One of the longer term projects I’ve been working on at the makerspace is a squirrel proof birdfeeder that uses machine learning to identify squirrels vs birds, then make a noise when those darn squirrels try to get a snack in the birdfeeder!
The project goes back a few years when I stumbled across a Google project to make a design using a Coral development board to accomplish the same thing. Since we had the equipment to make most of the components at the makerspace, except the development board, I started making one and looking into improving their idea.
First minor change was an error in the CAD files, which I’ve corrected on my GitHub. It also seemed like the board they chose was overkill, I had wondered about using a Raspberry Pi, but then Brian pointed out it might need a board with a better GPU. Ian mentioned the ESP32 board/chip could run TensorFlow Lite & might do the job. Cost of the ESP32 boards was appealing, so I started going down that path even though I didn’t know much about the subject at the time!
I made was the birdfeeder housing and seed holder well over a year ago, which turned out quite nice. But the ESP-EYE housing got complicated between mounting locations, how to power the device, how to create the noise and lots of other distractions, so it got put aside a while ago until I could get back to it. Earlier this month, with some help from a student working at the makerspace, I started working on it again creating a prototype housing for the ESP-EYE v2.1 board in Fusion360 and restarting the work on the software.
It’s now at the point where it’s taking pictures and we’ve begun training the software to identify squirrels (cat pictures work pretty well). It still needs more work, and I’ve gotten other projects that need finishing first. But I’ll put a page on our website when more details are finished. Or stop by our makerspace & we can show you the progress in person!