You are viewing an archive of the old fxhome.com forums. The community has since moved to hitfilm.com.

Color tracking effects - brennanmceachran joins - 65%

Posted: Sat, 21st Jan 2006, 6:03pm

Post 1 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Hello!

Maybe someone has already suggested this feature, but it would be REALLY nice if there were some way to make an effect's position track a specific color. (I assume this is the right forum to post suggestions?)

For example, let's say say I want to do the standard light saber fight. Let's say I paint my light saber green and shoot some footage.

In VisionLab (or EffectsLab), it would be really nice to position the neon light effect over my green light saber in the first frame and then tell the software to make the quad track the green light saber.

To me, this feature could have a great number of uses. Just imagine being able to carry a torch around with a green (or other color) end and having EffectsLab or VisionLab automatically move a flame to follow the torch end!

Last edited Wed, 5th Apr 2006, 2:21am; edited 9 times in total.

Posted: Sat, 21st Jan 2006, 9:50pm

Post 2 of 40

Serpent

Force: 5426 | Joined: 26th Dec 2003 | Posts: 6515

CompositeLab Pro User EffectsLab Pro User FXhome Movie Maker FXpreset Maker MacOS User

Gold Member

In that case it wouldn't work, but it would work in CLab for tracking, and in ELab for stock footage and particles. Lightsabers can't track, deal with it. They have already talked about this, and it isn't tracking to color, it is tracking to pixels. Example, if you have a corner in your shot, track it to the corner so the trac square covers that corner, it will move to keep that pixel style in the track square.
Posted: Sun, 22nd Jan 2006, 5:39am

Post 3 of 40

Axeman

Force: 17995 | Joined: 20th Jan 2002 | Posts: 6124

VisionLab User VideoWrap User PhotoKey 5 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXhome Movie Maker FXpreset Maker MacOS User

SuperUser

Yeah, lightsabers are near impossible to track generally, because the blades are so thin and move so fast; they often become nothing more than a faint blur, which can't really be tracked. As far as the torch, it could work well for that. It would be nice. In the meantime, though, you can simplify your tracking by shooting with that in mind, and using markers. If you have a torch stick with a very definite point marked on it (like half of a ping-pong ball, or a spot of paint) then you can place the origin point of your effect on that point in each frame, to simplify getting the movement right.

Remember to, that it is often worth it to put forth the extra effort to get practical effects. In the example of the torch, even if you did have motion tracking to plop te flame in for you, you would still need to individually mask each frame to get the light cast properly onto the actor who is holding it, and the surrounding environment. There is no way to automatically track where the light from a torch would land on the guy holding it. So it would probably be easier to just make a torch and light it, and all of those compositing difficulties and issues would not even come up.

Don't get me wrong, I think motion tracking is great, and would love to see it, but it is not always the best solution.
Posted: Tue, 24th Jan 2006, 4:42pm

Post 4 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Light saber tracking... I think I've almost figured out how to do it. (Yesterday, I decided to write a small program to see if I could do it.) So far, I can correctly identify the location of the light saber, even with a fast-moving, low-contrast saber, I just have to get the quad calculations finished.

Here's some tests I've done: (Even though this saber looks pretty high-contrast, it's really not. There's an awful lot of yellow in the background.)
http://www.streamload.com/roozer/test1.jpg
http://www.streamload.com/roozer/test2.jpg
http://www.streamload.com/roozer/test3.jpg

The red circle was drawn by the software to indicate the rough center of the light saber. If I can get the rest of this figured out, are any of you willing to play around with it and give me your feedback?

Yes, torch tracking... If you put one of those 25W yellow party light bulbs (or higher wattage) at the top of your torch, wouldn't it light your face, etc. properly? (That way, you wouldn't have to mask/grade each frame. Plus, it would be really easy to track!)

Well, tell me what you guys think!
Posted: Tue, 24th Jan 2006, 6:17pm

Post 5 of 40

Joshua Davies

Force: 25400 | Joined: 21st Mar 2001 | Posts: 3029

VisionLab User VideoWrap User PhotoKey 5 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXpreset Maker Windows User MacOS User

FXhome Team Member

Hi Roozer,

Good to see someone having a go - I look forward to seeing how you progress. There are some possible problems I can see from what you've uploaded.

