CN108694349A - A kind of pantograph image extraction method and device based on line-scan digital camera - Google Patents
A kind of pantograph image extraction method and device based on line-scan digital camera Download PDFInfo
- Publication number
- CN108694349A CN108694349A CN201710223798.XA CN201710223798A CN108694349A CN 108694349 A CN108694349 A CN 108694349A CN 201710223798 A CN201710223798 A CN 201710223798A CN 108694349 A CN108694349 A CN 108694349A
- Authority
- CN
- China
- Prior art keywords
- pantograph
- image
- checked
- gradient
- positive
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/2163—Partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
Abstract
The present invention relates to pantograph running state monitoring fields.In view of the problems of the existing technology the present invention, provides a kind of pantograph image extraction method and device based on line-scan digital camera.According to the pantograph image of extraction, whether pantograph working condition is monitored extremely, to guide maintenance.The present invention carries out feature detection and pantograph to image to be checked according to established pantograph disaggregated model and just positions, if in image to be checked, there are pantographs, then pantograph is repositioned, judge pantograph pan and contact line position in image to be checked, is judged to whether there is real pantograph in image to be checked according to pantograph pan and contact line relative position;Otherwise, positioning step at the beginning of executing pantograph to next image to be checked.
Description
Technical field
The present invention relates to pantograph running state monitoring fields, are related to a kind of pantograph image zooming-out based on line-scan digital camera
Method and device.
Background technology
Pantograph pan monitoring device predominantly detects pantograph operating status.Camera is mounted on train by way of position, is determined
Point shooting pantograph image, can be monitored pantograph carbon slide abnormality, to guide maintenance.
Before, pantograph image taking is mainly area array cameras shooting, is mainly had the disadvantage that:1, area array cameras is fixed point
The case where triggering only shoots one to two photos, and hardware trigger mechanism is susceptible to false triggering, causes in captured image
There is no pantograph or only part pantograph;2, since pantograph monitoring is real time on-line monitoring, high speed rail train operation
Excessive velocities, and area array cameras frame per second does not catch up with train running speed, and the pantograph picture of shooting is caused to be easy smear.
Camera is fixedly mounted on train by way of position, roof of train is shot using line-scan digital camera, can be taken whole
A train top image, to extract complete pantograph image.
Pantograph image captured by line-scan digital camera has following feature, such as:1, light conditions differ, pantograph monitoring
Device is round-the-clock detection device, is influenced by sunlight daytime, is susceptible to the reflective serious conditions of pantograph, cause it is captured by
Pantograph is too bright;The general fast train in part is coal goods train, and roof is partially black, to which pantograph is also partially black;2, line-scan digital camera line frequency
It is mismatched with speed, in general, before monitoring device work, parameter setting can be carried out to camera line frequency and experience speed matches,
But the case where inevitably encountering acceleration suddenly in train travelling process, slowing down, line frequency and speed mismatch will lead to institute
The pantograph image fault of shooting, becomes large-sized or becomes smaller.3, each monitoring device installation point scene objective circumstances are different, due to
The installation personnel of different monitoring devices is different, and installation site is different, is easy so that the captured position of pantograph in the picture
It is not quite similar.Before, image captured by line-scan digital camera extracts pantograph extraction generally use traditional images processing method.
For the characteristics of image, traditional images processing method needs the algorithm parameter being arranged excessive, to ring captured by the above line-scan digital camera
Border is sensitive, and different installation points, which is likely to require, is arranged different parameters, and workload is very big, and the setting of parameter has experience
Property, image procossing professional is removed, parameter can not be preferably arranged in remaining personnel.
Invention content
The technical problem to be solved by the present invention is to:In view of the problems of the existing technology, it provides a kind of based on linear array phase
The pantograph image extraction method and device of machine.According to the pantograph image of extraction, it is whether abnormal to pantograph working condition into
Row monitoring, facilitates pantograph to repair.
In order to solve the above technical problems, the present invention using a kind of technical solution be to provide it is a kind of based on line-scan digital camera by electricity
Bending image extraction method includes:
Pantograph disaggregated model foundation step:After being pre-processed to the pantograph positive and negative samples image of collection;To all
Pantograph positive and negative samples image carries out characteristics extraction, obtains pantograph positive and negative samples image direction histogram of gradients feature;
It is trained study using the positive and negative sample image histograms of oriented gradients feature extracted, determines two kinds of training samples of segmentation
Optimal classification hyperplane, i.e. pantograph disaggregated model;
The first positioning step of pantograph:Histograms of oriented gradients feature extraction is carried out to image to be checked, is classified according to pantograph
Model carries out classification and Detection to it, judges to whether there is pantograph in the image to be checked for the first time;
Pantograph is accurately positioned step:If there are pantographs in image to be checked, pantograph is repositioned, is judged
Go out pantograph pan and contact line position, is determined whether really by electricity according to pantograph pan and contact line relative position
Bow;Otherwise, positioning step at the beginning of executing pantograph to next image to be checked.
Further, pantograph positive and negative samples image preprocessing includes:
Each pixel (x, y) in pantograph positive and negative samples image is traversed, it is the pixel 8 to take the new value of corresponding pixel points
The median of all pixels in neighborhood, for removing random salt-pepper noise, i.e. filtering and noise reduction in the positive negative image of pantograph;
The positive and negative image of pantograph after filtering and noise reduction is subjected to histogram equalization processing;
Denoising wherein can also be filtered by mean filter, gaussian filtering, bilateral filtering or Steerable filter.
Further, the method for feature extraction includes:
According to all pantograph positive and negative samples images of formula (1) normalized;
I (x, y)=I (x, y)gamma (1)
The gradient that the directions x and the directions y are carried out to pantograph positive and negative samples image calculates, and correspondence obtains every sample respectively
Pixel (x, y) in image is in x direction gradients and y direction gradients:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy (x, y)=H (x, y+1)-H (x, y-1) (3)
According to formula (2) (3), the gradient magnitude at pixel (x, y) and gradient direction are obtained;
Each pixel in every sample image is traversed, the gradient magnitude and gradient direction of all pixels point is calculated;
All gradient magnitudes and gradient direction are combined again, then carry out statistics with histogram calculating, obtain the direction gradient of whole figure
Histogram feature, and then obtain the histograms of oriented gradients feature of all pantograph positive and negative samples images.
