A scalable algorithm for monte carlo localization using an incremental E2LSH-database of high dimensional features
Kanji Tanaka, Eiji Kondo · 2008
In recent years, high-dimensional descriptive features have been widely used for feature-based robot localization. However, the space/time costs of building/retrieving the map database tend to be significant due to the high dimensionality. In addition, most of existing databases are working well only on batch problems, difficult to be built incrementally by a mapper robot. In this paper, a scalable localization algorithm is proposed for incremental databases of high dimensional features. The Monte Carlo localization (MCL) algorithm is extended by employing the exact Euclidean locality sensitive hashing (LSH). The robustness and efficiency of the proposed algorithms have been demonstrated using the radish dataset.