KHz Real-Time PIV Using GPU

Time-resolved real-time Particle Imaging Velocimetry (PIV) has potential in fluid flow applications where feedback control or active monitoring is neccessary. It also offers the possibility to reduce storage capacity requirements in experiments. However, it has historically been limited to just dozens, or more recently a few hundred, frames per second. This has limited its usefulness to low speed flows such as in water or oil. We present, for the first time, real-time 2D PIV in the KHz regime (at 1MP resolution input and 1 vector per 4x4 pixel window output) using the optical flow hardware accelerator built into modern Nvidia GPU's. The method can be scaled to workloads shared across multiple GPU's, and is easily cross-platform transferable due to its ability to be programmed in Python. Preliminary benchmark performance experiments are performed, and tradeoffs to using this method are discussed.