You're test image is rather flawed when it comes to testing tracking on video. Its best to switch to the worst possible footage as early as you can as it will often provide a totally different set of problems compared to good footage like yours. The main problems I see are:-

1) There is a very clear difference between your background and foreground.
2) Your footage appears to be progressive or deinterlaced rather than interlaced which nearly all video cameras output and nearly all TVs need (to be considered broadcast quality).

From what I can tell your point isn't really central as much as just being inside the yellow area - have you progressed on this at all?

Tracking (or finding) a point inside your bright yellow blur is not the same as finding the outer points of the quad. The shape won't even be a quad most of the time or have any significant edge as yours does.

After much experimentation we've found, just like ILM before us, that the only effective way to draw a lightsword is by hand.

Coloured sticks can make it easier for you to see the stick and remember which stick belongs to which character. But the situations where you will be able to get a tracked sword which looks half as good/smooth as a hand drawn one are few and far between.

I look forward to seeing how you progress with your experiments. If you come up with something that works really well (which the community at FXhome find useable) maybe we can try and integrate it in to the programs at a future date?
Posted: Tue, 24th Jan 2006, 6:55pm

Post 6 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Thanks for the suggestions!

One problem... I can't seem to get an interlaced frame... I'm using Pinnacle Studio (I know, it's not very good :-) and it de-interlaces video when you capture it... I tried using AmCap and extracting frames in VisionLab, but that didn't work either. (The camera shoots interlaced.)

I know ideally, a tracker would work on any footage. Obviously, some footage could be quite difficult to work with. Would people still find this feature useful even if it required them to shoot in front of a greenscreen? I know it kind of limits your camera angles/etc... Humm... I guess it would only work in VisionLab too.

I guess I'll try for the bad footage and then it might work in front of reasonable backgrounds... (Large contrast between stick and background. Boy, this would be easy if people used Black&White sets like they did for TRON :-)

Last edited Thu, 2nd Feb 2006, 6:40am; edited 1 times in total.

Posted: Tue, 24th Jan 2006, 7:05pm

Post 7 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Hey, would there be anything wrong with requiring a stick to be a color quite different from anything else? (I.E. Use a green stick if you were shooting in a desert, use a yellow stick if you're shooting in the dark with a background that doesn't contain yellow, etc.)
Posted: Tue, 24th Jan 2006, 7:17pm

Post 8 of 40

Joshua Davies

Force: 25400 | Joined: 21st Mar 2001 | Posts: 3029

VisionLab User VideoWrap User PhotoKey 5 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXpreset Maker Windows User MacOS User

FXhome Team Member

The problem is, at anything other than very close distance, the colours on the stick will not be picked up enough to make tracking work very well.

To the eye there might be a relative colour difference, but camera store colour at fairly low resolution made even worse by compression. Couple this with a background which will have multiple colours (sky, grass, rock, shadow, sun) and there is no one colour which will stand out.

At a distance where you can see both fighting characters from head to toe the sticks might only be 4-8 pixels wide, couple this with interlacing on the vertical and colour sampling and the actual stored image of the stick might only be a couple of pixels for the width of the stick!

Then you add in blur, back lighting or reflection and the stick can show almost none of its own colour compared to whats around it.
Posted: Tue, 24th Jan 2006, 8:35pm

Post 9 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Thanks! I'll shoot some footage from a distance and see what I can come up with.

I have a new approach to the problem, so we'll see. (Before, I was doing more of a statistical analysis. (Find points within a certain range of your target color, perform a linear regression on the points (which fairly accurately obtains the saber axis), find the highest density of target points along the regression line, call that the center, and attempt to trace the edge of the saber.)

Now, I'm going to try to get the computer to find what characteristics define your saber... Could be kind of hard. Anyway, after that, I'll hopefully get a list of my target points, group them by regions, attempt to find the region with the best least-squares fit (most linear set of points), find the length of the saber (along the regression line), find the saber edge, find the angle between the saber edge and the regression, repeat for the other edge, and then give you a polygon.

I'm sure that didn't make any sence, but I have to go to school.

We'll see...

