Forum

Author Topic: "Align Chunks" detailed errors  (Read 3068 times)

Yoann Courtois

  • Sr. Member
  • ****
  • Posts: 316
  • Engineer in Geodesy, Cartography and Surveying
    • View Profile
"Align Chunks" detailed errors
« on: March 06, 2017, 03:44:45 PM »
Hi everyone !

I have asked some informations about this topic somewhere else, but without any answer I propose a new topic here.

I would like to use more often the "Align Chunks" tool but for the moment, the lack of alignment detailed errors restrict its possibilities.

My objectives:
- Firstly, I have created a global 3D model of an area (2000m2) which is georeferenced on about 10 GCPs (GNSS points). => subcentimeter accuracy
- Then, I have created small chunks (several m2) on parts of the same area (Holes dug after the creation of the global model.
- Those small chunks don't include GCPs used at the first step. But they present enough covering area with the main model (around holes)
- I have used "Align Chunks" tool to georeference my holes in order to make mesurements.

=> The big problem is that I don't have any alignment errors at the end of the chunks alignement. So I cannot have any idea on the accuracy of the alignement and then of the georeferencement of my holes.
=> The only thing I found is a value in the console which talk about "average projections" (4.31855 in my exampl). What does that mean ? pixels ? but which pixels: main model of aligned small chunks ?

I would like to know if it would be possible to have more. This is the biggest "black box" remaining in all the calculation process.

Thanks for your anwers

Code: [Select]
[...]
2017-03-06 11:58:33 photo 166: 19132 points
2017-03-06 11:58:34 photo 167: 20715 points
2017-03-06 11:58:34 points detected in 84.7 sec
2017-03-06 11:58:34 Matching points...
2017-03-06 12:08:04 103571 matches found in 569.623 sec
2017-03-06 12:08:04 matches combined in 0.026 sec
2017-03-06 12:08:04 setting point indices... 23871 done in 0.022 sec
2017-03-06 12:08:04 generated 23871 tie points, 4.31855 average projections
2017-03-06 12:08:04 removed 37096 multiple indices
2017-03-06 12:08:04 removed 11 tracks
2017-03-06 12:08:04 finished matching in 654.392 sec
2017-03-06 12:08:04 adding 953 points, 31 far (10 threshold), 18 inaccurate, 99 invisible, 279 weak
2017-03-06 12:08:04 adding 696 points, 14 far (10 threshold), 0 inaccurate, 14 invisible, 34 weak
2017-03-06 12:08:04 Aligning groups by 44 points
2017-03-06 12:08:04 iteration 0: 16 points, 0.0178772 error
2017-03-06 12:08:04 iteration 1: 16 points, 0.0178772 error
2017-03-06 12:08:04 iteration 2: 16 points, 0.0178772 error
2017-03-06 12:08:04 iteration 3: 16 points, 0.0178772 error
2017-03-06 12:08:04 iteration 4: 16 points, 0.0178772 error
2017-03-06 12:08:04 Finished processing in 654.48 sec (exit code 1)
2017-03-06 12:09:52 Generating report...
2017-03-06 12:10:11 Analyzing chunk...
2017-03-06 12:10:11 Finished processing in 20.047 sec (exit code 1)
2017-03-06 12:12:10 Generating report...
2017-03-06 12:12:14 Analyzing chunk...
2017-03-06 12:12:14 Finished processing in 6.955 sec (exit code 1)
>>>
« Last Edit: March 07, 2017, 11:09:42 AM by Yoann C. »
--
Yoann COURTOIS
R&D Engineer in photogrammetric process and mobile application
Lyon, FRANCE
--

Yoann Courtois

  • Sr. Member
  • ****
  • Posts: 316
  • Engineer in Geodesy, Cartography and Surveying
    • View Profile
Re: "Align Chunks" detailed errors
« Reply #1 on: March 14, 2017, 01:35:08 PM »
UP !

I got one chunk alignment (point based) which has worked (manual controls give me 2-3cm error which is acceptable regarding the survey process)

As I would like to repeat it with more confidence, I would like to know if somebody can help me to dissect this following console (given during the chunk alignment).
Here is what, to my mind, I understand:

- After matching between the two chunks, 1 100 898 points were found.             => OK
- Then matching are combined and, as there is many matching mistakes, only 420 723 points were kept. (Average projections 2.9 pixels, which represent 12mm on my project)       => OK
- removed 100 487 multiple indices         => what does it mean ?
- removed 81 tracks                                 => What's a track ?
- adding 13 716 and 11 872 points          => From ?
- But then the black box came... 420 000 points - 100 000 should give me around 320 000 points. But there is "Aligning groups by 1003 points" !  => Please where are the 319 000 others ???
- Then alignement iterations talk about 100 points...One more time divided by 10
                                                                     => What is the signification of all of that ?

