Simultaneous Localization And Mapping Based On Tightly Coupled Deapth Camera And IMU

Currently,most of the mobile sweeping robots that can be seen on the market are using lasers for positioning navigation,but the product effects and user experience are very poor which is manifested in that it often collides with obstacles.Moreover,the laser does not work effectively in certain situations,such as in the case of glass reflection.In addition,in some environments,such as kitchens,bathrooms,etc.,it is not desirable for the sweep robot,because this environment often has a lot of things that do not want sweeping robots to touch,such as water,oil,etc.Therefore,the sweeping robot is required to recognize these scenes.In this paper,the SLAM system combined with vision and IMU is used to solve the above problems.During the development of SLAM,the accuracy and robustness of the SLAM system are gradually enhanced from the initial laser positioning scheme to the recent visual scheme and then to the multi-sensor fusion scheme.The evolution of SLAM from a filter-based approach to an optimization-based approach benefits from the development of hardware and the discovery of new mathematical structures.This article addresses the issue of robot positioning and mapping in unfamiliar environments.It uses the tight integration scheme of vision and IMU for positioning,builds a complete SLAM system,and fully introduce the front-end tracking,back-end optimization,and loop-detection of the system.The system discards some speeds in order to consider the accuracy,and decides to use the method of graph optimization to optimize the camera pose or map point in different parts of the system to obtain precise poses and map points.Compare with the traditional SLAM system,the front-end point depth acquisition,front-end tracking strategy design,and pre-integration of IMU data were modified and optimized.Finally we compare this solution with the traditional system through experimentsIn order to solve the problems of “low texture”,“lighting change” and “motion blur” in traditional visual SLAM,this paper proposes a scheme for extracting line features and point features at the front end of the system.Different tracking and matching strategies are used for point features and line features to increase the accuracy and robustness of the system;in the IMU processing part of the system front end,the scheme of trapezoidal integration on the rotation matrix is adopted to ensure the integration speed and accuracy;in order to maintain the computational complexity and precision at the back end of the system,the local map size is fixed and marginalization techniques are used.