Projects and News

Pal2

Computer Vision for VR and AR

The AR and VR market is expected to grow to $150 Billion by 2020. This won’t happen without powerful computer vision. Once the preserve of specialist fields such as medical imaging, machine vision is the next revolution in computing and will be at the heart of a new generation of consumer devices. Augmented and virtual reality are two domains where computer vision is now fundamental to device manufacturers and application developers.

We predict that computer vision will be the limiting-factor for most, if not all AR and VR devices. The ultimate winners in this industry will, quite simply, be those with access to the best vision technology. With the largest companies in the world vacuuming up vision talent, there’s a serious risk of an expertise shortage.

Deep-learning has grabbed all the headlines recently, but it’s important to understand that these methods are just one of many and not suited to every type of vision problem. There are a growing number of talented vision engineers, but – relative to the growing need – experts remain in limited supply. Creating robust machine vision systems is notoriously hard. Typically every individual solution is engineered by large teams of specialists – an expensive and time-consuming business.

The team at Cubic Motion (and no doubt similar groups out there) believe vision-based applications should be accessible to everyone, not just the world’s largest organizations. Therefore, we’re excited that our team is soon to launch its Perception and Learning division providing a groundbreaking computer vision platform that enables developers without vision expertise to quickly create virtual and augmented reality software. This new division is led by internationally recognized computer vision experts – Dr Gareth Edwards, Dr Steve Caulkin, and Dr Steven Dorning.

We specialize in model-based vision – accurate analysis of the visual world based on machine learning. Indeed, Dr Edwards and former colleagues Professor Tim Cootes, and Professor Chris Talyor OBE (one of our advisors) invented the Active Appearance Model (AAM), one of the most successful inventions ever seen in computer vision. It was recognized by the IEEE with the Test of Time award because it: “… inspired many (if not all) subsequent works on deformable models for face analysis”. The AAM and its variants have been deployed in countless applications, including the Microsoft Kinect Face Tracker. However, even the AAM doesn’t make building vision systems accessible to non-experts.

So, we’ve created a new, faster, and more robust generation of technology that aims to change this. Our platform will give every developer the ability to quickly create and deploy advanced technical solutions for their applications.

As world leaders in transforming human performances into digital animation for video games and movies, the core of our success is the ability of our machine-learning technology to understand images and video. However, our technology isn’t specific to faces – we can track and analyze a vast range of complex objects. Indeed, developers can train our systems to recognize and measure items of their choice, using any of the visual data streams (including depth) available from the device sensors.

Our algorithms can analyze objects and scenes at incredible speed, even when the objects are deformable, such as faces, the human eye, clothing, documents, consumer goods, even foodstuffs! Developers don’t need vision expertise and can create deployable software in a matter of hours.