Support Website Contact Support Blog

Large data Project

Hi…

I will process large image to produce DEM and orthomozaik (about 21.000 photos). I have to split my project (3000 photos each project) and each project consist 3-4 GCP. I will ask some question.

  1. first step is I have to do step 1(initial processing) in each project.

  2. Then input GCP in each project and then “reoptimized”

  3. what’s the next step? merged all project into 1 project and then do step 2(point cloud and mesh) or go to step 2 and 3 and then merged the project.

i hope this forum will give me idea…

 

thanks a lot

Hi Hamemayu,

Here is the article you were probably looking for about processing large datasets: https://support.pix4d.com/hc/en-us/articles/202558579

Let us know if something remains unclear.

Regards,

Dear Julie!

We are having issues with moderately large datasets. One half of the project seems to be OK, the other is tilted 30 deg down (see the attached picture). We have 2200 24Mpix images, which is not that much. Any suggestion besides splitting the project into 2?

Hi Gyula,

It is highly recommended to add more GCPs in the project, especially where the flights intersect in order to ensure that the flight gets connected together and in order to avoid blocks. As a general rule, between 5-10 GCPs per flight are recommended. 

If possible, placing additional GCPs and MTPs in the project and the intersecting area would be beneficial.

Hi Blaz,

Yes, we had 7 or 8 GCPs, which are not included in this particular run, but we have tried to proecees the dataset with GCPs in all different combinations. When we have GCPs the project gets bent instead of broken. I just attached a picture showing strange behavior what you might have experienced before. Last time splitting the project into two parts helped, just makes processing more time consuming.

There is no general rule on how to proceed, but when dealing with corridor projects, uniform and flat surfaces, the breaking, and bending of projects is more likely to happen.

I would recommend the following:

  • Splitting the projects in two part and process them separately.
  • Using GCPs in MTPs in both subprojects.
  • Once the subprojects are successfully processed, merging them.
  • I would also recommend checking the camera optimization in the Quality Report. In case that the initial and optimized values vary a lot, using the All Prior option could solve the bending issue. 

Thanks Blaz!

Is it the same in Standard and Alternative Calibration process (optimize camera calibration parameters with All Prior)?

We had the experience, that alternative calibration is less precise when having Checkpoints.

Best,

Gyula

 

Alternative Processing Mode is recommended for processing aerial nadir images of projects where:

  • The ground looks essentially flat in the images due to the proportion between the height of the objects and the height of the flight.
  • There is good image geolocation.

If both of the conditions are fulfilled the Alternative Processing Mode can be used.

As for the All Prior, it is a way to force the calculated internal camera parameters to stay close to the initial parameters. This options should be used when large differences between the initial and calculated parameters are detected, regardless of the processing mode (Standard, Alternative or Accurate Geolocation and Orientation).

In case that you have a dataset with both accurate geolocation  and  orientation, you could also give it a try with the accurate geolocation pipeline as described in the FAQ on the accurate geolocation pipeline.