CN106679630B - A kind of contact net positioner slope detection system - Google Patents
A kind of contact net positioner slope detection system Download PDFInfo
- Publication number
- CN106679630B CN106679630B CN201710029167.4A CN201710029167A CN106679630B CN 106679630 B CN106679630 B CN 106679630B CN 201710029167 A CN201710029167 A CN 201710029167A CN 106679630 B CN106679630 B CN 106679630B
- Authority
- CN
- China
- Prior art keywords
- locator
- image
- camera
- gradient
- contact net
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C9/00—Measuring inclination, e.g. by clinometers, by levels
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
- G01C11/36—Videogrammetry, i.e. electronic processing of video signals from a single source or from different sources to give parallax or range information
Abstract
The invention discloses a kind of contact net positioner slope detection systems, the system comprises video acquisition units and main computer unit, wherein: video acquisition unit is installed on roof of train, image acquisition units contact Running State video for acquiring in real time, and video image is real-time transmitted to interior main computer unit and carries out data analysis;Main computer unit is installed on interior, receives the vision signal of video acquisition unit and carries out compression storage, and calculates the locator gradient according to bow net image data is obtained, and realizes the technical effect of efficiently and accurately detected to the locator gradient.
Description
Technical field
The present invention relates to electric railway safety testing fields, and in particular, to a kind of locator slope detection system.
Background technique
Develop the inexorable trend that electric railway is railway modernization construction.And electric railway is all made of electric propulsion,
Electric locomotive must reliably obtain electric energy under the conditions of high-speed cruising from contact net, otherwise will affect train operation and electrical
The performance of drive system.In order to reduce the abrasion in contact net during train operation, contact net is usually along rail overhead
What "the" shape was set up.Catenary erection system is generally by contact suspension, support device, positioning device, several portions of pillar and basis
It is grouped as.
Positioning device is fixed contact line, guarantees the important device that contact line is set up in the reasonable scope, its installation essence
Degree directly affects the geometric parameter of contact line, is related to that contact net is continual and steady to power to pantograph.Locator is positioning dress
The component directly contacted with contact line in setting, the locator gradient are the closely related contact net self structures of pantograph operational safety
Parameter.In order to avoid during pantograph sliding and running with locator collision and caused by beat bow event, to locator gradient model
Enclosing should there are certain requirements.
It is more in railway systems at present to use contact net static measuring instrument, it is general by hand-held or vehicle-mounted adjustable
The method of the laser measurement and positioning device two o'clock upright projection height difference of detecting distance processed calculates the locator gradient.However, this survey
Amount mode needs to carry out the work in the railway non-operation period, and measurement efficiency is low.Due to China express railway fast development, contact
Net installation accuracy requires to be continuously improved, growing to contact net line service.Therefore, traditional contact net static detection
Method is unable to satisfy the contact net system locator detection demand of current fast-developing High-speed Railway Network.
In conclusion present inventor has found above-mentioned technology extremely during realizing the present application technical solution
It has the following technical problems less:
In the prior art, there are the poor technologies of detection efficiency and accuracy rate to ask for existing locator slope detection method
Topic.
Summary of the invention
The present invention provides a kind of contact net positioner slope detection systems, solve existing locator slope detection side
The method technical problem poor there are detection efficiency and accuracy rate, realizes the skill of efficiently and accurately detected to the locator gradient
Art effect.
In order to solve the above technical problems, this application provides a kind of contact net positioner slope detection system, the system
Including video acquisition unit and main computer unit, in which:
Video acquisition unit is installed on roof of train, and image acquisition units for acquiring contact Running State view in real time
Frequently, and video image is real-time transmitted to interior main computer unit and carries out data analysis;
Main computer unit is installed on interior, receives the vision signal of video acquisition unit and carries out compression storage, and root
The locator gradient is calculated according to bow net image data is obtained.
Further, video acquisition unit includes: mounting seat, camera, supplementary lighting sources;Mounting seat and roof interface are solid
Fixed connection;Supplementary lighting sources and camera are both secured in the mounting seat, and the injection optical registration contact of camera and supplementary lighting sources
Net system frame and locator.
Further, the system also includes power management modules and communication control module;The power management module is used
It powers in for video acquisition unit and main computer unit;The communication control module is used for the video figure for acquiring video acquisition unit
As being transferred to main computer unit.
