Does anyone have experience with those:
If you guys have any tips of camera systems for live IR Tracking - that go beyond Kinect and PS - let me know. Beyond means more Resolution, More FPS, just cooler in any way :)
Im currently in Research mode - and didnt worked with industrial cameras before. Would be cool to hear your insight
Hi Mio, so far, I’ve used Flea3 USB3.0 with Elliot’s imagePack (4k RGB, 2K Mono, 2k RGB) and I was always very satisfied with the stability, vast configuration options, lens choice, responsivity, performance and customer support. With IR tracking, as you probably know it is often tricky to get good quality of input image, setting the lights right, so I use depth tracking whenever possible.
hey man thanks for the fast answer…this sounds very good…how is your experience with the fps?
i saw that the flea has this option:
Mono 1.3 MP Sony IMX035 CMOS, 1/3", 3.63 µm Rolling shutter 1328 x 1048 120 FPS
i mean thats quite nice reso and super cool 120 FPS.
Concerning the image pack of Elliot - does that mean you use ex9 on 32 bit? - or DX9toDx11 somthing like that.
Anyway cool to hear that someone tried the flea …as it really looked very promising to me.
as you see there, check page 3:
the sensitivity of this sensor in IR spectrum isn’t that great- 15%
you can compensate it with good IR lighting, but it’s not always easy.
I use IDS UI-1240LE-NIR-GL
just compare sensitivity curve:
@DiMiX good point on sensitivity, on the other had, this camera does not have bayer filter and it can give you 12bit of luminance. with most IR setups you going to boost IR lighting as much as possible anyway.
@princemio yep, it will give you 120fps (if the mainloop and cpu load allows for it), when not required, i use 32bit, if you need 64bit, it is easy to recompile elliot’s source codes from github. there is astexture dx11 node.
it’s true, if you have possibilities to achieve good EQUAL IR lighting, the sensitivity is of secondary importance. I’ve never tried rolling shutter, but maybe @120fps is OK? https://www.ptgrey.com/KB/10028
Hey DimiX thanks for your insight. And very good point about the sensitivity. Did not even thought about this. I just thought they had a good or bad sensor - and then you have your result. But that sensor vary in the way the “accept” different wavelength is very interesting - thank you.
Concerning your suggestion. The camera looks very nice. but 25 fps in my humble opinion is quite slow, isnt it? its slower then kinect. Dont you get artefacts or stutter. Also its less frames - and less information for interpolation. Hence 25 fps isnt only a flow problem - it might also be simply not enough information. If you use kalman or even running average for filtering, then 120 information per second deliver smoother results then 25. Please correct me if im wrong. Really curious to hear more of your insight.
Yo id! 120 sounds stunning …how do you feel about the latency…i imagine it also to be quite brutal! i think getting the pics in could be possible - if the rest of the performance is handled on GPU maybe
IMHO resolution and fps in this case might be not so important, but
it’s very depends what you want to do after you get camera data in v4.
I can imaging it would be hard to track blobs with 1328 x 1048 @120 FPS
I use my camera 640x512 @58 fps (with binning mode) and capture floor area 11x7m.
for tracking I have to go down to 256x256 to keep speed.
there are some other ways you can get good picture beside HD.
e.g. if you can get rid of noise (good equal lighting), you can switch on sharpness and edge enhancement.
FPS: probably make less sense to run camera @higher speed then patch itself?
To justify 120 fps you have to capture really fast movements of objects.
with butoh dancer even with 15 fps you will be fine.
me, personally, latency disturbs me much more then fps speed.
I am looking to upgrade to this one: http://eu.ptgrey.com/grasshopper3-41-mp-mono-usb3-vision-cmosis-cmv4000-3e12-2-eu
90fps 2048x2048 usb3 NIR
The thing is that such a high resolution will eat a lot of resources, but may it has format7 and you can lower the resolution and increase the FPS. Lenses are very important as well as some usb3 repeater in case you need cable length more then 5m.
just my 5cents ;)
yes, lens. try to get IR corrected lens:
I use tamron: http://www.tamron.biz/en/data/ipcctv/vm_index.html
they are cheap, but OK.
