Context: I'm an experienced programmer with a graduate education in AI and previous CUDA programming experience. I'm versed in Machine Learning but am out of the loop -- I've not used any of the modern software packages of the last 10 years.
Question: Is it possible using modern AI software to easily create a face-tracking application that can use a webcam to track the amount of time spent at one's desk.
My environment is Fedora Linux. I also have an NVidia GTX 1660 for acceleration.
To make this question and answer precise, I've narrowed it to the following sub-questions:
As of June 2021,
Is there existing software that one can simply "set up" with a small amount of programming work (or none at all) that would facilitate training a video classifier from webcam recordings?
How does one provide training examples to this software, or how is data labeled? Does it provide some sort of GUI or accessory tool to label still frames or video sequences?
Does said software provide "hooks" or an event API so that one may invoke code on the event of e.g. a classifier edge?
Finally (and consider this optional), would it be realistic for a seasoned programmer to accomplish such a project using said software in about 30 hours? I understand that this is subjective -- just assume a graduate student in the AI field and ballpark terms. Or, answer in terms of the software's intended audience.
First posting in this community, so just offer guidance if you'd like to see this question refined.