WO2023231262A1 - 基于视觉振频识别的提升钢丝绳张力检测方法 - Google Patents

基于视觉振频识别的提升钢丝绳张力检测方法 Download PDF

Info

Publication number
WO2023231262A1
WO2023231262A1 PCT/CN2022/123341 CN2022123341W WO2023231262A1 WO 2023231262 A1 WO2023231262 A1 WO 2023231262A1 CN 2022123341 W CN2022123341 W CN 2022123341W WO 2023231262 A1 WO2023231262 A1 WO 2023231262A1
Authority
WO
WIPO (PCT)
Prior art keywords
wire rope
data
rectangle
array
formula
Prior art date
Application number
PCT/CN2022/123341
Other languages
English (en)
French (fr)
Inventor
彭玉兴
朱真才
杜庆永
常向东
胡长华
曹国华
周公博
卢昊
唐玮
任一男
Original Assignee
中国矿业大学
徐州煤矿安全设备制造有限公司
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 中国矿业大学, 徐州煤矿安全设备制造有限公司 filed Critical 中国矿业大学
Publication of WO2023231262A1 publication Critical patent/WO2023231262A1/zh

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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
    • G01L5/00—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes
    • G01L5/04—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes for measuring tension in flexible members, e.g. ropes, cables, wires, threads, belts or bands
    • G01L5/042—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes for measuring tension in flexible members, e.g. ropes, cables, wires, threads, belts or bands by measuring vibrational characteristics of the flexible member
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]

