What is behind OSCAR’s AI?

OSCAR detects floating objects by day, as well as and above all by night, when the human eye is unable to see obstacles. With its day and night vision, OSCAR augments the crew’s vigilance 24/7 and boosts the safety of a vessel by reducing the risk of it colliding with floating objects.

But how does it do that ?
OSCAR AI is in fact mainly based on computer vision, a field of study which enables computers to replicate the human visual system. It’s a subset of artificial intelligence which collects information from digital images or videos and processes them to recognize objects.

The entire process involves image acquiring, screening, analysing, identifying and extracting information. This extensive processing helps OSCAR to detect and identify floating objects in front of the boat and warn the crew to prevent collision. OSCAR translates the visual and thermal information acquired thanks to its colour and thermal cameras into useful information for the crew.

Computer vision is all about automatic pattern recognition. So to train OSCAR to recognise horizon lines and distinguish waves, sun or moon reflections, from mammals to boats or floating objects, you need to feed it images, lots of images, thousands, millions if possible, that have been labeled by us, humans.

The green line around the objects being OSCAR’s prediction. The blue line marks a manual annotation

To acquire those millions of images, we rely on recordings of months of navigation of the best world sailing racing teams and increasing number of OSCAR users. And to labelize this increasing number of floating objects acquired by OSCAR, we trained a team of methodical (and most of all very patient) annotators to indicate which pixels or part of the picture represent which type or category of object. This complicity between OSCAR users and the BSB team has led to the creation of an impressive maritime database of floating objects which is constantly growing, navigation after navigation. This database is then subject to various software techniques like deep learning, a self-learning process based on artificial neural networks inspired by the human brain to allow OSCAR to detect and recognize specific visual signatures of object categories in any given situation.

As an example, we already fed OSCAR with million images of “sailing boats” and it has subject them all to algorithms that let it analyze the colors, the shapes, the temperatures, the distances between the shapes, and so on, so that it can now identify the profile of what “sailing boat” means and can warn the crew that a “sailing boat” is in front of their vessel.

What’s great with OSCAR is that its intelligence continuously grows as the system is exposed to new data inputs from the worldwide OSCAR user community and that its detection and identification performances improve at each software update. OSCAR’s artificial vision is of course only one part of the technology behind OSCAR and we reveal you more in coming articles…

We hope this helped to explain how OSCAR’s AI works. If you still have some questions, please feel free to contact us. We will be pleased to answer you.