Symmetry Signatures for Image-Based Applications in Robotics
Kai Huebner, Jianwei Zhang · 2007
In this work, bilateral symmetry has been proposed as a concept for the extraction of features from the visual data and their application to robot navigation tasks. Symmetry in shape and vision is strongly motivated by biological and psychophysical aspects. It is a natural feature that can be found in many scenes, whether they show structured indoor or unstructured outdoor environments. We conclude with a review on the three main topics: 1. Symmetry as a Feature: Symmetry has been investigated in several domains like biology, psychophysics, architecture and art. Accordingly, symmetry has also been applied as a valuable attentional feature for the extraction of regions of interest or for object description by symmetric properties in computer vision. Motivated by psychophysical experiments on symmetry perception, a fast and compact onedimensional operator was supposed earlier (Huebner, 2003) to handle horizontal and vertical bilateral symmetry measures only. The operator overcomes the problem of symmetry detection methods in literature that use large operators which are mostly unsuitable for robotic real-time tasks. However, each of these strength-based operators returns a relative, commonly normalized value of symmetry for each image element. For