The cat door closes for 15 minutes and sends pictures of the “killer” and its prey to the owner.
Amazon engineer Ben Hamm was tired of the fact that his cat named Metrik constantly brought home dead birds and rats. Therefore, he taught the algorithm not to let the pet with the prey in the mouth. First it drew attention to the publication The Verge.
At the presentation of Ignite Seattle, Hamm said that approximately every tenth night at three o’clock in the morning Metrics brought dead small animals, and sometimes half dead — in this case the engineer had to kill them himself. Once a pet brought a “gift” when its owner had sex.
Hamm noted that, because of his anxiety, he could not leave the cat on the street all night, but he also did not want to lock Metric at home, since he “has a unique talent to go to the toilet on the street.”
To solve his problem, Hamm decided to create an algorithm that locks the cat door when Metrics returns with the loot. To do this, he bought a camera with support for deep learning Amazon DeepLens and installed it above the door. It took the engineer several months to take 23,000 photos and manually divide them into four categories to train the neural network.
The engineer loaded all the images into the service for machine learning and created an algorithm that closes the cat door. The AI determines whether the cat is leaving or is returning, and if it comes with prey, the system locks the door for 15 minutes, sends photographs to Hamm and sends a donation to the environmental organization Audubon Society, which is engaged in the protection of birds.
For five weeks of use, the system five times did not let the cat in with the prey in the mouth, and once by mistake locked the “innocent” Metric on the street. Subsequently, Hamm hopes to collect more data and train the algorithm on other cats so that people with a similar problem can use his system.