Research On Visual Inertial Navigation SLAM Based On ORB-GMS

In recent years,computer vision technology combined with intelligent robots to achieve autonomous positioning and navigation of robots is one of the current research hotspots,and Simultaneous Localization and Mapping(SLAM)is an effective way for robots to achieve positioning and navigation.Because the visual sensor has the characteristics of simple structure,low price,and abundant information,the development of the visual SLAM system framework has attracted the attention of researchers.This paper is based on the fusion of inertial sensor and visual SLAM system as a research topic,and optimizes and improves the shortcomings of the existing system,mainly including the following aspects:In terms of visual data,the current feature matching algorithm based on the feature point method of visual SLAM has low accuracy and speed that cannot meet the real-time scene.An improved ORB-GMS feature matching method based on grid motion statistics is proposed.First,optimize the fast corner detection(FAST)algorithm based on image entropy,and construct an adaptive threshold function to realize the global distribution of feature points;then optimize the grid scoring model to ensure a higher feature matching accuracy rate while reducing feature allocation.Accurate calculation time;Finally,through experimental comparison,it is proved that the improved method has improved speed and performance.In the aspect of inertial data,pre-integration technology is used to realize the synchronization and fusion of visual-inertial information,and the joint initialization of the system is analyzed as the initial value of the system.In view of the problem of feature point method failure or decreased matching accuracy in some scenes,after in-depth understanding of the back-end optimization of the current visual SLAM system,combined with the robustness of the direct method in the feature missing scene,the photometric error weight is used as the state estimation objective function In addition,a more robust state estimation method is proposed based on the mainstream nonlinear optimization method framework.Finally,based on the open source SLAM framework,the open source SLAM framework was partially extended and improved around the improvement methods of this work,and the visual-inertial SLAM system experimental platform was built.In order to verify the effectiveness of the system improvement,the performance indicators of similar algorithms are compared in the public data set and real life scenes.The results show that the improved algorithm has the accuracy of trajectory estimation in scenes such as blurred images,fast motion,and missing visual features.And robustness is better than similar algorithms;in scenes with good visual conditions,it is consistent with similar algorithms in trajectory estimation accuracy and tracking speed.