Definitions

  • the invention relates to the field of hoisting wire rope tension detection based on visual vibration frequency recognition, and specifically relates to a hoisting wire rope tension detection method based on visual vibration frequency recognition.
  • Machine vision technology provides a new method to solve this problem, that is, using a computer to analyze the vibration image of the wire rope to be tested to obtain its vibration parameters, and then calculate the value of the tension on the wire rope.
  • Existing detection algorithms based on machine vision have poor anti-interference and are sensitive to abnormal phenomena such as afterimages and shadows.
  • existing methods for cleaning dead pixels in wire rope vibration data also have limitations and cannot correctly screen out and correct sample data. The hysteresis error in the detection results affects the accuracy of the final detection results.
  • the present invention aims to solve the above problems and provide a lifting wire rope tension detection method based on visual vibration frequency recognition, which overcomes the problems caused by the above traditional detection method that the equipment is not easy to install and is easy to affect the wire rope to be tested, and improves the efficiency of detection. It also improves the anti-interference performance of the wire rope image recognition algorithm and improves the accuracy of tension detection.
  • the lifting wire rope tension detection method based on visual vibration frequency identification includes the following steps:
  • Step A collect the vibration image of the lifting wire rope in real time, preprocess the image, and obtain a black and white image with good separation effect between the wire rope and the background;
  • Step B Perform edge feature extraction and contour fitting based on the tilted rectangle method on the black and white image segmentation results obtained in step A.
  • the specific steps are as follows:
  • B2 perform inclined rectangular fitting on the detected edges to eliminate shooting afterimage errors caused by problems such as the wire rope moving quickly or the image acquisition equipment failing to meet the requirements; also includes:
  • Step C Perform feature screening and geometric centroid positioning on the inclined rectangle fitted in step B to determine the time domain vibration data of the wire rope;
  • Step D further process the time-domain vibration data of the steel wire rope obtained in step C, and use the absolute mean method based on difference theory to filter the bad data points detected by the algorithm under abnormal working conditions to improve the accuracy of the data. Specifically, it includes the following sub-steps :
  • D1 determine the original array a o to be processed, and calculate its difference array a d ;
  • Step E perform Fourier transform on the cleaned data to obtain its vibration natural frequency, and calculate the tension of the steel wire rope based on the steel wire rope tension-vibration frequency formula.
  • the abnormal working conditions include strong wind, lighting failure and human interference.
  • step A specifically includes the following sub-steps:
  • A1 select the ROI area based on the active area of the wire rope in the image to reduce the total calculation amount of the algorithm
  • A2 perform grayscale conversion on the ROI area
  • step A3 perform adaptive threshold segmentation on the ROI grayscale image obtained in step A2;
  • step A4 perform Gaussian filtering operation on the black and white image obtained in step A3 to reduce background interference;
  • step A5 perform morphological processing on the filtering results obtained in step A4 to fill the internal holes in the wire rope target.
  • step C specifically includes the following sub-steps:
  • step B2 define the three characteristic parameters of the tilted rectangle in step B2: compactness, vertex height, and aspect ratio;
  • C2 combine the cutting logic of the ROI area to define the confidence intervals of the three characteristic parameters in C1, and filter out the undesired target fitting rectangles outside the intervals to eliminate interference caused by background shadows, oil stains or scratches;
  • step C3 Determine the geometric centroid of the target rectangle screened in step C2 as the position of the current frame of the steel wire rope, and then finally determine the time domain vibration data of the steel wire rope by collecting the centroid frame by frame.
  • step B1 includes the following sub-steps:
  • step B11 add white edges with a width of two pixels to the upper and lower edges of the black and white image obtained in step A, so that there is a closed outline in the image that can completely cover all edges of the wire rope target;
  • Step B2 includes the following sub-steps:
  • step B21 For the target edge extracted in step B1, fit the outline using a tilted circumscribed rectangle.
  • the fitting principle of the inclined rectangle is: select the point with the largest y value on the edge of the wire rope as the lower vertex of the circumscribed rectangle, Point 0, and draw two rays L 1 and L 2 from it as the vertex in the direction parallel to the x-axis. 0 as the center, rotate L 1 clockwise, and stop when L 1 touches the edge of the wire rope to be measured. Point 1, the intersection point of the edge and L 1 at this time, is used as the second vertex of the rectangle. This determines the first side of the rectangle. .
  • step C1 is as follows:
  • a is the number of pixels occupied by the length of the inclined rectangle, and b is the number of pixels occupied by the width of the inclined rectangle;
  • S 1 is the total number of pixels occupied by the tilted rectangle
  • S 2 is the total number of pixels in the ROI area
  • h 1 is the ordinate value of the highest vertex of the tilted rectangle
  • h 2 is the number of height pixels in the ROI area
  • step C2 The specific steps of step C2 are as follows:
  • the confidence interval Q2 of the compactness F2 of the tilted rectangle is [0.05,0.7]
  • the confidence interval Q3 of the vertex height F3 of the tilted rectangle is [0,0.1]; if the characteristic parameters of the tilted fitting rectangle are outside the confidence interval Then it will be excluded as interference contour;
  • step C3 The specific steps of step C3 are as follows:
  • k 1 is the ordinate of the upper left corner vertex when intercepting the ROI area
  • x P1 is the abscissa value of centroid P 1 ,
  • x P2 is the abscissa value of tracking point P 2 .
  • y P1 is the ordinate value of the centroid P 1 .
  • y P2 is the ordinate value of tracking point P 2 ;
  • step C32 initialize an original data array a 0 , add the tracking points P 2 obtained in step C31 in the order of image frames one by one to the original data array a 0 , and complete the information recording of the tracking target within one cycle for subsequent processing. data processing.
  • step D1 is as follows:
  • P d i represents the value of the element numbered i in the difference array a d .
  • P o i represents the value of the element with serial number i in the original array a o ,
  • n is the number of elements in the difference array a d ;
  • step D2 The specific steps of step D2 are as follows:
  • the filtering threshold T is defined as:
  • k is an empirical coefficient, taking a value between 4 and 5. According to the degree of dispersion of different parts of the data, different k values are selected for different wire rope vibration conditions.
  • the degree of dispersion of the data is determined by the standard deviation ⁇ of the original data before sorting by size.
  • ⁇ the standard deviation
  • the calculation method of each element value in a e is as follows The formula is determined:
  • Pe i represents the value of the element numbered i in the residual array a e ;
  • step D3 The specific steps of step D3 are as follows:
  • P o m is considered to be a bad point in the data, and P o (m+1) is judged in the same way;
  • the correction method is:
  • P o (w-1) is the element value with serial number w-1 in the array a o ,
  • P d (w-1) is the element value with serial number w-1 in array a d ;
  • the present invention adopts the above technical solution and has the following technical effects:
  • the present invention can effectively reduce detection errors caused by shooting afterimages. This is because the present invention uses an inclined rectangular fitting method to perform contour fitting and positioning of the steel wire rope without involving the area where the wire rope afterimages are generated in the image. Complete the positioning of the wire rope.
  • the present invention avoids false detection caused by oil stains, scratches in the detection background, or background shadows caused by uneven lighting. This is because the present invention filters images through the characteristic parameters of tilted rectangular fitting. The background interference targets are filtered out, and only the target outline of the wire rope to be measured is retained.
  • the present invention adopts the absolute mean filtering method based on the difference theory.
  • the advantages of this method are: first, the method is based on the fluctuation of the wire rope vibration data. Different filtering thresholds are set to different degrees, which has a better filtering effect according to the different fluctuation degrees of the wire rope vibration data in different service stages; then, this method can filter the absolute value of the wire rope vibration data and its adjacent points based on the difference method based on the idea of the difference method. , but the potential bad pixels are smaller in absolute value. Therefore, the bad pixel filtering method of the present invention can effectively identify the diffusive, concealed and hysteretic errors in the wire rope vibration data, and has better applicability for the non-normal distribution and strong randomness of the wire rope vibration data.
  • Figure 1 is a flow chart of the algorithm of the present invention.
  • Figure 2 is the basic principle diagram of the tilted rectangle fitting method.
  • Figure 3 is a comparison chart of the detection effects of the traditional fitting algorithm and the tilted rectangular fitting method when there is a shooting afterimage and when there is no shooting afterimage.
  • Figure 4 is a picture of a wire rope with shadows and an image segmentation effect diagram.
  • Figure 5 is a diagram showing the misdetection phenomenon of the traditional algorithm and the screening results of the algorithm of the present invention.
  • Figure 6 is a flow chart of the absolute mean correction method based on difference theory.
  • the present invention provides a hoisting wire rope tension detection method based on visual vibration frequency recognition, including image acquisition and preprocessing, target edge extraction and inclined rectangle fitting, feature screening and centroid positioning, data processing and results
  • visual vibration frequency recognition including image acquisition and preprocessing, target edge extraction and inclined rectangle fitting, feature screening and centroid positioning, data processing and results