Thanks!
Posted: Wed, 25th Jan 2006, 4:34pm

Post 10 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Well, I shot some footage outside and ran it through my new software:

http://www.streamload.com/roozer/In.bmp
http://www.streamload.com/roozer/Out.bmp

(Sorry that they're bmp... I just wanted to let people actually see them, not a compressed version.)

The main problem right now is there's snow outside. I tried a white stick so I could really have some bad footage to work with, but I could barley see the stick with my own eye, so I moved to an orange roll of wrapping paper :-)

Does anyone have an real footage? Just a few seconds of video would be great. (I may try tracking based on movement between frames.)

Here's what the software did for the above test: (It is highly unoptimized right now.)

1. It loads the red, green, and blue values for each pixel into an array.

2. The user clicks on the light saber and the software obtains the color of the pixel the user clicked on. (An average of several points would probably be better.)

3. The user types in the maximum %difference.

4. The software scans the all the pixels and tags the ones with a % color difference less than the %difference entered in step 3. (%difference = ((PixelColor - TargetColor)/TargetColor) * 100. Any pixel with a %difference less than that entered in step 3 is tagged.)

5. The software scans all the tagged pixels and groups them by relative distance. In other words, it places the first tagged pixel in group 1. It then looks at tagged pixel #2. Is the distance between it and any pixel in group 1 less than 25 pixels? If yes, it places it in group 1. If no, it places it in a new group (group 2).

6. The software places red dots on the picture to indicate all the pixels in group 1, a different colored dot on all the pixels in group 2, etc.

7. The software would run a linear regression on all the groups to see which one has the highest R^2 value. (It finds the group that most forms a line. (I haven't added this yet. In my example above, the linear regression would draw a line pretty much right down the center of the stick.)

Anyway, I won't know if this version is close to working or not until I get some real footage :-) (Which I can't get with all this snow outside :-)

If this version doesn't work, I currently have two more ideas to try before I give up.

Oh, would you guys as FXHome be interested in source code for tracking something like a torch (like I was talking about earlier)? A modified version of this current saber tracker would probably work. If you want it, just tell me where to upload/email it.

Thanks!
Posted: Wed, 1st Feb 2006, 11:30pm

Post 11 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Just to keep you all updated, I’m still slowly working on this software… It kind of got put on the back burner these last couple days as I got too involved with another “Conquer the World” in “Rise of Nations”… Now that that’s over, I should be able to get to bed at a more reasonable hour. smile

In order to continue this software, I’m now working with these new requirements: (Hey, if people have to work to shoot greenscreen, they should have to work to shoot saber fights, too, shouldn’t they? smile )

1. The light sabers must be at least 1” in diameter

2. The light sabers must be a solid color, highly contrasting the background. (Florescent orange is a great color for out-door scenes while florescent yellow is a great color for indoor scenes.)

3. The lighting must be such that the sword never inverts colors. (In other words, there must be enough lighting that the saber stays approximately the same basic color. Variances in intensity are no problem.)

4. The footage must not move too fast… I believe this is more a function of light than anything else.

Now, for the news…

I’ve re-written the framework for the program, allowing the software to process a frame at about 1 per second on my Athlon64 3700+. Once I actually compile the software, I’m hoping for a dramatic speed increase. (Right now, it’s being interpreted.)

Currently, the only interaction the user needs to make with the software is loading a picture file and clicking on the light saber. (No, you would not need to click on the saber for each frame.)

The software automatically adjusts boundary levels (internal values used for locating the saber) based on the current frame. (Theoretically, changing lighting and speed should require no user interaction.)

In my tests, the software can almost always accurately locate a point on the saber even in footage where the saber is difficult to see. I’ve gotten the software to accurately locate the saber in footage with a fast-moving, low-contrast saber, but this can require a lot of user intervention.

I’ve almost gotten the software to accurately locate the saber edges in good footage… I just need to work on it some more.

I would include some screenshots, but it doesn't look much different. It sure works better, though!

