I was thinking about something just now. We see all of this AI b*****t marketing from allot of manufacturers these days. They are basically calling on board video analytics AI.
In my mind video analytics is not AI, you make a rule and setup some boxes and the camera tries to detect an object in a box, or moving between boxes. It can try to classify the object but not intelligently, it just guesses based on proportion of VMD blocks in a given area.
For something to be intelligent it has to be able to learn either by itself or with some teaching. If you can go through 1000 event's on an analytics system and tell the system where it classified a human that wasn't a human and then moving forward it makes the system more likely to accurately classify a human you could argue there is some amount of intelligence, or you could call it calibration? Detecting a blob of moving VMD blocks moving from one side of a screen to another is not AI in my view!
I know we have the systems that try to do slightly more sophisticated classifications with face recognition and stuff like like sex, age, hair style. So far it seems none of them actually work but assume for a minute they do actually work! Is that AI? What is the camera learning that makes it better next time? To the bast of my knowledge its just detecting stuff and trying to classify things on size, color and maybe movement patterns, those functions are hard coded into the camera's firmware, its not like the camera is learning to do this stuff as it goes.
The idea of AI on a camera is to me a ridiculous statement. If somebody comes out with a camera with 20 TB of storage on board and 10 GHz of CPU cores to really store and methodically analyse every frame of video and learn from its own detection with minimal human intervention or a cloud platform fed with information from thousands of other cameras then they can market it as AI.
What am I missing here?