Import Cloud for DSM - Speed of processing Step 3 and reverting to Step 2 Cloud

Hello,

I have a couple of questions about the “Import Cloud for DSM” process, and how it affects processing speed of step 3.

I have a narrow 67km long canal corridor that we have flown photogrammetry and Lidar. Initially we processed the captured photography through steps 1, 2 and 3, to generate a point cloud and ortho photo, with the orthophoto component of the processing taking approx 11 hours total to generate 5cm resolution tiles. This resulted in some gaps in the orthophoto, because in some areas we did not have sufficient photo overlap.

I am currently re-processing Step 3, having imported the processed Lidar LAZ file using Import Point Cloud for DSM Generation. After 48 hours of processing 10cm tiles, and including a processing area, we are only 56% through the processing.

The imported lidar is UAV-captured, and has approximate density of ~100 points per m², classified for ground, and vegetation. I am confident that the lidar is in the same coordinate terms as the Pix4D project and the tiles that have been generated to date are correctly geolocated. It’s just super slow.

The processing is running on the same computer as before so should be running quicker, I would have thought.

So my first question:
Does the Ortho generation run quicker if the point cloud was produced by Pix4D than if using an imported point cloud? Is there any recommendations for optimal point cloud characteristics for import?

And my second question:
Is there a way that I can revert to the point cloud that was generated in Pix4D (Step 2), once I have imported a an point cloud for dsm generation?

Thanks very much.
Phil Dewar
Fox and Associates, Christchurch NZ.

Hi Phil,

Thank you very much for the detailed explanation of your project. Welcome to our Community!

Does the Ortho generation run quicker if the point cloud was produced by Pix4D than if using an imported point cloud?

That is a very interesting question. The speed of Step 3 DSM, Orthomosaic, Index relies heavily on the hard disk. The speed of the Hard Disk defines the processing speed. If there is an SSD available, it is usually recommended to process your projects on the SSD and to store the images on the SSD while processing. This can help very much. Hardware components usage when processing with PIX4Dmapper For step 2, CPU and RAM are the most important as they are being fully used. You can also check here to see if your computer spect reaches recommended or High-end Build Components. System requirements: Minimum and recommended computer specifications

Is there any recommendations for optimal point cloud characteristics for import?

A few things to keep in mind for the imported point cloud are stated in this article: How to import a Point Cloud delivered by an External Source into PIX4Dmapper for processing

  • The external point would need to be in the SAME coordinate system as the output coordinate system.
  • The point cloud generated by PIX4Dmapper and the external point cloud should be aligned. Mark common GCPs in both the model generated by Pix4Dmapper and the external point cloud to align them. If this is not possible, extract points from the Pix4Dmapper model and mark them as GCPs in the external point cloud: How to align projects.
  • If the external point cloud is a DTM, the Orthomosaic may present distortions.
  • All imported points will be treated as a unique group meaning that if the external point cloud is classified (e.g. some points have a terrain or object label), Pix4Dmapper will not take the classification into account. If only points corresponding to the terrain class are imported then the DTM will be generated accordingly.

Is there a way that I can revert to the point cloud that was generated in Pix4D (Step 2), once I have imported a an point cloud for dsm generation?

At the moment there is no reverting feature. We always welcome our users to share their idea of new features here: PIX4Dmapper feature request
For now, you can regenerate step 2 but step 3 would also have to be regenerated.

One more tip to increase the speed and performance when working during step 1 and working with rayCloud when it comes to high usage on GPU is to select high performance in the graphic setting.

Also if the dataset is more than 1000 images you may also give PIX4Dmatic a try. We developed PIX4Dmatic which is optimized for large-scale projects. It is designed to handle and process thousands of images with ease while maintaining survey-grade accuracy. For Windows OS, there is the hardware accelerated feature in the Densify step for generating point cloud. However, PIX4Dmatic can only process both LiDAR and photogrammetry point clouds in the same project that has been produced with PIX4Dcatch.

I hope this information is helpful. Please feel to share more with us if you are comfortable such as log files or a quality report for us and our peers here to have a look at.

Sincerely,
Rosana