A New Approach to Obtain Corner Point Coordinates in Subpixel Accuracy

Abstract

This paper deals with a proposing a new approach to obtain a corner point coordinates in subpixel precision. The describing algorithm here uses the regular cornerness map found by any of traditional corner detectors as the initial step. The main idea is to consider the cornerness values from neighboring area around found corner point as a 3D profile object where the key is to find its centroid. The part of this presented article is dedicated to verify the performance of our proposed approach where the comparison with another pixel and subpixel detector is done by applying the application in 3D reconstruction area. The all results were statistically analyzed, listed in a table and turned into specific findings and conclusions.