reference point/color for aero comparison

[vyselegend]

While browsing through our numerous topics that aim to compare cars, I was regretting that differences in cars position and liveries makes the comparisons so hard.

Obviously the idea of virtually painting all the cars in the same color (please not red ), and positioning them exactly the same is not original, it must have striken each of you, at least once.
This lead me to think it must be harder than it looks, otherwise such forum as ours would be overflowed by "standardised" pics like that. So what is the problem? Can't we give it a try? (By "we" I mean "you", of course, as I'm totally useless in this field)

Finding pics of different cars from same viewpoint (angle, distance, and shot with same apperture speed and wideness) is possible, but then we nearly never have the whole field in those pics. (bar555 often manage to find some for his "photo comparison" topics). For exemple on this dutch site there is a good deal of pics of different cars from same viewpoint, particulary side view.

Then, changing the colour of the cars to a "reference" one, from a bitmap image, should be feasible, given how easily certain members (manchild comes to mind- remember last year Renault's livery topic) can play with the cars schemes and colors.
In an ideal world the original pics should have the same exposition to light, because placing the shadows on the same places would be the top! Otherwise moving the shadows must be hard, but I guess people used to photoshop work can deal with this task, no?

It would really be great to have such standardised pics from every new contender in the same thread, for optimal comparison.(organised like bar555's threads) I think Honda's virginal white could be the best possible color, because of it's clarity.
Then the obvious reference point for car size and position should be the Bridgestones tyres.

Well, I'm shamefully asking for others to do the work, but I'm worthless for this task. I have in mind members who have proved photoshoping abilities, but anybody can help, particulary for finding the pics. Is someone interested?

[manchild]

That can be done but since such comparison is of no real importance or value no one would spend many hours to make photo-realistic comparison especially since teams change aero so frequently during the season. If I've made such comparison between BMW F08 and some other car after the launch, by now it would mean nothing since BMW added completely new nose and additional wings between suspension elements.

All those cars may look interesting and intriguing since they are new to us but in a year or two they'll be just statistics no one cares about. In a month, original launch aero will have no importance since most of the teams will change it. So basically, I'd do it or someone else would do it on regular basis only for the money but not for the fun because it takes a lot of time and wastes resources without any practical value.

[checkered]

There's a range

of suitable tools in Photoshop and similar programs. To make the task simpler, perhaps it'd be prudent to discard all colour information first and work in grayscale only. I can imagine employing adjustment layers, using saved selections made with a combination of the magic wand, range selections, masking and (in the worst case scenario) going over the shapes manually. One could achieve what you're suggesting, it's not impossible. In 2D, each operation done in this "humanly fuzzy" way will be "information lossy" to a degree, though, resulting in an "educated approximation".

At least I don't know of filters that would automatically look at continuities in lightness and contrast values between adjoining "RGB/CMYK value borders" in order to discard and replace different colours with a correspondingly adjusted monotone. That shouldn't be impossible either, since there are many filters that are used to find a variety of user defined limits ... this would quite likely have to involve some sort of a dynamic limit value search and combining the functions of several filters in this way. This might even exist somewhere, it's been a while since I last had to take a halfway serious interest in heavyhanded 2D image processing. But I guess by now you have some idea of the effort involved ... these are incredible tools, no doubt, but the subjectivity sometimes makes it challenging still. Reflective surfaces and distortions would present an even greater challenge.

In the "spy" documentation McLaren's staff made references to teams taking very high resolution images of competitors' cars. This would actually make the editing process easier, since small imperfections will "vanish in the noise" with a little dimensional downsizing of the image. I'm afraid that the web images are of such poor quality (even the "large" ones) though, that all border areas have a relatively high percentage of "hue mixed" pixels to the complete area that cannot be disregarded for quality and actually increase the amount of detailed "manual" work. Would this "monotoning" ease the comparisons significantly? I don't know. It's an intriguing thought I have to admit, but I wouldn't recommend anyone doing it unless someone uncovered a fairly universal method of achieving this, so that the effort could be shared. Eleven teams ...

