Research On Some Key Techniques Of Fringe Projection 3D Measurement

Three-dimensional(3D)measurement technology is the key technology in machine vision,and it has important applications in reverse engineering,image recognition,artificial intelligence,Virtual Reality and biomedical aspects.The traditional contact 3D measurement method has the disadvantage of measurement time-consuming and cannot be measured elastic material limitations and so on,so it cannot satisfy the demand of industrial production.Fringe projection three-dimensional measurement technology is the non-contact method,it has the advantages of fast speed and high precision.Therefore,it has become the research hotspot in this field.In this paper,some key technologies in fringe projection are studied and summarized.Such as the 3D reconstruction method based on phase-matching of multi-sensor system,binocular stereo vision technology based on the spherical mode and 3D data registration.The main works of this paper are as follows:1.The basic principle of 3D measurement technology with fringe projection and the research status are presented.The basic methods of PMP(Phase Measurement Profilometry),Gray code,camera imaging model and the processing of 3D point cloud data are described.2.Due to the measurement range limitation of the single sensor,the multi-sensor measurement system has become an essential method in obtaining the complete 3D data of the object.However,for the traditional multi-sensor measurement system,they are separated from each other in the process of system calibration.The entire measurement system calibration has the disadvantage of time-consuming and complex operation,so it cannot satisfy the needs of field calibration.Aiming at the shortcomings,this paper presents a multi-sensor measurement system based on the phase-matching.The calibration of each measurement system is related to each other system.What's more,after all the systems are calibrated,the 3D point cloud data obtained by each measurement system is translated into a unified coordinate system,so it automatically achieves 3D point cloud global registration.3.In order to satisfy the 3D scene measurement needs of the large field of view(FOV),this paper designed two sets of 3D data acquisition system.The first system is the line-structure light system,which is used to obtain the indoor environment 3D data by rotation scanning method.The system can achieve 360°scene measurement,but the measurement speed is slow and low efficiency.The second system takes the advantages of the wide FOV and the large depth of fisheye lens,it achieves the binocular stereo vision of the 3D measurement with large FOV.However,due to the characteristics of fisheye,the 3D reconstruction method based on the pinhole model is not satisfy for fisheye lens.Therefore,this paper proposes a feasible method based on the spherical model to obtain the 3D data by using fisheye lens.The new spherical model enables the world coordinate system directly transferred into the unit spherical coordinate system.In addition,it uses the spherical perspective projection model replaces the plane perspective projection model,so it will no longer be restricted in terms of the FOV.At the same time,aiming at the problem of feature matching in binocular measurement system,this paper proposes a diamond matching algorithm which greatly improves the speed of searching and the precision of measurement.4.After obtaining the 3D data of the measurement object,the obtained 3D point cloud data is also necessary to be processed.This paper presents a 3D point cloud registration algorithm based on two-dimensional(2D)image.First,by calculating the affine transformation matrix between two adjacent 2D images,the rough registration between two adjacent 3D point clouds is completed.Then,the search-area of the corresponding 3D points is determined by the 2D coordinates.Finally,the 3D coordinates of the corresponding 2D coordinates are used for global registration.Through experimental verification,this method greatly improved the speed of registration and completed multi-view 3D point cloud data registration.