SLAM Research On Fusion Of Monocular Camera And IMU

Vision-based Simultaneous Localization and Mapping(SLAM)technology refers to the process of using visual sensors to locate the motion subject and map the environment at the same time.In view of the fact that SLAM(Simultaneous Localization and Mapping)system is vulnerable to environment illumination,texture,and dynamic environment,the problem of initial scale ambiguity still exists in monocular SLAM system.In this paper,the scale ambiguity in monocular SLAM system is effectively solved by the integration of monocular camera and IMU(Inertial Measurement Unit).The fusion of the two sensors significantly improves the robustness of the system and achieves higher positioning accuracy.Based on the VINS-mono system with the best fusion performance of monocular vision and IMU,this paper improves the initialization scheme and increases the initialization of acceleration bias so that it can be applied to low-cost IMU sensors.Referring to ORB-SLAM scheme,this paper applies the front-end based on ORB feature points for VINS-mono,and combines the improved initialization method to achieve high precision real-time location and sparse mapping.Through experiments and analysis in different scenarios in EuRoc dataset,it is proved that the improved VINS-mono has improved the accuracy and robustness compared with the previous one.The specific contents of this paper are as follows:(1)The transformation relation between Lie group and Lie algebra and the theory of optimal estimation in state estimation are studied.(2)The camera imaging process and the non-linear optimization algorithm for IMU information fusion are studied.By parameterizing the two sensors,the parameterized objective function and its Jacobian matrix are derived.(3)Based on current best VINS-mono system,the initialization process is improved,the bias of accelerometer is initialized,and the robustness of system initialization is improved.(4)In VINS-mono system,the Hariis corner + LK optical flow tracking method in the process of image processing thread is changed to ORB feature point + feature matching method,which improves the accuracy of system state estimation.(5)The hardware platform and software system of the experiment are introduced,and the improved monocular VINS system is tested on the open data set EuRoc.The experimental results show that the IMU initialization can restore the scale factor of the system well,solve the scale uncertainty of monocular SLAM,complete the initialization quickly and robustly,and accurately estimate the real-time motion.Sparse mapping,accuracy and robustness have been improved.