I have one more expensive KOWA, but there is no diffrence at the end
thank you so much for dropping and helping. I learned a lot reading!
@Dimix you are right latency is also very important.
To me 30 FPS was never really enough for dance.
120 bpm is regular radio dance music (8 counts in 4 seconds) = 0.25 seconds from "Boom"to “Clap”.
If you hang out in a club and step from side to side to that rythm, then you capture only 3.5 pics per move (15 FPS) or 7.5 pics per move (30fps).
I do believe that 60 FPS or higher gives a more intimate and detailed insight into whats going on.
“probably make less sense to run camera @higher speed then patch itself?”
Hmm ya good question. I would think that it could make sense but not sure. Hence you have room for optimization. In OF (vsync off) you can read the kinect 1 camera input (multithreaded) and still have less then 2 ms per render cycle in the main thread (>500FPS). Hence you particle flow can continue to develope for example on CPU or GPU and might be feeded with new control info every i iteration. But not sure here - maybe i talk complete BS now.
What is a IR-corrected lense. Sorry if that question is stupid - but i never heard of that. Do normal lenses limit the amount of IR passing through?
“I use my camera 640x512 @58 fps (with binning mode) and capture floor area 11x7m”
sounds actually quite nice…means you have roughly 2 cm resolution across 11 meter. propably completly enough. But isnt that also very close to PS3?
@zeos thanks for the link - definitly will check this one out too!
“What is a IR-corrected lense. Sorry if that question is stupid - but i never heard of that. Do normal lenses limit the amount of IR passing through?”
IR has refracts differently than visible light, therefore it has a diffenent focal point. This would lead to blurry images when used both for visible and ir light. ir corrected lenses compensate this…
as long as you either just work with visible OR ir light you dont need a corrected lens…
@elektromeier in normal lenses IR and visible light have different focal planes. which means that you need to adjust focus if you want to look at the same thing in both visible and IR light.
If you have an IR-corrected lens then visible and IR light have the same focal plane so no need to adjust focus.
this is only relevant if you plan on using the lens for both visible and IR light “at the same time”
yup. youre right “at the same time” is the keyword. adjusting the focus doesnt help if you have ir + visible light the same time (which is always the case for sunlight) as you just can focus either on the visible or the ir light. means you always have a blurry image in this case.
or it could be an installation the looks at sunlight during daytime and IR light at night time and you can’t adjust focus every day, like a surveillance camera.
EDIT: Not look at sunlight, but sunlit objects
Camera FPS higher then main loop: even some of algorithms can theoretically benefit from higher speed, i don’t think that practically you will notice any difference in visual output. but I might be wrong, perhaps devs can give better answer.
IR-corrected lens: i think it has to do with special glass coating. And the difference should be noticeable in sharper focus. Maybe it refers to Day/Night switching ONLY, I don’t know exactly. I personally did not find a big difference in focus between using IR-corrected and normal lenses. maybe cause I do focus in IR mode only and my LowRes. It can be different if you capture 2048x2048.
2 cm resolution: yes- it’s my capture resolution, but I upscale it before FX chain, which runs on higher resolution
I am thinking that tracking at higher framerate can be beneficial in the same way that sampling mouse position in a drawing program usually work at much higher sample rate than framerate so that you can make nice mouse curves for drawing instead of those jagged polygons when moving the mouse fast in a curve.
Hmm so why not going all in
2k imaging at 150 FPS…and then wait for software solutions to come…
hehe just guessing … but what is the problem of picking just the …sickest
The NIR version of course…wait the price is on request - so the problem might be hidden there.
Any Thoughts on that one?
Ah yes thanks for the hint with ir corrected lense - since i want greyscale pattern calibration - i think that could be the way to go. Thank you guys!!!