Finding Fido With AI-Powered Pet Detection
Artificial Intelligence is a complex technology that’s at its best when it makes our lives simpler. That’s why we’re so excited about the new AI-powered features of our Xfinity cameras which can spot your dog, and make it super easy to find clips where he’s doing something cute…or something naughty.
Today, we announced the addition of a filter that allows customers to easily check in on any movement caused by their pets. So, for the 94 percent of camera users in a recent survey who said checking in on their pets was the best part of their day, simple pleasures just got simpler.
This is just the latest AI innovation in our cameras. In 2017, we launched better thumbnails for customers who opt to use our 24/7 video recording feature that dramatically improved our customers’ ability to find relevant motion events. Last year, we refined that feature further with enhancements to our computer vision models that allow Xfinity cameras to automatically detect when movement is triggered by people or vehicles … and now, pets.
The “pet filter” feature itself is simple and fun – at least according to our engineers, who have been testing it out on their own pets for the past several months – but the technology behind it is the result of a lot of hard work.
Computer vision PhDs from our Applied AI team in Washington D.C. worked with engineers from our camera team in Philadelphia to build the feature. Using an advanced form of machine learning called “deep learning,” we analyzed hundreds of thousands of home security camera clips donated by our employee testers. The goal was to build an algorithm that could quickly and consistently detect movement from a pet, and clearly distinguish it from a car, or a person, or anything else that might trigger motion detection.
Initially, we looked into commercial solutions that might work for our cameras, but we weren’t satisfied with the accuracy we were seeing. Detecting a dog photographed from five feet away, standing still in perfect lighting isn’t the same as detecting one running across the backyard 30 feet away, and we needed an algorithm that could do both – so the cameras could reliably spot … Spot.
Matt Zelesko is Chief Technology Officer for Comcast Cable.