Apparatus and method for multi-object detection in a digital image

摘要

Method for object detection employing finding characteristic points in at least one pattern and an input image; for each pattern image matching the pattern's characteristic points against the characteristic points in the input image in order to define points correspondences; the method further comprising the steps of: considering each points correspondence as a vote and for each correspondence points determining an adjacency feature which denotes a level of match between the corresponding points whereas votes of each pattern create a voting space; discarding votes having the adjacency feature value below a threshold; using the remaining votes to create a votes image that graphically represents votes; detecting characteristic points in the votes image; considering these characteristic points as potential propositions for centers of the object instance of a corresponding pattern; executing objects detection based on the votes image.