Further, pantograph disaggregated model specific implementation process includes:
Classification linear equation setting steps:If { (xi,yi), i=1 ..., N } it is N number of sample point, wherein xiFor each
Sample image, yiIndicate xiAffiliated class, X ∈ RN, Y={ -1,1 }, the corresponding desired output of training sample set is yi∈{+1,-
1 }, x is indicatediAffiliated class (affiliated class:Refer to xiBelong to the class where positive sample or the class where negative sample, such as can will
The desired output of class where positive sample is 1, and the class desired output where negative sample is -1), X indicates all xi, for training number
According to;The classification logotype for belonging to positive sample and negative sample that desired output+1 respectively represents sample image with -1, X, Y are known
Amount carries out positive and negative sample labeling before creating positive and negative sample image to all sample images;W indicates that weights, b are offset
Amount, w, b are unknown quantity;And the classification linear equation of this linear space is;
Y=wX+b (6)
It finds an optimal classification line to separate different classes, while keeping class interval maximum, make two class samples while expiring
$ |Y|>=1 can be expressed as the condition of Y, and class interval is equal to 2/| at this time;|w||, because in two-dimensional space, Y=0 be just,
The classification line of negative sample, as Y=1 or -1, the distance for calculating point to Y=0 straight lines on the straight lines of Y=1 or -1 is Y/||w||,
That is 2/||w||;Class interval, which maximizes, is equivalent to following optimization problem:
MinG (w)=s ||w||2 (7)
Make yi[(w·xi)+b]- 1 >=0, i=1 ..., N, meet yi[(w·x)+b]The training sample of -1=0 is called
Supporting vector;
Lagrangian constitution step:Following Lagrangian is constructed, to be converted into quadratic programming problem:
In formula:A=(a1,a2,...,aN) it is Lagrange's multiplier;Minimum value in formula (7) is the saddle point of formula (8), can
The dual problem of formula (7) is converted into the partial derivative operation of w and b by L, finds a function the maximum of φ (a);
Constraints isIfFor its optimal solution, then local derviation letter is carried out to formula (8)
Number can be calculated:Then
Linear classification optimal classification construction of function step:Constructing optimal classification function f (x) according to formula (6) is:
Wherein,Then
Nonlinear Classification optimal separating hyper plane constitution step:It is non-with one for the sample characteristics point of Nonlinear separability
Sample characteristics point control is mapped to the feature space of a higher-dimension by linear function φ, and linear classification is carried out in feature space;If
Mapping phi can be found, then inner product operation (xiX) (φ (x can be usedi) φ (x)) replace;Usually use kernel function K (xi,x)
=(φ (xi) φ (x)) indicate inner product operation;
Therefore the optimal separating hyper plane function of linear classification and Nonlinear Classification is represented by:
Wherein,Then
Pantograph disaggregated model construction step:According to formula (10) and formula (11), a pantograph point is finally obtained
Class model.
Further, judge that whether there is pantograph detailed process in the image to be checked is for the first time:
Input the pantograph sequence image captured by line-scan digital camera;By adjacent two images head and the tail in pantograph sequence image
It is connected, forms image to be checked;
One region ROI to be checked is set in image to be checked, constantly moves ROI, moving step length n, according to feature extraction
Method extracts the histograms of oriented gradients feature of each ROI region image;
After the histograms of oriented gradients feature extraction of ROI region image, according to pantograph disaggregated model to every ROI
Area image carries out pantograph classification, detection, judges to whether there is pantograph in the image to be checked;
When determining a certain region ROI to be checked, there are pantographs, are no longer carried out to the pictures subsequent ROI region image to be checked
Pantograph just positions, and continues to carry out pantograph just positioning to next image to be checked.
Further, real pantograph detailed process is determined whether according to pantograph pan and contact line relative position
It is:
The pantograph image of pantograph is just oriented in input, and the calculating of X-direction gradient or the side Y are carried out to pantograph image
It is calculated to gradient, correspondence obtains X-direction and Y-direction gradient magnitude respectively;
Binary conversion treatment is carried out to X-direction and Y-direction gradient magnitude figure respectively, respectively corresponding X-direction and the Y-direction of obtaining
Gradient binary map;
All connected regions in X-direction and Y-direction gradient binary map are extracted, it is corresponding respectively to calculate in X-direction and Y-direction
The length and width and size of each connected region, and length and width and the ineligible connected region of size are filtered out, it obtains parallel
In the contact line suspicious region of y-axis, it is parallel to the pantograph carbon slide suspicious region of x-axis;
Whether change the contact line being pin-pointed to according to contact line suspicious region gray value;
Based on the contact line gradient binary map and pantograph carbon slide suspicious region gradient binary map being pin-pointed to, root
According to both contact line and pantograph carbon slide corresponding position relationship in practice, judge whether the pantograph carbon slide is real
Pantograph carbon slide;Realize that pantograph image is accurately positioned, extracts.
In order to solve the above technical problems, the present invention using a kind of technical solution be to provide it is a kind of based on line-scan digital camera by electricity
Bending image acquiring apparatus includes:
Pantograph model creation module:After being pre-processed for the pantograph positive and negative samples image to collection;To all
Pantograph positive and negative samples image carries out feature extraction, obtains pantograph positive and negative samples image direction histogram of gradients feature;Profit
It is trained study with the positive and negative sample image histograms of oriented gradients feature extracted, determines two kinds of training samples of segmentation most
Good Optimal Separating Hyperplane, i.e. pantograph disaggregated model;
The first locating module of pantograph:For carrying out histograms of oriented gradients feature extraction to image to be checked, according to pantograph
Disaggregated model carries out classification and Detection to it, judges to whether there is pantograph in the image to be checked for the first time;
Pantograph pinpoint module:For if there are pantographs in image to be checked, pantograph is repositioned,
Judge pantograph pan and contact line position, is determined whether really according to pantograph pan and contact line relative position
Pantograph;Otherwise, locating module at the beginning of executing pantograph to next image to be checked.
Further, the method for extraction includes:
According to all pantograph positive and negative samples images of formula (1) normalized;
I (x, y)=I (x, y)gamma (1)
The gradient that the directions x and the directions y are carried out to pantograph positive and negative samples image calculates, and correspondence obtains every sample respectively
Gradient of the pixel (x, y) in x direction gradients and the directions y in image be:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy (x, y)=H (x, y+1)-H (x, y-1) (3)
According to formula (2) (3), the gradient magnitude at pixel (x, y) and gradient direction are obtained;
Each pixel in every sample image is traversed, the gradient magnitude and gradient direction of all pixels point is calculated;
All gradient magnitudes and gradient direction are combined again, then carry out statistics with histogram calculating, obtain the direction gradient of whole figure
Histogram feature, and then obtain the histograms of oriented gradients feature of all pantograph positive and negative samples images.
