4 ms·
How hard would it be to add AI that could distinguish between human and dog/cat motion?
by jonahx 11y ago
How hard would it be to add AI that could distinguish between human and dog/cat motion?
- StavrosK 11y agoImpossible right now, as the sensor just gives you motion/no motion readouts. I think it has a way to configure for mass, though, so it only considers large masses moving. If anyone has any information on that, please let me know, because my blind cat keeps making my lamp turn on :P
- jonahx 11y agoYeah, I think you'd have to have an actual camera image, and then you'd probably want a night vision lens of some sort... I'd imagine the logic for doing the detection might be available open-source somewhere...
- StavrosK 11y agoYeah, sounds like it. A much easier solution would be to adjust the sensitivity and put it somewhere your cat can't reach (like on a high shelf).
- squeaky-clean 11y agoI really don't know much, but was looking up exactly this because of your post. It seems like there are two ways. With a single device, you're limited to just smart placement (orienting it away from the floor or surfaces the animal climbs on) and adjusting the sensitivity of the detector, which will reduce the range and still trigger if the animal gets very close. The Adafruit PIRs come with a pot for this [0][1]. The other method, which most of the ones that seemed high quality used is a dual-sensor system, tuned for different ranges and sensitivity. Most of the commercial ones on Amazon seemed to use a combination of PIR and something else, like sonar or microwave detection. My idea for a quick "hacky" way to do it, would be double up on the sensors and place one 3-4 feet above the other and maybe use some tape on the lenses so the detection areas don't overlap. This way it would trigger for a person walking by, but not a cat walking on the ground, or jumping past a higher sensor. Awesome project by the way. I'm also working on my own home automation project as well, I'm mostly avoiding low-level hardware stuff for now, but definitely want to play with it in the future. Bookmarking this and your other posts for inspiration. [0] https://learn.adafruit.com/assets/13829 https://learn.adafruit.com/assets/13829 [1] https://learn.adafruit.com/pir-passive-infrared-proximity-motion-sensor?view=all https://learn.adafruit.com/pir-passive-infrared-proximity-mo...
- StavrosK 11y agoThank you for that information! I'm definitely going to follow the first method initially, as it's super simple to do. Good luck on your project, and stop avoiding low-level hardware, it's lots of fun and easy to learn. Here are my notes, which I'm working on, but may help you: https://www.stavros.io/notes/hardware/ https://www.stavros.io/notes/hardware/
- TheOtherHobbes 11y agoPIRs measure temperature changes, not motion. I set one up in my kitchen and it got very confused whenever I turned on the kettle or the toaster. :) Sharp make a range of IR distance trackers that are more useful for checking whether or not there's an object/person in front of the sensor. https://acroname.com/articles/sharp-infrared-ranger-comparison https://acroname.com/articles/sharp-infrared-ranger-comparis... Adafruit sell one of the short range models, but the GP2Y0A710 is better for domestic applications. If you're clever you could probably track the level of multiple WiFi receivers in an area and watch as people and pets change the distribution pattern. I think there was a commercial system that did that. It sounds quite hacky and potentially unreliable to me, but I haven't tried it in practice. I would have expect some kind of domestic super-RFID system to have appeared by now, but it doesn't seem to. There are issues with RFID and longer ranges but they should be solvable, and the potential market could be huge.
- bliti 11y agoNo need for AI. You can just use an rfid sensor on the animal's collar.