Research And Implementation Of RGBD-SLAM System For Indoor Wheeled Robot

With the development of technology,intelligent robots have entered almost every aspect of human life.Ubiquitous intelligent robots are changing people's lives.Facing the growing market demand for service robots,SLAM(Simultaneous Localization And Mapping),as one of the core technologies in the field of robot perception,has attracted more and more attention.This paper has studied RGBD-SLAM of indoor wheeled robot,and the main work is as follows:(1)This paper investigated the development of visual SLAM algorithm,and summarized the key technologies of SLAM.The observation model of the camera was analyzed,and the mapping process of the object from 3d space to the camera imaging plane was modeled.The 3d point cloud was established through the pinhole camera model and the RGBD camera model to restore the environmental information in the image.(2)A visual odometry module is built,which takes the color image and depth image of the RGBD camera as input,estimates the camera's motion trajectory based on the pixel information between adjacent images,and optimizes the visual odometry by minimizing the reprojection error and local map feature point matching.(3)The method of image matching is used to realize the loop detection of key frame sequence.Then the pose graph is constructed according to the constraints obtained by the visual odomety's estimation and loop detection,and the robot's pose is globally optimized in the way of graph optimization,so as to reduce the increasing cumulative errors in the process of SLAM.(4)In the indoor environment,human body is a common dynamic obstacle for the robot,but if the map containing human body is directly used for path planning,the robot will regard it as a static obstacle to avoid.In general,people do not stay in the same position,which causes a lot of unnecessary travel costs.Aiming at this problem,this paper based on u-net,vgg-16 and image data augmentation technology realized the semantic segmentation of human body in RGB image,thus creating the indoor environment point cloud map without human body.(5)Experiments were carried out in the indoor environment based on ROS(Robot Operating System)to verify the reliability of the proposed rgbd-slam algorithm.