I built an expert system (what we used to think of as AI back then) that could read the circuit diagram of a complex electronic circuit, figure out what it was meant to do, and set up the test gear to test it and diagnose manufacturing errors.
In 1985, using Vax/VMS and OPS5.

More recently, I was on a project (can't claim to have done it all myself this time) that used a neural network to detect patients in a care home that fell over/fell out of bed and alert the nurses' station.

How would you train a model which detect a patient have fallen down?

Do you train with a lot of images of people falling down (if you don't mind me asking) ? I think having some kind of sensor beside the bed will do well than neural net, although you have to add a lot of sensor (which I think will cost a lot).

Since I think image distribution (old people falling down) of people falling which you want to use is hard to gather , data argumentation is the only way to goes. But how realistic is those data argumentation until it become bias? Oops look like I run off my mouth (finger in this case) here.

@mr-user

Yes, we employed a bunch of actors and videoed them, as well as searching for as many useful images as we could find. What we were looking for was a change in detecting a person in the scene shortly followed by a prone person in the scene.

@RobAtPaxMondeo

Oh you solve it with people detector. I think it will work something like this

1) Detect people

2) Detect when people change position in the image

3) Check if people in the new image is flat on the ground.

Asking an actors to fall down in different position look like a fun task :) Don't forget to ask actors to fall down in funny position since people don't fall down flat on the ground.

I am sure you have already thought about it but don't forget the add the distribution of people grabbing part of the bed on the ground or when the sheet fell down upon the people.

I am sure people detector will not work when the bed sheet cover their whole body (since I think the outline have changes). Maybe you could just think of it as irregularity and ignore it.

@mr-user Yeah, we pretty much hoped if somebody fell out of bed, they would leave the sheets behind. :-)

@RobAtPaxMondeo

Here is a reality of AI development : being hopeful that thing will goes well :)

expert system????! how old are you?!? that is not important
(i-must treasure-you)

@mr-user @RobAtPaxMondeo I guess it's a bit like pose estimation...

@WeAreMany

Yeah it is kind of post estimation with other part. He also needed to determine the change in position of a person.

A person on the ground and on the bed will have the same pose (flat) but he only wanted to detect when a person is on the ground.

It may even be a simple neural network which determine whether a person is on a bed.

He already give us a lot of information , it may be rude to ask him for more.

@AleCx04 Only 64... :-)