Detection probabilities

[FlightControl]

Guys, let discuss some additional filters that could contribute to make detection more "realistic".

 

The first is DistanceProbability for visual detection only. The further the target is from the detector, the less likely it will be that it gets detected, right? Why this factor? I found that FAC, and JTAC have incredible eyes. Even with binoculars or de-interference zoomers, the speed and accuracy of detection is phenomenal in DCS World. The are able to detect targets with one second that is 5 km away, including type information, all visually. That is why this additional probability factor comes in. Specify a %-tage probability that if a target is detected at 5km distance, that it really would get detected...

 

The second is AngleProbability. The higher the detector is located, the more likely it can detect targets. I found that Recces are able to detect targets 5 km away within seconds when on the ground. Bullshit. There are always some clutter objects in between. Also, targets will likely hide behind clutter and camouflage... Certain areas or terrain you're should not allow this.

 

The third is Foggy Zones... A ZoneProbability. In DCS World, trees, buildings are not accounted for in DCS World forum LOS calculation. With *ZoneProbability*, the mission designer can define zones where detection will be very difficult. Each zone can be given a probability factor. An array of zones with probability factor will be accepted. Of course, the more zones, the more CPU overhead, so this is to be used for very specific detection scenarios.

 

Now, my question, ... Are there other probabilities that need to be taken into account for visual detection....

 

And are there any probabilities that need to be added for radar, ir, drink etc detections???

 

Fc

[Bushmanni]

I'm assuming you are talking about the ability of the forward controller to detect targets. Detection probability within area that can be seen is mostly about how attention is directed and what kind of cues the target is providing to be detected. This can be very situation dependent. For example detecting a moving tank is easy from 5km or even 10k away but stationary infantry can be essentially invisible until you step on the guy. You should take into account the unit type, if it's moving and if it's shooting.

 

You generally have some kind of idea where the enemy can or is likely to come so you direct more attention to that area and the smaller the area is the faster it is to detect a target within it. In Steel Beasts mission designer can assing a primary search sector to a unit in defensive position and it will detect targets faster within that sector. That is pretty good implementation of this effect. It would be nice to also dynamically alter the primary sector based on recent detections so that if one of the units of a larger formation is detected, it will improve the detection probability of the other units close to the detected unit.

 

Looking from a hill will increase the observable area and hence give you more chances of seeing something interesting. You will also be more visible to others in the low ground though. Bigger area to search will also mean it takes more time to detect units within the visible area.

[FlightControl]

I'm assuming you are talking about the ability of the forward controller to detect targets. Detection probability within area that can be seen is mostly about how attention is directed and what kind of cues the target is providing to be detected. This can be very situation dependent. For example detecting a moving tank is easy from 5km or even 10k away but stationary infantry can be essentially invisible until you step on the guy. You should take into account the unit type, if it's moving and if it's shooting.

 

You generally have some kind of idea where the enemy can or is likely to come so you direct more attention to that area and the smaller the area is the faster it is to detect a target within it. In Steel Beasts mission designer can assing a primary search sector to a unit in defensive position and it will detect targets faster within that sector. That is pretty good implementation of this effect. It would be nice to also dynamically alter the primary sector based on recent detections so that if one of the units of a larger formation is detected, it will improve the detection probability of the other units close to the detected unit.

 

Looking from a hill will increase the observable area and hence give you more chances of seeing something interesting. You will also be more visible to others in the low ground though. Bigger area to search will also mean it takes more time to detect units within the visible area.

 

So, how realistic is dcs then?

[razo+r]

look in your other thread for that ;)

[FlightControl]

look in your other thread for that ;)

 

Ok. Sometimes I have the impression that my posted are not read. Thanks for your answer.

[FlightControl]

look in your other thread for that ;)

 

Where is the answer?

[razo+r]

the answer is in the post #2 of your thread

 

as i said there, it depends...

[FlightControl]

look in your other thread for that ;)

 

the answer is in the post #2 of your thread

 

as i said there, it depends...

 

? Bushmanni answered that. And... It depends is not good enough.

