We focus on nine societal areas.
TNO cooperates with companies, the public sector and other organisations, to apply our knowledge and expertise with and for others.
We develop knowledge not for its own sake, but for practical application.
TNO offers you the chance to do groundbreaking work and help customers and society with innovative, practical and smart solutions.
This invention is designed to improve the resolution attributed to a small object (bundle of pixels) moving across a series of digital frames. It is typically difficult to attain good resolution for this type of object in the context of a complicated scene. The invention improves a process called Super-Resolution (SR) for gathering data about the object from across multiple frames. Existing methods describe the use of SR to improve resolution for an object that would be blurry in a single given frame; however, they fail when tracking particularly small objects across the sequence.
The particular advance of this method involves using an iterative optimization function to improve resolution for a single object. First the process creates a high-resolution model of the image background. The object is then detected and registered frame by frame. The method then calls for an algorithm to define the boundary and intensity of the object. The algorithm minimizes "costs" associated with describing the object by prioritizing consistency and penalizing irregularity—for example, "spikes" from the background into the object.
This method can be used as part of long-range detection devices, such as visual and infrared cameras. Other suggested uses include alarm systems, and vehicle alerts. Although the method represents a particular advancement for the resolution of small objects, it also works to improve resolution for larger objects. It is particularly useful for creating a high-resolution image of an object moving at relatively constant velocity.
The method calls for defining a geometrically precise polygon around the object to be detected. The polygon defines the difference between object (foreground) and background. The method independently defines camera blurring and noise in order to cancel them. It uses a cost function to locally optimize decisions about which pixels to include as part of the object or part of the background. Protrusions into the object ("spikes") are penalized in this algorithm, while regularity is preserved. This method also assumes minimal acceleration on the part of the small moving object.