The hardest part so far is figuring out how to make everything relative... (How to get the computer to automatically adjust parameters so it can still detect the saber in varying conditions.) I'm still working on this for the saber edge detection. I think I figured out how to use a mathematical model for locating the saber, but I still haven't figured out how to find the edges. (Well, actually, I have, and it pretty much works, but I don't like it.)
Posted: Tue, 21st Feb 2006, 4:31am

Post 12 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

Holy crap, I'm so impressed. I'm loking forward for future updates. If you need anyhelp say with some testing or somthing, I'll be here. I hope you get further with this very impressive so far keep up the good work
Posted: Thu, 23rd Feb 2006, 11:13pm

Post 13 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Sorry... I got side-tracked. I'm trying to design a control system for a 4-rotor helicopter... I found out last night that I have to spend more money to get it to work. (Insert coin to continue...) I think I'm going to try gyros instead of accelerometers for stabilization. Also, hardware PWM (HPWM) would really be nice.

Ok, probably none of you know what I'm talking about. Those of you who do might be interested in www.rctoys.com You might be interested in www.rctoys.com anyway. You could get some really cool footage and a cool toy if you have a grand or two in your pocket. smile

Anyway, I hope to get working on this project again sometime next week. I kind of lost interest in the project once I got the program mostly working. (I loose interest in any project once I figure out how and just have the “manual labor” (programming) left.) But, since several people seem to be interested in the project, I'll see if I can actually get myself interested enough in it again to finish it. The main problem is I’ll probably never use it smile (Light sabers are getting pretty boring to me. True, you could use this for just about any effect, but I’d probably only use it for simple tracking effects such as a torch.

By the way, how many people out there have more than one DV camera? (I’m trying to figure out how to shoot footage in 3D… Bye bye, masking! Hello, 3-D footage! (Possibly even bye-bye greenscreen? (More on this later if enough people have two cameras.)) Hey, they do this in robotics. (DARPA, etc.) Why not use this for movie-making? It would sure cut down on masking effects, 3D objects, etc.
Posted: Thu, 23rd Feb 2006, 11:19pm

Post 14 of 40

Hendo

Force: 13107 | Joined: 16th Sep 2004 | Posts: 848

VisionLab User VideoWrap User PhotoKey 5 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXhome Movie Maker FXpreset Maker Windows User MacOS User

FXhome Team Member

Rating: +1

Roozer wrote:

By the way, how many people out there have more than one DV camera? I’m trying to figure out how to shoot footage in 3D…
This may be of interest:

http://www.21stcentury3d.com/press/pr-060117-3dvx3.html
Posted: Mon, 27th Feb 2006, 1:23am

Post 15 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

Awsome.. Be sure to post regualr updates!
Posted: Mon, 13th Mar 2006, 7:01pm

Post 16 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Sorry it's been so long... I just started working on this again Saturday... I'm not any closer yet. (So, sorry to make you all look and then find nothing smile

Edge detection has proved to be quite a bit harder than I originally anticipated. I did come out with a version which worked, but it wasn't very robust... at all... Like totally bad... Well, at least, it would get your points roughly in the right place for you smile Well, sometimes.

After a bit of playing around, I think I discovered a new approach. I'll try to use edge detection techniques commonly used in graphic processing (VisionLab has this too), but give the algorithm something additional to work with. Since the software can almost always locate a point on the saber, I'm hoping I can give the edge detection techniques some hints on what to look for... (Ok, they're not human smile ) I also realized the other day that defined edges are rare occurrences. (Well, duh!) I need to look for an edge and/or a gradient. (A blurred edge is a gradient from one color to another.) But, where on the gradient is the edge? You tell me. I'm really sick of this project... I'm so close but so bored.

Oh, in case any of you have seen my posts on the "3D background generator for filmmakers" (http://fxhome.com/forums/viewtopic.php?p=207670&highlight=#207670), I'm still planning on May 1st for the beta release. I worked on that project most of my free time last week and made some significant progress.

Thanks for your support, guys! I would never get any of these projects done without it.
Posted: Mon, 13th Mar 2006, 7:43pm

Post 17 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Oh, I realized that the last screenshot was the old software. Why not a new one!

http://www.streamload.com/Roozer/NewScreen.jpg

The magenta circle is a point on the light saber (obviously).
The magenta line was an old edge-detection test.
The saber was located in about 1/3 of a second.
There are no user-entered parameters except for the initial "click on the saber so the software knows what to look for".

I'm not too confident that this particular footage would work too well in my software. I would go for a bight-yellow saber that's about twice as big in diameter.

