Orthofoto with roof overhang

[Airgo]

I have to compute an orthofoto from a UAV image set. Since there is a big roof overhang ( and I need the ground to the walls of the building) I have to eliminate the roof.
Classifying my point cloud and only respecting the ground points gave me a very blurry wall-ground transition. A similar result gave me the elimination of the corresponding polygons of the mesh.
Good results I get with masking the roof. But that is very time consuming since I have 4 roos in about 150 images.
Is there another way to achieve a good orthofoto or is there a way to mask the roof only in some images and "translate" the mask into the other?
Many thanks,
Mathias

[Yoann Courtois]

Hi Mathias !

Orthophotos under overhangs are pretty common problem in aerial photogrammetry.

As you said, to be able to get the most accurate orthophotos of the masked floor, you will firstly need the classify the ground points to be able to get a DTM (digital terrain model) instead of the DEM (digital elevation model).
Then, you will have to compute the orthomosaic. But, as PhotoScan create it using pieces of images which have been taken from the best place (i.e. the most vertically above a point), it will probably project the roofs on the floor...
Two solutions to correct that:
- As you said, previously the orthomosaic generation, by masking on pictures that have been mostly taken at the vertical above the roof. It may take a time and it may be difficult to copy a mask from a picture to another one (as the UAV moved) and also difficult to use the magic wand we can use to mask the sky on terrestrial imagery (as the roof isn't a plain color)...
- Subsequently the orthomosaic generation, by editing the produced orthophoto using polygon shapes and assigned images.

Hope it helps !
Regards