Halcon 18.11图像处理指南：22章详解一维测量与blob分析

需积分: 10 192 浏览量更新于2024-07-15 收藏 7.14MB PDF 举报

"SolutionGuideI.pdf" 是一份深入学习 Halcon（霍克）机器视觉软件的指南文档，版本为18.11.0.0。Halcon是一款由德国MVTec Software GmbH开发的强大工业视觉软件，主要用于自动化图像处理和机器视觉任务。该文档共分为22个章节，涵盖了广泛的一维测量、blob分析（二值区域分析）以及各类高级图像处理技术，旨在帮助用户理解和掌握如何有效利用Halcon在实际生产环境中进行物体检测、识别、定位等操作。在文档的开篇，强调了版权信息和使用许可，所有内容未经MVTec官方授权，不得以任何形式复制、存储或传输。版权归属MVTec Software GmbH，自2007年至2018年，并提到了其持有的多项专利，如US7,062,093, US7,239,929等，表明了公司在技术上的领先地位和对知识产权的保护。文档中提到的几个关键品牌商标包括Microsoft（Windows、Windows Server系列、.NET框架等）、AMD、Intel、Linux、macOS、OpenCL、NVIDIA、CUDA、cuBLAS和cuDNN，这些都展示了与Halcon集成的相关技术和平台环境，使得用户能够在多样化的硬件和操作系统上实现高效的工作流程。每一章的内容详尽且实用，从基础概念入手，逐步引导读者了解如何设置环境，导入图像数据，执行一维特征检测（例如直线、边缘检测），到高级功能如blob分析（用于识别和测量图像中的目标区域）、模板匹配、光学字符识别（OCR）、机器学习应用等。此外，还可能涉及性能优化、错误处理和调试技巧，确保用户能够充分利用Halcon的功能来提升产品质量和生产线效率。 "SolutionGuideI.pdf" 是一本宝贵的资源，对于希望深入研究和应用Halcon机器视觉技术的工程师、开发者和专业人士来说，它不仅提供了理论知识，也包含了丰富的实践案例和实战指导，有助于提升在工业视觉领域的专业技能。

16 Guide to HALCON Methods

1.1 Color Inspection

For color inspection, see the descriptions for Color Processing on page 157.

1.2 Completeness Check

Completeness checks can be realized by different means. Common approaches are:

• Object and position recognition 2D/3D (see page 17 ff), which is suitable, e.g., when inspecting objects on

an assembly line.

• Variation Model on page 133, which compares images containing similar objects and returns the difference

between them considering a certain tolerance at the object’s border.

1.3 Identiﬁcation

Dependent on the symbols or objects you have to identify, the following methods are suitable:

• Identify symbols or characters

– Bar Code on page 189

– Data Code on page 205

– OCR on page 213

• Identify general objects

– Object and position recognition 2D/3D (see page 17 ff)

1.4 Measuring and Comparison 2D

For measuring 2D features in images, several approaches are available. In the Solution Guide III-B, section 3.1 on

page 27, a graph leads you from the speciﬁc features you want to measure and the appearance of the objects in the

image to the suitable measuring approach. Generally, the following approaches are common:

• Blob Analysis on page 35 for objects that consist of regions of similar gray value, color, or texture.

• Contour Processing on page 85 for objects that are represented by clear-cut edges. The contours can be

obtained by different means:

– Edge Filtering on page 59 if pixel precision is sufﬁcient.

– Edge and Line Extraction on page 67 if subpixel precision is needed.

• Matching on page 97 for objects that can be represented by a template. Matching comprises different ap-

proaches. For detailed information about matching see the Solution Guide II-B.

• 1D Measuring on page 49 if you want to obtain the positions, distances, or angles of edges that are measured

along a line or an arc. More detailed information can be found in the Solution Guide III-A.

1.5 Measuring and Comparison 3D 17

1.5 Measuring and Comparison 3D

For measuring in 3D, the following approaches are available:

• The approaches used for measuring and comparison 2D (see page 16) in combination with a camera cali-

bration (see Solution Guide III-C, section 3.2 on page 59) for measuring objects that are viewed by a single

camera and that lie in a single plane.

• Pose estimation (Solution Guide III-C, chapter 4 on page 89) for the estimation of the poses of 3D objects

that are viewed by a single camera and for which knowledge about their 3D model (e.g., known points,

known circular or rectangular shape) is available.

3D reconstruction is an important subcategory of 3D measuring and comprises the following methods:

• Stereo for measuring in images obtained by a binocular or multi-view stereo system on page 239. Further

information can be found in the Solution Guide III-C, chapter 5 on page 115.

• Laser triangulation using the sheet-of-light technique (Solution Guide III-C, chapter 6 on page 145) for

measuring height proﬁles of an object by triangulating the camera view with a projected light line.

• Depth from focus (Solution Guide III-C, chapter 7 on page 161) for getting depth information from a se-

quence of images of the same object but with different focus positions.

