Animation problem

[Devin Johnston]

I've got a situation that I've never been in before, I'm working on a golf course animation and the surrounding environment is very important.  Normally I would use a drone to shoot video of the camera path and then drop my 3d model into the footage.  In this case there are lots of existing trees in the way and there's no way to easily remove them from the footage so I think I'm going to have to do an all 3d animation.  The course is surrounded by hundreds of houses so modeling them is out of the question.  I was thinking of taking photographs of all of them from the same angle the animated camera will be moving and creating one long billboard.  I was also considering trying to use pictometry to create models of the houses but I'm not sure the process is going to create models that are of the right quality.  I'm just looking for alternative suggestions that I haven't thought of, any advice is welcome.    

[James Vella]

My thinking is shoot the drone footage, use realitycapture or alternative to turn the footage into a 3D photogrammetry scan, delete the trees and replace with grass. 

 

Edited February 22, 2021 by James Vella

[Devin Johnston]

It's been a while since I used reality capture, my experience in the past was it did a pretty good job but the mesh was very dense.  I'm concerned that the final model may be too large to use and will likely not have all the detail I need.  It will also take a long time to photograph 100+ houses even if I use something like Litchi to automate the process.  

[James Vella]

Regarding the mesh you have decimate options - I havent used them since I normally just throw it into zbrush and set the amount of triangles/faces I want, export the low poly back into realitycapture and reproject the texture, looks the same basically. Im not saying this is how you should approach it this was just an idea that came to mind since you were looking for some alternatives. 

 

Would you need to photograph all the houses if the camera path follows the drone? That was why I suggested reality capture in the first place, you only model what you see from the view, My thinking was its easier to remove mesh trees than from 2D footage.

 

Never used Litchi for automation to be honest.

[Francisco Penaloza]

it been a while since this project was posted, I would be very interested to know what route you took to develop this.
Using 2D images is kind of tricky, you really get limited to what the camera can show, but it is always better to create many small chunks than using a large single image, small chunks will help to create parallax for those fake planes houses.  Other option could be creating proxy houses and building, general looking, again if your camera work is smart enough to focus on the main project, the details of the other houses is not that important.

Using 3D scanning is a great option to get lots of graphic data, again you will have to subdivide in chucks and then proxies so your render engine load them as need it, instead loading a single large site.

I have used google earth video tool, in some cases it work decent enough, you can draw a camera path then export that to your comp software and then your 3D software. But this tool is limited to what google earth has in that area, sometimes is nothing.

If this project is done, It would be very interesting if you could share your approach to it.

 

 

[Devin Johnston]

I decided to go with the video backplate option, I shot the site from between 100 & 75 feet and from that height I can overlay my new site without the trees getting in the way.  I did photograph the edges of the site with the drone incase I needed to fake something, hopefully I won't need to do that.  I didn't want to put all of my hopes onto a pictometry model, there's no telling how it would have turned out and since I had to fly to the location I would be able to go back if I missed something.

[Francisco Penaloza]

Great, thank you for the update