number of input photos. In other case, when reconstruction is poor, the number is close to the number of input photos. I’m not sure yet if it’s a general rule, but I’m investigating it.


Any photogrammetry software finds difficult to homogeneous photos included ODM.
So, personally, I recommend to increase min_num_features and matcher-neighbors.

Yes, it more resources consuming, But, I find it significant for better point cloud and orthophoto’s reconstruction

(AFAIK, got better orthophoto because orthophoto was based on point cloud)

1 Like

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