I hope there's a lot of variability in ground lighting

[FAZZ3 815]

Monotonous yellow ground lights which all look the same can make night flights boring. I hope Asobo have implemented, or plan to implement, some variation in the type, color, shape and intensity of city lights to make night flying more enjoyable. Here's an example video,
 

 

[n4gix 4,953]

Wow! Great lighting! It almost looks real... Oh, wait - it is real. 😹

[Alec 370]

I agree this would make the world more alive!

[GlideBy 324]

There are about six different types of street lights with different colors.  And these colors affect the color of the textures around them.  (Look up CRI for more on this.)

• Metal Halide Street Lights : They produce very white light and have good color performance/rendering, meaning that objects under these lights look their true color.
• High Pressure Sodium (HPS) Street Lights : They take few minutes to turn on completely and produce a yellow-orange kind of glow.
• Low Pressure Sodium (LPS) Street Lights : They also take a few minutes to turn on and LPS lamps produce exclusively yellow/warm yellow light. 
• Light Emitting Diode (LED) Street Lights :  LEDs are newer, from the last 8-9 years, and they emit a lot of blue/white blue light
• Phosphor-Converted Amber (PCA) LED Street Lights : Even newer, 5-6 years, have good color representation, but are still rather expensive.  (These are modern and come in cool or warm tints)
• Narrow-Band Amber (NBA) LED Street Lights :  Rather than emitting pure white light (all the seven colors of the rainbow) and blue light, they emit mostly in the yellow/ warm yellow color. They still have good color representation, meaning that they do not make objects look grey like LPS lamps can do. Because this technology is very much new, these bulbs are not commonly available and as such, and are still expensive.

The article also includes sample photos, about half way down the page.  (Sourced from https://www.linkedin.com/pulse/street-lighting-urban-design-ashish-batra)

Edited March 17, 2020 by GlideBy

[Claviateur 789]

This variety of street lighting types is dynamically possible and easily customized through a config file in the other simulator. Many addons tuned the config file to render such effects...

Certainly Asobo's engine is capable of doing the same and perhaps users could also be able to tune it as they wish...

 

 

[GlideBy 324]

I was thinking the AI could look at the satellite images, find the lights, identify the street lights, then look at the color value.  Then based on location (age of neighborhood), guess the type of street light.  Then it can render the look of the light.

[MatthewS 761]

Also a small percentage of lights that are blinking or broken.

[mp15 138]

I have RGB lights on my swimming pool, I hope lights that hue from Red, Green and Blue are also available, specially for sophisticated villas like mine or like famous buildings in South Korea.

[badderjet 181]

That video is a nice catch! I'm not so much concerned about city lights, way more about airport lights. This video is one of the few that show exactly how the rabbit and REILs should be - and they should be like this in the sim! You can see it at the very beginning of the video and then again and better around minute 12:00. Please, Asobo, get this right... ~2 Hz frequency for the rabbit with the REILs flashing once the rabbit hits the threshold. No fading in and out, just flashes. I've commented on this several times and I still hope they will do this right, as it's never been anything close to real life in any version of FS before.

[Claviateur 789]

On 3/17/2020 at 3:53 PM, GlideBy said:

I was thinking the AI could look at the satellite images, find the lights, identify the street lights, then look at the color value.  Then based on location (age of neighborhood), guess the type of street light.  Then it can render the look of the light.

Ummm I do not see how any AI can identify the street lights.from satellite imagery taken during the day.

We are having a lot of expectation from the Azure AI. Or from the AI role in MSFS.

The Azure AI scanned vegetation from Satellite imagery as far as the dev mentioned. Also, the footprint shape identification is the AI task eventually. Yet as far as the Alpha screens show, for now at least, The generated infrastructure worldwide do not seem to be all based on an in-house scanning like MS did for US and Canada a couple of years ago... We see that it is rather a mixed bag of data including OSM and then sometimes we see a purely procedural generation when no footprint data is available for a specific area...