Further, judge that whether there is pantograph detailed process in the image to be checked is for the first time:
Input the pantograph sequence image captured by line-scan digital camera;By adjacent two images head and the tail in pantograph sequence image
It is connected, forms image to be checked;
One region ROI to be checked is set in image to be checked, constantly moves ROI, moving step length n, according to feature extraction
Method extracts the histograms of oriented gradients feature of each ROI region image;
After the histograms of oriented gradients feature extraction of ROI region image, according to pantograph disaggregated model to every ROI
Area image carries out pantograph classification, detection, judges to whether there is pantograph in the image to be checked;
When determining a certain region ROI to be checked, there are pantographs, are no longer carried out to the pictures subsequent ROI region image to be checked
Pantograph positions, and continues to carry out pantograph just positioning to next image to be checked.
Further, real pantograph detailed process is determined whether according to pantograph pan and contact line relative position
It is:
The pantograph image of pantograph is just oriented in input, and the calculating of X-direction gradient or the side Y are carried out to pantograph image
It is calculated to gradient, correspondence obtains X-direction and Y-direction gradient magnitude respectively;
Binary conversion treatment is carried out to X-direction and Y-direction gradient magnitude respectively, correspondence obtains the ladder of X-direction and Y-direction respectively
Spend binary map;
All connected regions in X-direction and Y-direction gradient binary map are extracted, it is corresponding respectively to calculate in X-direction and Y-direction
The length and width and size of each connected region, and length and width and the ineligible connected region of size are filtered out, it obtains parallel
In the contact line suspicious region of y-axis, it is parallel to the pantograph carbon slide suspicious region of x-axis;
Whether change the contact line being pin-pointed to according to contact line suspicious region gray value;
Based on the contact line gradient binary map and pantograph carbon slide suspicious region gradient binary map being pin-pointed to, root
According to both contact line and pantograph carbon slide corresponding position relationship in practice, judge whether the pantograph carbon slide is real
Pantograph carbon slide;Realize that pantograph image is accurately positioned, extracts.
In conclusion by adopting the above-described technical solution, the beneficial effects of the invention are as follows:
1, because before carrying out pantograph image zooming-out, pretreatment is filtered to image first.It, can be with after being filtered
Some noises caused by filtering out in imaging process, reduce image contamination, and lifting feature extracts accuracy, improving pantograph just
Determine bit rate.
2, parameter modification is convenient in this method or device;During pantograph model foundation, it is not necessarily to professional, other
Layman can carry out the training of relevant parameter, modification, the pantograph disaggregated model for being suitble to oneself to need be obtained, after facilitating
Continuous image zooming-out.I.e. so that this method or device are applicable under multiple and different field conditions, the extraction of different type pantograph.
3, because this method or device carry out histogram equalization processing to pretreated image, image comparison is improved
Degree, and be normalized before feature extraction, therefore, uneven illumination is even, the figure of the relatively low situation of picture contrast
After carrying out histogram equalization processing so that this method or device change light source no longer quick in pantograph image zooming-out
Sense;And it ensure that pantograph extraction accuracy is high in positioning and pantograph fine positioning process at the beginning of subsequent pantograph, accidentally carry
Rate, leakage recovery rate is taken to reduce.This method or device are applicable to the pantograph image zooming-out under different light conditions, daytime, evening
On can use.
4, it is tentatively fixed to carry out pantograph by pantograph disaggregated model to testing image for this method or device
Position, in order to improve the accuracy of pantograph extraction, further, in conjunction with pantograph pan and contact line physical location, to be measured
The parameters such as pantograph pan position further measure in image, finally judge whether really there is pantograph in testing image.By
Pantograph initial survey is accurately positioned two steps with pantograph and is combined, and accurately located the pantograph in testing image, greatly improves
Detection efficiency, accuracy rate.
5, high in pantograph image zooming-out accuracy, in the case of accidentally recovery rate, leakage recovery rate reduce.To pantograph work shape
Whether exception monitoring also becomes simple to state, accelerates pantograph maintenance work.
Specific implementation mode
All features disclosed in this specification or disclosed all methods or in the process the step of, in addition to mutually exclusive
Feature and/or step other than, can combine in any way.
Any feature disclosed in this specification unless specifically stated can be equivalent or with similar purpose by other
Alternative features are replaced.That is, unless specifically stated, each feature is an example in a series of equivalent or similar characteristics
.
Related description of the present invention:
1, pantograph pan is parallel with the directions x;Contact line is parallel with Y-direction;X, Y-direction is pantograph two dimensional image pixel
The x-axis direction and the vertical direction vertical with x-axis direction (y-axis) of point;
2, under normal circumstances, sequence image is entire roof of train image;
Concrete operating principle of the present invention is:
Step 1:Collect the relevant positive and negative samples image of pantograph picture;
Collect photo captured by pantograph monitoring device (line-scan digital camera), it is extraction various types, various shooting angle, various
Pantograph picture under light conditions is positive sample, and non-pantograph region picture is negative sample image.
Step 2:Training pantograph positive and negative samples, pantograph algorithm model is established with machine learning algorithm;To positive and negative sample
This image carries out image preprocessing (image denoising, image enhancement), then carries out feature extraction first, reaches and is dropped to image
Dimension processing, reduces the purpose of redundancy, is trained study to image feature information using machine learning algorithm, obtains by electricity
Bend algorithm model.
1, image preprocessing:Pantograph positive and negative samples image is daytime, round-the-clock shooting in 24 hours at night, therefore can be deposited
Uneven illumination is even, the relatively low low situation of picture contrast;In addition, line-scan digital camera own electronic interferes, and in order to promote figure
The efficiency of transmission of picture, when acquisition, the jpeg format of use carries out pantograph original image some originals such as to compress image
Cause is easy so that there are noises in captured image.If do not pre-processed to the original image of array camera shooting, it will
Subsequent feature extraction accuracy is influenced, to influence pantograph locating accuracy.
The image preprocessing process taken herein is filtering and noise reduction, histogram equalization.
(1) filtering and noise reduction:(line-scan digital camera, line-scan digital camera are mounted on the hard crossbeam of train or branch to image scene collecting device
On column, in real time monitor pantograph working condition, to judge the pantograph work it is whether normal) acquisition pantograph image
It is middle to there is a large amount of random salt-pepper noise, for removal salt-pepper noise, take following method:Traverse each pixel in image (x,
Y), it is the median of all pixels in 8 neighborhood of pixel to take its new value, as follows:
G, (x, y)=median g (x-1, y-1), g (x, y-1), g (x+1, y-1), g (x-1, y), g (x, y), g (x+1,
y)
,g(x-1,y+1),g(x,y+1),g(x+1,y+1)}
Certain above-mentioned medium filtering can be replaced with mean filter or gaussian filtering;
(2) histogram equalization:That there are uneven illuminations is even due to the round-the-clock captured image of image scene collecting device,
The low situation of contrast, it is therefore desirable to enhancing processing is carried out to image, promotes the contrast of image, the enhancing processing used herein
Method is histogram equalization.The histogram of original graph is transformed to equally distributed form, which adds pixel gray levels
The dynamic range of value is to can reach the effect of enhancing image overall contrast ratio.If gray scale of the original image at (x, y) is f,
And the gray scale of the image after changing is g, then can be expressed as the gray scale f at (x, y) being mapped as g to the method for image enhancement.