Further, the system also includes data memory modules;The data memory module is for saving video acquisition
The video image of unit acquisition.
Further, the first CPU module is equipped in the main computer unit;First CPU module is for passing through depth
It practises algorithm and rough detection is carried out to locator, specifically include:
Step 1: acquisition localizer image information first is then based on localizer image information architecture deep learning model;
Step 2: being based on bow net image, construct locator training sample;
Step 3: based on the training sample training deep learning model in step 2;
Step 4: based on the deep learning model after training, the image containing locator to be detected being handled, is obtained
The location information of locator to be detected;
Step 5: based on the location information in step 4, to the local image region comprising locator, to the positioning in image
Device is accurately positioned, and locator straight line is fitted;
Step 6: projective transformation being carried out to localizer image, coordinate of the locator under world coordinate system is calculated, obtains generation
Boundary's coordinate system lower locating device gradient.
Further, the step 1 specifically includes:
Step 1.1: building include several hidden layers depth convolutional network, in which: including convolutional layer, down-sampled layer and
Full articulamentum;The local feature of image is extracted by convolutional calculation;
Step 1.2: down-sampled processing is done by local feature of the pond to image;
Step 1.3: further successively feature extraction processing is done by full articulamentum;
Step 1.4: class probability and the exact position of locator are predicted by classifier layer
Step 1.5: defining classification device loss function, including classification loss and position loss.
Further, the step 4 specifically includes:
Step 4.1: the depth convolutional network after inputting Optimal Parameters for image to be detected;
Step 4.2: obtaining the network output valve of test image
Step 4.3: locator classification and location information can be obtained according to network output valve;
Further, the second CPU module is additionally provided in the main computer unit;Second CPU module is used for locator
Region carries out examining survey, and calculates the gradient of the locator under world coordinate system by monocular vision solid geometry relationship, specifically
Include:
To the local image region comprising locator, using based on hough line detection algorithm to the locator in image
It is accurately positioned, is fitted locator straight line, specifically includes:
Firstly, Canny extracts image border:
(a) image is come smooth using the Gaussian filter with specified value deviation;
(b) partial gradient and edge direction are calculated at every bit;
(c) for the image ridge occurred in step (b), consider that Canny algorithm tracks the top of all ridges, and by it is all not
Pixel at the top of ridge is set as zero, provides a line in the output;Ridge pixel uses two threshold value T1And T2Threshold process is done,
Wherein, T1<T2, pixel value is greater than T2Ridge pixel become strong edge pixel, T1And T2Between ridge pixel become weak edge pixel;
(d) by the way that the weak pixel of 8 connections is integrated into strong pixel, edge link is executed;
Then, Hough is fitted locator straight line:
(a1) in the image Jing Guo Canny operator transformation, the transformation of hough straight line is carried out, obtains all possible straight lines
Section;
(b1) length is chosen in whole straightways and be greater than threshold value, and angle meets the straight line of preset condition as locator
Fitting is realized.
Further, the step 6 specifically includes:
Firstly, being demarcated to camera:
(a2) draw camera calibration disk, by camera roof position fixed camera;
(b2) from different distance and angle to calibration disk imaging, and meet calibration disk and camera plane angle less than 45 °;
(c2) image that will acquire imports computer, obtains the intrinsic parameter (f of camerax,fy,u0,v0) and outer parameter (R, t);
Then, locator coordinate is calculated using projection matrix:
(a3) transformational relation of image coordinate system and world coordinate system is established:
(b3) intrinsic parameter of camera calibration acquisition and outer parameter are brought into transition matrix, obtains locator world coordinates and turns
Change relationship;
(c3) locator is sampled, coordinate of the locator in real world coordinates system is calculated by image coordinate,
In, locator lower extreme point coordinate is (Xw1,Yw1) and upper extreme point coordinate be (Xw2,Yw2);
Then, the gradient is obtained according to locator coordinate, the locator gradient is calculated are as follows:
Further, the system also includes abnormal alarm modules and control system module;The abnormal alarm module is used
In output locator gradient exception information;The control system module is for integrating video acquisition unit and main computer unit
Control.