So I think the AI unfortunately can't identify street lights from any satellite imagery. Let's just hope it will be initiated to identify most of the world footprints by the time MSFS is in Beta. It is a huge task...

Street lights can be done procedurally with some flexible rules that eventually users can customize as they do with the other simulator: Light types and coverage based on Ubran vs Rural vs World Region...

[GlideBy 324]

2 hours ago, Claviateur said:

Ummm I do not see how any AI can identify the street lights.from satellite imagery taken during the day.

The satellite are in orbit and do travel over the earth at night time.  So that would give a picture at night, and clearly show where the lights are.  It then can pull street data, so it can find lights next to streets.  It is a bit more complicated than that.  

Here is a photo of Paris at night - 

See all the orangish colors/lights, those appear to be high pressure sodium lamps.  Now, see there are some yellow lights, those appear to be low pressure sodium lamps.  Next, see the whitish lamps, I would guess those are metal halide lamps.  So in each of those areas you street lamps that match those designs.

I don't even think you need deep learning AI to do the light identification.  You just do it based on colors in night time photos.  That detects the areas, and what lamps and lamp colors.  Placement can be done by the autogen next to streets.  

Could I be wrong on what exact lamps are where?  Yes, I am not an expert.  But, we also know the existing AI can and is often wrong about the buildings it places.  But as long as it looks realistic, then it is fine.  And doing it this way the colors would match the above photo.  Not using this information, means we get night scenes that could look very different than the real city.

Would this cover the world?  Probably not.  But we have the same issues with photogrammetry.  But if they have the images, why not use them?  And over time they could add more.

[Claviateur 789]

20 hours ago, GlideBy said:

The satellite are in orbit and do travel over the earth at night time.  So that would give a picture at night, and clearly show where the lights are.  It then can pull street data, so it can find lights next to streets.  It is a bit more complicated than that.  

Here is a photo of Paris at night - 

See all the orangish colors/lights, those appear to be high pressure sodium lamps.  Now, see there are some yellow lights, those appear to be low pressure sodium lamps.  Next, see the whitish lamps, I would guess those are metal halide lamps.  So in each of those areas you street lamps that match those designs.

I don't even think you need deep learning AI to do the light identification.  You just do it based on colors in night time photos.  That detects the areas, and what lamps and lamp colors.  Placement can be done by the autogen next to streets.  

Could I be wrong on what exact lamps are where?  Yes, I am not an expert.  But, we also know the existing AI can and is often wrong about the buildings it places.  But as long as it looks realistic, then it is fine.  And doing it this way the colors would match the above photo.  Not using this information, means we get night scenes that could look very different than the real city.

Would this cover the world?  Probably not.  But we have the same issues with photogrammetry.  But if they have the images, why not use them?  And over time they could add more.

For sure we have aerial imagery taken at night but I am not sure the night imagery database is comprehensive enough to cover the majority of the world urban areas?  

And again, even if we do have that, no matter what level of scanning is involved, it remains a huge task. Even with common automated algorithms that detect pixels (like Photoshop does since a long time now) it remains, when automated, a huge task for the scanning / coverage we want.

OSM data has a fair amount of street light tags that could be also available in the actual Bing map database who knows. And when coupled with procedural rules to generate the global type of installations in major regions, the result could be very satisfying as an out of the box night lighting experience.

As I mentioned previously, dynamic lighting in modern engines can be tuned easily with some parameters. Even us, end-users, could do this. Among the many freeware dynamic lighting tweaks/addons for the other simulator, there is one that fixes the light types on highways for Europe. 

As for the AI being wrong about the buildings it places as you mentioned, I am not sure it's the AI issue here. If it's the type of building that is wrong, then this has to do with extra tags or rules or algorithms involved in the generation of the buildings.

There are 2 types of assets one can use for infrastructure. 1) Footprint shapes (OSM, AI scanning or other databases). The footprints are extruded based on tags or on procedural algorithms. or 2) ready made generic 3D library objects that are placed there...

