Nvidia Jetson for shooting club

[Fristad]

Hello

 

I'm currently in the process of building a coaxial camera system for my pistol shooting club. The purpose of this system is to show our members where on their targets they've hit without changing shooting positions and walking 25 metres up to the target to check.

 

Everything is sorted regarding what cameras, cable, converters and monitors we're gonna buy, but i wonder if in the future we could feed all cameras through an AI-board such as a Nvidia Jetson that will mark each hit with a red ring around it so that older people or those who have a hard time focusing on electronic monitors have an easier time spotting their hits. I've seen demonstrations we're they've been used for counting people on CCTV by the silhouette of people's bodies, so I can't imagine this can be too hard to implement?

 

And if someone knows of people who've done this sort of thing before feel free to share, I don't doubt someone already has programs for this sort of thing up on Github or similar repositories, I just haven't been able to find any projects. 

[Eigenvektor]

35 minutes ago, Fristad said:

I've seen demonstrations we're they've been used for counting people on CCTV by the silhouette of people's bodies, so I can't imagine this can be too hard to implement?

I don't doubt the board is powerful enough to run a neural network that has been trained to do such a thing. But unless you know what type of hardware was used to train the AI and for how long, and what type of information was needed, you can't really say how difficult it was.

 

This might be a starting point:

https://www.reddit.com/r/computervision/comments/cyarfn/bullet_hole_detection/

[igormp]

2 hours ago, Eigenvektor said:

I don't doubt the board is powerful enough to run a neural network that has been trained to do such a thing. But unless you know what type of hardware was used to train the AI and for how long, and what type of information was needed, you can't really say how difficult it was.

You don't even need a neural network for that, plain OpenCV should do, and even a raspberry would be enough for it.

 

3 hours ago, Fristad said:

I've seen demonstrations we're they've been used for counting people on CCTV by the silhouette of people's bodies, so I can't imagine this can be too hard to implement?

Not that hard if you already have some python knowledge, googling for stuff like bounding boxes and image subtraction with opencv.

[Eigenvektor]

10 minutes ago, igormp said:

You don't even need a neural network for that, plain OpenCV should do, and even a raspberry would be enough for it.

Yeah, I was mostly trying to point out that using AI may not be as easy as it sounds, even though NNs have been used for more complex stuff. Running a NN is not the problem, creating one is.

 

The link I posted above goes in the same direction you do. If you know what the target looked like before you shot at it and compare that to it's current state you have a pretty good idea where the holes are.