CN102708378B - Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic - Google Patents

Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic Download PDF

Info

Publication number
CN102708378B
CN102708378B CN201210133557.3A CN201210133557A CN102708378B CN 102708378 B CN102708378 B CN 102708378B CN 201210133557 A CN201210133557 A CN 201210133557A CN 102708378 B CN102708378 B CN 102708378B
Authority
CN
China
Prior art keywords
image
row
abnormal
value
white light
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
Application number
CN201210133557.3A
Other languages
Chinese (zh)
Other versions
CN102708378A (en
Inventor
高飞
张元鸣
肖刚
韩政高
袁晓阳
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.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN201210133557.3A priority Critical patent/CN102708378B/en
Publication of CN102708378A publication Critical patent/CN102708378A/en
Application granted granted Critical
Publication of CN102708378B publication Critical patent/CN102708378B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The invention discloses a method for diagnosing the fault of intelligent traffic capturing equipment based on an image abnormal characteristic. An aim of diagnosing the fault of the capturing equipment is fulfilled by a method for intelligently identifying an abnormal image by a computer, so that the complicated and low-efficiency manual diagnosis method is avoided. The method comprises the following steps of: firstly, establishing a mapping relation from the abnormal image to the fault of the equipment, and implementing a high-practicability license plate positioning method based on a color characteristic and a character texture characteristic according to the requirements of normal characteristics; secondly, adopting an abnormal image identification method based on multi-characteristic combination, and researching the adaptability of an identification algorithm; and finally, performing high-observability visualization processing on fault information of complicated comprehensive equipment. According to an equipment fault diagnosis system, fault diagnosis for the capturing equipment can meet requirements on high instantaneity, high accuracy and high efficiency; and a humanized and scientific fault diagnosis method is supplied.

Description

The method of equipment fault diagnosis is captured in a kind of intelligent transportation based on image abnormity feature
Technical field
The present invention relates to a kind of method of capturing equipment fault diagnosis.
Background technology
Intelligent transportation refer to one based on hyundai electronics infotech the service system towards communications and transportation.In the field of numerous intelligent transportation application, all need to obtain vehicle image or other transport information images by candid photograph equipment, for example: high way super speed detection system, crossing red light running supervisory system, Intelligent charging system used in parking lot etc.
It is all in outside work that the equipment overwhelming majority is captured in intelligent transportation, will be subject to its test complicated and protean environmental factor; Meanwhile, equipment frequency of operation is large, and especially, in the situation that road traffic is very large, equipment is almost the duty in a kind of overload; In addition, the impact of the factors such as restriction and equipment quality in equipment serviceable life itself, causes capturing equipment failure and takes place frequently.If can not, in time to capturing diagnosing malfunction the maintenance of equipment, will badly influence fairness, science and the high efficiency of intelligent transportation system.Whether can candid photograph equipment normally work and be directly connected to intelligent transportation system and can normally work, so the fault diagnosis that intelligent transportation is captured to equipment has very important significance.
Existing intelligent transportation candid photograph equipment fault diagnosis mainly rests on the mode of artificial treatment, and it is divided into two kinds: the first is: manually investigate mode, arrange patrol officer to investigate one by one the failure condition of every complete equipment to outdoor candid photograph equipment working site.Another is: manually seek the mode of looking into, abnormal view data is found and found to the center image lane database obtaining at candid photograph equipment, then arrives the fault of concrete a certain equipment according to abnormal image data correlation.
First kind of way obviously wastes time and energy.First, candid photograph equipment is mounted on the portal frame of certain altitude, makes troubles to investigation work; Secondly, when investigation work, being subject to various boisterous impacts, is a kind of challenge of difficulty to staff; Finally, on-the-spot investigation equipment failure situation may affect the normal operation of traffic.The second way has improved work efficiency comparatively speaking, but is subject to its asthenopic impact of people of bringing that work long hours down, can cause the situation of undetected and flase drop to occur.These two kinds of modes are high especially to people's dependence, it is one compensatory diagnostic form afterwards, cause the phenomenon of the retardance of its fault diagnosis report to occur, it from job specification, is also a kind of thing extremely dry as dust, under the ITS Information epoch, the fault diagnosis of its equipment also needs to possess intelligent, realizes informationization, liberation labour.The efficiency of these two kinds of methods and scientific or cannot reach and capture that equipment fault diagnosis needs in time, requirement accurately on the whole.
