Computer Vision News - June 2019

Their first product in development is the smart co-pilot which helps the crew on board navigate safely and efficiently. One of the functions supports watch keeping and provides technology such as a camera with 360 views and 24/7 monitoring. This functionality proves more reliable than humans , who have limited visibility and may fall asleep or get sick while on watch. The crew of a cargo ship can also have difficulties navigating in crowded waterways. Shone is developing sensor fusion between different data sources that improves consistency, helps prioritize dangerous targets, and provides a recommendation of what the crew should do in terms of safety. Cargo ships face similar types of problems as autonomous driving for cars, but with a totally different set of constraints. On the technical side and in terms of retrofitting cargo ships, the existing technology is much different on cargo ships than in cars. Whereas a car only needs to see 15 meters (about 50 feet) ahead, cargo ships need to see 500 meters (1,640 feet) away . Seeing 15 meters away proves completely useless for a ship because, at that point, the ship cannot avoid an accident . Long distances involve a lot of estimation and require the use of different types of cameras. On the algorithm side, they face the same challenges. As Ugo explains: “ I guess the difference is seeing not far away at a very high FpS (frames per second) versus seeing really far away at a lower Fps; because you have so much inertia, but you have also a lot of real time issues in the sense that you need to detect from really far away . Otherwise, you won’t be able to prevent it. ” On the business side, autonomous shipping faces many different challenges than self-driving cars, primarily, because it doesn’t involve the same type of vehicle. Meanwhile, on the control side, self-driving cars operate on roads, which do not move. On their side, autonomous shipping must navigate on moving waterways. Autonomous shipping does share similarities with self-driving cars in terms of the algorithms . “ The algorithms are pretty much the same as self-driving cars. You need object detection, classification tracking, segmentation to know what’s water and what’s not water. We use neural nets which are state of the art to do that. 20 Application: Autonomous shipping Computer Vision News Application “Autonomous shipping does share similarities with self-driving cars in terms of the algorithms”