| The robot can move autonomously and premise for the commercial service environment is that it needs to know its location at all times.At present,the mainstream robot positioning methods include visual positioning and inertial positioning.The visual positioning works well in a stable environment,but the robustness is poor in the case of severe motion and poor light,inertial positioning is not affected,but the error of this method will increase with the accumulation of time,this bias is even greater when using low cost inertial devices.Combining these two kinds of sensor information and adopting the method of visual inertial combined navigation to improve the positioning accuracy and reliability of mobile robots has become a research hotspot of robot positioning.Based on this program,the specific research content of this paper is as follows:Firstly,the distortion and noise sources of the camera and IMU are analyzed.At the same time,the camera,IMU and related coordinate system are established,which provides a theoretical basis for the system construction.Secondly,the visual feature extraction and tracking algorithm are analyzed.The traditional LK optical flow method is improved by gradient pyramid.Then the mismatching problem in the tracking process is analyzed.The polar geometry constraint and RANSAC algorithm are respectively used for the mismatched of stereo camera's left and right tracking and back and forth motion tracking,which improves the accuracy of subsequent system observation models.Thirdly,considering the shortcomings of the traditional EKF filter framework,an extended Kalman filter framework based on multi-state constraints is proposed.Firstly,the observation model is improved by the sliding window method,and the accuracy of the observation model is improved.The point position error makes the improved observation model suitable for the filtering framework.This improves the positioning accuracy and robustness of the system.Finally,the paper conducts experimental tests and precision analysis on the constructed visual inertial positioning system through the open reference data set,and verifies the effectiveness of the positioning system proposed in this paper.The paper also analyzes the error distribution of the system from several aspects and compares it with the current mainstream optimization framework VINS.The results show that the designed visual inertial positioning system has better accuracy than the VINS method under the obvious characteristics,and the calculation amount is reduced by 330%,the average error ratio is 0.216%,and the average absolute error is 0.225 meters.The root mean square error of the absolute trajectory error is 0.249 meters on average.The test results in the real environment also show that the system fully meets the positioning requirements of indoor mobile robots and has a good engineering application prospect. |