Domestic many scholars have done a lot of research to the equipment fault diagnosis based on image processing, and its main research concentrates on the fault diagnosis of single equipment or a certain part.Xia Yong (engineering college of the Second Artillery Force, 2001) Fault Diagnosis of Internal Combustion Engine with neural network based on image processing has been proposed, utilize engine cylinder head system vibration signal to produce image, in its image, extract characteristics of image and carry out Fuzzy processing, realized the fault diagnosis to valve train with this.Cai Yanping (engineering college of the Second Artillery Force,) proposed a kind ofly to cut apart and the Diagnosis Method of Diesel Fault of Fuzzy Pattern Recognition based on video spectrogram, image, propose vibration signal image geometry feature and parameters for shape characteristic, can correctly diagnose valve fault.Zhang Yong (National University of Defense Technology, 2006) has proposed Fault Diagnosis of Aeroengines mesopore and has visited image feature extraction techniques applied research, utilizes hole spy image feature extraction techniques to rule out accurately the various faults of engine critical component.Wang Peizhen (Anhui University of Technology, 2010) has proposed the photovoltaic array fault analysis based on infrared image.
In sum; utilize imaging device to generate the fault picture of equipment, in addition various image processing techniquess are carried out signature analysis and feature extraction to it and are reached image processing method with this and in equipment fault diagnosis, have support and do not shut down the advantage that diagnosis, noncontact diagnosis, non-dismounting diagnosis, on-line intelligence are efficiently diagnosed.Application in a lot of fields will get more and more, especially some specific area, and the method can fully guarantee the safety of testing personnel and equipment.
The domestic fault diagnosis research about intelligent transportation candid photograph equipment or other equipment is less.Jiang Yongchi has proposed the electronic police crossing equipment guarantee design based on temporal characteristics in 2011, receive that according to electronic police system processing enter the length of the direction industrial computer data time at a certain crossing judges the fault of industrial computer and router.This kind of method proposed a kind of fault diagnosis settling mode of intelligent transportation candid photograph equipment of unartificial processing, principle is comparatively simple, it is mainly the fault diagnosis to industrial computer and router, do not relate to the fault diagnosis of video camera, inductive coil and these nucleus equipments of flashlamp in candid photograph equipment, but proposed a kind of new approaches that solve intelligent transportation candid photograph equipment fault diagnosis to people: the mode of utilizing computing machine processing.Meanwhile, the application of the various image processing techniquess of introducing before being referred from equipment fault diagnosis, we also can utilize the method for image processing to go to solve the fault diagnosis of candid photograph equipment, meanwhile, can guarantee the comprehensive accomplished of this fault diagnosis.
Summary of the invention
Capture equipment fault diagnosis for current intelligent transportation and depend on dehumanization and inefficient artificial treatment mode, the problem that equipment fault diagnosis is intended to science, solves accurately and efficiently candid photograph equipment fault diagnosis difficulty is captured in the intelligent transportation that the present invention is based on image abnormity feature.
The technical solution adopted for the present invention to solve the technical problems is:
1), set up the mapping relations of capturing between equipment failure and abnormal image a method of capturing equipment fault diagnosis based on the intelligent transportation of image abnormity feature, comprises the following steps:;
Can the accessed vehicle image of candid photograph equipment be divided into two classes according to correctly obtain our needed effective information in image: normal picture and abnormal image.Take high way super speed detection system as example, in the time of monitoring place of vehicle process hypervelocity, when recording the real-time speed of vehicle, this system also need obtain by candid photograph equipment the realtime graphic of vehicle, in this image, the information that we need is exactly complete license board information, the equipment of namely capturing need successfully capture the vehicle image that comprises clear car plate give computing machine next step carry out car plate identification work, we are defined as normal picture this image.But in the time that candid photograph equipment breaks down, in its image obtaining, license board information is lost clearly, we are defined as abnormal image the image that cannot see clear license board information.Abnormal image can be according to whether having captured vehicle and license board information can be divided into again three kinds of situations: but have vehicle car plate unintelligible, have vehicle but without car plate, without vehicle.
