基于机器视觉的智能制造产品识别与检测方法研究
哈尔滨工业大学工学硕士学位论文
Abstract
The world manufacturing giants have put forward the strategies to revitalize the local manufacturing industry in order to develop its own manufacturing industry. From "Industry 4.0" and "Industrial Internet" to "Made in China 2025", the essence is to develop the technology, industry and application of intelligent manufacturing. In the intelligent manufacturing environment, it is required that manufacturing companies can provide customized products, and then realize large-scale, customized and flexible production. With transformation of automation and digital, factories have introduced machine vision to classify and identify products in production lines. However, the production methods of these enterprises are still push-type production methods, and the product lines produced by the production line are few, so their application machines Visual counts are also less adaptable to product categories. That cannot meet the requirements for the identification of multiple products and small batches of customized products under the conditions of smart manufacturing. What’s more, this article researches product identification of machine vision in flexible production line under intelligent manufacturing environment, this paper deeply researches the matching of product and cloud data with machine vision in the flexible manufacturing environment, which mainly including improved research on matching template generation, product image preprocessing and outline extraction, research and improvement of outline feature description method, and the platform of detection and classification.
Firstly, based on the background of the transition to smart manufacturing production mode, this paper proposes a method to directly use the 3D model of cloud products to generate matching templates: analyze the 3D model data of the target object and to simulate the shooting of the product under the real shooting environment with OpenGL. Generate a matching template from the perspective. This method of image matching template generation can be used to improve the method of shooting production template images in current industrial production lines, and to increase the adaptability of machine vision technology to challenges in product types to meet the need to produce small-volume and multi-type products in the same production line.
- II -
哈尔滨工业大学工学硕士学位论文
Secondly, this paper compares various preprocessing algorithms such as image filtering and edge detection, and chooses bilateral filtering and Canny edge detection for image preprocessing. After that, perform contour extraction on the processed binary image. Afterwards, based on the classification of features by SVM, this paper proposes and compares the Fourier description, the Fourier description of the characteristics with the combined moment, and the elliptic Fourier description algorithm to extract the contour features for classification. Generate feature descriptors of MPEG-7 graphics library with three algorithms, and then identify or predict the verification with SVM. Among that, the correct rate of SVM-based elliptical Fourier description algorithm is 90%.
Finally, based on the aforementioned theoretical research, a visual inspection platform was built, and several products produced by the experiment were matched with the cloud storage data. Through experiments, the proposed SVM-based elliptic Fourier description algorithm is verified. This algorithm can meet the requirements for the classification of small and batch quantities of customized products in flexible production lines, and realize the matching of actual product and cloud product model data.
Keywords:Intelligent manufacturing, Machine vision, Fourier feature descrip-tor, Product identification
- III -
哈尔滨工业大学工学硕士学位论文
目录
摘要....................................................................................................................... I ABSTRACT ............................................................................................................ II 第1章绪论.. (1)
1.1课题来源 (1)
1.2研究目的与意义 (1)
1.3机器视觉的国内外研究现状 (2)
1.3.1 机器视觉 (2)
1.3.2 国外研究现状 (3)
1.3.3 国内研究现状 (4)
1.3.4 国内外文献综述的简析 (5)
1.4本文主要研究内容 (6)
第2章基于OPENGL的产品匹配模板生成 (8)
2.1相机标定 (8)
2.1.1 线性成像模型分析 (9)
2.1.2 相机畸变模型分析 (11)
2.2产品3D数据模型分析 (12)
2.2.1 STL数据格式分析 (13)
2.2.2 STL的读取与显示 (13)
2.2.3 STL模型轮廓提取 (14)
2.2.4 STL模型的面区域分割 (15)
2.2.5 STL模型的外轮廓提取 (16)
2.2.6 投影与消隐 (17)
2.3匹配模板生成 (18)
2.4本章小结 (18)
第3章图像预处理及轮廓提取 (19)
3.1图像滤波方法 (19)
3.1.1 方框滤波与均值滤波 (20)
3.1.2 中值滤波 (20)
3.1.3 高斯滤波 (21)
3.1.4 双边滤波 (22)
- IV -
哈尔滨工业大学工学硕士学位论文
3.2图像边缘检测算子 (23)
3.2.1 Roberts边缘检测算子 (23)
3.2.2 Perwitt边缘检测算子 (24)
3.2.3 Sobel边缘检测算子 (25)
3.2.4 Laplacian边缘检测算子 (25)
3.2.5 Canny边缘检测算子 (26)
3.2.6 边缘检测算子的效果对比 (26)
3.3轮廓提取原理 (28)
3.3.1 基于区域增长的轮廓提取 (28)
3.3.2 基于边缘追踪的轮廓提取 (30)
3.4本章小结 (31)
第4章基于傅里叶变换的轮廓特征描述 (32)
4.1轮廓描述方法 (32)
4.1.1 基于轮廓的形状描述方法 (33)
4.1.2 基于区域的形状描述方法 (35)
4.1.3 轮廓特征提取方法的选择 (37)
4.2傅里叶描述算子 (38)
4.2.1 傅里叶描述子的计算 (38)
4.2.2 傅里叶描述子不变性 (39)
4.2.3 傅里叶描述子归一化 (40)
4.2.4 通用相似度量法 (40)
4.2.5 SVM向量机 (41)
4.2.6 基于傅里叶描述子的SVM分类检测 (42)
4.2.7 基于矩特征的傅里叶特征描述 (43)
4.3椭圆傅里叶变换 (44)
4.3.1 椭圆傅里叶特征子的不变性构造 (46)
4.3.2 基于SVM椭圆傅里叶分类识别 (47)
4.4本章小结 (47)
第5章产品识别与实验分析 (49)
5.1工业相机标定 (49)
5.1.1 相机选型 (49)
5.1.2 镜头选择 (50)
5.1.3 视觉检测平台 (50)
- V -
哈尔滨工业大学工学硕士学位论文
5.1.4 标定实验 (51)
5.2产品的类型识别实验 (53)
5.2.1 模板库的生成及轮廓提取 (53)
5.2.2 特征生成 (55)
5.2.3 SVM类别预测 (55)
5.3本章小结 (55)
结论 (57)
参考文献 (58)
附录 (63)
哈尔滨工业大学学位论文原创性声明和使用授权说明 (66)
致谢 (67)
- VI -