Computer Vision News - January 2019

At Osmo , they make games for children, but with a twist! Rather than using objects with electronics inside, instead, the game uses computer vision to detect the movement and placement of objects as the children play with them in front of a tablet screen or smartphone. Everything runs on the device and it doesn’t require sending images to the cloud. This makes their product good for feedback as well as for privacy. Brejeon heads a small team of four people in a company of about 50 employees altogether. Their team overcomes many challenges. First, they must meet the high expectations of game designers. At the same time, kids themselves seek a high level of accuracy in the games. Otherwise, they will lose interest and give up. That means that Brejeon and his team must create games with a very high accuracy, between 95% to 98% , that kids will find interesting to play. The third challenge involves figuring out how to run algorithms on the devices with enough speed and accuracy. The technology behind their product works in the area of robotics and uses ad hoc algorithm pipelines tailored for each game . In some cases, games can have false positives while other games cannot. Prior to creating each game, Brejeon and his team discuss with game designers to decide whether a game can include false positives, along with other features such as the hardware. Osmo started in 2013 and released their first game in 2014. Since, they try to launch three to four games each year. “ We’ve been quite busy! ” laughs Brejeon. In one of their earliest games called Words , two players compete to guess and spell on-screen images. If an elephant appears on the screen, for example, each player tries to spell the word first. The device will recognize the words written on the pages to decide which player gets a point. In some of the harder rounds, the word that they need to spell may not seem so obvious. Perhaps a mother elephant appears, and the kids will need to spell the word “ mom ” or “ mother ”. Osmo is an award-winning digital gaming system that allows children to play games using their hands and physical objects, while interacting with a smartphone or tablet. For the past four years, Arnaud Brejeon has led the Computer Vision team at Osmo. Brejeon spoke with Computer Vision News about his experience in developing this fascinating new way for children to learn and play digitally. 28 Application Computer Vision News Application “A mix of legacy computer vision and machine learning”