Stereo Visual-inertial Odometry With Point And Line Features

The po int feature is widely used in feat ure-based visual simultaneous localization and mapping(V-SLAM)or visual odometry(VO)systems,whereas the line feature is rarely explored in these systems.The main reason for this is that the line feature is a kind of high-dimensio nal feature,making it more difficult for a line feature to be extracted,described and parameterized.However,the line feature contains more structural informat ion about environments than the point feature,these structural informat ion could improve the robustness and trajectory accuracy of a V-SLAM or VO system.Furthermore,in low-textured,light-varying and motion-blur scenarios or man-made environments where po int features are not sufficient or evenly distributed,line features are usually abundant and can be reliably extracted.As a result,line features could be the do minant features for pose est imat ion.In addit ion,the image represents the informat ion of the environments,while the inert ial measurement unit(IMU)measures the motion informat ion o f itself like the 3-axis angular velocit y and linear accelerat ion velocit y.By fusing the two kinds o f sensors,more constraints for the mot ion can be applied to improve the performance o f a system.Taking the above-ment ioned benefit s into account,in this paper,a stereo visualinert ial odo metry using po int and line features is designed.The implemented system is able to work robustly and accurately in co mplex environments and achieve real-t ime performance.In summary,the main contribut ions of this paper are as fo llows.1.Some careful designs for extract ing and describing line featues as well asparameterizing a line feature.Besides,this paper proposes the uncertainty modelof a spat ial line and designs a novel line posit ion refining algorithm according tothe uncertaint y model.These techniques speed the implemented system andimprove the accuracy of the reconstructed local map o f the environment.2.By fusing visual and inertial sensors,a mult i-sensor-fusion localizat ionframework based on t ightly-coupled optimizatio n is constructed.The frameworkcould improve the accuracy o f a system and work only wit h visual measurements.3.The inclusion o f line features improves the accuracy o f a system.In so meco mplex environments,the system designed in this paper is able to acquire amore accurate trajectory co mpared with the system relying on po int features.