[Paper Review] From Particle Tracks to Velocity and Acceleration Fields Using B-Splines and Penalties

Sébastian Gesemann|arXiv (Cornell University)|Oct 30, 2015

Fluid Dynamics and Turbulent Flows1 references20 citations

TL;DR

This paper presents a B-spline-based method to reconstruct continuous velocity and acceleration fields from noisy, scattered particle tracks in fluid flows. By using penalized least squares optimization with physical constraints—such as divergence-free flow for incompressible conditions—it achieves high spatial resolution while minimizing low-pass filtering effects inherent in traditional correlation-based methods like TomoPIV.

ABSTRACT

In this work a method for reconstructing velocity and acceleration fields is described which uses scattered particle tracking data from flow experiments as input. The goal is to reconstruct these fields faithfully with a limited amount of compute time and exploit known flow properties such as a divergence-free velocity field for incompressible flows and a rotation-free acceleration in case it is known to be dominated by the pressure gradient in order to improve the spatial resolution of the reconstruction.

Motivation & Objective

• Address the challenge of reconstructing high-resolution velocity and acceleration fields from sparse, noisy particle tracking data in 3D flows.
• Overcome the spatial lowpass filtering limitation of correlation-based methods like TomoPIV, which rely on windowed cross-correlation.
• Leverage known physical properties—such as divergence-free velocity fields in incompressible flows and curl-free acceleration fields dominated by pressure gradients—to improve reconstruction accuracy and resolution.
• Develop a continuous, smooth representation of particle tracks and flow fields using B-splines, enabling sub-sampling and derivative computation at arbitrary points.
• Minimize computational cost and memory usage compared to tomographic reconstruction methods by avoiding volume discretization and large constrained linear systems.

Proposed method

• Use B-spline curves to model particle tracks as continuous functions of time, enabling smooth interpolation and exact derivative computation (velocity and acceleration) via linear combinations of control points.
• Apply Tikhonov-like regularization (penalization of higher-order derivatives) to reduce noise in particle tracks, with the penalty parameter λ chosen based on signal-to-noise spectral characteristics.
• Formulate the track fitting as a weighted least squares problem that minimizes the sum of squared residuals between observed particle positions and the B-spline curve, with a penalty on the third derivative to control smoothness.
• Extend the B-spline framework to 3D spatial vector fields by defining a Cartesian lattice of control points, where each field component (e.g., velocity) is represented as a weighted sum of 3D B-spline basis functions.
• Incorporate physical constraints into the flow field reconstruction by adding linear equations to the least squares system that penalize divergence (for incompressible flow) or curl (for pressure-dominated acceleration fields).
• Solve the resulting large, sparse linear system using standard numerical methods, maintaining computational efficiency while enforcing physical consistency and improved spatial resolution.

Experimental results

Research questions

• RQ1 Can B-spline-based reconstruction of particle tracks reduce noise and preserve high-frequency content better than correlation-based methods like TomoPIV?
• RQ2 To what extent can physical constraints—such as divergence-free velocity or curl-free acceleration—improve the spatial resolution and accuracy of reconstructed flow fields?
• RQ3 How does the choice of penalty parameter λ affect the trade-off between noise suppression and fidelity to the original particle track data?
• RQ4 Can a continuous, differentiable representation of particle tracks and flow fields be efficiently computed with minimal memory and CPU usage compared to tomographic reconstruction?
• RQ5 Does the proposed method outperform traditional methods in terms of spatial resolution and accuracy when applied to real experimental particle tracking data?

Key findings

• The method successfully reconstructs continuous velocity and acceleration fields from scattered, noisy particle tracks using B-splines, enabling sub-sampling and exact derivative computation at any point in time or space.
• By penalizing the third-order derivative of the B-spline curve, the approach approximates an optimal Wiener filter, achieving noise reduction while preserving high-frequency content.
• The use of physical constraints—specifically divergence-free velocity fields and curl-free acceleration fields—improves spatial resolution and enforces physical consistency in the reconstructed fields.
• The reconstruction process is computationally efficient, requiring significantly less CPU and RAM than TomoPIV, due to direct particle tracking and avoidance of volume discretization.
• Variance in B-spline coefficient estimates is higher at the temporal edges of particle tracks, indicating reduced reliability in velocity and acceleration estimates at track boundaries.
• The method enables accurate computation of velocity and acceleration at arbitrary time instants through temporal super-sampling, which is not feasible with standard windowed cross-correlation techniques.