The calculated point cloud model is a mess

Hello,
We have created a mission to fly a vertical cell tower with our Phantom 4 Pro without any ground control points, just to be able to “try” how the model looks.
We have flown the mission using ground station pro and collected about 680 images flying around the 200ft cell tower.
This is what we got so far. Is there anyway to get a useful output out of this flight?

tower2_report.pdf (1.1 MB)

Here is a sample collected photo resized by 400%


I’d call an overall quality good.
What are we doing wrong?

Hi Adam,

When it comes to capturing images of a tower, there are three aspects you need to consider:

  1. The zoom level has to be large enough in order to capture more details of the antenna.
  2. Reduce the proportion of background elements, which relates to the zoom level mentioned at point 1. What is the radius of the mission that you flew?
  3. Ensure that your dataset has sufficient frontal and side image overlap: How to verify that there is enough overlap between the images.

The Phantom 4 Pro drone should be able to generate optimal tower reconstructions. I would strongly recommend reading this article: How to map and measure Pole and Tower Structures.

P.S. Regarding the first screenshot, you need to uncheck the Cameras, they are obscuring the point cloud. Also check the Point Clouds, as shown below:

Cheers,
Teodora

Thank you Theodora,
I was able to uncheck the cameras and the model has shown up, but the quality is not good.

I have created few folders online with our photos, would you mind having a look and telling us what is wrong with our data?

  1. https://1drv.ms/u/s!AlwmeFgzf0XeqQ20Pc_8ksTHWaFy?e=BsX06W

  2. https://1drv.ms/u/s!AlwmeFgzf0XerkibkyHOkL8R3dOu?e=k9T5IS

  3. https://1drv.ms/u/s!AlwmeFgzf0Xerkn1xM-SjxcuVjDR?e=wA1dni

We’d appreciate any pointers

I had a little extra time on my hand so I downloaded your dataset and processed it.
I managed to get a pretty decent point cloud out of it.

2 Likes

Jaakko,
Thank you very much for your help and using your resources to try to evaluate our images . Can you share your work flow? Did it require a lot of cleaning?
My “monstrum” looks like it’s generated by Commodore 64 and we are using the same set of photos.

Thanks to Jaakko’s kind assistance with the workflow I was able to generate point cloud that is similar to his result. Since then, I was able to process yet another tower using this template with a good result.

We collected the data with Phantom 4 Pro, no manual GCPs, and used Ground Station Pro to generate a flight plan with good overlap but, I think we were flying around too far. I think 2 m (7ft) closer would be perfect.

Here are my screenshots. Make sure to uncheck the “cameras” and “point size” of the point cloud has to be the smallest setting or very close to it.

As to workflow, here is one that works for me:

  1. Import the photos.
  2. Do the step 1 of processing.
  3. Add Manual Tie Points all around the tower - more IS better but do at least 20 - each with minimum of 10 marks.
  4. Process the settings (step 2) according to this screenshot below. This worked the best for me. Tipped again by Jaakko

1 Like

Hi.
I just got a pix4d licence. I tried to rebuild the same workflow but still got not good quality as you. I made 20 manual tie points and proceed the point cloud with the same settings as on picture above. What could be wrong?

![Huomautus 2020-04-08 083010|688x500]

As I am a new user I cant put much photos per time. So the others are below.

And settings

Make sure to scale your point size down (adjust with the slider under “display properties”)
And also scale down the Automatic tie points down (with the slider)
If you think that some part of the tower is still not reconstructed good in a specific spot, add more MTPs there…

Hi Jaakko. Did you cut parts of the pont cloud close to the Tower? Cause it looks like so clear. How many MTps you aded to this part that shown on the Picture ? Thanks

Nice tip. Thanks!
Now it looks a bit better

Hi there

Seems that you figured it out :slight_smile:

Not totally. Because your Picture looks much sharper. Thats why it is interesting to figure out how you got this result.

Add 10 more MTPs where you think model needs improvement (with at least -10 “marks” per MTP), do step 1 again - but I like to do “rematch and optimize”, and finally do step 2. More MTPs more precision…

1 Like

Why do you use “rematch and optimize” instead of just reoptimize? I read the difference between them and se that rematch and optimize can be used if some operations was made with photos. Is it right?