Maker Tennis Smith has constructed a Raspberry Pi-powered “cat doorbell,” designed to make use of on-device machine studying (ML) to determine when a cat desires to return again in from the chilly — and to ship its people a textual content message alert to allow them to come to its help.
“We’ve got a cat that likes to exit on our enclosed patio. When he is able to are available in, he’ll stand subsequent to the door and meow (yell). We open the door and let him in. No drawback,” Smith writes by the use of background to the undertaking. “However, he regularly stays out lengthy sufficient for us to neglect. Greater than as soon as he was exterior yelling and we have been oblivious. That is Drawback #1.”
A pair of TensorFlow examples flip right into a human-alerting “cat doorbell” on this intelligent undertaking. (📷: Tennis Smith)
“It’s also difficult by the truth that he likes to yell anyway,” Smith continues. “He’ll lay on the bottom wallowing and yelling only for the enjoyable of it. That causes ‘false alarms.’ We might suppose he desires in, however all he was actually doing is having fun with life. Irritating to us (however I believe it secretly amuses the cat). That is Drawback #2. By some means, we would have liked a tool that will: alert us when the cat needed inside (repair Drawback #1); make sure that was actually his intention (repair Drawback #2). Clearly, we would have liked a Cat Doorbell.”
Smith’s preliminary model of the undertaking labored on sound alone, listening out for the cat’s miaow and triggering an alert accordingly — however was liable to false alarms. The answer: the addition of a digicam, triggering alerts provided that the cat is each seen and heard. The key sauce: a Python-powered program which adopts two TensorFlow instance tasks to determine the cat by sight and sound utilizing on-device machine studying.
“That is primarily a small state machine,” Smith writes. “The Doorbell listens passively for the sound of a cat meowing. When it hears that sound, it allows a digicam. If the on-board mild sensor detects darkness, an LED strip will likely be turned on. For 45 seconds the Doorbell makes use of the digicam in an try to determine a cat. If no cat is recognized, the Doorbell goes again to passively listening.
The ML triggers on each sight and sound, with an LED strip behind a diffuser so the system nonetheless works at evening. (📷: Tennis Smith)
“If a cat is recognized through the 45-second window,” Smith continues, “a textual content message is distributed to me. The system then pauses for 2 minutes to forestall triggering a brand new alarm. If darkish, the LED mild stays on till after the two-minute pause is over. The Doorbell then goes again to listening.”
Smith’s full undertaking write-up is offered, alongside the supply code below an Apache 2.0 license, on his GitHub repository.