Problem with missing points on low feature surfaces

HI everyone.

I am testing Pix4D for some time now and the software is brilliant.

I have a contract with BIM company, to scan the roofs of the structures they are scanning with LIDAR, as it is hard for them to do this using laser scanner.

My problem is that when it comes to scan roofs covered with black bitumic insulation Pix leaves me point cloud with areas where there is no information, and i bellieve its because thisis low feature surface.

Correct me if im wrong…

Previously i used Photoscan and had better results, but decided to swith to P4D because its more robust and less resource consuming.

The best results as for now i achieved with settings as follow:

Initial Processing: Image size 1, Calibration Standard, Aerial grid

Densification: Image size 1, Optimal density, NoM 2, Multiscale No

Above this i get alot of noise on the edges of HVAC infrastructure located on the roof,

Thanks in advance



Hi Lukasz,

Would you mind sharing the dataset with me so we could take a look? Like this, I could better comprehend the limitations here.

To share the dataset we could use our Cloud platform as described here.  



Ok, i will upload the dataset later this day,

Meanwhile have another question…

If i want to process dataset that includes homogenous surfaces like tarmac etc. where very little points is generated. Then can i tell the software to create the surface from points in desired area, based on those few generated points ? Is there a tool for this?

I hope i made myself clear :slight_smile:



Hi Lukasz,

The point densification is taking part in step 2 and it can be done only for a specific part of the project by using a processing area. However, the point densification cannot be set as a specific option but rather as a set of them (please see: )

When doing the point cloud densification the process will always start from the ATP’s that are extracted in step 1. 

I hope this will answer your questions. In case not please do not hesitate to write back here.



This is a common photogrammetry issue because the algorithms search for changes/differences to make points.  LiDAR just always has equal spacing but photogrammetry can have better edge definition but worse monotone, smooth surface density.

I wish I could give specifics but each project is different and I can’t freely divuldge my trade secrets and pending patent information…but I can say the technology exists, using Pix4D, to take a monotone, smooth surface and make a great point cloud.  You just have to give Pix4D enough detail so that it isn’t so monotone or smooth as I did for this putting green below.

Thanks Adam,

I know it is possible because ive seen it done. At the moment the thickness of my point cloud in this black roof area is about 0,4-0,5m doesnt matter what initial and densification image size i choose. Its just to thick to accurately build surface in Revit.


Hi Karol,


Could you post here some screenshots of the point cloud so I could better visualize your current result?