Chapter 10 Phase Unwrapping Work Done via Graphic Processing Unit M.J. Huang and Y.C. Liu Abstract Phase-shifting technique is widely used in phase detection field. With the aid of this technique, hundreds to thousands of wavelength accuracy result can be achieved. Further calculating the grabbed phase-stepping frames yields the phase map, which is in wrapped format and has to be further treated to convert into a continuous (i.e., an unwrapped) mode. Different algorithms—path-dependent or path-independent, temporal or spatial, point-by-point or regional, noisesensitive or noise-immune, etc., have been proposed for solving problems in different applications. In present study, graphic processing unit (GPU), which has been developing rapidly in recent years, is utilized in addition to shorten and accelerate the processing time needed. By the aid of GPU parallel processing technique, the retrieving work is much more time effective. Since temporal phase unwrapping deals wrapped data from load-stepping or wavelength-stepping basis, it can be easily benefit from the parallel processing of GPU. Experimental work of photoelasticity is verified by the present study. Keywords Photoelasticity • Temporal phase unwrapping • Parallel processing • Isoclinic • Isochromatic 10.1 Introduction GPU consists of thousands of cores which can handle multiple tasks simultaneously, comparing with CPU (central processing unit) which has only a few cores to deal with serial processing tasks. On usage of GPU, the processing time can be dramatically reduced, especially while temporal wrapped data are treated. Temporal phase unwrapping is an effective way of phase retrieving method, like load stepping method [1] and different wavelength method [2, 3]. This technique later on is applied on the isoclinic as well as isochromatic wrapped phase retrieving and gets quite well results. Kihara [4] proposed a coincidence method with three wavelengths (RGB) to retrieve the phase jump 2πN for wrapped isochromatic, which can get unwrapped isochromatic correctly. This algorithm is implemented via parallel pixels processing and can reduce the consumed time significantly. 10.2 Parallel Processing Parallel processing is the ability to handle multiple tasks simultaneously, and GPU has highly parallel structure that makes it more effective than CPU for algorithms where processing of an amount of data is done in parallel. CUDA (the Compute Unified Device Architecture), is a parallel computing platform created by NVIDIA, and carried out by the GPUs that they produce. In CUDA, the data must be copy to the memory of GPUs for parallel processing, and after that, the result will be M.J. Huang (*) • Y.C. Liu Department of Mechanical Engineering, National Chung Hsing University, 250, Kuo-Kuang Road, Taichung, Taiwan 40227, Republic of China e-mail: mjhuang@dragon.nchu.edu.tw N. Sottos et al. (eds.), Experimental and Applied Mechanics, Volume 6: Proceedings of the 2014 Annual Conference on Experimental and Applied Mechanics, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-06989-0_10, #The Society for Experimental Mechanics, Inc. 2015 75
RkJQdWJsaXNoZXIy MTMzNzEzMQ==