eduzhai > Applied Sciences > Engineering >

Unsupervised Learning of Particle Image Velocimetry

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: Particle Image Velocimetry (PIV) is a classical flow estimation problem whichis widely considered and utilised, especially as a diagnostic tool inexperimental fluid dynamics and the remote sensing of environmental flows.Recently, the development of deep learning based methods has inspired newapproaches to tackle the PIV problem. These supervised learning based methodsare driven by large volumes of data with ground truth training information.However, it is difficult to collect reliable ground truth data in large-scale,real-world scenarios. Although synthetic datasets can be used as alternatives,the gap between the training set-ups and real-world scenarios limitsapplicability. We present here what we believe to be the first work which takesan unsupervised learning based approach to tackle PIV problems. The proposedapproach is inspired by classic optical flow methods. Instead of using groundtruth data, we make use of photometric loss between two consecutive imageframes, consistency loss in bidirectional flow estimates and spatial smoothnessloss to construct the total unsupervised loss function. The approach showssignificant potential and advantages for fluid flow estimation. Resultspresented here demonstrate that our method outputs competitive results comparedwith classical PIV methods as well as supervised learning based methods for abroad PIV dataset, and even outperforms these existing approaches in somedifficult flow cases. Codes and trained models are available atthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×