Is AI learning of falsely identified object possible?

nbraude
nbraude
in SecuritySpy

I find that the AI analysis of triggered moving objects works (in general) very well. However I have a persistent nocturnal trigger from a black cat that is recognised as a human. No amount of fiddling with the sensitivity and other controls has been able to stop this. Interestingly, other visiting cats don’t seem to have this effect.

Would it be possible to have a facility to identify this as an animal so that the AI makes a correct identification?

Comments

  • Ben
    Ben

    This can happen sometimes - the AI is very accurate (around 97% for human detection), but no algorithm is perfect and sometimes objects will be mis-classified. This cat must look very human in some important way!

    We currently don't have a mechanism for the AI to learn based on user feedback. My best advice would be to allow images from your camera(s) to be shared with us anonymously, using the option in the General Preferences. This way, images from your cameras will be fed back into our training data that we use to train future versions of the AI, improving its accuracy.

Categories