Research On SLAM Technology Based On Vision/IMU Fusion

With the development of unmanned autonomous driving,AR / VR and other technologies,visual SLAM technology has become one of the hot frontier technologies.Traditional visual SLAM solutions have some shortcomings,such as monocular vision SLAM cannot measure depth,binocular vision SLAM calculation is complex,and deep vision SLAM is not applicable.To solve the limitations of these solutions,one of the many solutions is to perform fusion calculations of multiple sensors to complete the pose estimation of the system.The vision / IMU fusion-based solution designed in this paper draws on the open source framework of VINS to achieve typical applications on specific sensors.The fused sensor system completes the fusion of the visual information and the attitude information of the inertial measurement unit IMU.The KLT sparse optical flow method,triangulation,and PNP algorithm are used to calculate the positional relationship between the sensors.The inertial measurement unit IMU obtains the pose estimation through a pre-integration method,and then combines the visual information and the attitude information of the IMU through a tightly coupled algorithm to obtain a more accurate and robust pose estimation.In the back-end optimization part,loop estimation and four-degree-of-freedom pose map optimization are used to optimize the system's estimated path to obtain more accurate and robust path estimation results.Although the SLAM system can run stably in most environments,it still has disadvantages for harsh environments and dynamic scenarios.Therefore,this paper introduces an improved algorithm to try to solve the problems of low system accuracy and poor robustness caused by image blur,noise,exposure,and fog blur.At the same time,a dynamic scene optimization algorithm based on deep learning is used to solve the problem of mismatching feature points of dynamic scenes in the SLAM system,which improves the accuracy,robustness and environmental applicability of the system.Otherwise,experiments are performed on various algorithms of the SLAM system based on vision / IMU fusion and the operating state of the system in various environments.The running test was performed in various indoor and outdoor environments,and good test results were obtained.For the SLAM system of vision / IMU fusion,this paper studies the data transmission of the system's sensing sensors,the front-end algorithm of the visual inertial odometer,the back-end optimization processing of the SLAM system,and the optimization and improvement algorithms for images and dynamic scenes.Compared with other systems,the improved system designed in this paper has better environmental adaptability and higher accuracy of path estimation.Combining the above-mentioned various algorithms and various test results,we can find that the vision / IMU fusion-based SLAM system designed in this paper has good computational complexity,applicability,accuracy,and robustness,which can meet the system requirements in most environments Operational requirements.