摘要
基于机器视觉的输送带跑偏检测方法检测的输送带边缘特征中包含伪边缘,现有研究难以识别伪边缘,且多场景适应性差。针对该问题,对输送带监控图像提取感兴趣区域(ROI)并进行归一化,采用较大阈值区间的Canny算法提取边缘特征点,以提高算法的场景适应性,并采用形态学滤波方法处理部分杂质及伪边缘;对于Canny算法无法检测到有效边缘的图像,对提取的ROI进行伽马变换和45,135°方向的梯度滤波,以增强边缘特征,之后进行基于Canny算法的特征点提取和形态学滤波。以边缘点像素值关系、邻域特征、紧密性特征,以及边缘线长度、相对位置、斜率等作为约束条件,采用基于分治搜索思想的直线筛选排序算法对提取的边缘特征点进行筛选及拟合,得到输送带实时边缘。将实时边缘的像素值与未发生跑偏时边缘像素值做差,得到当前跑偏的像素值。试验结果表明,针对多种场景下的输送带监控图像,基于Canny算法和直线筛选排序的输送带跑偏检测方法检测误差小于3个像素值,百张图像检测时间为6.9451 s,边缘计算机处理4路视频图像的CPU占有率为132%,满足现场输送带边缘检测的准确性、实时性要求。
The machine vision-based conveyor belt deviation detection methods detect conveyor belt edge features.The features contain false edges.The existing research is difficult to identify false edges and has poor adaptability to multiple scenes.To solve this problem,the region of interest(ROI)is extracted from the conveyor belt monitoring image and normalized.The Canny algorithm with a larger threshold range is used to extract edge feature points to improve the scene adaptability of the algorithm.Morphological filtering methods are used to deal with some impurities and false edges.For images where the Canny algorithm cannot detect effective edges,gamma transform and gradient filtering in the 45°and 135°directions are performed on the extracted ROI to enhance edge features.The feature point extraction and morphological filtering based on the Canny algorithm are carried out.The pixel value relationship of edge points,neighborhood features,compactness features,as well as the length,relative position,and slope of edge lines are taken as constraints.The line filtering and sorting algorithm based on the idea of divide and conquer search is used to filter and fit the extracted edge feature points to obtain a real-time edge of the conveyor belt.The pixel value of the real-time edge is subtracted from the pixel value of the edge when no deviation occurs,and the pixel value of the current deviation is obtained.The test results show that for the conveyor belt monitoring images under various scenes,the detection error of the conveyor belt deviation detection method based on the Canny algorithm and line filtering and sorting is less than three pixel values.The detection time of 100 images is 6.9451 s.The CPU occupancy of the edge computer processing four video images is 132%,which meets the accuracy and real-time requirements of on-site conveyor belt edge detection.
作者
王锴
曾祥进
黎新
张锐
徐成
WANG Kai;ZENG Xiangjin;LI Xin;ZHANG Rui;XU Cheng(School of Computer Science and Engineering,Wuhan Institute of Technology,Wuhan 430205,China;Hubei Three Gorges Laboratory,Yichang 445804,China)
出处
《工矿自动化》
CSCD
北大核心
2023年第3期23-30,52,共9页
Journal Of Mine Automation
基金
国家自然科学基金资助项目(61502355)
湖北省三峡实验室创新基金资助项目(SC215001)
武汉工程大学研究生教育创新基金资助项目(CX2021246)。
关键词
带式输送机
跑偏检测
边缘检测
特征点排序
直线筛选
分治搜索
belt conveyor
deviation detection
edge detection
sorting of feature points
line filtering
divide and conquer search
作者简介
王锴(1997-),男,山东菏泽人,硕士研究生,研究方向为图像处理和计算机视觉,E-mail:1181640871@qq.com。
