CN110261436A - Rail deformation detection method and system based on infrared thermal imaging and computer vision - Google Patents

Rail deformation detection method and system based on infrared thermal imaging and computer vision Download PDF

Info

Publication number
CN110261436A
CN110261436A CN201910509281.6A CN201910509281A CN110261436A CN 110261436 A CN110261436 A CN 110261436A CN 201910509281 A CN201910509281 A CN 201910509281A CN 110261436 A CN110261436 A CN 110261436A
Authority
CN
China
Prior art keywords
image
gray
value
track
foreign matter
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910509281.6A
Other languages
Chinese (zh)
Other versions
CN110261436B (en
Inventor
李伟华
张敏
佘佳俊
杨皓然
梁祖懿
雷英佳
张泽恒
谭铭濠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jinan University
University of Jinan
Original Assignee
Jinan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jinan University filed Critical Jinan University
Priority to CN201910509281.6A priority Critical patent/CN110261436B/en
Publication of CN110261436A publication Critical patent/CN110261436A/en
Application granted granted Critical
Publication of CN110261436B publication Critical patent/CN110261436B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N25/00—Investigating or analyzing materials by the use of thermal means
    • G01N25/72—Investigating presence of flaws
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10048—Infrared image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Abstract

The rail deformation detection method and system based on infrared thermal imaging and computer vision that the invention discloses a kind of, steps of the method are: unmanned plane carries out tramway Image Acquisition；Ground station reception high-definition camera image data carries out image preprocessing；It carries out multi-threshold orbital region twice to brighter areas outside darker area in slot and slot to divide, the distance feature adjacent according to brighter darker area divides orbital region, and extraction obtains orbital image；Infrared Thermogram carries out gray processing, extracts the high-temperature region on track using relative temperature difference method；Pretreated image is superimposed with track detecting window, and exposure mask obtains area-of-interest, and area-of-interest carries out edge closure judgement and filling obtains connected region, and screening connected region obtains doubtful track foreign matter；Doubtful track foreign matter input BP neural network is identified, foreign matter classification results are obtained.Real-time perfoming track foreign matter identification of the present invention and temperature detection, reduce the accident rate of rail traffic, improve electric car safety in operation.

Description

Rail deformation detection method and system based on infrared thermal imaging and computer vision
Technical field
The present invention relates to track fields, and in particular to a kind of rail based on infrared thermal imaging and computer vision Road fault detection method and system.
Background technique
Modern tram is come into people's lives, but is compared with environmentally protective, safety and comfort, flexibly convenient feature For subway, modern tram does not have completely self-contained right of way, and tramway is Chong Die with car lane or intersects, when rail electricity Vehicle travels with faster speed, the biggish passenger traffic volume when, the foreign matter on track will cause very big prestige to the traffic safety of electric car The side of body.Streetcar track fault detection technique at present relies primarily on manual work mode and detects and safeguard --- detection speed is slow, Consuming time is long, safety is low, human cost is high, and tramway fault detection has even influenced the day-to-day operation of electric car, Tramway road section traffic volume is hindered, and blocks urban transportation.The technology of existing detection of obstacles, according to detection method master It is divided into: obstacle detection method based on computer vision and the obstacle detection method based on radar, the barrier based on radar Hinder object detecting method to launch signal by radar, and by measurement transmitting signal and reflects the time difference etc. between signal Physical quantity obtains the range information of sensor and target.The method stability of radar detection is poor, commissioning device is complicated, cost compared with It is high.Obstacle detection technology based on computer vision relies primarily on the figure that the camera being installed on vehicle obtains vehicle front As information, and barrier is detected with digital image processing techniques, although having certain warning function for detection zone, But detection zone is narrow, is only capable of detecting barrier when vehicle travels, Daily Round Check when can not handle track clear and urgent In the case of failure inspection.
In conclusion existing detection technique has certain limitation, therefore, how tramway is efficiently accurately detected Failure become urgent problem to be solved.
Summary of the invention
In order to overcome shortcomings and deficiencies of the existing technology, the present invention provides one kind to be based on infrared thermal imaging and calculating The rail deformation detection method and system of machine vision carry unmanned plane and carry out inspection, using temperature detection and image recognition to electricity Vehicle rail fault carries out comprehensive detection, and detection and track foreign matter including track power supply system short circuit thermal (as stopped vehicle in violation of rules and regulations , discarded bicycle, ratchel) detection, carried out using short circuit thermal situation of the unmanned plane to track foreign matter and power supply system Automatic detection, can carry out efficiently monitoring in real time in a sudden situation, reduce the accident rate of rail traffic, improve track Safety.
In order to achieve the above object, the invention adopts the following technical scheme:
The present invention provides a kind of rail deformation detection method based on infrared thermal imaging and computer vision, including following steps It is rapid:
S1: high-definition camera and infrared thermal imager are mounted on unmanned plane, will acquire during unmanned plane inspection Orbital image pass earth station back in real time；
S2: image preprocessing: ground station reception high-definition camera image data carries out image preprocessing, and described image is pre- Processing includes image gray processing, image filtering, image reinforces and edge detection；
S3: it extracts orbital image: high-definition camera image data being carried out using darker gray threshold in track groove primary After segmentation, then brighter gray threshold is split outside using track groove, the last distance feature adjacent according to brighter darker area Divide orbital region, extraction obtains orbital image；
S4: infrared temperature detection: according to position of the orbital image extracted in the collected original image of high-definition camera Information obtains corresponding track position in Infrared Thermogram in conjunction with the position and angular relationship of infrared thermal imager and high-definition camera It sets, and by the infrared Infrared Thermogram gray processing received at thermal imaging system, extracts gray value, track is judged using relative temperature difference method It is upper to whether there is high-temperature region, high-temperature region and zoning area and maximum temperature point are then extracted if it exists；
S5: doubtful track foreign matter screening: pretreated image is overlapped with obtained orbital image is extracted, exposure mask Area-of-interest is obtained, edge closure judgement and filling are carried out to area-of-interest, obtain connected region, connected region is carried out Screening, obtains doubtful track foreign matter；
S6: the identification of track foreign matter: doubtful track foreign matter being inputted in BP neural network and is identified, obtains foreign matter classification knot Fruit.
Image preprocessing described in step S2 includes image gray processing, image filtering, image as a preferred technical solution, Reinforcement and Image Edge-Detection, specific steps are as follows:
S21: high-definition camera collects color image, after carrying out gray processing processing, obtains image Pgray, indicate are as follows:
Pgray=0.30R+0.59G+0.11B；
Wherein, R indicates that the pixel value of the red component in color image, G indicate the pixel of color image Green component Value, B indicate the pixel value of the blue component in color image；
S22: image filtering is carried out using discrete Gaussian filter function, image is weighted and averaged, using Gaussian template Each pixel in scan image, it is described discrete with the gray value at the weighted mean substitution Gaussian template center of neighborhood of pixels Gaussian filter function H (i, j) are as follows:
Wherein, (i, j) indicates that any coordinate in neighborhood, δ indicate standard deviation；
S23: changing image pixel gray level value and carry out image reinforcement, and treated, and image pixel value is g (x, y), indicates are as follows:
G (X, y)=[f (x, y)]2；
Wherein, f (x, y) is indicated through image gray processing and image filtering treated pixel value of the image at (x, y) point, Image grayscale range is [0,255], if the result g (x, y) calculated is set as 255 more than 255；
S24: Canny detective operators are chosen and carry out Image Edge-Detection: first with Gaussian mask and through image gray processing and figure As the image after filtering processing does convolution algorithm, the Information invariability of single pixel, then with single order local derviation Difference Calculation gradient Amplitude and direction, then the inhibition of non-maximum is carried out with gradient magnitude, image border, institute are finally detected and connected with dual-threshold voltage The amplitude and direction for stating gradient respectively indicate are as follows:
Wherein, Sx、SyRespectively represent x, the partial derivative of the image grayscale on the direction y.
As a preferred technical solution, described in step S3 using darker gray threshold in track groove to high-definition camera figure After carrying out once segmentation as data, then brighter gray threshold is split outside using track groove, specific calculation formula are as follows:
Wherein, f (x, y) indicates pretreated gray level image, TLIndicate darker area minimum gray value in slot, THIt indicates The outer brighter areas gray scale maximum value of slot,Respectively indicate TLAnd THThe gray scale fluctuated up and down Grade range；
The adjacent distance feature of the brighter darker area of foundation divides orbital region, and extraction obtains orbital image, specifically Step are as follows:
To darker area binary map g in slotL(x,y)With brighter areas binary map g outside slotH(x,y)It is expanded to obtain corresponding area Regional partition binary mapWithAnd intersection is solved, obtain orbital region segmentation binary map gu(x, y) is indicated Are as follows:
In track region segmentation binary map guThe starting pixels point of two siding tracks is determined on (x, y), tracking obtains on track Multiple pixels, obtain a plurality of trajectory line, the trajectory line of two siding tracks extracted from a plurality of trajectory line, using least square Segmentation carries out quadratic fit, constructs orbit equation, the orbital image extracted.
Relative temperature difference method described in the detection of step S4 infrared temperature as a preferred technical solution, specific steps are as follows:
S41: temperature value on the display screen of infrared thermal imager is read；
S42: by the processing of infrared thermal imaging figure gray processing, the information matrix of brightness value is obtained；
S43: temperature value and gray value setting mapping relations are expressed as:
Wherein, G indicates that gray value, T indicate temperature；
S44: temperature value when working normally to the rail temperature result detected with track compares, quasi- using curve Conjunction obtains rail temperature variation tendency, according to the mapping relations of temperature and gray scale, replaces temperature value with gray value, obtains gray scale threshold Value；
S45: being split fault zone by gray threshold, sets different gray values, and mention using edge detection The fault zone more than gray threshold is taken out, the pixel counts in fault zone are obtained into region area, compare the big of gray value It is small to obtain maximum gradation value pixel, obtain maximum temperature point.
The orbital image that pretreated image and extraction obtain is carried out described in step S5 as a preferred technical solution, Superposition, Superposition Formula are as follows:
S (i, j)=R (i, j) &ROI；
Wherein, R (i, j) indicates that pretreated image, ROI indicate that area-of-interest, S (i, j) indicate operation result figure Picture；
It is described that connected region is screened, screen formula are as follows:
DArea≥S；
DHeight≥D∩DWidth≥Dlow∩DWidth≤Dhigh；
Wherein, DAreaRepresent the number of pixel shared by connected region, DHeightRepresent the height of the outer area-encasing rectangle of connected region Degree, DWidthRepresent the width of the outer area-encasing rectangle of connected region, S, D, Dlow、Dhigh、DRatioRespectively represent the face of doubtful track foreign matter Diagonal line length, the rectangular degree of the length of long-pending, external minimum rectangle, wide, the external minimum rectangle of external minimum rectangle；
When screening formula while setting up, the connected region after screening is doubtful track foreign matter.
The training step of the BP neural network as a preferred technical solution, are as follows:
S60: numerical value initialization: setting BP neural network input layer number n, hidden node number l and output layer Number m, if the weight of input layer to hidden layer (hidden layer) is ωij, hidden layer to output layer weight be ωjk, input layer is to implicit The threshold value of layer is aj, hidden layer to output layer threshold value be bk, learning rate is η and excitation function is g (x), the excitation function G (x) uses Sigmoid function, indicates are as follows:
Wherein x is input matrix；
S61: input training sample: using the orbital image of high-definition camera shooting as original image, acquisition includes wait know The image pattern of other foreign matter carries out image gray processing and binary conversion treatment, the binary image of sample is obtained, by the sample of acquisition The unified size for arriving same ratio, is input in BP neural network；
S62: whether training of judgement sample, which is loaded into, finishes, and finishes if being loaded into, and performs the next step suddenly, finishes, hold if being not loaded with Row step S61；
S63: the output of hidden layer is set as Hj, calculate the output of hidden layer neuron:
Wherein, n is input layer number, ωijFor the weight of input layer to hidden layer, xiFor input matrix, ajIt is defeated Enter layer to hidden layer threshold value；
S64: the output of output layer is set as Ok, calculate the output of output layer neuron:
Wherein, l is hidden layer node number, ωjkFor the weight of hidden layer to output layer, bkOutput layer is arrived for hidden layer Threshold value；
S65: error is calculated:
Wherein, ekFor error, m is output layer node number, YkFor desired output, OkFor the output of output layer；
S66: right value update:
ωjk=ωjk+ηHjek
Wherein, ωijFor the weight of input layer to hidden layer, ωjkFor the weight of hidden layer to output layer, η is study speed Rate, HjFor the output of hidden layer, xiFor input matrix, m is output layer node number, ekFor error；
S67: threshold value updates:
bk=bk+ηek；
Wherein, ajFor the threshold value of input layer to hidden layer, bkFor the threshold value of hidden layer to output layer, ωjkIt is arrived for hidden layer The weight of output layer, η are learning rate, HjFor the output of hidden layer, xiFor input matrix, m is output layer node number, ekFor Error；
S68: judging whether the difference between adjacent error twice is less than setting value, if being less than setting value, BP neural network Training terminates, if being not less than setting value, recycles and executes step S63-S67.
The rail deformation detection system based on infrared thermal imaging and computer vision that the present invention also provides a kind of, comprising: nothing Man-machine and earth station, the unmanned plane include main control module, flight control modules, navigation module, wireless communication module and take photo by plane Module, main control module control navigation module, the wireless communication module and module of taking photo by plane, the flight control modules are for controlling The state of flight of unmanned plane, the navigation module is used to provide navigation to unmanned plane, the wireless communication module is used for unmanned plane It is communicated wirelessly with earth station, for the module of taking photo by plane for obtaining orbital image, the module of taking photo by plane includes high-definition camera And infrared thermal imager；
The earth station includes image pre-processing module, orbital image extraction module, extracts high-temperature region module, doubtful track Foreign matter screening module and track foreign matter identification module, described image preprocessing module is for carrying out high-definition camera image data Pretreatment, orbital image extraction module extract orbital image, the extraction high-temperature region for carrying out multi-threshold orbital region segmentation Module extracts high-temperature area and maximum temperature point in Infrared Thermogram, the doubtful track for carrying out relative temperature difference judgement Foreign matter screening module obtains doubtful track foreign matter for screening connected region, the screening connected region for area-of-interest into It is obtained after the judgement of row edge closure and filling, the track foreign matter identification module is equipped with BP neural network, is used for doubtful track Foreign matter input BP neural network is identified to obtain foreign matter classification results.
The BP neural network is equipped with input layer, hidden layer and output layer, the input layer as a preferred technical solution, Input node quantity be set as 20, the output node quantity of the output layer is set as 3, the hidden layer node of the hidden layer Quantity is set as:
Wherein, n is hidden layer node quantity, niFor input node number, noFor output node number, a takes [1,10] range Interior constant.
Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) unmanned plane is applied to rail polling by the present invention, in electric car non-operation period or catastrophic failure, to tram rail Road carries out inspection, has the characteristics that precise and high efficiency, unobscured, when unmanned plane is navigated by water along track routes, camera and ground base This is parallel, and image background does not change much, obtained dynamic background can approximation regard static state as, it is dry to considerably reduce background It disturbs, convenient for extracting effective image information.
(2) present invention extracts orbital image by the thought of multi-threshold method and skeletal extraction, using in track groove compared with dull gray Spend threshold value to carry out after once dividing original image, brighter gray threshold outside track groove recycled to be split, it is last according to it is brighter compared with The adjacent distance feature of dark areas divides orbital region, extracts accurate orbit information, effectively prevents the dry of redundant information It disturbs.
(3) present invention effectively reduces pseudo- foreign matter by doubtful track foreign matter screening step using computer vision technique It influences, and carries out target detection with artificial neural network, improve the accuracy of track foreign matter identification.
(4) present invention uses infrared thermal imaging temperature detection technology, will be warm using the mapping relations between temperature and gray value Degree threshold value is converted into gray threshold, can quick and precisely position and mark the excessively high position of track power supply system temperature, by fault point Information returns to earth station, understands fault condition in time convenient for staff, follow-up work is unfolded.
Detailed description of the invention
Fig. 1 is that the process of rail deformation detection method of the present embodiment based on infrared thermal imaging and computer vision is illustrated Figure；
Fig. 2 is that the infrared temperature of rail deformation detection method of the present embodiment based on infrared thermal imaging and computer vision is examined Flow gauge schematic diagram；
Fig. 3 is that the doubtful track of rail deformation detection method of the present embodiment based on infrared thermal imaging and computer vision is different Object screening process schematic diagram；
Fig. 4 is the present embodiment BP neural network structural schematic diagram；
Fig. 5 is the present embodiment BP neural network training flow diagram；
Fig. 6 is the unmanned plane structure of rail deformation detection system of the present embodiment based on infrared thermal imaging and computer vision Schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and It is not used in the restriction present invention.
Embodiment
As shown in Figure 1, the present embodiment provides a kind of rail deformation detection side based on infrared thermal imaging and computer vision Method carries unmanned plane, carries out comprehensive detection, including track power supply system to tramway failure using temperature detection and image recognition The detection of the detection of system short circuit thermal and track foreign matter (violation parked vehicles, discarded bicycle, ratchel), has installation side Just, debugging is simple；Unobscured, precise and high efficiency detects failure；The low advantage of human cost.
Rail deformation detection method provided in this embodiment based on infrared thermal imaging and computer vision is in python ring It is developed under border, has used OpenCV computer vision library, and carry out to image by temperature threshold conversion method and BP neural network Analysis processing, detailed process comprise the following steps:
S1: high-definition camera and infrared thermal imager are mounted on unmanned plane, adjust angle and focusing parameter information, control Unmanned plane processed carries out inspection, the tramcar rail for getting camera during unmanned plane inspection according to preset route Road image passes image data back earth station by 5G network in real time；
S2: image preprocessing, ground station reception realtime image data, first progress image preprocessing, including image grayscale Change, image filtering, image reinforcement, edge detection etc., it is possible to reduce what the variation of weather, road conditions, lighting angle identified track Significant impact improves anti-interference ability, and prominent orbit information rapidly and accurately detects errant foreign matter, improves image, specific to walk It is rapid as follows:
S21: collecting color image from ground station reception to high-definition camera, includes a large amount of such as object color information, The time is handled in order to shorten, calculation amount is reduced by the way of image gray processing, the useful information of prominent image uses formula:
Pgray=0.30R+0.59G+0.11B (1)
Wherein, R represents the pixel value of the red component in color image, and G represents the pixel of color image Green component Value, B represent the pixel value of the blue component in color image, PgrayIt represents through the gray level image after conversion；
S22: for improve picture quality, improve the image quality issues due to caused by noise jamming, using gaussian filtering into Row image denoising, for the present embodiment using each pixel in the Gaussian template scan image of 9*9, Gaussian filter is adjacent with pixel The gray value at the weighted mean substitution Gaussian template center in domain, since each neighborhood territory pixel point weight is with the point and central point Distance and reduce, enable to image more smooth,
The discrete Gaussian filter function that the present embodiment uses are as follows:
Wherein, H (i, j) is filter function, and (i, j) is the coordinate of any in neighborhood, and δ is standard deviation；
In the present embodiment, any coordinate (i, j) in neighborhood, substitutes intoObtained Gaussian function Coefficient of the numerical value as template；
S23: it is compared to improve image definition and contrast directly by being changed to image pixel gray level value Degree stretches to enhance image effect:
G (X, y)=[f (x, y)]2(3)；
Wherein, f (x, y) is through image gray processing and image filtering treated pixel value of the image at (x, y) point, g (x, y) is treated pixel value, and image grayscale range is 0-255, if the result calculated is more than 255, is set as 255, The method that image is reinforced is that the brightness and contrast in dark images part weakens, in the brightness and contrast of relatively bright part Reinforce；
S24: Canny detective operators are chosen and obtain complete image profile；First with Gaussian mask and through image gray processing and figure As the image after filtering processing does convolution algorithm, the Information invariability of single pixel, secondly, with single order local derviation Difference Calculation gradient Amplitude and direction, use Sx、SyX is represented, the partial derivative of the image grayscale on the direction y, then the amplitude of gradient and direction respectively indicate Are as follows:
The inhibition of non-maximum is carried out with gradient magnitude again, finally detects and connect edge with dual-threshold voltage.
S3: extracting orbital image, and the detection range of tramway failure is mainly between rail and above track, and tramway Using spill rail, have the characteristics that it is bright outside undercut in slot, using multi-threshold orbital region split plot design, to darker area in slot and slot Outer brighter areas is divided twice, accurately extracts orbit information；
Between the multi-threshold extracts orbital image to be according to the detection range of tramway failure be mainly rail and track The feature of top will accurately extract orbit information, first carry out orbital region segmentation, then extract trajectory characteristic point.Due to rail Tramway uses spill rail, darker in concave tracks slot, and brighter outside slot, grey value difference is obvious, and in slot darker area and The outer brighter areas of slot is adjacent, so the high-definition camera received using gray threshold darker in track groove to earth station is original Image carries out after once dividing, and brighter gray threshold outside track groove is recycled to be split, last according to brighter darker area phase Adjacent distance feature divides orbital region, obtains clear orbital image.Defining pretreated gray level image is f (x, y), is determined Darker area minimum gray value T in slotLWith brighter areas gray scale maximum of T outside slotH.Darker area segmentation and slot are brighter outside in slot Region segmentation carries out as follows respectively:
Wherein,It is T respectivelyLAnd THThe grey level range fluctuated up and down, to slot Interior darker area binary map gL(x,y)With brighter areas binary map g outside slotH(x,y)It is expanded to obtain corresponding region segmentation binary mapWith
It seeks common ground, is shown below to the two width region segmentation binary maps obtained after expansion, obtain what interference largely reduced The more complete region segmentation binary map of orbital region.
On the basis of Threshold segmentation binary map, first according to stringent " detection criterion " search track starting point pixel, then Other pixels on target object are found using the thought of skeletal extraction according to the position of these points and positional relationship, are finally utilized Certain priori knowledge rejects interference line segment, using the method construct orbit equation of least square segmentation quadratic fit. Detailed process is as follows:
(1) in track region segmentation binary map guThe starting pixels point of left and right sides track is determined on (x, y)；
In guThe possible starting point x of two siding tracks is determined on (x, y)Ln,xRm, n, m=1,2,3 ..., it defines left rail and rises Initial point search range [xLs,xLe], right rail starting point search range [xRx,xRe], Y row is searched for, Y starts from image base, The midpoint x in cross connection region is found in search rangeLn∈{xLs,xLe, and xRm∈{xRS,xRe, all points are considered as Possible starting point, if searching for Y-1 row without starting point in search range, and so on, if until YminRow also fails to look for To starting point, then the search of this siding track starting point is abandoned；
(2) tracking obtains other pixels on track, obtains a plurality of trajectory line；
A trajectory line can be obtained using tracking each starting point of criteria track, with xL1For, if xL1Locating Cross connection region is [xL1s,xL1e], it is defined as the initial value of track-while-scan range, locating line number is yL1, this region is prolonged It opens up as [xL1s-Te,xL1e+Te] it is used as new a line yL1- 1 search range, TeFor the pixel number of left and right continuation, the area is searched for The midpoint x of several connected regions in domainL11,xL12,…,xL1k, these points are considered as the tracing point traced into, if in region not Tracing point can be found, then searches for yL1- 2 rows, and so on, if continuous YrowRow fails to find tracing point, then abandons this trajectory line Search, if having traced into tracing point in current line it is necessary to the track-while-scan range of the new a line of determination；If yL1- 1 has traced into rail Mark point, then by xL11The left end point x of locating connected regionL11sAnd xL1kThe right endpoint x of locating connected regionL1keIt is subject to continuation conduct New track-while-scan range [xL11s-Te,xL1ke+Te], it repeats step before this and carries out the search of new a line tracing point, until meeting Limit requirement；
(3) longest most complete left and right two trajectory line is extracted from a plurality of trajectory line；
It is most complete that longest is chosen from a plurality of trajectory line, as railway line.The x that pixel is trackedL1, xL2,…,xLn Pixel is the largest number of in trajectory line is used as left side rail diatoms, the x that pixel is trackedR1,xR2,…,xRmIn trajectory line Pixel is the largest number of to be used as right side rail diatom；
(4) using the method construct orbit equation of least square segmentation quadratic fit；
Search order segmentation is pressed to collected left and right track characteristic, it is secondary using least square method progress per N number of point Fitting.The case where limit requires is not able to satisfy for track starting point and ending point, respectively to N number of point of beginning and last N A point using least square method carry out quadratic fit and extend until meet limit requirement, N specific size view actual conditions and It is fixed；
S4: infrared temperature detection: according to location information of the orbital image extracted in high-definition camera original image, in conjunction with Certain error is eliminated in position and angular relationship between infrared thermal imager and high-definition camera, and it is right in Infrared Thermogram to obtain The orbital position answered, the Infrared Thermogram gray processing that infrared thermal imager is received extract gray value, using relative temperature difference method Judge then to extract high-temperature region and zoning area and maximum temperature point if it exists with the presence or absence of high-temperature region on track, save note Staff is recorded and be sent to, next frame image is otherwise continued to test；
As shown in Fig. 2, the infrared temperature detection of the present embodiment is with the temperature value of image and the associated mapping of gray value Based on feature, the temperature value and gray value of each pixel of Infrared Thermogram are acquired, preset temperature threshold is converted to gray scale Threshold value determines temperature rise region；According to the feature of Infrared Thermogram fault zone, fault zone area and mass center, specific steps are calculated It is as described below:
S41: temperature value is directly read out from the display screen of infrared thermal imager；
S42: Infrared Thermogram gray processing is handled, and obtains the information matrix of brightness value, and value range is [0,255]；
S43: since Infrared Thermogram has the characteristics that edge blurry, poor contrast, it is unfavorable for equipment analysis, the present embodiment The contrast that image is improved using the mapping function of temperature value and gray value, since gray value data is by direct gray processing Reason obtains, so there are mapping relations between temperature value and gray value, it is quasi- to temperature value and gray value implementation data to choose sample point It closes；
Wherein, T and G respectively indicates temperature and gray value；
S44: using longitudinal comparison (i.e. relative temperature difference method) method, to the rail temperature result and the normal work of track detected Temperature value when making is analyzed, and obtains rail temperature variation tendency using the method for curve matching, according to temperature and gray scale Mapping relations formula replaces temperature value with gray value, obtains gray threshold；
S45: the segmentation of fault zone, zoning area and fever center are carried out by gray threshold；It will be more than ash The gray value of degree threshold portion is set as 255 (being shown as white), and the gray value of other parts is set as 0 (being shown as black), And extracted based on edge detection more than gray threshold part (i.e. fault zone), the present embodiment uses in MATLAB software Region-props function obtains region area to the pixel counts in fault zone, compares gray value size and obtains maximum gray scale It is worth pixel, i.e. fever central point；
In the present embodiment, the line anomalies of track power supply system, are often accompanied by fever phenomenon, in switchgear contact, lead When wire terminal etc. poor contact, after being passed through electric current, increased since thermal losses will lead to local temperature, for track and two rails Range between road carries out high temperature detection, can further analyze and determine whether route is abnormal, solve in time abnormal；
S5: pretreated image is overlapped, exposure mask by doubtful track foreign matter screening with obtained orbital image is extracted Area-of-interest is obtained, edge closure judgement and filling are carried out to area-of-interest and is further screened using formula, is excluded dry It disturbs；
As shown in figure 3, the doubtful track foreign matter screening technique of the present embodiment is will be pretreated using the method for exposure mask Image is superimposed with the orbital image extracted, Superposition Formula are as follows:
S (i, j)=R (i, j) &ROI (10)
Wherein R (i, j) is pretreated image, and ROI is area-of-interest, passes through the logic of R (i, j) and ROI region Operation makes only to retain the edge detecting information in area-of-interest in operation result image S (i, j)；
Judge whether the edge of the edge-detected image in area-of-interest is closed, closure image is filled, as Connected region.The statistics such as area, size, duty ratio are carried out to connected region to screen, and screen formula are as follows:
DArea≥S (11)
DHeight≥D∩DWidth≥Dlow∩DWidth≤Dhigh； (12)
Wherein, DAreaRepresent the number of pixel shared by connected region, DHeightRepresent the height of the outer area-encasing rectangle of connected region Degree, DWidthRepresent the width of the outer area-encasing rectangle of connected region, S, D, Dlow、Dhigh、DRatioRespectively represent the shape of doubtful track foreign matter Shape feature constant value: be respectively area, the length of external minimum rectangle, wide, the external minimum rectangle of external minimum rectangle it is diagonal Wire length, rectangular degree；
When (11) (12) (13) three formula meet simultaneously, gained connected region is doubtful track foreign matter.
S6: the identification of track foreign matter, it is inverse using multilayer feedforward neural network and error most widely used in artificial neural network Learning algorithm is propagated, i.e. BP neural network, the doubtful track foreign matter that previous step is filtered out is put into trained sample database in advance In identified, obtain specific foreign matter as a result, save data and report to staff.
In the present embodiment, the identification of track foreign matter uses multilayer feedforward neural network and error Back-Propagation learning algorithm, i.e., BP neural network, as shown in figure 4, BP neural network is by up of three-layer, input layer, hidden layer and output layer.
Track foreign matter is divided into three classes by this example: violation parked vehicles, discarded bicycle, ratchel, so input is saved Point is set as 20, and output node is set as 3, and the selection of hidden layer node determines according to the following formula:
Wherein, n is node in hidden layer；niFor input node number；noFor output node number；A takes normal between 1-10 Number.
As shown in figure 5, the training step that the present embodiment carries out neural network is as follows:
S60: weight and threshold value are initialized:
Determine BP neural network input layer number n=20, hidden node number l=8 and output layer number m=3, if The weight of input layer to hidden layer is ωij, hidden layer to output layer weight be ωjk, input layer to hidden layer threshold value be aj、 The threshold value of hidden layer to output layer is bk, learning rate is η and excitation function is g (x), wherein excitation function g (x) takes Sigmoid function indicates are as follows:
Wherein x is input matrix, random assignment ωij, ωjk, aj, bk；
S61: input training sample；
Multiple images that UAV system inspection is passed back obtain the image sample including foreign matter to be identified as original image This, the foreign matter image pattern to be identified of the present embodiment uses the RGB color figure for parking cars with violation, discarding bicycle etc., Image gray processing and binary conversion treatment are carried out, the binary image of sample is obtained, for trained accuracy and rapidity, is protecting Hold vehicle and bicycle shape feature it is distortionless under the conditions of, by the unified size to same ratio of the sample of acquisition: 300 pictures Plain high, 200 pixels are wide, and the method that the present embodiment uses is to fill background colour, i.e. black picture element in the periphery of all samples；
S62: whether training of judgement sample, which is loaded into, finishes, and finishes if being loaded into, and performs the next step suddenly, finishes, hold if being not loaded with Row step S61；
S63: the output of hidden layer is set as Hj, the output of each neuron of hidden layer is calculated according to the following formula:
Wherein, n is input layer number, ωijFor the weight of input layer to hidden layer, xiFor input matrix, ajIt is defeated Enter layer to hidden layer threshold value；
S64: the output of output layer is set as Ok, the output of each neuron of output layer is calculated according to the following formula:
Wherein, l is hidden layer node number, ωjkFor the weight of hidden layer to output layer, bkOutput layer is arrived for hidden layer Threshold value；
S65: error is calculated:
Wherein, ekFor error, m is output layer node number, YkFor desired output, OkFor the output of output layer；
S66: right value update:
ωjk=ωjk+ηHjek
Wherein, ωijFor the weight of input layer to hidden layer, ωjkFor the weight of hidden layer to output layer, η is study speed Rate, HjFor the output of hidden layer, xiFor input matrix, m is output layer node number, ekFor error；
S67: threshold value updates:
bk=bk+ηek
Wherein, ajFor the threshold value of input layer to hidden layer, bkFor the threshold value of hidden layer to output layer, ωjkIt is arrived for hidden layer The weight of output layer, η are learning rate, HjFor the output of hidden layer, xiFor input matrix, m is output layer node number, ekFor Error；
S68: judging whether the difference between adjacent error twice is less than specified value, i.e., whether reaches training objective, if Reach, then training terminates, if not up to, recycling and executing step S63-S67.
After neural metwork training is good, doubtful track foreign matter image obtained in previous step is put into network and is identified, really Determine foreign matter type.
The present embodiment also provides a kind of rail deformation detection system based on infrared thermal imaging and computer vision, comprising: Unmanned plane and earth station；
As shown in fig. 6, unmanned plane includes main control module, flight control modules, navigation module, wireless communication module and takes photo by plane Module, main control module control navigation module, the wireless communication module and module of taking photo by plane, the flight control modules are for controlling The state of flight of unmanned plane, the navigation module is used to provide navigation to unmanned plane, the wireless communication module is used for unmanned plane It is communicated wirelessly with earth station, for the module of taking photo by plane for obtaining orbital image, the module of taking photo by plane includes high-definition camera And infrared thermal imager, high-definition camera and infrared thermal imager and parallel track, for obtaining orbital image；
In the present embodiment, earth station includes image pre-processing module, orbital image extraction module, extracts high-temperature region mould Block, doubtful track foreign matter screening module and track foreign matter identification module, described image preprocessing module are used for high-definition camera Image data is pre-processed, and orbital image extraction module extracts orbital image, institute for carrying out multi-threshold orbital region segmentation It states and extracts high-temperature region module for carrying out relative temperature difference judgement, high-temperature area and high temperature dot in extraction Infrared Thermogram are described Doubtful track foreign matter screening module obtains doubtful track foreign matter, the screening connected region is to feel emerging for screening connected region Interesting region obtains after carrying out edge closure judgement and filling, and the track foreign matter identification module is equipped with BP neural network, and being used for will Doubtful track foreign matter input BP neural network is identified to obtain foreign matter classification results.
The present embodiment by unmanned plane, can within more than ten minutes to carrying out comprehensive detection above tramway, The safety that people and fault point are kept apart to effective guarantee testing staff reduces work risk, while time-consuming short raising The image that UAV system is sent back is carried out temperature detection and image recognition is in real time handled by the efficiency of fault detection, and the One time announcement maintenance personal reports abort situation, failure cause, helps to carry out maintenance work, accuracy high-timeliness is good.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention, It should be equivalent substitute mode, be included within the scope of the present invention.

Claims (8)

1. a kind of rail deformation detection method based on infrared thermal imaging and computer vision, which is characterized in that including following steps It is rapid:
S1: high-definition camera and infrared thermal imager are mounted on unmanned plane, the rail that will acquire during unmanned plane inspection Road image passes earth station back in real time；
S2: image preprocessing: ground station reception high-definition camera image data carries out image preprocessing, described image pretreatment Including image gray processing, image filtering, image reinforces and edge detection；
S3: it extracts orbital image: high-definition camera image data once being divided using gray threshold darker in track groove Afterwards, then using brighter gray threshold outside track groove it is split, the last distance feature segmentation adjacent according to brighter darker area Orbital region, extraction obtain orbital image；
S4: infrared temperature detection: according to location information of the orbital image extracted in the collected original image of high-definition camera, Corresponding orbital position in Infrared Thermogram is obtained in conjunction with the position and angular relationship of infrared thermal imager and high-definition camera, and By the infrared Infrared Thermogram gray processing received at thermal imaging system, extract gray value, judged using relative temperature difference method be on track No there are high-temperature regions, then extract high-temperature region and zoning area and maximum temperature point if it exists；
S5: doubtful track foreign matter screening: pretreated image is overlapped with obtained orbital image is extracted, exposure mask obtains Area-of-interest carries out edge closure judgement and filling to area-of-interest, obtains connected region, sieve to connected region Choosing, obtains doubtful track foreign matter；
S6: the identification of track foreign matter: doubtful track foreign matter is inputted in BP neural network and is identified, foreign matter classification results are obtained.
2. the rail deformation detection method according to claim 1 based on infrared thermal imaging and computer vision, feature It is, image preprocessing described in step S2 includes image gray processing, image filtering, image reinforces and Image Edge-Detection, tool Body step are as follows:
S21: high-definition camera collects color image, after carrying out gray processing processing, obtains image Pgray, indicate are as follows:
Pgray=0.30R+0.59G+0.11B；
Wherein, R indicates that the pixel value of the red component in color image, G indicate the pixel value of color image Green component, B Indicate the pixel value of the blue component in color image；
S22: image filtering is carried out using discrete Gaussian filter function, image is weighted and averaged, is scanned using Gaussian template Each pixel in image, with the gray value at the weighted mean substitution Gaussian template center of neighborhood of pixels, the discrete Gauss Filter function H (i, j) are as follows:
Wherein, (i, j) indicates that any coordinate in neighborhood, δ indicate standard deviation；
S23: changing image pixel gray level value and carry out image reinforcement, and treated, and image pixel value is g (x, y), indicates are as follows:
G (x, y)=[f (x, y)]2；
Wherein, f (x, y) is indicated through image gray processing and image filtering treated pixel value of the image at (x, y) point, image Tonal range is [0,255], if the result g (x, y) calculated is set as 255 more than 255；
S24: Canny detective operators are chosen and carry out Image Edge-Detection: first filtered with Gaussian mask and through image gray processing and image Then wave treated image does convolution algorithm, the Information invariability of single pixel use the amplitude of single order local derviation Difference Calculation gradient And direction, then the inhibition of non-maximum is carried out with gradient magnitude, image border, the ladder are finally detected and connected with dual-threshold voltage The amplitude of degree and direction respectively indicate are as follows:
Wherein, Sx、SyRespectively represent x, the partial derivative of the image grayscale on the direction y.
3. the rail deformation detection method according to claim 1 based on infrared thermal imaging and computer vision, feature It is, high-definition camera image data is carried out after once dividing using darker gray threshold in track groove described in step S3, It is split again using brighter gray threshold outside track groove, specific calculation formula are as follows:
Wherein, f (x, y) indicates pretreated gray level image, TLIndicate darker area minimum gray value in slot, THIt indicates outside slot Brighter areas gray scale maximum value,Respectively indicate TLAnd THThe gray level model fluctuated up and down It encloses；
The adjacent distance feature of the brighter darker area of foundation divides orbital region, and extraction obtains orbital image, specific steps Are as follows:
To darker area binary map g in slotL (x, y)With brighter areas binary map g outside slotH (x, y)It is expanded to obtain corresponding region point Cut binary mapWithAnd intersection is solved, obtain orbital region segmentation binary map gu(x, y) is indicated are as follows:
In track region segmentation binary map guThe starting pixels point of two siding tracks is determined on (x, y), tracking obtains multiple on track Pixel obtains a plurality of trajectory line, and the trajectory line of two siding tracks is extracted from a plurality of trajectory line, using least square be segmented into Row quadratic fit constructs orbit equation, the orbital image extracted.
4. the rail deformation detection method according to claim 1 based on infrared thermal imaging and computer vision, feature It is, relative temperature difference method described in the detection of step S4 infrared temperature, specific steps are as follows:
S41: temperature value on the display screen of infrared thermal imager is read；
S42: by the processing of infrared thermal imaging figure gray processing, the information matrix of brightness value is obtained；
S43: temperature value and gray value setting mapping relations are expressed as:
Wherein, G indicates that gray value, T indicate temperature；
S44: temperature value when working normally to the rail temperature result detected with track compares, and is obtained using curve matching To rail temperature variation tendency, according to the mapping relations of temperature and gray scale, temperature value is replaced with gray value, obtains gray threshold；
S45: being split fault zone by gray threshold, sets different gray values, and extract using edge detection More than the fault zone of gray threshold, the pixel counts in fault zone are obtained into region area, the size for comparing gray value obtains To maximum gradation value pixel, maximum temperature point is obtained.
5. the rail deformation detection method according to claim 1 based on infrared thermal imaging and computer vision, feature It is, is overlapped pretreated image with obtained orbital image is extracted described in step S5, Superposition Formula are as follows:
S (i, j)=R (i, j) &ROI；
Wherein, R (i, j) indicates that pretreated image, ROI indicate that area-of-interest, S (i, j) indicate operation result image；
It is described that connected region is screened, screen formula are as follows:
DArea≥S；
DHeight≥D∩DWidth≥Dlow∩DWidth≤Dhigh；
Wherein, DAreaRepresent the number of pixel shared by connected region, DHeightThe height of the outer area-encasing rectangle of connected region is represented, DWidthRepresent the width of the outer area-encasing rectangle of connected region, S, D, Dlow、Dhigh、DRatioRespectively represent doubtful track foreign matter area, Diagonal line length, the rectangular degree of the length of external minimum rectangle, wide, the external minimum rectangle of external minimum rectangle；
When screening formula while setting up, the connected region after screening is doubtful track foreign matter.
6. the rail deformation detection method according to claim 1 based on infrared thermal imaging and computer vision, feature It is, the training step of the BP neural network are as follows:
S60: numerical value initialization: setting BP neural network input layer number n, hidden node number l and output layer number m, If the weight of input layer to hidden layer is ωij, hidden layer to output layer weight be ωjk, input layer to hidden layer threshold value be aj, hidden layer to output layer threshold value be bk, learning rate is η and excitation function is g (x), the excitation function g (x) uses Sigmoid function indicates are as follows:
Wherein x is input matrix；
S61: input training sample: using the orbital image of high-definition camera shooting as original image, it includes to be identified different for obtaining The image pattern of object carries out image gray processing and binary conversion treatment, obtains the binary image of sample, and the sample of acquisition is unified To the size of same ratio, it is input in BP neural network；
S62: whether training of judgement sample, which is loaded into, finishes, and finishes if being loaded into, and performs the next step suddenly, finishes if being not loaded with, execute step Rapid S61；
S63: the output of hidden layer is set as Hj, calculate the output of hidden layer neuron:
Wherein, n is input layer number, ωijFor the weight of input layer to hidden layer, xiFor input matrix, ajIt is arrived for input layer The threshold value of hidden layer；
S64: the output of output layer is set as Ok, calculate the output of output layer neuron:
Wherein, l is hidden layer node number, ωjkFor the weight of hidden layer to output layer, bkFor the threshold of hidden layer to output layer Value；
S65: error is calculated:
Wherein, ekFor error, m is output layer node number, YkFor desired output, OkFor the output of output layer；
S66: right value update:
ωjk=ωjk+ηHjek
Wherein, ωijFor the weight of input layer to hidden layer, ωjkFor the weight of hidden layer to output layer, η is learning rate, HjFor The output of hidden layer, xiFor input matrix, m is output layer node number, ekFor error；
S67: threshold value updates:
bk=bk+ηek；
Wherein, ajFor the threshold value of input layer to hidden layer, bkFor the threshold value of hidden layer to output layer, ωjkFor hidden layer to output layer Weight, η is learning rate, HjFor the output of hidden layer, xiFor input matrix, m is output layer node number, ekFor error；
S68: judging whether the difference between adjacent error twice is less than setting value, if being less than setting value, BP neural network training Terminate, if being not less than setting value, recycles and execute step S63-S67.
7. a kind of rail deformation detection system based on infrared thermal imaging and computer vision characterized by comprising unmanned plane And earth station, the unmanned plane include main control module, flight control modules, navigation module, wireless communication module and module of taking photo by plane, Main control module control navigation module, the wireless communication module and module of taking photo by plane, the flight control modules are for controlling nobody The state of flight of machine, the navigation module are used to provide navigation, the wireless communication module to unmanned plane for unmanned plane and ground Face station communicates wirelessly, and for the module of taking photo by plane for obtaining orbital image, the module of taking photo by plane includes high-definition camera and red Outer thermal imaging system；
The earth station includes image pre-processing module, orbital image extraction module, extracts high-temperature region module, doubtful track foreign matter Screening module and track foreign matter identification module, described image preprocessing module for being located high-definition camera image data in advance Reason, orbital image extraction module extract orbital image, extraction high-temperature region module for carrying out multi-threshold orbital region segmentation For carrying out relative temperature difference judgement, high-temperature area and maximum temperature point in Infrared Thermogram, the doubtful track foreign matter are extracted Screening module obtains doubtful track foreign matter, the screening connected region is that area-of-interest carries out side for screening connected region It is obtained after the judgement of edge closure and filling, the track foreign matter identification module is equipped with BP neural network, is used for doubtful track foreign matter Input BP neural network is identified to obtain foreign matter classification results.
8. the rail deformation detection system according to claim 7 based on infrared thermal imaging and computer vision, feature It is, the BP neural network is equipped with input layer, hidden layer and output layer, and the input node quantity of the input layer is set as 20 A, the output node quantity of the output layer is set as 3, and the hidden layer node quantity of the hidden layer is set as:
Wherein, n is hidden layer node quantity, niFor input node number, noFor output node number, a is taken in [1,10] range Constant.
CN201910509281.6A 2019-06-13 2019-06-13 Rail fault detection method and system based on infrared thermal imaging and computer vision Active CN110261436B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910509281.6A CN110261436B (en) 2019-06-13 2019-06-13 Rail fault detection method and system based on infrared thermal imaging and computer vision

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910509281.6A CN110261436B (en) 2019-06-13 2019-06-13 Rail fault detection method and system based on infrared thermal imaging and computer vision

Publications (2)

Publication Number Publication Date
CN110261436A true CN110261436A (en) 2019-09-20
CN110261436B CN110261436B (en) 2022-03-22

Family

ID=67917947

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910509281.6A Active CN110261436B (en) 2019-06-13 2019-06-13 Rail fault detection method and system based on infrared thermal imaging and computer vision

Country Status (1)

Country Link
CN (1) CN110261436B (en)

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110991377A (en) * 2019-12-11 2020-04-10 辽宁工业大学 Monocular visual neural network-based front target identification method for automobile safety auxiliary system
CN111223097A (en) * 2020-03-13 2020-06-02 中冶长天国际工程有限责任公司 Method and system for detecting degree of blockage caused by grate bar pasting of sintering machine
CN111256841A (en) * 2020-03-26 2020-06-09 深圳市永达电子信息股份有限公司 Track state detection method and detection system thereof
CN111260906A (en) * 2020-03-18 2020-06-09 云境商务智能研究院南京有限公司 Intelligent agricultural system based on embedded mode
CN111476759A (en) * 2020-03-13 2020-07-31 深圳市鑫信腾机器人科技有限公司 Screen surface detection method and device, terminal and storage medium
CN111488868A (en) * 2020-03-27 2020-08-04 贵州电网有限责任公司 High-temperature area identification method and system based on transformer infrared image
CN111751024A (en) * 2020-06-09 2020-10-09 江苏雷威建设工程有限公司 Intelligent track temperature control method, system and device and storage medium thereof
CN111967526A (en) * 2020-08-20 2020-11-20 东北大学秦皇岛分校 Remote sensing image change detection method and system based on edge mapping and deep learning
CN111986168A (en) * 2020-08-07 2020-11-24 中国农业大学 Method and system for detecting binaural root and optimal ear root temperature measurement frame of T-shaped branch
CN112001963A (en) * 2020-07-31 2020-11-27 浙江大华技术股份有限公司 Fire fighting channel investigation method, system and computer equipment
CN112051298A (en) * 2020-09-09 2020-12-08 飞础科智慧科技(上海)有限公司 Steel ladle surface fault diagnosis method and equipment
CN112115776A (en) * 2020-08-07 2020-12-22 长江三峡通航管理局 Method for monitoring lubrication state of large open gear of ship lock hoist
CN112255141A (en) * 2020-10-26 2021-01-22 光谷技术股份公司 Thermal imaging gas monitoring system
CN112927223A (en) * 2021-03-29 2021-06-08 南通大学 Glass curtain wall detection method based on infrared thermal imager
CN112947588A (en) * 2021-03-01 2021-06-11 中国南方电网有限责任公司超高压输电公司贵阳局 Unmanned aerial vehicle electric wire netting patrols line system
CN113110577A (en) * 2021-04-15 2021-07-13 中国南方电网有限责任公司超高压输电公司柳州局 Unmanned aerial vehicle flight route planning management system is patrolled and examined to electric wire netting
CN113591251A (en) * 2021-08-17 2021-11-02 中冶北方（大连）工程技术有限公司 Equipment fault temperature analysis and diagnosis method
CN113640308A (en) * 2021-08-31 2021-11-12 郑州铁路职业技术学院 Track abnormity monitoring system based on machine vision
CN113673614A (en) * 2021-08-25 2021-11-19 北京航空航天大学 Machine vision-based subway tunnel foreign matter intrusion detection device and method
CN114019879A (en) * 2021-11-08 2022-02-08 河南数字中原数据有限公司 High-low voltage power distribution cabinet monitoring system
CN114383735A (en) * 2021-12-17 2022-04-22 暨南大学 Thermal power generating unit air cooling array temperature field monitoring method and device based on machine vision
CN114670899A (en) * 2022-04-20 2022-06-28 北京运达华开科技有限公司 Image acquisition device for track detection system
CN114937040A (en) * 2022-07-22 2022-08-23 珠海优特电力科技股份有限公司 Train inspection method, device and system for rail transit vehicle section and storage medium
CN115188091A (en) * 2022-07-13 2022-10-14 国网江苏省电力有限公司泰州供电分公司 Unmanned aerial vehicle grid inspection system and method integrating power transmission and transformation equipment
CN115460895A (en) * 2022-11-10 2022-12-09 武汉至驱动力科技有限责任公司 Electronic water pump controller heat dissipation method based on temperature field image information
CN116468729A (en) * 2023-06-20 2023-07-21 南昌江铃华翔汽车零部件有限公司 Automobile chassis foreign matter detection method, system and computer
CN116630304A (en) * 2023-07-18 2023-08-22 东莞市京品精密模具有限公司 Lithium battery mold processing detection method and system based on artificial intelligence
CN116721095A (en) * 2023-08-04 2023-09-08 杭州瑞琦信息技术有限公司 Aerial photographing road illumination fault detection method and device
CN111626104B (en) * 2020-04-13 2023-09-08 国网上海市电力公司 Cable hidden trouble point detection method and device based on unmanned aerial vehicle infrared thermal image
CN116758085A (en) * 2023-08-21 2023-09-15 山东昆仲信息科技有限公司 Visual auxiliary detection method for infrared image of gas pollution
CN117341781A (en) * 2023-12-04 2024-01-05 深圳市鼎善信息科技有限公司 Rail transit fault processing method and device, computer equipment and storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0394932B1 (en) * 1989-04-24 1998-03-04 Siemens Aktiengesellschaft Photothermal inspection method, arrangement for its working out, and utilisation of the method
CN102305664A (en) * 2011-05-19 2012-01-04 中国农业大学 Thermal imaging temperature measurement and fault location inspection system
CN102508110A (en) * 2011-10-10 2012-06-20 上海大学 Texture-based insulator fault diagnostic method
CN102608162A (en) * 2012-02-20 2012-07-25 首都师范大学 Threshold segmentation method for ultrasonic infrared thermograph
CN103308521A (en) * 2012-08-29 2013-09-18 中国人民解放军第二炮兵工程大学 Method for enhancing infrared thermal wave detection image defect contrast
CN206132684U (en) * 2016-10-21 2017-04-26 上海工程技术大学 Subway contact net node internal defect detection device
CN108932721A (en) * 2018-06-28 2018-12-04 上海电力学院 A kind of infrared Image Segmentation and fusion method for crusing robot

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0394932B1 (en) * 1989-04-24 1998-03-04 Siemens Aktiengesellschaft Photothermal inspection method, arrangement for its working out, and utilisation of the method
CN102305664A (en) * 2011-05-19 2012-01-04 中国农业大学 Thermal imaging temperature measurement and fault location inspection system
CN102508110A (en) * 2011-10-10 2012-06-20 上海大学 Texture-based insulator fault diagnostic method
CN102608162A (en) * 2012-02-20 2012-07-25 首都师范大学 Threshold segmentation method for ultrasonic infrared thermograph
CN103308521A (en) * 2012-08-29 2013-09-18 中国人民解放军第二炮兵工程大学 Method for enhancing infrared thermal wave detection image defect contrast
CN206132684U (en) * 2016-10-21 2017-04-26 上海工程技术大学 Subway contact net node internal defect detection device
CN108932721A (en) * 2018-06-28 2018-12-04 上海电力学院 A kind of infrared Image Segmentation and fusion method for crusing robot

Cited By (51)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110991377A (en) * 2019-12-11 2020-04-10 辽宁工业大学 Monocular visual neural network-based front target identification method for automobile safety auxiliary system
CN110991377B (en) * 2019-12-11 2023-09-19 辽宁工业大学 Front mesh identification method of automobile safety auxiliary system based on monocular vision neural network
CN111476759A (en) * 2020-03-13 2020-07-31 深圳市鑫信腾机器人科技有限公司 Screen surface detection method and device, terminal and storage medium
CN111223097A (en) * 2020-03-13 2020-06-02 中冶长天国际工程有限责任公司 Method and system for detecting degree of blockage caused by grate bar pasting of sintering machine
CN111223097B (en) * 2020-03-13 2024-04-19 中冶长天国际工程有限责任公司 Trolley grate bar paste blocking degree detection method and system of sintering machine
CN111260906A (en) * 2020-03-18 2020-06-09 云境商务智能研究院南京有限公司 Intelligent agricultural system based on embedded mode
CN111256841A (en) * 2020-03-26 2020-06-09 深圳市永达电子信息股份有限公司 Track state detection method and detection system thereof
CN111488868B (en) * 2020-03-27 2023-06-30 贵州电网有限责任公司 High-temperature area identification method and system based on transformer infrared image
CN111488868A (en) * 2020-03-27 2020-08-04 贵州电网有限责任公司 High-temperature area identification method and system based on transformer infrared image
CN111626104B (en) * 2020-04-13 2023-09-08 国网上海市电力公司 Cable hidden trouble point detection method and device based on unmanned aerial vehicle infrared thermal image
CN111751024A (en) * 2020-06-09 2020-10-09 江苏雷威建设工程有限公司 Intelligent track temperature control method, system and device and storage medium thereof
CN112001963A (en) * 2020-07-31 2020-11-27 浙江大华技术股份有限公司 Fire fighting channel investigation method, system and computer equipment
CN112115776B (en) * 2020-08-07 2024-03-12 长江三峡通航管理局 Method for monitoring lubrication state of large open gear of ship lock hoist
CN112115776A (en) * 2020-08-07 2020-12-22 长江三峡通航管理局 Method for monitoring lubrication state of large open gear of ship lock hoist
CN111986168B (en) * 2020-08-07 2024-03-15 中国农业大学 T-branch binaural root detection and optimal auricular root temperature measurement frame detection method and system
CN111986168A (en) * 2020-08-07 2020-11-24 中国农业大学 Method and system for detecting binaural root and optimal ear root temperature measurement frame of T-shaped branch
CN111967526B (en) * 2020-08-20 2023-09-22 东北大学秦皇岛分校 Remote sensing image change detection method and system based on edge mapping and deep learning
CN111967526A (en) * 2020-08-20 2020-11-20 东北大学秦皇岛分校 Remote sensing image change detection method and system based on edge mapping and deep learning
CN112051298A (en) * 2020-09-09 2020-12-08 飞础科智慧科技(上海)有限公司 Steel ladle surface fault diagnosis method and equipment
CN112255141A (en) * 2020-10-26 2021-01-22 光谷技术股份公司 Thermal imaging gas monitoring system
CN112255141B (en) * 2020-10-26 2021-05-11 光谷技术有限公司 Thermal imaging gas monitoring system
CN112947588A (en) * 2021-03-01 2021-06-11 中国南方电网有限责任公司超高压输电公司贵阳局 Unmanned aerial vehicle electric wire netting patrols line system
CN112927223A (en) * 2021-03-29 2021-06-08 南通大学 Glass curtain wall detection method based on infrared thermal imager
CN113110577A (en) * 2021-04-15 2021-07-13 中国南方电网有限责任公司超高压输电公司柳州局 Unmanned aerial vehicle flight route planning management system is patrolled and examined to electric wire netting
CN113591251A (en) * 2021-08-17 2021-11-02 中冶北方（大连）工程技术有限公司 Equipment fault temperature analysis and diagnosis method
CN113591251B (en) * 2021-08-17 2024-04-23 中冶北方（大连）工程技术有限公司 Equipment fault temperature analysis and diagnosis method
CN113673614A (en) * 2021-08-25 2021-11-19 北京航空航天大学 Machine vision-based subway tunnel foreign matter intrusion detection device and method
CN113673614B (en) * 2021-08-25 2023-12-12 北京航空航天大学 Metro tunnel foreign matter intrusion detection device and method based on machine vision
CN113640308A (en) * 2021-08-31 2021-11-12 郑州铁路职业技术学院 Track abnormity monitoring system based on machine vision
CN113640308B (en) * 2021-08-31 2024-03-29 夏冰心 Rail anomaly monitoring system based on machine vision
CN114019879A (en) * 2021-11-08 2022-02-08 河南数字中原数据有限公司 High-low voltage power distribution cabinet monitoring system
CN114019879B (en) * 2021-11-08 2024-03-15 河南数字中原数据有限公司 High-low voltage power distribution cabinet monitoring system
CN114383735B (en) * 2021-12-17 2024-03-26 暨南大学 Thermal power generating unit air cooling array temperature field monitoring method and device based on machine vision
CN114383735A (en) * 2021-12-17 2022-04-22 暨南大学 Thermal power generating unit air cooling array temperature field monitoring method and device based on machine vision
CN114670899A (en) * 2022-04-20 2022-06-28 北京运达华开科技有限公司 Image acquisition device for track detection system
CN115188091A (en) * 2022-07-13 2022-10-14 国网江苏省电力有限公司泰州供电分公司 Unmanned aerial vehicle grid inspection system and method integrating power transmission and transformation equipment
CN115188091B (en) * 2022-07-13 2023-10-13 国网江苏省电力有限公司泰州供电分公司 Unmanned aerial vehicle gridding inspection system and method integrating power transmission and transformation equipment
CN114937040B (en) * 2022-07-22 2022-11-18 珠海优特电力科技股份有限公司 Train inspection method, device and system for rail transit vehicle section and storage medium
CN114937040A (en) * 2022-07-22 2022-08-23 珠海优特电力科技股份有限公司 Train inspection method, device and system for rail transit vehicle section and storage medium
CN115460895B (en) * 2022-11-10 2023-02-17 武汉至驱动力科技有限责任公司 Electronic water pump controller heat dissipation method based on temperature field image information
CN115460895A (en) * 2022-11-10 2022-12-09 武汉至驱动力科技有限责任公司 Electronic water pump controller heat dissipation method based on temperature field image information
CN116468729B (en) * 2023-06-20 2023-09-12 南昌江铃华翔汽车零部件有限公司 Automobile chassis foreign matter detection method, system and computer
CN116468729A (en) * 2023-06-20 2023-07-21 南昌江铃华翔汽车零部件有限公司 Automobile chassis foreign matter detection method, system and computer
CN116630304B (en) * 2023-07-18 2023-09-19 东莞市京品精密模具有限公司 Lithium battery mold processing detection method and system based on artificial intelligence
CN116630304A (en) * 2023-07-18 2023-08-22 东莞市京品精密模具有限公司 Lithium battery mold processing detection method and system based on artificial intelligence
CN116721095B (en) * 2023-08-04 2023-11-03 杭州瑞琦信息技术有限公司 Aerial photographing road illumination fault detection method and device
CN116721095A (en) * 2023-08-04 2023-09-08 杭州瑞琦信息技术有限公司 Aerial photographing road illumination fault detection method and device
CN116758085B (en) * 2023-08-21 2023-11-03 山东昆仲信息科技有限公司 Visual auxiliary detection method for infrared image of gas pollution
CN116758085A (en) * 2023-08-21 2023-09-15 山东昆仲信息科技有限公司 Visual auxiliary detection method for infrared image of gas pollution
CN117341781A (en) * 2023-12-04 2024-01-05 深圳市鼎善信息科技有限公司 Rail transit fault processing method and device, computer equipment and storage medium
CN117341781B (en) * 2023-12-04 2024-03-22 深圳市鼎善信息科技有限公司 Rail transit fault processing method and device, computer equipment and storage medium

Also Published As

Publication number Publication date
CN110261436B (en) 2022-03-22

Similar Documents

Publication Publication Date Title
CN110261436A (en) Rail deformation detection method and system based on infrared thermal imaging and computer vision
CN110532889B (en) Track foreign matter detection method based on rotor unmanned aerial vehicle and YOLOv3
CN105550665B (en) A kind of pilotless automobile based on binocular vision can lead to method for detecting area
CN104392212B (en) The road information detection and front vehicles recognition methods of a kind of view-based access control model
CN109670376B (en) Lane line identification method and system
CN105512720B (en) A kind of public transit vehicle passenger flow statistics method and system
CN106652465A (en) Method and system for identifying abnormal driving behavior on road
CN108596129A (en) A kind of vehicle based on intelligent video analysis technology gets over line detecting method
CN106951879A (en) Multiple features fusion vehicle checking method based on camera and millimetre-wave radar
CN107031623A (en) A kind of road method for early warning based on vehicle-mounted blind area camera
DE102009050492A1 (en) Travel's clear path detection method for motor vehicle i.e. car, involves monitoring images, each comprising set of pixels, utilizing texture-less processing scheme to analyze images, and determining clear path based on clear surface
CN107609522A (en) A kind of information fusion vehicle detecting system based on laser radar and machine vision
CN105488454A (en) Monocular vision based front vehicle detection and ranging method
CN105373135A (en) Method and system for guiding airplane docking and identifying airplane type based on machine vision
CN102902974B (en) Image based method for identifying railway overhead-contact system bolt support identifying information
CN107491720A (en) A kind of model recognizing method based on modified convolutional neural networks
CN106022243B (en) A kind of retrograde recognition methods of the car lane vehicle based on image procossing
DE102009050504A1 (en) Clear path of travel detecting method for motor vehicle i.e. car, involves combining clear path of travel and determined flat surface to describe enhanced clear path of travel, and utilizing enhanced clear path of travel to navigate vehicle
CN109766746A (en) A kind of track foreign matter detecting method of unmanned plane video
CN109190483B (en) Lane line detection method based on vision
DE102009048892A1 (en) Clear traveling path detecting method for vehicle e.g. car, involves generating three-dimensional map of features in view based upon preferential set of matched pairs, and determining clear traveling path based upon features
DE102009048699A1 (en) Travel's clear path detection method for motor vehicle i.e. car, involves monitoring images, each comprising set of pixels, utilizing texture-less processing scheme to analyze images, and determining clear path based on clear surface
CN107066968A (en) The vehicle-mounted pedestrian detection method of convergence strategy based on target recognition and tracking
CN104778444A (en) Method for analyzing apparent characteristic of vehicle image in road scene
DE102013205952A1 (en) Reconfigurable system for detection of a clear path

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