I also suspect that certain teams have ways of making fairly high fidelity 3D digitisations, either from 2D images or even with dedicated equipment. This would be easier to do with objects with clear borders, such as wings. To a person allocating resources within the team it must be a tough call to decide how much to take away from one's own R&D to keep track of other teams' doings.

[vyselegend]

Ok, now I have a clue of the amount of work involved.

And you're right MC, cars are often changing through a season, although overall they keep a general philosophy.

But I kind of disagree with you when you say such comparisons would be irrelevant because of that. I think if you take a comparison between let's say a MP4/22 and a F2007 from winter test, then the same comparison between those two cars, but from Interlagos, you should be able to do the same remarks about what differs in those two. Both designs will have evolved through the season, but each following it's original philosophy.

I hope you get what I mean, maybe I'm a little confusing, but I try to insist about the gain of doing it.

I also suspect that certain teams have ways of making fairly high fidelity 3D digitisations, either from 2D images or even with dedicated equipment. This would be easier to do with objects with clear borders, such as wings. To a person allocating resources within the team it must be a tough call to decide how much to take away from one's own R&D to keep track of other teams' doings.

I also think so! And not only F1 teams actually. Look how often TV channels show up with satisfying 3D models of the whole cars, in those animations before the GP. Surely they don't mesh up the totallity of a F1 car from studying photos from every imaginable angle...

So I guess they have some kinds of softwares able to deduce the depth of an element (by reference with another), just by calculating the distance between elements, (and "scanning" the density of the shadows?) from very high-res pics taken at different angles. (LOTS of pics I admit).

It has to be that way, otherwise it means team does provide all done 3D models (labeled "satisfyingly incorrects" in order to protect secrets) to the medias.

Which is the most credible

Anyway thanks guys for your answears.

[bar555]

Quite challenging I would say , we can work on one example to see the possible potential .

I am in if there is help from you guys

[checkered]

I've had something to

do with stereographic images in some past activities, that is one possible platform to digitise 3D information "on the fly". Btw., with clever maths I think it's within the realm of possibilities that shapes can be approximated very accurately from fairly low resolution moving i.e. sequential images. I've sometimes toyed with the possibility of bringing old film footage into 3D this way ... give some reference points and that's it.

But what I found with a very superficial search into recording and processing raw 3D data from physical objects really shocked me. Now there's absolutely no doubt for me that all F1 teams have this capability - it's just too easy to acquire and make use of. If they have the processing resources available, they're bound to test their competitors' cars with CFD. This also raises questions about McLaren's "metallic" livery as a "3D countermeasure" (I did float this idea at its introduction while in the BBC 5Live messageboard a long time ago).

In fact, it's so easy it might well be possible for some F1T members to acquire 3D data from "pitwalks" or such, clean the mesh with shareware programs easily available and supply the data here for public appraisal! There are many technologies available. No more tinkering with photo editing software ... "live", editable, testable 3D data. How's that? Below, I've provided a very disorganised link list. You'll easily find more with search engines. Much more.

http://www.photomodeler.com/index.htm
http://www.3d3solutions.com/flexscan3d.php
http://www.thinglab.co.uk/compare.php?SubCatID=30
http://www.david-laserscanner.com/
http://www.simple3d.com/

This is a very fast moving field. I'd expect 3D features in basic digital cams very soon.

[Carlos]

A Freeware 2D>3D photo-converter. No experience. Just found it a few minutes ago, after reading checkered's post. OT - Anyone found high-res R28 pictures for bar555?

http://software.techrepublic.com.com/do ... cid=334841

"Overview: The 2D to 3D Converter application was designed to convert a 2D picture into a quasi-3D false-height relief in (almost) true perspective. Just select any .JPG or .BMP picture and wait for the 3D image to take shape."

45 Freeware Graphics/Photo/Animation/Conversion/Font/Flash Programs
http://www.snap2objects.com/2007/07/20/ ... n-programs

[vyselegend]

Me again. I'm glad there is some enthusiasm finally, maybe it will work in the end. I registered on the site Carlos linked to, in order to download the program "3-D shade" , and had a try at it.

It's weird.

