A Novel Image Similarity Measure for Place Recognition in Visual Robotic Navigation

In this work we tackle the issue of visually recognising a place without any prior knowledge of its position, even in a world where the same place can look different or many places can look identical.
To achieve a fast and robust image similarity measure for place recognition, we use the concept of quadtree decomposition combined with a number of standard image distance measures to create a novel image similarity method. Unlike the majority of current image comparison methods that use feature extraction and matching, our approach is a direct pixel-wise comparison of two images [1] gaining robustness through the incorporation of the quadtree concept. Quadtrees not only provide a noise resistant, fast, and easy to use comparison method, but also allow us to identify those image regions that genuinely represent changes within the environment.