wrong again ! try and use specific words to give meaning to what you are trying to say, instead of this meaningless superficial self gratifying banter your producing which doesnt contribute anything

let me try and rephrase it for you correctly the statement you just made " some contacts are very difficult to spot under some specific circumstances and visibility conditions ". and that is the starting point of the discussion, not the end :)

hint: as a starting point the comparison has to be made in near perfect visibility conditions. once that is done other observations can be made as to how correctly certain weather and atmospheric conditions are represented and how correctly/badly it affects visibility in CoD compared to real life (but that is much more complex, first we need comparisons under near perfect visibility conditions)

try and at least read the previous posts before you "contribute". the difference between static and moving target identification is very significant, and has been listed as a variable that needs to be differentiated when comparing different observations or historical account. we already knew that much, thanks !

I understand what's you are thinking and I think it's valid!

We can still live with the real distance gap. The only problem might be for the purist of the ingame immersion. Think about a seaport: all the ships would be bigger than the whole port, probably of the city too.

This problem can be solved by a using this visualization as a flight mode, for example pressing a key button; because of their size relative to the map, ships and planes would still be icons but at least their model and their vector can be recognized at long distances.

I didn't, Zapatista. I think it's not useful to test ingame circumstances since having the "black dot" as visual target rappresentation is IMO enough to discard the validity of every simulated "visibility" process.

In IL2 you see black dots on every surface or condition: reflections, camos and the rest are not taken in account. I've not really much experience with CloD, but ships and ghost contacts let me assume that nothing is changed since 1946.

The Earth is flat!!

FLAT!!! I TELL YOU!!!

:-)

Ok, I've spent some time to explain my idea.

First of all it's only an idea and I would like to discuss it with you all: I know that I can be wrong (but still I know that the earth is not flat).

This is not about having a 100% truthful visual spotting algorytm: probably it's a 70% realistic method, but still better than the original IL2's one. Here I'm not claiming that 1c, Luthier and his men suck! This is an idea about something that NOBODY did in years of combat sims and could be the one that only IL2 has and nobody else.

It's not a easy thing to develop.

Let's start from the beginning: speaking about both world wars air warfare one of the most used statement is "Beware The Hun In The Sun".
Before the radar appearance the sun was a very important variable. In IL2 we usually try to have energy advantage and a partial positional advantage (being on enemy's rear sector for example), but do we really fly to gain a useful position to not be spotted by the enemies (and so ambush them)?
Do we really care to have the sun behind us?
Do we follow the correct guidelines for our camouflaged plane? Time of day, altitude and surface tipology?
Why was the "Finger-four" a great idea but we use it only because it's "cool"?

For years many of us have been great lone wolfs because of the IL2's easied SA. We are still searching for black dots since they appear at all events.
A white plane over a cloud? -> A black dot.
A plane in front of the sun? -> A black dot.
A light brown plane on the desert surface? -> A black dot.

The only useful thing about this contact representation was for lower planes: a black dot between a myriad of moving pixels could not be easily seen (remember: watching a picture on the monitor our eyes' functions are limited).

And more the visual recognition of the dots is actually instantaneous: Do we really have to patiently scan every point of the sky searching for the enemies? We move our head and quickly we see the threats; all we need is to remember of looking behind us every X seconds... So of course being ambushed is really difficult: how could real pilots still been unawares of enemies when they were flying in multiple plane flights when we, alone, can do better?

So these issues made me think about a correct algorytm about visual contact. Please read again the studies in the first pages of this thread about the target visibility variables and think about what we currently have ingame.

I know that some of you claim IL2 the most realistic WW2 flight sim, but how much realistic is it? Above all about the visual search that was the first and most important task of the fighter pilot? Add to this the sound radar too... I would easily favour average graphic but a great realistic combat sim than the opposite: I know... it's always the same problem: CoD3 or Arma2? But I agree with most of the software developers when they head themself for the greatest market... I'm part of a little market's corner and I know that a developer can't live with so few customers.

Part 1:
The real fighter pilots have a method of scanning the sky... they don't act like us: a plane can be seen if they know where to search even at long distances and, of course, if the conditions are not a malus for them (fog, sun ect). The planes don't appear istantly at their eyes but it have to be on the sky's area they are focusing on.

