Stereo Visual Inertial Odometry Using Point And Line Features And Its Application

Visual odometry(VO)is a technique which estimate the pose of mobile platform only using camera.It has gradually become an important way for mobile robot ego-motion estimation because of its low cost and wide range of applications.The traditional point-based VO algorithm is limited by point features,and it is difficult to provide reliable positioning results in the absence of texture,illumination changes or dynamic objects.Considering that there are many line features in man-made constructed environment,and inertial measurement units(IMUs)and cameras have complementary natures.The visual odometry using line features and visual inertial odometry(VIO)fusing inertial information have gradually become research hotspots in recent years.In order to,improve the positioning ability of VO in complex environments,in this thesis,we add line features into the traditional point-feature-based VO algorithm and fuse it with IMU measurement information.A stereo VIO algorithm based on point and line features which is called PLSVIO algorithm is proposed.The main works include:1)The parameterization methods of spatial straight line and its camera projection model are discussed.Two different representations are used to parameterize the spatial straight line.One is Plucker coordinates for coordinate transformation and camera projection,and the other is orthonormal representation for parameter updating of 3D line in the states optimization.2)A visual odometry based on point and line features is designed.In order to improve the accuracy and efficiency in the data association,the stereo match and match-by-projection methods for line features are designed.According to the definition of reprojection error of line features and the orthogonal representations of spatial lines,the Jacobian matrix of reprojection error is derived.3)The visual observations and IMU measurements are fused in a tightly-coupled nonlinear optimization framework.In order to avoid repeated integration of IMU measurements in the back-end optimization,the pre-integration algorithm is used to pre-integrate the IMU measurements.After the IMU bias update,the pre-integrated measurements are corrected by a first-order approximation.In addition,the loosely-coupled stereo-IMU initialization procedure is designed to estimate the initial state of VIO.Finally,the proposed algorithm is tested by not only the EuRoC datasets but also real-world experiments.The results show that the proposed algorithm is able to estimate camera pose with high accuracy and robustness.