Forum

Author Topic: Non-Coded targets automatic detection  (Read 1498 times)

3DWinter

  • Full Member
  • ***
  • Posts: 103
    • View Profile
Non-Coded targets automatic detection
« on: September 20, 2022, 03:56:40 AM »
Hello,
The two examples of Non-coded targets (seen in the attachment) are black and white with a black edge.
I have a few questions:
1. Does the algorithm care for color?
2. I assume the black edge is not critical, or is it!?
3. What is the definition of "Tolerance"?
4. What is the definition of Maximum Residual (pix)?
Thanks

BenW22

  • Newbie
  • *
  • Posts: 49
    • View Profile
Re: Non-Coded targets automatic detection
« Reply #1 on: January 26, 2023, 06:53:50 AM »
Hi All,

I think 3DWinter has asked dome good questions. Would anyone have any answers?

I too am looking at making some non-coded targets and am looking for details on what is best. I'm flying a DJI m300 with Zenmuse P1 (35mm and 50mm lens). Could anyone tell me if they have had success with non-coded targets? If so what pattern, colour, size etc did use use. Also what GSD, flight height, camera where you using.

Any help appreciated.

Regards
Ben

dpitman

  • Full Member
  • ***
  • Posts: 236
    • View Profile
Re: Non-Coded targets automatic detection
« Reply #2 on: January 27, 2023, 02:17:41 AM »
Depending on the nature of the project, I wouldn't waste your time with this tool in it's current condition.  It will detect targets, and miss others. It will give you many more false hits than actual hits.  The colors of the targets are not critical.  They just need to be contrasted in the expected shape. 

For example, on this project there were 3 actual markers.  The tool was set to it's most restrictive tolorance and residual, (1,1), and the tool found 2 of the 3 targets and found 5 non-targets.

Maybe if the tool was shown an example of a target in the dialog box, then it could be more effective. Perhaps, if it was looking for blue pixels as in my targets, it could more easily eliminate outliers.

« Last Edit: January 27, 2023, 06:59:48 AM by dpitman »

Kiesel

  • Sr. Member
  • ****
  • Posts: 332
    • View Profile
Re: Non-Coded targets automatic detection
« Reply #3 on: January 29, 2023, 05:34:41 PM »
For good detection, you need high-contrast and big enough targets without glare in your photos. The success of detecting none coded targets (cross type) also depends on the surrounding, in sand and grass surroundings we have very good detection rates but in stone surroundings, it could be that stones and their shadows are falsely detected as none coded targets because they look similar to the algorithm.

To my understanding, with the 'Tolerance' value you can choose how much the target can differ from the ideal pattern for example in oblique photos, and with the 'Maximum residual' how large the residual of the detected center is.

Regards

Kiesel
« Last Edit: May 09, 2023, 07:59:11 PM by Kiesel »

BenW22

  • Newbie
  • *
  • Posts: 49
    • View Profile
Re: Non-Coded targets automatic detection
« Reply #4 on: May 01, 2023, 06:04:37 AM »
Thank you both for your responses.

Cheers
Ben