| |
| |
| |
|
Page: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Comments:
<0> yo
<1> my mum says hi
<0> to me?
<0> <grin>
<2> TByte - YOUR mom says i'm better in bed than you
<3> Hello everyone....
<4> hi
<3> box_ Were you the one working on that commercial zapper?
<4> yep
<3> How's that going?
<4> real good, taking a breather from it a moment after i just got past the most recent hurdle
<5> hoooohaaaaa
<4> i was migrating from VFW to directshow and it was going very poorly
<5> new monitor is here :)
<3> Something I was wondering - what were you using for filtering? I got videolab recently, but I didn't see a filter that did what you were doing in it.
<4> it took over 3 monthes before i could access a tv tuner card the same basic way i was accessing my other video capture card
<4> by filter what do you mean?
<4> i run a few diff processes over the image
<3> You had something that filtered out the static by working across like 3 frames to find the garbage.
<4> oh
<4> thats all custom
<4> nothing prefab like it
<4> basically its an averaged frame
<3> You may be able to make some $$$ if you interfaced it to videolab and sold it to them.
<4> huh what does their product do for users?
<4> video post processing?
<3> Videolab is a set of controls for Delphi that allow you to do a huge pile of stuff.
<4> looks neat
<3> Lot of post-processing, but also other stuff
<3> Also piles of audio and other things.
<3> In other labs that is
<4> what sort of license is it?
<4> damn i should have been using video lab it looks like
<4> i just reinvented a lot of this
<3> Free for non-commercial - like $1500 if you use it in a commercial product.
<4> wow
<4> pretty pricey to profit from
<3> They got a huge pile of stuff though. If I were using it in a product it'd probably be worth the dough.
<4> sweet, brb
<3> What I really want is something that works better than histogram equilization with my camera for seeing in bad light.
<3> That was one cool thing I did with it though - I took my plain digital camera and got it to make viewable images just by the light of my computer monitor.
<3> However I couldn't get it to work well with the actual stuff I wanted to use it for - collision detection - because automobile headlights and taillights caused m***ive flares after EQing the image.
<4> sorry phone i'll brb
<3> Bad because something could be in the flare and I'd never see it. However I did get some stuff working such that I could detect moving objects in daylight.
<4> ah
<4> back
<4> yeah actually i bet i got something for that
<4> it'd be a hack of this median filter
<4> i dont actually use a standard histogram with my analysis right now
<4> i use a hue-istogram
<4> and a variance-istogram :)
<4> but i think what you would want is some sort of brightness spread where you could map how much if at all to amp the light
<4> so you could set the light spots to not amp at all the the dim spots to amp considerably
<4> yeah videolab looks cool, lil rich for my blood though
<4> maybe i'll rewrite it to use it when i get some VC
<5> this is just the most stupidly large monitor ever
<5> i am in love
<3> Not sure if that would work now that I think about it - the problem is that the lights create a field around them of dim light - reflection off of dust and such. Thus I amplify that rather than what may be behind it. The physical optics of this is fairly complex before it ever gets to the camera....
<4> hmm
<4> yeah dust could be a problem
<3> So to do this right, I need to figure out what the flare around a light actually is, and remove it entirely leaving any slight variance still there.
<4> what you could do is some sort of primitive pattern recognition
<4> and recognize the basic shape/form of a flare
<3> The other problem with histogram, and would probably be a problem with median etc. is that there's a certain level of static from the camera itself that creates ghost images. I THINK That your filter that goes across frames to eliminate noise would fix that.
<4> and remove its brightness value at any given point from the source image at that point
<4> curious though why not use sonar?
<6> box, is your project available, sounds very interesting, i'd be interested in trying it out with my tv capture card. been using dscaler but the video quality is quite poor
<3> For what I have in mind, moving WAY too fast.
<4> i think its better suited for collision detection than images
<4> bigstar no not yet but tell ya what if you msg me an email i'll put you on my beta list (just a list of emails by my computer atm)
<3> box_ Using sonar at 500mph ain't gonna work.
<4> it'll be ready to demo here pretty soon
<4> wtf are you doing at 500mph?!?
<3> Or transsonic speeds if I I build something that fast.
<4> i dont think you should be polling your environment in realtime at that speed
<3> While I'm experimenting on the ground, this is anticollision for drone aircraft and maybe some of my rockets.
<4> just seems impossible to get decent valid data
<3> Being in the air actually simplifies the problem a LOT.