So scanning the sky needs time (look at visual searching sectors on pg 10 of the PDF, and more after): ingame after 5 seconds we have scanned all around us and have actually seen a contact when in reality you need, as baseline, 5 seconds to scan only a sector of 90° x 45° (Hor x Ver).

I have divided the pilot's sight in focus areas. These are dinamic based on the target distance. Look at the PDF: visual acuity decrease with the distance.

http://img20.imageshack.us/img20/3902/fasce1.jpg

So the accuracy starts from those values who can be increased or decreased by range plus the variables exposed in the thread ( http://forum.1cpublishing.eu/showpos...3&postcount=44 ).

The application run the algorytm every X msec: it computes the value needed to spot a contact and compare it with a random number (0 to 100): if the random number is lower then the plane can be seen (about the needed value: x<0 plane is always invisible, x>100 plane is always visible).. it's difficult to spot a plane? Keep your sight for 10 seconds and maybe you can see it (if you're lucky).
What I think here is to not let the "dot" appears (call it "dot" so for now) if there is not the visual capability to see it. A plane at 3km can be seen, but normally not by peripheral visual acuity (as, instead, we have now). You can't scan the sky around a mountain's peak at your 12oc and spot a distant plane at your 2oc. Human FOV is larger than the game's one, but the game does not distinguish between visual acuity.
So you don't only need to look at the right way, but you need also the time to actually spot something. If you can concentrate to one spot because they told you something is there, then you can see it like it is in real life. Probably you'll need time to do it, but you can spot a plane at 20km (look the data in the PDF) as missing another at 1km.

Does the plane emit smoke? Greatly increase the value. Does the plane fly in the right camo circustances? Greatly decrease it. What about sun and clouds? What about relative motion? Compare and increase/decrease at every loop (starting with distance and acuity to save cycles of cpu).
In the case of windshield reflections (random event if the plane position and sun are ok) increase drammatically the number also OUT of the ingame camera to simulate the peripherical acuity using a temporary disappearing arrow (PDF pg 9).
So the sum of the variables and the comparative method can reproduce a semirealistic visibility. Think about having to scan the rear sector for many seconds instead of the 200 msec we use now (above all with the ugly wide view: and then they tell my about "immersion".. yes, like a fish in the water!).
Think about it having to watch instruments too because of the CEM (maybe a better one I hope).

http://img12.imageshack.us/img12/9458/fasce2.jpg

The tweaking work behind this algorytm can be hard and time comsuming: I never stated it was easy but even if it's not accurate I think it could the greatest news in the world of the combat sims. To me this is like passing from the easy bombing mode of the old IL2 to the newer detailed one: both are not 100% accurate but who can state that the first was more realistic?

Part 2:
This part is about the limit caused by our monitor resolution and the FOV of the game. Many hate the word "LABEL" but IMO they don't understand that the black dots are labels too (BTW look at the ingame ships). We can discuss all you want about them as "immersion killers" and I agree. So what about using labels in a selective way?
Pay attention; this is not about the magic labels of IL2, the ones that appear instantly at distance, the ones who are detailed on model, distance and appear like arrow on your screen!

Simply combine the labels and what explained in part 1 of this post.

You have to scan the sky for enemies or friendly contacts; but because of the resolution issue with FOV you still can's see them if not as a dot. So you push a KEY BUTTON and start to fly in search mode (I would like to add that visual acuity should be linked to a tiredness value to simulate the actual visual skill of a pilot: think about the 4° mission of the day...).
In this MODE labels appear on the screen IF you have spotted a plane, only in that event.
How can be these labels?
IMO something like this:

http://img804.imageshack.us/img804/628/dot1.jpg

Big and visible (customizable?).

Remember: they appear only if you are in search mode and they DISAPPEAR if you lost sight of the contact. So you have to scan again if you lost them, like in RL. Of course if you have to lose direct visual sight for only 2 seconds it's probable that you still know its position and you will search in the same spot... but if in those seconds the guy has changed its position then you lose him.

