CN110567383A - pantograph abrasion early warning system and detection method based on structural forest and sub-pixels - Google Patents

pantograph abrasion early warning system and detection method based on structural forest and sub-pixels Download PDF

Info

Publication number
CN110567383A
CN110567383A CN201910693615.XA CN201910693615A CN110567383A CN 110567383 A CN110567383 A CN 110567383A CN 201910693615 A CN201910693615 A CN 201910693615A CN 110567383 A CN110567383 A CN 110567383A
Authority
CN
China
Prior art keywords
image
pixel
pantograph
edge
forest
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.)
Pending
Application number
CN201910693615.XA
Other languages
Chinese (zh)
Inventor
周竹萍
孙攀
朱周
彭云龙
张嘉彤
杨康乐
林天婵
刘嘉明
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.)
Nanjing Tech University
Original Assignee
Nanjing Tech 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 Nanjing Tech University filed Critical Nanjing Tech University
Priority to CN201910693615.XA priority Critical patent/CN110567383A/en
Publication of CN110567383A publication Critical patent/CN110567383A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
    • G01B11/06—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness for measuring thickness ; e.g. of sheet material
    • 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
    • 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
    • G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • 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
    • G06T2207/20024—Filtering details
    • G06T2207/20032—Median filtering
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30204—Marker
    • G06T2207/30208—Marker matrix

Abstract

The invention provides a pantograph abrasion early warning system and a detection method based on a structural forest and sub-pixels, wherein the system comprises an image acquisition module, an image preprocessing module, a structural forest rapid edge detection module, a sub-pixel detection module, a camera calibration module and a judgment module; the method comprises the following steps: the image acquisition module acquires images and transmits the images to the image preprocessing module to preprocess the images, the rapid edge detection module and the sub-pixel detection module of the structural forest perform edge detection on the preprocessed images, the camera calibration module performs camera calibration and comparison calculation on the images, and the judgment module judges whether the wear of the pantograph is out of limit or not. The invention relates to a fast edge detection algorithm of a structural forest, which ignores local interference information, strengthens boundary extraction and has good robustness and high accuracy; the sub-pixel edge detection algorithm obviously improves the detection precision of the system on the basis of the whole pixel detection; the early warning system can monitor the pantograph in real time, and is high in automation degree and efficiency.

