wouldn’t it be great if we could get a stitched picture of the complete garden? So we would be able to use better pictures and don’t have big jumps from one picture to another.
Even more the weed detection farmware could run on the stitched garden picture and thus would be more reliable. I also could image to exclude certain areas from being automatically scanned (e.g. outlines of the garden bed…).
Any ideas on that? Is that a big effort to implement?
I recently finished my XL build and was trying to set up weeding, and while it seems to detect weed fine for single camera shots I was quite surprised to see that the documentation for weeding the entire garden is so… unhelpful, even more so since there is no loop construct.
So I’d like to tune in and I have an additional suggestion/idea:
Have the weed detector do the whole garden with much overlap of the single pictures. Doing this would allow the software to detect taller plants and exclude them from weeding (e.g. apply basic stereo vision principes; it should probably suffice to compare the shape and displacement of the detected weeds and only consider “overlapping” weeds).