A Sequence-Based Visual Place Recognition Technique with Segmented Database and Compact Sequence List

Abstract

Sequence-based visual place recognition algorithms have been proven to be able to handle environmental changes caused by illumination, weather, and time of the day with handcrafted descriptors. However, an exhaustive search for all images in a query sequence is computationally expensive. In this paper, we propose a technique that can significantly reduce the size of searching space for sequence matches while remaining state-of-the-art accuracy. Firstly, we managed to achieve a better selection of reference candidates of images in query sequence by segmenting the database according to similarity. Then, a much more informative and compact query sequence is designed by removing all the unnecessary images in the original query sequence. State-of-the-art performance is reported on a public dataset with challenging environmental changes. Our algorithm shows comparable accuracy with other current best results and exceeds all the other methods in the dataset with illumination variation. In addition, the decrease in execution time and higher success rate for selecting candidates of query images for sequence match is also provided.