The classification of carrying out according to vehicle and license board information can only remove to hold abnormal image on the whole, in the process of the actual identification of computing machine abnormal image, we also need the abnormal characteristic of abnormal image to carry out trickleer division, and the abnormal characteristic of image is divided into: color exception, abnormal behavior and other abnormal three classes.
2), the car plate position fixing process based on color characteristic and textural characteristics:
First can be according to whether having complete license board information to be divided into two classes at the accessed all images of candid photograph equipment: have license plate image and without license plate image, and the abnormal image overwhelming majority belongs to without license plate image; Secondly license board information can distinguish two kinds similar abnormal, car plate as dark in figure kine bias is visible and the dark car plate of figure kine bias is invisible, the invisible explanation flashlamp of the dark car plate of figure kine bias quits work completely, this situation needs preferentially timely its fault to be processed, and the dark car plate of figure kine bias is to be caused by exposure of flash lamp quantity not sufficient as seen; The last the most significant feature of certain image abnormity is exactly without car plate;
Judge in image whether have effective license board information according to whether can successfully navigate to car plate in image, can navigate in car plate key diagram picture and have license board information, otherwise there is no license board information; .
The feature of car plate mainly concentrates on color and above character, the color characteristic of car plate has four kinds of wrongly written or mispronounced character of the blue end, yellow end surplus, white gravoply, with black engraved characters and black matrix wrongly written or mispronounced characters, character is that the additional cut-point symbol of seven character group being made up of Chinese character, English alphabet and arabic numeral respectively forms, and the interval between character has unified standard.
Car plate location based on color characteristic and textural characteristics is mainly the candidate region that utilizes these four kinds of color characteristics of car plate to carry out Primary Location to obtain car plate, thereby then in its binary image, carrying out secondary location according to the fixed intervals feature between gray scale jumping characteristic and character between characters on license plate obtains the pinpoint result of car plate;
3), the abnormal image recognition methods that combines based on many features, process is:
3.1) identification colors abnormal image
The color exception characteristic of image is mainly reflected in the half-tone information in statistical law and the gray level image of R, G in former RGB coloured image, tri-passages of B;
3.2) identification abnormal behavior image
Abnormal behavior image is because video camera candid photograph abnormal behavior causes, the target vehicle comprising in the present image of property list of this image abnormal, comprises lacking in image in target vehicle, image, having comprised the three kinds of situations of target vehicle malposition in abnormal target vehicle, image.Analyze thisly when abnormal, mainly find and best embody this abnormal rule according to zoning method;
3.3) identify other abnormal images
The unique point that other off-notes of image mainly comprise from size and the image of image shows, and analyzes this abnormal characteristic and mainly adopt the method for feature extraction.
Further, described step 3) in, the adaptivity of recognizer refers to that algorithm can adjust the order of the various abnormal images of identification in time, guarantees the lasting high efficiency of algorithm.At this, we are divided into internal cause and external cause the factor that affects efficiency of algorithm.
Further again, the method for described fault diagnosis also comprises: 4) resultant fault information is visual:
Along with the operation of the fault diagnosis system a period of time in this invention, can accumulate a large amount of device fault information.How to unify and remove easily to express complicated and diversified a large amount of failure message, need to do a visualization processing to comprehensive device fault information, provide the visualization result of a human to be used for carrying out decision-making to user.
Resultant fault information visual can be respectively from region, time, three aspects of failure rate decides.
A Fault Diagnosis of Mechanical Equipment is captured in intelligent transportation based on image abnormity feature, and this system is mainly made up of configuration module, identification module, data analysis module.
