Pix4D Mavic 2 Processing Issues

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
61

Hello Alejandro,

This whole community topic refers to the wrong georeferencing when using Mavic 2 and that has been solved.

Please see the official comment for more information.

Your problem is different, your project is properly geolocated but you get a flat DSM.

I have looked at the Quality Report and it clearly shows what is going on:

As you can see, the difference between the initial and computed camera parameters is almost 100%.

If I continue reviewing the Quality Report, I see:

 

Due to this big difference, the computed camera has a focal length of 0.05 mm and that is the reason why you get a flat point cloud and DSM. Such a short focal length makes the whole project go wrong.

Before continuing with Step2 and Step3 is recommened to check whether something is wrong in Step1 as it is the case here.

When this happens (camera changing so much after computation), you should use “All prior” for camera optimization to force the software to not change the camera internal parameters too much.

I have downloaded your project and applied “All prior” and the project works perfectly as you can see from the following screenshots:

From the Quality Report, I can see that the ortho and DSM preview look correct:

 

A snapshot of the point cloud shows that it is not flat and looks also correct:

A snapshot of the DSM and Ortho also show that everything is also ok:

Answering your questions,  Mavic 2 is fully supported by Pix4D, both mapper and capture. I just processed your data (using Pix4Dmapper).

As for the other projects that you mention, the same issue is probably happening. Please check the Quality Report and see how much the internal camera parameters change and in case they change much, please apply “All prior” and run the process again.

If that is not the case (or the other projects still fail with “All prior”) please, do not hesitate to open a support ticket or a new community post and send us: p4d file, log file, and Quality Report.

Thank you very much.

Regards.

 

 

62

Hi Daniel,

I have this problem with ‘flat’ point clouds with all my projects flown with Mavic 2. I already found out how to solve it and always use the All Prior, it works fine for me. But it means that with this setting, the camera is not calibrated at all which could have some effect on the results. 

Is is something that will be solved by the Pix4D dev team so it will be possible to use the camera calibration when processing the Mavic 2 data?

63

Hi Nikola,

When you use “All Prior” the camera is calibrated, the only difference is that the software is forced to not change the internal parameters too much but they do change. You can see it in the Quaility Report.

This is a Mavic 2 project processed with “All prior” and you can see how the internal parameters change after Step1:

 

The only way to not calibrate the internal parameters is by selecting “None”:

If you select “None”, the internal camera parameters will not be touched at all and they will remain the same.

That is only recommended when using large format photogrammetric cameras which have a much more precise calibration than the ones used in drone photogrammetry.

Please also consider that in order to get the maximum accuracy when processing with no precise geotags (as with the Mavic 2), we recommend using GCPs.

Regards.

 

64

Hello.

I have tried the BR EXIFextracter and the image geolocation works well.

However the Camera Optimization is showing very large differences in the initial processing.(between 50 and 150% in separate flights) 

Has anyone else had this problem and knows a method to correct it?

Thanks.

65

Hi Darryn,
Yes, I have this problem with every project. Use All prior for camera optimisation, it works.

66

Thanks Nikola…it worked a treat.

67

I am now finding the final surfaces are very noisy and not uniform over relatively flat surfaces. There are 15 GCP’s used across the image to fix to a local coordinate system.

Any suggestions to how to improve the DSM and remove the “divots” ?

68

Hello Darryn,

As Nikola says, All prior will help you when the camera optimization shows large differences.

There is a message by me on March 20th explaining it in detail.

As for your problem, I would need to have a look at the Quality Report to analyze what is going on. Would you share it?

I would say that perhaps the area you are mapping has a very low texture. If that is the case, increasing the Image Scale to 1/2 or 1/4 instead of 1 (default value) in Step 1 might help.

Regards.

 

69

Hi Daniel

I have sent the Quality Report via the Support page. (too large for here as a pdf)

I have followed your suggestions however the surface still appears noisy in the sparse DSM preview.

 

70

Hi Darryn,

The DSM preview is generated with the automatic tie points only which do not have much density compared to the point cloud so it is possible that it is not sharp.

You should look at the final point cloud or DSM. They should be clean and sharp.

Regards.

71

Ok. I followed the above suggestion and the Quality report passed the initial Quality Check requirements. Continuing processing the issue appears that there are differing surfaces created within the final results, seemingly independent of the surrounding surface. The point cloud is far worse than the triangle mesh shown above.

Depending on where I click the surface to create a boundary for volume calculation, I am getting variances of up to 10% for the same stockpile, which is difficult to report.

Is there a method for removing outlier points to create a more uniform surface?

