Subpixel Displacement Measurement at 784 FPS: From Algorithm to Hardware System
Subpixel Displacement Measurement at 784 FPS: From Algorithm to Hardware System
复制标题
784 FPS 的亚像素位移测量：从算法到硬件系统
DOI：
10.1109/tim.2022.3162290
复制
发表时间：
2022
影响因子：
5.6
通讯作者：
Ikenaga Takeshi
中科院分区：
文献类型：
--
作者：
Du Songlin;Gu Kaidong;Ikenaga Takeshi
基于亚像素级精度的物体位移测量由于具有补偿相机分辨率的巨大潜力，在许多基于计算机视觉的应用中越来越受到关注。虽然超高速测量在许多领域都有很高的要求，但现有的亚像素位移测量研究主要集中在精度上，很少有研究试图达到高测量速度。提出了一种超高速亚像素位移测量方法。在算法方面，提出了一种基于特征融合的仅方向符号相关、一种无拟合的亚像素位移测量方法和一种双对角线强度间隙量化方法。从硬件系统的角度出发，为所提出的硬件友好算法设计了一个系统级架构，以实现高度并行处理。通过集成高速工业相机、现场可编程门阵列（FPGA）和计算机，将所提出的硬件架构实现为一个实际系统。实验结果表明，该方法在图像处理核心以每秒784帧（FPS）的速度实时工作，延时为0.75 ms /帧，处理能力指数（PCI）量化的平均测量稳定性比单符号相关的离散余弦变换（DCT）提高了5.75。
Measuring object displacement with subpixel-level accuracy is attracting increasing attention in numerous computer-vision-based applications, because of its high potential in compensating for camera resolution. Although ultrahigh-speed measurement is highly desired in many fields, existing researches on subpixel displacement measurement mainly concentrate on accuracy, and few works attempt to reach high measurement speed. This article proposes an ultrahigh-speed subpixel displacement measurement approach. From algorithm perspective, a feature fusion-based directional sign-only correlation, a fitting-free subpixel displacement measurement method, and a dual-diagonals intensity-gap quantification are proposed. From hardware-system perspective, a system-level architecture for the proposed hardware-friendly algorithm is designed for highly-parallel processing. The proposed hardware architecture is implemented as a practical system by integrating a high-speed industrial camera, a field programmable gate array (FPGA), and a computer. Experimental results show that the proposed approach works in real-time at 784 frames per second (FPS) in image processing core with a delay of 0.75 ms per frame, while the average measurement stability quantified by the process capability index (PCI) improves 5.75 compared with discrete cosine transform (DCT) sign-only correlation.