That is a consultant answer. Ib guess only those people can answer this question who have used the detection API of dcs world...

Not sure if there are people around who have?

 

Los going directly through villages and forests is not what I would call realistic.

 

Fc

[razo+r]

https://forums.eagle.ru/showthread.php?t=182817

[FlightControl]

https://forums.eagle.ru/showthread.php?t=182817

 

I confused you.

 

That other one is an open question, to get input from military experts. There are quite some people there...

 

The post here is for mission designers... Maybe a bit more technical.

 

Wanted to grasp both views.

 

Fc

[FlightControl]

@bushmanni. You've introduced an interesting concept. Recce van be "informed" ok target locations likeliness. Also, advanced processes may lead to faster and accurate detection? How do recce deal with camouflage?

[shagrat]

Keep in mind, that DCS AI has the same "uber" capabilities when spotting/identifying targets.

I recommend to have a look at the ../scripts/AI/Detection.lua to get an idea what basic factors influence spotting of targets.

 

Usually the JTAC is setup in missions with "FAC engage Group" which comes with an option to know where the group to attack is! It basically deactivates the detection algorithms.

If you change the detection of targets for the JTAC only, you may imbalance the whole AI, as the enemy still sees through trees, buildings, and if you are close enough and it is not dark he will spot you in seconds whatever you do. :dunno:

 

If you want to make a realistic spotting algorithm, you need to take into account, weather (fog, humidity, rain, snowfall), time of day (light available, angle of light), backdrop behind the target (contrast, color-difference), reflections (metal, glass), movement (both speed, but also angle), obscuring terrain features (basically an LOS check) and last but not least equipment (Binos, IR-sensors, etc.), even fatigue/concentration after being on watch for 2 hours plays a major role.

Finally the time of day/night relates to the biorhythm, so it is very difficult to stay alert between 3-5 in the morning... ;)

[Bushmanni]

One factor I forgot to mention is that detection probability is about detection time, it's not a can see - can't see situation. So as long as you have enough time you will eventually detect all the targets that can be detected, the ones that are harder to see just take more time on average (you might be able to spot a hard to see target quickly because you just happened to stumble your binoculars on it or you might miss an obvious target for quite some time because you are looking somewhere else).

 

DCS environment is too abstract in the ground fight level to give accurate representation of real world detection distances. You have to decide if you model the detection distances based on DCS terrain or real world terrain. If you model by DCS terrain then you model the detection distance based on detection distances in real world terrain that is equally lacking of terrain detail (almost open desert). If you model by real world then you have to factor in how the terrain would look like in real life and model the distances based on that. This has the problem that real world Caucasus (and even NTTR) has pretty bushy landscape in some places which is not represented in DCS in any way and so infantry might be pretty much invisible in some situations, especially if they decide to lay low. So realistic detection distances might be few hundred meters for moving infantry in real life but in DCS terrain they are visible from kilometers away which will be confusing for the player as he is wondering why the recce unit can't detect the obviously visible targets.

 

The "it depends" problem is heavily based on the actual terrain in question when you are tryin to detect a target. Detection always boils down to contrast regardless of the system in use (eye, thermal camera, radar, etc.) but modeling detection based on contrast is very hard. The moment you step away from contrast modeling, you need to make assumptions on the situation and environment where the detection is assumed to happen. The color of the background vs. color of the target is one assumption that has heavy influence on the contrast of the target and hence it's visibility. Other big factor is if the shadow of the object is visible or not as dark shadow can break out an object from it's background despite perfect camouflage color. Real world detection distances can vary wildly because of the effect of contrast which is heavily dependent on the details of the environment.

 

So I'm afraid there's no concrete easy to apply data that you can use but you are required to make some design decisions based on what you want to simulate.

[Bushmanni]

My advice about these design decisions would be to give the mission creator some way to adjust the amount of terrain detail simulated which mostly affects infantry. Bushes and tall grass doesn't help tanks and other big vehicles that much but has a huge influence on infantry. Firing should make units detected (maybe only momentarily or only as long as they are stationary) even beyond the range they are actually visible.