My idea is that it's not possible to fight at 2 km from a contact and all you see is a dot... in no way! Look again at the real contact size with the one at FOV 70. Talking about IL2 how many times I had to wait for a shooting plane since it was the only thing I could see... "Oh look! somebody is shooting at someone else (maybe, probably, boh)... can't see them DIRECTLY at 2km but I'm sure there is somebody there"
You see the gunfire and you scan that exactly that place and still you don't see anything.

Use this label method: it's not a cheat since you HAVE TO SEARCH for the enemy, but still it renders more realistic the fight because it take away the resolution issues (and nobody have to fly at lesser resolution anymore!).

But there is more. I've already told the issue about FOV: in many games (starting from OFP IIRC, or maybe it was ArmA?) we have the zoom function... we all know that its not like having a robotic eye but it needs to see the true size of a 3D model. All it does is to decrease the FOV (IL2 zoom is at FOV 30).
This work on an infantry sim, where speed is really low compared to planes: A10 and Falcon4 have this feature but it's easier since you have a radar and you actually know where to search in zoomed mode. Instead in IL2 you don't know it... you can search a tank column in zoomed mode, but it's no possible inflight.

How can we use the zoom function? Simply when you are in search mode and you see the dots you can move the visual over one of these so that the green central dot it over it (or near: inside the invisible circle): you press a key button and the game LOCK and ZOOM on that dot giving the real 3D model's size (maybe with labels on the details: model and number)
You release the key button and the visual return to the normal one. The dot get smaller if the planes 3D model is very visible (but not invisible): the more bigger is the lod, me more smaller is the dot.

http://img403.imageshack.us/img403/4396/dot2.jpg

As I said before this is an idea: I would like to discuss with you about the usefulness of this.

Knowing that very probably 1C will not put resources in something similar.

EDIT: I ask the mods if kindly can change the thread's title adding "UPDATE at post #97". Thanks.

it is a great idea, actually very ingenious and clever. It could be a unique feature of Il2: Cliffs of Dover, something that no other game has ...

One minor problem with your description of this feature is that it requires the player to move the view centre to actually move the detailed search field, where often players will simply move their eyes - it's a bit counterintuitive, but maybe it would only be to start with.

I think it would help make a lot of this unnecessary if simply the far distance dot changed colour to suit the prevalent visible colour of the aircraft depending upon the angle at which it was viewed. The furthest visible LOD must simply be a cube textured with a really small mipmap of the aircraft skin. Or even just a billboard of the closest to perpendicular face of this cube relative to the camera. This also makes the dots a lot more difficult to see without relying on a lot of scripting to artificially "set" their visibility.

Dear Manu,

this is an interesting search and I am glad your pursuing your own idea.

However let me just give you a quick hint :
1. Speed and size of the con are of utmost importance here regarding distance at witch you can spot a bandit.
In the diag above, you show us that the cone of visual lock (let's call it CoVL) is of a complex shape : widely expended and then narrowing itself when the distance is increased to a point that it is only a narrow pencil.
At first I wld like to make a remark. Tunnel of vision is what a fighter pilot experience flying low at high speed in a jet. The vision in neat only in the far distance in front of the pilot and the surrounding area is blurred.
Rapidly, without demonstration man can understand that the time available for the retina to "print" an object is the leading factor for the neat vision. If you turn your head 30° out of your axis of view, the object on focus will travel beside you at V/sin30 speed (=2*V). The time for the "retina print" (let's call it RP) will be divided per two. Compared to an object directly in front of you, teh time for your brain to act in reconnaissance/analysis mode will be divided per 8 if we consider that it's a volume print (both eyes are simultaneously focusing on the same points).

You see now why it's so difficult to distinguish a rapid object even if he is traveling right nearby.

Now that we hve introduced the volume factor leading to a full recognition, let's hve a look of the variable ctrling an identification of a shape (the leading cause to the "there is something there" sentence)

In order to get a (positive) identification (let's call it PI), you hve only to see that a point (unreal : there is no point in mother nature) or a surface is right where you are looking at. Looking at the RP factor and the difference in speed in the above example, we see that the time required for the reconnaissance is 4 time superior at 30 deg than at 0deg.