There's no proper installation, as soon as you click the executable file you downloaded you land on this interface:



I then loaded this pic (the same I just uploaded on the R28 thread). After a few times waitng it gave this:



I guess my PC config is a bit outdated, maybe the program need more resources, (I run on an old P4 3.2 mounted on an Asus P5-P800 motherboard, 1Go RAM, and my graphical card is an ATI Radeon 9700 pro) at this state I couldn't do anything but print screen, then escape the program, as none of the keys the interface spoke about would work... I tried with other pics, with similar result.

Did anyone else try it?

[checkered]

Did anyone else try it?

No ... and frankly

I'd be flabbergasted if anyone had come up with a program that could universally "3Dimensionalise" any 2D picture with no other references whatsoever (than a single picture). Interesting effort, nonetheless. I know of facial image recognition/reconstruction software that does operate with very little 2D data, but even those rely on a certain set of given parameters (the facial geometry will vary only so much within and between its elements). To my knowledge the best ones achieve well over 90% 3D accuracy from a 2D reference.

While first searching for information I fleetingly saw a couple of academic websites that had something to do with 3D modelling from pictures, those might also contain links to freeware. Apart from that, I guess we'll have to make enquiries about the costs of commercial programs ... right now I've got no idea whatsoever what the range might be. One could of course do an "inverse" perspective by hand also, but that's quite labour intensive. I'll continue looking into this within my means, this is interesting by its own right.

[Treead]

@checkered: Computer imaging happens to be my field. You're right, there's webcams out there that can do basic 3D imaging. The cost of these systems are reasonable too, but they wouldn't be able to generate anything with high enough resolution for use in F1.

http://www.tyzx.com/
http://www.3dvsystems.com/index.html

The first system uses stereo camera, 2x 2D images with a known distance to calculate the z-axis. The second (might be more scalable to F1 use) uses flashing IR light with an IR camera to get the z-axis info.

[vyselegend]

Did anyone else try it?

No ... and frankly

I'd be flabbergasted if anyone had come up with a program that could universally "3Dimensionalise" any 2D picture with no other references whatsoever (than a single picture). Interesting effort, nonetheless.

Well, you sounded so surprised about the level of 2D to 3D modeling softwares that I got a little too transported. I myself was very disturbed that it might be possible to do, but my naiveness took the uperside of me.

Still, if some algorythm can make a software modelise a human face from a single pic, why couldn't it do the same with a F1 nose?
There is still more room for investigation, I just tried this soft, but there are some others in the links you provided. When I'll have some time for me, I'll try them.

[checkered]

@checkered: Computer imaging happens to be my field. You're right, there's webcams out there that can do basic 3D imaging. The cost of these systems are reasonable too, but they wouldn't be able to generate anything with high enough resolution for use in F1.

http://www.tyzx.com/
http://www.3dvsystems.com/index.html

The first system uses stereo camera, 2x 2D images with a known distance to calculate the z-axis. The second (might be more scalable to F1 use) uses flashing IR light with an IR camera to get the z-axis info.

Wow, computer imaging

can certainly be a tremendously challenging and theory intensive field - from what I can tell. But I guess that also means it's not short of opportunities. Thank you for posting the links, I've bookmarked them and the technologies also got me to try and grasp the "playing field" of visual information derived 3D modeling. Not an easy task. My initial surprise was mainly because I had toyed with such ideas some years ago without really going into it and now that I actually took a slightly more active approach it seemed like the field had "exploded".

What I was going for here was visual "non-contact passive" approach, perhaps being able to use only a few high-res images to derive 3D shapes with a significant accuracy. User-assisted image based modeling thingies certainly seem to exist and seem promising up to a degree, with some limitations. In the case of wings which have shapes with complex tangents that don't correspond to the shapes of the extremities, additional info would have to be extracted from somehow being able to compare different reflections etc. of a stationary object from at least two angles. Highly detailed and complex liveries would yield information more easily. Real stereoscopy actually requires dedicated equipment to record the object. The existence of these triangulation based methods wasn't such a surprise and perhaps the only development I'd have expected to emerge even sooner.