Description

pantograph abrasion early warning system and detection method based on structural forest and sub-pixels
Technical Field
the invention relates to a pantograph abrasion early warning system and a pantograph abrasion early warning detection method based on structural forests and sub-pixels, and belongs to the technical field of intelligent traffic early warning.
Background
the pantograph is the very important equipment of getting electricity on the locomotive, and it obtains the electric energy through the sliding contact between with the contact net and provides the locomotive, and in the middle of the train operation process, continuous contact slip between pantograph slide and the contact net can make the pantograph slide wear and tear constantly, when slide thickness is less than the specified value, must change the pantograph, otherwise can produce serious traffic accident, consequently, for the operation safety of train, it is very necessary to develop real-time effectual detection to the pantograph.
along with the rapid development of scientific technology and the increasingly developed traffic industry, the equipment detection of the locomotive gradually moves to automation and intellectualization, the traditional online detection technology has large labor capacity and low efficiency, the detection accuracy cannot be ensured, and potential safety hazards are brought to locomotive equipment. The invention provides a pantograph abrasion early warning system based on a structural forest and sub-pixels, a structural forest rapid edge detection algorithm related to the system can well ignore local interference information, enhances boundary extraction, has good robustness and high accuracy, and has positive significance for automatic maintenance of a pantograph. Compared with a manual detection mode, the detection mode of image processing has the characteristics of high automation, high precision, high efficiency, non-contact and low cost.
Disclosure of Invention
the invention aims to provide a pantograph abrasion early warning system and a pantograph abrasion early warning detection method based on structural forests and sub-pixels, which can be used for quickly and accurately extracting the upper edge and the lower edge of a pantograph so as to judge whether the abrasion of the pantograph is over-limit or not, replace the pantograph in time and prevent vehicle safety accidents.
In order to solve the problems, the invention provides a pantograph abrasion early warning system based on a structural forest and sub-pixels, which has the following specific technical scheme:
the pantograph abrasion early warning system based on the structural forest and the sub-pixels comprises an image acquisition module, an image preprocessing module, a structural forest rapid edge detection module, a sub-pixel detection module, a camera calibration module and a judgment module; the image acquisition module acquires an image and transmits the image to the image preprocessing module to preprocess the image, the structure forest rapid edge detection module obtains an integral pixel level edge from the preprocessed image by using a structure forest rapid edge detection algorithm, the sub-pixel detection module constructs an edge function model to perform least square fitting to obtain a sub-pixel edge point, the camera calibration module performs camera calibration and comparison calculation on the image to obtain the actual thickness of wear of the pantograph, and the judgment module judges whether the wear of the pantograph is over-limit or not.
Further, the image acquisition module acquires an image of the pantograph in a system measurement area by using a high-speed industrial camera and a photoelectric sensor.
furthermore, the image preprocessing module preprocesses the image by using median filtering, reduces interference information in the image and retains boundary information.
Further, the fast edge detection module of structure forest carries out the fast edge detection of structure forest and includes:
a. Training a random decision forest:
Single decision tree ft(x) According to a binary decomposition function h (x, theta)j) E {0,1} branches repeatedly to the left or right of the tree to classify the sample X e X until a leaf node is reached, if h (X, θ)j) If the node j is 0, the node j sends x to the left, otherwise, the node j sends x to the right, and the process is terminated at the leaf node; recursive training is performed to find the decomposition function h (x, theta)j) Until a set tree depth or threshold value of information gain is reached, the form of the information gain criterion:
Ij＝I(Sj，Sj L，Sj R)
Wherein Sj L＝{(x，y)∈Sj|h(x，θj)＝0}，Sj R＝Sj/Sj R，θjTo make IjGain parameter at maximization, using Sj Ltraining left node, using Sj Rtraining the right node, and defining the standard information gain as follows:
wherein H (S) ═ Sigmaypylog(py) Denotes Shannon entropy, pyIs training data Sjprobability with label y; the method has the advantages that a forest is formed by training a plurality of irrelevant trees to form a forest, the problem that a single decision tree is unstable and over-fitted is solved, a BSDS data set is used as training data for training, and the accuracy of decision forest can be improved by inputting randomly sampled pixel blocks x or feature classes as the training data;
b. Inputting an image: after training a structural forest edge detection model, inputting a preprocessed RGB image for edge detection;
c. And (3) random forest structured output:
Each structural label Y in the label set Y has certain similarity and information gain IjCalculated by measuring the similarity of the structure labels y, which leads toso as to be not easy to define Ijto facilitate calculation of IjDefining a node j, discretizing and mapping all labels y on the node to discretization labels c, and calculating I by using c instead of yj，
