Research On SLAM Technology Of Mobile Robot In Dynamic And Complex Scene

Simultaneous Localization and Mapping(SLAM)technology has a wide range of applications in robot navigation,augmented reality,unmanned driving and other fields.However,most current SLAM methods are only suitable for static environments,such as warehouse robots and robots for the production of automotive parts.With the rapid development of artificial intelligence,the demand for SLAM methods suitable for dynamic environments is increasing.At the same time,mobile robot SLAM technology often has problems such as reduced positioning accuracy and even failure to work normally in dynamic complex and low-texture scenarios.Therefore,this thesis mainly focuses on the positioning and mapping of mobile robots to carry out the following research:(1)This thesis proposes an improved SLAM algorithm based on vision and deep learning to solve the problem that the use of feature points may discard important information of the image in the scenario of feature loss,resulting in inaccurate algorithm positioning.The algorithm uses deep learning network to replace the traditional ORB feature extraction method as the extraction of image feature information.Through experiments on the Open LORIS-Scene dataset,the results show that the RMSE index of the absolute trajectory error of the proposed algorithm is reduced by 66.34% compared with ORB-SLAM2,which improves the positioning accuracy of the robot.(2)In this thesis,an improved SLAM algorithm based on fast object detection is proposed to solve the interference problem of moving objects on traditional visual feature point extraction.The algorithm uses the advantages of the fast object detection algorithm to remove the feature points on movable objects without affecting the real-time requirements of the algorithm.Through experiments on outdoor KITTI datasets,the results show that the absolute trajectory error RMSE index of the proposed algorithm is reduced by 53.59% compared to the Dyna SLAM algorithm.In order to further reduce the influence of movable objects on robot positioning,this thesis proposes an improved SLAM algorithm based on vision and laser.The algorithm combines lidar with vision,uses the target detection algorithm to detect the image,and uses the laser point cloud algorithm to convert the point cloud into a bird’s-eye view,and then fuses the two algorithms to reduce the detection of the moving target area.Through experiments on outdoor KITTI datasets,the results show that the improved algorithm proposed in this thesis is superior to ORB-SLAM2 and Dyna SLAM algorithms,and has better robustness and positioning accuracy in outdoor dynamic environment.(3)In this thesis,a 3D reconstruction SLAM algorithm based on fast object detection and plane matching is proposed to solve the problem of missing image feature information in low-texture scenes.The algorithm adopts plane matching technology to improve the positioning accuracy in dynamic and complex environments.At the same time,dense 3D point cloud maps are generated,which can be used for applications such as robot environment analysis.Through experiments on the TUM dataset of indoor robots,the results show that the RMSE index of the absolute trajectory error of the proposed algorithm is reduced by 70.18% compared with the Point-Plane SLAM algorithm.Moreover,the algorithm runs at a rate of 22.20 fps in the test sequence,indicating that the algorithm can take into account positioning accuracy,running efficiency and robustness at the same time.