Computer vision method for recognizing objects and/or areas in digital image in environment by autonomous, programmable, medium-sized humanoid robot, involves retaining one of interest points of current image by filter

摘要

The method involves extracting interest points (I, J) in reference and current images according to an approach by descriptors e.g. scale invariant feature transform descriptors. The points of the current image are mapped with the points of the reference image, and filtered by a spatial coherence filter. One of the points of the current image is retained by the filter if the point belongs to a pair of interest points of the current image according to geometric constraint determined based on the descriptors and a pair of the reference image, where the constraint is valid in projective geometry. Independent claims are also included for the following: (1) a computer program comprising a set of instructions for performing a computer vision method (2) a humanoid robot for recognizing objects in an environment.