[shagrat]

One factor I forgot to mention is that detection probability is about detection time, it's not a can see - can't see situation. So as long as you have enough time you will eventually detect all the targets that can be detected, the ones that are harder to see just take more time on average (you might be able to spot a hard to see target quickly because you just happened to stumble your binoculars on it or you might miss an obvious target for quite some time because you are looking somewhere else).

 

DCS environment is too abstract in the ground fight level to give accurate representation of real world detection distances. You have to decide if you model the detection distances based on DCS terrain or real world terrain. If you model by DCS terrain then you model the detection distance based on detection distances in real world terrain that is equally lacking of terrain detail (almost open desert). If you model by real world then you have to factor in how the terrain would look like in real life and model the distances based on that. This has the problem that real world Caucasus (and even NTTR) has pretty bushy landscape in some places which is not represented in DCS in any way and so infantry might be pretty much invisible in some situations, especially if they decide to lay low. So realistic detection distances might be few hundred meters for moving infantry in real life but in DCS terrain they are visible from kilometers away which will be confusing for the player as he is wondering why the recce unit can't detect the obviously visible targets.

 

The "it depends" problem is heavily based on the actual terrain in question when you are tryin to detect a target. Detection always boils down to contrast regardless of the system in use (eye, thermal camera, radar, etc.) but modeling detection based on contrast is very hard. The moment you step away from contrast modeling, you need to make assumptions on the situation and environment where the detection is assumed to happen. The color of the background vs. color of the target is one assumption that has heavy influence on the contrast of the target and hence it's visibility. Other big factor is if the shadow of the object is visible or not as dark shadow can break out an object from it's background despite perfect camouflage color. Real world detection distances can vary wildly because of the effect of contrast which is heavily dependent on the details of the environment.

 

So I'm afraid there's no concrete easy to apply data that you can use but you are required to make some design decisions based on what you want to simulate.

Please have a look at the Detection.lua!

DCS already takes a lot of these factors into account.

Weather/Visibility, Contrast against backdrop, LOS and time to detect. Plus neat details as lighting your aircraft with position lights, flares etc. and sun position...

 

Rather ensure that JTAC uses normal detection rules.

 

Edit: it is very useful to hop into a vehicle (Combined Arms) and check the LOS / visibility of units on the ground.

Often the amount of ground clutter that is rendered when low enough is very different from what you see in a plane at 3.000ft

Also the "flat desert terrain" in Nevada is far from flat!

 

To understand LOS from a JTAC perspective it is a good practice to actually get into the JTAC and verify its view from that position.

Just 2-3m elevation makes a huge difference sometimes.

Edited February 12, 2017 by shagrat

[FlightControl]

Keep in mind, that DCS AI has the same "uber" capabilities when spotting/identifying targets.

I recommend to have a look at the ../scripts/AI/Detection.lua to get an idea what basic factors influence spotting of targets.

 

Thanks, i will look into that.

I've been heavily testing the Detection API of DCS, for different units, different weather conditions, hence this post.

I am working on a new class to improve the detection realism a bit, which uses the default DCS World detection API, but adds additional filters to the detection. Of course, the normal AI detection behaviour cannot be influenced that easy. The new DETECTION derived classes are for Recce assignments, which report detected targets to interested subscribers...

 

Usually the JTAC is setup in missions with "FAC engage Group" which comes with an option to know where the group to attack is! It basically deactivates the detection algorithms.

If you change the detection of targets for the JTAC only, you may imbalance the whole AI, as the enemy still sees through trees, buildings, and if you are close enough and it is not dark he will spot you in seconds whatever you do. :dunno:

 

The "seeing" through trees and building is a big issue. But that that, i am making a solution. The challenge is to get the solution working without overloading CPU and influence LOS based on "cloudy zones"...

 

If you want to make a realistic spotting algorithm, you need to take into account, weather (fog, humidity, rain, snowfall), time of day (light available, angle of light), backdrop behind the target (contrast, color-difference), reflections (metal, glass), movement (both speed, but also angle), obscuring terrain features (basically an LOS check) and last but not least equipment (Binos, IR-sensors, etc.), even fatigue/concentration after being on watch for 2 hours plays a major role.

