摘要
针对软管装配的视觉检测问题,定义了软管装配视觉检测问题和图像采集约束,开发了相应的检测系统。基于五层结构模式,设计了系统架构模型及相关的训练和检测流程;利用投影分析和色谱分析,讨论了发动机软管装配中的目标形状和颜色约束;结合类平行曲线特征,提出了边界点拟合、候选区域融合和目标特征判断等图像处理技术,研究了目标识别和缺陷检测方法体系。实验表明,目标识别平均准确率达到94%以上,缺陷全部检出,每图运行时间不到30s,能同时满足在线检测的精度和效率要求。
Aiming at the visual inspection problem of hoses assembly, the definition of visual inspection problem and the constraints of image acquisition were defined, and a corresponding inspection system was developed. Based on the pattern of five-layer structure, the system schema and related training and inspection processes were designed. With projection analysis and chromatography, shape and color constraints of objects for hose assembly of engines were discussed. According to the characteristics of perceptually parallel curves, image processing techniques of edge- points fitting, candidate regions merging and object feature judging were put forward. Furthermore, a methodology of object recognition and defect inspection was studied. Experiments showed that average object recognition accura- cies were over 94%, and all defects could be inspected with a running time of less than half an hour for each figure, which could fulfill requirements of accuracy and efficiency of online inspection concurrently. The presented system could be generalized to inspect other hose-like parts.
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2015年第6期1486-1496,共11页
Computer Integrated Manufacturing Systems
基金
中央高校基本科研业务费资助项目(CDJZR11110001)~~
关键词
目标识别
软管装配
类平行曲线
视觉检测
计算机视觉
object recognition
hoses assembly
perceptually parallel curves
visual inspection
computer vision