An experiment on living microorganisms is conducted to gain insight into their motion and fluid exchange characteristics. Biocompatible microscopic particle image velocimetry (PIV)-systems are used to capture images of seeded particles in the induced fluid flows. To enhance the abilities of these devices we present a model-based approach for the reconstruction of admissible flow fields from captured images. A priori knowledge of the physical model of the flow is used to iteratively refine a predicted flow field. A physics-based filter operation generates a velocity field that is consistent with the model of incompressible laminar flows described by the Navier–Stokes equations. Interactive steering of the reconstruction process is achieved by exploiting programmable graphics hardware as a co-processor for numerical computations. To validate our method, we estimate velocity vector fields from synthetic image pairs of flow scenarios for which ground truth velocity fields exist and real-world image sequences of the flow induced by sessile microorganisms.