π:y∈Y→c∈C{1,2,...,k}
The similarity of a label Y is discretized and mapped to a discretization label C by adopting second-order mapping, a mapping pi is defined, Y → Z is used for coding pixel blocks with the label Y into binary vectors, Euclidean distance between Z is calculated in Z to distinguish whether the pixel blocks with the similar label Y belong to the same partition, m-dimensional features are taken in Z to form a low-dimensional mapping pi, phi, Y → Z, and pi, phi, Z → C is defined, principal component analysis dimension reduction quantization is adopted, a specific label C (1,2, a.
d. Image binarization processing: setting the gray value of a pixel point on the image to be 0 or 255 to obtain a black-white image;
e. Extracting upper and lower edges: the upper edge of the pantograph contour diagram is composed of pixel points of which the first pixel in each row of the image is 1, the lower edge is composed of pixel points of which the second pixel in each row of the image is 1, the upper edge is extracted by adopting a row search method, the pixel points form an image matrix T, the image reflected by the matrix T is subjected to Hough straight line detection, the position of the detected straight line is on the lower edge of the pantograph, and the lower edge of the pantograph is obtained by expanding the straight line.
Further, the sub-pixel detection module performing sub-pixel detection includes:
a. determining the gradient direction: establishing a rectangular coordinate system by taking the gradient direction as an x-axis positive axis and the corresponding pixel gray value as a y-axis;
b. Constructing an edge function model by simulating the blurring effect of the camera: selecting a quadratic polynomialto simulate a low-pass filter with blurring, a step function is selectedFor the ideal step edge function model, the edge model I (x) ═ f (x) × h (x) ═ ax is constructed by convolving h (x) with f (x)3+bx2+cx+d；
c. Finding the position of the sub-pixel edge point: order toObtaining sub-pixel edge point positions
d. Carrying out least square fitting on the constructed edge function model I (x) and pixel points of the area where the edge is located, setting the sum of squares of residual errors returned by the least square method as S,
Wherein I (x) represents the estimated gray value in the edge model, y (x) represents the real image gray value, S should be minimized, so that the first derivative of S to each parameter is equal to 0, and the second derivative is greater than 0, and a set of parameters a, b, c, d is calculated accordingly, wherein the parameter is a cubic polynomial I (x) ax3+bx2+ cx + d edge model parameters to compute the sub-pixel edgethe position of (a).
Further, the camera calibration module performs camera calibration by adopting a checkerboard calibration method, the checkerboard calibration paper and the pantograph slide plate surface are placed on the same plane for image acquisition, the side length of each square of the checkerboard is len, the vertexes of the squares are determined by using an angular point detection method, the coordinates of the vertexes are image coordinates, and the obtained image coordinate matrixes of the vertexes are set as follows:
Defining the corresponding world coordinate matrix as:
n is more than or equal to 4, so that n space points with known world coordinate system coordinates are determined, m' can be obtained, the conversion relation between the world coordinate system and the image coordinate system is also determined, the image coordinates of the sub-pixel points are known, the corresponding world coordinates are obtained through the image coordinates, the distance between each point on the upper edge and each point on the lower edge is determined, the distance between the upper edge and the lower edge is the thickness of the pantograph, and the wear degree of the pantograph is known by comparing the distance with the initial thickness of the pantograph.
Further, the judging module judges whether the wear of the pantograph is out of limit, if the wear of the pantograph is out of limit, a new pantograph is replaced, and if not, the pantograph returns to the image preprocessing module.
the pantograph abrasion early warning detection method based on the structural forest and the sub-pixels comprises the following steps:
Step 1: image acquisition, namely accurately acquiring an image of a pantograph in a system measurement area by using a high-speed industrial camera and a photoelectric sensor;
step 2: image preprocessing, namely judging whether the acquired image pixel points are edge pixel points, if so, directly outputting the pixel values of the points, otherwise, performing median filtering on the points, and outputting the pixel values after the median filtering;
And step 3: performing fast edge detection on the structural forest, and processing the preprocessed image by using a fast edge detection algorithm of the structural forest to obtain an image of the whole pixel level edge;
And 4, step 4: the method comprises the following steps of performing sub-pixel detection, namely determining the gradient direction of an image, constructing an edge function model by using the fuzzy action of a high-speed industrial camera, searching the position of a sub-pixel edge point, and performing least square fitting by using the constructed edge function model and a pixel point of a region where an edge is located to obtain the parameter of the edge function model so as to obtain the sub-pixel edge point;
And 5: calibrating a camera, finding out the corresponding relation between the geometric position of one point on the surface of the pantograph and the image, and calculating by image comparison to obtain the actual thickness of the pantograph;
