Scholarlink

<p>Background-oriented schlieren (BOS), owing to its simple optical setup, low cost, and applicability to large-field-of-view measurements in complex environments, has become a widely used technique for flow-field visualization. However, current BOS methods still face several limitations, including insufficient real-time capability, difficulties in quantitative analysis, and high computational cost. To address these limitations, this study proposes a real-time gradient-based BOS method which enhances optical flow estimation by introducing a constant gradient assumption in place of the conventional brightness-constancy constraint. This assumption is incorporated into both the Dense Inverse Search (DIS) and Farneback algorithms, based on the premise that local image structures remain relatively stable over short temporal intervals despite intensity variations. Furthermore, the preprocessing pipeline integrates subpixel interpolation and three-layer progressive temporal smoothing to improve image quality while preserving schlieren details. To enable real-time processing, a GPU-based parallel architecture is developed together with an optimized data pipeline. This design enables concurrent execution of multiple processing modules and improves GPU resource utilization, thereby accelerating the overall BOS workflow. Experimental results show that the proposed method achieves 30 frames per second in tested scenarios, indicating its potential for real-time industrial BOS applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Subpixel-accurate real-time BOS measurement via GPU-optimized optical flow algorithms

  • 期刊论文

  • 发表日期：

    2026年7月2日

查看全文
  • Borui Zheng,
  • Shixian Yang,
  • Jiayin Chen,
  • Hanyang Shen,
  • Wengang Zhang,
  • Liangyue Ji

摘要

Background-oriented schlieren (BOS), owing to its simple optical setup, low cost, and applicability to large-field-of-view measurements in complex environments, has become a widely used technique for flow-field visualization. However, current BOS methods still face several limitations, including insufficient real-time capability, difficulties in quantitative analysis, and high computational cost. To address these limitations, this study proposes a real-time gradient-based BOS method which enhances optical flow estimation by introducing a constant gradient assumption in place of the conventional brightness-constancy constraint. This assumption is incorporated into both the Dense Inverse Search (DIS) and Farneback algorithms, based on the premise that local image structures remain relatively stable over short temporal intervals despite intensity variations. Furthermore, the preprocessing pipeline integrates subpixel interpolation and three-layer progressive temporal smoothing to improve image quality while preserving schlieren details. To enable real-time processing, a GPU-based parallel architecture is developed together with an optimized data pipeline. This design enables concurrent execution of multiple processing modules and improves GPU resource utilization, thereby accelerating the overall BOS workflow. Experimental results show that the proposed method achieves 30 frames per second in tested scenarios, indicating its potential for real-time industrial BOS applications.