Reconstruction

Reconstruction – recovery of 3D shape from images has been a primary problem addressed by the CVG under different perspectives. A number of successful methods have been developed using camera pose variation, stereo camera setting and lighting variation. Retrieving accurate geometry of both static and dynamically moving objects have been challenges that CVG has embraced since the beginning of its foundation. Cutting-edge approaches have been delivered for more than two decades, starting from classical optimisation based methods to exploiting deep neural network for inverting computer graphics generated data.

Recognition

Recognition – a very basic and intuitive task for vision based systems is the capability of understanding the content of a digital image. It is a fundamental building block of machine reasoning, essential for the ability of distinguishing objects from a scene. The CVG is committed to push the boundaries of latest research in object detection and segmentation in 2D and 3D, and make autonomous systems work on a wide variety of cameras. This includes planar and omnidirectional cameras, enabling CVG's technology to be applied to latest robotics and virtual reality. As reliability for such computer vision tasks translates into speed and accuracy for embodied systems, CVG is committed to develop state-of-the-art methods to overcome challenges in real industrial and societal settings.

Registration

Registration – with the aim of developing intelligent vision systems, the CVG has been working on interpreting multisensorial data. As such, the CVG has presented cutting-edge results in rigid and dynamic registration tasks. Early works enabled the fusion between speech technology with facial expression warping in order to recreate a virtual talking head emotionally mimicking both vocal and facial languages. Another fundamental tool for full scene understanding is object and camera pose estimation. Here the CVG systematically addresses the core challenges of localization, solving visual variation, view invariance and dynamics, supporting a variety of cameras, from classical planar to omnidirectional images.

Reasoning

Reasoning – with CVG's latest theme, the group advances from perception systems to interactive systems. Like in animals, vision plays a primary role at allowing intelligent biological systems to interpret and understand the world around them. With the rise of Artificial Intelligence, digital cameras have taken the place of primary vision receptors which enable capturing of images, and their interpretation. The growing adoption of vision for autonomous artificial systems is driving CVG research to applications such as visual navigation, autonomous exploration and general embodied AI, leveraging CVG's core strength in perception systems for intelligent decision making.