Thanks for your help !



Code: [Select]
[...]
2017-03-14 08:27:35 photo 425: 39943 points
2017-03-14 08:27:36 photo 426: 39998 points
2017-03-14 08:27:37 photo 427: 39964 points
2017-03-14 08:27:37 points detected in 400.841 sec
2017-03-14 08:27:37 Matching points...
2017-03-14 10:45:50 1100898 matches found in 8292.62 sec
2017-03-14 10:45:50 matches combined in 0.146 sec
2017-03-14 10:45:50 finished matching in 8693.62 sec
2017-03-14 10:45:50 setting point indices... 420723 done in 0.14 sec
2017-03-14 10:45:50 generated 420723 tie points, 2.92693 average projections
2017-03-14 10:45:50 removed 100487 multiple indices
2017-03-14 10:45:50 removed 81 tracks
2017-03-14 10:45:50 adding 13716 points, 335 far (10 threshold), 167 inaccurate, 760 invisible, 1488 weak
2017-03-14 10:45:51 adding 11872 points, 170 far (10 threshold), 98 inaccurate, 290 invisible, 322 weak
2017-03-14 10:45:51 Aligning groups by 1003 points
2017-03-14 10:45:51 iteration 0: 127 points, 0.0258083 error
2017-03-14 10:45:51 iteration 1: 117 points, 0.0262112 error
2017-03-14 10:45:51 iteration 2: 114 points, 0.0263664 error
2017-03-14 10:45:51 iteration 3: 112 points, 0.0264231 error
2017-03-14 10:45:51 iteration 4: 111 points, 0.0264425 error
2017-03-14 10:45:51 Saving project...
2017-03-14 10:45:51 saved project in 0.124 sec
2017-03-14 10:45:51 Finished processing in 8694.93 sec (exit code 1)
2017-03-14 10:46:28 SaveProject
2017-03-14 10:46:28 Saving project...
2017-03-14 10:46:28 saved project in 0.036 sec
2017-03-14 10:46:28 Finished processing in 0.037 sec (exit code 1)
« Last Edit: March 14, 2017, 01:36:42 PM by Yoann C. »
--
Yoann COURTOIS
R&D Engineer in photogrammetric process and mobile application
Lyon, FRANCE
--

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14816
    • View Profile
Re: "Align Chunks" detailed errors
« Reply #2 on: March 14, 2017, 02:43:25 PM »
Hello Yoann,

I can suggest the following approach, if you need to align the smaller chunks overlapping with the large model for the areas that were modified since the original (large model) capture:
- you can place markers in the common points (in every pair of main chunk and smaller chunk) with the same labels,
- get estimated coordinates for those markers in the large chunk,
- put those estimated coordinates to the source values tab of the smaller chunk and uncheck the markers,
- align using point based or marker based approach using large chunk as a reference chunk,
- chunk the error values on the markers in the smaller chunks.

Such approach should give you some quantitative  estimation of the chunk alignment results.

As for the values from the log, not actually sure, if they would give you any information you can use. However, as for the "missing" 319 000 points, the explanation is the following: when you are running point based chunk alignment operations all the photos are being matched again. The found matching points that are related to the images from the single chunk only are then filtered, so you've got a huge amount of tie points filtered out by that condition. After than some of the matching points corresponding to the images from the different chunks may be also considered invalid, so the final number of the tie points used would be lower.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Yoann Courtois

  • Sr. Member
  • ****
  • Posts: 316
  • Engineer in Geodesy, Cartography and Surveying
    • View Profile
Re: "Align Chunks" detailed errors
« Reply #3 on: March 14, 2017, 03:41:29 PM »
Hello Alexey,

Indeed, it is exactly the protocol I used to get quantitative estimation of the chunk alignment.
However, I have no solution if it didn't work correctly (experiences during which I got 10 cm errors on created check points). The only conceivable solution to correct it could be using those markers created for control, by checking them in the reference panel and then update the model. The chunk alignment would be then only based on those markers. But my objective is to skip this stage and create a workflow only based on SIFT matching between chunks.
Do you imply that you may don't know how to get residues of that tool either ?

About the missing matches, I have to thank you about your answer that I did't thing about that.
However, which stage is correponding with thoses points delected because they linked pictures from the same chunk ? In this exampl I got 300 906 tie points on the main chunk and 66 511 on the small one, which give 367 417 points in total.
So those points might be detected during the chunk alignment and then deleted, but this number isn't even close from my "missing points"... Then I would ask you more about that !  :)
--
Yoann COURTOIS
R&D Engineer in photogrammetric process and mobile application
Lyon, FRANCE
--