University of Rostock

Particle Image Velocimetry (PIV) and Particle Tracking Velocimetry (PTV) are important imaging techniques for flow characterization. They are applied in many different areas from the optimization of combustion processes to propulsion systems of ships. Particles injected into the liquid flow have to be detected and tracked by using high speed CCD or CMOS cameras with high spatial resolution. Meanwhile, three-dimensional motions of several thousand particles can be detected and tracked with a spatial resolution of 0.1 pixels allowing the characterization of instationary and turbulent flows. The basic limitation of PIV and other multi-dimensional imaging techniques is the lack of real-time processing for a high temporal resolution (frame rates above 100Hz). Hence, imaging techniques cannot be applied for process measurement technology. Furthermore, the required data rates for frame rates in the kHz range are extremely high so that the limited size of state-of-the-art RAMs restricts the process observation time to only a few seconds which is not sufficient for a meaningful analysis. Therefore, this project pursues two different approaches.

On one hand, a multidimensional estimation of the motion vector field has to be developed by using stochastically sampling of the image date. The motion vector field is directly estimated from two or more images without analyzing their individual features or particles. Traditional approaches consider complete images and using phase correlation, cross-correlation (PIV) or the filtering with periodic spatial structures (spatial filtering technique). In this project, these approaches have to be generalized to stochastically sampled images. The real-time solution comes at the expense of a reduced data rate.

On the other hand, the recording duration shall be extended by extreme subsampling while keeping the spatial resolution in a subsequent offline processing high. Compared to the resolution of the camera sensor only a few particles are to be detected. Hence, the signal to be detected is sparse and techniques like compressed sensing or spectral estimation can be applied. They shall be extended and improved for the problem at hand and compared with state-of-the-art techniques.

German National Science Foundation (DFG), 2013 - 2015