Wherein, the contact net positioner slope detection method in the application includes:
Firstly, obtaining train contact network operation image information using image collecting device;
Then, train contact network operation image information is handled, obtains localizer image information;
Then, it is based on localizer image information, analytical calculation is carried out to locator, obtains world coordinate system lower locating device slope
Degree.
Wherein, carrying out analytical calculation to locator is passed through using the operation host of vehicle-mounted contact net operation detection device
Locator line detection algorithms based on deep learning carry out detection positioning to the locator in image, then utilize three-dimensional monocular
Vision carries out gradient calculating to locator.
Wherein, the image collecting device in the application can be for camera, video camera, camera etc., such as: being by being installed on
The high-resolution high speed video camera of roof acquires contact net running environment video (as shown in Figure 3).
Also, a kind of contact net positioner slope detection system is provided based on the method in the application is corresponding, this is
System includes:
Vehicle-mounted contact net video acquisition device comprising in the high-resolution high-speed camera dress of high speed motor car roof installation
It sets, the Running State video of acquisition contact in real time, and video image is real-time transmitted to interior host and carries out data analysis;
Main computer unit, including power management, net cast, operation and control interface, abnormal alarm, Communication Control, data analysis with
And control system module.Wherein in data analysis module, image is analyzed using deep learning algorithm, obtains locator
Accurate location and be fitted locator straight line, after successfully obtaining locator fitting a straight line, handled using three-dimensional monocular vision
To the gradient of the locator under world coordinate system.Abnormal alarm module is analyzed for the aforementioned obtained locator gradient, when
The locator gradient be more than defined threshold when export warning message, and provide the locator gradient it is exceeded when high speed motor car position letter
Breath.
Wherein, the installation site of photographic device, direction, visual field size, picture-taken frequency meets system can be for each
Catenary mast collects sufficient amount, is appropriate for the image of locator detection, and imaging model meets using three-dimensional single
Visually feel and carries out locator gradient design conditions.
Wherein, the method in the application is combined using camera and computer, is detected, is avoided to the locator gradient automatically
Traditional man-hour manually hand-held equipment is detected, and detection efficiency is higher, and when detecting between upper there is no limit in train without stopping
It is detected when fortune, detection efficiency is higher, and combines and calculated using deep learning model, image procossing, can obtain standard
True testing result.
One or more technical solution provided by the present application, has at least the following technical effects or advantages:
Locator slope detection system in the application carries out locators different under complex environment based on deep learning
Accurate detection;And the thinking based on straight line fitting can accurately be fitted locator straight line in regional area;Further
The solution based on projection equation, exact value of the locator under world coordinate system can be provided;So efficiently solving
The existing locator slope detection method technical problem poor there are detection efficiency and accuracy rate, and then realize efficiently and accurately
The technical effect that the locator gradient is detected.
Detailed description of the invention
Attached drawing described herein is used to provide to further understand the embodiment of the present invention, constitutes one of the application
Point, do not constitute the restriction to the embodiment of the present invention;
Fig. 1 is the flow diagram of locator slope detection method in the application;
Fig. 2 is the composition schematic diagram of locator slope detection system in the application;
Fig. 3 is the video acquisition device scheme of installation of locator slope detection system in the application.
Specific embodiment
The present invention provides a kind of contact net positioner slope detection systems, solve existing locator slope detection side
The method technical problem poor there are detection efficiency and accuracy rate, realizes the skill of efficiently and accurately detected to the locator gradient
Art effect.
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the case where not conflicting mutually, the application's
Feature in embodiment and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
Implemented with being different from the other modes being described herein in range using other, therefore, protection scope of the present invention is not by under
The limitation of specific embodiment disclosed in face.
Fig. 1-Fig. 3 is please referred to, this application provides a kind of locator slope detection methods, which comprises
Firstly, obtaining train contact network operation image information using image collecting device;
Then, train contact network operation image information is handled, obtains localizer image information;
Then, it is based on localizer image information, analytical calculation is carried out to locator, obtains world coordinate system lower locating device slope
Degree.
Wherein, carrying out analytical calculation to locator is passed through using the operation host of vehicle-mounted contact net operation detection device
Locator line detection algorithms based on deep learning carry out detection positioning to the locator in image, then utilize three-dimensional monocular
Vision carries out gradient calculating to locator.
