Optical flow algorithms and background oriented schlieren for a supersonic wind tunnel setup

Abstract

The work details a comparison between various GPU accelerated optical flow algorithms that the OpenCV open source computer vision library provides as implemented in a Background Oriented Schlieren (BOS) setup for a supersonic wind tunnel. The chosen test case is the visualisation of supersonic flow (Mach 2.5) over a conical wind tunnel model. The theory for predicting the distortion of the images is developed and used for comparison, as well as a correlation based brute-force block matching algorithm, which is typical of current BOS setups. Additionally, two background images were designed and tested with all algorithms to investigate the interaction between these backgrounds and the various algorithms. The theory behind the optical flow calculations is outlined and discussed in the BOS context and in particular, how it can be used for future background designs. It is also shown that there exists a general compromise between optical flow detail and noise rejection in the processed BOS images and that the more modern algorithms that have been implemented in the OpenCV toolbox generally outperform the traditional correlation based block-matching methods when their accuracy is compared.