Research On Indoor Autonomous Navigation And Localization Using Monocular Vision And Inertial Sensors

Indoor navigation is one of the hotspots in the field of navigation.GPS positioning technology won't work well in indoor situation,for which the satellite signal being blocked.Therefore,the indoor navigation scheme instead should perform well in all kinds of complex indoor environment.Inertial navigation and visual navigation have good complementarity and autonomy,which is suitable for indoor navigation because the navigation performance is immune to environment factors.The vision inertial integrated navigation system has the advantages of small size,inexpensive cost,simple structure and wide applicability.It can be widely used in mobile robots,drones,unmanned vehicles,augmented reality and other fields.This thesis mainly studies vision inertial integrated navigation system.Based on epipolar geometry,robust estimation and other theories,monocular visual odometry algorithm is designed.Through the analysis of inertial navigation system error model,Kalman filter is built as a fusion approach of image sensor and inertial sensor.The image output result is used to prevent navigation calculation from divergence.Several threshold judgements are proposed to eliminate measurement failures,which improve the system reliability.An initial alignment method of camera coordinate system and body coordinate system through rotation platform experiment is put forward.In final the feasibility and effectiveness of the algorithm is verified by experimentFirst of all,in monocular visual odometry method,select SURF operator as feature detection method considering the indoor application situation.Partition image into a grid before feature detection to avoid feature points concentration.Taking into account the computation cost,we repeat SURF detecting and KLT tracking method to get feature points correspondences between frames.Then,considering the influence of image noise and motion blur,the RANSAC method combined with five-point method is used to eliminate external points and estimate the essential matrix.Finally,the unique solution of pose matrix is extracted from the estimated essential matrix.The pose matrix is regarded as the output of the image sensor,which makes it easier to fuse with the inertial sensor.Secondly,Kalman filter is designed based on inertial navigation.Take error of navigation parameters as research object.Using inertial navigation and other related theory to deduce the relation model between error parameters.Then design Kalman filter using quaternion error,angular velocity error and acceleration error as state variables,and using acceleration,magnetometer,and image sensor for measurement of Kalman filter.On the basis of it,a new Kalman filter was built,which uses velocity and position as state variables and output of image sensor as measurement to prevent navigation parameters from divergence.To solve the problem of image quality,external acceleration and ferromagnetic interference affecting navigation result,several threshold conditions is used to eliminate bad measurement,which improves the system reliability.Finally,the experiment method of initial alignment between camera and inertial sensors is proposed,which can change the camera coordinate to body coordinate.The feasibility and effectiveness of the algorithm are verified by experiments.