Wherein, the image collecting device in the application can be for camera, video camera, camera etc., such as: being by being installed on
The high-resolution high speed video camera of roof acquires contact net running environment video.
Also, a kind of locator slope detection system is provided based on the method in the application is corresponding, referring to FIG. 2,
The system comprises video acquisition units and main computer unit, in which:
Video acquisition unit is installed on roof of train, and image acquisition units for acquiring contact Running State view in real time
Frequently, and video image is real-time transmitted to interior main computer unit and carries out data analysis;
Main computer unit is installed on interior, receives the vision signal of video acquisition unit and carries out compression storage, and root
The locator gradient is calculated according to bow net image data is obtained.
Further, video acquisition unit includes: mounting seat, camera, supplementary lighting sources;Mounting seat and roof interface are solid
Fixed connection;Supplementary lighting sources and camera are both secured in the mounting seat, and the injection optical registration contact of camera and supplementary lighting sources
Net system frame and locator.
Wherein, the system also includes power management modules and communication control module;The power management module is for being
Video acquisition unit and main computer unit power supply;The video image that the communication control module is used to acquire video acquisition unit passes
It is defeated to arrive main computer unit.
Wherein, the system also includes data memory modules;The data memory module is for saving video acquisition unit
The video image of acquisition.
Wherein, the first CPU module is equipped in the main computer unit;First CPU module is used to calculate by deep learning
Method carries out rough detection to locator, specifically includes:
Step 1: acquisition localizer image information first is then based on localizer image information architecture deep learning model;
Step 2: being based on bow net image, construct locator training sample;
Step 3: based on the training sample training deep learning model in step 2;
Step 4: based on the deep learning model after training, the image containing locator to be detected being handled, is obtained
The location information of locator to be detected;
Step 5: based on the location information in step 4, to the local image region comprising locator, to the positioning in image
Device is accurately positioned, and locator straight line is fitted;
Step 6: projective transformation being carried out to localizer image, coordinate of the locator under world coordinate system is calculated, obtains generation
Boundary's coordinate system lower locating device gradient.
Wherein, the step 1 specifically includes:
Step 1.1: building include several hidden layers depth convolutional network, in which: including convolutional layer, down-sampled layer and
Full articulamentum;The local feature of image is extracted by convolutional calculation;
Step 1.2: down-sampled processing is done by local feature of the pond to image;
Step 1.3: further successively feature extraction processing is done by full articulamentum;
Step 1.4: class probability and the exact position of locator are predicted by classifier layer
Step 1.5: defining classification device loss function, including classification loss and position loss.
Wherein, the step 4 specifically includes:
Step 4.1: the depth convolutional network after inputting Optimal Parameters for image to be detected;
Step 4.2: obtaining the network output valve of test image
Step 4.3: locator classification and location information can be obtained according to network output valve;
Wherein, the second CPU module is additionally provided in the main computer unit;Second CPU module is used for locator region
Examining survey is carried out, and the gradient of the locator under world coordinate system is calculated by monocular vision solid geometry relationship, is specifically included:
To the local image region comprising locator, using based on hough line detection algorithm to the locator in image
It is accurately positioned, is fitted locator straight line, specifically includes:
Firstly, Canny extracts image border:
(a) image is come smooth using the Gaussian filter with specified value deviation;
(b) partial gradient and edge direction are calculated at every bit;
(c) for the image ridge occurred in step (b), consider that Canny algorithm tracks the top of all ridges, and by it is all not
Pixel at the top of ridge is set as zero, provides a line in the output;Ridge pixel uses two threshold value T1And T2Threshold process is done,
Wherein, T1<T2, pixel value is greater than T2Ridge pixel become strong edge pixel, T1And T2Between ridge pixel become weak edge pixel;
(d) by the way that the weak pixel of 8 connections is integrated into strong pixel, edge link is executed;
Then, Hough is fitted locator straight line:
(a1) in the image Jing Guo Canny operator transformation, the transformation of hough straight line is carried out, obtains all possible straight lines
Section;
(b1) length is chosen in whole straightways and be greater than threshold value, and angle meets the straight line of preset condition as locator
Fitting is realized.
