| This paper focuses on the problem that mobile robots cannot cruise autonomously due to the loss of BDS(Bei Dou Navigation Satellite System)signals in specific indoor environments,proposed based SLAM(Simultaneous Localization and Mapping)technology,a visual-inertial fusion localization method is implemented using the theory of interactive multi-sensor data fusion.The main research contents of this paper are as follows:(1)Visual information and inertial measurement data acquisition: A visual odometer based on I-LK(Improve LK)is proposed to track the corners of FAST.to solve the real-time problem of mobile robot feature extraction tracking and large displacement movement and mismatching of local feature points caused by dynamic objects;The IMU(Inertial Measurement Unit)manifold pre-integration discrete model is constructed by Inertial orientation and pose calculation.Through four scenes image comparison experiment,it shows that the single frame processing speed of I-LK algorithm can meet the real-time acquisition of visual scene information,Moreover,it maintains excellent tracking stability of object features in dynamic scenes;The IMU random error is analyzed,The prior noise and random walk bias constraints are obtained,to improve the response speed of inertial measurement unit and the accuracy of pose solution.(2)The fusion positioning of visual information and IMU measurement data: A key frame selection strategy based on parallax and tracking quality and an online calibration method for visual-inertial rotation external parameters are proposed;Then the loosely-coupled constraints of visual pose and IMU pre-integration term are established,The precise initial parameters of the visual inertia system to accelerate the convergence of the system are solved by iterative optimization method.Finally,A joint nonlinear optimization model of visual reprojection error,IMU pre-integral residuals and marginal prior residuals was constructed in a keyframe-based sliding window,It realizes the tight coupling of visual-inertial data and accurate positioning function of the system.Through normal,medium,and hard mode scenario initialization tests in the EuRoc dataset,,to verify the success of visual-inertial initialization and data fusion localization.Experiments have shown that under normal conditions,System reaches stable tracking status in 13 seconds.However,as the difficulty of the scene increases,the time for the system to reach the convergence state increases.(3)Experiment and analysis of vision-inertial fusion positioning:Building an experimental platform to test the data set and real environment of the constructed visual and inertial system.Through the comparison of data set positioning performance with excellent open source algorithms,the RMSE of visual and inertial system localization is 0.201 m,which comprehensive performance is better than OKVIS and slightly inferior to VINS-Fusion.In the real scene,the trajectory positioning accuracy of the laboratory is 0.212 m,The accuracy of building floor trajectory positioning is 0.216 m.Experimental results show that the autonomous localization accuracy of visual and inertial system is 0.20-0.22 m in an unknown indoor environment,It can meet the positioning accuracy requirements of existing indoor VI-SLAM. |