If any of you want to play with the software at its current state, PM me and I'll post it on the forums.
Posted: Tue, 14th Mar 2006, 7:44pm

Post 18 of 40

Redhawksrymmer

Force: 18442 | Joined: 19th Aug 2002 | Posts: 2620

VisionLab User VideoWrap User PhotoKey 3 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXhome Movie Maker FXpreset Maker Windows User MacOS User

SuperUser

Ahh...think about not having to keyframe a lightsaber battle at all. Just click a button and lightsabers magically appear within a few seconds (complete with masking).

Now that would be Mission Impossible!
Posted: Tue, 14th Mar 2006, 10:48pm

Post 19 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

It won't work THAT well smile It would be cool, though!
Posted: Thu, 16th Mar 2006, 2:45am

Post 20 of 40

Fill

Force: 1257 | Joined: 1st Jul 2005 | Posts: 1652

CompositeLab Lite User EffectsLab Lite User Windows User

Gold Member

Your pictures aren't loading on firefox.
I'm really excited to see what progress you have made!
Posted: Thu, 16th Mar 2006, 5:15am

Post 21 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

Yes, it looks awsome so far, please keep it up. What language are you coding this in? So far so good keep it up!!!!
Posted: Wed, 29th Mar 2006, 8:48pm

Post 22 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Hey, guys!

I need alpha testers! (Only runs on PC.)

I've been working on the edge detection for some time now, and I've recently stumbled across something. I just need a bunch of people to test this out and see if they get the same results.

Here's what I need you to do:
1. Download this file: http://www.streamload.com/Roozer/Saber.zip
2. Extract the zip file to a location on your computer.
3. Get some footage with these requirements:

1. Saber must be at least 1" (2.25cm) in diameter
2. Saber must be a bright color which isn't too common in the rest of the footage. (Bright-yellow is awesome.)
3. Saber must not travel too fast. (You can play with this. Basically, if you can easily see the edges of the saber with your eye, it's probably ok. I would like to know how the software performed on both fast and slow footage.)

4. Extract a couple different frames from the footage (I.E. Fast and slow motion) and place these pictures in the directory where you extracted the program.
5. Start the program and click "Load Picture". Select your photo and open it.
6. You should see your picture in the main window now. Click on a point in your saber which displays the most defining characteristic. (I.E. the brightest part, the most yellow, etc.)
7. A circle should show up where you clicked. The target color is generated from an average of the colors inside that circle. So, if the circle overlaps the edge of the saber, this is not a good thing smile
8. Click "Process".
9. You should be able to see a red box drawn around your saber. If the red box contains red patches drawn over the background, decrease the "Edge %" value. If the red box doesn't contain the entire saber, increase the "Edge %" value. (You're trying to get as much of the saber covered as possible without getting any of the background. It's better to not have enough saber than too much.) Adjust "Edge Border" as necessary to include the entire saber.
10. Examine all the lines originating from a point on the saber. In general, do the lines change from yellow to red at the approximate edge of the saber? (That's what I need to know.)

I know there's a lot of bugs in the software. For example, you'll most likely get a different result if you click "process" more than once. You'll also get a bunch of "subscript out of range", "division by zero", etc. errors if you play with it too much smile

In future, you shouldn't have to adjust any of these values after your first frame is setup correctly. The part of the program which detects the location of the saber should also detect the "visibility" of the saber.

Thanks, guys!

Last edited Wed, 29th Mar 2006, 11:02pm; edited 3 times in total.

Posted: Wed, 29th Mar 2006, 10:44pm

Post 23 of 40

Hendo

Force: 13107 | Joined: 16th Sep 2004 | Posts: 848

VisionLab User VideoWrap User PhotoKey 5 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXhome Movie Maker FXpreset Maker Windows User MacOS User

FXhome Team Member

Rating: +1

Roozer, your app depends on the "comdlg32.ocx" ActiveX control which doesn't come by default with Windows, so you'll either need to include that file in a setup package, or replace that control with a call directly to the Win32 shell.

Some users will already have the control if they have installed other VB apps. But for anyone who doesn't have it, you can download it here. Save it to "C:\Windows\System32" (or wherever your system directory is) and then at the Command Prompt, type:

regsvr32 C:\windows\system32\comdlg32.ocx
Posted: Wed, 29th Mar 2006, 10:50pm

Post 24 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Thanks, Hendo! I forgot about comdlg. I'll try to remember to fix that tonight.

