Subpixel-Accurate Real-Time Background Oriented Schlieren via Optimized Optical Flow Algorithms

Background Oriented Schlieren (BOS) technology, known for its simple setup, low cost, and applicability to visualize complex flow fields over large areas, has become a crucial method for flow field visualization. However, current BOS technology has several limitations, such as inadequate real-time performance, challenges in quantitative analysis, and high computational resource requirements. To address these limitations, this study presents the Real-time Gradient-based BOS (RTG-BOS) method, which enhances optical flow performance by substituting the traditional constant brightness assumption with a constant gradient assumption. This gradient assumption is applied to both the Dense Inverse Search (DIS) and Farneback algorithms, leveraging the principle that local image structures remain stable over short time intervals, despite variations in brightness. Additionally, the preprocessing pipeline includes sub-pixel interpolation and a three-layer progressive temporal smoothing to enhance image quality and retain schlieren image details. To achieve real-time processing, a GPU-based parallel architecture is employed with an optimized data pipeline. This design allows for concurrent module execution and full GPU resource utilization, thereby accelerating the entire BOS workflow. Experimental validation shows that the proposed method achieves 30 frames per second in most scenarios, fulfilling the real-time requirements for industrial applications.