Research And Application Of Key Technologies Of RGBD-SLAM

In recent years,with the increasing ability of robots to perceive space environment,combining RGB-D information to realize the perception and recognition of space environment has become an important trend in the development of robots.Simultaneous localization and mapping(SLAM)technology is one of the key technologies of intelligent mobile robots,and RGBD-SLAM has been a research hotspot because of its comprehensive access to information.Generally,intelligent mobile robots are considered to be able to perceive the environment,cognitive objects and movement.The main technical difficulties include:object recognition and location,3D environment point cloud data construction,and 3D data-based path planning.In this paper,a wheeled mobile robot with RGB-D vision sensor is taken as the research object.Based on the key technology of RGBD-SLAM and combined with existing object recognition and location methods,an object recognition and location framework considering point cloud information,geometric size and RGB information is proposed,which realizes the recognition of space environment for intelligent mobile robot.Then,according to the projection principle of three-dimensional point cloud data,a solution of path planning based on three-dimensional data is proposed,which realizes the path planning of wheeled mobile robot based on three-dimensional data.Finally,the feasibility of the overall scheme is verified by building a mobile robot experimental platform.The main contents and conclusions of this paper are as follows:1.Combining inception-v3,Graham scanning and point cloud matching algorithm,an object recognition and location framework based on RGB-D information is proposed.In the process of robot vision realization of object recognition,the difference of surface color information between some objects is small,and it is difficult to distinguish objects by relying on the RGB information acquired by the vision sensor.This paper proposes a method for extracting the three-dimensional geometric size of an object using the Graham scanning method.Then,the data construction data set is extracted from the 3D point cloud image constructed by RGBD-SLAM,and object recognition is constructed by combining point cloud information,RGB information and geometric size.The positioning frame first uses the point cloud information to match the target point cloud in the data set to obtain the region of interest,and then identifies the object by combining the three-dimensional geometric size and color information.2.According to the projection principle of three-dimensional point cloud data,a solution of path planning for wheeled mobile robot based on three-dimensional data is proposed.The information acquired by wheeled mobile robots based on two-dimensional data only relates to the information of a certain height plane in three-dimensional scene.The planned path based on the plane information can not ensure that it is also applicable to other planes in space.In this paper,RGBD-SLAM is used to construct the three-dimensional point cloud map of the environment,and the three-dimensional point cloud map is projected to construct the navigation map based on the three-dimensional data.In path planning,the AD*algorithm and elastic band algorithm are used to realize path planning by matching the two-dimensional data projected from the current point cloud data with the navigation map.Based on the Robot Operating System(ROS),the virtual environment simulation of the path planning algorithm proves the feasibility of the path planning method.3.The experimental platform of mobile robot was built,and the object recognition and positioning frame and 3D path planning were further verified on the physical robot.The whole process of 3D point cloud image construction,2D projection map generation,object recognition and positioning,self-positioning and path planning navigation is realized on the mobile robot experimental platform.The experimental results show that the accuracy of the robot positioning object is within 34.8mm.Through the comparison of two-dimensional data navigation and three-dimensional data navigation by robots,it is proved that the solution of robot path planning combined with three-dimensional data can avoid the problem of collision between obstacles due to insufficient information in two-dimensional data navigation.