High-Performance Vision

Tasks for which the human eye is too slow are the challenges addressed by the "High-Performance Vision" research field. The research focus is on ultrafast line scan sensors and cameras, scalable embedded vision systems, computational imaging and deep learning

Here, technology for image processing is researched with a focus on real-time capabilities at extreme data rates and high complexity. To develop this type of image processing systems successfully in the highest range of performance, both fast components and a carefully balanced overall architecture are essential. In this sense, a hardware/software co-design is the most successful approach.

New concepts for still more powerful cameras and hardware also require new concepts for software and algorithms. Especially for new multiprocessor and future many-core systems, knowledge of a correct distribution of data and evaluation loads on the individual components of the hardware architecture are the key to a successful high-speed image processing system. The research topics in this research field thus include both cameras as well as hardware and software, along with their interplay in image processing. An additional important aspect is the development of user-friendly interfaces allowing these complex systems to be configured safely and simply, and used successfully.

Our high-speed image processing technologies are typically used for two- and three-dimensional surface inspection. Examples are quality testing in printing applications. Here, printed paper is moved past the cameras at high speed and even the smallest of printing errors detected. Another example is the quality testing of rails. Here, rails are examined for safety-relevant defects from a moving vehicle.

There is a need for high-speed image processing technologies in high-speed production processes in which quality is a safety-relevant property of the products, but also in computer-supported testing of our infrastructure, such as rails and streets.