Handheld 3D Structured Light Measurement System Based On The Digital Speckle

In order to obtain the complete three-dimensional expression,a single 3D measurement system usually needs motion-assistant devices such as the mechanical arm and the turntable in case of the object with large scale,occlusion,or internal and external surfaces.Sometimes a 3D sensor network is employed.These methods are expensive and inflexible.Therefore,it is of great significance to develop a low cost,portable and flexible 3D measurement device that can register data in real time.In this paper,a portable 3D measurement system is developed based on the principles of 3D speckle measurement and point cloud registration in the binocular stereo vision system,which can realize rapid and real-time measurement.The main research contents are summarized as follows:The key techniques of 3D measurement based on speckle structure light are described in this paper,mainly including: speckle correlation,corresponding point search,sub pixel optimization and so on.Firstly,pixel level corresponding points are acquired with the digital correlation principle.Combined with epipolar constraint,parallax constraint,the corresponding point search is decreased from two-dimension to one-dimension,which reduces the computational complexity.Secondly,the sub-pixel location of the corresponding point is optimized.In this paper,N-R(Newton-Raphson)iterative optimization methods based on one order and two orders parallax transformations,and optimization method based on curve fitting are studied.The three algorithms are compared in terms of algorithm performance and reconstruction effect,respectively.Finally,the 3D reconstruction results are given out with the system calibration parameters.This paper also introduces the registration method of 3D point cloud data.In order to achieve real-time point cloud data registration,we study the algorithm structure in Microsoft Kinect Fusion,and focus on the registration algorithm based on Kinect Fusion developed in DIP(Depth Image Processing).The algorithm is different from the traditional ICP(Iterative Closest Point).When registering the corresponding points between the target point cloud and the source point cloud,it will project the source point cloud onto the image plane of the target point cloud,and establish the correspondence of point clouds from the image plane.As for the objective function definition,it follows the definition of the point to surface distance proposed by the predecessors.In addition,the algorithm applies pre-defined spatial cube to TSDF(Truncated Signed Distance Functions)model to render the target point cloud in ICP registration,and implements a point to point registration method,which improves registration accuracy.In the iterative strategy,a multilevel point cloud is established by sampling the original depth data to achieve a coarse to fine iterative method,which improves the registration speed.Based on the framework of the C/C++ and the Qt interface development tool,this paper develops a hand-held three-dimensional measurement system.The system integrates the camera calibration,real-time scanning,point cloud reconstruction,real-time registration and so on.After finishing the scanning,we can use the global registration,point cloud removal,point cloud filtering and other functions to further process the scanning point cloud.The optimized point cloud model is finally generated.