[Edit] I just included the ocx in the zip file. Should work now.

Last edited Thu, 30th Mar 2006, 11:49pm; edited 1 times in total.

Posted: Thu, 30th Mar 2006, 9:07pm

Post 25 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

wow great progress. With the image sent with the program it works great, but with some of the other image i have found (just googled sword fighting) alot of the lines are yellow around the face area and the body, the images may not be as contrasty as need tho... but it works awsome! keep it up
Posted: Thu, 30th Mar 2006, 11:43pm

Post 26 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Thanks, Brennanmceachran!

Yea, I didn’t figure it would work, but I have some other ideas.

Here’s what I was testing in the alpha:
When all those lines are drawn on the saber from the single point, the line is drawn red if (pixel %difference from the target color) is >40%. It’s drawn yellow if the pixel is <40%. If you noticed, the program generates a .csv file when you process a picture. This file contains the pixel %differences along the line at the various angles it was drawn. I plotted a bunch of these %differences and found the graphs look almost identical: http://www.streamload.com/Roozer/SaberGph.gif In all the pictures I tested, the saber edges pretty much always fell at 40% difference. (I tried various lighting, etc.)

I didn’t think that would work, but I got hopeful results smile

Well, I have some other ideas which seem a lot more robust. Maybe I should take all the % errors, plot them in 3D, slice the mountain at a certain elevation, and then grab it smile

My biggest problem is figuring out how to ignore extraneous data. It’s easy to visually see the edges by just looking at the data, but it’s hard to figure out how to get the computer to do it. I guess I just need to take a statistics class.

So, until then, thanks!
Posted: Tue, 4th Apr 2006, 10:59pm

Post 27 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

When Roozer had first started his project i emailed him offering suggestions, recently I have started to help Roozer with his project. I created my own different verson that lies basically on the same ideas as his. Even with us bouncing ideas back and forth our programs are very different.

Hopefuly we can continue our sucesses on our little project, and maybe it will work some day. I have no doubt we can do it.

I think the plan is to combine the two programs together taking the best of mine and the best of his.. adding in some more new ideas along the way.

Anyways we are both really excited about this project, i'll put up my version of the program soon.

Here's a screenshot of what i've done so far:
http://www.dataffinity.com/~brennan/bm_productions/images/sabertrackerscreen1.gif

Basically mine has a fairly good chance of finding the edges of the saber, even if the saber isn't a different colour than the background (although it helps). You can move sliders so the calculations will fit your footage. It still has some major problems, but its gettting there. Also you can overlay a image of the saber in the predicted position.

Anyways, I dont want all the credit for my program Roozer basically gave me most of the theories behind all of the edge detections and stuff, I just programmed it in. He has a new idea for a better edge dectection part, and if we can get it to work the way we hope to we'll be well on the way to getting to where we want to be.

Check back later for the link to download my program.
Posted: Wed, 5th Apr 2006, 2:14am

Post 28 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

As promised, here's brennanmceachran's program:
http://www.streamload.com/Roozer/BSaberTracker.zip

Here's some pictures to test with:
http://www.streamload.com/Roozer/ExPics.zip

If you don't already have it, you'll need the .NET framework V2:
http://www.microsoft.com/downloads/details.aspx?FamilyID=0856eacb-4362-4b0d-8edd-aab15c5e04f5&displaylang=en

If you don't have it, .NET framework v2 requires Windows Installer 3:
http://www.microsoft.com/downloads/details.aspx?FamilyID=5fbc5470-b259-4733-a914-a956122e08e8&DisplayLang=en
Posted: Wed, 5th Apr 2006, 5:17pm

Post 29 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Sorry, forgot to tell you how to use it smile

Start the program and click "Load"

Select "4.bmp" from the example pictures. (Most impressive smile )

Left-click at one end of a saber and right-click at the other end.

Click "Find". A line is drawn on the saber.

Move the "Width" tolerance slider to 4 and click "Find Width" A quad is displayed.

Click "Show Saber".