Finally the time of day/night relates to the biorhythm, so it is very difficult to stay alert between 3-5 in the morning... ;)

 

So, as told working on it, let me list the points that you've mentioned, and that i have currently developed in the DETECTION class:

 

weather (fog, humidity, rain, snowfall): This is covered by the DCS API I believe, well, sort of ... For sure overcast clouds and rain, snowfall is covered, but not sure if LOS through Individual clouds (none overcast weather) is covered. Anyway, can't do all, do I will assume that DCS world is handling these parameters correctly.

 

time of day (light available, angle of light): Definitely proven that DCS world is handling these correctly. Interesting, the repeated detection at sunrise detects more and more targets as the sun is coming up.

 

backdrop behind the target (contrast, color-difference): Parking that one. I assume impossible to handle this in DCS world using lua scripting.

I mean impossible to know if certain vehicles or planes have "fluorecent yellow" skins or not :-)

 

reflections (metal, glass): Same as the previous.

 

movement (both speed, but also angle): This one is an interesting ...

1. High speed targets are more easily detectable than low speed targets?

2. Targets driving perpendicular to the observer are more easily detected than targets approach front?

3. Formation of multiple vehicles ...

Will see what i can do with this. Maybe add an option later.

 

obscuring terrain features (basically an LOS check): Hence the 3 categories of additional probabilities:

1. Distance Probability. Handles an additional filter that assumes ground clutter objects. In DCS, the floor seems flat at certain areas, however, in real life there will be trees, cages, high corn fields, little deviations in altitude of 2-3 meters etc.

2. Angle probability: This is an interesting one. If the observer is a few feet higher than the target, the target i assume is more easily detected. That i why I've created a model that calculates the alpha angle between the floor and the observer, target vector. With a little formula the angle is calculated and an angle probability treshold is calculated...

3. Cloudy zones: Working on it. Got the zone check working. But need to do additional ray casting algorithms to check if the target is visible if the target is behind the cloudy zone. It would be wonderful to have an observer observing, sees nothing, but suddenly infantry runs out of a forest and is observed, and corrective action can be taken, right?

 

equipment (Binos, IR-sensors, etc.): That is handled by the Detection API of DCS World. Options can be given what equipment to take into account when detecting using the DCS World API. Of course, the above

probability parameters are only evaluated when a detection is done visually.

 

fatigue/concentration + time of day: That is an interesting parameter. Will take that into account... Indeed a Recce that has been observing for hours may get tired, especially at night ...

 

Thanks shagrat for this comprehensive reply.

A few good points have been added that i am happily incorporating into the new MOOSE capability. I'll mention your name as one of the contributors of the class if that is okay. If you like, you can help with some testing. Just do a ping on slack. I have some prototype models and missions working to demonstrate how the detection is working and what value it brings etc.

 

great stuff!

 

FC

[FlightControl]

Hi bushmanni,

 

Thanks for your answer. The many parameters you discuss are further documented in the previous post. Note that the location of operations for detection will be very specific

recces, facs or jtac will operate at locations defined by the mission designer. As such, direct parameters for distance, angle and zone probability can be defined. These will differ from location per location.

So within a mission, multiple detection objects can be created governing the detection of recces at different locations. As such, each detection object can be given specific parameters.

[shagrat]

So, as told working on it, let me list the points that you've mentioned, and that i have currently developed in the DETECTION class:

 

weather (fog, humidity, rain, snowfall): This is covered by the DCS API I believe, well, sort of ... For sure overcast clouds and rain, snowfall is covered, but not sure if LOS through Individual clouds (none overcast weather) is covered. Anyway, can't do all, do I will assume that DCS world is handling these parameters correctly.

Yep, correct.

 

time of day (light available, angle of light): Definitely proven that DCS world is handling these correctly. Interesting, the repeated detection at sunrise detects more and more targets as the sun is coming up.

