Research On Simultaneous Localization And Mapping Algorithm For Indoor Environment Of Mobile Robots

With the new round of artificial intelligence technology changes and the acceleration of social industrialization,mobile robotics has received more widespread attention and application,robot simultaneous localization and mapping algorithms(SLAM)is one of the key technologies.The current mainstream visual SLAM technology mainly uses point features in the environment as its visual observation,but in the indoor weak texture environment,it is often easy to cause low localization accuracy or localization failure due to insufficient visual observation.Therefore,in this paper,we study and construct a visual inertial SLAM system based on line features to improve the localization accuracy and robustness of the system in indoor environment.For the point features in the image,the Shi-Tomasi corner detection algorithm and KLT optical flow method are used for detection and tracking,and the point features are reconstructed using the triangulation principle;for the line features in the image,the LSD line segment detector and LBD descriptor are used for detection and matching,and the Plücker Line Coordinates and orthonormal representation are used to represent the 3D spatial lines,which facilitate the coordinate transformation and nonlinear optimization of the line coordinates.In addition,re-projection error models of point features and line features are constructed for pose optimization.For the IMU measurement,its integral is used to obtain an estimate of the initial pose,which is used as the initial value for the back-end pose optimization,and then the IMU pre-integration is calculated to construct the IMU measurement error in the back-end optimization.In back-end pose optimization,the number of keyframes is controlled using a sliding window model,two marginalization strategies are adopted to update the keyframe in the window,and the removed keyframe information is retained as a prior information using the marginalization approach.In the optimization process,t the sum of cost terms include the IMU measurement error,the re-projection error of the point features and line features,and the prior information.The local pose is iteratively optimized using the nonlinear optimization method to obtain more accurate position.In order to reduce the cumulative drift error of the system over a long period of time,bags of words model are used for loop closure detection,loop-closure edges are constructed and added to the pose graph for global pose graph optimization,the cumulative drift error of the system is calculated using the latest keyframe poses before and after optimization to correct the real-time pose estimation.Finally,the system is tested on the open source dataset Euroc.The error between the estimated pose and the real trajectory is calculated and compared with the mainstream visual inertia SLAM algorithm VINS-Mono.The experimental results show that the SLAM system built in this paper can obtain more accurate position estimation and meet the real-time requirements,and the localization accuracy is slightly higher than VINS-Mono under the indoor weak texture condition.