It seems to work pretty well when the values are adjusted correctly. We're hoping to find a way to integrate our methods. (Mine can hopefully find an "adjustment" value and his can actually find the saber.)
Posted: Wed, 5th Apr 2006, 7:51pm

Post 30 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

Does anyone have ideas or suggestions?
Posted: Wed, 5th Apr 2006, 8:15pm

Post 31 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Brennanmceachran just uploaded a new version which has a better saber image:

http://www.streamload.com/Roozer/BSaberTracker.zip

Again, comments would be great!
Posted: Tue, 11th Apr 2006, 6:33am

Post 32 of 40

Redhawksrymmer

Force: 18442 | Joined: 19th Aug 2002 | Posts: 2620

VisionLab User VideoWrap User PhotoKey 3 Pro User MuzzlePlug User PowerPlug User PhotoKey 3 Plug-in User FXhome Movie Maker FXpreset Maker Windows User MacOS User

SuperUser

Now THAT was completely stunning. Using the correct footage this could actually work!
Posted: Wed, 26th Apr 2006, 7:27pm

Post 33 of 40

Remco Gerritsen

Force: 578 | Joined: 4th Mar 2005 | Posts: 517

EffectsLab Lite User FXpreset Maker Windows User

Gold Member

How is the progress ?
Posted: Wed, 26th Apr 2006, 7:49pm

Post 34 of 40

Roozer

Force: 1440 | Joined: 29th Jul 2004 | Posts: 94

VisionLab User Windows User

Gold Member

Well, both brennanmceachran and I have been quite busy recently and haven't had time to work on it. brennanmceachran is running for treasurer and I'm trying to finish up a movie and finals are almost here. Sorry!

Do keep an eye out, though, because, I'm going to try and apply some newly feasible technology to this application... If everything goes as hoped, a mere $100 may solve A LOT of our problems.

Brennanmceachran, I'll let you know later once I have time to try it out and see if it is worth pursuing.
Posted: Sat, 27th May 2006, 3:34am

Post 35 of 40

Penguindude14

Force: 800 | Joined: 30th Dec 2004 | Posts: 5

CompositeLab Pro User EffectsLab Lite User

Gold Member

tard How's It comin'? I'm really looking foward this project. I also have a question. I have to green screen guys in moving footage. Would it be able to track a certain pixel, would it move with it?
Posted: Sat, 27th May 2006, 4:06am

Post 36 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

Squirl Pirate1, Roozer and I have been quite busy these last couple of weeks, but don't fret we are slowly making progress. We haven't worked on it too much lately but I'm just getting free now so i'll start coding again soon. What we are working on now is an apropriate edge detection for the saber, once we get past that, I think we have no problems... knock on wood...

To answer your other question, currently the tracker looks for a certain colour, so if your footage has a green screen and your lightsaber is red, blue, somthing like that it would be able to find it easy. Once we get past the edge detection the program would hopefully cycle through all frames and find the quad for the saber, so in short Yes, and Yes.

I might continue this weekend you've inspired me alittle smile

P.S. A little screen of my current edge detection.
Posted: Sat, 27th May 2006, 4:28am

Post 37 of 40

Penguindude14

Force: 800 | Joined: 30th Dec 2004 | Posts: 5

CompositeLab Pro User EffectsLab Lite User

Gold Member

tard That's Great. I really look foward to it. GOOD LUCK!!! hugegrin
Posted: Sat, 27th May 2006, 4:41pm

Post 38 of 40

NickD

Force: 2163 | Joined: 10th Sep 2003 | Posts: 1224

EffectsLab Lite User FXpreset Maker Windows User MacOS User

Gold Member

Keep it up guys. This might work smile Still, sometimes you can't beat the old frame-by-frame method, but I'm excited to see what else you can come up with.

NickD
Posted: Sat, 27th May 2006, 5:32pm

Post 39 of 40

brennanmceachran

Force: 330 | Joined: 14th Oct 2005 | Posts: 176

CompositeLab Lite User Windows User

Gold Member

Thanks for the support. I don't think there will ever be anyway better than frame-by-frame but its a fun little project.
Posted: Fri, 2nd Jun 2006, 7:26pm

Post 40 of 40

Penguindude14

Force: 800 | Joined: 30th Dec 2004 | Posts: 5

CompositeLab Pro User EffectsLab Lite User

Gold Member

tard How's it Commin' confused: