Question about Image quality

Any idea what causes the aberrations please? And what do I need to do to avoid them?

Many Thanks

Flight details here:

image

Image detail here:

image

What elevation and overlaps did you use?

Did you process this map in Structures Mode? This has been happening allot since the new 3D interface which may mean that something changed in processing as well. I just did one yesterday that I took over 300 oblique images in an orbit pattern in addition to the map images where the point cloud looks pretty good, but the 3D model looks terrible.

If you processed in Terrain Mode it was probably not enough tie-points and you could try Structure Mode…

@Erika_Houseman @Kaitlin_Blando

Hi MIchael,

Thanks for your reply
Overlap settings:
image
Basic Flight settings
image

It’s like the stiching process has missed a batch of images all down one side.
Apart from a few blurry bits

Here is the flight plan


And the orthomosaic

I’m not happy with the finished article.And neither will the client. And it’s not the first time I have questioned the quality. My renewal is due in a couple of weeks and I’m not keen. It’s a lot of money for this level of quality

Another example from the same day, different roof flown at 70 ft, same overlap settings

image

Did you capture any oblique images? You should really use Structure flight and do two of them at different altitudes if you want a really crisp map and model. The reason the seams in the roof didn’t line up is because it looks like you flew too low. If your altitude was 40ft then you must take into account the height of the structure. I am guessing that building is 25ft at the peak so your altitude should be no lower than 75ft. You can do a lower flight to capture more detail, but in addition to the main flight.

Recommended:
Structure mode at 100ft AGL with 85% overlap.
Structure mode at 150ft AGL with 75% overlap. Rotate this second flight so that the path is 90 degrees to the first.

Secondary Recommendation is to fly the Auto-Mode Crosshatch at -60deg gimbal pitch. I recommend this as an alternative because this flight alone could have the same problems with the roof seams unless you fly a standard nadir map flight in addition. Also, if you use crosshatch make sure to extend your perimeter far enough beyond the subject so that the images coming out of the turn back towards the subject actually capture the structure.

Here’s an example of the map from yesterday that I just reran because I figured out that I had a couple of images in the sequence that were not meant to be in the map. Still, because of the intersecting braces and tiltwall joints the system had a hard time processing and it is garbage. Note how nice the crane is though. A test map for the DD engineers… I am also running a test with my Pix4D machine and will post that when it is done. Two days later, lol. :crazy_face:

Speak of the devil… The point cloud from Pix4D looks a little better than DD’s, mainly because it caught all of the panel braces, but… The 3D textured mesh failed!

Michael,

If you send me a link to the photos, I can give them a go with Metashape in order to provide another comparison.

Regards,
Terry.

Will do. Thanks for the offer!

So how did we get from here …'The easiest way…

To this

From the above email

"Recommended:
Structure mode at 100ft AGL with 85% overlap.
Structure mode at 150ft AGL with 75% overlap. Rotate this second flight so that the path is 90 degrees to the first.

Secondary Recommendation is to fly the Auto-Mode Crosshatch at -60deg gimbal pitch. I recommend this as an alternative because this flight alone could have the same problems with the roof seams unless you fly a standard nadir map flight in addition. Also, if you use crosshatch make sure to extend your perimeter far enough beyond the subject so that the images coming out of the turn back towards the subject actually capture the structure."

It doesn’t ‘say this on the tin’ when you buy into the product

I have checked back on earlier orthomosaics and it’s only recently I have seen this type of issue. I have also tried using the ‘location marker’ and it seems that in some cases the 4 images displayed don’t match with location in the image.

We still haven’t heard what processing mode was used?

There is this departure in the literature and what I am saying because those are the basics of creating a map. my recommendation is going to get you a better map, but also a good 3D model.

Bottom line though is that your map failed because your altitude was too low and there was not enough coverage. That is proven by the fact that the system had to search for images outside of the range of your location marker. The blank area was also on the very edge of your flight plan meaning that it got even less coverage. The way to ensure that it would have had coverage would be to make the flight plan a little larger, but whenever dealing with vertical faces oblique images will make the edges sharper and will also provide more images to fill in gaps so that holes like that aren’t created.

Hi Michael, thanks for your reply and time. This particular mission was photgraphed at 50’ AGL and I have notice that planning to fly lower than 50’ chucks up the message in Red 'Warning, Low Altitude. However, not to give up on a good mystery, I have rerun the upload in ‘terrain’ mode and mapped the image locations (blue dots) and done some further digging. The resultant ortho was a definite improvement, but it was the overlay of the blue dots on the ortho that revealed the possible reason why the quality is poor. Have a look at the overlay. It seems that the camera misses a beat, I would say every 6th or 7th image. Not surprising that that seems to be where the issues are. I can only think that the SDHC card write speed is causing this or the mission speed is too fast.

That makes sense. Terrain mode is a simpler processing and is focus more on flatter triangles. What does the 3D model look like? It’s possible that it could stitch from the roof of the building or some point higher on the wall back to the ground just to fill in the gap.

The missing/uneven image spacing has been a problem that I have brought up and has been discussed at length. We ended up breaking the discussion into two particular problems. Missing Images and Uneven Image Spacing. Feel free to checkout the threads below. It just so happened in your case (as has happened to me in the past) that two missing shots were exactly where you needed them most. One of them appearing to be coming out of a corner which is a new problem that has arisen in the last month.

Hi Paul,

Thanks for sharing your results. I’m James, product director here at DroneDeploy. The reason you are seeing these issues is due to a combination of your overlap and your altitude.

Overlap and alititude
First thing is to know that overlap is greatly diminished as you fly over objects that are high relative to your fight altitude (full help article):

So by taking off from the ground, and flying low over the roof, the effective overlap on the roof is greatly diminished.

The result, on your maps, is as follows:

These images are from our new accuracy report that is in early testing.
Where you see yellow or red, there will certainly be accuracy issues.
Anywhere you see green, there may be accuracy issues,
Blue should be well reconstructed.

Michael’s other point was about 3D coverage.
On the left is an image of a 3D model taken from the viewpoint of the drone in flight.
On the right is the same model from a view the drone never saw.

This is the difference between normal mapping and structures mode.

So how can you ensure you get good results?

Some options:

  1. If you are primarily flying roof maps, use the Precision Roof app
  2. For other map types, use the overlap optimizer

I hope those help.

1 Like

@MichaelL - the difference in points you’re seeing is because our depth map fusion algorithm is not currently optimised to pick out thin structures in the point cloud - we have to set it up to be robust for a broad range of input imagery. We’ll likely be using lower resolution depth maps than the ones you selected manually in Pix. We’re also about to double our point cloud density, so watch out for that in then next couple of weeks.

James

2 Likes

Exciting! Can’t wait to pull one down.

1 Like