CN111127384A - Strong reflection workpiece vision measurement method based on polarization imaging - Google Patents

Strong reflection workpiece vision measurement method based on polarization imaging Download PDF

Info

Publication number
CN111127384A
CN111127384A CN201910396592.6A CN201910396592A CN111127384A CN 111127384 A CN111127384 A CN 111127384A CN 201910396592 A CN201910396592 A CN 201910396592A CN 111127384 A CN111127384 A CN 111127384A
Authority
CN
China
Prior art keywords
image
highlight
polarization
pixel
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
CN201910396592.6A
Other languages
Chinese (zh)
Inventor
祝振敏
王心韵
裴爽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
East China Jiaotong University
Original Assignee
East China Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by East China Jiaotong University filed Critical East China Jiaotong University
Publication of CN111127384A publication Critical patent/CN111127384A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B11/00—Measuring arrangements characterised by the use of optical techniques
    • G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Quality & Reliability (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Processing (AREA)

Abstract

The invention provides a strong reflection workpiece vision measurement method based on polarization imaging. Aiming at the influence of the loss of image texture details and color information of a high-light area in an image on the stereoscopic vision non-contact measurement precision, a pixel value model of the high-light area of the image is established according to the reflection characteristic of the high-light area and the polarization imaging principle in application physics. The two-dimensional images of the target object corresponding to different polarization angles are obtained by installing the polarization lens on the lens of the CCD camera and rotating different polarization angles, and a BP neural network is adopted to carry out model building on the pixel value of the image highlight area, so that the pixel value of the image highlight area in the model is effectively reduced, the effective detection and elimination of the image highlight area are realized, the pixel distribution of the corresponding image line is smooth, and the high-light area pixels are well limited. The research of the strong reflection workpiece vision measurement method based on polarization imaging eliminates the high-brightness area of the image part, can be used for a stereoscopic vision measurement system, and has good application prospect in the fields of industrial manufacturing technology and the like which need high-precision camera calibration under the natural light condition.