And there is where the nasty camo plays its part.

Remember that we are talking here with surface and that this surface is nothing else that the averaged contrast with the background. If I am flyiong above a forest and look at a bright orange triangle, the time the brain has with a given RP depend only to it's size and shape. Imagine now, that the surface is painted with a finely tuned pattern of color that blend quite well with the background. Here the averaged contrast ratio is perturbed by the difficulty to make an identification of what is the surrounding background and was is not in that very specifically part of the image. This add time that hev to be subtracted from the available RP to get a PI.This can be summed in a blend factor (BF) analytically like :
If Cammo : RP:=RP/k with k being the BF

All the above is fairly basic for you and just an other way to say what you hve alrdy explained to all of us in your excellent post.

Now let's take the problem upside down.

Now we are not flying extremely fast low but at WWII patrol speed that said 1/3rd to 1/4th of the speed we were talking in the above.

Time is a function of distance and speed with t=d/v and reciprocally distance is a function of time and speed d= v*t

without looking at anything less and focusing ONLY on the above eq. we can understand that
RP= f(d/v) | eq. (i)
PI= f{(v*t)*(v*t)} | eq. (ii)

Then we we can say that at a given time (and same RP) traveling at 4v while sighting at d is "like" traveling at v and sighting at 4d.... (Wew what a big step ! I hope you are still there - but remember, don't look at the significance of the value, just at the variable and the way they interact each others)

In the above conditions the PI at v (let's say PI(v)) would be "reached" 16 time faster.

What does is says is that your sensor (eyeball) scanned zone for a given surface would be insistently 16 time bigger in volume than at 4 time the speed. We hve seen (or more honestly admitted or even more made the hypothesis) that the axis vision is not imparted by the speed at witch your pilot travel giving is head is contently up and his eyes are focused right on his vector of motion. Hence the corresponding volume is 4 time bigger at the base - the volume of a cone being a function of PIr² the circular base and 1/3 of h its height.

So when traveling at 1/4teh the speed, your PI planar zone for a given RP depending of the time available for the scan is 4 time bigger.

In your picture, the 100% "look zone" wld be 4 time bigger.

Now the hardest part (I 'll let those that hve read the above lines so far solve the volumetric case):

What abt the scanning distance that obviously impart the shape of the CoVL with target being acquired sooner than an other one just right beside her?

Saying that we give the answer. hehe ;)

There is one thing that we are all aware of is the target aspect ratio. We know that teh bigger the shape present a target the more probable our canon will score a hit. Hence size matter.

Regarding the visual acquisition things are similar. The biggest the apparent shape, the earlier it will be spotted.
But let's see what is that apparent shape about.
We saw earlier that de-cluttering the background is what ease the brain for a given RP. Let's imagine a bright orange surface traveling at 30° offset at a speed v. The "de-cluttered" surface is of a size function of it's shape (aspect ratio) and the distance traveled for a given RP. Let's say now that the same surface in the same geo condition is traveling 4 time faster, the de-cluttered surface wld be more than 4 time bigger ("more" and not equal as there is the trapeze effect well known of any artist - but the diff is negligible).

Hence my PI with a target traveling 4 time faster is 1/4th of a time.

Hummm not weird... In the above we concluded that the faster we are moving the less RP we hve hence the decreased PI [huge case of sneaky editing - sry]

In fact in both conditions the speed is the relative speed of both our pilot and his target. Let's say that I am flying slower and slightly above of my target. For a given time t the corresponding RP will be "'bigger" due to the de-cluttered shape swept by the faster move of the target. Hence a lowered PI time

Now let's see what happens if while still flying slightly above my target, I am flying now much faster than the potential contact. As I am travelling faster than him regarding the de-cluttered zone will be smaller and the target shape will evolve more in a volumetric manner - what we hve seen takes averagely more time to be interpreted (8x).

We can admit that it's a matter of balance averaged by a typical Gaussian curve (the rounded mountain like curve).

This is where Manu, your dispersiveness in target acquisition came from ; the relative speed of the observer and the target.

