Support Website Contact Support Blog

Point Cloud Clarity

I generate 3d models / point clouds for our studio projects here at UT School of Architecture. After flying the UAV on the site to capture the project, and process the model, i’m able to offer the point cloud data in a number of useful formats for the students to work with. 

I’m looking for some guidance in how to further refine the 3d point clouds I’ve generated.  I’m using Pix4D Mapper Pro 2.0.81, DJI Phantom 2 Vision+ with the onboard HD camera. The system performing the processing is Xeon CPU, 32GB Ram, Dual 2GB Nvidia Quadro cards, SSD main hard drive.  I’ve included a few screen shots of the latest models i’ve produced. 


Notice some of the holes in the building’s facade. I’ve processed the point cloud using 2 matches.

Thank you for checking this out.

Rob

Instructional Technologies

UT School of Architecture

Some things that maybe helpful,

Imaging in flat light (Nature is hard to schedule this).

Adding: MTP, GCP, Processing area, Orthoplanes on facades/roofs .

Spending time in phase 1 or initial processing set-up can yield much better results in phase 2 processing.

I’m with Gary, and it might just be the texture of the facade that Pix4D can’t find enough keypoints. Check some of the images that have that part of the facade and see if there were any automatic keypoints in that area. If not it might just be too homogeneous.

Another option is to change to processing options for Step 2 so that point density is set to High so you have more points to work with.

Update:  I’ve added MTP, GCP and spent more time in phase 1 of the process.  The point cloud has improved some, but, not quite as much as i was hoping.  It’s as if there is more than one layer of wall points when the UAV passed in front of the facades.  I also ran the processing for step 2 at high density, and ended up with more than 320 million points in the cloud.  There are 244 camera images in this project.  I split that into 3 sub projects to get a better grasp on the size of the model.   Is there a way to select just a certain area of the model to export and then run Step 2 on a smaller portion?   The goal is to focus only on this end portion of the building.

Thank you

Additional insight on the above update:  The project I ran through prior to this one turned out an exceptional point cloud. I find i’m always looking for better resolution when utilizing the file types in the 3d_mesh directory using Rhino & 3DS Max.  Recap has been a challenge, where Pix4D is far better. 

Thanks for the look.