Research And Application On SLAM Algorithm Based On The Fusion Of Vision And Inertia

Simultaneous Localization and Mapping is a key technology in the field of mobile robots.In recent years,due to the rapid development and large-scale application of autonomous mobile service robot technologies such as autonomous driving,drones,and food delivery robots.Therefore,Although the visual slam that relies on a single sensor can obtain rich texture information,the camera cannot provide the scale information of motion.due to the low frequency of the camera,the camera tends to dropping frame.while it is maneuvering quickly or the objects observed by it are in quick motion.In order to offset the flaws of camera,Inertial Measurement Units comes to rescue.The IMU is able to measure the speed and angular velocity of the camera during its movement at higher frequency and is also capable of obtaining the scale information of the camera.In addition,because of the low price of IMU,the fusion of vision and inertia o SLAM is relatively cost-effective.This paper focus on the fusion of visual inertial SLAM system,which analyzes the mathematical model of SLAM,studies the physical characteristics of the camera,the method of image feature extraction,the error model of the IMU,the joint initialization of vision and inertial measurement units,the nonlinear optimization algorithm,pose solving algorithm based on the sliding window and marginalization.On the basis of the techniques and theories mentioned above and the study of the open source algorithm Vins Mono,this paper's algorithm makes an improvement to the front end and back end of Vins Mono respectively.Vins Mono only employ light flow method for image feature extraction and tracking,which is flawed in the environment in which textures varies.Hence,what a new feature is introduced in this paper's algorithm is that the combination of multiple algorithms for the feature extraction and tracking is designed for the enhancement of the robustness and reliability of the system.Due to the physical characteristics and design flaws of the camera,the captured image will be noisy.This paper adopts a loop detection method based on key frame noise reduction to improve the precision of loop detection and pose estimation.The experiment on vision inertia SLAM system of this paper is done on the platform of high performance gaming laptop.by comparing the results of this paper's system with the performance of VINS Mono,which is one of the most prominent of visual inertia SLAM designed by Hong Kong University of Science and Technology.The results of the experiment,by comparing the results of this paper's system with the performance of VINS Mono which is one of the most prominent of visual inertia SLAM designed by Hong Kong University of Science and Technology,demonstrates that the positioning precision of the SLAM system of this paper has attained the centimeter level.The experiment on the open data set of EuRoc demonstrates that the minimum absolute pose error of this paper's algorithm is 0.0059m,and the minimum relative pose error is 0.00021m,outperforming the VINS Mono system.As the result.this paper's system has the excellent positioning precision and robustness.