Computer Vision

“Cubic Motion’s computer vision technology is awesome – without it none of this would have been possible.”Kim Libreri – CTO Epic Games

“Cubic Motion’s real-time facial performance capture is awe-inspiring. In two years or so their computer vision technology will add a real-time element to enable streaming of live performances into interactive entertainment. This will change the way video games are made and played forever.”Tim Sweeney Founder and CEO – Epic

Scientific expertise with a reputation for results.

Cubic Motion creates world-class technology for the analysis of images, video, and depth data using a wide range of machine-learning technologies. The multidisciplinary scientific team is led by computer vision specialists Dr. Steve Caulkin and Dr. Gareth Edwards, and former research nuclear physicist Dr. Steven Dorning. Our scientific advisers include Professor Christopher Taylor OBE FREng, one of the world’s most eminent researchers in machine vision, formerly Head of Computer Science at the University of Manchester. Dr. Edwards, with Professor Taylor and Professor Tim Cootes, previously invented the Active Appearance Model – one of the most widely used (for example, as part of the Microsoft Kinect face tracking system) and widely cited methods in the analysis of video and medical images.

The research team at Cubic Motion has created a new generation of vision technologies – much more powerful methods than AAMs – and continues to lead research into state-of-the-art algorithms and software platforms for complex model-based analysis of images, video and depth data.

From data to productivity.

We specialize in converting complex data (e.g. video or depth streams) into meaningful signals. This information is then used to solve challenging problems, transforming productivity. Cubic Motion is of course already well-known for a major application of our technology platform – high-fidelity transformation of human performance into digital animation. This greatly accelerates the process of consistently producing world-class results. In turn, this allows leading developers to handle the unrelenting demand for ultra high-quality content.

Of course, our computer vision technology can be applied to a huge variety of problems and data sources, including low quality web-cam footage and in AR applications:

Deep analysis of images, video or depth.

Extracting ‘meaning’ from images and video is a big challenge. Research groups around the world attack the problem from many different angles. All recognize the value to so many industries of computers that can make sense of spatial data just like the human brain. Many research teams address challenges like: “find me all the images of bicycles”, or “show me all the photos of Uncle Dave”. In the age of social media, with huge volumes of images and video, these capabilities are full of potential.

Cubic Motion address an even more challenging problem. We seek detailed measurement of the objects in images, video and other spatiotemporal data. To stick with the face example – we’re not just asking for a ‘photo of Uncle Dave’. Instead, we want detailed description of every detail of his face. We need the location of all the facial features, in accurate detail. (We even need an accurate assessment of the location of hidden features such as teeth). There are two reasons for accurate measurement. The first is that in applications such as digital animation, detailed measurement is essential to map motion from a human face to a digital character. Simple detection isn’t useful. The second reason is even more important. If your system is not able to provide detailed location of key features, then it just isn’t an ‘image understanding’ system. Even a small child can look at a face and point at the lips, or for that matter, look at a bicycle and point at the handlebars. What this tells us is that the child understands the image (or video). Once you – or a computer – understand data, that data becomes a powerful driver of automation.

Always seeking the fastest solution.

Recent years have seen big advances in deep learning and its application to complex analysis problems. We’ve spent a lot research effort investigating and developing new approaches to feature detection and measurement. Our research team also tracks some of the great academic ideas in vision – some proposed even decades ago. Many powerful ideas in vision have gathered dust, seemingly impractical at the time of their proposal. Often, modern computing technology can transform their efficacy. Even the simplest-sounding ideas can be powerful when structured and matched with skill to a given problem domain. In fact, our overriding guideline is to look always to reduce the complexity of a proposed vision solution wherever possible. We apply the resulting systems to the measurement of complex and variable objects (such as faces) – in images, video, and other data sources such as 4D scans.

Advanced data analysis is a critical component of our facial animation pipelines. We also offer video and other data tracking services, along with real-time tracking capabilities.