Deep learning-based particle tracking velocimetry (PTV) for spherical and non-spherical particles: Application to granular flow and sediment transport

Transport of granular materials on Earth and planetary surfaces are at the heart of landscape dynamics and geohazards. These transport phenomena are controlled by particle-scale mechanisms, including particle motion, collisions, and interactions with the ambient fluid, which highlights the importance of particle-resolved measurements in physical experiments. However, despite recent progress in particle tracking velocimetry (PTV) for spherical (and regularly shaped) particles, there still lacks a robust technique in tracking and analyzing the motion of non-spherical particles, particularly because conventional PTV cannot identify moving objects of an arbitrary shape. This limitation largely compromises our particle-scale understanding of the transport of natural granular materials with a wide range of shapes and sizes. To tackle this issue, we propose a novel deep learning-based PTV framework for arbitrarily shaped and sized particles, which consists of a real-time computer vision algorithm called YOLO (you only look once) and an accurate inter-frame matching algorithm based on Kalman filtering. The proposed PTV framework is validated in various granular flow and sediment transport scenarios, using high-resolution data obtained from discrete element method simulations and small-scale physical experiments. Using this new technique, we are able to precisely analyze the kinematics information of spherical, non-spherical, and mixed particles with different concentrations in a series of open channel bedload transport experiments. Scaling relations are obtained between the sediment flux and bed shear stress to reveal the effects of particle shape and composition on the sediment transport dynamics across bedload and sheet flow conditions. The proposed PTV technique and its potential applications are expected to provide a new avenue for future research on the micromechanical aspects of geophysical granular flow and sediment transport.