  • A1 select the ROI area based on the active area of the wire rope in the image to reduce the total calculation amount of the algorithm
  • A2 perform grayscale conversion on the ROI area
  • step A3 perform adaptive threshold segmentation on the ROI grayscale image obtained in step A2;
  • step A4 perform Gaussian filtering operation on the black and white image obtained in step A3 to reduce background interference;
  • step A5 perform morphological processing on the filtering results obtained in step A4 to fill the internal holes in the wire rope target;
  • Step B Perform edge feature extraction and contour fitting based on the tilted rectangle method on the image segmentation results obtained in step A.
  • the specific steps are as follows:
  • Step C Perform feature screening and geometric centroid positioning on the tilted rectangle fitted in step B, which specifically includes the following steps:
  • step B2 define the three characteristic parameters of the tilted rectangle in step B2: compactness, vertex height, and aspect ratio;
  • C2 combine the cutting logic of the ROI area to define the confidence intervals of the three characteristic parameters in C1, and filter out the undesired target fitting rectangles outside the intervals to eliminate interference caused by background shadows, oil stains or scratches;
  • step C3 Determine the geometric centroid of the target rectangle screened in step C2 as the position of the current frame of the steel wire rope, and then finally determine the lateral vibration displacement curve of the steel wire rope by collecting the centroid frame by frame.
  • Step D perform data processing on the transverse vibration parameters of the steel wire rope obtained in step D, and calculate the tension of the steel wire rope, which specifically includes the following steps:
  • Step D further process the detected data, and use the absolute mean method based on the difference theory to filter the bad pixels of the data detected under abnormal working conditions (strong wind, lighting failure, human interference, etc.) to improve the accuracy of the data.
  • abnormal working conditions strong wind, lighting failure, human interference, etc.
  • D1 determine the original array a o to be processed, and calculate its difference array a d ;
  • Step E perform further data processing on the cleaned data to obtain the wire rope tension.
  • the specific steps are as follows:
  • E1 perform Fourier transform on the data to obtain the natural frequency of wire rope vibration
  • step A2 The specific steps of step A2 are:
  • R i , G i , and B i are the values of the red, green, and blue channels of the i-th pixel of the three-channel color image. Then, assign the grayscale value to this pixel and traverse the entire image to complete the grayscale image conversion.
  • step A3 The specific steps of step A3 are:
  • A31 Calculate the grayscale histogram of the current picture input image to determine a mean M. Use the M value as the dividing line to divide the histogram into two parts, MA and MB, and define the number of pixels occupied by the MA part as PA and the pixels in the MB part. The number is PB, defined by the between-class variance:
  • ICV PA ⁇ (MA-M) 2 +PB ⁇ (MB-M) 2
  • ICV is the inter-class variance
  • the threshold M that achieves the maximum value is the optimal threshold
  • threshold M As the dividing line to segment the image into two parts: target and background.
  • step A4 The specific steps of step A4 are:
  • ⁇ is the Gaussian distribution parameter, which determines the width of the Gaussian function.
  • the present invention uses a two-dimensional discrete function to smooth filter the image. First, the image is convolved with the above-mentioned one-dimensional Gaussian function, and then convolved with a one-dimensional Gaussian function of the same direction but perpendicular to the convolution result to obtain a two-dimensional Gaussian function:
  • A42 obtains the weights at discrete points by discretizing the two-dimensional Gaussian function. For example, to obtain a filter template with a size of (2n+1) ⁇ (2n+1), use the pixel at the center of the template as The central element is used to create a matrix A[2n+1,2n+1]. The value of A[i,j] in the matrix is determined by the following formula:
  • the weighted average of each pixel in the image can be performed to achieve a smooth filtering effect.
  • step B1 The specific steps of step B1 are as follows:
  • step B11 add white edges with a width of two pixels to the upper and lower edges of the black and white image obtained in step A, so that there is a closed outline in the image that can completely cover all edges of the wire rope target;
  • the first-order partial derivative matrices P and Q in the x and y directions, the gradient amplitude M and the direction ⁇ are expressed as:
  • step B2 The specific steps of step B2 are:
  • step B21 For the target edge extracted in step B1, fit the outline using a tilted circumscribed rectangle.
  • the fitting principle of the inclined rectangle is: select the point with the largest y value on the edge of the wire rope as the lower vertex of the circumscribed rectangle, Point 0, and draw two rays L 1 and L 2 from it as the vertex in the direction parallel to the x-axis. 0 as the center, rotate L 1 clockwise, and stop when L 1 touches the edge of the wire rope to be measured. Point 1, the intersection point of the edge and L 1 at this time, is used as the second vertex of the rectangle. This determines the first side of the rectangle. .
  • step C1 The specific steps of step C1 are as follows:
  • a is the number of pixels occupied by the length of the tilted rectangle, and b is the number of pixels occupied by the width of the tilted rectangle;
  • S 1 is the total number of pixels occupied by the tilted rectangle
  • S 2 is the total number of pixels in the ROI area
  • h 1 is the ordinate value of the highest vertex of the tilted rectangle
  • S 2 is the height pixel number of the ROI area
  • step C2 The specific steps of step C2 are as follows:
  • step C3 The specific steps of step C3 are as follows:
  • k 1 is the ordinate of the upper left corner vertex when intercepting the ROI area
  • x P1 is the abscissa value of the centroid P 1
  • x P2 is the abscissa value of the tracking point P 2
  • y P1 is the ordinate of the centroid P 1 .
  • y P2 is the ordinate value of tracking point P 2 ;
  • step C32 initialize an integer array a 1 , add the tracking points P 2 obtained in step C31 in the order of video frames to a 1 one by one, and complete the information recording of the tracking target for subsequent data processing.
  • step D1 The specific steps of step D1 are as follows:
  • step D2 The specific steps of step D2 are as follows:
  • the filtering threshold T is defined as:
  • k is an empirical coefficient, usually taking a value between 4 and 5. This method selects different k values for different wire rope vibration conditions based on the degree of dispersion of different parts of the data.
  • the degree of dispersion of the data is determined by the standard deviation ⁇ of the original data before sorting by size.
  • ⁇ the standard deviation
  • the calculation method of each element value in a e can be as follows The formula is determined:
  • Pe i represents the value of the element numbered i in the residual array a e .
  • step D3 The specific steps of step D3 are as follows:
  • the problem data in the difference array a d can be checked according to the filtering threshold T. If a certain element P d m in the array a d satisfies:
  • P o m is a bad pixel of data.
  • the same method is used to judge P o (m+1).
  • the correction method is:
  • P o w is the data bad point, which is corrected by adding the previous value in the original data array a o sequence to the corresponding element in the difference array a d .
  • step E1 The specific steps of step E1 are as follows:
  • Y( ⁇ ) is the frequency domain signal function obtained after transformation, and the peak value of the function waveform is calculated to obtain the n-order vibration natural frequency ⁇ n of the steel wire rope;
  • step E2 The specific steps of step E2 are as follows:
  • T s is the tension of the steel wire rope
  • ⁇ s is the linear density of the steel wire rope
  • l is the length of the steel wire rope.
  • Figure 2 shows the basic principle of the inclined rectangle fitting method.
  • First select the point with the largest y value on the edge of the wire rope as the lower vertex of the circumscribed rectangle, Point 0, and draw two rays L 1 from it as the vertex in the direction parallel to the x-axis.
  • L 2 rotate L 1 clockwise with Point 0 as the center, stop when L 1 contacts the edge of the wire rope to be measured, and use Point 1, the intersection point of the edge and L 1 at this time, as the second vertex of the rectangle, thus determining the first side of the rectangle.
  • the computer When there is an afterimage phenomenon in shooting, the computer will mistakenly regard the afterimage as part of the wire rope target, causing detection errors.
  • the inclined rectangular fitting method can be used to determine the circumscribed rectangle of the wire rope outline without touching the area where the afterimage may occur, and then use the geometric center of the rectangle to determine the circumscribed rectangle. Position the wire rope to be measured to reduce detection errors caused by residual images.
  • Figure 3 shows a comparison of the detection effects of the traditional fitting algorithm and the tilted rectangular fitting method when there is a shooting afterimage and when there is no shooting afterimage.
  • the light color in the figure is the fitting effect of the traditional method, and the dark color is the fitting effect of the tilted rectangle method proposed by the present invention.
  • Figure 1 is a comparison of the detection effect in the case of high resolution (1224 ⁇ 2448) and shooting afterimage;
  • Figure 2 It is a comparison of the detection effect at high resolution (1224 ⁇ 2448) without shooting afterimage;
  • Figure 3 is a comparison of detection effect at low resolution (480 ⁇ 640) and without shooting afterimage;
  • Figure 4 is at low resolution (480 ⁇ 640) and no afterimage detection results comparison.
  • Figure 4 shows the image segmentation results when there is background shadow interference in the captured image. Because the background shadow interference is obvious and the area is large, it cannot be filtered out through image preprocessing.
  • the misdetection phenomenon occurs under background shadow interference by the traditional method. This is because the traditional method will also position the background interference as a wire rope; the right side of the figure shows the correct detection result of the present invention after filtering by rectangular characteristic parameters. Filter out the interference contours and only retain the detection effect of the target contours.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Multimedia (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开一种基于视觉振频识别的提升钢丝绳张力检测方法，对正交式双目相机采集到的提升钢丝绳运行振动图像进行处理，通过提升钢丝绳的振动固有频率来计算其所受张力大小，从而达到完成对提升钢丝绳张力的实时检测。本发明通过倾斜矩形拟合的方式提高了算法的抗干扰性能、通过基于差分理论的绝对均值数据坏点清洗方法提高了算法的检测精度，获得了良好的检测效果。