I had seen structured light and modulated light equipment featured in science mags, websites and programs but had only the most superficial of ideas as to how those worked. It's highly unlikely (well, impossible actually) I'd have the possibility to go all "James Bond" on this anyway, stealthily going about pitlanes projecting various low intensity carbon dioxide laser grids on F1 cars and recording the results, for example. But I'm taking it as a given that no team would be impressed by a stranger appearing by their car and brazenly aiming strange equipment at it anyway. Time-of-flight (i.e. a kind of a photonic sonar) I didn't even imagine as very promising/possible given the infinitesimally small (well, near relative) time frames involved. These devices seem tremendously accurate for the "ultimate speed challenge" they are set to utilise.

Some of the theory involved is as good as SciFi, if not better. The ramifications seem to go as far as being able to make electromagnetic wormholes with metamaterials. I've only leafed through a couple of pages of Inverse Problems and Imaging Mathematics and feel humbled above all, coupled with some childlike awe. Fortunately modeling F1 cars doesn't propably require dedicating oneself to this depth of theory ... my impression that digital cams will soon feature 3D has only been strengthened. When it happens, pointing a lense and pushing a button doesn't really do justice for the technology. But perhaps the inventors' account balances will make up for the commercialised trivialisation of their effort.

Well, you sounded so surprised about the level of 2D to 3D modeling softwares that I got a little too transported. I myself was very disturbed that it might be possible to do, but my naiveness took the uperside of me.

Still, if some algorythm can make a software modelise a human face from a single pic, why couldn't it do the same with a F1 nose?
There is still more room for investigation, I just tried this soft, but there are some others in the links you provided. When I'll have some time for me, I'll try them.

I think you just beat me to it. I'm a bit careful with software trials, so usually I take a few days to appraise whether I really want to do these sorts of things. It's propably one of my shortcomings, not being too empirical with stuff. Good point with comparing a "template" face and a "template" F1 nose - that could be one starting point.

[Treead]

Found a cool post about 2D to 3D imaging.

http://www.techeblog.com/index.php/tech ... ser-video-
(make sure there's a '-' at the end of the link or it won't work)

Check out the video, it's a software that takes a large samples of 2D images and create a 3D world. Letting the user browse the photos in 3D. Sorry about getting a bit off topic here because I think it's still a long way to go before this could be used for getting dimensions for F1 aero parts.

[bazanaius]

Hey guys, new to the forum but think I might have an interesting idea someone could try.
Was sat next to a guy at dinner the other night - he is doing his fourth year engineering project on almost exactly what you guys are talking about! He is attempting to recreate 3d worlds using photos taken at high altitude. He had some interesting points that you guys might be interested in.

Firstly, things like cars are inherently difficult to work with becuase of their reflective nature seemingly distorting the depth calculations (McLaren is particularly shiney).

From photos a lot of factors need to be considered, such as focal length, lighting etc. and generally in order to successfully model a part you'd need 11 points of commonality over as many photos as possible.

For this reason, he suggested that video is often the best way to do this - a panning shot provides high numbers of images, lots of points of commonality and fairly uniform picture characteristics.

Thoughts?

[vyselegend]

Hi bazanaius, and sorry for the late welcome on the forum. Usually our fellow members are warmer, but I guess nobody had anything to say on the topic, and I was away for some times... This week-end I had a conversation with my cousin who is graphist, and he held the same point of view as your friend about the compexity encountered by a software to cope with reflective surfaces and to "understand" the subtilities of a shape composed of slight and worked curvatures.

He told he's sure it would be achievable with the powerfull software/hardware combo checkered reffered to (I mean those used by surgeons for facial reconstruction simulations).
He joked about the fact we just had to buy some insanely expensives silicon stations and hack the "facial 3D" soft's source to adapt it to deal with F1 shapes, which, he said, should be piece of cake...

Anyway, no matter my cousin sucks at trying to be ironic, the point is we are limited by our means, but the project is achievable in the absolute.

You're right that a video clip should provide more chance to comput some 3D image of the model, but I haven't seen any free soft proposing to work with animated media instead of pics, in the end it might be our job to extract every pics from the clip so it means no differences. There would be an advantage, I think, if we get some of those "old school" tools used for motion capture in the 90's.(where you put some little "reference balls" on a model, for the soft to follow the movements of given points, hence to identify same elements viewed from different angles.)

Double-thinking about it, I'm sure it must exist. Maybe I'll search for some later.