step 6: judging whether the abrasion is over-limit: and judging whether the wear of the pantograph is out of limit according to the actual thickness of the pantograph obtained by calibrating the camera, immediately replacing a new pantograph if the wear of the pantograph is out of limit, and returning to the image preprocessing module if the wear of the pantograph is out of limit.
Has the advantages that:
1. In the whole pixel detection part, compared with a common Canny algorithm, the structural forest fast edge detection algorithm has the advantages that interference edges are obviously reduced, robustness is good, and accuracy is high;
2. The edge detection algorithm based on the sub-pixels based on fitting is adopted, the detection result is accurate to the sub-pixel level on the basis of the whole pixel detection, and the detection precision and accuracy are improved again;
3. The automatic detection method has the advantages that the automation degree is high, the safety is good, compared with a traditional manual detection method, the image processing method saves a large amount of labor and material cost, and meanwhile, the safety of the maintenance process is improved;
4. The convenience is good, the train can be detected without stopping, and the link of stopping is avoided;
5. but real-time detection, detection go on in real time, and the wearing and tearing transfinite can report to the police at once, improves factor of safety.
Drawings
figure 1 is a schematic flow chart of the present invention,
FIG. 2 is a flow chart of the operation of the forest edge detection module of the present invention,
FIG. 3 is a flow chart of the operation of the sub-pixel detection module of the present invention,
FIG. 4 is a top and bottom edge graph of the pantograph extracted by the forest detection algorithm of the structure of the present invention,
FIG. 5 is a checkerboard for the camera target of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the scope of protection of the present invention.
As shown in fig. 1, the pantograph wear warning system based on the structural forest and the sub-pixels comprises an image acquisition module, an image preprocessing module, a structural forest fast edge detection module, a sub-pixel detection module, a camera calibration module and a judgment module; the embodiment also provides a pantograph abrasion early warning detection method based on the structural forest and the sub-pixels, which comprises the following steps:
Step 1: image acquisition, namely accurately acquiring an image of a pantograph in a system measurement area by using a high-speed industrial camera and a photoelectric sensor;
step 2: image preprocessing, namely judging whether the acquired image pixel points are edge pixel points, if so, directly outputting the pixel values of the points, otherwise, performing median filtering on the points, and outputting the pixel values after the median filtering;
And step 3: performing fast edge detection on the structural forest, and processing the preprocessed image by using a fast edge detection algorithm of the structural forest to obtain an image of the whole pixel level edge;
And 4, step 4: the method comprises the following steps of performing sub-pixel detection, determining the gradient direction of an image, constructing an edge function model by utilizing the fuzzy action of a high-speed industrial camera, searching the position of a sub-pixel edge point, and performing least square fitting by using the constructed edge function model and pixel points of a region where an edge is located to obtain the parameters of the edge function model so as to obtain the sub-pixel edge point;
and 5: calibrating a camera, finding out the corresponding relation between the geometric position of one point on the surface of the pantograph and the image, and calculating by image comparison to obtain the actual thickness of the pantograph;
Step 6: judging whether the abrasion is over-limit: and judging whether the wear of the pantograph is out of limit according to the actual thickness of the pantograph obtained by calibrating the camera, immediately replacing a new pantograph if the wear of the pantograph is out of limit, and returning to the image preprocessing module if the wear of the pantograph is out of limit.
as shown in fig. 2, the performing of the structure forest fast edge detection by the structure forest fast edge detection module includes:
a. training a random decision forest:
single decision tree ft(x) According to a binary decomposition function h (x, theta)j) E {0,1} branches repeatedly to the left or right of the tree to classify the sample X e X until a leaf node is reached, if h (X, θ)j) If the node j is 0, the node j sends x to the left, otherwise, the node j sends x to the right, and the process is terminated at the leaf node; recursive training is performed to find the decomposition function h (x, theta)j) Until a set tree depth or threshold value of information gain is reached, the form of the information gain criterion:
Ij＝I(Sj，Sj L，Sj R)
Wherein Sj L＝{(x，y)∈Sj|h(x，θj)＝0}，Sj R＝Sj/Sj R，θjto make IjGain parameter at maximization, using Sj LTraining left node, using Sj RTraining the right node, and defining the standard information gain as follows:
wherein H (S) ═ Sigmaypylog(py) Denotes Shannon entropy, pyIs training data Sjprobability with label y; the method has the advantages that a forest is formed by training a plurality of irrelevant trees to form a forest, the problem that a single decision tree is unstable and over-fitted is solved, a BSDS data set is used as training data for training, and the accuracy of decision forest can be improved by inputting randomly sampled pixel blocks x or feature classes as the training data;