Description

基于视觉振频识别的提升钢丝绳张力检测方法 技术领域
本发明涉及一种基于视觉振频识别的提升钢丝绳张力检测领域，具体是一种基于视觉振频识别的提升钢丝绳张力检测方法。
背景技术
在矿井提升系统的运行过程当中，钢丝绳会不可避免地产生振动，这将导致钢丝绳承受张力的大小发生变化，严重时会造成钢丝绳的疲劳变形甚至断裂，给矿井提升系统的安全运行造成严重的威胁。因此，如何准确高效地检测钢丝绳的张力对于预防矿井事故具有重大意义，是煤矿生产领域亟需解决的科学问题。
然而，矿井提升结构复杂，提升系统的张力检测难度较大，现有技术多使用传感器对钢丝绳进行张力检测，这些方法存在着一些局限性，例如设备不易安装、传感器会对待测钢丝绳的运行产生影响等问题，不利于钢丝绳张力的长期检测。
机器视觉技术为解决该问题提供了新的方法，即通过计算机来分析待测钢丝绳的振动图像以获取其振动参数，进而计算得到钢丝绳所受张力的值。现有基于机器视觉的检测算法抗干扰性较差，对残影、阴影等异常现象较为敏感；另一方面，现有钢丝绳振动数据坏点清洗方法也存在局限性，不能正确筛除修正样本数据中的迟滞性误差，影响最终检测结果的准确性。
发明内容
本发明旨在解决上述问题，提供一种基于视觉振频识别的提升钢丝绳张力检测方法，克服了上述传统检测方法带来的设备不易安装、易对待测钢丝绳产生影响等问题，提高了检测的效率与安全可靠性，同时提高了钢丝绳图像识别算法的抗干扰性，提高了张力检测的精确性。
为了实现上述目标，本发明采用如下技术方案：
基于视觉振频识别的提升钢丝绳张力检测方法，包括以下步骤：
步骤A，实时采集提升钢丝绳的振动图像，对图像进行预处理，获得钢丝绳与背景分割效果良好的黑白图像；
步骤B，对步骤A获得的黑白图像分割结果进行边缘特征提取与基于倾斜矩形法的轮廓拟合,具体步骤如下：
B1，对步骤A得到的提升钢丝绳目标图像进行边缘检测；
B2，对检测到的边缘进行倾斜矩形拟合，排除因钢丝绳移动较快或图像采集设备达不到使用要求等问题导致的拍摄残影误差；还包括：
步骤C，对步骤B中拟合的倾斜矩形进行特征筛选与几何形心定位，确定钢丝绳的时域振动数据；
步骤D，对步骤C得到的钢丝绳的时域振动数据进行进一步处理，采用基于差分理论的绝对均值法过滤算法异常工况下检测到的数据坏点，提高数据的准确性，具体包括以下子步骤：
D1，确定待处理原始数组a o，并计算其差分数组a d；
D2，定义过滤阈值T作为数据坏点筛选依据；
D3，对数据坏点进行识别与修正；
步骤E，对清洗后的数据进行傅里叶变换，获得其振动固有频率，结合钢丝绳张力-振频公式计算得到钢丝绳所受张力。
进一步的，所述异常工况包括强风、照明失效和人为干扰。
进一步的，步骤A具体包括以下子步骤：
A1，以钢丝绳在图像中活动区域为基准选定ROI区域，减少算法总计算量；
A2，对ROI区域进行灰度转化；
A3，对步骤A2获得的ROI灰度图进行自适应阈值分割；
A4，对步骤A3获得的黑白图像进行高斯滤波操作，减少背景干扰；
A5，对步骤A4获得的滤波结果进行形态学处理，填补钢丝绳目标中的内部空洞。
进一步的，步骤C具体包括以下子步骤：
C1，定义步骤B2中倾斜矩形的紧凑度、顶点高度、横纵比三个特征参数；
C2，结合ROI区域的截割逻辑定义C1中三个特征参数的置信区间，并对区间外的非期望目标拟合矩形进行筛除，排除因背景阴影、油污或划痕造成的干扰；
C3，确定步骤C2中筛选得到的目标矩形的几何形心，作为当前帧钢丝绳的位置，再通过逐帧采集形心最终确定钢丝绳的时域振动数据。
进一步的，所述步骤B1包括以下子步骤：
B11，为步骤A中得到的黑白图像的上下边缘添加宽度为两个像素的白边，使图片中存在一条闭合的轮廓能够完整覆盖钢丝绳目标的所有边缘；
B12，使用Canny边缘算子对提升钢丝绳目标区域进行边缘提取；
步骤B2包括以下子步骤：
B21，对步骤B1提取的目标边缘，采用倾斜外接矩形的方式对轮廓进行拟合。倾斜矩形的拟合原理为：选取钢丝绳边缘中y值最大的点作为外接矩形的下顶点Point 0，以其为顶点沿与x轴平行的方向做两条射线L 1与L 2，以点Point 0为中心顺时针旋转L 1，当L 1接触到待测钢丝绳边缘时停止，以此时边缘与L 1的交点Point 1作为矩形的第二个顶点,这样就确定了矩形的第一条边。然后做这条边的垂线，与L 2的夹角为θ 2，然后在该方向确定点Point 3，使Point 0到Point 3的距离为钢丝绳的直径d 1，从而完成倾斜矩形的建立。
进一步的，所述步骤C1的具体步骤如下：
C11，计算步骤B2得到的倾斜矩形的横纵比F 1，计算公式为：
Figure PCTCN2022123341-appb-000001
式中，a为倾斜矩形的长所占像素数，b为倾斜矩形宽所占像素数；
C12，计算步骤B2得到的倾斜矩形的紧凑度F 2，计算公式为：
Figure PCTCN2022123341-appb-000002
式中，S 1为倾斜矩形所占总像素数，S 2为ROI区域的总像素数；
C13，计算步骤B2得到的倾斜矩形的顶点高度F 3，计算公式为：
Figure PCTCN2022123341-appb-000003
式中，h 1为倾斜矩形最高顶点的纵坐标值，h 2为ROI区域的高度像素数；
所述步骤C2的具体步骤如下：
C21，设定倾斜矩形的横纵比F 1的置信区间Q 1为[1,5]，
倾斜矩形的紧凑度F 2的置信区间Q 2为[0.05,0.7]，倾斜矩形的顶点高度F 3的置信区间Q 3为[0,0.1]；如果倾斜拟合矩形的特征参数于置信区间外则将其作为干扰轮廓进行排除；
所述步骤C3的具体步骤如下：
C31，记录筛选得到的钢丝绳目标轮廓的几何形心，记录形心在ROI区域中的坐标P 1作为追踪依据，将其换算为原图坐标P 2，原图坐标P 2的横纵坐标的值与形心在ROI区域中的坐标P 1横纵坐标的值如下：
Figure PCTCN2022123341-appb-000004
Figure PCTCN2022123341-appb-000005
式中，k 1为截取ROI区域时的左上角顶点纵坐标；
x P1为形心P 1的横坐标值，
x P2为追踪点P 2的横坐标值，
y P1为形心P 1的纵坐标值，
y P2为追踪点P 2的纵坐标值；
C32，初始化一个原始数据数组a 0，将步骤C31按照图像帧的顺序得到的追踪点P 2的逐个添加到原始数据数组a 0中，完成对追踪目标一个周期内的信息记录，以供后续进行数据处理。
进一步的，所述步骤D1的具体步骤如下：
D11，将步骤C3中检测得到的钢丝绳振动位移按照帧数的顺序依次添加到原始数据数组a o中，然后依照数值由小到大的规则对其重新排序为a o[P o1,P o2,P o3,…,P on]，其中P o1≤P o2≤P o3≤...≤P on；
D12，接着，计算数组a o的差分数组a d，其计算方式由下式表达：
P di＝P o(i+1)-P oi
式中，P di代表差分数组a d中序号为i的元素的值，
P oi代表原始数组a o中序号为i的元素的值，
接着，计算差分数组a d的绝对均值
Figure PCTCN2022123341-appb-000006
其计算方式可由下式表示：
Figure PCTCN2022123341-appb-000007
式中，n为差分数组a d中元素个数；
所述步骤D2的具体步骤如下：
D21，过滤阈值T的定义方法为：
Figure PCTCN2022123341-appb-000008
式中，k为经验系数，取4～5之间的值，根据数据不同部分的分散程度，为不同的钢丝绳振动情况选取不同的k值，
数据的分散程度由未按照大小排序前的原始数据的标准差σ所决定，为计算标准差σ，需先确定数组a o的残差数组a e，a e中各元素值的计算方式由下式确定：
Figure PCTCN2022123341-appb-000009
式中，Pe i代表残差数组a e中序号为i的元素的值；
计算标准差σ为：
Figure PCTCN2022123341-appb-000010
