Volumetric Flow Estimation for Incompressible Fluids Using the Stationary Stokes Equations

Katrin Lasinger, Christoph Vogel, Konrad SchindlerICCV 2017 · DOI

Introduction

Bridging the gap between experimental fluid dynamics and computer vision, this work applies variational motion estimation to 3D fluid volumes. In experimental fluid mechanics, researchers use Particle Imaging Velocimetry (PIV) to visualize flow by injecting tracer particles into a fluid and reconstructing 3D volumes of their distributions across time steps.

The primary challenge lies in the matching step: accurately estimating the 3D velocity field between two time-steps. Traditional PIV methods often rely on local window matching and heuristic smoothing, which struggle with high particle densities and ignore the physical properties of the fluid, such as incompressibility. Furthermore, processing high-resolution 3D volumes is computationally intensive, scaling cubically with resolution.

Overview of the volumetric flow estimation process, showing input particle distributions at times $t_1$ and $t_2$ and the resulting estimated velocity components $V_x, V_y, V_z$.

Methodology

The authors propose a variational energy minimization framework: $E(\mathbf{v}) = \lambda E_D(\mathbf{v}) + E_S(\mathbf{v})$, where $E_D$ is the data term and $E_S$ is the regularizer.

Physics-Based Regularization

Instead of generic smoothness priors like Total Variation (TV), the method derives a regularizer from the stationary Stokes equations. This approach penalizes the squared gradient of the flow while strictly enforcing a divergence-free constraint ($\nabla \cdot \mathbf{v} = 0$), ensuring the estimated motion is physically plausible for incompressible fluids.

Semi-Dense Formulation

To manage massive 3D datasets, the resolution of the flow field is decoupled from the data resolution. Flow vectors are estimated on a coarser grid (e.g., every 4th voxel), while the data term (particle matching) is evaluated at full resolution to preserve sharp geometric details. This semi-dense approach provides a $64\times$ reduction in variables with minimal loss in accuracy.

Optimization and Data Term

The problem is solved using a first-order primal-dual algorithm, suited for handling the hard constraints of the divergence-free requirement. For the data term, a simple Sum of Squared Differences (SSD) in a local 3D neighborhood was found to be most effective, outperforming more complex robust measures in the high-contrast environment of PIV.

Key Results and Insights

The proposed Stokes-based regularizer achieved an Average Endpoint Error (AEE) of 0.369, significantly outperforming standard quadratic regularization (0.408) and Total Variation (0.462). Despite being a two-frame method, it produces results competitive with highly engineered commercial PIV software that utilizes multiple time-steps and temporal smoothing.

A key insight is that for fluids, the best regularizer is the actual equations of motion. Enforcing the divergence-free constraint acts as a powerful prior that resolves ambiguities in particle matching. Additionally, the semi-dense approach with a step size of $h=2$ outperformed full-resolution matching ($h=1$), suggesting the sparser grid helps avoid overfitting to noise.

Limitations and Future Work

Current limitations include the two-frame constraint and the approximation of using stationary Stokes equations rather than full Navier-Stokes for highly dynamic flows. Future work may involve a "tightly coupled" approach that simultaneously estimates 3D reconstruction and flow to mitigate artifacts like "ghost particles" created during the initial reconstruction phase.

Summary updated

BibTeX
@inproceedings{lasinger2017volumetric,
  author = {Katrin Lasinger and Christoph Vogel and Konrad Schindler},
  title = {{Volumetric Flow Estimation for Incompressible Fluids Using the Stationary Stokes Equations}},
  booktitle = {IEEE/CVF International Conference on Computer Vision},
  year = {2017},
  doi = {10.1109/iccv.2017.280},
}
Figure