[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [APD] web cam

Alan and Kim Van Nevel wrote:
> Doing the Gaussian blur is reasonable in trying to avoid any
> mis-registration problems due to motion, but the downsample would not be
> needed, imho.

It's needed to hold down computation power if a stream would be
processed. It's done in Java and on a 1 Ghz PIII.

> The white balance is just a camera calibration, and would be
> needed for any type of algorithm development.  Assuming a fixed camera
> position and a stable camera calibration , you'll get a filtered image
> which will combine information from many pixels into a single pixel.

There are other variations too. Daylight, room lights etc, so a fixed
white balance is not enough if you want to cancel those variations

If I add a gray or white pad somewhere in the image I can take
that area and cancel out those variations. Will be doing that.

> You
> could hand segment the image into 'uniform' regions that should only
> contain one plant type.

I don't think so, but it would be easier ofcourse to just shoot M. umbrosum
do get the green-nitrogen correlation.

> If the area is truly uniform, and depending on the
> kernel size for the Gaussian blur, you should wind up with a small
> distribution of pixel intensities.  if you dont, then the kernel needs to
> be bigger (5x5 or maybe even 7x7)  or the problem may not be feasible.

Exactly - I use an algorithm where I calculate whatever kernel-size I
want. But then again, it's easier to just unfocus the cam *lol* ;)

> If
> the distribution varies spatially within a 'uniform'  plant section, it
> indicates that the color intensities are dependent upon the orientation of
> the leaves, other reflections from the tank, etc, complicating the
> diagnostic process.

I just want a hue of the total green and see how it changes. You know
planted tank nerds say "the greens was unbelievable when I started
dosing this and that". That is what I want to have numbers on.

> So lets say the distribution is nice and tight.  Then
> multiple measurements need to be made over time.

Exactly what I do. It seems you didn't see the chart. I save
all the samples in a MySQL-db.

> Variance in the
> measurements or distribution may be an indicator of plant stress, or 
> simply
> changing conditions, eg watre changes, fertilizer dosing, CO2 changes
> etc.  You're simply looking at a single measurement for subtle changes
> which may be highly dependent upon the water chemistry and parameters.

Exactly. I just want a measure that a human, or a bit better, would see and
have a mathematical value for it instead of a human "I think it's a bit 
today, I might need to dose".

I get a real value instead of using my experience as a plant grower.

> If
> you were talking multiple wavelengths then perhaps you could build more
> robustness into the system, but for a simple color or IR camera, I'd be
> surprised.  In my research I have not found any papers on using single 
> band
> imagery for this type of problem.

The green-measure-method is already used by some high-tech-farmers
here in Sweden. They put in the green value + GPS-value in a db in then
their auto-doser-computer on their tractor dose the right amounts of
nitrogen, so I'm fairly sure it can be used on a planted tank. Especially
if you turn the camera to shoot nitrogen indicator plants like M. umbrosum.

Best regards
Daniel Larsson
Aquatic-Plants mailing list
Aquatic-Plants at actwin_com