W.I.P. Lime Rock

[Pixelchaser]

getting there, here is editor view so far.

 

[Pixelchaser]

ingame happy with progress, just not entirely happy with my work. shit loads still to do and tidy up manually.

 

[LilSKi]

I know it's early still but it looks like the road texture is repeating a bit too much.

 

[Pixelchaser]

its not set up for tiling yet. just flung it all together. everything will get remapped after the model is a solid dawg! still not happy with some alignments tbh, so after doing this all twice today, ill probably go back and do it again tomorrow

 

[Pixelchaser]

errors and error margins. what is acceptable, and what is not acceptable ?


its my bet no modder ever talks about this stuff. so if you don't notice then you don't know any better. is that how it really goes I wonder ?

opinions ?

 

[LilSKi]

More or less for me yeah. If I compare to the real deal via videos or pictures and can't see any perceivable difference then it is fine by me.

Also iracing would be a good reference for this. I'm sure there are many videos of it on YT with comparable FOV when compared to real life go-pros.

 

[Pixelchaser]

yeah but you cant really read an error of say 2 metres over the distance of 1000 m. I cant even deduce where an error has come from tbh, is it even an error, is google imaging correct ? is the lidar based image correct. total cluster **** is how I feel about it, of course I can STFU about it and pretend its perfect when its not, what to do, what to do.

is the fabled 1% error margin acceptable, that is the real question. I reckon I'm hitting 2% at present prior to a new build.

 

[LilSKi]

Do not ever trust Google imaging. It didn't even come close to fitting the lidar intensity data.

 

[Pixelchaser]

ok, so what was its error ? mine seems dimensional, but not necessarily a precision area error. I guess its like the data taken wasn't a square but they chose to fit it into a square.

 

[LilSKi]

Corner radius on a few corners were all screwed up. There was simply no way to get it fully lined up with the data no matter how I rotated, moved, and scaled it. I recommend the track is built on the lidar intensity data and only use GE images for object placement. Do not rely on it for the layout.

 

[Mr Whippy]

Google maps images aren't orthorectified, because orthorectified images are many factors more expensive and only useful for GIS types (or us), not just for generic online maps.

Take a look at the new OS Open Data ortho-rectified image tiles, they're perfect and drape over their DSM or DTM grids perfectly (1km squares)



As for how accurate is accurate enough? You have LIDAR data. That gets you 99% of the way there.

If you filter for noise (by eye, unless you have super accurate data to reference against that's all you've got) then I'd just run with what comes out of the other end unless you have really strong evidence that the data is wrong.

I built a few engineering tracks a few years back and it was all done using what data I had to hand and just processing it and trusting what came out of the other end.



I'd drape a spline over the lidar mesh you have.

Create a nice spline that follows the centreline of the course (and maybe one for the inner and outer edges)

Create 1m knot intervals (using normalise spline), then use a smoothing script (scriptspot) to get rid of the small noise but keep the big noise (the by eye bit, you can get an idea of what is right/wrong), the 1m knot intervals make any filtering consistent vs distance.

Use these splines for surfacing, lofting, or 'sweeping' along, and then set the vertical data of the mesh to the splines z data.

Then just work out from that.

 

[Pixelchaser]

yeah it be nice if OS did cross the pond to america though

I'm already doing that style of build...

lidar sits under the splines and game mesh sits above and tied to the splines for smoothness.it seems to capture major height relief, but I don't trust any of it. 99% is only achievable if the talent for it is 100% and its not .

 

[Willy Wale]

When I overlayed a google image of my home town on the OS master map data and lined up one side, the other (2km away) was more than a road width out.

That said I would agree with @LilSKi, if you compare the in-game to a video and can't see a problem then there isn't one. I would bet that if you took your entire circuit and scaled it up or down by 2% you'd struggle to know which one you were driving unless you did a back to back drive.

 

[liquido]

Great work Ali, I miss this thread.

love what are you doing. Sure you tell me more in discord.

 

[Mr Whippy]

When I overlayed a google image of my home town on the OS master map data and lined up one side, the other (2km away) was more than a road width out.

That said I would agree with @LilSKi, if you compare the in-game to a video and can't see a problem then there isn't one. I would bet that if you took your entire circuit and scaled it up or down by 2% you'd struggle to know which one you were driving unless you did a back to back drive.

I agree too, this is the 'is it right before I build everything else' jitters.

At some point you just have to trust it

yeah it be nice if OS did cross the pond to america though

I'm already doing that style of build...

lidar sits under the splines and game mesh sits above and tied to the splines for smoothness.it seems to capture major height relief, but I don't trust any of it. 99% is only achievable if the talent for it is 100% and its not .

Go for it



Dave's story for the day...

I worked on a project 4 years ago now (wow!) that was a mobile laser scanned UK road, but I didn't get the road mesh for months into the build. I didn't even get the point cloud data for a week or two before starting.

So to get going I just got the 50m DTM (all you could get back then from OS) and just dropped a spline over it, lofted it, and built the verges and terrain out from it to get me going placing everything else.

Once the actual road meshes started coming in, yes it was accurate, yes some details were more right, and yes the finer 'bumps' were apparently real.
But driving it side by side with my original loft with random noise applied for the bumps, generally I didn't really find either one better or worse than the other.

Yes the laser scan was technically right and better, but in spirit of the real place, and in the absence of superior data to use any way, the rough DTM version was pretty damn good!





So have faith in the data... it'll be 95% right with LIDAR of the quality you have...

Keep up the good work

 

[Pixelchaser]

mapping side transfers to grass. 4 variants where the blue is. and red will be one to surround the rumbles and kerbs. but everywhere has transfers now and mapped road etc. I might make variation for that too. lime rock is kinda the same all around so....

 

[luchian]

From this height looks brilliant. Very impressed with the job so far .
Keep it up !

 

[Pixelchaser]

starting to look ok in game I think. the main point is. it all fits now. game mesh outer low rez lidar, everything has side textures and its all mapped. just kept the google on the multilayer diffuse for now with a plain red mask and 1x basic grass. ,all the inside terrain will be remapped later and no design on the tarmac specular yet, bit shiny there etc. but that will be painted over. still shit loads to do but it is going forwards. \o/ the first real race is in may, so I hope to finalise it with footage from the may races.

 

[Pixelchaser]

and being honest, I'm failing with my currect processes. I'm having to restart this project. 1 its my first time playing with real data. 2. without orthographic corrected imagery I'm just not willing to fudge anything. 3. I now have orthographic imagery.

the differences in data used are utterly amazing. but when you build such flaws into your track, I must know when to stop and restart and its that time.

 

[Pixelchaser]

got some new data to work with. and the data I was using was majorly flawed, I just didn't know. I had ordered only "ground" data. this was clearly processed data and not raw data. so back to the drawing board. good thing is between the ortho corrected imagery, the intensity map delivered with the lidar and the mesh, it all fits now with raw data and better knowledge.