G=EQ (f) may be defined as to the mapping function of image in gray-level histogram equalizationization processing, this mapping function EQ (f) must
Two conditions must be met (wherein L is the number of greyscale levels of image):(1) in 0≤f≤L-1, (L is maximum ash in whole image to EQ (f)
Degree) it is a monotonically increasing function in range.Ensure that image enhancement processing does not upset the gray scale ordering of original image, it is former
Scheme each gray level and still keeps arrangement from black to white (or from white to black) after the conversion.(2) there is 0≤g≤L- for 0≤f≤L-1
1, this condition ensure that the consistency of the front and back gray value dynamic range of transformation.Specifically mapping function is
Wherein k=(0,1,2 ... ..., L-1), n are sum of all pixels in image, njIndicate that gray scale is the sum of all pixels of j, according to the equation
The gray value of each pixel after histogram equalization can be directly obtained by each grey scale pixel value of original image.
2, feature extraction, with the method for feature extraction, extracts pantograph positive and negative samples before positive and negative samples image modeling
The feature of image obtains the histograms of oriented gradients feature of whole figure of the positive negative sample of pantograph;This process carries out dimensionality reduction to image
Processing reduces image redundancy information, improves algorithm process efficiency.The method detailed process of feature extraction is:
Feature extraction is carried out to every image, dimension-reduction treatment is carried out to image, reduces image redundancy information, is improved at algorithm
Manage efficiency;This method extraction is characterized as histograms of oriented gradients feature;
(1) it is primarily due to captured pantograph image irradiation situation to differ, to reduce the shadow of illumination factor and local shades
It rings, needs every figure being normalized.
I (x, y)=I (x, y)gamma (1)
(2) influence shone for further weakened light, need to carry out pantograph image in the directions x and the gradient in the directions y calculates.Point
The pixel (x, y) that Dui Ying do not obtain in every sample image is in x direction gradients and the directions y:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy (x, y)=H (x, y+1)-H (x, y-1) (3)
According to above equation, the gradient magnitude and gradient direction at pixel (x, y) can be obtained;
(3) feature synthesizes:Each pixel in every sample image is traversed, the gradient magnitude of all pixels point is calculated
And gradient direction;All gradient magnitudes and gradient direction are combined again, then carry out statistics with histogram calculating, obtain whole figure
Histograms of oriented gradients feature, and then obtain the histograms of oriented gradients feature of all pantograph positive and negative samples images.
Wherein, gradient magnitude and gradient direction anabolic process are:Whole image is divided into several cells cell first,
For example each cell is 4*4 pixel.Gradient direction is divided into 9 directions for 360 degree by us, using the histogram of 9 bin come
Count the gradient information of this 4*4 pixel.Such as:If the gradient direction of this pixel is 20-40 degree, the 2nd bin of histogram
Counting just add 1, in this way, being weighted projection in histogram with gradient direction to each pixel in cell (is mapped to fixation
Angular range), so that it may be exactly corresponding 9 dimensional feature vectors of the cell to obtain the gradient orientation histogram of this cell
(because having 9 bin).In addition, we are using gradient magnitude as the weights of projection.Such as:The gradient direction of this pixel is 20-
40 degree, it is assumed that its gradient magnitude is 2, then the counting of the 2nd bin of histogram is then to add 2.
The variation of the variation and foreground-background contrast shone due to local light so that the variation range of gradient intensity is non-
Chang great.This just needs to normalize gradient intensity.Normalization can further compress illumination, shade and edge.
Multiple cell are formed a block by us, in this way, the feature vector of all cell is together in series and just obtains in a block
The histograms of oriented gradients feature of the block.Finally, the feature vector of all block in image is together in series and is just somebody's turn to do
Histograms of oriented gradients feature in image.
Above-mentioned pantograph image can be the positive negative sample of pantograph or pantograph image to be measured etc..
3, machine learning:After obtaining the feature of all images, these features can be instructed by machine learning method
Practice study, a pantograph algorithm model is obtained, herein using support vector machines (svm) machine learning algorithm.Support to
Amount machine is mainly used for classifying to data, looks for a divisional plane (line) that sample point is separated in space, divisional plane (line)
Optimal conditions is exactly that class interval maximizes, and class interval is the distance based on point to plane (straight line) to calculate.
(1) related description non-linear during pantograph model foundation and linear classification:
With the development of science and technology, people often encounter various types of mass datas in actual application,
Such as securities market transaction data, multimedia graphic image/video data, space flight and aviation gathered data, biological attribute data, this
A little data are commonly referred to as high dimensional data in statistical disposition
About data whether linear separability:For low-dimensional data, we can draw image, can intuitively find out
Come.But for whether being high dimensional data, generally judge by the way that whether calculating convex closure intersects.Convex closure is exactly that a convex closure is bent
Line (curved surface), and it has just encased all data.When we draw the convex closure of two classes, if the two is not overlapped, that
The two linear separability, it is on the contrary then be not linear separability.
