cmucam lighting

[muy dificil]

We are finding trouble getting the right lighting to calibrate all three color values for the camera, we only can get one, maybe two. Does anyone have any suggestion of lighting that is easy to obtain? Thanx.

[Dave Scheck]

We set up a few halogen work lights around the room to get consistent lighting (our field's in a warehouse). I think we ended up having to recalibrate at competition, but this should get your by for local development.

[Kingofl337]

Last year we painted plexi glass and used a halogen light to illuminate it from the rear. But, sometimes even that didn't work in general the Java CMU utility was horrible.

[Eldarion]

After playing with the CMUCam quite a bit, I noticed that the lighting only mattered when using the Java calibration utility. The only thing that the camera needs is a constant source of light, IE it could even be standard arc lighting and it would still work, as long as the lighting stayed that way. However, the only way to make this work was to bypass the Java utility's calibration and just use the raw RGB values from a captured frame.

Originally Posted by Kingofl337

But, sometimes even that didn't work in general the Java CMU utility was horrible.

I agree.

As an aside, the blue channel of the camera (not just the CMU board but the actual OmniVision module) is horrible. The amount of noise on that particular channel is just incredible. The blue tracker will not work with either of the two cameras I've tried due to the noise; has anyone else been able to get it to track blue?

[Al Skierkiewicz]

Different types of light produce different amounts of color. Your eye/brain doesn't see the changes unless you have been trained to see them. A camera pickup device is based roughly on the human response to color and as such only about 11 % of the overall video is a blue signal. As such, most designers will try to normalize the gains on all stages and the blue signal will need about 6 times the gain of the green channel and three time the gain of the red channel just to get the signals the same level. Gain comes at a cost and that is an increase in noise, both thermal noise introduced at the pickup and electrical noise from the added gain stage. Bandwidth limiting can correct some of the noise in later stages and the encoder (where the colors are added into a composite color signal) can eliminate some more but blue will produce more noise than the other two colors. Please remember that fluorescent lighting produces two colors, one when the lamp is "on" and a different one when on the lamp is "off" occuring at 120 Hz. Your eye intergrates this strobe action above about 20 frames per second so you only see one color, the camera sees both. The best bet is to force a white balance with the light provided. Additionally, objects will reflect differing amounts of color depending on surface texture, dirt and the angle of incident light.

[John Gutmann]

Instead of just lights on stands try over head flourescent shop lights.