The quantification of the deformations presented by mechanical parts is a useful tool for several applications in engineering; regularly this quantification is performed a posteriori. In this work, a digital holographic interferometer for measuring micro-deformation at video rate is presented. The interferometer is developed with the use of the parallel paradigm of CUDA™ (Compute Unified Device Architecture). A commercial Graphics Processor Unit (GPU) is used to accelerate phase processing from the recorded holograms. The proposed method can process record holograms of 1024x1024 pixels in 48 milliseconds. At the best performance of the method, it processes 21 frames per second (FPS). This benchmark surpasses 133-times the best performance of the method on a regular CPU.
The numerical reconstruction of digitally recorded holograms has constituted the bottle neck for real-time digital
holography. The reconstruction process can be understood as the diffraction that undergoes a wavefront as it illuminates
the digitally recorded hologram. As this process is done numerically, the reconstruction of a M × N pixels hologram into
an image of similar dimensions is an operation with a Ο (M × N)2 complexity. The diffraction process can be represented
by a Fresnel transform or a scalable convolution of the recorded hologram. In these representations the numerical
reconstruction has a complexity of Ο (M × log N)2, still quite demanding computationally if the holograms are of 2048 × 2048 pixels. In this work, the power provided by a Graphics Processing Unit (GPU) is used to accelerate the numerical
reconstruction of digitally recorded holograms. The methodology is supported on the parallelization of typical Fresnel
transform and scalable reconstruction algorithms. On reconstructing holograms of 2048 × 2048 pixels, the reconstruction
is speeded up 20 times for the former method and 11 times for the scalable convolution. For holograms of 1024 × 1024,
the accelerated reconstruction methods allow for real-time digital holography.