There are 74 million house cats in the United States, four million of which, according to Amazon senior product manager Benjamin Hamm, like to hunt. Hamm’s cat Metric is one such hunter, and while this behavior was great for dealing with a rat infestation in his apartment, the fact that he brings back dead or wounded animals one out of 10 nights is a problem.
Ideas like locking the cat in or out all night or getting him a noisy collar to hamper his abilities weren’t the correct solution here, so Hamm turned to — you guessed it — machine learning. His solution uses an Amazon DeepLens camera mounted above a ramp to the cat door, which continuously monitors the conditions outside. It first asks, “Is there a cat?” progressing to, “Is it coming or going?” and finally, “Does it have some sort of prey?”
If there is, in fact, have a dead animal in tow, the cat door is electronically locked with the help of an Arduino board for 15 minutes, which seems like enough time for Metric to get the point. It then texts pictures to Hamm, and if that seems a bit cruel, it also donates money to the Audubon society as “blood money.” The system was trained with over 23,000 cat images, and although it currently only works with Metric, it’s been able to stop five out of six attempts at “extra” animal entry, with Metric only being unfairly locked out once.