Research On Monocular Visual Odometry For Indoor Robot

With the continuous development of the robot domain,the problem of robot navigation and positioning has become a current research hotspot.This paper designs and implements a visual odometry algorithm based on a monocular camera for a mobile robot platform in an indoor scene,and builds an experimental system of it.The main research contents are as follows:(1)The pinhole camera model and camera distortion model are studied,and the relevant model parameters are derived.Aiming at the problem of low accuracy of traditional camera calibration methods,based on Zhang’s calibration method,a camera calibration method based on improved Harris corner detection are studied.Based on the Harris corner algorithm,the extracted corner coordinates are further precise,which provides more accurate data for subsequent calibration.Through experimental comparison,the improved camera calibration method is superior to the traditional camera calibration method.(2)An improved ORB feature extraction algorithm is proposed,which fully considers the image’s own factors to set the extraction threshold,and uses the quadtree algorithm to make the feature points evenly distributed in the image,which is beneficial for subsequent feature matching.Through experimental comparison,the improved algorithm is superior to the traditional algorithm;An optimized feature matching algorithm based on motion smoothness and RANSAC algorithm is proposed.Motion smoothing constraints are introduced for matching,and RANSAC algorithm is used for optimization.Experiments show that the optimized matching algorithm can not only improve the matching quality but also ensure the speed of matching.(3)The motion estimation algorithm based on monocular vision is studied.First,the basic pose estimation algorithm is introduced,and then for the particularity of monocular vision,the monocular initialization method based on environment selection and the improved key frame selection strategy are studied.In order to further improve the efficiency of the algorithm,a motion tracking algorithm based on the combination of reference frame and unified motion model is proposed.Aiming at the error accumulation problem of visual odometer,the graph optimization back-end global optimization algorithm is studied to solve the problem of longterm trajectory drift.(4)A monocular visual odometer experiment system under indoor environment is built,and public data set experiments and natural environment experiments are conducted under this system.In the data set simulation experiment,the indoor office scene data set and the warehouse scene data set are selected for the experiment,the feasibility of the system is verified by comparing with the true value;In the natural environment experiment,linear motion experiment,right-angle turning experiment and closed-loop experiment are carried out.The experiment proved that the system performed better in indoor scenes and could better complete the navigation function in indoor scenes.