Technical conceive of the present invention is: a kind of mode of utilizing computer intelligence to identify efficiently abnormal image to reach equipment fault diagnosis.This method is the thorough change of the second in artificial treatment mode manually being sought to the mode of looking into, manually seeking the mode of looking into is to rely on manpower to remove to distinguish and extract abnormal image, the technical solution adopted in the present invention substitutes people with computing machine just and removes to identify abnormal image, solve all not enough problem in artificial treatment mode with this, also can guarantee efficiency in failure diagnostic process and the accuracy of diagnostic result simultaneously.Manually seeking the mode of looking into is rule of thumb to judge whether image exists extremely, and computing machine identification is the one application of digital image processing techniques, the namely abnormal characteristic of analysis image, and utilize image processing method to this feature extraction and identification.To process application in equipment fault diagnosis different from traditional image: the disposal route of the various forms image of above-mentioned introduction solves the troubleshooting issue of equipment all need to be by the fault picture of imaging device forming apparatus in next life, extract be the multiple microscopic features such as image middle distance, size, area, solution be the troubleshooting issue of a certain equipment.And intelligent transportation capture equipment itself be a set of imaging device, the view data that we can directly utilize candid photograph equipment to get is carried out fault diagnosis, solution be the troubleshooting issue of plurality of devices.The method is to get on to identify the abnormal image in all view data that candid photograph equipment obtains in macroscopic aspect.In the time that breaking down, candid photograph equipment can cause its image taking to occur abnormal, namely the fault of equipment is to be indirectly reflected among the abnormal characteristic of image, therefore, in candid photograph equipment center image data base, identify abnormal image and just can find corresponding equipment failure.Artificial treatment mode relies on people's experience and understanding to see that it is which equipment has been out of order that abnormal image just can be judged, and to make computing machine reach the object of " knowledge figure knows fault ", need to set up a mapping relations net covering between existing all candid photograph equipment failure type and abnormal image type.Candid photograph equipment is to capture vehicle image, is in order to obtain the important information in vehicle image: license board information.In the time that candid photograph equipment breaks down, in the abnormal image obtaining, license board information is a very important basis of characterization, thus, in identification abnormal image process, need to process accordingly license board information.What the diversity of equipment failure was brought is the diversity of image abnormity characteristic, in the image abnormity characteristic procedure of identification non-singularity, need to take diverse ways and process.Here said process refers to gray scale processing procedure or the threshold process process in image processing, image processing method in most of other equipment fault diagnosiss normally carries out rim detection, morphologic region growing and feature extraction etc. in its corresponding bianry image, and image abnormity feature in candid photograph equipment fault diagnosis shows respectively in the original color image of image or in gray level image and bianry image, so for the processing procedure of this non-uniformity, the image abnormity characteristic of identification complicated variety need to be in conjunction with several different methods.Efficientibility in order to guarantee that algorithm is lasting, need to do corresponding adjustment to the recognition sequence of algorithm simultaneously, namely studies a kind of realization of algorithm adaptivity.Finally, for the equipment fault diagnosis result of a, hommization the most directly perceived is provided to user, at this, the resultant fault information of equipment is carried out to a visual processing.
Beneficial effect of the present invention is mainly manifested in: real-time is good, accuracy is high, efficiency is high.
Accompanying drawing explanation
Fig. 1 is candid photograph equipment principle of work schematic diagram.
Fig. 2 captures equipment failure classification chart.
Fig. 3 is abnormal image classification chart.
Fig. 4 captures equipment failure and abnormal image mapping relations figure.
Fig. 5 is solution design drawing of the present invention.
Fig. 6 is the algorithm of locating license plate of vehicle process flow diagram based on color characteristic and textural characteristics.
Fig. 7 is the abnormal image recognition methods design drawing combining based on many features.
Fig. 8 is the method figure of the abnormal line ball image of identification.
Fig. 9 is the method figure of the reflective abnormal image of identification.
Figure 10 is system function figure.
Figure 11 is system identification abnormal image precedence diagram.
Embodiment
Below in conjunction with accompanying drawing, the invention will be further described.
With reference to Fig. 1~Figure 11, the method for equipment fault diagnosis is captured in a kind of intelligent transportation based on image abnormity feature, in the present embodiment,
1, capture the most common failure of equipment, with reference to accompanying drawing 2.
1) video camera most common failure has: the image that white balance arranges mistake, take occurs that reflective phenomenon, video camera are arranged on the angle mistake on portal frame.
2) flashlamp most common failure has: flashlamp is bad, stops light filling work, exposure is too high, exposure is too low.
3) inductive coil most common failure has: sensitivity, reset too slow etc.
2, the Common Abnormity characteristic of image, with reference to accompanying drawing 3.
1) thus color exception refers to that in the image of being captured cannot Normal appearances going out its color property is unfavorable for computing machine identification car plate.Three kinds of color exception images of main existence: figure kine bias indigo plant, refer to that above entire image, the film just as blue layer shrouds, though do not affect car plate identification, too large with the color distortion of normal picture; Figure kine bias is dark, refers to that entire image brightness is inadequate, and wherein again in two kinds of situation, the visible and partially dark car plate of partially dark car plate is invisible; Figure kine bias is bright, and especially license plate area brightness is too high to refer to car face.
