SLAM of Mobile Robot Based on the Joint Optimization of LiDAR and Camera

Abstract

Simultaneous localization and mapping (SLAM) is a key technology for intelligent mobile robots. SLAM solution based on a single sensor has limitations. The mapping based on 2D LiDAR SLAM cannot contain the complete structure of the environment, and the localization based on camera SLAM has low accuracy. To cope with this problem, a graph optimization framework of SLAM through the combination of 2D LiDAR and RGB-D camera is proposed. The scan data and visual data are added to the same framework for joint optimization, a cost function combining scan and visual data is proposed. On the premise of preserving the richness of visual data, the accuracy of localization is improved by adding the restriction of scan data, and in the loop closure detection stage, the Bag of Words (Bow) model based on visual feature point is applied, then finally, a better 3D point cloud map is obtained. Experiment results show that this scheme has better performance than using LiDAR or camera only.