We can analytically summed this with the generalized form of equations :

PI(t) = f(RP, k, q) with k alrdy defined above and q(v) a factor for the Gaussian effect.

Please that note that k and q are kfor a give pilot a given days in given meteorological condition (ok ok you know that just wanted to add some sophisticated word to my rather simplistic demo :oops: )

My guess would be something like PI = k/q*RP (t) with q like 1/v²1 - 1/v²2 and v1 and v2 the relative speed of the pilot and the target.

SO what about the CoVL : That where I thing my demo has some points of interest in the fact that the CoVL is the integrand in time of the PI with the time being the actual time of scanning. Obviously the instantaneous time disappear with the (i) form of the RP like

CoVL = Int.[k/q*RP] from -15° to 15° in a second = Int.1

hence

CoVL= Int.[Int.1]in distance and speed (=Int.2)

or using IntVol [div f] = IntSurf[f] and some mixing magics

we've got something like IntVol = RP (instantaneous)/(ScanedVol in 1 sec)
with RP (inst) = k/q*(d/v) with v the speed of the pilot and d the decreasing dist btw them
and ScannedVol being calculated with the scanned surface and the distance traveled per sec and a cumulative factor.

The cumulative factor hving been discussed somehow by you earlier and translate the persistence of the target in the brain being correlated with it's actual position with its estimated trajectory.

Something like Cf =c*{1-(TAR(t0)-TAR(t))} * PI(t0)/{(t-t0) *R(Ttrj)}
with TAR being the Target Aspect Ratio at a given time
t0 being the time of the initial scan in the zone of the target to correlate
c reflecting the pilot consciousness
R(Ttr) is the rate of change of target trajectory


2. What the hell is the interest to add so much new variables (my demo lack some and prob not in negligible number) ?

At first, let's remind that a sim world is a world where there is no hazard and everything is known and dully characterized.
Secondly let's remind that using the random hard fction consume time.

In your demonstration 6S.Manu, the target acquisition range is supported by a series of calls to the random fction at each frame that will invariably impact the FPS IMHO.

As we know each plane position, the TAR, the speed and all the value needed, it cld be more profitable to use something in the form of the (very long - sry) demonstration.

Once a PI is scored a target reconnaissance process can begin that as I hve alrdy advocated cld be some form of image dilution from blur to neat (a pre-processed sprite ?). But I know certainly nothing on that relatively to our prof devs.

Only my 2 cents ! :rolleyes:

~S!

@Tomcat VIP

Could I ask you to explain this i layman terms!?
I sounds interesting but you lost me which make it hard for me to get the point.
Please!?

Very impressive paper you put together there 6S.Manu, I respect the amount of thought and work you put in there !!!!!

I must admit, it looks rather compicated to implement (in programming terms) taking also into account the difference of Graphics Cards (or rather say graphics drivers) that exist on the market.
I somehow have the feeling that the "optimisations" NV & ATI put in the different versions of their drivers will have a massive effect on the result.

Following the tip I found in the ATAG forum I set my AntiAlliasing to 0 and was surprised to see how much easier I can recognise the dots in the distance. And with the other tip (MeshDot...something= 1) the combination is perfect.
Yes, I lose sight of planes very often. This makes it particularly difficult. But it is also what makes the game much more interesting... I can no longer stay at 7Km and watch the fighters crawling at ground level.

Besides, if I want to play easy, I can go somewhere with labels turned on.


I repeat, I admire the amount of work and clarity you put into this concept.
And mabe we will see it one day.
It reminds me the big discussions we had years ago about the ammo belts in IL2 and the people who tried to develop the concept and those days we said "forget it! to difficult to implement" and suddently we have it in CoD so, nothing is impossible! :)

At the same time, I am extremely happy and I enjoy what I have now; Which is a lot more than what I had with IL2FB :)

With the settigs I mentioned above (and the ones I posted in some other threads) I fly my 3072x1024 resilution (three monitors) on a single GTX570 with 80fps high, 50fps low on the ATAG server (big map) and enjoy flying and dogfighting without struggling to recognise dots far and near (and without having to restart my gme after every every mission because of the memory leak).


