Research On Visual-inertial Tight Coupling Positioning Method Based On Point And Line Feature

As the development of information technology and sensor technology,mobile robots have been widely used in human production and daily life.Simultaneous Localization and Mapping(SLAM)is a key technology for achieving autonomous robot localization.However,conventional visual SLAM systems based on point features have difficulty extracting enough feature points in scenes with low light or sparse texture,leading the failures of localization.Considering the favorable performance of line features in such senses,the aim of this study is to specialize point-line features of space and to design a visual-inertial tightly-coupled localization system,with the combination of inertial measurement units(IMU).The main research content of this study is as follows:(1)To address the problem that visual SLAM is unstable in extracting environmental feature information in scenes with low light or sparse texture,the ORB(Oriented Fast and Rotated BRIEF)feature point extraction method is improved by dividing the image grid and dynamically adjusting the feature point extraction threshold based on the image grayscale information.The LSD line feature is introduced,and a scheme for removing short line segments and merging adjacent line segments is designed.The improvement of RANSAC algorithm is conducted by the careful selection of matching pairs with initial input data,i.e.,smaller hamming distances.The verification of improved algorithms has been executed by the comparison of conventional algorithms.(2)Build a visual inertial tight coupling system.Firstly,a loose coupling scheme is designed to realize the joint initialization of vision and inertia,including using SFM method to initialize the pure vision system separately,calculating the external parameter data of camera and IMU,calculating the bias data of gyroscope.Then,the visual inertial tightly coupled positioning scheme is designed.The sliding window is used to limit the amount of system data,and the marginalization method is used to retain the constraint relationship between the eliminated data frames.The visual bag of words model containing point and line features is established and improved to realize loop closure detection.(3)The design of simulation experiments is dependent on the Eu Ro C open-source dataset;then,the accuracy of the developed algorithm in this study is compared with conventional opensource algorithms.Besides,the developed algorithm has been transplanted to the laboratory robot platform for conducting the tests in indoor scenes.Consequently,a better accuracy of localization has been presented in the developed algorithm,rather than that in the conventional algorithm,and it has good practicability in dim light and sparse texture scenes.