Description

Strong reflection workpiece vision measurement method based on polarization imaging
Technical Field
The invention belongs to the field of measurement technology and machine vision, and particularly relates to a strong-reflection workpiece vision measurement method based on polarization imaging.
Background
The rapid development of modern industrialization puts forward the requirement of high precision on the real-time online detection of the processing machine workpiece of a production line, and how to rapidly and accurately carry out non-contact measurement on a measured space object is always a research hotspot in the field of machine vision. The traditional contact type measuring device cannot meet the requirements of diversity and surface complexity of processing and producing workpieces, a stereoscopic vision non-contact type measuring system acquires images of space objects at different visual angles, extracts pixel coordinates of two-dimensional characteristic points, converts the pixel coordinates into a corresponding three-dimensional space coordinate system, then carries out three-dimensional information reconstruction on the measured space objects, realizes non-contact type measurement of object sizes, and improves high-efficiency production of an industrial processing assembly line and guarantees product production safety. Under the influence of the change of natural illumination conditions, when the image information of the space object is collected, the inherent color and texture characteristics of the object are hidden by highlight, and certain interference is caused to the subsequent stereoscopic vision information reconstruction. The stereoscopic vision non-contact measurement technology based on polarization information is researched, and the purpose is to improve the non-contact measurement precision of the space dimension of a target object.
Disclosure of Invention
The invention aims to overcome a plurality of defects and restrictions in the prior art, and designs a strong-reflection workpiece vision measurement method based on polarization imaging to realize the detection and elimination of an image highlight area. The invention is realized by the following technical scheme:
a strong reflection workpiece vision measurement method based on polarization imaging comprises the following steps:
(1) and establishing an image highlight removing system based on polarization information.
(2) The highlight area of the image is obtained by processing a color image containing the highlight area.
(3) An image highlight removing method based on a polarization device for collecting target object information is provided.
A strong reflection workpiece vision measurement method based on polarization imaging comprises the following steps:
and the CCD camera is used for shooting images in a vision system. The corresponding maximum resolution/frame rate is: 1624 × 1224@29fps, pixel size: 4.4 μm. The linear polarizer PL is used to change the light balance in the photograph. The experiment platform is convenient for the CCD camera to acquire images. The high-reflectivity metal workpiece is used for acquiring a highlight image under a natural illumination condition. Image acquisition software and a PC host;
according to the invention, an image highlight region detection and elimination system based on polarization information is set up, target object images under different polarization angles are collected, dark channel primary color images and a Mean-Shift segmentation algorithm are adopted to detect highlight regions, a BP neural network model is set up, highlight region pixel sets in images corresponding to different polarization angles are used as input parameters, Gaussian distribution functions are adopted to process highlight region pixel points and normalize the highlight region pixel points to a diffuse reflection pixel region, and experimental results show that image pixel distribution under the highlight detection elimination model provided by the invention is smooth and the highlight region pixels are well limited.
The strong reflection workpiece vision measurement method based on polarization imaging is compiled into calculation and determined through the following steps:
1. after the two-dimensional image information of the target object is acquired in multiple angles based on the stereoscopic vision measuring system, the invention provides an image highlight detection method based on a dark channel prior algorithm in an objective sense. The dark channel prior algorithm solves the minimum value of a certain pixel in a color image in the RGB three channels.
Pixel values of a certain Pixel point in R, G, B three channels in a color image are respectively set as Pixel _ R, Pixel _ G and Pixel _ B, a dark and black channel prior algorithm firstly compares the sizes of the three Pixel values and then selects the minimum Pixel value to carry out local area minimum filtering, and the mathematical model of the corresponding dark primary color of the image is obtained and expressed as:
Figure BDA0002057327100000021
in the formula (1), JdarkFor dark primary images, the corresponding image pixel intensity value is very low and approaches 0, i.e. Jdark→0，JcFor the R, G, B channel image, x is the image corresponding pixel point location, and Ω (x) represents the local area module centered at x.
Compared with the originally acquired color image, the image obtained by white balance algorithm processing has the inherent color characteristics of the object in visual effect, and the corresponding color histogram characteristics are distributed uniformly, so that the method is more suitable for subsequent image highlight area detection. In order to better perform statistics on highlight area pixels in a color image, a Mean-Shift algorithm is introduced to perform clustering processing on the image so as to highlight the highlight area existing in the image. The principle of the Mean-Shift algorithm lies in solving a local optimal solution of probability density, pixel point distribution of an image corresponds to a probability density function f (x), and the estimation of a Parzen window corresponding to n sampling points xi, i ═ 1,2,3, …, n, f (x) in a known d-dimensional space is as follows:
Figure BDA0002057327100000022
where K (x) is a kernel function, ω (x)i) Equal to or greater than 0 corresponds to the weight of sample point xi, i equal to 1,2,3, …, n. Gradient of probability density function f (x)
Figure BDA0002057327100000023
Comprises the following steps:
Figure BDA0002057327100000024
let g (x) -k' (x), g (x) g (| | x | | | non-calculation phosphor)2) Then, there are:
Figure BDA0002057327100000025
order to
Figure BDA0002057327100000031
I.e., the Mean-Shift vector, which is in direct proportion to the probability density gradient.
The following modifications are made to N, namely:
Figure BDA0002057327100000032
then there is Mh(x)＝mh(x) -x, which is related to the gradient
Figure BDA0002057327100000035
The process of rising is consistent.
2. On the premise of effectively detecting the highlight area of the image, the invention provides an image highlight removing method based on the information of a target object acquired by a polarizing device. This patent uses image information under the different polarization angles as the research focus, extracts the image highlight region pixel under the different polarization angles of correspondence, makes the pixel value that corresponds the highlight region pixel effectively reduce with the gaussian distribution function form through adopting BP neural network to build the model, and the image highlight region pixel value that will be based on gaussian distribution reduces the model and is applied to the target object image under the optimum polarization angle at last, can weaken the highlight region that exists in the image deeply.
1) Rotating a polarized lens in front of a lens of a CCD camera, acquiring an optimal target object polarization information image (the brightness of the surface of the corresponding object imaging surface is minimum) within a range of rotating 360 degrees, sequentially rotating the polarized lens for n times and acquiring the target object polarization information image under different polarization angles theta i (i is 1,2,3, …, n), applying the highlight region detection algorithm provided by the invention to an image set, extracting a corresponding highlight region pixel value, and recording the pixel value as:
Figure BDA0002057327100000033
2) similarly, the highlight region detection algorithm provided by the invention is applied to the optimal polarized image, corresponding highlight pixel values are extracted, and the set P and the matrix Q are normalized, so that the matrix of the pixel values corresponding to the highlight region pixels of the polarized image collected by the CCD camera is as follows:
Figure BDA0002057327100000034
3) recording a high light region pixel value interval corresponding to the polarization image set and the optimal polarization image, generating random numbers obeying Gaussian distribution in the corresponding interval, and recording the dimension of a random number matrix Y as follows, wherein the dimension of the random number matrix Y is consistent with that of the matrix X:
Figure BDA0002057327100000041
y is an (N +1) xN matrix;
4) the BP neural network model is divided into three layers, namely an input layer, a hidden layer and an output layer, firstly, input data is a set and corresponds to the input layer of the BP neural network, the input data depends on a random connection weight value between the input layer and the hidden layer as the input of the hidden layer, and then reaches the output layer through the random connection weight value between the hidden layer and the output layer, the BP neural network is a forward propagation process of the BP neural network, a data error generated in the process is defined as a certain partial derivative function, the connection weight value between the layers is adjusted in the backward propagation process of the BP neural network, so that a global error coefficient is minimum, and the convergence of the BP neural network model is improved. And taking the first n row vectors of the matrix X as input data of the BP neural network, and taking the first n row vectors of the matrix Y as output data of the BP neural network, and building a neural network model and training for multiple times to obtain the optimal net. And taking the n +1 th row vector of the matrix X as input data of the BP neural network, and obtaining corresponding output data PixelBP in net. And replacing the matrix Q by using a matrix formed by the first m data of the vector PixelBP, namely, replacing high-light area pixels in the Itarget image to realize a deep highlight removing effect, extracting pixel values of a certain random row in the image after highlight removal, and checking whether the pixel values corresponding to the high-light area pixels are effectively reduced.
Drawings
FIG. 1 is a diagram illustrating the effect of detecting highlight region of an image based on polarization information according to the present invention
FIG. 2 is a diagram of the present invention for eliminating the effect of high light in an image based on polarization information
FIG. 3 is a comparison graph of pixel distribution of highlight image and no-highlight image based on polarization information
Detailed Description
The following drawings
The invention provides a strong reflection workpiece vision measurement method based on polarization imaging, which is applied to a stereoscopic vision non-contact measurement technology of polarization information and aims to improve the space dimension non-contact measurement precision of a target object.
The following detailed description of specific embodiments of the invention is provided in connection with the accompanying drawings.
The first embodiment is as follows: the method comprises the steps of building an image highlight region detection and elimination system based on polarization information, collecting target object images under different polarization angles, detecting highlight regions by adopting a dark channel primary color image and a Mean-Shift segmentation algorithm, building a BP neural network model, taking highlight region pixel sets in images corresponding to different polarization angles as input parameters, processing highlight region pixel points by adopting a Gaussian distribution function and normalizing to a diffuse reflection pixel interval, and indicating that image pixel distribution under the highlight detection elimination model provided by the invention is smooth and the highlight region pixels are well limited.
After the highlight pixel interval of the image is extracted, the pixel value of the pixel point of the original color image falling in the interval is 1, the pixel value of the pixel point falling outside the interval is 0, and the highlight detection effect graph of the image is shown in figure 1. On the premise of effectively detecting the highlight area of the image, the invention provides an image highlight removing method for collecting target object information based on a polarizing device. Because target object images under a plurality of polarization angles need to be shot and high-light area pixel points corresponding to different polarization images are extracted, the BP neural network model is introduced to carry out optimization training on a high-dimensional high-light area pixel matrix, so that the high-dimensional high-light area pixel matrix is effectively reduced in a Gaussian distribution function form, and further deeper removal of the high-light area of the image and image quality optimization are realized.
Fig. 2 (a) shows the target object image with the highlight region removed, and (b) shows the pixel distribution of the corresponding image line. In this document, pixel values of a corresponding line in the image are randomly extracted, a certain degree of improvement in the image quality after highlight removal can be observed, and in addition, the highlight image is compared with the image after highlight areas are removed based on polarization information, and the pixel values of a certain line in the image are randomly extracted for comparison, and a comparison graph is shown in fig. 3. The pixel distribution of four random rows of pixels in the image is shown in fig. 3, and it can be seen that: the image pixel distribution after the highlight region is removed is smooth, and the bright pixel corresponding to the highlight region is improved, so that the feasibility and the practicability of the image highlight removal algorithm based on the polarization information provided by the invention can be verified.

Claims (1)

1. A strong reflection workpiece vision measurement method based on polarization imaging is characterized by comprising the following steps:
(1) an image highlight removing system based on polarization information is established, and the image highlight removing system based on a polarization device mainly comprises the following parts: the CCD camera has the following corresponding maximum resolution/frame frequency: 1624 × 1224@29fps, pixel size: 4.4 mu m, a linear polaroid PL, an experimental platform, a high-reflectivity metal workpiece, image acquisition software and a PC host;
(2) and converting a color image containing a highlight area into an HSV color space after white balance processing, further obtaining a highlight image on a V channel, and analyzing pixel points of the highlight image to obtain a brightness distribution map corresponding to the highlight image. An image white balance processing algorithm based on a dynamic threshold is adopted, and the algorithm carries out two steps of white point detection and white point adjustment on a specified image. In order to better perform statistics on highlight area pixels in a color image, a Mean-Shift algorithm is introduced to perform clustering processing on the image so as to highlight the highlight area existing in the image.
The principle of the Mean-Shift algorithm lies in solving the local optimal solution of probability density, the pixel point distribution of an image corresponds to the probability density function f (x), and n samples in the known d-dimensional spaceSample point xiThe Parzen window corresponding to i ═ 1,2,3, …, n, f (x) is estimated as:
Figure FDA0002057327090000011
(3) on the premise of effectively detecting the highlight region of the image, an image highlight removing method based on the information of a target object collected by a polarizing device is provided. Taking image information under different polarization angles as research focus, extracting image highlight region pixels corresponding to different polarization angles, adopting a BP neural network building model to effectively reduce pixel values of the corresponding highlight region pixels in a Gaussian distribution function form, and finally applying the image highlight region pixel value reduction model based on Gaussian distribution to a target object image under an optimal polarization angle, so that highlight regions existing in the image can be deeply weakened, and the calculation and the determination are carried out through the following steps:
1) the polarization lens in front of the CCD camera lens is rotated, an optimal target object polarization information image (the brightness of the surface of the corresponding object imaging surface is minimum) is acquired within the range of rotating 360 degrees, the polarization lens is rotated for n times in sequence, the target object polarization information images under different polarization angles are acquired, and the highlight region detection algorithm provided by the invention is applied to an image set and corresponding highlight region pixel values are extracted.
2) The highlight region detection algorithm provided by the invention is applied to the optimal polarization image, corresponding highlight pixel values are extracted, and the set P and the matrix Q are normalized.
3) And recording a highlight region pixel value interval corresponding to the polarization image set and the optimal polarization image, and generating random numbers obeying Gaussian distribution in the corresponding interval, wherein the dimensionality of a random number matrix Y is consistent with that of the X matrix.
4) And taking the first n row vectors of the matrix X as input data of the BP neural network, and taking the first n row vectors of the matrix Y as output data of the BP neural network, and building a neural network model and training for multiple times to obtain the optimal net. And taking the n +1 th row vector of the matrix X as input data of the BP neural network, and obtaining corresponding output data PixelBP in net. And replacing the matrix Q by using a matrix formed by the first m data of the vector PixelBP, namely, replacing high-light area pixels in the Itarget image to realize a deep highlight removing effect, extracting pixel values of a certain random row in the image after highlight removal, and checking whether the pixel values corresponding to the high-light area pixels are effectively reduced.
CN201910396592.6A 2018-10-15 2019-05-13 Strong reflection workpiece vision measurement method based on polarization imaging Withdrawn CN111127384A (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2018111967524 2018-10-15
CN201811196752 2018-10-15

Publications (1)

Publication Number Publication Date
CN111127384A true CN111127384A (en) 2020-05-08

Family

ID=70495674

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910396592.6A Withdrawn CN111127384A (en) 2018-10-15 2019-05-13 Strong reflection workpiece vision measurement method based on polarization imaging

Country Status (1)

Country Link
CN (1) CN111127384A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112069974A (en) * 2020-09-02 2020-12-11 安徽铜峰电子股份有限公司 Image recognition method and system for recognizing defects of components
CN112557413A (en) * 2020-11-24 2021-03-26 中国科学院西安光学精密机械研究所 Photovoltaic cell panel subfissure detection camera and detection method
CN112672008A (en) * 2020-12-14 2021-04-16 杭州联吉技术有限公司 Lens adjusting method and device
CN112785491A (en) * 2021-01-20 2021-05-11 北京航空航天大学 Image highlight reflection separation calculation method based on polarization guidance

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108596849A (en) * 2018-04-23 2018-09-28 南京邮电大学 A kind of single image to the fog method based on sky areas segmentation
CN113554575A (en) * 2020-04-23 2021-10-26 华东交通大学 High-reflection object surface highlight removing method based on polarization principle

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108596849A (en) * 2018-04-23 2018-09-28 南京邮电大学 A kind of single image to the fog method based on sky areas segmentation
CN113554575A (en) * 2020-04-23 2021-10-26 华东交通大学 High-reflection object surface highlight removing method based on polarization principle

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
祝振敏 等: "《基于偏振信息的强反射工件高光去除及视觉测量方法研究》", 《HTTP://KNS.CNKI.NET/KCMS/DETAIL/31.1252.O4.20180706.1608.002.HTML》 *
陈伟 等: "《基于机载LiDAR和Mean Shift算法提取森林密度》", 《西南交通大学学报》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112069974A (en) * 2020-09-02 2020-12-11 安徽铜峰电子股份有限公司 Image recognition method and system for recognizing defects of components
CN112069974B (en) * 2020-09-02 2023-04-18 安徽铜峰电子股份有限公司 Image recognition method and system for recognizing defects of components
CN112557413A (en) * 2020-11-24 2021-03-26 中国科学院西安光学精密机械研究所 Photovoltaic cell panel subfissure detection camera and detection method
CN112672008A (en) * 2020-12-14 2021-04-16 杭州联吉技术有限公司 Lens adjusting method and device
CN112672008B (en) * 2020-12-14 2022-09-06 杭州联吉技术有限公司 Lens adjusting method and device
CN112785491A (en) * 2021-01-20 2021-05-11 北京航空航天大学 Image highlight reflection separation calculation method based on polarization guidance
CN112785491B (en) * 2021-01-20 2022-10-04 北京航空航天大学 Image highlight reflection separation calculation method based on polarization guidance

Similar Documents

Publication Publication Date Title
CN108549873B (en) Three-dimensional face recognition method and three-dimensional face recognition system
CN111127384A (en) Strong reflection workpiece vision measurement method based on polarization imaging
CN104850850B (en) A kind of binocular stereo vision image characteristic extracting method of combination shape and color
CN111640157B (en) Checkerboard corner detection method based on neural network and application thereof
CN109086675B (en) Face recognition and attack detection method and device based on light field imaging technology
CN104597057B (en) A kind of column Diode facets defect detecting device based on machine vision
CN111461036B (en) Real-time pedestrian detection method using background modeling to enhance data
CN112561899A (en) Electric power inspection image identification method
TWI745204B (en) High-efficiency LiDAR object detection method based on deep learning
CN111027415A (en) Vehicle detection method based on polarization image
Liu et al. Iterating tensor voting: A perceptual grouping approach for crack detection on EL images
CN114170165A (en) Chip surface defect detection method and device
CN114387262A (en) Nut positioning detection method, device and system based on machine vision
CN113628170A (en) Laser line extraction method and system based on deep learning
CN112102379B (en) Unmanned aerial vehicle multispectral image registration method
CN106355576A (en) SAR image registration method based on MRF image segmentation algorithm
CN115937021A (en) Polarization defogging method based on frequency domain feature separation and iterative optimization of atmospheric light
CN115170832A (en) Weak texture surface microstructure feature extraction method based on visible light single image
CN115760549A (en) Processing method for flattening 3D data of curved surface
CN110956640B (en) Heterogeneous image edge point detection and registration method
CN110473246B (en) Distance measurement method of multiple shielding targets based on binocular vision
Arunkumar et al. Estimation of vehicle distance based on feature points using monocular vision
Zhao et al. Steel plate surface defect recognition method based on depth information
Xu et al. A Few-shot Learning Method for the Defect Inspection of Lithium Battery Sealing Nails
Bi et al. Multi-object Paper Money Recognition Technology Based on AlexNet Model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WW01 Invention patent application withdrawn after publication

Application publication date: 20200508

WW01 Invention patent application withdrawn after publication