Given a temporal sequence of images of fluids we will use local polynomials to regularise obser-vations of normal flows into smooth flow fields. This technique furthermore allows us to give a qualitative local description of the flow field and to estimate the position of stagnation points. The algorithm is applied to two data sets. First a series of Meteosat images are processed with the purpose of estimating cloud motion. Secondly, the airflow in livestock buildings is estimated using the technique on images recorded of smoke patterns illuminated by a laser sheet. Here the purpose is the evaluation of the ventilation system.