Research On GPU-based Two-dimensional Ultrasonic Speckle Motion Tracking Algorithm

The method of motion/displacement estimation of biological tissues plays a pivotal role in ultrasound elastography,which makes it a hot topic in elastography.Among them,normalized cross-correlation(NCC)has been proved to be one of the best algorithms for motion estimation.However,its disadvantages are also very significant.The computational cost of the algorithm is high,especially when it is applied to 2D or 3D scenes,obvious.To solve this problem,many experts and scholars have proposed their own solutions,one of which is based on the cross-correlation algorithm with the table,which reduces the computational cost without increasing the motion estimation error and sacrificing spatial resolution.Based on the existing research of cross-correlation algorithms based on tables and tables,this paper has conducted more in-depth research and systematically analyzed the traditional cross-correlation algorithm and the table-based cross-correlation algorithm under CPU and GPU conditions in two-dimensional scenes.Under the performance.In this paper,the application of traditional cross-correlation algorithm and table-based correlation algorithm in two-dimensional scene is first implemented,and the GPU parallelization of traditional two-dimensional cross-correlation algorithm and table-based two-dimensional cross-correlation algorithm is completed based on CUDA platform..Then through computer simulation experiments,the computational efficiency and motion estimation accuracy of the two algorithms under the conditions of CPU and GPU hardware acceleration are compared.The results show that under complete CPU conditions,the 2-D cross-correlation algorithm based on the table can obtain 5 times faster than the traditional 2-D cross-correlation algorithm.In the case of the GPU,the acceleration effect is affected by the cross-correlation window size.And the search range has a great influence,but in the motion estimation accuracy,both algorithms can obtain high-quality lateral and axial displacement maps.Finally,in order to further improve the accuracy of motion estimation,GPU texture interpolation technology is used to interpolate the matrix of cross-correlation matrix obtained by traditional two-dimensional cross-correlation algorithm.The experimental results show that the bilinear interpolation algorithm has higher computational efficiency and three times of B-samples.The interpolation algorithm can significantly improve the imaging quality.To sum up,the specific work of this paper is as follows:1.Describe the implementation details and characteristics of these two algorithms in two-dimensional scenes.Use computer simulation experiments to analyze the computation performance of the two algorithms on the CPU side in two-dimensional scenes.Compare the two techniques under the premise of obtaining high-quality elasticity imaging maps.Calculation efficiency and motion estimation accuracy.2.Analyze the possibility of computationally intensive parallel computing of these two algorithms,and implement the parallel acceleration of the traditional two-dimensional cross-correlation algorithm and the two-dimensional cross-correlation algorithm based on the table through the CUDA platform.Finally,through the computer simulation experiment,the performance of the two algorithms in the GPU side and the CPU side performance analysis were comprehensively analyzed based on the calculation efficiency and motion estimation accuracy.3.Using the GPU-based two-line interpolation algorithm and cubic B-spline interpolation algorithm to interpolate the cross-correlation matrix obtained by the traditional two-dimensional cross-correlation algorithm,and systematically compare the computational efficiency and image quality of the two methods.