将未按照大小排序前的原始数据平均分割为N部分，以同样的方法分别计算每一部分的标准差σ 1,σ 2,σ 3,…,σ N，然后分别确定每一部分的经验系数k为：
Figure PCTCN2022123341-appb-000011
所述步骤D3的具体步骤如下：
D31，确定k值之后，根据过滤阈值T来检查差分数组a d中的问题数据，若数组a d中某元素P dm满足：
Figure PCTCN2022123341-appb-000012
则认为其对应的原始数据数组a o中的P om、P o(m+1)两点至少存在一个数据坏点；
接着，分别对P om、P o(m+1)两个点进行是否为数据坏点的判断，若点P om满足：
|[P om-P o(m-1)]-[P o(m-1)-P o(m-2)]|>σ
则认为P om是数据坏点，对P o(m+1)的判断方式同理；
D32，对判断后认定的数据坏点进行修正，修正方式为：
P ow＝P o(w-1)+P d(w-1)
式中，P ow为数据坏点，
P o(w-1)为数组a o中序号为w-1的元素数值，
P d(w-1)为数组a d中序号为w-1的元素数值；
若局部数据标准差σ i近似为0，则认为这部分数据出现了迟滞误差，算法会将这部分数据予以警报并从样本中剔除，避免对后续参数计算结果造成影响；
将修正后的数组数值还原为原始采集序列，即完成钢丝绳横向振动数据的数据清洗工作。
有益效果：
本发明采用以上技术方案与现有技术相比，具有以下技术效果：
第一点：本发明能够有效减少由拍摄残影造成的检测误差，这是因为本发明采用了倾斜矩形拟合的方式对钢丝绳进行轮廓拟合与定位，在不涉及图像中钢丝绳残影生成区域的情况下完成对钢丝绳的定位。
第二点：本发明避免了检测背景中存在的油污、划痕或由光照不均导致的背 景阴影等现象导致的错检现象，这是因为本发明通过倾斜矩形拟合的特征参数筛选对于图像中的背景干扰目标进行了筛除，仅保留了待测钢丝绳的目标轮廓。
第三点：本发明具有更好的钢丝绳振动数据坏点过滤效果，这是因为本发明采用了基于差分理论的绝对均值滤波法，该方法的优势在于：首先，该方法根据钢丝绳振动数据的波动程度设置不同的过滤阈值，针对钢丝绳振动数据在不同服役阶段波动程度不同的特点具有更好的过滤效果；然后，该方法能够基于差分法思想过滤钢丝绳振动数据中绝对值与其相邻点相差较大，但本身绝对值较小的潜藏坏点。因此，本发明的坏点过滤方法能够有效识别钢丝绳振动数据中的扩散性、隐蔽性与迟滞性误差，针对钢丝绳振动数据非正态分布、随机性强等特点具有更好的适用性。
附图说明
图1是本发明算法流程图。
图2是倾斜矩形拟合法的基本原理图。
图3是出现拍摄残影情况与未出现拍摄残影情况时，传统拟合算法与倾斜矩形拟合法的检测效果对比图。
图4是存在阴影的钢丝绳图片与图像分割效果图。
图5是传统算法的误检现象与本发明算法的筛选结果图。
图6是基于差分理论的绝对均值修正方法流程图。
具体实施方式
下面结合附图对本发明的技术方案做进一步的详细说明：
如图1所示，本发明提供一种基于视觉振频识别的提升钢丝绳张力检测方法，包括图像采集和预处理、目标边缘提取与倾斜矩形拟合、特征筛选与形心定位、数据处理与结果输出四个部分，其具体实施步骤如下：
A1，以钢丝绳在图像中活动区域为基准选定ROI区域，减少算法总计算量；
A2，对ROI区域进行灰度转化；
A3，对步骤A2获得的ROI灰度图进行自适应阈值分割；
A4，对步骤A3获得的黑白图像进行高斯滤波操作，减少背景干扰；
A5，对步骤A4获得的滤波结果进行形态学处理，填补钢丝绳目标中的内部空洞；
步骤B，对步骤A获得的图像分割结果进行边缘特征提取与基于倾斜矩形法的轮廓拟合,具体步骤如下：
B1，对步骤A得到的提升钢丝绳目标图像进行边缘检测；
B2，对检测到的边缘进行倾斜矩形拟合，排除因钢丝绳移动较快或图像采集设备达不到使用要求等问题导致的拍摄残影误差；
步骤C，对步骤B中拟合的倾斜矩形进行特征筛选与几何形心定位，具体包括以下步骤：
C1，定义步骤B2中倾斜矩形的紧凑度、顶点高度、横纵比三个特征参数；
C2，结合ROI区域的截割逻辑定义C1中三个特征参数的置信区间，并对区间外的非期望目标拟合矩形进行筛除，排除因背景阴影、油污或划痕造成的干扰；
C3，确定步骤C2中筛选得到的目标矩形的几何形心，作为当前帧钢丝绳的位置，再通过逐帧采集形心最终确定钢丝绳的横向振动位移曲线。
步骤D，对步骤D获得的钢丝绳横向振动参数进行数据处理，计算得到钢丝绳的张力大小，具体包括以下步骤：
步骤D，对检测到的数据进行进一步处理，采用基于差分理论的绝对均值法过滤算法异常工况下(强风、照明失效、人为干扰等)检测到的数据坏点，提高数据的准确性，该方法流程如图6所示。
D1，确定待处理原始数组a o，并计算其差分数组a d；
D2，定义过滤阈值T作为数据坏点筛选依据；
D3，对数据坏点进行识别与修正；
步骤E，对清洗后的数据进行进一步的数据处理，以得到钢丝绳张力，具体步骤如下：
E1，对数据进行傅里叶变换，获得钢丝绳振动固有频率；
E2，根据钢丝绳振动固有频率，计算其承受张力的值。
步骤A2具体步骤为：
A21，计算图像第i个像素点的灰度值为：
Gray i＝0.299R i+0.587G i+0.114B i
式中R i、G i、B i为三通道彩色图像第i个像素点的红、绿、蓝通道的值。然后，将灰度值赋给此像素点，并遍历整幅图像，就完成了灰度图转化。
步骤A3具体步骤为：
A31.计算当前图片输入图像的灰度直方图确定一个均值M，以M值为分界线将直方图分为前后两个部分MA和MB，并定义MA部分所占像素数为PA，MB部分像素数为PB，通过类间方差定义：
ICV＝PA·(MA-M) 2+PB·(MB-M) 2
式中ICV为类间方差，使其取得最大值的阈值M即为最佳阈值；
A32.以阈值M为分界线将图片分割为目标和背景两部分。
步骤A4具体步骤为：
A41.定义一维离散零均值高斯滤波器函数g(x)为：
Figure PCTCN2022123341-appb-000013
式中σ为高斯分布参数，它确定了高斯函数的宽度。本发明采用二维离散函数对图像进行平滑滤波，首先将图像与上述一维高斯函数进行卷积，之后再与一个相同但方向与卷积结果互相垂直的一维高斯函数卷积，得到二维高斯函数：
Figure PCTCN2022123341-appb-000014
A42.通过对二维高斯函数进行离散化来获取离散点上的权值，例如，要获得一个大小为(2n+1)×(2n+1)的滤波模板，则以模板中心位置的像素作为中心元素来建立一个矩阵A[2n+1,2n+1]，矩阵内A[i,j]的值由如下公式确定：
Figure PCTCN2022123341-appb-000015
之后对计算得到的各点的值进行归一化处理以得到各点权重V，计算过程为：
Figure PCTCN2022123341-appb-000016
之后就能对图像各像素点进行加权平均以取得平滑滤波的效果。
步骤B1的具体步骤如下：
B11，为步骤A中得到的黑白图像的上下边缘添加宽度为两个像素的白边，使图片中存在一条闭合的轮廓能够完整覆盖钢丝绳目标的所有边缘；
B12，使用Canny边缘算子对提升钢丝绳目标区域进行边缘提取，本发明采用Canny算法中较为简单的一种卷积算子，表达如下：
Figure PCTCN2022123341-appb-000017
式中x方向与y方向的一阶偏导矩阵P、Q，梯度幅值M和方向θ表达为：
P[i,j]＝(f[i,j+1]-f[i,j]+f[i+1,j+1]-f[i+1,j])/2
Q[i,j]＝(f[i,j]-f[i+1,j]+f[i,j+1]-f[i+1,j+1])/2
Figure PCTCN2022123341-appb-000018
θ[i,j]＝tan -1(Q[i,j]/P[i,j])
通过求解矩阵来进行进一步的运算，获得提升钢丝绳目标的边界图像。
步骤B2具体步骤为：
