Research On Multi-sensor Integrated Navigation Algorithm Based On Monocular Vision

Navigation and positioning technology is a crucial issue in robotics and driverless fields.However,the robustness applied to indoor and outdoor highly dynamic and unstructured scenes is not very good so far.This thesis proposes and designs a multi-sensor integrated navigation algorithm based on monocular vision.It will integrate ORB_SLAM2 monocular vision and INS,compared with the traditional method to IMU and visual data separated filtering,we designed a general nonlinear optimization framework could bring visual feature points and the noise of the IMU data into it to optimize the camera position.Not only could scale and sensor bias be accurately estimated,could also improve the robustness and position precision of the algorithm.When the GPS signal is better,the algorithm combines the GPS data with the kalman filter,which could be used to estimate the position of the GPS data more accurately,and can also assist the construction of visual map.The main work of the thesis includes:(1)This thesis proposes a Visual/INS integrated navigation algorithm based on nonlinear tight coupling.Firstly,the key technologies of monocular vision based on keyframe are studied,including automatic initialization based on model selection and keyframe selection strategy.Then,the pre-integration model of IMU and the initialization of VINS are derived.In addition,the visual optimization based on nonlinear optimization Inertial fusion scheme is studied.Finally,the algorithm is validated on the Euroc dataset and the actual scene to verify its performance.(2)Proposed and designed Visual/GPS/INS integrated navigation algorithm.Firstly,the GPS and INS data are fused through the kalman filter design.Secondly,the GPS/INS fusion algorithm is used to study the monocular vision scale estimation.In addition,the GPS/INS fusion algorithm is used to study the pose estimation of the VINS.Finally,The vehicle test verify the algorithm,which proves that the algorithm has high positioning accuracy and robustness,the error is less than 0.2 meters and has strong practical value.