【作者】 杨洋；

【导师】 田裕鹏；

【作者基本信息】 南京航空航天大学 ， 工程硕士（专业学位）， 2019， 硕士

【摘要】 机器视觉在工业检测领域的应用广泛,其检测精度与效率相比人工检测有了质的提升。本文针对汽车保险盒内孔缺陷的工业检测需求,研究了基于机器视觉的汽车保险盒内孔缺陷检测系统。首先,本文根据汽车保险盒的检测需求,对机器视觉成像系统的镜头、相机、光源进行了理论分析与计算选型,并对成像系统的支撑机构进行了设计,以提高系统成像质量。其次,针对汽车保险盒在实际检测时易发生误检的问题,从工件热胀冷缩和制造模具两方面对问题原因进行了分析,提出了不同模具生产的工件进行针对性分类再检测的方法。针对检测时同一缺陷多次检出的问题,设计了自排查的方法予以解决。与原保险盒缺陷检测系统相比,本系统有效降低了误检率。再次,针对保险盒内孔图像在检测过程中匹配效果不佳的问题,本文提出了一种基于区域图像特征的图像分割策略,采用了基于位置因素的区域配准策略对匹配进行约束,设计了基于二维高斯分布的模板滤波方法。对比实验表明,本文方法较已有方法在匹配精度方面提高了8%。然后,针对目前匹配耗时过长的问题,本文在算法方面研究了基于双边投影直方图的匹配方法,在硬件设备方面研究了GPU加速的技术,对匹配进行了实时性优化。对比实验表明,本文的方法耗时较已有方法减少70%。最后,针对实际检测过程中易出现误检的问题,研究了一种能够对检测系统进行故障自诊断的算法,减少人工排查故障的工作量。本文还研制了保险盒内孔缺陷检测系统,同时对系统进行了实验研究,实验表明,系统对保险盒内孔的检测准确率较已有系统至少提高了13%,达到工业检测要求。更多还原

【Abstract】 Machine vision is widely used in the field of industrial detection,and its detection accuracy and efficiency have been improved qualitatively compared with manual detection.In this paper,aiming at the need of industrial inspection of the inner hole defect of automobile insurance box,the defect detection system of the inner hole of automobile insurance box based on machine vision is studied.Firstly,according to the inspection requirement of automobile insurance box,the lenses,camera and light source of machine vision imaging system are theoretically analyzed and calculated,and the supporting mechanism of the imaging system is designed to improve the imaging quality of the system.Secondly,Aiming at the problem that the automobile insurance box is easy to be mischecked in actual inspection,this paper analyses the causes of the problem from two aspects: the thermal expansion and cold contraction of the workpiece and the manufacture of the modules,and puts forward the method of classifying and re-testing the workpieces produced by different modules.Aiming at the problem that the same defect is detected repeatedly during the detection,a self-checking method is designed.Compared with the prior insurance box defect detection system,the mistaken detection rate of the system designed in this paper is reduced.Thirdly,aiming at the problem that the matching effect of the inner hole image of insurance box is not good in the process of detection,this paper proposes an image segmentation strategy based on the regional image features,uses the regional registration strategy based on location factor to restrict the matching,and designs a template filtering method.The experimental results show that the matching accuracy of the proposed method is 8% higher than that of the existing method.Then,aiming at the problem that the matching time is too long at present,this paper studies the matching method based on bilateral projection histogram in algorithm,GPU acceleration technology in hardware equipment to optimize the matching step in real time.The experimental results show that the time-consuming of the proposed method is 70% less than that of the existing methods.Finally,aiming at the problem of error detection in actual detection process,an algorithm for fault self-diagnosis of detection system is studied to reduce the workload of manual troubleshooting.This paper develops a defect detection system of the insurance box,and carries out an experimental study on the system.The experimental results show that the accuracy of the system for the detection of the inner hole of the insurance box is at least 13% higher than that of the existing system.更多还原