| With the progress and development of science and technology,the application fields of mobile robots and other industries are becoming more and more extensive.In complex task scenarios,mobile robots can complete complex path planning and environmental exploration,and at the same time greatly reduce the consumption of labor costs.With the outbreak of the new crown pneumonia,it is more necessary to reduce the close contact between people,so mobile robots can play a huge role in people’s daily life.Mobile robots need to construct maps in unknown environments and obtain their relative positions on the constructed maps.Therefore,Simultaneous Localization and Mapping(SLAM)technology is the core technology to realize autonomous motion of mobile robots.The positioning information of the robot and the built environment map of the SLAM technology enable the robot to make better action decisions and path planning in the follow-up.Since most SLAM technologies are based on the assumption that the environment is static,the actual environment is dynamic,resulting in low positioning accuracy of mobile robots in complex dynamic scenarios,and large errors and noises in map construction.This greatly limits the application of SLAM technology in mobile robots.In order to adapt SLAM technology to various complex dynamic scenes,the main research content of this paper is to improve the positioning and mapping accuracy of mobile robots in complex dynamic scenes,so that they can run smoothly in complex dynamic scenes.The main work of this paper is as follows:1)This paper studies some basic visual-inertial SLAM models,introduces and derives the visual measurement model,inertial measurement model,and visual and inertial sensor information fusion process in detail,and introduces the basic framework of the visual-inertial SLAM system in this paper.The traditional ORB feature point extraction unevenness and clustering phenomenon in the visual odometry are improved,the LK optical flow method is used to track the feature points,and the feature matching optimization algorithm is used to solve the mismatching phenomenon that occurs in the feature point matching.Eliminate the feature points to make the matching of the two frames before and after more accurate.2)Considering that when the visual-inertial SLAM algorithm is running in a dynamic scene,a large number of dynamic features will be extracted in the front-end of visual measurement,and the positioning accuracy in a dynamic environment is poor.In this paper,the initialization and feature extraction algorithm of the visual-inertial system are improved,and the dynamic feature point filtering algorithm based on geometric constraints and semantic information is used to eliminate the influence of dynamic objects on the positioning accuracy of the visual-inertial SLAM system.Through experiments and result analysis,it can be proved that the improved SLAM algorithm in this paper can effectively filter out dynamic feature points and improve the positioning accuracy of SLAM in dynamic scenes.3)This paper introduces the initialization of the visual-inertial SLAM system,uses nonlinear tightly coupled joint optimization after the visual-inertial initialization,and introduces the back-end loop detection method and closure constraints.For the interference of the dynamic environment to the operation of the visual-inertial SLAM system,this paper proposes to use an adaptive factor to adjust and optimize the proportion of visual information and inertial information in the back-end optimization of the visual-inertial SLAM system.The improved algorithm effectively improves the robustness of the visual-inertial SLAM system in dynamic scenes.4)In this paper,based on the Ubuntu16.04 system,the algorithm in this paper is tested and analyzed using the static and dynamic scenes in the TUM-VI visual-inertial data set.First,the effectiveness of the feature uniform extraction algorithm proposed in this paper is verified and the feature matching performance test is carried out on the dataset using the feature matching optimization algorithm in this paper.Then,the dynamic feature points extracted by the visual measurement front-end in dynamic scenes are filtered out.The same experimental equipment and conditions are compared with the VINS-MONO algorithm.The experimental results show that the algorithm in this paper can run smoothly in dynamic scenes,effectively improving the robustness of the visual-inertial SLAM system.The positioning accuracy of the algorithm in this paper is not only higher than that in static environments.The VINS-Mono algorithm is good,and the positioning accuracy of the algorithm in this paper is also better than that of the VINS-MONO algorithm in a dynamic environment. |