• Photometric Stereo (Reference Manual, chapter “3D Reconstruction . Photometric Stereo”) for getting

information about an object’s shape because of its shading behavior (e.g., by the operator photomet-

ric_stereo).

1.6 Object Recognition 2D

For ﬁnding speciﬁc objects in images, various methods are available. Common approaches comprise:

• Blob Analysis on page 35 for objects that are represented by regions of similar gray value, color, or texture.

• Contour Processing on page 85 for objects that are represented by clear-cut edges. The contours can be

obtained by different means:

– Edge Filtering on page 59 if pixel precision is sufﬁcient.

– Edge and Line Extraction on page 67 if subpixel precision is needed.

• Matching on page 97 for objects that can be represented by a template. Matching comprises different ap-

proaches. For detailed information about matching see the Solution Guide II-B.

• Classiﬁcation on page 143 for the recognition of objects by a classiﬁcation using, e.g., Gaussian mixture

models, neural nets, or support vector machines. For more detailed information about classiﬁcation see the

Solution Guide II-D.

• Color Processing on page 157 for the recognition of objects that can be separated from the background by

their color.

• Texture Analysis on page 171 for the recognition of objects that can be separated from the background by

their speciﬁc texture.

• Movement detection (see section 1.13 on page 19) for the recognition of moving objects.

1.7 Object Recognition 3D

For the recognition of 3D objects that are described by a 3D Computer Aided Design (CAD) model, see the

descriptions for 3D Matching on page 121. For the recognition of planar objects that can be oriented arbitrarily

in the 3D space, see the descriptions for perspective, deformable matching and descriptor-based matching in the

chapter about Matching on page 97 for the uncalibrated case and in the Solution Guide III-C, chapter 4 on page 89

for the calibrated case.

Guide to Methods

18 Guide to HALCON Methods

1.8 Position Recognition 2D

In parts, the approaches for measuring 2D features are suitable to get the position of objects. In the Solution Guide

III-B, section 3.1 on page 27, a graph leads you from several speciﬁc features, amongst others the object position,

and the appearance of the objects in the image to the suitable approach. For position recognition, in particular the

following approaches are common:

• Blob Analysis on page 35 for objects that consist of regions of similar gray value, color, or texture.

• Contour Processing on page 85 for objects that are represented by clear-cut edges. The contours can be

obtained by different means:

– Edge Filtering on page 59 if pixel precision is sufﬁcient.

– Edge and Line Extraction on page 67 if subpixel precision is needed.

• Matching on page 97 for objects that can be represented by a template. Matching comprises different ap-

proaches. For detailed information about matching see the Solution Guide II-B.

• 1D Measuring on page 49 if you want to obtain the positions of edges that are measured along a line or an

arc. More detailed information can be found in the Solution Guide III-A.

1.9 Position Recognition 3D

For the position recognition of 3D objects, the following approaches are available:

• The approaches used for measuring and comparison 2D (see page 16) in combination with a camera cali-

bration (see Solution Guide III-C, section 3.2 on page 59) for measuring objects that are viewed by a single

camera, and which lie in a single plane.

• Pose estimation (Solution Guide III-C, chapter 4 on page 89) for the estimation of the poses of 3D objects

that are viewed by a single camera and for which knowledge about their 3D model (e.g., known points,

known circular or rectangular shape) is available.

• Stereo for measuring positions in images obtained by a binocular stereo system on page 239 or a multi-view

stereo system. Further information can be found in the Solution Guide III-C, chapter 5 on page 115.

• 3D matching on page 121 for objects that are searched for based on a 3D Computer Aided Design (CAD)

model.

1.10 Print Inspection

For print inspection, the suitable method depends on the type of print you need to inspect:

• Optical character veriﬁcation (Reference Manual, chapter “Inspection . OCV”) for the veriﬁcation of char-

acters.

• Data Code on page 205 for inspecting the quality of a printed 2D data code symbol. Further information can

be found in the Solution Guide II-C, section 6 on page 43.

• Variation Model on page 133, which compares images containing similar objects and returns the difference

between them considering a certain tolerance at the object’s border.

• Component-based matching (part of Matching on page 97), if the print is built by several components which

may vary in their relation to each other (orientation and distance) within a speciﬁed tolerance.

1.11 Quality Inspection 19

1.11 Quality Inspection

How to inspect the quality of an object depends on the features describing the quality. The following application

areas and the methods used by them can be applied for quality inspection:

• Surface inspection (see page 19) if, e.g., scratches in an object’s surface have to be detected.

• Completeness check (see page 16) if you need to check an object for missing parts.

• Measuring and comparison 2D/3D (see page 16 ff) if an object has to fulﬁll certain requirements related to

its area, position, orientation, dimension, or number of parts.

Additionally, Classiﬁcation on page 143 can be used to check the color or texture of objects. For more detailed

information about classiﬁcation see the Solution Guide II-D.

