Projects
The Computer Vision Group at the Toshiba Research Lab is working on new methods for recognising, reconstructing and registering objects, leading to entirely new interactions between the digital and real worlds.
3D Modelling
Reconstructing 3D shapes from images has many potential applications. 3D modelling from static images is now very accurate. CRL is now focusing on the harder task of reconstructing faces and the human body in motion. We have developed a technique using a single camera with three lights of different colours. When these three lights illuminate the surface from different angles, there is a one-to-one mapping between the red-green-blue colour reflected by the surface and its orientation. CRL uses the depth-map produced from these lights to capture the changing shape of an object in motion. Surface bends, wrinkles and temporary folds can be captured with a high level of detail.
Gesture User Interfaces
Computer vision allows touch-free input via gestures which are complementary to other input devices for computers and televisions. CRL is developing simple and reliable gesture interfaces which can identify a single user in a group using an attention mechanism. It can then reliably recognise and track the user's gestures even under rapid motion or changing illumination. This is achieved by extracting multiple cues, including colour, motion and appearance, within the tracking framework. Since different cues are less reliable in different situations this multiple-cue approach is key to robustness.
Motion Analysis
Motivated by applications such as pedestrian detection for car safety, CRL is investigating methods to analyse motion patterns. Motion is a powerful visual cue, which can give a lot of information about the objects in the scene. The framework that is currently being developed extracts point trajectories from the video and subsequently analyses these using an efficient random forest classifiers.
Terms and Conditions
