基于CUDA的光纤传像束光纤数量优化检测

CUDA based Optimal Detection of Fiber Number in Optical Fiber Image Transmission Bundle

光纤传像束是利用一定数量且规则排列的光纤实现传输图像的无源器件。为保证其成像质量,文章基于机器视觉技术研究了一种对光纤数量的优化检测系统。针对光纤传像束端面图像采集不完整或边缘有缺陷的情况,文章提出了一种基于欧拉数计算的边缘优化处理算法,可以识别、标记并分类图像中所有的边缘轮廓,不仅实现了对光纤数量的精确检测,而且通过图形处理器(GPU)处理提高了算法性能。文章算法均在Visual Studio 2013、Opencv 3.1和CUDA 10.1的环境下调试运行。实验结果表明,该系统对图像中光纤总数的检测误差率约为0.27%,优化检测的误差率约为0.03%,基本满足设计要求。不仅如此,与常规的检测方法相比,文章所提优化算法可大幅度降低对阈值的选取要求,并有更高的检测效率。在统一计算设备架构(CUDA)的支持下,双边滤波提速约20倍,优化算法及整体运行减少近1/2时间。

Optical fiber image transmission bundle is a passive device that using a certain number of regularly arranged optical fibers to transmit images.In order to ensure its imaging quality,the detection system for the number of optical fibers is optimized based on the machine vision technology.In view of the incomplete image acquisition of the end face of the optical fiber image transmission bundle or the defect of the edge,this paper proposes an edge optimization processing algorithm based on Euler number calculation,which can identify,mark and classify all the edge contours in the image.It not only realizes the accurate detection of the number of optical fibers,but also improves the performance of the algorithm through Graphic Processing Unit(GPU)processing.The algorithms in this paper are debugged and run in the environment of visual studio 2013,Opencv 3.1 and CUDA 10.1.The experimental results show that the detection error rate of the total number of optical fibers in the image is about 0.27%,and the error rate of optimized detection is about 0.03%,which basically meets the design requirements.Moreover,compared with the conventional detection methods,the optimization algorithm proposed in this paper can greatly reduce the selection requirements of threshold and have higher detection efficiency.With the support of Compute Unified Device Architecture(CUDA),the speed of bilateral filtering is increased by about 20 times,and the optimization algorithm and overall operation time are reduced by nearly 1/2.