B21，对步骤B1提取的目标边缘，采用倾斜外接矩形的方式对轮廓进行拟合。倾斜矩形的拟合原理为：选取钢丝绳边缘中y值最大的点作为外接矩形的下顶点Point 0，以其为顶点沿与x轴平行的方向做两条射线L 1与L 2，以点Point 0为中心顺时针旋转L 1，当L 1接触到待测钢丝绳边缘时停止，以此时边缘与L 1的交点Point 1作为矩形的第二个顶点,这样就确定了矩形的第一条边。然后做这条边的垂线，与L 2的夹角为θ 2，然后在该方向确定点Point 3，使Point 0到Point 3的距离为钢丝绳的直径d 1，从而完成倾斜矩形的建立。
步骤C1的具体步骤如下：
C11，计算步骤B2得到的倾斜矩形的横纵比F 1，计算公式为：
Figure PCTCN2022123341-appb-000019
式中a为倾斜矩形的长所占像素数，b为倾斜矩形宽所占像素数；
C12，计算步骤B2得到的倾斜矩形的紧凑度F 2，计算公式为：
Figure PCTCN2022123341-appb-000020
式中S 1为倾斜矩形所占总像素数，S 2为ROI区域的总像素数；
C13，计算步骤B2得到的倾斜矩形的顶点高度F 3，计算公式为：
Figure PCTCN2022123341-appb-000021
式中h 1为倾斜矩形最高顶点的纵坐标值，S 2为ROI区域的高度像素数；
步骤C2的具体步骤如下：
C21，设定F 1的置信区间Q 1为[1,5]，F 2的置信区间Q 2为[0.05,0.7]，F 3的置信区间Q 3为[0,0.1]；如果倾斜拟合矩形的特征参数于置信区间外则将其作为干扰轮廓进行排除。
步骤C3的具体步骤如下：
C31，记录筛选得到的钢丝绳目标轮廓的几何形心，记录形心在ROI区域中的坐标P 1作为追踪依据，将其换算为原图坐标P 2，P 2的横纵坐标的值与P 1横纵坐标的值如下：
Figure PCTCN2022123341-appb-000022
Figure PCTCN2022123341-appb-000023
式中k 1为截取ROI区域时的左上角顶点纵坐标；x P1为形心P 1的横坐标值，x P2为追踪点P 2的横坐标值，y P1为形心P 1的纵坐标值，y P2为追踪点P 2的纵坐标值；
C32，初始化一个整形数组a 1，将步骤C31按照视频帧的顺序得到的追踪点P 2的逐个添加到a 1中，完成对追踪目标的信息记录，以供后续进行数据处理。
步骤D1的具体步骤如下：
D11，将C3中检测得到的钢丝绳振动位移按照帧数的顺序依次添加到原始数据数组a o中，然后依照数值由小到大的规则对其重新排序为a o[P o1,P o2,P o3,…,P on]，其中P o1≤P o2≤P o3≤...≤P on；
D12，接着，计算数组a o的差分数组a d，其计算方式可由下式表达：
P di＝P o(i+1)-P oi
式中，P di代表差分数组a d中序号为i的元素的值，P oi代表原始数组a o中序号为i的元素的值。接着，计算差分数组a d的绝对均值
Figure PCTCN2022123341-appb-000024
其计算方式可由下式表示：
Figure PCTCN2022123341-appb-000025
步骤D2的具体步骤如下：
D21，过滤阈值T的定义方法为：
Figure PCTCN2022123341-appb-000026
式中k为经验系数，通常取4～5之间的值，本方法根据数据不同部分的分散程度，为不同的钢丝绳振动情况选取不同的k值。
数据的分散程度由未按照大小排序前的原始数据的标准差σ所决定，为计算标准差σ，需先确定数组a o的残差数组a e，a e中各元素值的计算方式可由下式确定：
Figure PCTCN2022123341-appb-000027
式中Pe i代表残差数组a e中序号为i的元素的值。
进一步地，可计算标准差σ为：
Figure PCTCN2022123341-appb-000028
将未按照大小排序前的原始数据平均分割为N部分，以同样的方法分别计算每一部分的标准差σ 1,σ 2,σ 3,…,σ N，然后分别确定每一部分的经验系数k为：
Figure PCTCN2022123341-appb-000029
步骤D3的具体步骤如下：
D31确定k值之后，即可根据过滤阈值T来检查差分数组a d中的问题数据，若数组a d中某元素P dm满足：
Figure PCTCN2022123341-appb-000030
则认为其对应的原始数据数组a o中的P om、P o(m+1)两点至少存在一个数据坏点。
接着，须分别对P om、P o(m+1)两个点进行是否为数据坏点的判断。若点P om满足：
|[P om-P o(m-1)]-[P o(m-1)-P o(m-2)]|>σ
则认为P om是数据坏点。对P o(m+1)的判断方式同理。
D32，对判断后认定的数据坏点进行修正。修正方式为：
P ow＝P o(w-1)+P d(w-1)
式中P ow为数据坏点，通过其原始数据数组a o序列中的上一个值与差分数组a d中的对应元素相加来进行修正。
此外，若局部数据标准差σ i近似为0，则认为这部分数据出现了迟滞误差，算法会将这部分数据予以警报并从样本中剔除，避免对后续参数计算结果造成影响。
将修正后的数组数值还原为原始采集序列，即完成钢丝绳横向振动数据的数据清洗工作。
步骤E1的具体步骤如下：
E11，对钢丝绳横向振动时域函数y(t)进行傅里叶变换：
Figure PCTCN2022123341-appb-000031
式中Y(ω)即为变换后得到的频域信号函数，计算函数波形峰值以获得钢丝绳n阶振动固有频率ω n；
步骤E2的具体步骤如下：
E21，根据步骤E11中得到的钢丝绳振动固有频率ω n计算其所受张力为：
Figure PCTCN2022123341-appb-000032
式中T s为钢丝绳所受张力，ρ s为钢丝绳线密度，l为钢丝绳绳长。通过该计算公式即可获得钢丝绳张力检测结果。
如图2所示为倾斜矩形拟合法的基本原理，首先选取钢丝绳边缘中y值最大的点作为外接矩形的下顶点Point 0，以其为顶点沿与x轴平行的方向做两条射线L 1与L 2，以点Point 0为中心顺时针旋转L 1，当L 1接触到待测钢丝绳边缘时停止，以此时边缘与L 1的交点Point 1作为矩形的第二个顶点,这样就确定了矩形的第一条边。然后做这条边的垂线，与L 2的夹角为θ 2，最后在该方向确定点Point 3，使Point 0到Point 3的距离为钢丝绳的直径d 1，从而完成倾斜矩形的建立。
当拍摄出现残影现象时，计算机会误把残影也当作钢丝绳目标的一部分，从而造成检测误差。但由于残影必定出现在钢丝绳运动方向的反方向，因此可以采用倾斜矩形拟合方式，在不接触可能产生残影的区域的情况下确定钢丝绳轮廓的外接矩形，然后通过矩形的几何形心来对待测钢丝绳进行定位，以减少残影现象带来的检测误差。
如图3所示为出现拍摄残影情况与未出现拍摄残影情况时，传统拟合算法与倾斜矩形拟合法的检测效果对比图。图中浅色为传统方法拟合效果，深色为本发明提出的倾斜矩形方法拟合效果，图①为高分辨率(1224×2448)且有拍摄残影情况下的检测效果对比；图②为高分辨率(1224×2448)且无拍摄残影情况下的检测效果对比；图③为低分辨率(480×640)且有拍摄残影情况下的检测效果对比；图④为低分辨率(480×640)且无拍摄残影情况下的检测效果对比。由图3可以看出，在分辨率不同的情况下，两种拟合方式在无残影现象时定位结果基本一致，但在有残影现象时出现了明显的定位偏差，这说明采用倾斜矩形拟合的方式对于抑制残影现象带来的定位偏差具有重要作用。
如图4所示为当拍摄图像存在背景阴影干扰时的图像分割结果。由于背景阴影干扰较为明显，面积较大，无法通过图像预处理的方式对其进行滤除。
如图5左侧所示为传统方法在背景阴影干扰下出现的误检现象，这是因为传统方法会将背景干扰也作为钢丝绳进行定位；图右侧为本发明经过矩形特征参数过滤后，正确筛选掉干扰轮廓，只保留目标轮廓的检测效果。
以上结合附图对本发明的具体实施方式作了说明，但这些说明不能被理解为限制了本发明的范围，任何在本发明权利要求基础上的改动都是本发明的保护范围。

