Research On Indoor Localization And Mapping Of Mobile Robot Based On RGB-D Data

As we all know, there is no GPS signal in the building. How to effectively achieve the information of the robot's position indoor becomes a major challenge in the study of intelligent mobile robots, but SLAM (simultaneous localization and mapping) aims to provide a good solution. In recent years, the research on the 3D map building based on RGB-D sensor has formed a research boom. How to apply it to SLAM has attracted the attention of scholars both at home and abroad.This paper mainly conducts research on the simultaneous localization and mapping of indoor mobile robot based on the RGB-D data. According to the background of the subject,considering the significant structural characteristics of the indoor environment and human demand on the map navigation and human-computer interaction, this paper conducts the study of two aspects including visual odometry considering ground constraints and navigation and cognitive map building.Firstly, it has elaborated the background and the significance both in the simultaneous localization and mapping and Artificial Intelligence. The research status of simultaneous localization and mapping was summarized, and RGB-D sensor was introduced. And then this paper determined the researching details combined with projecting background.Secondly, this paper introduced the basic theory of SLAM based on RGB-D data,including common framework based on graph optimization and details of the key technology.According to the discuse we chose the suitable strategies and methods of the research for this topic. And these introductions of basic knowledge would contribute to the following evolution and verification.Thirdly, through analyzing the structure of the mobile robot's RGB-D sensor system, this paper built the appropriate model under the coordinate system. According the probabilistic description of the SLAM problem, we established the mathematical model based on the graph,which made the foundation of following research.Fourthly, considering the structural characteristics of the indoor environment, through the depth image detection we can get point cloud surface. Combining the spatial information of high precision obtained by plane fitting method with feature extraction and matching based on the ORB algorithm,this paper established more accurate visual odometry model for subsequent graph optimization process. And the simulation basing a public data set of RGB-D data has verified the validity of the improved algorithm.Finally, the 3D point cloud map composed of disordered points causes the lack of description of the indoor scene and it's difficult to apply it in robot navigation and mission planning. On the basis of simple point cloud map, a new map building method is proposed for robot navigation and environment cognition. First of all, the use of octree data model established a three-dimensional grid map for navigation and experimental verification of the effectiveness of the algorithm has been done; then,based on bags of visual words model we established new forms of semantic map for scene recognition, and tested the effects of the algorithm in the simulation.