(2) the specific implementation process of pantograph model is:
Classification linear equation setting steps:If { (xi,yi), i=1 ..., N } it is N number of sample point, wherein xiFor each
Sample image, yiIndicate xiAffiliated class, X ∈ RN, Y={ -1,1 }, the corresponding desired output of training sample set are yi∈{+1,-
1 }, x is indicatediAffiliated class (affiliated class:Refer to xiBelong to the class where positive sample or the class where negative sample, such as can will
The desired output of class where positive sample is 1, and the class desired output where negative sample is -1), X indicates all xi, for training number
According to;The classification logotype for belonging to positive sample and negative sample that desired output+1 respectively represents sample image with -1, X, Y are known
Amount carries out positive and negative sample labeling before creating positive and negative sample image to all sample images;W indicates that weights, b are offset
Amount, w, b are unknown quantity;And the classification linear equation of this linear space is;
Y=wX+b (6)
It finds an optimal classification line to separate different classes, while keeping class interval maximum, make two class samples while expiring
C_SUB_SEM[AJ |Y|>=1 can be expressed as the condition of Y, and class interval is equal to 2/| at this time;|w||, because in two-dimensional space, Y=0 be just,
The classification line of negative sample, as Y=1 or -1, the distance for calculating point to Y=0 straight lines on the straight lines of Y=1 or -1 is Y/||w||,
That is 2/||w||;Class interval, which maximizes, is equivalent to following optimization problem:
MinG (w)=s ||w||2 (7)
Make yi[(w·xi)+b]- 1 >=0, i=1 ..., N, meet yi[(w·x)+b]The training sample of -1=0 is called
Supporting vector;
Lagrangian constitution step:Following Lagrangian is constructed, to be converted into quadratic programming problem:
In formula:A=(a1,a2,...,aN) it is Lagrange's multiplier.Minimum value in formula (7) is the saddle point of formula (8), can
The dual problem of formula (7) is converted into the partial derivative operation of w and b by L, finds a function the maximum of φ (a);
Constraints isIfFor its optimal solution, then local derviation letter is carried out to formula (8)
Number can be calculated:Then
Linear classification optimal classification construction of function step:Constructing optimal classification function f (x) according to formula (6) is:
Wherein,Then
Nonlinear Classification optimal separating hyper plane constitution step:It is non-with one for the sample characteristics point of Nonlinear separability
Sample characteristics point control is mapped to the feature space of a higher-dimension by linear function φ, and linear classification is carried out in feature space.If
Mapping phi can be found, then inner product operation (xiX) (φ (x can be usedi) φ (x)) replace.Usually use kernel function K (xi,x)
=(φ (xi) φ (x)) indicate inner product operation;
Therefore Nonlinear Classification optimal separating hyper plane function is represented by:
Wherein,Then
Pantograph disaggregated model construction step:A pantograph disaggregated model is finally obtained according to formula (10), (11),
Middle nonlinear data regards pantograph disaggregated model with Nonlinear Classification optimal separating hyper plane function (formula 11);Linear data
Use linear classification optimal classification function (formula 10) as pantograph disaggregated model.
Step 3:Pantograph just positions:Input pantograph sequence image (under normal circumstances, the sequence captured by line-scan digital camera
Row image is entire roof of train image), adjacent two images in pantograph sequence image are joined end to end, figure to be checked is formed
Picture.Because being possible to only part pantograph in single image, for example, a part for pantograph is in the bottom of previous image,
Another part, need to be by adjacent two image mosaics at the image to be checked of a pantograph to be extracted at the top of latter image;
One particular size region ROI to be checked is set in image to be checked, constantly moves ROI, moving step length n, according to special
Levy the feature of extracting method extraction ROI region image.Feature extraction finishes, can be to the figure that extracts according to pantograph disaggregated model
As feature is classified, judge to whether there is pantograph in the image to be checked.Because of the image to be checked of adjacent two images synthesis
A certain region ROI to be checked may be determined there are one pantograph, to save efficiency if there is pantograph, it is no longer to be checked to this
Image ROI region image is subsequently positioned, and continues to carry out pantograph just positioning to next image to be checked.
Step 4:Pantograph fine positioning:In roof of train image, due to some regions (such as compartment junction) and
The characteristic similarity in pantograph region is higher, flase drop is had so being tentatively set in the image, in order to reduce pantograph
It accidentally positions, secondary positioning, filtering need to be carried out to the pantograph image that Primary Location goes out;The thinking of secondary positioning is to judge pantograph
The position of carbon slipper and contact line;In train travelling process, pantograph is to need to take electricity to the contact line of Along Railway, therefore
In image, pantograph carbon slide and contact line there are certain location-prior relationship (in a set of pantograph image picking-up apparatus,
Pantograph rising bow takes electricity, so rising bow and contact line are at cross reference;Pantograph drop bow does not work and contact line is non-intersecting, but
It is fixed with the positional distance of contact line.In addition, in the case that hardware condition is constant, the position of contact line in the picture is relatively fixed
It is constant).
Pantograph pan Primary Location:The pantograph image that input Primary Location goes out, to overcome illumination effect, first to figure
Gradient as carrying out the directions y calculates.It is special can be regarded as pantograph primary structure for pantograph carbon slide in pantograph structure
Sign.From image, pantograph carbon slide is almost perpendicular to car body, perpendicular to y-axis, so we are to the figure just oriented
Gradient as carrying out the directions y calculates, specific as shown in formula 3, to filter out the gradient information in other directions in image.Due to ladder
Degree amplitude illustrates that the intensity of pixel mutation, pantograph pan edge are that pixel is mutated stronger region, we can be to gradient magnitude
Binary conversion treatment is carried out, a certain threshold value is set, by GyThe pixel placement that (x, y) is more than threshold value is 255, and rest of pixels is set as 0.
Many noises are had after gradient image binaryzation, can be carried out morphologic filtering expansion and corrosion treatment to binary map, be filtered out noise.
Finally, all connected regions in image are extracted, length, width and size of each connected region are calculated.Set certain
One threshold value filters out the ineligible connected region of length, width and size, be finally parallel to x-axis by electricity
Bend carbon slipper suspicious region.
Contact line positions:The pantograph image of pantograph is just oriented in input, to overcome illumination effect, first to image
The gradient for carrying out the directions x calculates.From image, contact line is perpendicular to x-axis, so we carry out x to the image just oriented
The gradient in direction calculates, specific as shown in formula 2, to filter out the gradient information in other directions in image.Due to gradient magnitude
Illustrate the intensity of pixel mutation, we can carry out binary conversion treatment to gradient magnitude, a certain threshold value be set, by Gy(x, y) is big
It is 255 in the pixel placement of threshold value, rest of pixels is set as 0.Then, binary map is carried out at morphologic filtering expansion and corrosion
Reason, filters out noise.Finally, all connected regions in image are extracted, length (length), the width of each connected region are calculated
(width) and size (size).It sets certain threshold value and filters out length (length), width (width) and size (size) no
Qualified connected region obtains the final contact line suspicious region for being parallel to y-axis.Since image is clapped by line-scan digital camera
It takes the photograph, so having the interference region similar to contact line in image.But contact line usage time is permanent, has abrasion, spot
Situations such as appearance, the contact line grey scale pixel value in this sampled images can change that (other interference regions do not have gray scale change
Change), so we can filter out other interfering lines according to whether final contact line suspicious region gray value changes, to
To the contact line being pin-pointed to.
After obtaining pantograph carbon slide suspicious region gradient binary map and the contact line gradient binary map that is pin-pointed to, root
According to both contact line and pantograph carbon slide corresponding position relationship in practice, you can judge the pantograph carbon slide whether be
Real pantograph pan realizes pantograph image fine positioning, extraction to filter out the pantograph image of accidentally positioning.
The invention is not limited in specific implementation modes above-mentioned.The present invention, which expands to, any in the present specification to be disclosed
New feature or any new combination, and disclose any new method or process the step of or any new combination.
Claims (10)
1. a kind of pantograph image extraction method based on line-scan digital camera, it is characterised in that including:
Pantograph disaggregated model foundation step:After being pre-processed to the pantograph positive and negative samples image of collection;To all by electricity
Bend positive and negative samples image and carry out characteristics extraction, obtains pantograph positive and negative samples image direction histogram of gradients feature;It utilizes
The positive and negative sample image histograms of oriented gradients feature extracted is trained study, determines the best of two kinds of training samples of segmentation
Optimal Separating Hyperplane, i.e. pantograph disaggregated model;
The first positioning step of pantograph:Histograms of oriented gradients feature extraction is carried out to image to be checked, according to pantograph disaggregated model
Classification and Detection is carried out to it, judges to whether there is pantograph in the image to be checked for the first time;
Pantograph is accurately positioned step:If there are pantographs in image to be checked, pantograph is repositioned, judge by
Pantograph slider and contact line position, determine whether real pantograph according to pantograph pan and contact line relative position;
Otherwise, positioning step at the beginning of executing pantograph to next image to be checked.
2. a kind of pantograph image extraction method based on line-scan digital camera according to claim 1, it is characterised in that by electricity
Bending positive and negative samples image preprocessing includes:
Each pixel (x, y) in pantograph positive and negative samples image is traversed, it is 8 neighborhood of pixel to take the new value of corresponding pixel points
The median of interior all pixels, for removing random salt-pepper noise, i.e. filtering and noise reduction in pantograph positive and negative samples image;
The positive and negative image of pantograph after filtering and noise reduction is subjected to histogram equalization processing;
Denoising wherein can also be filtered by mean filter, gaussian filtering, bilateral filtering or Steerable filter.
3. a kind of pantograph image extraction method based on line-scan digital camera according to claim 1, it is characterised in that feature
The method of extraction includes:
According to all pantograph positive and negative samples images of formula (1) normalized;
I (x, y)=I (x, y)gamma (1)
The gradient that the directions x and the directions y are carried out to pantograph positive and negative samples image calculates, and correspondence obtains every sample image respectively
In pixel (x, y) be in x direction gradients and y direction gradients:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy (x, y)=H (x, y+1)-H (x, y-1) (3)
According to formula (2) (3), the gradient magnitude at pixel (x, y) and gradient direction are obtained;
Each pixel in every sample image is traversed, the gradient magnitude and gradient direction of all pixels point is calculated;It is right again
All gradient magnitudes and gradient direction are combined, then carry out statistics with histogram calculating, obtain the direction gradient histogram of whole figure
Figure feature, and then obtain the histograms of oriented gradients feature of all pantograph positive and negative samples images.
4. a kind of pantograph image extraction method based on line-scan digital camera according to claim 3, it is characterised in that by electricity
Bow disaggregated model implements process:
Classification linear equation setting steps:If { (xi,yi), i=1 ..., N } it is N number of sample point, wherein xiFor each sample
Image, yiIndicate xiAffiliated class, X ∈ RN, Y={ -1,1 }, the corresponding desired output of training sample set is yi∈ {+1, -1 }, table
Show xiAffiliated class, X indicate all xi, it is training data;Desired output+1 and -1 respectively represent sample image belong to positive sample
With the classification logotype of negative sample, X, Y are known quantities, are carried out to all sample images before creating positive and negative sample image
Positive and negative sample labeling;W indicates that weights, b are offset, and w, b are unknown quantity;And the classification linear equation of this linear space is;
Y=wX+b (6)
It finds an optimal classification line to separate different classes, while keeping class interval maximum, make two class samples while meeting Y|
>=1 can be expressed as the condition of Y, and class interval is equal to 2/| at this time;|w||, because in two-dimensional space, Y=0 is positive and negative samples
Classification line, as Y=1 or -1, calculate the straight lines of Y=1 or -1 on point to Y=0 straight lines distance be Y/||w||, and i.e. 2/|
|w||;Class interval, which maximizes, is equivalent to following optimization problem:
MinG (w)=s ||w||2 (7)
Make yi[(w·xi)+b]- 1 >=0, i=1 ..., N, meet yi[(w·x)+b]The training sample of -1=0 is called support
Vector;
Lagrangian constitution step:Following Lagrangian is constructed, to be converted into quadratic programming problem:
In formula:A=(a1,a2,...,aN) it is Lagrange's multiplier;Minimum value in formula (7) is the saddle point of formula (8), can be passed through
L is converted into the partial derivative operation of w and b the dual problem of formula (7), finds a function the maximum of φ (a);
Constraints isIfFor its optimal solution, then partial derivative meter is carried out to formula (8)
It calculates available:Then
Linear classification optimal classification construction of function step:Constructing optimal classification function f (x) according to formula (6) is:
Wherein,Then
Nonlinear Classification optimal separating hyper plane constitution step:It is non-linear with one for the sample characteristics point of Nonlinear separability
Sample characteristics point control is mapped to the feature space of a higher-dimension by function phi, and linear classification is carried out in feature space;If can look for
To mapping phi, then inner product operation (xiX) (φ (x can be usedi) φ (x)) replace;Usually use kernel function K (xi, x) and=(φ
(xi) φ (x)) indicate inner product operation;
Therefore the optimal separating hyper plane function of linear classification and Nonlinear Classification is represented by:
Wherein,Then
Pantograph disaggregated model construction step:According to formula (10) and formula (11), a pantograph classification mould is finally obtained
Type.
5. a kind of pantograph image extraction method based on line-scan digital camera according to claim 3, it is characterised in that first
Judge that whether there is pantograph detailed process in the image to be checked is:
Input the pantograph sequence image captured by line-scan digital camera;By adjacent two images head and the tail phase in pantograph sequence image
Even, image to be checked is formed;
One region ROI to be checked is set in image to be checked, constantly moves ROI, moving step length n, the method according to feature extraction
Extract the histograms of oriented gradients feature of each ROI region image;
After the histograms of oriented gradients feature extraction of ROI region image, according to pantograph disaggregated model to each ROI region
Image carries out pantograph classification, detection, judges to whether there is pantograph in the image to be checked;
When determining a certain region ROI to be checked, there are pantographs, no longer carry out the pictures subsequent ROI region image to be checked by electricity
Bow just positioning continues to position at the beginning of carrying out pantograph to next image to be checked.
