Simultaneous Localization And Mapping Of Mobile Robot Based On Multi-Sensor Fusion

Mobile robot positioning technology is the basis and key technology for robots to carry out intelligent autonomous movement,Simultaneous Localization and Mapping(SLAM)is the mainstream method for solving robot positioning problems.Among them,the monocular vision SLAM has the advantages of low sensor cost and simple structure.However,there are still some shortcomings:when the camera is moving or rotating,the accuracy of the positioning system will be reduced,even tracking failure;monocular camera can only achieve relative scale estimation,and can't obtain absolute scale information;trajectory drift is easy to occur when it runs for a long time.Aiming at these problems,this paper studied a SLAM system that combines inertial navigation and visual information.The positioning data of the inertial navigation was used to constrain the visual data to improve the accuracy and robustness of the positioning.The specific research contents are as follows:Firstly,the basic theory of visual inertial fusion SLAM was studied.Including the basic theory of positioning of inertial navigation systems and the basic theory of monocular vision SLAM.In the inertial navigation technology,the common coordinate system of inertial navigation was introduced.The advantages and disadvantages of three attitude characterization methods of INS and the conversion between the three methods were analyzed.The attitude calculation and dead reckoning formulas of SINS based on quaternion were derived;In the basic theory of monocular visual SLAM,the framework of SLAM is introduced,and the ORBSLAM based on indirect method is focused on,and the roles and mutual relations of the three threads are analyzed.It laid the theoretical foundation for the next step of research.Aiming at the real-time and invariance requirements of feature point detection in visual SLAM algorithm,several feature detection algorithms are analyzed in this paper,and a comparison experiment of detection performance is designed.After analysis,the stability and real-time performance of the FAST corner+FREAK descriptor in the camera motion state is better than other algorithms,so it is chosen as the feature detection algorithm in this paper.On this basis,in order to improve the detection speed of the algorithm and improve the uniformity of the detection results in the image distribution,this paper designs a feature selection strategy based on quadtree,so that the SLAM system can make full use of the acquired image features.For the vision-only SLAM algorithm,the tracking loss is easy to occur when the camera is moving fast or rotating.This paper proposes a visual inertial multi-sensor tightly coupled SLAM algorithm based on ORBSLAM,improve the three threads of ORBSLAM to program the algorithm:firstly,aiming at the output frequency of the inertial navigation data is much larger than the visual sensor,the INS data is pre-integrated in the measurement pre-processing stage to align with the video frame.Secondly,In order to obtain a good initial value,a calculation formula for the tight coupling of visual inertia is derived,which is used to calibrate the initial offset,scale factor,gravity vector and velocity of the inertial navigation;Then improved the three threads of ORB-SLAM,and add the result of INS pre-integration as a constraint to nonlinear optimization to realize INS and visual data tight coupling.And introduced a marginalization strategy to keep the optimization matrix sparse to keeps the optimization matrix sparse,which reduces the computational complexity.Finally,in order to evaluate the rationality of the visual inertial navigation system,experiments were carried out using public data sets and actual environments.The experiment selected the pure visual typical algorithm ORBSLAM and VINS-mono which is latest research results of visual inertial fusion SLAM algorithm as a comparison target.Data comparison shows that the algorithm proposed in this paper has better robustness compared with ORBSLAM.It is hard to lose in rotating or fast motion.Compared with VINS-mono,the speed is slightly insufficient but with better accuracy.Comprehensive full-text work,this method has good rationality and accuracy,and can be widely applied to the positioning and navigation of ground intelligent robot systems.