Just started with this- in a while I could put up a Python example. Right now I've got an Ubuntu box running this modified version of the "Darknet" recognizer GitHub - saiprabhakar/darknet-modified at v0 in batch mode that spits out a file with (detection, probability, x1,y1,x2,y2) data for each image, which I can then use to crop out items and assemble a montage. My low-powered system takes 25 seconds per frame, so not exactly "real time". It will sometimes detect a "person" in the street who is just driving by in a car, if it gets a good enough view in through the window. Maybe good for the backyard if you've got motion from tree branches, or animals that too often trigger simple motion detect.
For a glimpse of the future, check this out- running at 30 fps on 2k x 1k images and detecting and color-coding people, vehicles, signs, trees, and pavement areas, pretty accurately:
For a glimpse of the future, check this out- running at 30 fps on 2k x 1k images and detecting and color-coding people, vehicles, signs, trees, and pavement areas, pretty accurately:
Last edited: