This work attempts to develop a method for SLAM using semantics based on FastSLAM 2.0. Our approach to semantic mapping consists of segmenting images obtained from two sensors (optical and radar) aboard a UAV. We then identify landmarks within the segmented image, followed by the construction of relational trees with the landmarks; these trees are then used at consecutive time-steps of the robot’s motion for its localization as well as update of the landmarks. The term semantics has been used for region-landmarks which are validated with a look-up table (LUT) of the predefined surface type information, a superset of the robot’s actual environment. Finally, based on particle filters, the posterior density of the state of the robot is estimated and a 2-D semantics map is constructed. The methodology has been tested in a situation wherein the robot’s true environment and path have been simulated. For simulation we consider satellite images of optical and radar sensors of the robot’s environment. At different time-steps the robot’s images are cropped from these images, incorporating errors in the robot’s control information. Experiments carried out on the simulated environment have provided encouraging results.
SLAM Using Relational Trees and Semantics
Advanced Materials Research ; 452-453 ; 648-653
2012-01-24
6 pages
Article (Journal)
Electronic Resource
English
Semantics of Non-Deterministic Repairable Fault Trees
German Aerospace Center (DLR) | 2019
|Gamma-SLAM: Visual SLAM in unstructured environments using variance grid maps
British Library Online Contents | 2009
|British Library Conference Proceedings | 2011
|