Yes, including light/dark colors against backdrop ;)

backdrop behind the target (contrast, color-difference): Parking that one. I assume impossible to handle this in DCS world using lua scripting.

I mean impossible to know if certain vehicles or planes have "fluorecent yellow" skins or not :-)

 

reflections (metal, glass): Same as the previous.

That should be part of the detection mechanism already, as well.

 

movement (both speed, but also angle): This one is an interesting ...

1. High speed targets are more easily detectable than low speed targets?

2. Targets driving perpendicular to the observer are more easily detected than targets approach front?

3. Formation of multiple vehicles ...

Will see what i can do with this. Maybe add an option later.

1. Yes, human eye is trained to spot movement. The faster a target moves, the easier to spot.

2. the more perpendicular to the LOS, the easier to spot the movement.

3. Formations, if moving together, are more easy to spot, as they are a regular and more obvious than the background.

 

(...)

2. Angle probability: This is an interesting one. If the observer is a few feet higher than the target, the target i assume is more easily detected. That i why I've created a model that calculates the alpha angle between the floor and the observer, target vector. With a little formula the angle is calculated and an angle probability treshold is calculated...

(...)

I actually meant the angle targets drive towards the LOS as above, but yeah, higher ground for example covers a wider area of observation.

 

equipment (Binos, IR-sensors, etc.): That is handled by the Detection API of DCS World. Options can be given what equipment to take into account when detecting using the DCS World API. Of course, the above

probability parameters are only evaluated when a detection is done visually.

 

fatigue/concentration + time of day: That is an interesting parameter. Will take that into account... Indeed a Recce that has been observing for hours may get tired, especially at night ...

 

Thanks shagrat for this comprehensive reply.

A few good points have been added that i am happily incorporating into the new MOOSE capability. I'll mention your name as one of the contributors of the class if that is okay. If you like, you can help with some testing. Just do a ping on slack. I have some prototype models and missions working to demonstrate how the detection is working and what value it brings etc.

 

great stuff!

 

FC

Guess that helps to better understand the influences on detecting targets.

[FlightControl]

Shagrat,

 

In the moose slack channel func-detection, I posted 2 demo missions, including the script api usage.

Have a look :-) missions can he run without installing anything...

 

Fc

[DarkFire]

Something that I think DCS may not adequately take in to account, but that could probably be implemented at fairly minimal computational cost, would be the exact type of an observer and any equipment they might have.

 

Take for example a covert observation post manned by couple of Royal Marines Arctic & Mountain Warfare Cadre soldiers, equipped with 3rd generation thermal detection gear. Trying to hide a tank or soldiers from individuals who have literally spent months training to covertly observe other military units, equipped with top-of-the-line thermal gear, is going to be nearly impossible.

 

By contrast exchange the RM AMWCC soldiers for some conscripts who, if they're very lucky might have a pair of binoculars between them, and the results will likely be entirely different.

 

I would imagine that the DCS detection API already takes "unit skill" in to account, but I think it could do with perhaps being more granular in terms of assigning "sensor capability" numbers to individual units.

 

Making probability calculations is likely within the realms of possibility, but to do so accurately, especially for ground units v ground units, would likely take the calculation outside of what would be computationally acceptable. For example, visibility is something that even pretty good infantry simulators such as ARMA really struggle with. Calculating visibility and by extension detectability on that level for potentially dozens or even hundreds of ground units in a DCS mission would likely bring even the best gaming PC to its knees.

 

It'll be really interesting to see how DCS models LOS for ground units going forwards, particularly with regards to buildings and trees.

[FlightControl]

Something that I think DCS may not adequately take in to account, but that could probably be implemented at fairly minimal computational cost, would be the exact type of an observer and any equipment they might have.

 

Take for example a covert observation post manned by couple of Royal Marines Arctic & Mountain Warfare Cadre soldiers, equipped with 3rd generation thermal detection gear. Trying to hide a tank or soldiers from individuals who have literally spent months training to covertly observe other military units, equipped with top-of-the-line thermal gear, is going to be nearly impossible.

