Research And Implementation Of Mobile Robot SLAM Integrating Vision And IMU

With the rapid development of mobile robots,the underlying technology of SLAM(Simultaneous Localization and Mapping)for autonomous navigation motion planning has become a research hotspot.The complexity of industrial application scenarios and the diversity of social requirements make it impossible to meet the positioning requirements of mobile robots with high accuracy,robustness,and real-time based on a single sensor.Monocular vision has the advantages of low price,rich feature information,and extremely low error accumulation.However,there are also problems of scale uncertainty,image blur and strong rotation under fast motion.Inertial measurement unit(IMU)have high instantaneous accuracy and can provide scale constraints for monocular vision,but also have problems such as serious error accumulation.There is a good complementarity between the two,and the fusion of these two sensors can achieve a SLAM system that meets the needs.This thesis aims at this type of fusion method and combines its own research results to implement a VI-SLAM system.First,this thesis analyzes the IMU and visual relevant SLAM model,and designs a framework structure of the fusion algorithm that takes into account the loose coupling and tight coupling,which improves the positioning accuracy and robustness of the VI-SLAM system;for the problem of slow convergence speed of optimization,A Levenberg-Marquadt optimization and updating strategy was improved,and on the basis of this,an in-depth experimental verification analysis of the system was carried out.Then,through the fusion and experimental analysis of the First Estimated Jacobian and free gauge algorithms,the problem of state quantity drifting to zero space was solved,At the same time,it avoids the problem that the unobservable state quantity becomes observable due to the artificial introduction of error information during the linearization process;it was proposed to accelerate the actual project through OpenMP multithreading to improve the real-time performance of the system.Finally,through a number of complex EuRoC real data sets,VI-SLAM and mainstream open source systems are compared and analyzed in detail,and the VI-SLAM system designed in this thesis is experimentally analyzed through the built mobile robot experimental platform.