Computer Vision News - August 2016

The second thing he understood was that the idea was strongly connected to science, machine learning and computer vision. This technology would enable to expand the boundaries of human observation. At Cornell Tech, it was possible to develop the product next to scientists that understand machine learning, computer vision, health and also sleep science. But he also exchanged with many sleep experts: from the tech guys at Dana Children's Hospital in Tel Aviv, to Peretz Lavie, the President of Technion and a sleep guru. Lavie connected Glazer with Prof. Avi Sadeh from Tel Aviv University , which is also well known in the domain of sleep disorders in infants: he actually invented the gold standard on how to measure sleep for kids using a brace that you put on the leg with a medical device. He worked with Dr. Haviva Veler, who manages the Pediatric Sleep Center of the Weill Cornell Medical College in New York, one of the leading medical institutes in the world. This is how he defined the core scientific measurements that he wanted to observe: sleep quality, which is the percentage of time which an individual is capable of sleeping in bed; total sleep time; and number of visits during the night, since Nanit can detect when a parent goes to or leaves the crib; sleep onset, which is the time when baby falls asleep, aligned by Nanit with a sleep score. We asked Glazer about the algorithmic challenges his work needs to address: “ Challenges are analyzing and running computer vision algorithms in real time on the cloud in an efficient way that will also be affordable for parents. I don’t know of any company that has done it before. If you look at existing consumer cameras on the market, they detect motion or changing pixels. For example, if there is a movement or something in the background, there is an algorithm that will collect the synopsis like the time something happened. We are doing much more than this. A kid can move 10 times in an hour at night. It’s called arousals. It is not considered awake. It is an event if you use background subtraction models. If you do behavioral classification, and you look at the window and some other assumption, there are many different ways to do it. You can distinguish them. It would prevent the parents from having false alarms. You have much cleaner data and an experience that you are expecting from a product. The idea here is not to make a technology. The idea is to create a product which “ If the technology doesn’t meet the need, you cannot sell it ” 6 Computer Vision News Application Application is a natural extension of the human body. This is the idea. If the technology doesn’t meet the need, you cannot sell it.“