Research On Key Techniques Of Fringe Projection 3D Measurement

Fringe projection three-dimensional(3D)measurement technique has the characteristics of non-contact,high precision,high speed,low cost and full-field measurement.It has been widely used in industrial detection,reverse engineering,cultural heritage preservation,bionic design,human body modeling and medical diagnosis,etc.The key technologies of fringe projection 3D measurement mainly include the non-sinusoidal correction of grating fringes,the calibration of measurement system,the phase calculation and the 3D point cloud registration.This paper mainly focuses on the non-sinusoidal correction of grating fringes and the phase calculation technology.The key issues,such as the phase calculation accuracy,the measurement efficiency,the 3D measurement of special objects and dynamic objects,are studied.(1)Since the non-linear response of the measurement system,the grating fringes captured by the camera do not have good sinusoidal property,which reduces the phase calculation accuracy.Although the double N-step phase-shifting algorithm can reduce phase errors to a great extent,it needs to double the number of projection fringes.For this purpose,two error correction methods are proposed.One is phase error self-correction algorithm.After transforming the original wrapped phase,an additional wrapped phase similar to that in the double N-step phase-shifting algorithm is obtained,and we integrate the both wrapped phases to reduce phase error.The experimental results show that,compared with the traditional phase-shifting algorithm,the measurement error of the proposed algorithm is reduced by 34.2%.Compared with the existing double N-step phase-shifting algorithm,the measurement efficiency is improved by 48.4%.The second is the phase error correction technology based on color fringe projection.This method integrates the original and additional phase-shifted fringes into two channels of a color coded image,separates the two channels of th,e captured image,calculates the phase information separately and fuses them to reduce the error.The second method is also effective in reducing measurement error and improving measurement efficiency.(2)In the fringe projection 3D measurement techniques,the branch-cut algorithm is a path-dependent local algorithm,its branch-cuts search is local optimization,and its branch-cuts is related to the pre-set maximum search radius,but the maximum search radius is uncerrtain.In addition,the branch-cuts are easy to self-enclosed and form"islands" in the region with dense residual points.For this purpose,a branch-cut algorithm with fast search ability for the shortest branch-cuts based on modified genetic algorithm(GA)is proposed.This algorithm combines the nearest neighbor algorithm with improved GA,replaces the path in traveling salesman problem(TSP)with positive and negative residuals matching,and uses selection,crossover and mutation operators similar to GA to optimize the whole situation.It is not easy to appear self-closed branch-cuts,which provides a fast and robust method for solving the shortest branch-cuts problem.(3)The wavelet ridge extraction method by finding the maximum modulus of grating fringe two-dimensional wavelet transform coefficients thas the advantages of simple and fast,but it is susceptible to noise interference.Although the existing two-dimensional wavelet ridge extraction algorithm based on cost function is effective,it does not adjust the weights of the two indexes in cost function,and the extraction of wavelet ridge candidates is redundant,which increases the computation.Therefore,a wavelet ridge extraction method employing a weight optimized cost function in two-dimensional wavelet transform is proposed.It extracts the maximum point,and the local extreme points of 90%of the maximum points from the modulus of two-dimensional wavelet tran,sform to form wavelet ridge candidates.Logistic model is used to adjust the weights of the two indexes in the cost function to obtain the more reasonable cost estimates.This algorithm not only improves the accuracy of wavelet ridge extraction,but also reduces the computation.However,the cost function mentioned above only introduces the gradient of scale factor related to instantaneous frequency information,while the gradient of rotation factor related to orien,tation information of fringe structure is not considered.Therefore,a second algor ithm is proposed.It introduces the gradient of rotation factor into the original cost function,thus an improved algorithm is established.Meanwhile,the Logistic model is used to adjust the weights of the three indexes in the cost function.The experimental results show that,compared with the exiting algorithms,the measurement error using the second algorithm is reduced by 19.03%,but the processing time is increased by 17.4%.(4)There are many objects with a large range of reflectity variations in practice,which is very challenging for any optical measurement method.At present,the intensity of projected fringes or camera exposure are usually adjusted to obtain different brightness fringes image sequence,from which the maximum unsaturated value is selected to form the final fringes image.However,the above methods affect the signal to noise ratio(SNR)of frInge images,in addition,the camera exposure adjustment can not be quantified and the adjustment is very time-consuming.For this reason,a high-speed 3D measurement method for object surface with a large range of reflectivity variations is proposed.Firstly,the intensity of the projected fringes is set to the maximum and does no adj ustment in the proj ection process,which ensures a better SNR of the fringe images.Secondly,the 8-bit gray fringes are transformed to binary fringes to achieve high-speed projection,meanwhile,the projector is defocused to filter out the higher harmonics.Finally,by projecting the color light of the projector itself,the fringe sequence with different brightness can be obtained,and the pixels with the greatest gray value but unsaturated can be selected from a series of fringe image sequences to form the optimal fringe image for three-dimensional reconstruction.This method has the advantages of high-speed measurement,low cost in hardware,high measurement accuracy,simple algorithm and easy to extend.(5)The 3D measurement of dynamic objects is a very challenging task.In view of two specific situations,two solutions are proposed accordingly.One is for objects whose surface changes continuously.When high-speed camera is used for acquisition,the absolute value of the wrapped phase difference between adjacent two frames is far less than ?.When the wrapped phase can not be accurately unwrapped along the x and y coordinates,the phase can be unwrapped along the t coordinate.This method can realize 3D measurement of dynamic objects whose surface does not satisfy the continuity assumption.Two is for the 3D measurement of the rigid,fast moving and discontinuous contour products on the automated conveying line.When the phase-shifting algorithm and the three pitches heterodyne unwrapping algorithm are used for high-precision measurement,there are interframe errors between multiple fringes due to the movement of the objects,which makes it impossible to accurately reconstruct the 3D shape.For this purpose,a new method is proposed.Firstly,the projection fringes are phase shifted,and then the acquisition fringes are pixel shifted,which eliminates the interframe error between fringes images,thus obtaining a good result similar to the accuracy of 3D measurement of static obj ects.(6)The potential application demand of fringe projection 3D measurement technology is huge.In this paper,the following two engineering applications are studied and practiced:sign language learning based on high-speed 3D measurement technology and 3D measurement of small-size object.In addition,3D measurement of high temperature forging are also explored.