Computer Vision News - July 2016

The development was much more difficult than he expected and it took several years until Poseidon was able to bring a working system to the market. In fact, classic motionless detection models are not a viable solution, due to shadows and lights which might generate false positives of motionless bodies at the bottom of the pool . Thus, the main challenge is to be able to distinguish somebody in a real danger (typically, a motionless volume on the bottom of the basin). Poseidon worked instead on well- known computer vision models, the first one being stereovision : two cameras viewing the same scene from different angles. We also developed an algorithm able to differentiate between a volume and a projection of shadow or light: when a person swims on the surface, two points of view will give two different projections, while a person laying on the bottom of he pool gives the same projection to both cameras. This is how we are able to identify true positives. The most problematic factor in this domain is time: in order to save the life of somebody laying on the bottom of the basin, you cannot let more than one mi nut e pas s between t he immersion of the airways and the rescue by a lifeguard. After one minute, the consequences might become irreversible and after two “ Image processing is fully automated for hundreds of thousands of images per day ” 14 Computer Vision News Application Application