Research On Indoor Positioning Technology Of Mobile Robot Based On Fusion Of Binocular Vision And Inertial Navigation

With the development of mobile robots and autonomous driving,simultaneous localization and mapping(Simultaneous Localization and Mapping,SLAM)technology has attracted much attention.Compared with laser SLAM,visual SLAM has the characteristics of rich collection of information,light weight and low price.In recent years,visual SLAM technology has become a research hotspot.A visual SLAM system fused with a vision sensor and an inertial measurement unit(IMU)is one of the solutions to improve the positioning accuracy of the SLAM system.The inertial measurement unit has the characteristics of small sensor size,high data measurement frequency and high instantaneous measurement accuracy.Fusion of visual data and IMU data can use IMU's high-precision measurement data to make up for less robust visual data.At the same time,the relative robustness of visual data can be used to suppress the cumulative drift error of IMU data,making the system more complex.It has stronger robustness in the environment.This paper proposes a visual SLAM system that integrates binocular vision and inertial navigation.The main research work is as follows:1.Constructed a visual odometer combining feature method and direct method to improve the efficiency of motion tracking.First,the optical flow method is used to track the image feature points extracted from the image frames,and then the initial pose is calculated using the matching relationship of the feature points between the image frames.Then,the matching relationship between the map points and the feature points in the image frame is obtained through projection transformation,and the posture is optimized by the reprojection error.While using the direct method to improve the tracking efficiency of the system,the characteristic method is used to reduce the cumulative error caused by the direct method,thereby improving the robustness of the system.The experimental results show that although the visual odometer based on the fusion feature method and the direct method proposed in this paper has a slightly lower positioning accuracy than ORB-SLAM2 based on the feature method,the operating efficiency has been significantly improved compared with ORB-SLAM2.2.Construct a visual SLAM system that combines binocular vision and inertial navigation on the basis of visual odometry that combines feature method and direct method.Construct an error model of IMU data,use visual information for joint initialization of visual inertial navigation to estimate the gyroscope deviation and acceleration deviation in the IMU,and estimate the motion speed.At the same time,add IMU constraints to the back-end optimization method based on sliding window to further improve the camera The estimation accuracy of the pose.The experimental results show that the SLAM system fused with binocular vision and inertial navigation proposed in this paper can run well in the Eu Roc data set,and it is also better than the open source visual-inertial fusion OKVIS in terms of absolute trajectory error.