3D Simultaneous Localization And Mapping For Mobile Robot With Autonomous Operation In Unknown Environment

The autonomous operation of intelligent robots cannot merely help or replace human to efficiently complete repetitive and complicated works,but can enter unknown complex environments,e.g.toxic or dangerous to complete specific tasks.Therefore,various types of intelligent robots are now widely used in different fields and gradually enter people’s daily life.3D simultaneous localization and mapping(SLAM)is the key supporting technologies for robot in their process of autonomous operation.It enables the robot to complete its own position and pose estimation in real-time according to the environmental information perceived by its own sensors,and build a map consistent with the surrounding environment,which make the robot able to correctly understand the surrounding environment,and successfully complete the task of autonomous operation.Nevertheless,mobile robots during their process of autonomous operation in unknown environments are often interfered by dynamic objects,which will lead to a large amount of error accumulations in pose estimation;furthermore,when there are few texture features in the environment,the robot cannot obtain enough feature information,so as to cause the algorithm tracking failure.According to the application requirements of mobile robots during their process of autonomous operation in unknown environments,this paper employs Delaunay triangulation method to build a triangulation network that can represent the geometric relationship between matching map points of two frames to detect dynamic feature points.Considering that dynamic feature points may be misjudged,more feature points are extracted when matching two adjacent frames to realize the compensation of static feature points.Meanwhile,a combinatorial feature extraction method that combines point features and line features is adopted,so that when mobile robots encounter fewer textures in unknown environments,they can also obtain sufficient feature information in the environmental perception stage.Additionally,the feature matching and error matching elimination methods are performed on the extracted point line features,based on which the dynamic feature points are effectively detected and eliminated to reduce the error accumulations introduced by the dynamic objects,so as to realize the accurate estimation for the pose of mobile robot.To reduce the redundancy of keyframe and the computational complexity of closed-loop detection,a high-performance keyframe extraction strategy is proposed by introducing sliding window to improve the real-time performance of 3D SLAM system.Then,according to the extracted point-line features,efficient closed-loop detection and error closed-loop elimination are carried out to further improve the accuracy and consistency of 3D sparse mapping.On this basis,the RGB-D image obtained from depth camera and the keyframes pose are utilized to generate a dense 3D map,and further combine with the Octree map to build a dense 3D navigation map.Finally,the algorithms proposed in the previous study are integrated and optimized as a whole to form a complete 3D SLAM system.Through a series of simulations on multiple sets of public datasets and a comprehensive experiment under indoor real dynamic scene,the results show that the 3D SLAM algorithm developed in this paper has high positioning accuracy,the constructed 3D dense navigation map is consistent with the actual environment,also,the program runs smoothly and meets the real-time requirements in the experiment.