The AI fetches footprint shapes if asked to, perhaps also roof colors with margin of errors and maybe who knows, analyzes the shadows for height estimation (but this is not easy as it is not consistent among all imagery). Who knows, perhaps it brings also extra estimated data such as the location of the footprint in a specific area vs land use type (industrial, residential, etc.).

When we see a screenshot with a building that does not have the same shape as the footprint in the aerial image, it means usually it's not footprint that is extruded but rather a generic 3D asset placed there for different reasons. It could be that the footprint shape is not available or a 3D asset is used instead of the footprint to represent a specific type of infrastructure (place of worship etc.).

And then when it's a 3D asset that is placed (not an extruded footprint) and the 3D object is misplaced, it is usually the result of a purely procedural (random) process...

In fact what MS asked the Azure AI to do a couple of years ago when it released millions of footprints for the US and Canada is to demonstrate that it can do the same thing people did for many years now when they manually traced footprints and tagged them for their area of interest, in Open Street Maps (OSM). 

Yet if the Azure AI is scanning the whole world for MSFS, I think this process is not completed or done or applied for the whole world (yet) as far as the Alpha screens show... Because we see some purely procedural processes in some areas...

Now if this type of scanning is not completed yet (if it is ongoing), then I think street lights unfortunately are not something we could expect realistically from any type of in-house scanning.

Edited April 1, 2020 by Claviateur

[GlideBy 324]

3 hours ago, Claviateur said:

For sure we have aerial imagery taken at night but I am not sure the night imagery database is comprehensive enough to cover the majority of the world urban areas? 

 

23 hours ago, GlideBy said:

Would this cover the world?  Probably not.  But we have the same issues with photogrammetry.  But if they have the images, why not use them?  And over time they could add more.

 

3 hours ago, Claviateur said:

OSM data has a fair amount of street light tags that could be also available in the actual Bing map database who knows. And when coupled with procedural rules to generate the global type of installations in major regions, the result could be very satisfying as an out of the box night lighting experience.

Does this data cover the color and spectrum of the light output?  I just want the night view to have the same lighting variations that a real life night shot has.  Like in the Paris shot it seems there are at least three different types of lights being used on the streets.

 

3 hours ago, Claviateur said:

As for the AI being wrong about the buildings it places as you mentioned, I am not sure it's the AI issue here. If it's the type of building that is wrong, then this has to do with extra tags or rules or algorithms involved in the generation of the buildings.

The rest of what you talked about seems to be about buildings, and not lights.  (Just want to make sure I understood correctly)

They have aerial/satellite images of most places.  So this often means they don't know what the sides of the buildings look like.  So the sides are guesses that they try to fit the area.

Take a look at the two detailed castle pictures posted so far and compare to the real images.  You can see differences on the sides, like the number and placement of windows.  And those are very close matches.  Some screenshots some specific buildings don't even look close, even though they have the same footprint.  I think when you fly over an area you know well, you will see little differences.  But then again, if you don't know the area, it will look very realistic.

[Claviateur 789]

1 hour ago, GlideBy said:

Does this data cover the color and spectrum of the light output?  I just want the night view to have the same lighting variations that a real life night shot has.  Like in the Paris shot it seems there are at least three different types of lights being used on the streets.

No it does not globally but perhaps, in some cases tags concerning light type are added in OSM, not sure though.

But the dynamic light types can be created as configurations and applied through rules. I did so for my scenery in the other simulator, a small text file describing the position of the light, the spill, halo, RGB, intensity etc. One can create as many as desired. But of course these would be applied through rules based on visual observations coupled with the OSM data or any road data. So you won't get the 100% match with real life.

As for the Castle pictures or any landmark, these are custom made models and not a result of a global data processing. As for the global data processing it will never produce exact match of any architecture unless a huge database of facades assets are created and assigned to so many rules within specific regions.