Photogrammetry for large scale environments

Blubberblase

Hello guys,
There are a lot of ways to get the environment for an architectural project into unreal but what about civil engineering and bigger datasets? I am currently facing the task of getting around 50 houses + surroundings into unreal and i could really use some new ideas.

What i have tried so far:
modelling after Photoreference, texturing with baked camera mapping
pro:
- acceptable results for buildings
con:
- everything except the buildings is quite hard to do
- time consuming and boring after a while
- a lot of guesswork and trial and error
- might work for a project of this size, not quite sure if there were a few hundred houses

Photoscanning everything at once in reality scan using pictures from a drone and handheld camera, retopo in chunks in max, texturing in reality scan
pro:
- good results especially for stuff that is further away
- very high poly meshes i can use for normal baking
- could probably be automated to just work its way in the background
con:
- I have noticed that People don't like it if someone flies over their property or takes a bunch of pictures from their home 
- sometimes it takes a few tries to get a decent result
- if not computed in high settings worse results than the other method

I also tried to photoscan single buildings, but i never got reliable alignments that way and often had to do manual tracking of the alignment process in reality scan (i also tried photoscan and zephyr)

what i want to try:
- take a bunch of action cameras, tape them to my car and just record some 4k footage, put that into reality scan and see how it looks.
- take a single or two pictures of a building, blur them for color information, use substances for details, hope noone notices the probably quite glaring differences to reality.

stuff that won't work:
- everything drone related will propably be impossible very soon due to new laws here in Germany.

I am open for all Ideas, no matter if it's workflow related or software. 
Thank you guys

cabooes

I honestly think photoscanning might be the way to go  but that's just me. You could scan multiple buildings with one camera it would just take sometime. I don't know much about Reality Scan, however for Photoscan the alignment could be fixed if you edit the original images to the extent where they're properly exposed and increase the key points to 400,000 and the tie points 100,000 and have the accuracy set to high or highest. For me I have these options set as default and even scans that consist of 500-700 photos come out really good. I will say though if you don't have a powerful computer this may take some time because you are using more power to process the models. 

fearian

If this is work you are doing for a client, you might want to budget for a renting a lidar scanner. FARO scanners are the best entry level option here, although learning to use them and handle the data will be a big timesink.

all that might be out of reach if this is a smaller project. You need to ask yourself/client how important it is that all these buildings look accurate and to what detail level. Could you copy instances of houses around even if the dimensions vary by several metres?

But to answer your questions about photo scanning, I would recommend reality capture. Walk around with an video camera on the end of a 3m pole you can hold steady to get a higher angle. For key areas print out tracking markers and do a couple of static "scans" to get better alignment and quality. When you process the data you can process these static areas, then match and align them into your global scan.

you will probably still need to remodel buildings if you do a drive by scan, but you can always keep it simple or do a detailed scan of geometry you know will repeat.

Blubberblase

Thanks guys, i did not know that you can rent a lidar, i will try that. I actually don't like photoscan that much, mainly because i never got good results, but that may be my mistake. For a lidar scan i would need rc anyway if i am not mistaken?

fearian

yes that would probably be your best bet