~S~

I know it is off topic, but I really enjoy the ability to change ammo myself. I always wanted to try out a tracerless combination and now I got it. I am no longer distracted by my own tracers :grin:

@TomcatViP
Thanks for your post Tomcat!

I've to admit that I've some difficulties following your argument entirely, I'm asking to one of my teammates to explain some parts of it in Italian. Shame on me. :-)

Anyway as I said in the post above my target is not to have a 100% fidelity with reality, so I tried to keep the things simple. Of course a more detailed algorithm is gladly welcomed!

If I understand correctly CoVL heavily depend also on the observer's speed/altitude and these are the main factor for the tunnel vision. I think that it should be simulated. In terms of CPU usage we should test what it's better: a formula that uses all the real time values (CPU) or a model based on tables with defined and fixed values and approximations (RAM).

The part about the target shape is already explained in the first page of the thread, where is a formula (provided by one teammate of mine) to have the max distance based on aspect ratio.

Also camo is being taken in account in the initial analysis here: http://forum.1cpublishing.eu/showpos...3&postcount=44

For the part about "relative motion" I think your argument is valid (at least the part I've understood, but I've faith in your work) and its the matter I was mostly concerned about.

As you can note there most of the things are not detailed (for example I don't take care about camo design but only the colors of the skin in that aspect, but it can always be a parameter linked to the skin itself) but IMO they are enough to give us a semi realistic output.

If you focus at the part about contrast you see that it's a function only lightly based on real time values and it's not related to the video drivers (to answer at GRAthos about video driver issues): the algorithm I was thinking runs around a parallel 3d world (matrix here) there the data are mostly fixed and its affected in real time only by really few variables.

For example the part about "contrast":
We constantly have the matrix of plane aspects (A): this one contains the reflections value for each aspect and mostly the ratios of colors of the skin in that aspect. This are not real time calculated values and I would put it on directly the plane model package.

We have constantly the map matrix (B): the map is divided in sectors (area has to be defined) and each one has an array with "ratios of colors" similar to the one the plane has.

A and B (but really only a fraction of A, the array corresponding to the plane aspect) are affected by lighting in real time. Here you compare the resulting colors. The plane is not "spotted" if it exactly passes over a street or a lake like in RL, that would be very difficult to calculate IMO, but if the difference of the colors between actual aspect and map sector gives us a probability value (high difference = 100%, no difference in colors = 0% and some ratios).

Because we have 2 main problems: CPU usage and mainly there are too many variables to calculate in RL and both can be avoided by approximations (data tables) and sadly in this case you have to use the concept of "probability".

This is the reason I used rand() functions (or optimized ones): today is not anymore a cpu consuming function. I read that today it is possible to have 1,000,000 random numbers on less then 50ms. In my algorithm a rand() is called for every plane at the end of algorithm, and only if the conditions are the right ones (if the plane can be really be spotted and it inside the dynamic CoVL).

Then running this algorithm in a different thread (requested every 100ms for example, asynchronous at the main process) and on another core you can have good performance IMO. We actually have PC with high end CPU chipset and 6-8-12Gb of RAM: I think that performance are not a problem anymore.

We all know that today the majority of the issues of a videogame are on the video matter: I think that anything can be done in these days... look at that MODs did with IL2 1946. The biggest part I was annoyed about CloD after the release was the very limited CPU usage and really little RAM needed (but not VRAM...).

The most important thing to remember is that not everything can be simulated and so we need to work by approximations. Your analysis about realitive motion and RP seems very valid and useful for this matter, above all because is the part that mostly scared me.

@GRAthos

Of course this can be complicated above all because working by approximations needs a lot of tweaking and beta testing. But it's in no way related to video driver.

Sincerely it could be in some parts as the one about camo, but here the developers should work like they did for the epilepsy filter (reading the output IIRC): knowing the result of their effort I think that it's not the road to head on (or maybe it is... it could be that it's impossibile or that the developer was not skilled enough, who knows).

Anyway I hope you agree with me that something like this would take this WW2 sim to another level of realism.



I apologize with all the readers for my bad english: I use always the same words. I'm bad in Italian too... Again, shame on me...