Does anyone have: spectral sensitivity info (and other specs) for camera on little jumping sumo?


#1

Product: [Jumping Sumot]
Product version: [N/A]
SDK version: [N/A]
Use of libARController: [N/A]
SDK platform: [N/A]
Reproductible with the official app: [N/A]

hi, so, I posted earlier about cutting my developer’s teeth on the little jumping sumo, since its relatively inexpensive and has an ok battery life … I am trying to drive navigation from some cv routines, and now I wonder about some spectral sensitivity and other data on the camera in the little sumo … I have been using some blue painter’s tape to draw patterns on the floor, but, I find that when I have light reflections on the floor from a sunny day, the flares from the sun will appear in a blue hue to the camera, and, in low light, I get a sort of blue haze (I think that there is a name for this, and I’ll look for it and post it later) …

so, now, I’ve been wondering what the best lighting scenarios and colors I might look to match off of, and so, thought to read a bit of of the technical specifications for the camera, if such a thing could be found …

perhaps these differ in the various sumo models, so, perhaps I could read through the others, as well. thanks.


#2

What about using black and white fiducial markers, such as ARToolkit?
It works well even with crappy webcams so should do a decent job on the Sumo.


#3

hi arnaud: in a future version, I’ll give it a try;I was just wondering, at the time, whether or not there was a particular color tape I could find, if I could get some spectral sensitivity & perhaps temperature information about the camera’s lens. I was thinking that perhaps I could find some particular green tape that would work a bit better.

also, up to this moment, since I was doing everything “by hand,” and in a “naive” way, I didn’t know about either ROS or ARToolkit, but, as I make this project more sophisticated, I’ll look at these tools; definitely thanks for giving me a heads up on these projects.

in the version of this that I wrote, I was using opencv to chop the image into half and then look at the bottom part, since I was only examining things on the floor. Then I filtered the image using the HSV conversion to throw out anything except my H ranges, and then further filtered for S and V. The S & V were actually problems because, as I stated, light flares from reflections of sunlight on the floor, and noise produced in very low-light situations would appear blue. But those could be filtered out, because, in the end, the tape’s S & V values were, generally speaking, different from the camera artifacts.


#4

HSV filtering is generally a straightforward filtering method, but not very resilient to noise…
If you want to keep it simple, you could definitely improve your accuracy using a combination of HSV filters, for instance using a marker made of blue tape next to red one.


#5

oh hi arnaud, this is binary piece is a good idea … generally speaking though, in opencv, I use an iterator over the jpg image I’ve transferred from the device setting to black everything within my hsv filter range, and throwing out the rest … I later make a navigation decision on this “reduced” image …

I’ll have to look into what you’re asking for, w/ two colors, but, it seems that I’d have to, in examining some horizontal scanline (or some horizontal scanning window) set some condition to only consider some length of red pixels followed by blue pixels, or vice versa ?


#6

Yes, you could do such approach succesfully I guess.
Keep us updated!