Research On The Technology Of Visual-Inertial Odometry

SLAM(Simultaneous Localization and Mapping)is divided into front-end and back-end.The front end is also called visual odometry.When the motion is too fast,the image obtained by the visual odometry will be blurred or the overlapping area between the two frames is too small to perform feature matching,so the visual odometry is difficult to deal with fast motion.In addition,when the image content changes,the visual odometry cannot essentially judge whether the camera has moved or the external environment has changed.As an important inertial sensor,the inertial measurement unit not only has the advantages of small size,light weight and high cost performance,but also has a very high sampling frequency.It can provide good motion estimation for short-time fast motion.Therefore,using IMU can not only assist image feature matching,improve the accuracy of feature point matching,but also provide motion estimation for the system in the time when image features are lost.In addition,according to the measurement data of IMU and the stationary detection algorithm,it can be judged whether the camera is moving or not,which provides a solution to the problem of motion detection.Therefore,compared with the visual odometry,visual-inertial odometry can deal with more complex scenes and has higher robustness.The research contents of this paper are as follows:Firstly,in order to improve the accuracy of inertial navigation,the error source of the inertial measurement unit is analyzed and its error model is determined.A four-position calibration method is proposed and derived.Secondly,the related theory of image feature matching is introduced,and the method of inertial navigation aided image feature matching is proposed and deduced.Then the stationary detection algorithm is introduced,and the method of inertial navigation assisted dynamic object detection is proposed.Finally,an experimental device is built to verify the correctness of the theory.The experimental results show that the four-position calibration method can complete the calibration of the IMU quickly and accurately.The inertial navigation can not only assist image feature matching,improve the accuracy of image feature matching,but also assist the camera to detect dynamic objects.