Wherein, the step 6 specifically includes:
Firstly, being demarcated to camera:
(a2) draw camera calibration disk, by camera roof position fixed camera;
(b2) from different distance and angle to calibration disk imaging, and meet calibration disk and camera plane angle less than 45 °;
(c2) image that will acquire imports computer, obtains the intrinsic parameter (f of camerax,fy,u0,v0) and outer parameter (R, t);
Then, locator coordinate is calculated using projection matrix:
(a3) transformational relation of image coordinate system and world coordinate system is established:
(b3) intrinsic parameter of camera calibration acquisition and outer parameter are brought into transition matrix, obtains the locator world
Coordinate transformation relation;
(c3) locator is sampled, seat of the locator in real world coordinates system is calculated by image coordinate
Mark, wherein locator lower extreme point coordinate is (Xw1,Yw1) and upper extreme point coordinate be (Xw2,Yw2);
Then, the gradient is obtained according to locator coordinate, the locator gradient is calculated are as follows:
Wherein, the system also includes abnormal alarm modules and control system module;The abnormal alarm module is for defeated
Locator gradient exception information out;The control system module is used to carry out video acquisition unit and main computer unit comprehensive control
System.
Wherein, image is analyzed using deep learning algorithm, obtain the accurate location of locator and is fitted locator
Straight line handles to obtain locator under world coordinate system using three-dimensional monocular vision after successfully obtaining locator fitting a straight line
The gradient.Abnormal alarm module is analyzed for the aforementioned obtained locator gradient, when the locator gradient is more than defined threshold
When export warning message, and provide the locator gradient it is exceeded when high speed motor car location information.Fig. 3 is locator slope detection system
Acquisition device, is mounted on the top of train by the scheme of installation of vehicle-mounted contact net video acquisition device in system.
Referring to FIG. 1, locator detecting step handles video image and detection image using deep learning algorithm in real time
The locator of middle appearance;
1 building deep learning model:
As shown in Figure 1, building includes the depth convolutional network of several hidden layers, including convolutional layer, down-sampled layer,
Full articulamentum and output layer.Wherein convolutional layer, down-sampled layer and full articulamentum are for successively extracting characteristics of image;Output layer includes
Classifier is for obtaining framing device classification and region.
Step 1.1: extracting the local feature of image by convolutional calculation, and nonlinear transformation is carried out to local feature.
Step 1.2: being done by local feature of the pond to image down-sampled.
Step 1.3: further layer-by-layer feature extraction is done by full articulamentum.
Step 1.4: class probability and the exact position of locator are predicted by output layer.Wherein class probability includes positioning
The positive positioning and antidirection finding of device;Exact position includes the center point coordinate and length and width of locator region.
Step 1.5: defining output layer classifier loss function, including classification loss and position loss.Wherein loss function
The least mean-square error of the true classification of locator and position and depth convolutional network output valve.
The construction method of above-mentioned convolutional layer, down-sampled layer, full articulamentum and classifier layer was delivered at Lecun et al. 1998
Article " Gradient-based learning applied to document recognition " on IEEE has specifically
It introduces
2 selection samples.
Equipment is acquired by 3C and obtains a large amount of bow net image, and button goes locator sample, building locator instruction from image
Practice sample set.It selects to include locator region in image, selective positioning device classification: positive positioning and antidirection finding.Record locator exists
Position in image, the center comprising image block is in the coordinate of full figure and the length of image block and width.
3 training deep learning models;
Output layer classifier loss function is minimized using gradient descent algorithm, to roll up in an iterative manner to depth
Product network parameter is adjusted.Article " the Learning that method for solving was published on Nature at Rumelhart et al. 1986
There is specific introduction in representations by back-propagating errors ".
4 detection locators.
Step 4.1: the depth convolutional network after inputting Optimal Parameters for image to be detected,;
Step 4.2: obtaining the network output valve of the test image.
Step 4.3: the image block destination probability in output layer being ranked up, the image block that selection meets threshold requirement is made
For target area.
Step 4.4: obtaining target area according to output layer and calculate class probability, differentiate positive positioning and antidirection finding.
Step 4.5: obtaining the coordinate and zone length and width at the center of target area according to output layer.
5 straight line fitting steps, to the local image region comprising locator, using based on hough line detection algorithm pair
Locator in image is accurately positioned, and locator straight line is fitted;
Canny extracts image border
A image is come using the Gaussian filter with specified value deviation smoothly, so as to reduce noise;
B calculates partial gradient at every bitWith edge direction α (x, y)=arctan
(Gy/Gx).It is the point that its intensity is local maxima that marginal point, which is defined as gradient direction,.
C considers that algorithm tracks the top of all ridges for the image ridge occurred in step b, and by all not on the top of ridge
The pixel in portion is set as zero, to provide a filament in the output.Ridge pixel uses two threshold value T1And T2Threshold process is done,
Middle T1<T2.Pixel value is greater than T2Ridge pixel become strong edge pixel, T1And T2Between ridge pixel become weak edge pixel.
D executes edge link by the way that the weak pixel of 8 connections is integrated into strong pixel.
Hough is fitted locator straight line
It in the image Jing Guo Canny operator transformation, carries out the transformation of hough straight line [], obtains all possible straightways;
Length is chosen in whole straightways and is greater than threshold value, and angle is straight between [0 °, 45 °] and [135 °, 180 °]
Line is fitted as locator and realizes;(for just positioning and two kinds of Installation Modes of antidirection finding, the installation of locator and the folder of horizontal plane
Angle is between 5 °~20 °, since the setting angle of camera and the distance of Distance positioning device are different, the angle of locator in the picture
Usually be different, in order to avoid erroneous detection, it is desirable that the angle of locator fitting a straight line section in the picture be [0 °, 45 °] and
[135 °, 180 °])
The gradient calculates step, carries out projective transformation to localizer image, calculates coordinate of the locator under world coordinate system,
Obtain the world coordinate system lower locating device gradient.
Camera is demarcated:
Draw camera calibration disk, by camera roof position fixed camera;
From different distance and angle to calibration disk imaging, and meet calibration disk and camera plane angle less than 45 °;
The image that will acquire imports program, obtains the internal reference of camera using " CalibrateCamera2 " function in opencv
Several and outer parameter.
Locator coordinate is calculated using projection matrix;
As shown by the equation, the transformational relation of image coordinate system and world coordinate system is established:
Intrinsic parameter (the f that camera calibration is obtainedx,fy,u0,v0) and outer parameter (R, t) bring into transition matrix, determined
Position device world coordinates transformational relation;
Locator is sampled, coordinate of the locator in real world coordinates system is calculated by image coordinate, wherein fixed
Position device lower extreme point coordinate is (Xw1,Yw1) and upper extreme point coordinate be (Xw2,Yw2)。
The gradient is obtained according to locator coordinate
According to formula, the locator gradient is calculated are as follows:
Technical solution in above-mentioned the embodiment of the present application, at least have the following technical effects or advantages:
Locator slope detection system in the application carries out locators different under complex environment based on deep learning
Accurate detection;And the thinking based on straight line fitting can accurately be fitted locator straight line in regional area;Further
The solution based on projection equation, exact value of the locator under world coordinate system can be provided;So efficiently solving
The existing locator slope detection method technical problem poor there are detection efficiency and accuracy rate, and then realize efficiently and accurately
The technical effect that the locator gradient is detected.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (9)
1. a kind of contact net positioner slope detection system, which is characterized in that the system comprises video acquisition units and host
Unit, in which:
Video acquisition unit is installed on roof of train, and image acquisition units contact Running State video for acquiring in real time, and
Video image is real-time transmitted to interior main computer unit and carries out data analysis;
Main computer unit is installed on interior, receives the vision signal of video acquisition unit and carries out compression storage, and according to obtaining
It obtains bow net image data and calculates the locator gradient;
The second CPU module is additionally provided in the main computer unit;Second CPU module is used to carry out examining to locator region
It surveys, and the gradient of the locator under world coordinate system is calculated by monocular vision solid geometry relationship, specifically include:
To the local image region comprising locator, the locator in image is carried out using based on hough line detection algorithm
It is accurately positioned, is fitted locator straight line, specifically includes:
Firstly, Canny extracts image border:
(a) image is come smooth using the Gaussian filter with specified value deviation;
(b) partial gradient and edge direction are calculated at every bit;
(c) for the image ridge occurred in step (b), consider that Canny algorithm tracks the top of all ridges, and by all not in ridge
The pixel at top be set as zero, provide a line in the output;Ridge pixel uses two threshold value T1And T2Do threshold process, wherein
T1< T2, pixel value is greater than T2Ridge pixel become strong edge pixel, T1And T2Between ridge pixel become weak edge pixel;
(d) by the way that the weak pixel of 8 connections is integrated into strong pixel, edge link is executed;
Then, Hough is fitted locator straight line:
(a1) in the image Jing Guo Canny operator transformation, the transformation of hough straight line is carried out, obtains all possible straightways;
(b1) length is chosen in whole straightways and is greater than threshold value, and angle meets the straight line of preset condition as locator fitting
It realizes.
2. contact net positioner slope detection system according to claim 1, which is characterized in that video acquisition unit packet
It includes: mounting seat, camera, supplementary lighting sources;Mounting seat is fixedly connected with roof interface;Supplementary lighting sources and camera are both secured to institute
It states in mounting seat, and the injection optical registration contact net system bracket and locator of camera and supplementary lighting sources.
3. contact net positioner slope detection system according to claim 1, which is characterized in that the system also includes electricity
Source control module and communication control module;The power management module is used to power for video acquisition unit and main computer unit;Institute
Transmission of video images of the communication control module for acquiring video acquisition unit is stated to main computer unit.
4. contact net positioner slope detection system according to claim 1, which is characterized in that the system also includes numbers
According to memory module;The data memory module is used to save the video image of video acquisition unit acquisition.
5. contact net positioner slope detection system according to claim 1, which is characterized in that set in the main computer unit
There is the first CPU module;First CPU module is used to carry out rough detection to locator by deep learning algorithm, specifically includes:
Step 1: acquisition localizer image information first is then based on localizer image information architecture deep learning model;
Step 2: being based on bow net image, construct locator training sample;
Step 3: based on the training sample training deep learning model in step 2;
Step 4: based on the deep learning model after training, the image containing locator to be detected being handled, is obtained to be checked
Survey the location information of locator;
Step 5: based on the location information in step 4, to the local image region comprising locator, to the locator in image into
Row is accurately positioned, and is fitted locator straight line;
Step 6: projective transformation being carried out to localizer image, calculates coordinate of the locator under world coordinate system, obtains world's seat
The mark system lower locating device gradient.
6. contact net positioner slope detection system according to claim 5, which is characterized in that the step 1 is specifically wrapped
It includes:
Step 1.1: building includes the depth convolutional network of several hidden layers, in which: including convolutional layer, down-sampled layer and Quan Lian
Connect layer;The local feature of image is extracted by convolutional calculation;
Step 1.2: down-sampled processing is done by local feature of the pond to image;
Step 1.3: further successively feature extraction processing is done by full articulamentum;
Step 1.4: class probability and the exact position of locator are predicted by classifier layer
Step 1.5: defining classification device loss function, including classification loss probability and position loss probability.
7. contact net positioner slope detection system according to claim 5, which is characterized in that the step 4 is specifically wrapped
It includes:
Step 4.1: the depth convolutional network after inputting Optimal Parameters for image to be detected;
Step 4.2: obtaining the network output valve of test image;
Step 4.3: locator classification and location information can be obtained according to network output valve.
