CVG are developing new algorithms for recognising, reconstructing and registering objects from images. The rapid progress in software, sensor and processor technology will lead to new ways of capturing and processing visual data.

3D Modelling

Building 3D models of static scenes is now very accurate and can be used, for example, to create 3D prints of an object. CVG are working on extending these methods and applying them to faces and the human body in motion. CVG have created a computer vision technique using multiple cameras and three lights of different colours to capture deforming surfaces at high levels of detail.

Large-scale Reconstruction

In construction and architecture the acquisition of three-dimensional models of buildings is becoming increasingly common. With continuously improving image quality and novel algorithms, computer vision provides a practical solution that captures colour information as well as geometry. CVG are collaborating with the University of Cambridge to explore the application of image-based reconstruction to infrastructure construction and maintenance.

Object Recognition

Given a large set of 3D points captured with a stereo camera or a depth sensor one task is to recognise objects from this data. In this line of research CVG are inventing new methods for finding and registering objects by extracting corner points and computing descriptors, each of which votes for object pose and identity.

Motion-based Recognition

Motivated by applications such as pedestrian detection to improve the safety of tomorrow's cars, CVG are investigating methods to rapidly analyse large video data sets. Machine learning methods are used to discriminate between pedestrians and background with continuous reliability.

Interface Technology

CVG are working closely with the Speech Technology Group to enable user interfaces that go beyond the current input technology. This includes touch-free input to devices by hand tracking to more complex interaction scenarios including natural speech and gestures.