b. inputting an image: after training a structural forest edge detection model, inputting a preprocessed RGB image for edge detection;
c. And (3) random forest structured output:
each structural label Y in the label set Y has a certain valueSimilarity, information gain Ijcalculated by means of measuring the similarity of the structure labels y, this results in an undefined Ijto facilitate calculation of Ijdefining a node j, discretizing and mapping all labels y on the node to discretization labels c, and calculating I by using c instead of yj，
π:y∈Y→c∈C{1,2,...,k}
the similarity of a label Y is discretized and mapped to a discretization label C by adopting second-order mapping, a mapping pi is defined, Y → Z is used for coding pixel blocks with the label Y into binary vectors, Euclidean distance between Z is calculated in Z to distinguish whether the pixel blocks with the similar label Y belong to the same partition, m-dimensional features are taken in Z to form a low-dimensional mapping pi, phi, Y → Z, and pi, phi, Z → C is defined, principal component analysis dimension reduction quantization is adopted, a specific label C (1,2, a.
d. image binarization processing: setting the gray value of a pixel point on the image to be 0 or 255 to obtain a black-white image;
e. Extracting upper and lower edges: the upper edge of the pantograph contour diagram is composed of pixel points of which the first pixel in each row of the image is 1, the lower edge is composed of pixel points of which the second pixel in each row of the image is 1, the upper edge is extracted by adopting a row search method, the pixel points form an image matrix T, the image reflected by the matrix T is subjected to Hough straight line detection, the position of the detected straight line is on the lower edge of the pantograph, and the lower edge of the pantograph is obtained by expanding the straight line.
as shown in fig. 3, the sub-pixel detection module performing sub-pixel detection includes:
a. Determining the gradient direction: establishing a rectangular coordinate system by taking the gradient direction as an x-axis positive axis and the corresponding pixel gray value as a y-axis;
b. Constructing an edge function model by simulating the blurring effect of the camera: selecting a quadratic polynomialTo simulate a low-pass filter with blurring, a step function is selectedFor the ideal step edge function model, the edge model I (x) ═ f (x) × h (x) ═ ax is constructed by convolving h (x) with f (x)3+bx2+cx+d；
c. Finding the position of the sub-pixel edge point: order toobtaining sub-pixel edge point positions
d. Carrying out least square fitting on the constructed edge function model I (x) and pixel points of the area where the edge is located, setting the sum of squares of residual errors returned by the least square method as S,
Wherein I (x) represents the estimated gray value in the edge model, y (x) represents the real image gray value, S should be minimized, so that the first derivative of S to each parameter is equal to 0, and the second derivative is greater than 0, and a set of parameters a, b, c, d is calculated accordingly, wherein the parameter is a cubic polynomial I (x) ax3+bx2+ cx + d edge model parameters to compute the sub-pixel edgeThe position of (a).
As shown in fig. 5, the camera calibration module performs camera calibration by using a checkerboard calibration method, places checkerboard calibration paper and a pantograph slide plate surface on the same plane for image acquisition, where the side length of each square of the checkerboard is len, determines vertexes of a plurality of squares by using an angular point detection method, where coordinates of the vertexes are image coordinates, and sets the obtained image coordinate matrices of the vertexes as:
Defining the corresponding world coordinate matrix as:
n is more than or equal to 4, so that n space points with known world coordinate system coordinates are determined, m' can be obtained, the conversion relation between the world coordinate system and the image coordinate system is also determined, the image coordinates of the sub-pixel points are known, the corresponding world coordinates are obtained through the image coordinates, and the distance between each point on the upper edge and each point on the lower edge are determined, as shown in FIG. 4; the distance between the upper edge and the lower edge is the thickness of the pantograph, and the wear degree of the pantograph is known by comparing the thickness with the initial thickness of the pantograph.
Compared with a common Canny algorithm, the fast edge detection algorithm for the structural forest in the whole pixel detection part has the advantages that interference edges are obviously reduced, the robustness is good, the accuracy is high, the edge detection algorithm based on the fitted sub-pixels is adopted, the detection result is accurate to the sub-pixel level on the basis of whole pixel detection, the detection accuracy and the accuracy are improved again, the automation degree is high, the safety is good, compared with the traditional manual detection method, the image processing method saves a large amount of manpower and material cost, the safety of the maintenance process is improved, the convenience is good, the train can be detected without stopping, the link of stopping is avoided, the real-time detection can be carried out, the alarm can be immediately carried out when the abrasion is beyond the limit, and the safety coefficient is improved.
However, the present invention is not limited to the above-described preferred embodiments. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention should be determined by the appended claims. Although the present invention has been described with reference to the preferred embodiments, it is not intended to be limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims (8)

1. Pantograph wearing and tearing early warning system based on structure forest and subpixel, its characterized in that: the system comprises an image acquisition module, an image preprocessing module, a structure forest rapid edge detection module, a sub-pixel detection module, a camera calibration module and a judgment module; the image acquisition module acquires an image, the image is transmitted to the image preprocessing module to be preprocessed, the structure forest rapid edge detection module processes the preprocessed image by using a structure forest rapid edge detection algorithm to obtain an image of an integral pixel level edge, the sub-pixel detection module constructs an edge function model and performs least square fitting to obtain sub-pixel edge points, the camera calibration module performs camera calibration and comparison calculation on the image to obtain the actual thickness of wear of the pantograph, and the judgment module judges whether the wear of the pantograph is over-limit or not.
2. The structural forest and sub-pixel based pantograph wear warning system of claim 1, wherein: the image acquisition module acquires the image of the pantograph in the system measurement area by using a high-speed industrial camera and a photoelectric sensor.
3. The structural forest and sub-pixel based pantograph wear warning system of claim 1, wherein: the image preprocessing module preprocesses the image by using median filtering, reduces interference information in the image and reserves boundary information.
4. the structural forest and sub-pixel based pantograph wear warning system of claim 1, wherein: the quick edge detection module of structure forest carries out the quick edge detection of structure forest and includes:
a. training a random decision forest:
Single decision tree ft(x) According to a binary decomposition function h (x, theta)j) E {0,1} iteratively branches to the left or right of the tree to classify the sample X e X until reachingto a leaf node, if h (x, θ)j) If the node j is 0, the node j sends x to the left, otherwise, the node j sends x to the right, and the process is terminated at the leaf node; recursive training is performed to find the decomposition function h (x, theta)j) Until a set tree depth or threshold value of information gain is reached, the form of the information gain criterion:
Ij＝I(Sj，Sj L，Sj R)
Wherein Sj L＝{(x，y)∈Sj|h(x，θj)＝0}，Sj R＝Sj/Sj R，θjTo make IjGain parameter at maximization, using Sj LTraining left node, using Sj RTraining the right node, and defining the standard information gain as follows:
wherein H (S) ═ Sigmaypylog(py) Denotes Shannon entropy, pyis training data SjProbability with label y; the method has the advantages that a forest is formed by training a plurality of irrelevant trees to form a forest, the problem that a single decision tree is unstable and over-fitted is solved, a BSDS data set is used as training data for training, and the accuracy of decision forest can be improved by inputting randomly sampled pixel blocks x or feature classes as the training data;
b. inputting an image: after training a structural forest edge detection model, inputting a preprocessed RGB image for edge detection;
c. And (3) random forest structured output:
Each structural label Y in the label set Y has certain similarity and information gain IjCalculated by means of measuring the similarity of the structure labels y, this results in an undefined Ijto facilitate calculation of IjDefining a node j, discretizing and mapping all labels y on the node to discretization labels c, and calculating I by using c instead of yj，
π:y∈Y→c∈C{1,2,...,k}
the similarity of a label Y is discretized and mapped to a discretization label C by adopting second-order mapping, a mapping pi is defined, Y → Z is used for coding pixel blocks with the label Y into binary vectors, Euclidean distance between Z is calculated in Z to distinguish whether the pixel blocks with the similar label Y belong to the same partition, m-dimensional features are taken in Z to form a low-dimensional mapping pi, phi, Y → Z, and pi, phi, Z → C is defined, principal component analysis dimension reduction quantization is adopted, a specific label C (1,2, a.
d. Image binarization processing: setting the gray value of a pixel point on the image to be 0 or 255 to obtain a black-white image;
e. Extracting upper and lower edges: the upper edge of the pantograph contour diagram is composed of pixel points of which the first pixel in each row of the image is 1, the lower edge is composed of pixel points of which the second pixel in each row of the image is 1, the upper edge is extracted by adopting a row search method, the pixel points form an image matrix T, the image reflected by the matrix T is subjected to Hough straight line detection, the position of the detected straight line is on the lower edge of the pantograph, and the lower edge of the pantograph is obtained by expanding the straight line.
5. The structural forest and sub-pixel based pantograph wear warning system of claim 1, wherein: the sub-pixel detection module for sub-pixel detection comprises:
a. Determining the gradient direction: establishing a rectangular coordinate system by taking the gradient direction as an x-axis positive axis and the corresponding pixel gray value as a y-axis;