Claims (7)

  1. 基于视觉振频识别的提升钢丝绳张力检测方法，包括以下步骤：
    步骤A，实时采集提升钢丝绳的振动图像，对图像进行预处理，获得钢丝绳与背景分割效果良好的黑白图像；
    步骤B，对步骤A获得的黑白图像分割结果进行边缘特征提取与基于倾斜矩形法的轮廓拟合,具体步骤如下：
    B1，对步骤A得到的提升钢丝绳目标图像进行边缘检测；
    B2，对检测到的边缘进行倾斜矩形拟合，排除因钢丝绳移动较快或图像采集设备达不到使用要求等问题导致的拍摄残影误差；其特征在于，还包括：
    步骤C，对步骤B中拟合的倾斜矩形进行特征筛选与几何形心定位，确定钢丝绳的时域振动数据；
    步骤D，对步骤C得到的钢丝绳的时域振动数据进行进一步处理，采用基于差分理论的绝对均值法过滤算法异常工况下检测到的数据坏点，提高数据的准确性，具体包括以下子步骤：
    D1，确定待处理原始数组a o，并计算其差分数组a d；
    D2，定义过滤阈值T作为数据坏点筛选依据；
    D3，对数据坏点进行识别与修正；
    步骤E，对清洗后的数据进行傅里叶变换，获得其振动固有频率，结合钢丝绳张力-振频公式计算得到钢丝绳所受张力。
  2. 根据权利要求1所述的基于视觉振频识别的提升钢丝绳张力检测方法，其特征在于，所述异常工况包括强风、照明失效和人为干扰。
  3. 根据权利要求1所述的基于视觉振频识别的提升钢丝绳张力检测方法，其特征在于，步骤A具体包括以下子步骤：
    A1，以钢丝绳在图像中活动区域为基准选定ROI区域，减少算法总计算量；
    A2，对ROI区域进行灰度转化；
    A3，对步骤A2获得的ROI灰度图进行自适应阈值分割；
    A4，对步骤A3获得的黑白图像进行高斯滤波操作，减少背景干扰；
    A5，对步骤A4获得的滤波结果进行形态学处理，填补钢丝绳目标中的内部空洞。
  4. 根据权利要求1所述的基于视觉振频识别的提升钢丝绳张力检测方法，其特征在于，步骤C具体包括以下子步骤：
    C1，定义步骤B2中倾斜矩形的紧凑度、顶点高度、横纵比三个特征参数；
    C2，结合ROI区域的截割逻辑定义C1中三个特征参数的置信区间，并对区间外的非期望目标拟合矩形进行筛除，排除因背景阴影、油污或划痕造成的干扰；
    C3，确定步骤C2中筛选得到的目标矩形的几何形心，作为当前帧钢丝绳的位置，再通过逐帧采集形心最终确定钢丝绳的时域振动数据。
  5. 根据权利要求1所述的基于视觉振频识别的提升钢丝绳张力检测方法，其特征在于，所述步骤B1包括以下子步骤：
    B11，为步骤A中得到的黑白图像的上下边缘添加宽度为两个像素的白边，使图片中存在一条闭合的轮廓能够完整覆盖钢丝绳目标的所有边缘；
    B12，使用Canny边缘算子对提升钢丝绳目标区域进行边缘提取；
    步骤B2包括以下子步骤：
    B21，对步骤B1提取的目标边缘，采用倾斜外接矩形的方式对轮廓进行拟合；倾斜矩形的拟合原理为：选取钢丝绳边缘中y值最大的点作为外接矩形的下顶点Point 0，以其为顶点沿与x轴平行的方向做两条射线L 1与L 2，以点Point 0为中心顺时针旋转L 1，当L 1接触到待测钢丝绳边缘时停止，以此时边缘与L 1的交点Point 1作为矩形的第二个顶点,这样就确定了矩形的第一条边；然后做这条边的垂线，与L 2的夹角为θ 2，然后在该方向确定点Point 3，使Point 0到Point 3的距离为钢丝绳的直径d 1，从而完成倾斜矩形的建立。
  6. 根据权利要求4所述的基于视觉振频识别的提升钢丝绳张力检测方法，其特征在于，所述步骤C1的具体步骤如下：
    C11，计算步骤B2得到的倾斜矩形的横纵比F 1，计算公式为：
    Figure PCTCN2022123341-appb-100001
    式中，a为倾斜矩形的长所占像素数，b为倾斜矩形宽所占像素数；
    C12，计算步骤B2得到的倾斜矩形的紧凑度F 2，计算公式为：
    Figure PCTCN2022123341-appb-100002
    式中，S 1为倾斜矩形所占总像素数，S 2为ROI区域的总像素数；
    C13，计算步骤B2得到的倾斜矩形的顶点高度F 3，计算公式为：
    Figure PCTCN2022123341-appb-100003
    式中，h 1为倾斜矩形最高顶点的纵坐标值，h 2为ROI区域的高度像素数；
    所述步骤C2的具体步骤如下：
    C21，设定倾斜矩形的横纵比F 1的置信区间Q 1为[1,5]，
    倾斜矩形的紧凑度F 2的置信区间Q 2为[0.05,0.7]，倾斜矩形的顶点高度F 3的置信区间Q 3为[0,0.1]；如果倾斜拟合矩形的特征参数于置信区间外则将其作为干扰轮廓进行排除；
    所述步骤C3的具体步骤如下：
    C31，记录筛选得到的钢丝绳目标轮廓的几何形心，记录形心在ROI区域中的坐标P 1作为追踪依据，将其换算为原图坐标P 2，原图坐标P 2的横纵坐标的值与形心在ROI区域中的坐标P 1横纵坐标的值如下：
    Figure PCTCN2022123341-appb-100004
    Figure PCTCN2022123341-appb-100005
    式中，k 1为截取ROI区域时的左上角顶点纵坐标；
    x P1为形心P 1的横坐标值，
    x P2为追踪点P 2的横坐标值，
    y P1为形心P 1的纵坐标值，
    y P2为追踪点P 2的纵坐标值；
    C32，初始化一个原始数据数组a 0，将步骤C31按照图像帧的顺序得到的追踪点P 2的逐个添加到原始数据数组a 0中，完成对追踪目标一个周期内的信息记录，以供后续进行数据处理。
  7. 根据权利要求1所述的基于视觉振频识别的提升钢丝绳张力检测方法，其特征在于，所述步骤D1的具体步骤如下：
    D11，将步骤C3中检测得到的钢丝绳振动位移按照帧数的顺序依次添加到原始数据数组a o中，然后依照数值由小到大的规则对其重新排序为a o[P o1,P o2,P o3,…,P on]，其中P o1≤P o2≤P o3≤…≤P on；
    D12，接着，计算数组a o的差分数组a d，其计算方式由下式表达：
    P di＝P o(i+1)-P oi
    式中，P di代表差分数组a d中序号为i的元素的值，
    P oi代表原始数组a o中序号为i的元素的值，
    接着，计算差分数组a d的绝对均值
    Figure PCTCN2022123341-appb-100006
    其计算方式可由下式表示：
    Figure PCTCN2022123341-appb-100007
    式中，n为差分数组a d中元素个数；
    所述步骤D2的具体步骤如下：
    D21，过滤阈值T的定义方法为：
    Figure PCTCN2022123341-appb-100008
    式中，k为经验系数，取4～5之间的值，根据数据不同部分的分散程度，为不同的钢丝绳振动情况选取不同的k值，
    数据的分散程度由未按照大小排序前的原始数据的标准差σ所决定，为计算标准差σ，需先确定数组a o的残差数组a e，a e中各元素值的计算方式由下式确定：
    Figure PCTCN2022123341-appb-100009
    式中，Pe i代表残差数组a e中序号为i的元素的值；
    计算标准差σ为：
    Figure PCTCN2022123341-appb-100010
    将未按照大小排序前的原始数据平均分割为N部分，以同样的方法分别计算每一部分的标准差σ 1,σ 2,σ 3,…,σ N，然后分别确定每一部分的经验系数k为：
    Figure PCTCN2022123341-appb-100011
    所述步骤D3的具体步骤如下：
    D31，确定k值之后，根据过滤阈值T来检查差分数组a d中的问题数据，若数组a d中某元素P dm满足：
    Figure PCTCN2022123341-appb-100012
    则认为其对应的原始数据数组a o中的P om、P o(m+1)两点至少存在一个数据坏点；
    接着，分别对P om、P o(m+1)两个点进行是否为数据坏点的判断，若点P om满足：
    |[P om-P o(m-1)]-[P o(m-1)-P o(m-2)]|>σ
    则认为P om是数据坏点，对P o(m+1)的判断方式同理；
    D32，对判断后认定的数据坏点进行修正，修正方式为：
    P ow＝P o(w-1)+P d(w-1)
    式中，P ow为数据坏点，
    P o(w-1)为数组a o中序号为w-1的元素数值，
    P d(w-1)为数组a d中序号为w-1的元素数值；
    若局部数据标准差σ i近似为0，则认为这部分数据出现了迟滞误差，算法会将这部分数据予以警报并从样本中剔除，避免对后续参数计算结果造成影响；
    将修正后的数组数值还原为原始采集序列，即完成钢丝绳横向振动数据的数据清洗工作。
PCT/CN2022/123341 2022-06-01 2022-09-30 基于视觉振频识别的提升钢丝绳张力检测方法 WO2023231262A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210619490.8A CN115018785A (zh) 2022-06-01 2022-06-01 基于视觉振频识别的提升钢丝绳张力检测方法
CN202210619490.8 2022-06-01

Publications (1)

Publication Number Publication Date
WO2023231262A1 true WO2023231262A1 (zh) 2023-12-07

Family

ID=83072797

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/123341 WO2023231262A1 (zh) 2022-06-01 2022-09-30 基于视觉振频识别的提升钢丝绳张力检测方法

Country Status (2)

Country Link
CN (1) CN115018785A (zh)
WO (1) WO2023231262A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115018785A (zh) * 2022-06-01 2022-09-06 中国矿业大学 基于视觉振频识别的提升钢丝绳张力检测方法
CN116567438B (zh) * 2023-07-11 2023-09-05 中南大学 一种基于同步相机的钢丝绳张力监测方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020169555A1 (en) * 2001-05-11 2002-11-14 Yuji Hasegawa System, method, and program for detecting approach to object
CN110763388A (zh) * 2019-11-15 2020-02-07 国网四川省电力公司凉山供电公司 一种基于视频识别的钢缆绳张力在线监测装置及方法
CN113466947A (zh) * 2021-06-18 2021-10-01 嘉兴微磁超导科技有限公司 一种应用于超导瞬变电磁的坏点自动去除方法
CN114140384A (zh) * 2021-10-26 2022-03-04 徐州煤矿安全设备制造有限公司 基于轮廓拟合与形心跟踪的提升钢丝绳横向振动图像识别算法
CN115018785A (zh) * 2022-06-01 2022-09-06 中国矿业大学 基于视觉振频识别的提升钢丝绳张力检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020169555A1 (en) * 2001-05-11 2002-11-14 Yuji Hasegawa System, method, and program for detecting approach to object
CN110763388A (zh) * 2019-11-15 2020-02-07 国网四川省电力公司凉山供电公司 一种基于视频识别的钢缆绳张力在线监测装置及方法
CN113466947A (zh) * 2021-06-18 2021-10-01 嘉兴微磁超导科技有限公司 一种应用于超导瞬变电磁的坏点自动去除方法
CN114140384A (zh) * 2021-10-26 2022-03-04 徐州煤矿安全设备制造有限公司 基于轮廓拟合与形心跟踪的提升钢丝绳横向振动图像识别算法
CN115018785A (zh) * 2022-06-01 2022-09-06 中国矿业大学 基于视觉振频识别的提升钢丝绳张力检测方法

Also Published As

Publication number Publication date
CN115018785A (zh) 2022-09-06

Similar Documents

Publication Publication Date Title
CN108921176B (zh) 一种基于机器视觉的指针式仪表定位与识别方法
WO2023231262A1 (zh) 基于视觉振频识别的提升钢丝绳张力检测方法
CN109743879B (zh) 一种基于动态红外热像图处理的地下管廊渗漏检测方法
CN105447512B (zh) 一种精粗结合的光学表面缺陷的检测方法及装置
CN106960208B (zh) 一种仪表液晶数字自动切分和识别的方法及系统
CN110569857B (zh) 一种基于质心距离计算的图像轮廓角点检测方法
CN109409355B (zh) 一种新型变压器铭牌识别的方法及装置
CN104268872B (zh) 基于一致性的边缘检测方法
CN109993099A (zh) 一种基于机器视觉的车道线提取识别方法
CN114549981A (zh) 一种基于深度学习的智能巡检指针式仪表识别及读数方法
CN110414308B (zh) 一种针对输电线路上动态异物的目标识别方法
CN110335233B (zh) 基于图像处理技术的高速公路护栏板缺陷检测系统和方法
CN112862744B (zh) 一种基于超声图像的电容内部缺陷智能检测方法
CN108921831B (zh) 一种基于图像处理技术的污损硬币识别方法
CN109886932A (zh) 基于svm的轮速传感器齿圈表面缺陷检测方法
CN110189375A (zh) 一种基于单目视觉测量的图像目标识别方法
CN106846354B (zh) 一种基于图像分割和随机hough变换的架上图书清点方法
CN105787912B (zh) 一种基于分类的阶跃型边缘亚像素定位方法
CN109886931A (zh) 基于bp神经网络的轮速传感器齿圈表面缺陷检测方法
CN114140384A (zh) 基于轮廓拟合与形心跟踪的提升钢丝绳横向振动图像识别算法
CN110032946A (zh) 一种基于机器视觉的铝/铝泡罩包装药片识别与定位方法
CN115661110B (zh) 一种透明工件识别与定位的方法
CN114972575A (zh) 一种基于轮廓边缘的直线拟合算法
CN115205223A (zh) 透明物体的视觉检测方法、装置、计算机设备及介质
CN113537211A (zh) 一种基于非对称iou的深度学习车牌框定位方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22944570

Country of ref document: EP

Kind code of ref document: A1