This reminds me of Matthias Wandel, who also loves using the Raspberry Pi camera to monitor everything, including rodents [0], shooting squirrels [1], and keeping a heater aimed at him [2].<p>[0] <a href="https://www.youtube.com/watch?v=dytn2LypY_c" rel="nofollow">https://www.youtube.com/watch?v=dytn2LypY_c</a><p>[1] <a href="https://www.youtube.com/watch?v=y9OcHxiUNMA" rel="nofollow">https://www.youtube.com/watch?v=y9OcHxiUNMA</a><p>[2] <a href="https://www.youtube.com/watch?v=6kK3xF1REHo" rel="nofollow">https://www.youtube.com/watch?v=6kK3xF1REHo</a>
I'm working on a system for the deer that jump my fence and eat everything. Current plan is a servo to unlock the dog door, an audible ding to alert the dog, and a treat dispenser that releases a treat once the deer is gone. The treat would be in the house, and retrieving it would relock the door. I will just have to make sure to turn it off during rut!
I've always been facinated by the usage of the Pi. I think of the methods used when you have a pi, versus not having one. It looks like a fun project if you have all the parts, but I don't so I thought of this alternative!<p>A method I thought of was just a camera that will utilize yolo3 <a href="https://github.com/experiencor/keras-yolo3" rel="nofollow">https://github.com/experiencor/keras-yolo3</a> or just an always on security camera that you can just skip through the video to see all the activity with less setup, and seeing what animals cause issues. The jetson for faster 'edgey' visual ML models might be an option for those who want a stronger NPU, but using online GPU/NPU tokens and GPUs on computers if you don't need live feedback is very effective.
Why use motion detection sensor if you already have a camera?<p>sudo apt-get install motion<p>Raspberry PI is powerful enough to do object detection offline, in real time.
So you don't really need to detect any motion, just detect the animals in the video feed 24x7
I have the same problem, was hoping the end would tell me what to do once I discover that it is actually moles.<p>I’m finishing setting up a rig with Frigate for object detection but one thing it lacks is many common animals (although if a zebra ever shows up in my yard…). Maybe as a weekend project I’ll figure out a way to integrate MegaDetector.
Attaching a motion sensor and a camera to a rasberry pi is a little hey ho. But the Microsoft megadetector machine learning software to distinguish between motion by an animal rather than say a leaf is quite interesting. No doubt soon to be wrapped in an app for the rest of us...
> one of two animals in the world known to smell underwater<p>I hate being that person but... in fact all fishes and most aquatic invertebrates are able to smell underwater. Sharks can smell and track ridiculously small amounts of blood in water.<p>That said the project is simply brilliant