b. constructing an edge function model by simulating the blurring effect of the camera: selecting a quadratic polynomialto simulate a low-pass filter with blurring, a step function is selectedFor the ideal step edge function model, the edge model I (x) ═ f (x) × h (x) ═ ax is constructed by convolving h (x) with f (x)3+bx2+cx+d；
c. Finding the position of the sub-pixel edge point: order toObtaining sub-pixel edge point positions
d. Carrying out least square fitting on the constructed edge function model I (x) and pixel points of the area where the edge is located, setting the sum of squares of residual errors returned by the least square method as S,
wherein I (x) represents the estimated gray value in the edge model, y (x) represents the real image gray value, S should be minimized, so that the first derivative of S to each parameter is equal to 0, and the second derivative is greater than 0, and a set of parameters a, b, c, d is calculated accordingly, wherein the parameter is a cubic polynomial I (x) ax3+bx2+ cx + d edge model parameters to compute the sub-pixel edgethe position of (a).
6. The structural forest and sub-pixel based pantograph wear warning system of claim 1, wherein: the camera calibration module is used for calibrating a camera by adopting a checkerboard calibration method, checkerboard calibration paper and a pantograph slide plate surface are arranged on the same plane for image acquisition, the side length of each square of the checkerboard is len, the vertex of a plurality of squares is determined by using an angular point detection method, the coordinate of the vertex is an image coordinate, and the obtained image coordinate matrix of the plurality of vertices is set as follows:
defining the corresponding world coordinate matrix as:
N is more than or equal to 4, so that n space points with known world coordinate system coordinates are determined, m' can be obtained, the conversion relation between the world coordinate system and the image coordinate system is also determined, the image coordinates of the sub-pixel points are known, the corresponding world coordinates are obtained through the image coordinates, the distance between each point on the upper edge and each point on the lower edge is determined, the distance between the upper edge and the lower edge is the thickness of the pantograph, and the wear degree of the pantograph is known by comparing the distance with the initial thickness of the pantograph.
7. the structural forest and sub-pixel based pantograph wear warning system of claim 1, wherein: the judgment module judges whether the abrasion of the pantograph is out of limit or not, if the abrasion of the pantograph is out of limit, a new pantograph is replaced, and if not, the image preprocessing module returns.
8. Structural forest and sub-pixel-based pantograph abrasion early warning detection method is characterized by comprising the following steps of: the method comprises the following steps:
Step 1: image acquisition, namely accurately acquiring an image of a pantograph in a system measurement area by using a high-speed industrial camera and a photoelectric sensor;
Step 2: image preprocessing, namely judging whether the acquired image pixel points are edge pixel points, if so, directly outputting the pixel values of the points, otherwise, performing median filtering on the points, and outputting the pixel values after the median filtering;
And step 3: performing fast edge detection on the structural forest, and processing the preprocessed image by using a fast edge detection algorithm of the structural forest to obtain an image of the whole pixel level edge;
And 4, step 4: the method comprises the following steps of performing sub-pixel detection, namely determining the gradient direction of an image, constructing an edge function model by using the fuzzy action of a high-speed industrial camera, searching the position of a sub-pixel edge point, and performing least square fitting by using the constructed edge function model and a pixel point of a region where an edge is located to obtain the parameter of the edge function model so as to obtain the sub-pixel edge point;
And 5: calibrating a camera, finding out the corresponding relation between the geometric position of one point on the surface of the pantograph and the image, and calculating by image comparison to obtain the actual thickness of the pantograph;
step 6: judging whether the abrasion is over-limit: and judging whether the wear of the pantograph is out of limit according to the actual thickness of the pantograph obtained by calibrating the camera, immediately replacing a new pantograph if the wear of the pantograph is out of limit, and returning to the image preprocessing module if the wear of the pantograph is out of limit.
CN201910693615.XA 2019-07-30 2019-07-30 pantograph abrasion early warning system and detection method based on structural forest and sub-pixels Pending CN110567383A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910693615.XA CN110567383A (en) 2019-07-30 2019-07-30 pantograph abrasion early warning system and detection method based on structural forest and sub-pixels