6. a kind of pantograph image extraction method based on line-scan digital camera according to claim 3, it is characterised in that according to
Pantograph pan and contact line relative position determine whether that real pantograph detailed process is:
The pantograph image of pantograph is just oriented in input, and the calculating of X-direction gradient or Y-direction ladder are carried out to pantograph image
Degree calculates, and correspondence obtains X-direction and Y-direction gradient magnitude respectively;
Binary conversion treatment is carried out to X-direction and Y-direction gradient magnitude figure respectively, correspondence obtains the gradient of X-direction and Y-direction respectively
Binary map;
All connected regions in X-direction and Y-direction gradient binary map are extracted, are corresponded to respectively each in calculating X-direction and Y-direction
The length and width and size of connected region, and length and width and the ineligible connected region of size are filtered out, it obtains being parallel to y-axis
Contact line suspicious region, be parallel to the pantograph carbon slide suspicious region of x-axis;
Whether change the contact line being pin-pointed to according to contact line suspicious region gray value;
Based on the contact line gradient binary map and pantograph carbon slide suspicious region gradient binary map being pin-pointed to, according to connecing
Touch both line and pantograph carbon slide corresponding position relationship in practice, judge the pantograph carbon slide whether be really by
Pantograph carbon slipper;Realize that pantograph image is accurately positioned, extracts.
7. a kind of pantograph image acquiring apparatus based on line-scan digital camera, it is characterised in that including:
Pantograph disaggregated model creation module:After being pre-processed for the pantograph positive and negative samples image to collection;To all
Pantograph positive and negative samples image carries out feature extraction, obtains pantograph positive and negative samples image direction histogram of gradients feature;Profit
It is trained study with the positive and negative sample image histograms of oriented gradients feature extracted, determines two kinds of training samples of segmentation most
Good Optimal Separating Hyperplane, i.e. pantograph disaggregated model;
The first locating module of pantograph:For carrying out histograms of oriented gradients feature extraction to image to be checked, classified according to pantograph
Model carries out classification and Detection to it, judges to whether there is pantograph in the image to be checked for the first time;
Pantograph pinpoint module:For if there are pantographs in image to be checked, pantograph is repositioned, is judged
Go out pantograph pan and contact line position, is determined whether really by electricity according to pantograph pan and contact line relative position
Bow;Otherwise, locating module at the beginning of executing pantograph to next image to be checked.
8. a kind of pantograph image acquiring apparatus based on line-scan digital camera according to claim 7, it is characterised in that feature
The method of extraction includes:
According to all pantograph positive and negative samples images of formula (1) normalized;
I (x, y)=I (x, y)gamma (1)
The gradient that the directions x and the directions y are carried out to pantograph positive and negative samples image calculates, and correspondence obtains every sample image respectively
In gradient of the pixel (x, y) in x direction gradients and the directions y be:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy (x, y)=H (x, y+1)-H (x, y-1) (3)
According to formula (2) (3), the gradient magnitude at pixel (x, y) and gradient direction are obtained;
Each pixel in every sample image is traversed, the gradient magnitude and gradient direction of all pixels point is calculated;It is right again
All gradient magnitudes and gradient direction are combined, then carry out statistics with histogram calculating, obtain the direction gradient histogram of whole figure
Figure feature, and then obtain the histograms of oriented gradients feature of all pantograph positive and negative samples images.
9. a kind of pantograph image acquiring apparatus based on line-scan digital camera according to claim 8, it is characterised in that first
Judge that whether there is pantograph detailed process in the image to be checked is:
Input the pantograph sequence image captured by line-scan digital camera;By adjacent two images head and the tail phase in pantograph sequence image
Even, image to be checked is formed;
One region ROI to be checked is set in image to be checked, constantly moves ROI, moving step length n, the method according to feature extraction
Extract the histograms of oriented gradients feature of each ROI region image;
After the histograms of oriented gradients feature extraction of ROI region image, according to pantograph disaggregated model to each ROI region
Image carries out pantograph classification, detection, judges to whether there is pantograph in the image to be checked;
When determining a certain region ROI to be checked, there are pantographs, no longer carry out the pictures subsequent ROI region image to be checked by electricity
Bow positioning continues to carry out pantograph just positioning to next image to be checked.
10. a kind of pantograph image acquiring apparatus based on line-scan digital camera according to claim 8, it is characterised in that according to
Pantograph pan and contact line relative position determine whether that real pantograph detailed process is:
The pantograph image of pantograph is just oriented in input, and the calculating of X-direction gradient or Y-direction ladder are carried out to pantograph image
Degree calculates, and correspondence obtains X-direction and Y-direction gradient magnitude respectively;
Binary conversion treatment is carried out to X-direction and Y-direction gradient magnitude figure respectively, correspondence obtains the gradient of X-direction and Y-direction respectively
Binary map;
All connected regions in X-direction and Y-direction gradient binary map are extracted, are corresponded to respectively each in calculating X-direction and Y-direction
The length and width and size of connected region, and length and width and the ineligible connected region of size are filtered out, it obtains being parallel to y-axis
Contact line suspicious region, be parallel to the pantograph carbon slide suspicious region of x-axis;
Whether change the contact line being pin-pointed to according to contact line suspicious region gray value;
Based on the contact line gradient binary map and pantograph carbon slide suspicious region gradient binary map being pin-pointed to, according to connecing
Touch both line and pantograph carbon slide corresponding position relationship in practice, judge the pantograph carbon slide whether be really by
Pantograph carbon slipper;Realize that pantograph image is accurately positioned, extracts.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710223798.XA CN108694349B (en) | 2017-04-07 | 2017-04-07 | Pantograph image extraction method and device based on linear array camera |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710223798.XA CN108694349B (en) | 2017-04-07 | 2017-04-07 | Pantograph image extraction method and device based on linear array camera |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108694349A true CN108694349A (en) | 2018-10-23 |
