Research On Mobile Robot Positioning And Mapping Base On Vision And Inertial Guidance Fusion

In recent years,indoor mobile robots have become more and more important in people’s lives.(Simultaneous Localization and Mapping,SLAM)is the basis and core of mobile robots to perform positioning and navigation tasks.Map construction and positioning can be done with a single vision sensor,but image features are easily lost in the case of weak texture or fast motion.Inertial Measurement Unit(IMU),with high sampling frequency,can make up for the problem that visual sensor image features are easy to track and lose,and improve the positioning accuracy and robustness of the system.In this paper,the SLAM technology of mobile robot integrating vision and inertial guidance sensors in indoor environment is studied,and the main research contents are as follows:Firstly,based on the theory of visual inertial guidance of mobile robots,the coordinate system and its transformation relationship of the system are established,and the camera model,IMU model and octree-map model are established,which lays a theoretical foundation for subsequent positioning and mapping research.Secondly,the visual odometer of mobile robot is studied,and a front-end visual odometry scheme based on depth camera is designed for the problems of feature point redundancy and reduction of matching rate of traditional ORB feature points,including ORB feature point adaptive threshold detection,homogenization extraction,grid motion statistical rejection of false matching between frames and pose estimation algorithms of EPNP and PL-ICP,and experimental comparison with the original scheme verifies the effectiveness of feature point extraction and matching and front-end visual odometer.All error values have decreased.Then,on the basis of visual odometry,the inertial guidance information is integrated in a tight coupling manner,the IMU pre-integration formula is derived,and the joint initialization scheme of the visual inertial guidance system is designed,the objective function of vision and IMU residuals is constructed,which is used to estimate the optimal pose,the sliding window and key frame are used to reduce the amount of calculation,and the constraint relationship between the old frame and the observation frame is established by using the marginalization,and the loopback detection of the bag-of-word model is introduced to optimize the trajectory drift problem.Finally,an experimental platform for the visual inertial guidance system of mobile robots is built,and the error comparison between the proposed system and other mainstream open source SLAM algorithms is tested in the public data set and experimental scenarios,and the experimental results verify that the proposed system can obtain better robustness and accuracy,and an octree-tree map is established to provide a navigation map for path planning for mobile robots.