Algorithm Implementation And Optimization Of Crack Detection Based On CUDA

Cloud computing, intelligent, automated technology has penetrated into all aspects of life and production, and real-time computing has become an urgent problem. Using a computer for testing positive for crack's safe, it's easy to operate as features are applied. In the field of digital image processing, GPU usage is booming, GPU-based digital image processing, can achieve faster processing speed, the results obtained earlier has a very high prospect in the automated production. In order to solve the existing crack detection method has the accuracy and timeliness can not meet the problem, according to the characteristics parts of the surface of the image, the paper-based CUDA implementation uses digital image processing method for detecting cracks in the process of implementation, precise high as the goal to find parallelizable algorithm, detection of cracks, and accelerating algorithms, real-time purposes,This article specifically for crack detection in the implementation of digital image processing algorithms to achieve GPU explored. In the paper, the detection algorithms used were different optimization acceleration, image processing stage, the use of shared memory for the histogram equalization algorithm to optimize the histogram and the shared memory optimized computing applied to two value segmentation, the segmentation algorithm parallel implementation is the maximum variance threshold, Gaussian filtering and morphological filtering and edge detection, the use of parallel processing on the image template, and the template used in the optimization. Image thinning algorithm used herein based on two times the scan to achieve an increase in the degree of parallelism, communication region detection algorithm and check the use of set further upgrade, each thread processing pixel is not associated independent write and random read results through increased parallelism lift up speed, in addition, the full text of more than one algorithm uses parallelism reduction strategies to achieve further optimization of the algorithm.In experiment, each crack detection algorithm uses a different strategy to accelerate, both to ensure the realization of parallelism, but also to achieve a significant reduction of the running time, in contrast to the final experiment, each parallel algorithms are implemented using CUDA compared with the CPU in the traditional standard algorithm. Experimental results show that, implemented on CUDA crack detection algorithm that greatly improves the efficiency of detection, faster and more efficiently to get the test results, for real-time implementation has great significance.