| CN108694349B CN108694349B (en) | 2020-03-06 |
Family
ID=63843046
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710223798.XA Active CN108694349B (en) | 2017-04-07 | 2017-04-07 | Pantograph image extraction method and device based on linear array camera |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108694349B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110763681A (en) * | 2019-08-23 | 2020-02-07 | 南京理工大学 | Pantograph abrasion area positioning detection method |
| CN110910443A (en) * | 2019-12-04 | 2020-03-24 | 成都唐源电气股份有限公司 | Contact net geometric parameter real-time measuring method and device based on single monitoring camera |
| CN111260631A (en) * | 2020-01-16 | 2020-06-09 | 成都地铁运营有限公司 | Efficient rigid contact line structure light strip extraction method |
| CN111260629A (en) * | 2020-01-16 | 2020-06-09 | 成都地铁运营有限公司 | Pantograph structure abnormity detection algorithm based on image processing |
| CN111666947A (en) * | 2020-05-26 | 2020-09-15 | 成都唐源电气股份有限公司 | Pantograph head offset measurement method and system based on 3D imaging |
| CN113487561A (en) * | 2021-07-02 | 2021-10-08 | 成都唐源电气股份有限公司 | Pantograph foreign matter detection method and device based on gray gradient abnormal voting |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745238A (en) * | 2013-11-15 | 2014-04-23 | 中国科学院遥感与数字地球研究所 | Pantograph identification method based on AdaBoost and active shape model |
| CN104374373A (en) * | 2014-10-15 | 2015-02-25 | 中铁电气化局集团有限公司 | Catenary status monitoring system based on pantograph image analysis |
| CN104463240A (en) * | 2013-09-23 | 2015-03-25 | 深圳市朗驰欣创科技有限公司 | Method and device for controlling list interface |
| CN105652154A (en) * | 2016-01-25 | 2016-06-08 | 成都国铁电气设备有限公司 | Safety monitoring analysis system for contact net running state |
| CN106052575A (en) * | 2016-08-02 | 2016-10-26 | 易讯科技股份有限公司 | Pantograph carbon slide plate wear online detection method based on train high speed running |
-
2017
- 2017-04-07 CN CN201710223798.XA patent/CN108694349B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463240A (en) * | 2013-09-23 | 2015-03-25 | 深圳市朗驰欣创科技有限公司 | Method and device for controlling list interface |
| CN103745238A (en) * | 2013-11-15 | 2014-04-23 | 中国科学院遥感与数字地球研究所 | Pantograph identification method based on AdaBoost and active shape model |
| CN104374373A (en) * | 2014-10-15 | 2015-02-25 | 中铁电气化局集团有限公司 | Catenary status monitoring system based on pantograph image analysis |
| CN105652154A (en) * | 2016-01-25 | 2016-06-08 | 成都国铁电气设备有限公司 | Safety monitoring analysis system for contact net running state |
| CN106052575A (en) * | 2016-08-02 | 2016-10-26 | 易讯科技股份有限公司 | Pantograph carbon slide plate wear online detection method based on train high speed running |
Non-Patent Citations (3)
| Title |
|---|
| SAML BARMADA ET AL.: "Arc detection in pantograph-catenary systems by the use of support vector machines-based classification", 《IET ELECTRICAL SYSTEM IN TRANSPORTATION》 * |
| 刘凤民 等: "《岩心扫描及应用》", 30 September 2010, 地质出版社 * |
| 肖怀铁 等: "《基于核方法的雷达高分辨距离像目标识别理论与方法》", 30 September 2015, 国防工业出版社 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110763681A (en) * | 2019-08-23 | 2020-02-07 | 南京理工大学 | Pantograph abrasion area positioning detection method |
| CN110763681B (en) * | 2019-08-23 | 2022-06-10 | 南京理工大学 | Pantograph abrasion area positioning detection method |
| CN110910443A (en) * | 2019-12-04 | 2020-03-24 | 成都唐源电气股份有限公司 | Contact net geometric parameter real-time measuring method and device based on single monitoring camera |
| CN110910443B (en) * | 2019-12-04 | 2023-03-21 | 成都唐源电气股份有限公司 | Contact net geometric parameter real-time measuring method and device based on single monitoring camera |
| CN111260631A (en) * | 2020-01-16 | 2020-06-09 | 成都地铁运营有限公司 | Efficient rigid contact line structure light strip extraction method |
| CN111260629A (en) * | 2020-01-16 | 2020-06-09 | 成都地铁运营有限公司 | Pantograph structure abnormity detection algorithm based on image processing |
| CN111260631B (en) * | 2020-01-16 | 2023-05-05 | 成都地铁运营有限公司 | Efficient rigid contact line structure light bar extraction method |
| CN111666947A (en) * | 2020-05-26 | 2020-09-15 | 成都唐源电气股份有限公司 | Pantograph head offset measurement method and system based on 3D imaging |
| CN111666947B (en) * | 2020-05-26 | 2023-08-04 | 成都唐源电气股份有限公司 | Pantograph head offset measuring method and system based on 3D imaging |
| CN113487561A (en) * | 2021-07-02 | 2021-10-08 | 成都唐源电气股份有限公司 | Pantograph foreign matter detection method and device based on gray gradient abnormal voting |
| CN113487561B (en) * | 2021-07-02 | 2023-06-30 | 成都唐源电气股份有限公司 | Pantograph foreign matter detection method and device based on gray gradient abnormal voting |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108694349B (en) | 2020-03-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108694349A (en) | A kind of pantograph image extraction method and device based on line-scan digital camera | |
| Wei et al. | Multi-target defect identification for railway track line based on image processing and improved YOLOv3 model | |
| CN107253485B (en) | Foreign matter invades detection method and foreign matter invades detection device | |
| CN107256225B (en) | Method and device for generating heat map based on video analysis | |
| CN104978567B (en) | Vehicle checking method based on scene classification | |
| WO2019196131A1 (en) | Method and apparatus for filtering regions of interest for vehicle-mounted thermal imaging pedestrian detection | |
| CN107909081B (en) | Method for quickly acquiring and quickly calibrating image data set in deep learning | |
| CN106530281B (en) | Unmanned plane image fuzzy Judgment method and system based on edge feature | |
| CN105404857A (en) | Infrared-based night intelligent vehicle front pedestrian detection method | |
| CN113324864B (en) | Pantograph carbon slide plate abrasion detection method based on deep learning target detection | |
| CN107944354B (en) | Vehicle detection method based on deep learning | |
| CN103530638A (en) | Method for matching pedestrians under multiple cameras | |
| Kim et al. | Autonomous vehicle detection system using visible and infrared camera | |
| CN110070545B (en) | Method for automatically extracting urban built-up area by urban texture feature density | |
| CN110414571A (en) | A kind of website based on Fusion Features reports an error screenshot classification method | |
| CN110222661A (en) | It is a kind of for motion estimate and the feature extracting method of tracking | |
| CN111461003A (en) | Coal-fired working condition identification method based on video image sequence feature extraction | |
| CN108629776B (en) | Mineral rock granularity detecting system | |
| CN109815784A (en) | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium | |
| CN106997583A (en) | A kind of underwater robot adaptive image enhancement and feature extracting method | |
| CN103578121A (en) | Motion detection method based on shared Gaussian model in disturbed motion environment | |
| Pan et al. | A super-resolution-based license plate recognition method for remote surveillance | |
| CN111089586A (en) | All-day star sensor star point extraction method based on multi-frame accumulation algorithm | |
| CN107330436B (en) | Scale criterion-based panoramic image SIFT optimization method | |
| CN105740834B (en) | To the high-precision detecting method of front vehicles under overnight sight |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |