| Digital Image Correlation Method (DIC) has been widely used in the engineering field owing to its non-contact, full-field, high-resolution, easy-realized automation and low environment requirement when measuring the displacement field of specimen. But with the demand of material and mechanics research, realizing the real-time measurement of DIC is becoming the urge. However, since calculation amount is too large when processing the data, it is hard to meet the practical demand by using CPU to process the data. Therefore, using GPU programming to improve the processing speed of DIC rapidly becomes the research hot-spot.This thesis, based on GPU programming, does some research on the specific application problem in DIC and gets some achievements in the following aspects:(1) Discussing the effect of various parameters on the efficiency of GPU programming in detailed, which providing the reference to GPU program optimization;(2)Comparing the speed-up ratio between Newton-Raphson algorithm and Inverse Compositional Gauss-Newton algorithm in the GPU and CPU programming mode, the result shows that the speed-up ratios are not the same in different programming mode;(3)Proposing a new way to determinate particles location in speckle pattern. And after being realized by GPU programming, the computation speed is140times faster than serial programming;(4) Proposing a new seed point search method, based on temporal deformation prediction, to solve the experiment situation that the specimen deforms continuously and largely, which works out one of the most difficult problems on evaluating the initial deformation value when realizing DIC by GPU programming;(5) Proposing a new way to solve the sub-pixel displacement. Differing from using spatial information, our fit the integer-pixel using the deformation law along the time axis to get the sub-pixel, the result shows that the new method can greatly improve the calculation speed while keeping the accuracy. |
PDF Full Text Request