Simultaneous localization and mapping using adaptive appearance based loop-closing detection

Visual appearance based scene recognition has been widely studied in the past decade. However, such systems still present failures in dynamic, heterogeneous urban environments. This paper presents a methodology in order to integrate an advanced Adaptive Appearance Based Loop-Closing Detection algorithm into a graph optimization based Simultaneous Localization and Mapping algorithm. The system described uses stereo visual data, captured in discrete locations in order to detect loop-closure situations and to build, respectively update the topological map of a large, complex environment.