 

By contrast exchange the RM AMWCC soldiers for some conscripts who, if they're very lucky might have a pair of binoculars between them, and the results will likely be entirely different.

 

I would imagine that the DCS detection API already takes "unit skill" in to account, but I think it could do with perhaps being more granular in terms of assigning "sensor capability" numbers to individual units.

 

Making probability calculations is likely within the realms of possibility, but to do so accurately, especially for ground units v ground units, would likely take the calculation outside of what would be computationally acceptable. For example, visibility is something that even pretty good infantry simulators such as ARMA really struggle with. Calculating visibility and by extension detectability on that level for potentially dozens or even hundreds of ground units in a DCS mission would likely bring even the best gaming PC to its knees.

 

It'll be really interesting to see how DCS models LOS for ground units going forwards, particularly with regards to buildings and trees.

 

DCS does not model vector ray casting between trees and villages.

A simple model is being created that tries to mitigate these issues. I try to keep the factors easy and understandable. Keeping it simple...

[Fri13]

One of the problems is that AI has perfect 360 spherical view. Even if you would sneak up to AI from position that it can't possibility see, it knows exactly what is coming and when.

 

Ground vehicles should have simulated FOV angles and then times when they are being viewed. Meaning a MBT gunner is not looking up to skies ever. A driver can't see anywhere else than in front, a loader is blind as bat. And commander is now and then looking around searching other targets for gunner, sometimes looking up to the skies if even that, through a small prism periscopes.

 

So having an MBT crew to be able engage even an helicopter is very lucky thing if they have no idea where such is coming. Why you would just seek for cover and hope that something else is coming to take that threat away.

 

Same thing should be for fighter pilots, let them get distrupted, have them limited field of view and time required they need to look for specific directions etc. This would make a single pilot work very difficult and possible to sneak up and get on better fighting position.

 

But that is all impossible for now as it would eat the CPU for simulating all that when the situation would call for it.

[FlightControl]

AI having perfect 360° view? Will check that... I was under the impression it didn't for some unit types.

[shagrat]

One of the problems is that AI has perfect 360 spherical view. Even if you would sneak up to AI from position that it can't possibility see, it knows exactly what is coming and when.

 

Ground vehicles should have simulated FOV angles and then times when they are being viewed. Meaning a MBT gunner is not looking up to skies ever. A driver can't see anywhere else than in front, a loader is blind as bat. And commander is now and then looking around searching other targets for gunner, sometimes looking up to the skies if even that, through a small prism periscopes.

 

So having an MBT crew to be able engage even an helicopter is very lucky thing if they have no idea where such is coming. Why you would just seek for cover and hope that something else is coming to take that threat away.

 

Same thing should be for fighter pilots, let them get distrupted, have them limited field of view and time required they need to look for specific directions etc. This would make a single pilot work very difficult and possible to sneak up and get on better fighting position.

 

But that is all impossible for now as it would eat the CPU for simulating all that when the situation would call for it.

No, they don't. It is in the Detection.lua as well.

If you come from behind they "get aware" of your presence much later. (It's a factor).

The issue is that a Group "knows" all what Group Units know (talking to each other, like IRL) so a Group of tanks in a circle formation has all angles covered and thus spots you from every side.

 

A real issue is, once they've spotted you they track you no matter where you hide until you leave the detection distance.

 

EDIT and the damn time factor means in the end they will spot you, it is just a clock ticking down. You have LOS (through trees) it is just a matter of the time running out and they "spot" you...

It is not too unrealistic, as the longer you scan an area, the more likely you are to detect the enemy, but in combination with trees and buildings obstructing your view, but not the AI's it is pretty unfair sometimes.

Edited February 12, 2017 by shagrat

[Pikey]

Anecdotally, 30 years ago I walked into a main battle tank in the dark when it was switched off and covered up (i was looking for it in a navigation excercise and actually hit it and it hurt)

 

And if you turn the buggers on they make a racket you can hear for miles.

 

So moving, not moving should/could be a massive difference in detection for armoured vehicles.