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910693615.XA CN110567383A (en) 2019-07-30 2019-07-30 pantograph abrasion early warning system and detection method based on structural forest and sub-pixels

Publications (1)

Publication Number Publication Date
CN110567383A true CN110567383A (en) 2019-12-13

Family

ID=68773588

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910693615.XA Pending CN110567383A (en) 2019-07-30 2019-07-30 pantograph abrasion early warning system and detection method based on structural forest and sub-pixels

Country Status (1)

Country Link
CN (1) CN110567383A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112762877A (en) * 2020-12-28 2021-05-07 成都国铁电气设备有限公司 Subway contact net rigid contact line abrasion early warning method and device
CN112991316A (en) * 2021-03-30 2021-06-18 中国空气动力研究与发展中心超高速空气动力研究所 Dynamic measurement technology for model edge ablation amount
CN113324864A (en) * 2020-02-28 2021-08-31 南京理工大学 Pantograph carbon slide plate abrasion detection method based on deep learning target detection
CN114463258A (en) * 2021-12-23 2022-05-10 中铁第一勘察设计院集团有限公司 Locomotive brake pad abrasion detection method based on deep learning technology
CN114463258B (en) * 2021-12-23 2024-04-30 中铁第一勘察设计院集团有限公司 Rolling stock brake pad abrasion detection method based on deep learning technology

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113324864A (en) * 2020-02-28 2021-08-31 南京理工大学 Pantograph carbon slide plate abrasion detection method based on deep learning target detection
CN112762877A (en) * 2020-12-28 2021-05-07 成都国铁电气设备有限公司 Subway contact net rigid contact line abrasion early warning method and device
CN112991316A (en) * 2021-03-30 2021-06-18 中国空气动力研究与发展中心超高速空气动力研究所 Dynamic measurement technology for model edge ablation amount
CN112991316B (en) * 2021-03-30 2022-07-01 中国空气动力研究与发展中心超高速空气动力研究所 Dynamic measurement technology for model edge ablation amount
CN114463258A (en) * 2021-12-23 2022-05-10 中铁第一勘察设计院集团有限公司 Locomotive brake pad abrasion detection method based on deep learning technology
CN114463258B (en) * 2021-12-23 2024-04-30 中铁第一勘察设计院集团有限公司 Rolling stock brake pad abrasion detection method based on deep learning technology

Similar Documents

Publication Publication Date Title
CN108519605B (en) Road edge detection method based on laser radar and camera
Gao et al. Detection and segmentation of cement concrete pavement pothole based on image processing technology
CN110567383A (en) pantograph abrasion early warning system and detection method based on structural forest and sub-pixels
CN110473187B (en) Object-oriented line scanning three-dimensional pavement crack extraction method
CN110910350B (en) Nut loosening detection method for wind power tower cylinder
CN112419297A (en) Bolt looseness detection method, device, equipment and storage medium
CN115797354B (en) Method for detecting appearance defects of laser welding seam
CN112419298B (en) Bolt node plate rust detection method, device, equipment and storage medium
CN103727930A (en) Edge-matching-based relative pose calibration method of laser range finder and camera
CN113436157A (en) Vehicle-mounted image identification method for pantograph fault
CN112800938B (en) Method and device for detecting occurrence of side rockfall of unmanned vehicle
CN110619258A (en) Road track checking method based on high-resolution remote sensing image
Kumar et al. An automated road roughness detection from mobile laser scanning data
CN114488190A (en) Laser radar 3D point cloud ground detection method
CN112950532A (en) Train pantograph state detection method
CN110348307B (en) Path edge identification method and system for crane metal structure climbing robot
CN116703895B (en) Small sample 3D visual detection method and system based on generation countermeasure network
TW202017784A (en) Car detection method based on LiDAR by proceeding the three-dimensional feature extraction and the two-dimensional feature extraction on the three-dimensional point cloud map and the two-dimensional map
CN109829511B (en) Texture classification-based method for detecting cloud layer area in downward-looking infrared image
CN113066041A (en) Pavement crack detection method based on stack sparse self-coding deep learning
CN116309817A (en) Tray detection and positioning method based on RGB-D camera
CN115755088A (en) Laser point cloud-based automatic measurement method for power transmission line engineering construction parameters
CN116242774A (en) Light three-dimensional detection method and device suitable for rural highways
CN111612087B (en) Method for generating image feature dictionary of EMUs TEDS system
Peng et al. Automatic Crack Detection by Multi-Seeding Fusion on 1mm Resolution 3D Pavement Images

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20191213

WD01 Invention patent application deemed withdrawn after publication