Based on Nonlinear Optimization and Keyframes Dense Mapping Method for RGB-D SLAM System

Abstract

This paper presents a RGB-D SLAM system, which can simultaneous local every pose of a camera motion and then construct an visualizable pointcloud map for the mobile robots. The system includes four main modules:visual odometry, optimization, loop closing and mapping. With the help of this system, a mobile robot in an unknown environment can acquire a global map and local its own position. In order to satisfy the requirement of high-accuracy location, we have used nonlinear optimization method with Bundle Adjustment based on the local map in optimization. Moreover, a new filter method has been used to remove redundant information points, it will speed up the mapping procedure. Finally we acquire a dense pointcloud map that could help robots finish its own job. The result of our experiment demonstrated the performance of our system in the speed and accuracy under these methods.