Accurate Elevations for France ?

Discussion in 'Tracks' started by r@m, Jun 16, 2017.

  1. Pixelchaser

    Pixelchaser Well-Known Member

    Joined:
    Jan 4, 2015
    Messages:
    1,541
    Likes Received:
    1,050
    that looks pretty good quality. just like some optimised lidars.
     
  2. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    brushing a mesh onto a terrain for elevations is bloody blissful :D
     
  3. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    I tend to grab everything I may possibly need in one pass.

    Quite often Google might change their reference materials. On some areas you even see the 3D model and textures change depending on the viewing angle.
    For cities stuff can actually change quite often too, and stuff doesn't always join up perfectly.

    Since each set is slightly different it just makes binding it all together a pain.

    Also because it's best to rotate/scale each 'reconstruction' over a larger area (error reduces the bigger you go)

    Trying to make sure each reconstruction you create is scaled/rotated the same amount is also a faff.

    Needless to say I just do it once and do it really nicely and grab everything now. It might be like a day of work, but I know I'll never need to go back to try get a small area in more detail, or whatever else.




    I also usually do a fairly zoomed in pass for the main details which are captured very accuratley, then a less accurate zoomed out pass which captures stuff like 1km away from the track, handy for cities with sky scrapers, hills etc...

    4k can be done, or any resolution, simply by using the scale up/down in your browser, and then using a tool like screengrab (firefox) to save the actual render buffer from the browser (not the screen).

    In practice you can zoom out more if you're happy with the detail you've got, and cover more area in less shots.



    For the quality you have above, I can see the mesh you got is quite soft at sharp edges, but for the roads/areas you're making it's perfectly good enough to get the heights.

    In my case I just figure while I'm getting data, I get it as sharp as possible because I can then use the whole data-set for everything from buildings to the roads etc.
    Infact if you really let the Google maps load up nice and sharp (on a native 4k screen, the software 'zoom' method isn't quite as good in my experience) on a good quality city like Montreal, you start to get the crown of the road coming through in the data!


    At least from a perspective of making a track quickly and under time constraints, just having everything there and not needing to interpret data later, or try recapture it and it's changed, is very reassuring when you have a client wanting to run it asap before a race day!
     
  4. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    Just remember that the surface is implict, not explicit.

    If you happen to pick up peaks/troughs on the biased mesh, your surface will be bumpier than reality.

    Averaging the mesh is then lossy too.

    I tend to use lower intervals like 10m grids, sub-d it down to 1.25m (3x sub-d), and then just double check the surface you've made lays 'through' the rough surface of the road you're copying.

    It's still lots of work by eye... but at least you've got a Z target :D

    Dave
     
  5. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Yeah I wouldn't even consider using the actual conformed mesh as the track, just select the center loop and create a spline shape, then normalize / smooth the spline and create the track mesh from that.
     
  6. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    Now you have to look out for why the 3D data doesn't line up with the 2D imagery. Good old parallax!

    Unless of course your 2D aerial imagery is orthorectified (not likely for a city map on Google etc)

    Another reason why just getting the 3D data as good as possible is nice because you can mostly abandon the 2D aerial imagery :)


    Either way it's all still workable, it's just more faff later.

    Oh for static lidar scans for free of the entire world :D
     
  7. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    Roger that.

    If you're just picking up the Z data of the middle of the road you won't go far wrong with the data as you've captured it :D


    I'm likely going overkill for your needs here... I just pushed this technique as far as it'd go for client projects and they just want it to be as close as reality as possible... but in practice, a fun track accuracy doesn't always make :D
     
  8. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Same as OSM data overlaying sat imagery?, manual intervention is always required :banghead:
     
  9. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Yup, I can create edge splines from the new spline mesh and camber the track with splinetrackmaker too if needed.
     
  10. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    Problem is some OS data is drawn in vector over old rasters, drawn over old aerial imagery.

    This is why the 3D data from photogrammetry is so nice, it's spatially perfect in xyz all together.

    Then orthorectified imagery (some I checked from OS is good, even on buildings etc in rural areas, not sure about cities) is nice too.

    But yeah, it's why I just get super HQ 3D if I can, then just use the very high res 2D aerial for measuring only (ie, pan it around to where I'm working etc)

    Yep splines are really nice if you're sticking to just a single route.

    But I abandoned splines for city tracks and just ran with sub-d stuff everywhere. After lots of playing it seemed the best way to manage things because of all the street intersections running all over.
    Ie, crowns/cambers through cross-roads etc, are just unmanageable with splines.

    With Formula E tracks they'd often just run all over the place on wide streets etc, so splines just were too inflexible vs just making the whole surface in a control mesh.

