webcam expose set_window

[brianmichalk1]

In this code,
https://github.com/espressif/esp32-camera/blob/7aa37d4f22503fdac9ccd449e4678c4894c40055/sensors/ov2640.c#L137
Is the function to set a window, which I think will set an ROI that is returned instead of the full resolution image.

First, can I submit a feature request for this to be exposed as on of the Wc commands?
Otherwise, this seems fairly easy to do myself. What files and structs should I change?

[Jason2866]

@Staars Honestly idk if you showed me the possibility of using parts of a Webcam image only or if you have provided this as PR already.
@brianmichalk1 Before you start coding wait for the answer from Staars

[brianmichalk1]

I have several applications, and one of them is to monitor water level in a cistern. It's the most simple, and gets me into a lot of the basics I want to apply to my wireless solar solution to see if its feasible.
I can do this through my office window where I have power and such. I'm more interested in exploring the capabilities of what I can push out to the edge, very much like your Tensor Flow conversion of digital characters.
The cistern has a big red marker on a string attached to a float. My thought was to extract a narrow ROI, say 1000h x 100w pixels. Average each row's red component, giving me a 1D array, apply a smoothing window, then report the peak location. I also have this data from a pressure sensor, so I will have a nice reference to compare against.
I'm looking through your code. I haven't yet found where you use the ROI.

[Staars]

The rough idea to work with ROIs in Berry is primarily based on static positions of ROIs in the image with a fixed cam setup.
To create the affine matrices you just take a demo picture and drag-n-drop it into the ROI editor:
https://tasmota.github.io/docs/ROI-editor/

There you create basically Berry maps, that must be converted into a byte buffer as shown here:

def roi_dsc(m)
var d = bytes(-24)
d.setfloat(0,m["scaleX"])
d.setfloat(4,m["shearX"])
d.setfloat(8,m["shearY"])
d.setfloat(12,m["scaleY"])
d.seti(16,m["transX"],2)
d.seti(18,m["transY"],2)
d.seti(20,m["width"],2)
d.seti(22,m["height"],2)
return d
end

This buffer can be use as an arg for img.get_buffer(roi_dsc).

If you need a „moving“ ROI, you have to do it in Berry, where transX and transY in the map from above are basically the pixel positions of the upper left pixel of the ROI.

[brianmichalk1]

Yes, I am reading the code in be_img_get_buffer now to see how it works. Thanks for the help.

[brianmichalk1]

I know next to nothing about Berry. I'm working on a simple example to get images, and the img() call is failing. I decided to load your example, and it's also failing at the same location. No, I don't have the Tensor Flow artifacts there.

https://tasmota.github.io/docs/TFL/#berry-implementation
line 46: self.cam_img = img()
BRY: failed to load '/t2.be' (syntax_error - /t2.be:46: 'img' undeclared (first use in this function))

[sfromis]

To have the img class available, I think that you need a build with USE_BERRY_IMAGE defined, in the same way as already needing USE_BERRY_TF_LITE