A Scan-to-Locality Map Strategy for 2D LiDAR and RGB-D Data Fusion

2D LiDAR and RGB-D camera are two widely used sensors in the task of simultaneous localization and mapping (SLAM). In spite of various map methods for SLAM, an effective strategy which is able to fuse 2D-LiDAR and RGB-D data in a uniform style is still expected. This work proposes a novel map strategy for 2D-LiDAR and RGB-D data. Different from traditional maps where the map information for different views is stored separately in their original two-dimensional (2D) grids or three-dimensional (3D) voxels, in the proposed map strategy, the data obtained by multiple sensors for current view are fused as feature vectors and stored in the current grids. We call this as Scan-to-Locality (STL) map strategy. In re-localization phase, the fusion vector obtained at the current view is used to find the similar ones from the STL map using distance based similarity evaluation and Siamese-network image matching technique. In this way, the re-localization strategy is optimized as a coarse-to-fine schema in the proposed STL map strategy. We validate the proposed method on the widely recognized in-door navigation database Robot@Home. The experiments indicate the proposed method own the abilities of accurate re-localization in nearly real time.