8. locator slope detection system according to claim 5, which is characterized in that the step 6 specifically includes:
Firstly, being demarcated to camera:
(a2) draw camera calibration disk, by camera roof position fixed camera;
(b2) from different distance and angle to calibration disk imaging, and meet calibration disk and camera plane angle less than 45 °;
(c2) image that will acquire imports computer, obtains the intrinsic parameter (f of camerax,fy,u0,v0) and outer parameter (R, t);
Then, locator coordinate is calculated using projection matrix:
(a3) transformational relation of image coordinate system and world coordinate system is established:
(b3) intrinsic parameter of camera calibration acquisition and outer parameter are brought into transition matrix, obtains the conversion of locator world coordinates and closes
System;
(c3) locator is sampled, coordinate of the locator in real world coordinates system is calculated by image coordinate, wherein fixed
Position device lower extreme point coordinate is (Xw1,Yw1) and upper extreme point coordinate be (Xw2,Yw2);
Then, the gradient is obtained according to locator coordinate, the locator gradient is calculated are as follows:
9. contact net positioner slope detection system according to claim 1, which is characterized in that the system also includes different
Normal alarm module and control system module;The abnormal alarm module is for exporting locator gradient exception information;The control
System module is used to carry out comprehensively control to video acquisition unit and main computer unit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710029167.4A CN106679630B (en) | 2017-01-16 | 2017-01-16 | A kind of contact net positioner slope detection system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710029167.4A CN106679630B (en) | 2017-01-16 | 2017-01-16 | A kind of contact net positioner slope detection system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106679630A CN106679630A (en) | 2017-05-17 |
CN106679630B true CN106679630B (en) | 2019-02-26 |
Family
ID=58860653
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710029167.4A Active CN106679630B (en) | 2017-01-16 | 2017-01-16 | A kind of contact net positioner slope detection system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106679630B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109115180A (en) * | 2018-07-02 | 2019-01-01 | 苏州大成电子科技有限公司 | A kind of contact net operation detection system |
CN109631800B (en) * | 2018-11-27 | 2020-11-06 | 中国铁道科学研究院集团有限公司 | Method and device for detecting dynamic lifting amount of contact line |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101361086A (en) * | 2005-12-22 | 2009-02-04 | 株式会社Gmg | Image transmission system of a slope |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102175219B (en) * | 2011-01-31 | 2013-01-23 | 中国铁道科学研究院基础设施检测研究所 | High-speed contact network locator gradient detection method and apparatus thereof based on video analysis |
CN203116843U (en) * | 2013-01-05 | 2013-08-07 | 西南交通大学 | Contact system vibration measuring device |
CN205292329U (en) * | 2015-12-18 | 2016-06-08 | 铁道第三勘察设计院集团有限公司 | Contact net locator |
-
2017
- 2017-01-16 CN CN201710029167.4A patent/CN106679630B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101361086A (en) * | 2005-12-22 | 2009-02-04 | 株式会社Gmg | Image transmission system of a slope |
Also Published As
Publication number | Publication date |
---|---|
CN106679630A (en) | 2017-05-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105373135B (en) | A kind of method and system of aircraft docking guidance and plane type recognition based on machine vision | |
CN108956640B (en) | Vehicle-mounted detection device and detection method suitable for power distribution line inspection | |
CN105426905B (en) | Robot barrier object recognition methods based on histogram of gradients and support vector machines | |
CN103824420B (en) | Fatigue driving identification system based on heart rate variability non-contact measurement | |
CN110487562A (en) | One kind being used for unpiloted road-holding ability detection system and method | |
CN109615870A (en) | A kind of traffic detection system based on millimetre-wave radar and video | |
CN202793389U (en) | On-line routing inspection device for high-speed rail overhead line system | |
CN108196260A (en) | The test method and device of automatic driving vehicle multi-sensor fusion system | |
CN104573646A (en) | Detection method and system, based on laser radar and binocular camera, for pedestrian in front of vehicle | |
CN109389829A (en) | A kind of illegal parking recognition methods and system | |
CN102980896B (en) | High ferro overhead contact line device auricle fracture detection method | |
CN103837087B (en) | Pantograph automatic testing method based on active shape model | |
CN108960060A (en) | A kind of automatic driving vehicle pavement texture identifying system and method | |
CN101915672A (en) | Testing device and testing method of lane departure warning system | |
CN112348034A (en) | Crane defect detection system based on unmanned aerial vehicle image recognition and working method | |
CN107187464A (en) | Track plates detection car, system and method | |
CN105205785A (en) | Large vehicle operation management system capable of achieving positioning and operation method thereof | |
CN107796373B (en) | Distance measurement method based on monocular vision of front vehicle driven by lane plane geometric model | |
CN108919367A (en) | Transmission line of alternation current inversion method based on current field | |
CN114037703B (en) | Subway valve state detection method based on two-dimensional positioning and three-dimensional attitude calculation | |
CN106679630B (en) | A kind of contact net positioner slope detection system | |
CN105913034A (en) | Vehicle identification method and apparatus, and vehicle | |
CN103422417A (en) | Dynamic identification system and method for detecting road surface damages | |
CN111855667A (en) | Novel intelligent train inspection system and detection method suitable for metro vehicle | |
CN113449688B (en) | Power transmission tree obstacle recognition system based on image and laser point cloud data fusion |
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 |