Research On 3D Surface Measurement And Benchmark Association Based On Dynamic Stereo Vision

In recent years,with the rapid development of the rail transit industry and manufacturing industry,the demand for dynamic measurement method of freeform surfaces in large-scale building surface and industrial scenes has become more and more urgent.The existing static measurement methods have the disadvantages of low measurement efficiency and limited measurement range,especially the splicing process in the large-scale surface measurement process is more complicated.New measurement methods and theories for complex curved surface measurement are urgently needed in industrial field.In order to solve the above problems,this paper proposes a dynamic push-broom surface measurement model based on binocular line structure light,which uses the push-broom motion of the line structure light to obtain the surface topography,and uses the visual positioning method to recover the motion trajectory.The key theoretical issue is the correlation method between the dynamic measurement reference coordinate system and the global reference in the motion state.In this study,we have studied various theories,including high-precision extraction of structured light strips,global unification of dynamic benchmarks during push broom process and deformation comparison,and field-of-view stitching during large-scale surface measurement.The key contributions of this thesis can be summarized as follows:(1)In the study of dynamic reference transfer process,a camera pose optimization algorithm based on binocular object space error is proposed.The algorithm makes full use of binocular information to calculate the pose information between the associated positions in the orthogonal iterative framework.The algorithm effectively improves the pose calculation accuracy during the process of movement.Furthermore,a global optimization method is proposed to calculate the poses of the camera relative to the global coordinate system at each shooting position.This algorithm effectively reduces the error accumulation and pose drift during visual positioning,an effective solution is provided for the suppression of errors in the benchmark transfer process.(2)In order to compare the surface deformation,a method of establishing the comparison reference using high-precision attitude sensor is proposed.This method associates the global reference coordinate system with the North East coordinate system,calculates the relative rotation matrix between the global reference coordinate systems during the two deformation comparison processes,and obtains the relative displacement between the global reference coordinate systems using a single static reference point.The initial alignment method of the attitude sensor system and the camera coordinate system is given.According to the characteristics of point cloud,a deformation comparison strategy based on minimum distance projection method is proposed.The experimental results show that the root mean square error of the deformation measurement is 0.591 mm and the relative error is 0.109%.(3)In order to improve measurement efficiency,large curved surfaces often require a large measurement field of view.In this study,multiple cameras stitching method is proposed for the field of view expansion.To solve the problem of camera extrinsic parameters calibration of small overlapping field of view between multiple systems in the splicing process,a multi-targets extrinsic parameters calibration method based on Lie algebra framework is proposed.This method does not require special targets and is robust.The point cloud of the structured light stripe obtained by different cameras are spliced using the calculated extrinsic parameters matrix.The experimental results show that the relative error of the stitching is less than 0.12%.