Computer Physics Communications

Cerda M, Navaro CA, Silva J, Waitukaitis SR, Mujica N, Hitschfeld N.

ABSTRACT
Many fields of study, including medical imaging, granular physics, colloidal physics, and active matter, require the precise identification and tracking of particle-like objects in images. While many algorithms exist to track particles in diffuse conditions, these often perform poorly when particles are densely packed together—as in, for example, solid-like systems of granular materials. Incorrect particle identification can have significant effects on the calculation of physical quantities, which makes the development of more precise and faster tracking algorithms a worthwhile endeavor. In this work, we present a new tracking algorithm to identify particles in dense systems that is both highly accurate and fast. We demonstrate the efficacy of our approach by analyzing images of dense, solid-state granular media, where we achieve an identification error of 5% in the worst evaluated cases. Going further, we propose a parallelization strategy for our algorithm using a GPU, which results in a speedup of up to 10x when compared to a sequential CPU implementation in C and up to 40x when compared to the reference MATLAB library widely used for particle tracking. Our results extend the capabilities of state-of-the-art particle tracking methods by allowing fast, high-fidelity detection in dense media at high resolutions.

LINK:
https://doi.org/10.1016/j.cpc.2018.02.010