“GPGPU Accelerated Flow Diagrams” by Bird and Laycock

  • ©Daniel Bird and Stephen Laycock



Entry Number: 10


    GPGPU Accelerated Flow Diagrams



    In today’s ’big data’ environment the sheer volume of data can cause problems in the creation of visualisations of trajectory data. An overabundance of data creates a number of issues not only tied to the computational complexity of the visualisation, but also with occlusion of data. This is particularly an issue for large datasets, where multiple data points overlap, causing some parts of the data to become obscured. More involved aggregation techniques for large movement datasets of GPS data are now beginning to be developed. This allows for the visualisation of not only where movement is occurring, but also what direction. For example, Andrienko and Andrienko [2011] developed a method of aggregation that creates a number of flows between ‘characteristic points’. These characteristic points are defined as relocations within a trajectory of particular interest, such as significant turning points or stopovers. Recent developments by Graser et al. [2020] improved upon this by creating flow maps for large maritime vessel movement datasets. This method however utilises the power of a distributed computing environment. We argue that with adjustments to the original methods used by Andrienko and Andrienko, it can be made more suitable for processing large amounts of data utilising the massively parallel processing of the GPU, without the need for networked clusters.

    Here we present a GPGPU accelerated aggregate flow method, used to aggregate the millions of data points found in today’s ecological movement datasets. We adapt the sequential method of Andrienko and Andrienko [2011], making it more suitable for the GPU architecture. This allows for interactive adjustment of parameters, removing the disconnect of adjustment and results that occurs when long computation times are required. This creates a fully interactive environment that can be used to investigate large amounts of movement data that can be used to examine phenomena at multiple scales.



    We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research. We would also like to thank Dr Aldina Franco for supplying the GPS data for the White Storks [Gilbert et al. 2016] and the discussions on the research.


ACM Digital Library Publication:

Overview Page: