CN105719305B - Component falls off defect identification method and system in contact net - Google Patents
Component falls off defect identification method and system in contact net Download PDFInfo
- Publication number
- CN105719305B CN105719305B CN201610049171.2A CN201610049171A CN105719305B CN 105719305 B CN105719305 B CN 105719305B CN 201610049171 A CN201610049171 A CN 201610049171A CN 105719305 B CN105719305 B CN 105719305B
- Authority
- CN
- China
- Prior art keywords
- target
- image
- connector
- component
- falls
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
It falls off defect identification method and system the invention discloses component in contact net, using component template image on the different scale of image to be detected slip scan target element, the band of position image of target element is matched in image to be detected, component template image includes the template image of various components in contact net；Target connector region is oriented in the area image of position according to edge analysis algorithm；According to the structural relation of target element and target connector, target connector is partitioned into the area image of position according to corresponding ratio；Obtain the grey level histogram and Gradient Features of target connector, judge that target connector whether there is the doubtful defect that falls off according to the grey level histogram of target connector, determine whether the doubtful defect that falls off is true further according to the Gradient Features of target connector, if if true, judging that the target connector falls off defect there are component, if if false, judging that the target connector is normal.The present invention can accurately identify that component falls off defect.
Description
Technical field
The present invention relates to contact net field, fall off defect identification method and system more particularly to component in contact net.
Background technology
Contact net is to be set up along rail overhead "the" shape in electric railway, take the high pressure of stream defeated for pantograph
Electric wire.Contact net is the main truss of railway electrification project, is powering to track vehicle for downline overhead erection
The transmission line of electricity of special shape.Generally it is made of contact suspension, support device, positioning device, several parts of pillar and basis.
A large amount of such as screw, bolt connection component has been used to fix relevant holder, locator, positioning in contact net
Bar etc., and once there is the case where loosening or falling off in such connection component, and has prodigious potential danger to contact net safe operation
Danger.
And existing detection mode is then to check corresponding video image by human eye, due to the continuous increasing with amount of video
Add, human eye checks that the connection components such as screw, bolt with the presence or absence of the defect that falls off, not only expend a large amount of manpower, it is also possible to
Because the fatigue test of people leads to missing inspection.Therefore component fall off defect automatic identification have very strong practical value.
Invention content
It is an object of the invention to overcome the deficiencies in the prior art, for manually to confirm that component falls off scarce for analysis video frame by frame
Sunken inconvenience, it is proposed that the component component in defect identification method and contact net that falls off falls off defect recognition system in a kind of contact net
System.
The purpose of the present invention is achieved through the following technical solutions：
(1) component falls off defect identification method in contact net, and the various components in contact net are made of connector connection,
The method the described method comprises the following steps for detecting the connector with the presence or absence of the defect that falls off：
S1 obtains image to be detected；
S2, positioning component：Using component template image on the different scale of image to be detected slip scan target element,
The band of position image of target element is matched in image to be detected, the component template image includes each group in contact net
The template image of part；
S3, positioning link：Target connector area is oriented in the area image of position according to edge analysis algorithm
Domain；
S4 divides connector：According to the structural relation of target element and target connector, according to corresponding ratio in position
Target connector is partitioned into area image；
S5, signature analysis：The grey level histogram and Gradient Features for obtaining target connector, according to the gray scale of target connector
Histogram judges that target connector whether there is the doubtful defect that falls off, and determines that this is doubtful further according to the Gradient Features of target connector
Whether the defect that falls off is true, if if true, judging that the target connector falls off defect there are component, if if false, judging the target
Connector is normal.
Further, the positioning component step S2 includes following sub-step：
S201, formation component subtemplate image：In component template image object, component template image object is intercepted
Subgraph as component subtemplate image object_sub；
S202, multiple dimensioned positioning target element：Carried out on the n scale of component template image object up-sampling and under
Sampling, obtains the component template image object of 2n+1 different scalek, at n of component subtemplate image object_sub
Up-sampling and down-sampling are carried out on scale, obtain the component subtemplate image object_sub of 2n+1 different scalek, wherein
k∈[1,2n+1]；
S203 extracts the gradient magnitude of image：Computation module template image objectk, component subtemplate image object_
subkAmplitude Amp is obtained with the gradient magnitude of image to be detectedk, amplitude Ampk_subWith amplitude Ampdetect；
S204 positions the band of position of target element：By amplitude AmpkWith amplitude Ampk_subRespectively in amplitude AmpdetectIn
Template matches are carried out, the optimal positioning image Loc of target element under current scale is obtainedkWith positioning image Lock_sub,
S205 calculates the similitude of positioning image and template image：
If positioning image Lock_subIt is positioning image LockSubgraph, i.e.,Then extraction is current respectively
Scale lower component template image objectkHOG features and positioning image LockHOG features, and calculate two HOG features it
Between Euclidean distance distk；
If positioning image Lock_subIt is not positioning image LockSubgraph, i.e.,Then define current ruler
Spend lower component template image objectkWith positioning image LockEuclidean distance distkFor a maximum MaxValue, i.e.,
distk=MaxVa；
S206 determines the band of position image of target element：From 2n+1 Euclidean distance distkIt is middle to select the European of minimum
Distance distk, and it is compared with threshold value MaxTh, if the Euclidean distance dist of the minimumk＞ MaxTh then judge that this is fixed
Bit image LockIn vain, if the Euclidean distance dist of the minimumk＜ MaxTh then judge positioning target LockFor the target element
Band of position image.
Further, the positioning link step S3 includes following sub-step：
S301 obtains the sobel edge graphs of band of position image；
S302 counts the horizontal edge histogram and vertical edge histogram of sobel edge graphs；
S303 is compared with threshold value Th1 according to horizontal edge histogram, orients target connector region for the first time
Coboundary and lower boundary are compared with threshold value Th2 according to vertical edge histogram, orient target connector region for the first time
Left margin and right margin, obtain the target connector region that positions for the first time；
S304 is repeated one or more times step S301-S303, after obtaining multiple bearing according to the positioning result of S303
Target connector region.
Further, the step S303 includes following sub-step：
S3031, according to horizontal edge histogram HistRpIt is compared with threshold value Th1, orients target connector region
Coboundary top：From horizontal edge histogram HistRpInitial value start to be compared with its mean value AvgR successively, Zhi Dao
P rows HistRpWhen being more than threshold value Th1 with the ratio of mean value AvgR, then the coboundary top=p in target connector region is enabled；
S3032, according to horizontal edge histogram HistRpIt is compared with threshold value Th1, orients target connector region
Lower boundary bottom：From horizontal edge histogram HistRpEnd value start to be compared with its maximum value MaxR successively,
Until pth row HistRpMore than mean value AvgR and HistRpIt is more than threshold value Th0 with the ratio of maximum value MaxR, then enables target connect
The lower boundary bottom=p in part region；
S3033, according to vertical edge histogram HistCqIt is compared with threshold value Th2, orients target connector region
Left margin left：From vertical edge histogram HistCqInitial value start to be compared with threshold value Th2 successively, when q arrange
HistCqWhen more than threshold value Th2, then the left margin left=q in target connector region is enabled；
S3034, according to vertical edge histogram HistCqIt is compared with threshold value Th2, orients target connector region
Right margin right：From vertical edge histogram HistCqEnd value start to be compared with threshold value Th2 successively, when q arrange
HistCqWhen more than threshold value Th2, then the right margin right=q in target connector region is enabled；
S3035 is oriented for the first time according to coboundary top, lower boundary bottom, left margin left and right margin right
Target connector region Ifirst。
Further, the positioning link step S3 further includes sub-step S305：The target after multiple bearing is excluded to connect
Wrong positioning result in fitting region.
Further, the structural relation includes relative position relation and area proportionate relationship.
Further, described in signature analysis step S5 judge target connector with the presence or absence of the doubtful defect that falls off include with
Lower sub-step：The grey level histogram of each target connector is calculated, and judges that the grey level histogram of each target connector is maximum
The difference between whether gray level identical or same grey level whether be less than threshold value Th3, if so, judging the target connector
There are the doubtful defects that falls off, and record connector and fall off and indicate flag=1, and otherwise judging the target connector, there is no doubtful de-
Defect is fallen, record connector, which falls off, indicates flag=0.
Further, determine whether the doubtful defect that falls off is true, including following sub-step described in signature analysis step S5：
S501 calculates separately the horizontal gradient and vertical gradient of each target connector, calculates the argument of each pixel
θ；
S502, counts the distribution situation of argument θ in entire target connector image, and place is normalized to argument histogram
Reason obtains argument features vector；
S503, if connector falls off, mark flag=1, calculates all targets that the target connector region is partitioned into
In connector, per the mean square deviation of the argument features vector between target connector two-by-two, if there is one group of mean square deviation to be more than threshold value
Th4, it is determined that whether the doubtful defect that falls off is true, judges that the target connector falls off defect there are component；
S504, the mark flag=0 if connector falls off, the argument principal direction of more all target connectors whether phase
Together；
If identical, it is determined that whether the doubtful defect that falls off is false, judges that the target connector is normal；
If differing, calculate in all target connectors that the target connector region is partitioned into, per target two-by-two
The mean square deviation of argument features vector between connector, if there is one group of mean square deviation more than threshold value Th5 and small containing one group of mean square deviation
In threshold value Th6, it is determined that whether the doubtful defect that falls off is true, judges that the target connector falls off defect there are component.
(2) component falls off defect recognition system in contact net, applies method as described above, the system comprises successively
Image to be detected acquisition module of connection, multiple dimensioned component locating module, connector locating module, connector segmentation module and company
Fitting characteristics analysis module.
Image to be detected acquisition module is used for image to be detected of securing component.
The multiple dimensioned component locating module is used for sliding on the different scale of image to be detected using component template image
Dynamic search target element, matches the band of position image of target element, the component template image packet in image to be detected
Include the template image of various components in contact net.
The connector locating module is used to orient target in the area image of position according to edge analysis algorithm
Connector region.
The connector segmentation module is used for the structural relation according to target element and target connector, according to corresponding ratio
Example is partitioned into target connector in the area image of position.
The connector characteristics analysis module is used to obtain the grey level histogram and Gradient Features of target connector, according to mesh
The grey level histogram judgement target connector for marking connector whether there is the doubtful defect that falls off, further according to the gradient of target connector
Feature determines whether the doubtful defect that falls off is true, if if true, judge that the target connector falls off defect there are component, if
It is false then judge that the target connector is normal.
The beneficial effects of the invention are as follows：
1) the various connectors of detection various components are present invention can be suitably applied to, only need to be directed to different components and different connections
Part, using different parameters, principle is essentially identical, can accurately orient the connection in each component and its various components
Part is effectively improved component and fallen off the accuracy of identification, subtracted by the technologies such as multiple dimensioned positioning, template matches and HOG signature analysis
Few false drop rate, can also exclude the environmental disturbances such as background, grove and tunnel.
2) realize that component falls off the intelligent measurement of defect through the invention, detect image exist fall off defect when, hair
The defect that goes out to fall off is alarmed, and export the presence and fall off the picture of defect, and staff only need to be to going out through intelligent recognition of the present invention
Picture carries out manual confirmation and analysis, greatly shortens the time that staff checks video, improves working efficiency.
Description of the drawings
Fig. 1 is that component falls off the flow diagram of defect identification method in contact net of the present invention；
Fig. 2 is that component falls off the system block diagram of defect recognition system in contact net of the present invention.
Specific implementation mode
Technical scheme of the present invention is described in further detail below in conjunction with the accompanying drawings, but protection scope of the present invention is not limited to
It is as described below.
(1) component falls off defect identification method in contact net
As shown in Figure 1, fall off defect identification method This embodiment describes component in a kind of contact net, it is each in contact net
A component is made of connector connection, and for the method for detecting the connector with the presence or absence of the defect that falls off, the connector can
Including screw, nut, bolt etc..The various connectors of detection various components are present invention can be suitably applied to, only need to be directed to different components
With different connectors, using different parameters, principle is essentially identical.
Method proposed by the invention includes the following steps：
S1 obtains image to be detected.
S2, positioning component：Using component template image on the different scale of image to be detected slip scan target element,
The band of position image of target element is matched in image to be detected, the component template image includes each group in contact net
The template image of part.
S3, positioning link：Target connector area is oriented in the area image of position according to edge analysis algorithm
Domain.
S4 divides connector：According to the structural relation of target element and target connector, according to corresponding ratio in position
Target connector is partitioned into area image.
S5, signature analysis：The grey level histogram and Gradient Features for obtaining target connector, according to the gray scale of target connector
Histogram judges that target connector whether there is the doubtful defect that falls off, and determines that this is doubtful further according to the Gradient Features of target connector
Whether the defect that falls off is true, if if true, judging that the target connector falls off defect there are component, if if false, judging the target
Connector is normal.
Further, the positioning component step S2 includes following sub-step：
S201, formation component subtemplate image：In component template image object, component template image object is intercepted
Subgraph as component subtemplate image object_sub.
S202, multiple dimensioned positioning target element：Carried out on the n scale of component template image object up-sampling and under
Sampling, obtains the component template image object of 2n+1 different scalek, at n of component subtemplate image object_sub
Up-sampling and down-sampling are carried out on scale, obtain the component subtemplate image object_sub of 2n+1 different scalek, wherein
k∈[1,2n+1]。
Wherein, up-sampling multiplying power is fup, down-sampling multiplying power is fdown。
S203 extracts the gradient magnitude of image：Computation module template image objectk, component subtemplate image object_
subkWith the gradient magnitude Amp of image to be detected, amplitude Amp is obtainedk, amplitude Ampk_subWith amplitude Ampdetect。
S204 positions the band of position of target element：By amplitude AmpkWith amplitude Ampk_subRespectively in amplitude AmpdetectIn
Template matches are carried out, the optimal positioning image Loc of target element under current scale is obtainedkWith positioning image Lock_sub。
Wherein, the matching way of the template matches includes difference of two squares matching, standard deviation matching, relevant matches, mark
Quasi- relevant matches, correlation coefficient matching method and canonical correlation coefficient matching etc..
S205 calculates the similitude of positioning image and template image：
If positioning image Lock_subIt is positioning image LockSubgraph, i.e.,Then extraction is current respectively
Scale lower component template image objectkHOG features and positioning image LockHOG features, and calculate two HOG features it
Between Euclidean distance distk。
If positioning image Lock_subIt is not positioning image LockSubgraph, i.e.,Then define current ruler
Spend lower component template image objectkWith positioning image LockEuclidean distance distkFor a maximum MaxValue, i.e.,
distk=MaxVa.Wherein, maximum MaxValue is theoretically infinitely great positive number, and the positive number that approach infinity can be used is retouched
It states, such as MaxValue=10000.
S206 determines the band of position image of target element：From 2n+1 Euclidean distance distkIt is middle to select the European of minimum
Distance distk, and it is compared with threshold value MaxTh, if the Euclidean distance dist of the minimumk＞ MaxTh then judge that this is fixed
Bit image LockIn vain, if the Euclidean distance dist of the minimumk＜ MaxTh then judge positioning target LockFor the target element
Band of position image.
In the present invention, threshold value can be adjusted correspondingly according to different components.
Further, the positioning link step S3 includes following sub-step：
S301 obtains the sobel edge graphs of band of position image.
S302 counts the horizontal edge histogram and vertical edge histogram of sobel edge graphs.
S303 is compared with threshold value Th1 according to horizontal edge histogram, orients target connector region for the first time
Coboundary and lower boundary are compared with threshold value Th2 according to vertical edge histogram, orient target connector region for the first time
Left margin and right margin, obtain the target connector region that positions for the first time.
S304 is repeated one or more times step S301-S303, after obtaining multiple bearing according to the positioning result of S303
Target connector region.
Further, the positioning link step S3 further includes sub-step S305：The target after multiple bearing is excluded to connect
Wrong positioning result in fitting region.
Further, the step S303 includes following sub-step：
S3031, according to horizontal edge histogram HistRpIt is compared with threshold value Th1, orients target connector region
Coboundary top：From horizontal edge histogram HistRpInitial value start to be compared with its mean value AvgR successively, Zhi Dao
P rows HistRpWhen being more than threshold value Th1 with the ratio of mean value AvgR, then the coboundary top=p in target connector region is enabled；
S3032, according to horizontal edge histogram HistRpIt is compared with threshold value Th1, orients target connector region
Lower boundary bottom：From horizontal edge histogram HistRpEnd value start to be compared with its maximum value MaxR successively,
Until pth row HistRpMore than mean value AvgR and HistRpWhen being more than threshold value Th0 with the ratio of maximum value MaxR, then target is enabled to connect
The lower boundary bottom=p in fitting region；
S3033, according to vertical edge histogram HistCqIt is compared with threshold value Th2, orients target connector region
Left margin left：From vertical edge histogram HistCqInitial value start to be compared with threshold value Th2 successively, when q arrange
HistCqWhen more than threshold value Th2, then the left margin left=q in target connector region is enabled；
S3034, according to vertical edge histogram HistCqIt is compared with threshold value Th2, orients target connector region
Right margin right：From vertical edge histogram HistCqEnd value start to be compared with threshold value Th2 successively, when q arrange
HistCqWhen more than threshold value Th2, then the right margin right=q in target connector region is enabled；
S3035 is oriented for the first time according to coboundary top, lower boundary bottom, left margin left and right margin right
Target connector region Ifirst。
Obtaining target connector region IfirstAfterwards, step S304 is executed, in general, need to only can be carried out to target connector
It positions twice, i.e., in step S304, only repeats a S301-S303.By boundary top, bottom, the left of positioning,
Right is truncated to the target connector area image I of second of positioningsceond.In the target connector area for obtaining second of positioning
Area image IsceondAfterwards, it can also carry out the positioning result operation of removal target connector mistake.
Further, the structural relation described in step S4 may include relative position relation and area proportionate relationship.For example,
For screw assembly, due to being identical according to the size of each screw in screw assembly, and the position of screw is in component
In be relatively-stationary.Therefore the segmentation of screw can be split according to this characteristic.The specific method is as follows：
1. according to screw proportion in assembly, the height Height*0.5-offset of single screw can be calculated, wherein
Height is IsceondPicture altitude.Since the scale visual of screw is square, you can know that the width of screw is Height*
0.5-offset, wherein taking offset=4.
2. divisible accordingly to go out corresponding single screw according to the relative position relation of screw in assembly.Such as three
Angular shape distribution screw it is divisible go out Atria apex screw.
Further, described in signature analysis step S5 judge target connector with the presence or absence of the doubtful defect that falls off include with
Lower sub-step：The grey level histogram of each target connector is calculated, and judges that the grey level histogram of each target connector is maximum
The difference between whether gray level identical or same grey level whether be less than threshold value Th3, if satisfied, then judging that the target connects
There are the doubtful defects that falls off for part, and record connector and fall off mark flag=1, fall off mark if not satisfied, then recording connector
Flag=0.
Further, determine whether the doubtful defect that falls off is true, including following sub-step described in signature analysis step S5：
S501 calculates separately the horizontal gradient and vertical gradient of each target connector, calculates the argument of each pixel
θ。
S503, counts the distribution situation of argument θ in entire target connector image, and place is normalized to argument histogram
Reason obtains argument features vector.
S504, if connector falls off, mark flag=1, calculates all targets that the target connector region is partitioned into
In connector, per the mean square deviation of the argument features vector between target connector two-by-two, if there is one group of mean square deviation to be more than threshold value
Th4, it is determined that whether the doubtful defect that falls off is true, judges that the target connector falls off defect there are component.
S505, the mark flag=0 if connector falls off, the argument principal direction of more all target connectors whether phase
Together.
If identical, it is determined that whether the doubtful defect that falls off is false, judges that the target connector is normal.
If differing, calculate in all target connectors that the target connector region is partitioned into, per target two-by-two
The mean square deviation of argument features vector between connector, if there is one group of mean square deviation more than threshold value Th5 and small containing one group of mean square deviation
In threshold value Th6, it is determined that whether the doubtful defect that falls off is true, judges that the target connector falls off defect there are component.
Experiment proves that the present invention can effectively identify that component falls off defect in the full background image of different gray scales, row
Except interference such as background, tunnel, groves, multiple dimensioned positioning objective result is good, and missing inspection is few, flase drop is less, wherein screw is accurately positioned
As a result also ideal.
(2) component falls off defect recognition system in contact net
As shown in Fig. 2, falling off defect recognition system This embodiment describes component in a kind of contact net, institute as above is applied
Method is stated, the system comprises sequentially connected image to be detected acquisition module, multiple dimensioned component locating module, connector positioning
Module, connector segmentation module and connector characteristics analysis module.
Image to be detected acquisition module is used for image to be detected of securing component.
The multiple dimensioned component locating module is used for sliding on the different scale of image to be detected using component template image
Dynamic search target element, matches the band of position image of target element, the component template image packet in image to be detected
Include the template image of various components in contact net.
The connector locating module is used to orient target in the area image of position according to edge analysis algorithm
Connector region.
The connector segmentation module is used for the structural relation according to target element and target connector, according to corresponding ratio
Example is partitioned into target connector in the area image of position.
The connector characteristics analysis module is used to obtain the grey level histogram and Gradient Features of target connector, according to mesh
The grey level histogram judgement target connector for marking connector whether there is the doubtful defect that falls off, further according to the gradient of target connector
Feature determines whether the doubtful defect that falls off is true, if if true, judge that the target connector falls off defect there are component, if
It is false then judge that the target connector is normal.
Component in contact net according to the present invention is described in an illustrative manner above with reference to attached drawing to fall off defect recognition side
Method and system.It will be understood by those skilled in the art, however, that in the contact net proposed for aforementioned present invention component fall off it is scarce
Recognition methods and system are fallen into, various improvement can also be made on the basis of not departing from the content of present invention, or to which part
Technical characteristic carries out equivalent replacement, all within the spirits and principles of the present invention, any modification, equivalent replacement, improvement
Deng should all be included in the protection scope of the present invention.Therefore, protection scope of the present invention should be by appended claims
Content determine.
Claims (8)
1. component falls off defect identification method in contact net, the various components in contact net are made of connector connection, the side
Method is for detecting the connector with the presence or absence of the defect that falls off, which is characterized in that the described method comprises the following steps：
S1 obtains image to be detected；
S2, positioning component：Using component template image on the different scale of image to be detected slip scan target element, waiting for
The band of position image of target element is matched in detection image, the component template image includes various components in contact net
Template image；
S3, positioning link：Target connector region is oriented in the area image of position according to edge analysis algorithm；
S4 divides connector：According to the structural relation of target element and target connector, according to corresponding ratio in the band of position
Target connector is partitioned into image；
S5, signature analysis：The grey level histogram and Gradient Features for obtaining target connector, according to the intensity histogram of target connector
Figure judgement target connector whether there is the doubtful defect that falls off, and determine that this doubtful falls off further according to the Gradient Features of target connector
Whether defect is true, if if true, judging that the target connector falls off defect there are component, if if false, judging that the target connects
Part is normal；
The positioning component step S2 includes following sub-step：
S201, formation component subtemplate image：In component template image object, the son of interception component template image object
Image is as component subtemplate image object_sub；
S202, multiple dimensioned positioning target element：Carried out on the n scale of component template image object up-sampling and under adopt
Sample obtains the component template image object of 2n+1 different scalek, in the n ruler of component subtemplate image object_sub
Up-sampling and down-sampling are carried out on degree, obtain the component subtemplate image object_sub of 2n+1 different scalek, wherein k ∈
[1,2n+1]；
S203 extracts the gradient magnitude of image：Computation module template image objectk, component subtemplate image object_subk
Amplitude Amp is obtained with the gradient magnitude of image to be detectedk, amplitude Ampk_subWith amplitude Ampdetect；
S204 positions the band of position of target element：By amplitude AmpkWith amplitude Ampk_subRespectively in amplitude AmpdetectMiddle progress
Template matches obtain the optimal positioning image Loc of target element under current scalekWith positioning subgraph Lock_sub,
S205 calculates the similitude of positioning image and template image：
If positioning subgraph Lock_subIt is positioning image LockSubgraph, i.e.,Current ruler is then extracted respectively
Spend lower component template image objectkHOG features and positioning image LockHOG features, and calculate two HOG features between
Euclidean distance distk；
If positioning subgraph Lock_subIt is not positioning image LockSubgraph, i.e.,Then define current scale
Lower component template image objectkWith positioning image LockEuclidean distance distkFor a maximum MaxValue, i.e. distk
=MaxValue；
S206 determines the band of position image of target element：From 2n+1 Euclidean distance distkThe minimum Euclidean distance of middle selection
distk, and it is compared with threshold value MaxTh, if the Euclidean distance dist of the minimumk＞ MaxTh then judge the positioning figure
As LockIn vain, if the Euclidean distance dist of the minimumk＜ MaxTh then judge positioning target LockFor the position of the target element
Set area image.
2. component falls off defect identification method in contact net according to claim 1, which is characterized in that described to be located by connecting
Part step S3 includes following sub-step：
S301 obtains the sobel edge graphs of band of position image；
S302 counts the horizontal edge histogram and vertical edge histogram of sobel edge graphs；
S303 is compared with threshold value Th1 according to horizontal edge histogram, orients the top in target connector region for the first time
Boundary and lower boundary are compared with threshold value Th2 according to vertical edge histogram, orient the left side in target connector region for the first time
Boundary and right margin obtain the target connector region positioned for the first time；
S304 is repeated one or more times step S301-S303 according to the positioning result of S303, obtains the target after multiple bearing
Connector region.
3. component falls off defect identification method in contact net according to claim 2, which is characterized in that the step S303
Including following sub-step：
S3031, according to horizontal edge histogram HistRpIt is compared with threshold value Th1, orients the top in target connector region
Boundary top：From horizontal edge histogram HistRpInitial value start to be compared with its mean value AvgR successively, until pth row
HistRpWhen being more than threshold value Th1 with the ratio of mean value AvgR, then the coboundary top=p in target connector region is enabled；
S3032, according to horizontal edge histogram HistRpIt is compared with threshold value Th1, orients the following of target connector region
Boundary bottom：From horizontal edge histogram HistRpEnd value start to be compared with its maximum value MaxR successively, until pth
Row HistRpMore than mean value AvgR and HistRpWhen being more than threshold value Th0 with the ratio of maximum value MaxR, then target connector region is enabled
Lower boundary bottom=p；
S3033, according to vertical edge histogram HistCqIt is compared with threshold value Th2, orients the left side in target connector region
Boundary left：From vertical edge histogram HistCqInitial value start to be compared with threshold value Th2 successively, as q row HistCqGreatly
When threshold value Th2, then the left margin left=q in target connector region is enabled；
S3034, according to vertical edge histogram HistCqIt is compared with threshold value Th2, orients the right in target connector region
Boundary right：From vertical edge histogram HistCqEnd value start to be compared with threshold value Th2 successively, as q row HistCq
When more than threshold value Th2, then the right margin right=q in target connector region is enabled；
S3035, according to coboundary top, lower boundary bottom, left margin left and right margin right, the mesh oriented for the first time
Mark connector region Ifirst。
4. component falls off defect identification method in contact net according to claim 2, which is characterized in that described to be located by connecting
Part step S3 further includes sub-step S305：Exclude positioning result wrong in the target connector region after multiple bearing.
5. component falls off defect identification method in contact net according to claim 1, it is characterised in that：The structural relation
Including relative position relation and area proportionate relationship.
6. component falls off defect identification method in contact net according to claim 1, which is characterized in that signature analysis step
Judgement target connector described in S5 includes following sub-step with the presence or absence of the doubtful defect that falls off：Calculate each target connector
Grey level histogram, and judge whether the maximum gray level of grey level histogram of each target connector identical or same grey level it
Between difference whether be less than threshold value Th3, if so, judging the target connector, there are the doubtful defects that falls off, and record connector
Fall off mark flag=1, otherwise judges that the doubtful defect that falls off is not present in the target connector, record connector, which falls off, indicates flag
=0.
7. component falls off defect identification method in contact net according to claim 1, which is characterized in that signature analysis step
Determine whether the doubtful defect that falls off is true, including following sub-step described in S5：
S501 calculates separately the horizontal gradient and vertical gradient of each target connector, calculates the argument θ of each pixel；
S502 counts the distribution situation of argument θ in entire target connector image, argument histogram is normalized,
Obtain argument features vector；
S503, the mark flag=1 if connector falls off calculate all targets connection that the target connector region is partitioned into
In part, per the mean square deviation of the argument features vector between target connector two-by-two, if there is one group of mean square deviation to be more than threshold value Th4,
It determines whether the doubtful defect that falls off is true, judges that the target connector falls off defect there are component；
S504, if connector falls off, whether mark flag=0, the argument principal direction of more all target connectors are identical；
If identical, it is determined that whether the doubtful defect that falls off is false, judges that the target connector is normal；
If differing, calculate in all target connectors that the target connector region is partitioned into, is connected per target two-by-two
The mean square deviation of argument features vector between part, if thering is one group of mean square deviation to be more than threshold value Th5 and being less than threshold containing one group of mean square deviation
Value Th6, it is determined that whether the doubtful defect that falls off is true, judges that the target connector falls off defect there are component.
8. component falls off defect recognition system in contact net, apply such as any one of claim 1-7 the method, it is special
Sign is：The system comprises sequentially connected image to be detected acquisition module, multiple dimensioned component locating module, connector positioning
Module, connector segmentation module and connector characteristics analysis module；
Image to be detected acquisition module is used for image to be detected of securing component；
The multiple dimensioned component locating module is searched for being slided on the different scale of image to be detected using component template image
Rope target element, matches the band of position image of target element in image to be detected, and the component template image includes connecing
The template image of various components in net-fault；
The connector locating module is used to orient target connection in the area image of position according to edge analysis algorithm
Part region；
The connector segmentation module is used for the structural relation according to target element and target connector, exists according to corresponding ratio
Target connector is partitioned into the image of the band of position；
The connector characteristics analysis module is used to obtain the grey level histogram and Gradient Features of target connector, is connected according to target
The grey level histogram judgement target connector of fitting whether there is the doubtful defect that falls off, further according to the Gradient Features of target connector
Determine whether the doubtful defect that falls off is true, if if true, judge that the target connector falls off defect there are component, if if false,
Judge that the target connector is normal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610049171.2A CN105719305B (en) | 2016-01-25 | 2016-01-25 | Component falls off defect identification method and system in contact net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610049171.2A CN105719305B (en) | 2016-01-25 | 2016-01-25 | Component falls off defect identification method and system in contact net |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105719305A CN105719305A (en) | 2016-06-29 |
| CN105719305B true CN105719305B (en) | 2018-10-19 |
Family
ID=56154081
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610049171.2A Active CN105719305B (en) | 2016-01-25 | 2016-01-25 | Component falls off defect identification method and system in contact net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105719305B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570857A (en) * | 2016-09-11 | 2017-04-19 | 西南交通大学 | High-speed railway overhead contact system lateral conductor fixation hook nut falling bad state detection method based on HOG features |
| CN106504238A (en) * | 2016-10-31 | 2017-03-15 | 成都交大光芒科技股份有限公司 | Railway contact line defect inspection method based on image procossing and convolutional neural networks |
| CN107230205A (en) * | 2017-05-27 | 2017-10-03 | 国网上海市电力公司 | A kind of transmission line of electricity bolt detection method based on convolutional neural networks |
| CN107680091B (en) * | 2017-10-12 | 2020-11-17 | 成都思晗科技股份有限公司 | Method for detecting falling defect of vibration damper of power transmission line |
| CN107588809B (en) * | 2017-10-24 | 2019-07-30 | 中国矿业大学 | A kind of frictional high-strength bolts in rod piece connecting node fall off recognition methods |
| CN109389566B (en) * | 2018-10-19 | 2022-01-11 | 辽宁奇辉电子***工程有限公司 | Method for detecting bad state of fastening nut of subway height adjusting valve based on boundary characteristics |
| CN110059631B (en) * | 2019-04-19 | 2020-04-03 | 中铁第一勘察设计院集团有限公司 | Contact net non-contact type monitoring defect identification method |
| CN112285111A (en) * | 2019-07-09 | 2021-01-29 | 株洲变流技术国家工程研究中心有限公司 | Pantograph front carbon sliding plate defect detection method, device, system and medium |
| CN110610516B (en) * | 2019-08-26 | 2023-03-24 | 湖南大学 | Railway fastener nut center positioning method |
| CN110910373B (en) * | 2019-11-25 | 2023-01-24 | 西南交通大学 | Identification method of orthotropic steel bridge deck fatigue crack detection image |
| CN111754508A (en) * | 2020-07-06 | 2020-10-09 | 中兴飞流信息科技有限公司 | Contact net limiting nail-stopping allowance state detection system |
| CN112365481B (en) * | 2020-11-13 | 2021-06-18 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting bolt loss of cross beam assembly based on image processing |
| CN113192071B (en) * | 2021-06-29 | 2021-11-23 | 南方电网数字电网研究院有限公司 | Insulator binding wire missing identification method and device and computer equipment |
| CN116385353B (en) * | 2023-02-10 | 2024-01-30 | 南通大学 | Camera module abnormality detection method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102567998A (en) * | 2012-01-06 | 2012-07-11 | 西安理工大学 | Head-shoulder sequence image segmentation method based on double-pattern matching and edge thinning |
| CN103400393A (en) * | 2013-08-21 | 2013-11-20 | 中科创达软件股份有限公司 | Image matching method and system |
| CN104318582A (en) * | 2014-11-14 | 2015-01-28 | 西南交通大学 | Detection method for bad state of rotating double-lug component pin of high-speed rail contact network on basis of image invariance target positioning |
| CN105139413A (en) * | 2015-09-25 | 2015-12-09 | 深圳大学 | Railway connecting piece quality detection method based on image characteristic extraction and detection system |
-
2016
- 2016-01-25 CN CN201610049171.2A patent/CN105719305B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102567998A (en) * | 2012-01-06 | 2012-07-11 | 西安理工大学 | Head-shoulder sequence image segmentation method based on double-pattern matching and edge thinning |
| CN103400393A (en) * | 2013-08-21 | 2013-11-20 | 中科创达软件股份有限公司 | Image matching method and system |
| CN104318582A (en) * | 2014-11-14 | 2015-01-28 | 西南交通大学 | Detection method for bad state of rotating double-lug component pin of high-speed rail contact network on basis of image invariance target positioning |
| CN105139413A (en) * | 2015-09-25 | 2015-12-09 | 深圳大学 | Railway connecting piece quality detection method based on image characteristic extraction and detection system |
Non-Patent Citations (3)
| Title |
|---|
| Automatic Fastener Classification and Defect Detection in Vision-Based Railway Inspection Systems;Hao Feng 等;《IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT》;20140430;第63卷(第4期);全文 * |
| 基于图像处理的接触网定位线夹目标检测;张春春 等;《铁道科学与工程学报》;20151231;第12卷(第6期);全文 * |
| 车载铁路路障智能图像检测技术研究;李东明;《中国优秀硕士学位论文全文数据库 工程科技II辑》;20071015;第2007年卷(第04期);第4.3.2节、图4.11 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105719305A (en) | 2016-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105719305B (en) | Component falls off defect identification method and system in contact net | |
| CN101625723B (en) | Rapid image-recognizing method of power line profile | |
| CN106919978B (en) | Method for identifying and detecting parts of high-speed rail contact net supporting device | |
| CN111814686A (en) | Vision-based power transmission line identification and foreign matter invasion online detection method | |
| CN105404867B (en) | A kind of substation isolating-switch state identification method of view-based access control model | |
| CN103512762A (en) | Image processing method and device and train fault detection system | |
| CN103400379B (en) | The detection of a kind of icing based on image and thickness method of estimation | |
| CN104318582A (en) | Detection method for bad state of rotating double-lug component pin of high-speed rail contact network on basis of image invariance target positioning | |
| CN108009591A (en) | A kind of contact network key component identification method based on deep learning | |
| CN105718964B (en) | A kind of visible detection method of power transmission line damper | |
| CN106679567A (en) | Contact net and strut geometric parameter detecting measuring system based on binocular stereoscopic vision | |
| CN106485694B (en) | A kind of high iron catenary double-jacket tube connector six-sided nut based on cascade classifier falls off defective mode detection method | |
| CN108680833B (en) | Composite insulator defect detection system based on unmanned aerial vehicle | |
| CN106407924A (en) | Binocular road identifying and detecting method based on pavement characteristics | |
| CN104794720A (en) | Method and system for positioning steady ears in catenary images | |
| CN111044197A (en) | Non-contact type cable force testing system and method based on unmanned aerial vehicle platform | |
| CN106570853A (en) | Shape and color integration insulator identification and defect detection method | |
| CN105976368A (en) | Insulator positioning method | |
| CN106570857A (en) | High-speed railway overhead contact system lateral conductor fixation hook nut falling bad state detection method based on HOG features | |
| CN106340019A (en) | Method for detecting adverse state of high-speed rail overhead line system inclined cable fixing hook | |
| Zhao et al. | Image-based comprehensive maintenance and inspection method for bridges using deep learning | |
| CN105354827A (en) | Method and system for intelligently identifying clamp nut shedding in catenary image | |
| CN106920240A (en) | A kind of insulator identification and method for diagnosing faults based on infrared image | |
| CN112508911A (en) | Rail joint touch net suspension support component crack detection system based on inspection robot and detection method thereof | |
| CN114030395B (en) | Foreign matter detection method and system for contact suspension dropper area |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |