Research On Visual-inertial Fusion Positioning System For UAV

With the rapid development of UAV technology,the application of UAV in transportation,agriculture,security,logistics and other fields is more and more extensive,and it can be engaged in patrol,inspection,service,photography and other tasks.As an important research direction in UAV field,positioning technology is an important prerequisite for UAV to complete the tasks of motion control and autonomous navigation.Aiming at the problems of poor visual feature information quality and easy missing of yaw Angle observation information of UAV-based visual inertial positioning system in low-texture scenes,this paper studies the visual inertial positioning technology by introducing line features and integrating magneto information to provide an accurate and robust positioning method for UAV.The main research contents are as follows:(1)In order to solve the problem that the quality of the feature points collected by the UAV in the low-texture scene is poor,which leads to the inaccurate pose estimation.Based on the open source system VINS-Fusion,a line feature tracker module is added to the front-end,and the EDLines algorithm is used to quickly extract line features from the image.The coordinates of line features in space are represented by the Plucker coordinate system and the orthogonal representation.The point and line features of two frames are matched and tracked jointly and pose estimation is completed by direct method.Meanwhile,line reprojection residuals are constructed in the back-end.Through iteration,the orthogonal representation of line features is updated to complete the minimization and optimization of pose,so as to achieve a more robust tracking effect and improve the positioning accuracy of the system.(2)In the visual inertial positioning system with line feature,the magnetometer sensor is integrated in the system by means of tight coupling for the lack of global observation information of yaw angle leading to the increase of time accumulation error.Before initialization,the magnetometer is corrected,and the yaw estimated by the system is corrected by using the measured value of the real geomagnetic field,and the global yaw based on the ENU coordinate system is obtained.In the back end of the system,the magnetometer residual is calculated,the pose state quantity is iteratively updated to minimize,the pose quantity is optimized,and the error accumulation of the system output is suppressed.Even in the absence of a loop,the system can still obtain a globally consistent trajectory.(3)Finally,since the magnetometer provides observation information for yaw Angle to achieve the suppression of cumulative errors,the yaw Angle does not need to be repeatedly optimized,so this paper constructs a three-degree-of-freedom displacement diagram optimization model in the system loop to re-optimize the displacement that is also missing observation.First,the key frame information in the back-end is obtained,and all the key frame information is put into a graph model.The displacement of all key frames will form constraints on each other in the graph.When the system detects a loop,all key frames between the loop frame and the current frame will be relocated,and then the displacement increment residual of all key frames will be constructed.Finally,the displacement of all key frames is minimized by iteratively updating,thus achieving a smaller optimization model than the four-degree-offreedom pose map and speeding up the system optimization speed.