In the human eye, the fovea centralis provides sharp central vision by packing in a high concentration of cone photoreceptors and connecting about half the nerve fibers from the optic nerve onto those cones. This gives high definition to the center 2 percent of the visual field, providing detailed vision and allowing us to read and track clearly the most important objects in the visual field.
TACO, a mangled acronym squeezed from Three-dimensional Adaptive Camera with Object Detection and Foveation, is a European research project applying high-definition 3D vision of a small centralized area to enhance robot vision. The TACO camera will emulate the human eye to increase the resolution of a small central area of the viewing field, refining the focus area to important objects, speeding recognition time and visual processing speeds with a three to 10 times increase in spatial recognition and frame rate. Advances in robust yet low-cost hardware, and the software necessary to rapidly detect objects in the environment, drive the project that started in February 2010 and will run three years.
Mimicking the Human Eye
While current 3D robotics cameras scan the environment to acquire images with a coarse level of 3D details and must evaluate all areas to find important objects, the TACO camera will follow a more human process. After scanning with the standard 3D resolution, TACO will apply fast object recognition techniques, then concentrate higher-resolution scans on the details of the primary object. Focusing on one area will reduce the optical processing load on the host system and generate higher-resolution scans of target objects more quickly.
Domestic and Industrial Service Robots
To more fully explore the TACO initiative, Robotics Business Review spoke with Jens T. Thielemann, lead technician on the project, via email. The primary new technology driving the TACO project is “advances within micromirror technology, allowing the production of optical MEMS (micro-electro-mechanical systems) devices. This enables the flexibility to continuously control the optical resolution (‘megapixels’) and frame rate of the camera.”
While no commercial applications for TACO technology have been identified at this time, “the project will include proof-of-concept studies that can later be evolved into commercial applications. These proof-of-concept studies will target domestic robots designed to support their owners with daily living routines and provide assistance in other ways. The technology can also be used in other classes of service robots, including those designed to work autonomously or semiautonomously in unstructured, often dangerous, environments, such as during the deconstruction of nuclear reactor sites.”
As for the results of their work, “individual technologies can be licensed from each individual partner in the project. Regarding the sensor as a whole, plans are to license this technology to a third-party camera manufacturing company.”
Co-financed by the European Commission, the TACO research consortium includes research facilities and robotics companies from Austria, Germany, the United Kingdom, and Norway. Project Partners: Technikon Forschungs und Planungsgesellschaft mbH (Austria), Shadow Robot Company Ltd. (U.K.), Oxford Technologies Ltd. (U.K.), TU Vienna (Austria), Fraunhofer (Germany), Stiftelsen SINTEF (Norway), and Carinthian Tech Research AG (Austria). Additional information regarding the TACO project can be found at www.taco-project.eu.