Dynamic precision surveying of railway tunnels based on MLS with robust constraints of track control network

Abstract

The inertial navigation system (INS)/odometer integrated dead reckoning (DR) is usually used for the positioning and orientation of mobile LiDAR system in the precision surveying of railway tunnels. However, due to the accumulation of pose errors and the creaky method of joint surveying of track control points, it is difficult to achieve the millimeter level absolute accuracy of point clouds relative to the track control network under dynamic conditions, especially when there are gross errors of some control points coordinates. This paper puts forward an algorithm to optimize accuracy of INS/odometer integrated DR with robust constraints of track control network. The errors of DR are divided into two types, which affect the relative accuracy and the absolute accuracy respectively; The concept of virtual mileage coordinate system is proposed, in this coordinate system, the influence of gyro bias error on position is modeled as a circular distribution and estimated indirectly by using the high-precision relative spatial relationship of control points; An absolute accuracy optimization method with constraints by sequence match between positions of surveying equipment and track control points is proposed, and the gross errors of control point coordinates are detected and processed based on reduced weight iteration method. The experimental results of North China high speed railway tunnel show that when the distance interval of control points of correction is 120 m, the root mean square (RMS) error of coordinates in X, Y and Z directions are 0.003, 0.003 and 0.005 m respectively. When the distance interval of control points of correction is 420 m, the RMS in X, Y and Z directions are 0.029, 0.030 and 0.010 m respectively. When there are gross errors of some control point coordinates and the distance interval of control points of correction is 120 m, the RMS of coordinates in X, Y and Z directions are 0.005, 0.006 and 0.006 m respectively, and the gross errors of 4 cm in one direction of the three-dimensional control point coordinates can be detected. The experimental results verified the high accuracy of the algorithm and its robustness to the gross errors of control point coordinates.