A GPU-accelerated particle-detection algorithm for real-time volumetric particle-tracking velocimetry under non-uniform illumination

Real-time volumetric particle tracking velocimetry (VPTV) equipped with field-programmable gate array (FPGA) cameras has been used for open-space, low particle density, and large-scale airflow measurements with long measurement periods. However, the particle detection accuracy of FPGA cameras is inevitably hindered by non-uniform illumination, resulting in a reduction in the particle detection ratio and positional accuracy. In this article, we propose to use both synchronized FPGA and grayscale cameras in a VPTV system, where grayscale cameras utilize a new algorithm based on two-frame centroid and corner extraction (TFCCE) under non-uniform white-light illumination. To keep the frame rate of the FPGA cameras the same, the TFCCE algorithm was accelerated by a graphics processing unit (GPU). The simulation results showed that the 2D particle detection ratio of TFCCE was enhanced to approximately 80% with a positional accuracy of 0.57 pixels, compared to 30% and 0.94 pixels for the single-frame centroid extraction used in the FPGA. The GPU version of TFCCE was 15.09 times faster than the CPU version, resulting in a calculation time of 4.55 ms per image, compared to 68.70 ms when using the CPU. This system was also validated by the measurement of a turbulent jet flow in real-time at 120 fps. The experimental results correspond well with data published in the literature. Therefore, this new algorithm can improve VPTV systems in terms of particle detection ratio and positional accuracy in real time under conditions of non-uniform illumination.