Improving Performance of Particle Tracking Velocimetry Analysis with Artificial Neural Networks and Graphics Processing Units

Flow analysis has a wide range of applications both in engineering and science, and many techniques have been developed over the years that allow the data extraction and measurement of the flow dynamics. In particular, Particle Tracking Velocimetry (PTV) techniques have shown good results when combined with Artificial Neural Networks (ANN) techniques, especially using Self-Organizing Maps (SOM). In order to improve the performance and reduce the time consumption of proposed SOMs for PTV analysis, the parallel nature of modern architectures such as Graphics Processing Units (GPU) can be used. In this paper we describe how the GPU architecture can be exploited for the implementation of the inherent parallelism of a SOM for PTV analysis, and measure the performance obtained with different optimizations techniques. We show that it is possible to gain a speedup of ∼ 5 when running in parallel.