Elastic orb: non-rigid transformation based slam

Abstract

We present Elastic ORB, an improved implementation of a novel visual SLAM approach. Our system uses point cloud data to build the map of the environment and for the localization of the robot, odometry data is used. A frame-to-frame loop closure method is used at the back of the SLAM system using the concept of deformation theory. When building the reduced model of the map, the ORB keypoint extraction algorithm and the FPFH feature descriptor are used. Our system is capable of mapping a larger environment overcoming the limitations of existing SLAM systems which use the deformation theory at the backend.