| The realization of autonomous navigation of unmanned vehicles requires technical support such as perception,decision-making and control.Environmental perception technology is the premise of autonomous navigation for unmanned vehicles,and positioning technology is the basis of environmental perception technology and one of the core technologies of unmanned vehicle navigation.Since positioning with a single sensor cannot meet the positioning requirements of unmanned vehicles,it has become an inevitable requirement to choose a multi-sensor fusion positioning method.Aiming at the problem of unmanned vehicle positioning,this paper proposes a visual inertial navigation SLAM positioning algorithm based on point-line feature fusion,which aims to solve the difficulty of unmanned vehicle positioning in complex environments.The main research contents of this paper are as follows:A visual odometry based on point-line feature fusion is proposed.In order to obtain a finer corner position in the point feature part,the Shi-Tomasi corner point algorithm is used for rough extraction to obtain pixel-level corner point coordinates.On this basis,the subpixel corner point is used.The point algorithm performs finer extraction to obtain the correct corner positions.The line feature part is to solve the problem of time-consuming line feature extraction.In this paper,the EDLine algorithm is used for line feature extraction and the LBD algorithm is used for line feature description.In view of the problem of wrong matching caused by a large number of short lines in the line features extracted by the EDLine algorithm,it is proposed to use length suppression.The strategy will eliminate short-term.At the same time,the reason of the reprojection error model of the point-line feature in the above process is analyzed,and the LM algorithm is used to optimize the reprojection error.Through the simulation test,the test results show that the point and line feature extraction and tracking algorithm proposed in this paper still has a good feature extraction effect in the weak texture environment,which ensures the accuracy and robustness of the system positioning.A visual-inertial navigation SLAM localization algorithm based on point-line fusion is proposed.Aiming at the problem that the data dimension of the back-end nonlinear optimization module is too large and the error accumulation exists,this paper constructs a new objective function,which integrates the point-line reprojection error and the IMU error,and adopts the sliding window strategy,which effectively reduces the operating time of the system.error and increased operating speed.In order to improve the robustness of the system positioning,the loop closure detection module based on the point-line word bag is used to improve the accuracy and recall rate of the loop closure detection.The loop closure detection algorithm in this paper is verified by the KITTI data set.The test shows that the loop closure detection algorithm in this paper is accurate.The SLAM localization algorithm in this paper is simulated and verified by the EUROC data set,and the test results show that the localization effect of the algorithm in this paper is better than that of the VINS＿Mono algorithm and the PL＿VINS algorithm in complex environments.Finally,an experimental platform is built to further verify the algorithm in this paper in a real scene.The Xiaomi camera and the unmanned vehicle platform are combined into a set of mobile platforms,and the loopback detection test and the positioning trajectory test are carried out in the real scene.The test results show that the algorithm in this paper has higher positioning accuracy and good portability. |
