Research On Stereo Visual-inertial Odometry Based On Structural Lines And Collinearity

Simultaneous localization and mapping(SLAM)is an indispensable core technology in mobile robot research and application.Vision SLAM and visual inertial odometry(VIO),which take the vision measurement collected by camera as the main input,are becoming the current research hotspot.At present,most popular visual SLAM/VIO systems usually use point features or general line features as landmarks to establish visual measurement constraints,and rarely consider the structural geometric characteristics of the scene and the correlation between point and line features.Based on this,this paper proposes a stereo visual inertial odometer system based on structural line features and collinearity between point and line features.The use of structural lines not only provides more visual measurement constraints for the system,but also introduces the orientation information of the scene into the system,thereby eliminating the accumulated orientation errors during positioning and mapping.The use of collinearity between point and line features allows the system to make full use of the prior geometric information in the environment,thereby further improving the accuracy of the system's positioning and mapping.This article first introduces the research background and development status of visual SLAM/VIO,and then also introduces the relevant background knowledge of general VIO systems,and then introduces in detail the structured line features proposed in this article and considers the collinearity between point and line features.The theoretical model and concrete realization of the binocular VIO system.In order to introduce structured line features into the VIO framework based on sliding window optimization,this paper defines the parameterization of the 2-parameter structure line and its visual measurement constraint model under the assumption of the Manhattan world model.In addition,for the point and line features recognized as collinear in the reconstructed map,a spatial distance constraint is established between them,so that the geometric characteristics of the reconstructed map are more consistent with the real situation.Finally,after testing on public datasets such as Euroc and Tum and physical experiments in real environments,the results show that the proposed stereo VIO system based on structural lines and collinearity can achieve higher positioning accuracy and better mapping than some popular VIO methods.