I wonder if someone would be kind enough to explain the scenery (specifically the terrain, rather than buildings) rendering technology that we believe is going on here. I have been googling photogrammetry but there seem to be different implementation options. Is it the same as ortho/satellite scenery or is it some sort of hybrid using dimensions  extracted from images and used to define areas which are then filled in with artificially tiled texture? Or is it a mixture of the two? I have been flying Aerofly FS2 for a couple of years so I’m used to ortho but haven’t spent much time in the other sims.

Edited August 23, 20196 yr by spit40

54 minutes ago, spit40 said:

(specifically the terrain, rather than buildings) 

Hey Phil!

Yeah, interesting question.

I believe that there's an underlying high resolution DEM mesh that dictates how to "drape" the photogrammery mesh, or maybe it's baked in to the photogrammetry mesh itself when it's processed. This gets you the land contours that photogrammetry can't "see".

In a Bing photogrammetry area the ground imagery does indeed look like one continuous mesh and texture rather than a photogrammetry model on top of satellite imagery. That makes sense because photogrammetry will be the highest resolution ground images you can provide.

Trees are the giveaway where they force other parts of the ground to be extruded upwards. There's no hard edge where the tree mesh stops and the ground begins, it's one continuous mesh. For example, when a car is parked near a tree you often see part of the car become part of the side of the tree.

On the Windows 10 Maps app, if you go to the edge of a photogrammetry enabled city, where the photogrammetry stops you see that the imagery quality changes where it switches to satellite or lower resolution orthos.

 

Edited August 23, 20196 yr by nickhod

My understanding is that it comes from two sources, high detailed LiDAR point clouds (These are lasers that are fired out of the plane, and the returns are recorded and a point cloud is created with elevations and other characteristics of the objects it hits). The point cloud can be turned into an extremely detailed mesh of what was seen below the plane.

The aerial imagery is taken from a plane at several angles and not just straight down, and some tech then stitches the point cloud mesh and imagery together to produce a basically highly detailed textured 3D representation of the terrain below. The buildings, trees, cars etc are all part of the same mesh, and not detachable, which can clearly be seen in most cases.

However, it's difficult and expensive to produce, so it's unfeasible that we'll be seeing this on a global scale anytime soon, but both Bing, Apple and Google are always increasing their coverage of cities and interesting areas, so they are expanding out.

6 minutes ago, tonywob said:

However, it's difficult and expensive to produce, so it's unfeasible that we'll be seeing this on a global scale anytime soon, but both Bing, Apple and Google are always increasing their coverage of cities and interesting areas, so they are expanding out.

Than would mean no 3rd parties photogrammetry. So if, let's say ORBX, wants to build an entire region with hand-placed objects like their True Earth series, how do they know Microsoft wont'be expanding this particular region with photogrammetry in the near future?

1 minute ago, Noooch said:

So if, let's say ORBX, wants to build an entire region with hand-placed objects like their True Earth series, how do they know Microsoft wont'be expanding this particular region with photogrammetry in the near future?

They don't.. Also bare in mind that photogrammetry cities don't have any night lighting, PBR etc.. so there are always pluses and minuses to any approach. But traditonally Orbx have done entire states or countries, something that is not feasible at all with photogrammetry (as awesome as it would be)

Thanks. So I can see that cities and interesting areas justify LiDAR and lots of processing. I'm wondering about a bunch of fields say in somewhere fairly boring, but perhaps near my house. With ortho I can imagine I'm looking at the real thing. From what we've seen so far, I wonder if this is more landclass / tiled generic texture or something that looks more like the real thing?

Yes, and photogrammetry works great until you get quite close and the effect falls apart. So it's pretty good for flight simulation, but don't expect lots of nice detail close up when it becomes a bit of a blocky blur (You can try this is Google Earth, especially in VR).

The important thing for me is to not necessarily the cities where only a small part of my flight is, but rather the detail in between. For it to work for me, it needs autogen and detailed aerial imagery, or accurate 3D landclass before it could replace what I currently have as my main sim. There are hints in the shots that it has this, and others seem to indicate it doesn't, so it's very hard to tell until MS verify it.

 

We need some sort of fractal thing, where the ortho is analysed and the patterns identified so its as good as the real image but when you zoom in it creates extra detail.

Way back in time when I converted Spitfire 40 from the BBC Micro to Amstrad 64 I also worked on a LucasFilms game called Koronis Rift - you could fly into a landscape and increasing levels of detail would be generated using a crude fractal algorithm. It was a sequel to Rescue on Fractalus - I concede the landscape was a bit more barren and simplistic than our lovely planet.

Edited August 23, 20196 yr by spit40

51 minutes ago, tonywob said:

My understanding is that it comes from two sources, high detailed LiDAR point clouds (These are lasers that are fired out of the plane, and the returns are recorded and a point cloud is created with elevations and other characteristics of the objects it hits). The point cloud can be turned into an extremely detailed mesh of what was seen below the plane.

What I find amazing is that there is software that can construct the photogrammetry mesh with just machine vision.

I saw a video of someone taking screenshots of a Google Earth photogrammetry building from different angles, feeding back into a photogrammetry app and it reconstructed the model pretty well.

I'm sure MS and Google throw every resource at the task and also use LiDAR, as you say.

Will someone find a way to decode Google's photogrammetry mesh tiles and bring them into MSFS? Long shot, but I wouldn't rule it out given enough time.

Pertaining to MFS, what I find very interesting is the pics of the sim we’ve gotten so far clearly point to Microsoft doing much more than simply placing sat images as ground textures.  They seem to be doing work to those sat images with regards to both coloring and building/autogen placements, either on the server end or possibly in game in real time.

Just compare that pic of Australia they gave us yesterday to how that exact location looks in both Bing and Google maps.  The coastlines and tree/field placements seem to match exactly, yet the sim pic looks much better as its coloring is vastly improved from the sat images and the tree/foliage density is outstanding.  This might be where the Azure AI is coming into play.

On ‎8‎/‎23‎/‎2019 at 6:47 AM, Mengy said:

Pertaining to MFS, what I find very interesting is the pics of the sim we’ve gotten so far clearly point to Microsoft doing much more than simply placing sat images as ground textures.

Edit: never mind, got the update.

Where is the MSFS Australia pic?

Edited August 24, 20196 yr by n4gix
REMOVED EXCESSIVE QUOTE!!! Please stop quoting the entire post you are replying to!

I'm by no means knowledgeable on this subject but is the use of OpenStreet Maps not an option as well?

2 hours ago, dongdongliushui said:

Where is the MSFS Australia pic?

In MFS:

 

In Bing Maps:

Azure AI will take care of all the implications and imperfections that come with photogrammetry.