1.12 Robot Vision

For robot vision, you can combine the approaches for object and position recognition 2D/3D (see page 17 ff) with

the hand-eye calibration described in the Solution Guide III-C, chapter 8 on page 173.

1.13 Security System

For a sequence analysis or movement detection you can use, e.g., one of the following approaches:

• Blob Analysis on page 35, e.g., by using the operator dyn_threshold to obtain the difference between two

images and thus detect moving objects. The approach is fast, but detects objects only as long as they are

moving.

• Background estimator (Reference Manual, chapter “Tools . Background Estimator”) for the recognition of

moving objects even if they stop temporarily. It adapts to global changes concerning, e.g., illumination.

• Optical ﬂow (Reference Manual, chapter “Filters . Optical Flow”) for the recognition of moving objects

even if they stop temporarily. Because of complex calculations, it is slower than the other approaches, but

additionally returns the velocity for each object.

• Scene ﬂow (Reference Manual, chapter “Filters . Scene Flow”) for the recognition of moving objects in 3D

even if they stop temporarily. In contrast to the optical ﬂow, which returns the information about moving

objects only in 2D, the scene ﬂow calculates the position and motion in 3D. Because of complex calculations,

it is even slower than the optical ﬂow, but additionally returns 3D information.

• Kalman ﬁlter (Reference Manual, chapter “Tools . Kalman Filter”) can be applied after the recognition of

moving objects to predict the future positions of objects.

For the recognition of, e.g., irises or faces, Classiﬁcation on page 143 may be suitable. For more detailed informa-

tion about classiﬁcation see the Solution Guide II-D.

Examples solving different tasks relevant for security systems can be found via the Open Example dialog inside

HDevelop for the category Industry/Surveillance and Security.

1.14 Surface Inspection

For surface inspection, several approaches are available:

• Via comparison with a reference image

Guide to Methods

20 Guide to HALCON Methods

– Variation Model on page 133, which compares images containing similar objects and returns the differ-

ence between them. This is suited especially if you need to detect irregularities that are placed inside

the object area, whereas small irregularities at the object’s border can be tolerated.

• Via comparison with a reference pattern or color

– Texture Analysis on page 171

– Color Processing on page 157

– Classiﬁcation on page 143 (for more detailed information about classiﬁcation see the Solution Guide

II-D)

• Via defect description for uniformly structured surfaces

– Blob Analysis on page 35 for objects that consist of regions of similar gray value, color, or texture.

– Contour Processing on page 85 for objects that are represented by clear-cut edges. The contours can

be obtained by an Edge Filtering on page 59 if pixel precision is sufﬁcient, or by an Edge and Line

Extraction on page 67 if subpixel precision is needed.

If a single image is not suited to cover the object to inspect, several approaches exist to combine images after their

acquisition:

• Calibrated mosaicking (Solution Guide III-C, chapter 9 on page 189) for a high-precision mosaicking of a

discrete number of overlapping images obtained by two ore more cameras.

• Uncalibrated mosaicking (Solution Guide III-C, chapter 10 on page 199) for a less precise mosaicking of a

discrete number of overlapping images obtained as an image sequence.

• Combination of lines obtained by a line scan camera to so-called pages (Solution Guide II-A, section 6.6 on

page 39) for inspecting continuous material on an assembly line.

1.15 Texture Inspection

Common approaches for texture inspection are:

• Fast Fourier Transformation (see section 15.6 on page 181)

• Classiﬁcation on page 143 (for more detailed information about classiﬁcation see the Solution Guide II-D)

• Texture Analysis on page 171

1.16 Text Processing

This chapter describes how HALCON processes text. Although HALCON is mainly an image processing library,

there is also some text processing in HALCON as the following examples demonstrate:

• Passing string parameters such as ﬁle names or data code contents to and from HALCON operators via

programming language interfaces.

• Files that are opened, read or written with HALCON operators.

• Strings that are transmitted via sockets.

• Reading characters with the help of OCR classiﬁers, which make use of user deﬁned class names.

• String processing with the help of HALCON’s tuple operators, e.g., by means of regular expressions.

剩余278页未读，继续阅读

笑笑子

粉丝: 3
资源: 3

Halcon 18.11图像处理指南：22章详解一维测量与blob分析

HALCON手册简体中文版.pdf

如何往sonirock_git@192.168.1.149:~/audio_solution_c90.git上面发送代码

Cannot find reference '' in '__init__.py'

selfie_segmentation_solution_simd_wasm_bin.js

改写solution_02函数。 函数的参数不固定，请返回所有参数的和： 例如： 调用：let ret = solution_02(1,2,3); 结果：ret = 6 调用：let ret = solution_02(1,2,3,4); 结果：ret = 10

最新资源

Cannot find reference '' in 'init.py'

改写solution_02函数。函数的参数不固定，请返回所有参数的和：例如：调用：let ret = solution_02(1,2,3); 结果：ret = 6 调用：let ret = solution_02(1,2,3,4); 结果：ret = 10