| Because 3D measurement technology can provide effective basic data support for material forming performance evaluation,forming process improvement,forming accuracy evaluation,etc.,it is now an important development direction in the field of material processing.Among them,digital speckle projection three-dimensional measurement is one of the more advanced three-dimensional measurement methods,which belongs to the single-frame surface structure three-dimensional measurement method.It only needs to project a single coded image during the measurement process,which is easy to encode and decode,and the measurement process is flexible and convenient,so it has great advantages in the field of dynamic measurement and handheld measurement.However,limited by the accuracy of the matching algorithm,it is often impossible to take both high-precision and high-efficiency measurements into consideration.Based on this,this paper studies a fast and high-precision 3D measurement method based on digital speckle projection.This method uses a single speckle pattern projection to increase the surface texture of the measured object,builds a dense and highprecision parallax acquisition algorithm based on digital image correlation technology,and performs parallel acceleration optimization on the GPU platform.The specific research content is as follows:The principle of the digital speckle 3D measurement method is explained in detail,including the principle of camera imaging and the principle of binocular stereo vision.The model of binocular stereo vision,the epipolar geometry,and the principle of disparity mapdepth map mapping are introduced.The process from image acquisition to restoration of the three-dimensional shape of the measured object is explained in detail.Pointed out the importance of the disparity map acquisition algorithm in the measurement process.In order to solve the problem of low measurement accuracy of digital speckle projection3 D measurement method due to the low accuracy of the disparity map acquisition algorithm,a high-precision disparity map acquisition method is proposed-semi-global stereo Matching algorithm based on Inverse Compositional Gauss-Newton iteration(ICGN-SGM).The algorithm uses the Census-based semi-global stereo matching algorithm to obtain the entire pixel disparity value.Combining left and right consistency checking,coupling uniqueness checking,and connected region method to achieve mismatch removal of disparity values.And add a classic digital image correlation(DIC)domain algorithm-reverse synthesis Gauss Newton iterative algorithm to optimize the initial matching disparity results at the sub-pixel level.The experimental results show that the absolute error of the standard sphere center distance reconstructed based on the ICGN-SGM algorithm is 0.057mm,and the relative error is 0.038%.In order to effectively improve the calculation efficiency of the above-mentioned highprecision dense disparity value algorithm,an efficient parallel computing architecture CUDA is introduced on the basis of the original algorithm to perform parallel optimization and acceleration.Taking full advantage of CUDA’s multi-core and high parallel performance,a high-precision,high-efficiency and dense disparity value parallel algorithm that can be applied to mainstream chips is designed.The experimental results show that the operating efficiency of the disparity algorithm under the GPU platform is 250 times higher than that on the CPU platform.Realize the disparity value data output of 5 frames per second for 1.2 million pixel images.High-efficiency three-dimensional calculations up to 6000 points per millisecond.In addition,the absolute error of the reconstructed standard sphere center distance is less than0.06 mm,and the relative error is less than 0.04%.The algorithm takes into account both efficiency and accuracy.Based on the above research,combining the dense disparity value acquisition algorithm with the digital speckle projection method can obtain a digital speckle projection threedimensional measurement method that can take into account both high precision and high efficiency.The method has a measurement resolution of 1296×966,a measurement efficiency of 5 frames per second,and a measurement accuracy of 0.06 mm.It has huge application potential in dynamic measurement and handheld measurement. |