Measurement of Fluid Flow Velocity by Using Infrared and Visual Cameras: Comparison and Evaluation of Optical Flow Estimation Algorithms

Due to the rising importance of energy conservation and emissions reduction, gas leak detection has become critical. Optical gas imaging (OGI) is widely used in this field. For the quantification of gas emissions, the images captured by OGI cameras can be analyzed by optical flow estimation algorithms to achieve gas velocity estimation. Three different algorithms have been applied in this field, namely Brox, Farnebäck, and Flownet2. This study investigates the performance of fluid flow velocity measurements by using three optical flow estimation algorithms. Meanwhile, the algorithms are also compared under different flow-generating substances that are visible under visual and mid-wave infrared OGI cameras. Due to the higher frame rate and resolution, the results of the three different algorithms from the visual data set prove better than those from the infrared data set. In addition, the effect of different parameters on flow velocity measurements is analyzed. The Brox and Farnebäck methods are sensitive to parameter choices. In contrast, the deep learning based Flownet2 algorithm exhibits greater robustness and is less affected by speed or frame rate variations.