Monocular robustness visual-inertial tight coupling localization method

Abstract

The invention discloses a monocular robustness visual-inertial tight coupling localization method which comprises the steps of: acquiring visual data by a camera, and acquiring inertial data by an IMU(Inertial Measurement Unit); carrying out IMU pre-integration and obtaining an IMU priori value; substituting the IMU priori value into a visual-inertial combined initialization model, and completinginitialization of a parameter; and in time required for initializing the parameter, calculating motion information by utilizing a transformation matrix among continuous key frames, and substituting the motion information into a rear-end module of a visual-inertial fused localization framework so as to implement tight coupling localization. After initialization of the parameter is completed, the parameter is substituted into the visual-inertial fused localization framework, and the motion information is calculated and is substituted into the rear-end module of the visual-inertial fused localization framework so as to implement tight coupling localization. By using the method disclosed by the invention, initialization time can be shortened to be within 10 seconds, and compared to localization accuracy of a conventional ORB_SLAM2 (ORB_ Simultaneous Localization and Mapping)-based monocular visual-inertial localization system, localization accuracy of the monocular robustness visual-inertial tight coupling localization method can be improved by about 30%.

Description

technical field [0001] The invention relates to a monocular robust visual-inertial tightly coupled positioning method, which belongs to the field of SLAM (Simultaneous Localization and Mapping, simultaneous positioning and mapping). Background technique [0002] With the rapid development of technologies such as autonomous flight, automatic driving, virtual reality and augmented reality of micro air vehicles, the realization of high-precision and robust positioning is an important prerequisite for completing the established tasks of autonomous navigation of mobile agents and exploration of unknown areas. Fusion of visual sensor and inertial measurement unit (IMU) can construct a visual-inertial fusion positioning system (VIO) with higher accuracy and stronger robustness. [0003] The traditional visual-inertial fusion positioning framework includes two modules, the front-end and the back-end. The front-end estimates the camera motion between adjacent images through the IMU a...

AI Technical Summary

Problems solved by technology

The OKVIS framework (Leutenegger S, Furgale P, Rabaud V, et al. Keyframe-based visual-inertialslam using nonlinear optimization [J]. Proceedings of Robotis Science and Systems (RSS) 2013, 2013.) proposed the front end of the visual-inertial fusion positioning framework model, which realizes the tight coupling of visual and inertial data, but does not recover data such as system scale and gravitational acceleration, and does not contain a pre-integration framework, so its positioning accuracy and robustness are poor; based on ORB_SLAM2 (Mur-Artal R,Tardós J D.Visual-inertialmonocular SLAM with map reuse[J].IEEE Robotics and Automation Letters,2017,2(2):796-803.) Monocular visual-inertial positioning system, its pre-integration model uses flow pattern The pre-integration algorithm takes about 15 seconds to initialize the visual inertial model. At this stage, real-time robust positioning cannot be achieved. At the same time, the tracking model uses a simple uniform speed model, so that the camera cannot be initialized correctly when the camera moves violently, and even tracking fails, which eventually leads to positioning. Ineffective; VINs (Qin T, Li P, Shen S. Vins-mono: A robust and versatile monocular visual-inertial state estimator [J]. IEEE Transactions on Robotics, 2018, 34 (4): 1004-1020.) initialization The model is relatively fast, but the accelerometer bias is not calibrated, and the initialization accuracy is slightly low, and the traditional reference frame tracking model limits its positioning accuracy

Images