72

Hello Darryn,

There are several things that you can do to improve the point cloud (which will result in a better DSM). Please have a look at this article: How to remove the noise from the point cloud

As for the DSM, you could also add surfaces to improve it as explained here: How to improve the DSM

Or you can also change the interpolation method and use:

 

I hope this helps you.

 

73

Hello,

Is there any update with the Mavic Air and the shifting bug?

I am using the 4.5.0 app version, and the 4.3.33 desktop mapper version.

 

Regards

Arik

74

Hi Arik, 

The new firmware v01.00.04.00 for Mavic 2 Pro/Zoom has been released on 2019.04.17. DJI claims to solve the issue of geotags in that upgrade.

I would also suggest you have a look at the official comment that has been made by our technical support on community post about the Mavic 2 Pro geotag issues. It will suggest two possible workarounds to apply about the issue you are experiencing. You could also upgrade to Pix4Dmapper 4.4.10, the issue is solved in this version.

75

Thank you for the response. I was asking about the Mavic AIR.
Could you please refer to the bug status with this drone?

Did anybody experience similar issues with it?

76

Hi Arik, 

Its the same issue as Mavic 2 Pro/ Zoom. You should be able to deal with it using the above recommendations. 

77

EVERYONE = I just noticed the same thing.  EASY FIX = Run your imagery through Lightroom and Export.  This gives you the chance to both boost shadows and reduce highlights uniformly or just leave as is and the resulting JPEGs have the right location.  My frames snapped perfectly into place after doing this every time.  

Hope this helps :slight_smile:

 

  • Jon
78

Hey all,

I have been trying to get our Mavic Pro 2 to correctly process for a few months now. While I have had, generally good results using some of the methods outlined in this post, I wanted to add that the new firmware really seems to make a difference.

I had a project that I flew prior to the firmware update. I couldn’t quite get the output to ‘lock’ into the panels (one panel was off vertically by about 3’ and some of the other panels were not registering by 0.2’ - 0.4’ horizontally). I tried most of the methods listed in this post and was pretty well stumped. While I was getting data from the output, in one area (namely around the area where the panel wouldn’t register properly vertically) that differed from the field verified data (hand collected QC survey-grade GPS points). I was not satisfied with these results, as the final surface and contours will be used for engineering/design purposes.

Last week, I decided to try and re-fly the project with the updated firmware. What a difference! I didn’t have to re-run the initial processing with updated camera specifications or remove/change the EXIF data. Every GCP that I had used in the prior flight registered with very tight (sub 0.01’) error accuracy numbers (vertically and horizontally) where, prior to the firmware update, the registration error varied from 0.02’ - 0.4’ horizontally.

When I compare my field QC shots against the Pix4D generated results, the contours and resulting surfaces compare very nicely to the field QC shots. The results that I am getting, thus far, (I have some massaging to do to the point cloud in Global Mapper) give me much better confidence that I can produce a high-quality surface model/contours for engineering/design purposes without needing a large quantity of field QC shots needing to be collected.

I had a decent QC collection process figured out for the Phantom 4 that I hoped would translate into the Mavic easily, but prior to the firmware update, the Mavic needed far more QC field coverage than I thought it should - I almost felt like I was just doubling the process of collecting topography, which undercuts the UAS data effectiveness and productivity yields. Now I think my older field QC process (i.e. major breaklines, random spot shots, etc) will better reflect the improved productivity from the updated Mavic firmware and generate more accurate surface models and contours.

I finally am gaining confidence that the Mavic 2 Pro will yield results similar to the Phantom 4 Pro, which is fortunate, as the Phantom 4 models are going to be phased out in favor of the Mavic models.

I should note that I’m using Pix4D 4.4.10, so I can’t quite speak to results in the prior Pix4D versions. Although I wouldn’t be surprised to see many of the abilities in 4.4.10 translate into future builds of Pix4D for general release.

79

Hello All,

I was experiencing the same problems on my DJI MP2 for the processing. 

I know there is a fix in the latest version, but as I am a Mac user, we don’t have it. Extracting XMP on Mac is a also difficult for a beginner like me. 

Anyway, I just find a quick fix by luck through Lightroom as **Jonathan Ellinger **mentioned it.

See bellow the problems I had in processing with my original file in PIX4d :

 

 

By a few fast step, you can’t find a fix.

  1. open Lightroom and import your picture folder

  2. select all the pictures in LR

  3. click on read metadata from file

  4. click on save metadata to file

  5. export them all in a new folder without compression.

 

You can now import the folder in PIX4d and it seems to work ( see bellow ) 

 

 

As said, I am an amateur, so I don’t know if it can work for you all in therms of quality.

But it is good enough for me.

 

See ya.