Research On Technique Of 3D Registration Based On Static Datum Conversion

In the large-scale three-dimensional(3-D)topography measurement,most of the 3-D measuring equipment cannot ensure the measuring accuracy together with the breadth.In order to obtain measured data of high accuracy,we need to measure the object with a short distance and small field of view.Moreover,one single measurement can only get a limited range of measurement data,which cannot meet the measurement needs.For instance,in practical production process,size inspection of the machining parts is needed,whose sizes range from few meters to tens of meters.Due to the excessively large measuring distance or occlusion of the situation,single measurement cannot get the desired data.Therefore,in order to obtain precise measured data covering the entire large scene,it is necessary to adjust the position and posture of the stereo measuring device and make several local measurements on the scene.However,the measured data obtained in different positions,or different postures are clearly not in the same coordinate system,which cannot be directly applied to data stitching to get the overall morphology.So,coordinate transformation is necessary for these independent measuring system coordinates to unify them into one coordinate system.Focusing on the coordinate conversion problem mention above,this paper proposes a 3-D stitching technology based on the static reference conversion,and some key technologies are studied deeply.This paper mainly includes the following aspects:(1)This paper introduced the static reference conversion system and its basic units,including local measurement unit,posture translation unit and static reference unit.This paper illustrated the component and function of each basic units in detail,and the method to obtain three-dimensional data and coordinate conversion principle were explained as well.Finally,the design scheme of static reference conversion system was given for several typical three-dimensional topography measurement scenes,and the calculation of the distance between two adjacent units is studied.(2)The key techniques of image feature point extraction and pose estimation were discussed in this paper.The central perspective projection model and aberration model were studied in detail.The sub-pixel corner point localization simulation results showed that the corner pixel extraction error is less than 0.01 pixel,which meets the high-precision requirement.Simulation experiments are carried out for several commonly used algorithms.The simulation results showed that the algorithms of RPn P and SP + LHM is of high accuracy and robust.(3)The camera calibration technology was studied in this paper,including the calibration of camera instinct parameters based on coplanar control points,calibration of camera fixed relationship,and camera stereo calibration.In this paper,the re-projection error of the internal calibration is about 0.1 pixel.Two feasible solutions for calibration of camera fixed relationship were proposed,full-field calibration and visual calibration.The camera calibration technique provides the solution of some invariable parameters in the process of camera serial-parallel network measurement.(4)The 3-D stitching experiments based on static reference conversion system were conducted in this paper.The results showed that,the stitching error for the measured object of 1 to 10 meters ranges from 1cm to 10 cm.The static reference conversion system proposed in this paper is of good flexibility and can adapt to different measuring scenes.The cost is relatively low,and it does not need expensive precision instruments.The system proposed in this paper avoids the generation of accumulated errors and the problem of surface occlusion and provides a new solution for the stitching of the measured data in the application of large-scale three-dimensional topography measurement.