A method for detection of known objects in visual search tasks
Andrzej Stefan Sluzek · 2008
The paper outlines a novel method for visual detection of known objects in images of arbitrary contents. The major improvement over the existing techniques using keypoint detectors (e.g. Harris-Plessey, SIFT) is that the areas around pre-detected keypoints are approximated by various geometric patterns. Matching can be, therefore, done by comparing local geometric and colour structures of keypoints and by testing the geometric consistency over sets of matched keypoint pairs. The method can be used primarily for vision-guided search in mobile autonomous devices. The paper presents the mathematical foundations of the method and illustrates its performances for exemplary problems.