    That said, I still used the spline for the initial sweep and Z data. I then draped a quad mesh onto that Z data, and then added camber/crowns into that quad mesh, which was then sub-divided, and then the actual visual surface was draped onto that haha :D
     
  11. Willy Wale

    Willy Wale Member

    Joined:
    Jan 12, 2017
    Messages:
    96
    Likes Received:
    96
    Location:
    UK
    Interesting. I've been wrestling with this on my city (well it's a town) circuit. I plan two layouts that a few of the same junctions but different routes through the junction. One layout turns right the other straight on etc... I've been trying to decide on a criteria for picking which layout takes priority and gets the spline. Your solution negates that but I've been extruding the road sections with kerb stones as Blender seems to struggle to do even the simplest CAD operation with any accuracy.

    So now I'm torn!

    WW
     
  12. luchian

    luchian Administrator Staff Member

    Joined:
    Jun 3, 2014
    Messages:
    3,007
    Likes Received:
    1,491
    Pics or it didn't happen :D
    /sorry, had to
     
  13. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Yup I'm finding OS data in certain areas totally unusable, and the stuff I can use needs a ton of TLC before it's acceptable.
    It's all very "loose" and needs a shitload of post work to truly match the sat imagery, so much so that sometimes it would be quicker to just manually build things with splines using the sat imagery as reference.

    re: cross roads and junctions I generally build the kerbs first, copy the splines build the road surface from them, I still haven't figured out the best way to get them evenly meshed.

    This one on the champs is a beauty, the islands are extruded directly from the OS data after re positioning, these need doing again, the interpolation on the curves is too much. it equates to around 3m in game which is very apparent.
    I generally normalise these to around 1m.


    upload_2017-7-12_0-3-1.png

    upload_2017-7-12_0-12-22.png

    upload_2017-7-12_0-6-27.png
     
    Willy Wale likes this.
  14. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    Yep, the old how to fill in the space between splines with a nice even 1m flowing grid issue :D

    On the last city track I was building (personal), I extruded the 1m interval spline outlines (normalised chunks to about 1m) to get a surface (top of the extrusion)

    Then it was just a case of cutting into it and flowing things and adding triangles into the curves to get nice kerbs.

    If you were a good coder you could probably write a tool to fill in the gaps elegantly. I bet there are tools in pro apps to do this too.
    I'm sure the Leica Cloudworx thingy has a nurbs outliner and in-fill polygon tool.

    I did try using nurbs to mesh a nice surface but I don't get on well with Max's nurbs :D

    Dave
     
    Willy Wale likes this.
  15. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Nobody does ;)

    Couple of tools for "Quad" generation on meshes in Max, afaik Pop Terrain only goes up to Max 2014.

    QuadCap Pro:



    Populate Terrain:

     
    Mr Whippy likes this.
  16. Mr Whippy

    Mr Whippy Active Member

    Joined:
    Jan 18, 2016
    Messages:
    199
    Likes Received:
    142
    Ah Quad Cap Pro... I didn't realise it worked on splines too. That could be quite promising for my needs actually.
    With just a few control cross-sections in complicated areas, the rest *should* fill in quite nicely maybe.



    I've bought a few of Marius' tools over the years.

    I wasn't so keen on turbosmooth pro all considered, the main issue I have with both it and opensub-div (same thing) is that as soon as you use any poly tools to add/move/tweak loops then the size of the creases you've set already change.
    Quad chamfer from Marius with the edge weighting is much more useful, vs doing support loops manually while doing the initial tweaking of sub-d meshes.

    Since I don't use his turbosmooth pro (was useless as soon as I realised creases were not persistently sized once set!), and I bought his quad chamfer, I might ask for a trust based freebie-swap to Quad Cap Pro :D


    Dave
     
  17. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Yup, make a spline shape, normalise the splines around it's edges to 1m and be amazed :cool:
     
  18. Pixelchaser

    Pixelchaser Well-Known Member

    Joined:
    Jan 4, 2015
    Messages:
    1,541
    Likes Received:
    1,050
    great modeling, but how are you going to reduce that road mesh for Assetto Corsa visual layer ? is it for AC ? my presumption. I think if its like that everywhere you might run into troubles in ac.
     
  19. r@m

    r@m Active Member

    Joined:
    Jun 7, 2017
    Messages:
    199
    Likes Received:
    141
    Which road mesh ?
     
  20. Pixelchaser

    Pixelchaser Well-Known Member

    Joined:
    Jan 4, 2015
    Messages:
    1,541
    Likes Received:
    1,050
    I just see the start of a 1 metre spaced road mesh, which great for starting the physics layers, but isn't it overkill for any visual mesh ? it all adds up that's all I know. post #73
     
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice