Research On Visual Inertial SLAM Using Point And Line Feature

With the gradual replacement of human by robots,factories are developing in the direction of highly intelligent and large-scale.It is of great significance to realize the autonomous navigation of mobile robots.In autonomous navigation task,simultaneous localization and mapping(SLAM)is one of the key technologies.It is a process of constructing a global consistent model through local observation of an environment.Due to the unique advantages of visual sensors,visual slam has become a research hotspot in recent years.However,it is easily affected by the inconsistency of ambient light and dark,lack of texture and so on.The factory indoor environment with too similar scenes will affect the relocation function of visual slam.Therefore,it is of great significance to improve the positioning accuracy and robustness of mobile robot in complex indoor environment.Point feature is the most commonly used feature in visual slam.However,in the artificial structured environment,there are abundant line features.Adding line features to form additional visual constraints can effectively improve the accuracy of the system,and the constructed map has a more intuitive geometric structure.However,the slam system based on point feature and line feature will be seriously affected by the appearance of blank or no edge feature in the environment.By introducing the idea of direct method and complementing the feature method,a semi direct slam system is formed,which can effectively improve the robustness of the system.In addition,the IMU sensor is added to fuse with the camera,which can be used in the case of rapid rotation or light mutation,It can still track and locate accurately.Aiming at the localization and mapping of mobile robot in complex environment,this paper uses the fusion of binocular camera and IMU sensor to extract point and line features at the same time,and effectively combines the feature method and direct method to propose a point and line vision inertial slam system based on semi direct method.The main achievements of this paper are as follows1.On the basis of DSO,a sparse direct slam system based on binocular camera is proposed.The scale drift problem is solved by using binocular camera,and the inverse depth of gradient point can be accurately estimated by using binocular matching algorithm.The loop detection and closed-loop optimization module is added to eliminate the cumulative error caused by pose estimation,and further improve the positioning accuracy and robustness of the system.2.On the basis of point feature and line feature,complete the tracking and optimization of point line integrated feature,and use IMU sensor and binocular camera fusion.In the back-end optimization,build a sliding window algorithm based on point line feature and IMU measurement error joint optimization,complete the visual inertial odometry(VIO)based on point line feature.3.Aiming at the shortcomings of the feature method,a semi direct slam system is designed by combining the feature method with the direct method,which makes it still be able to track and locate successfully in the auxiliary environment of texture missing and large illumination transformation.Loop detection and global pose optimization are added to form a complete and unified slam system.