<6> ok cool
<4> maybe some sort of distributed laser ruler
<4> randomly sampling its area
<3> box_ It can be done visually because what I'm looking for is at a distance.
<4> well damn
<3> I'm debating on whether or not to tell you how I'm doing it with video ... I may be willing to trade you how I'm doing it for your filters or some such, but I really don't want to tell everyone what I'm doing so I don't wind up with someone pirating my ideas.
<4> i have a really great image recog system in mind i've 'come up wiht' (prolly reinventing someone esle wheel but i dunno who)
<4> yeah i understand
<3> However the whole anticollision method does NOT depend on object recognition.
<3> In its most primative form, it doen't even require edge detection, though that can help.
<3> Basically what I do is search for motion against a more static background.
<4> yeah i got a fair bit of that going on with my commercial filter
<3> And if I do it from more than one camera I can tell distance as well as direction.
<4> mroe or less, sure
<4> better have those cameras really nicely synched
<4> heh
<4> well i dunno man
<4> you seem cool and all
<3> Ok, at the basis of this (which you can probably infer from what I'm telling you) is that rather than looking at a frame, I'm looking at the difference between two consecutive frames.
<4> but i'm not interested in helping make any sorta rockets
<4> for the good guys or the bad guys
<3> box_ Mine do NOT carry anything military.
<3> Never have, never will./
<3> They carry electronics.
<4> what are your rockets for then?
<3> I do it as a hobby.
<3> I like to fly stuff. I do rockets rather than RC aircraft, though the two fields overlap.
<3> ....
<3> Lemme show you the biggest project I helped on. I did some of the video work.
<4> cool
<4> yeah thats cool i like model rockets too
<3> Go to www.wsrocketry.com and click the Titan IV link
<3> The actual failure of this flight was an engine did not come up to pressure.
<3> Caused a huge amount of torque that the rest of the system could not stablize.
<3> There were other issues, but a stop-frame analysis shows that the engine power problem was the main culprit.
<3> The Delta II worked MUCH better, but I wasn't around for that one.
<4> those are some pretty cool rockets there
<3> Really REALLY ****ed that the Titan tore itself to pieces.
<4> yeah i'm so worried about some of the stuff i've got in mind
<4> this commercial filter could just as easily be recognizing people from a video stream
<4> that wont be happening if i can help it
<3> Yeah, but the cops already have that.
<3> I wouldn't worry about it too much because other people have developed it.
<4> i have a path nav algo i cooked up that sometimes feels like the next t2000
<4> and this other robotic sight project that a camera could use to concievably read and comprehend all kinds of written imagery
<3> Like with my anti-collision - the military already has drones that have weapons etc. The problem is that their methods are all cl***ified so I have to do my own work. Thus I'm not giving them anything they don't already have.
<4> like logos, signs, any sort of visual cues
<4> yeah
<4> thats true
<3> If you get that going, I'd love to have it for my nav system. As I said, doing it in the air is easier, but I'd LOVE to make a car that could pretty much drive itself.
<3> And that requires what you're talking about.
<3> Have to read signs....
<4> yeah
<4> my algo recognizes high contrast patterns (letters, logos, etc) in any orientation very quickly
<3> Can you tell me your base method for doing that? The only way I know to do it is with wavelets.
<4> and can abstract somewhat so it can recognize a variety of, say, capital A's from one basic capital A source pattern
<7> naked?
<4> yeah its wavelet based
<4> basically you emit a wave from the averaged center of a region that has defined edges
<4> and that wave produces feedback when it encounters contrast from its current tone to the next tone outwards from center along that given angle
<4> the output is accumulated and has a spectrum, one side is contrast variance close to the emiitter, the other side, farther away
<4> each degree incriment of the emitter wave would accumulated its own graph of contrast along that given line
<4> and then all those graphs are more or less stacked up
<3> Hrm. I'm kind of following that, but probably would need to step through it to really visualize it well.
<4> a given input source image will produce the same graph regardless of orientation
<4> upside down, 37 degrees to the right, doesnt matter
<3> Yeah, 'cause it's a frequency spectrum.
<3> Cleve.
<3> Clever.
<4> and if you normalize the size of the overall output image to the same sizes as your comparison images, size doesnt matter either
<3> Yeah, for recognition high resolution usually makes things harder.
<4> yeah
<4> havent actually used that method yet
Return to
#delphi
or
Go to some related
logs:
#cisco
#eggtcl
#windowsxp
skr0oface
#bsd
#freebsd
#computers
#windows
stockhouseusa
#worldcup
|
|