Aiming at the problem that pure visual SLAM is prone to feature tracking failure and positioning accuracy degradation in in-door scenes with obvious illumination changes, less environmental texture and rapid carrier motion, this paper proposes a visual iner-tial tight coupling method based on sliding window back-end optimization, which integrates IMU information to improve tracking accu-racy and system robustness. This method uses IMU pre-integration error and monocular vision SLAM re-projection error to construct a new loss function for state estimation, and uses a nonlinear optimization method based on sliding window for motion estimation to recov-er the pose of the combined system in real time. The experimental results on the measured data set show that the root mean square er-ror of the proposed method in the x-axis direction is 0.124 m, and the root mean square error in the y-axis direction is 0.113 m, which achieves centimeter-level positioning accuracy.
vision/inertialSimultaneous Localization and Mappingsliding windowtightly coupled
维普
万方数据
