Research on the Organization Method of Topological Nodes Based on Mapping Matrix for Place Recognition

Inspired by the spatial cognition and localization mechanism of animals, this paper focuses on the organization method of topological nodes for place recognition. The topological map is constructed on the basis of graph by assigning each node with a set of features for localization, in which the inertial/visual odometer provides the required motion parameters. The mapping matrix between nodes and various features is constituted to realize effective organization and utilization of empirical knowledge. The availability criterion of empirical knowledge and the corresponding location error are established based on analysis of the mapping matrix, which provides a priori probability model required in data fusion applications. The topological map of a campus is constructed with scene features and the place recognition result is discussed, which shows the effectiveness of the proposed method.