Research On SLAM Technology Based On Fusion Of Inertial Navigation And Binocular Vision

With the development of robot and the demand of intelligent surveying and mapping,the technology of navigation and detection and perception of scene information in the unknown environment has been widely researched in recent years,and the traditional of surveying is transform to mobile and real-time dynamic,among which the simultaneous localization and mapping(SLAM)is the hot research topic.Positioning is the key to the autonomous operation of the robot.The inertial navigation system can play a crucial role in navigation and positioning in indoor where GPS signal is limited.However,the continuous accumulation of errors will lead to inaccurate positioning results.Positioning through the camera is easily affected by lighting,texture,the scene moves too fast during the motion and other factors.Therefore,based on the SLAM technology of inertial navigation and vision fusion,this paper studies the navigation and positioning of a mobile robot in indoor.The main research results are as follows:(1)The performance of the fusion system relies on the initial value of the system state.To obtain the mathematical model's proper parameters,the pre-integral model,initialization model,and tightly coupled model of inertial measurement unit(IMU),the multi-sensor calibration are studied.For the inconsistency of the timestamp of the data output collected by the inertial sensor and the vision sensor,the time drift and calibration accuracy of the sensor are studied and analyzed respectively,and the time drift between the sensors is obtained.(2)To solve the problem of different working frequencies between data noise and SLAM sensors,the values of IMU are filtered by Hilbert-Huang transform and pre-integrated,which improves the accuracy and efficiency of the system.Under the condition of hardware synchronization,a joint initialization method is adopted.The initial value calculation models of linearized gyroscope zero-bias estimation,accelerometer zero-bias estimation,gravity optimization,velocity estimation,and IMU data noise are established respectively,and the initial parameters of the system are used to correct the camera pose and optimize the global beam adjustment.Compared with the VINS algorithm,the results show that the convergence speed of the gyroscope and accelerometer is improved about 10s on average,and the fast initialization effect is achieved.(3)For the information redundancy caused by many duplicate data,this paper proposes a keyframe screening strategy based on a sliding window to reduce information redundancy,reduce the loss of computer resources,and ensure the smooth operation of the system.Meantime,to improve the robustness and accuracy of mobile robot positioning and navigation in a complex indoor environment,the tightly coupled method of inertial navigation and visual information data fusion is studied,and a tightly coupled joint optimization model is established with local optimization and global optimization.The optimal pose is obtained,and the effectiveness of which is verified and analyzed.