2) abnormal behavior refers to that candid photograph equipment can not normally photograph vehicle image.As, line ball, the situation of being captured while referring to vehicle through road mark line; Disorderly clap, refer to the situation that electric motor car, tricycle or pedestrian etc. do not need the target of capturing to be captured; Empty bat, refers to the situation that does not have any target process and arrived by candid photograph.
3) other are caused by other factors extremely.As, image repeats, thereby refer to that capturing in a short period of time equipment has clapped continuously twice and be kept in same image and cause vehicle in this image to repeat; Image is reflective, refers at the zone line of image and obviously has the reflective light beam of one light tone to run through entire image; Phase place is not right, refer to that candid photograph equipment had photographed vehicle image but due to the position of video camera setting angle not to and do not photograph the situation of license plate area.
Obtain their mapping relations between the two according to the common device fault of above-mentioned analysis and image abnormity feature, with reference to accompanying drawing 4.
2, the algorithm of locating license plate of vehicle based on color characteristic and textural characteristics
With reference to accompanying drawing 6, the algorithm steps of this car plate location is described below:
Step (1), in original RGB coloured image, extract the color characteristic of car plate background and character and cut apart image and obtain the candidate region of car plate with this.
Step (2), original RGB coloured image is carried out to gray scale carry out again binary conversion treatment after processing.
The binary processing method of using in this localization method under this introduces:
(2.1) binaryzation of license plate image
Can not solve well by the inhomogeneous widely different problem of vehicle image light and shade causing of intensity of illumination according to single threshold process, at this, utilize two kinds of threshold process methods to obtain best car plate bianry image, the inhomogeneous various situations that refer under natural weather environment and under flash lighting of intensity of illumination, such as: fine day, cloudy day, the under-exposed bright-dark degree that all can affect image.In the normal situation of illumination, use maximum variance between clusters, when normally illumination figure kine bias not enough and that cause is dark in the abnormal situation of illumination, show that according to the statistical information of gray level image a threshold value nAverGray carrys out binary image, AverGray is average gray, and n is a coefficient.
Distinguish illumination normal picture and the undesired image of illumination with mAverGray, m is a coefficient.As:
AverGray > mAverGray illumination normal picture
The undesired image of AverGray < mAverGray illumination
Unify the optimum efficiency of binaryzation by two kinds of threshold methods, be beneficial to the processing of car plate location below.
Step (3), complete car plate up-and-down boundary determine, in the corresponding bianry image region, candidate region completing in its step (1), carry out line scanning and add up simultaneously the number of every row jump, in the time that the sum of this row jump reaches 12 times, the candidate row of changing one's profession to be defined as license plate area, scan behind all candidate regions, obtained continuous candidate row and in these continuous candidate rows, be decided to be respectively preliminary lower boundary and the coboundary of car plate with the row at minimum value place of candidate row and the row at maximal value place.
Versicolor license plate area obtains black matrix wrongly written or mispronounced character and two kinds of situations of white background black mole after threshold process, and between character and background, can have thus the jump between black and white is the jump of pixel value from 0 to 255 or 255 to 0 in bianry image.There is twice jump in character, seven characters value of keeping of going bail at least exists 12 jumps to I haven't seen you for ages.Can greatly reduce the quantity of license plate candidate area by step (3).
Step (4), the result of the preliminary up-and-down boundary of car plate completing according to step (3), in the corresponding bianry image region surrounding at these up-and-down boundaries, carry out column scan, at this, in the time of column scan license plate candidate area, first first find out stain and character " 1 ", then remove again stain and " character 1 " widened to the numerical value as for the same width of other characters, so, spacing distance between just remaining two kinds of characters between seven characters in license plate area, and this interval location is also fixed, the spacing distance between second character and three characters is D1, spacing distance between other characters is D2.Finally remove to travel through one group of spacing distance between character with D1 and these stack features data of D2, the traversal successfully Far Left columns of first character and the rightmost columns of last character of this group character is respectively final border, car plate left and right, the border, left and right that can obtain car plate represents that car plate locates successfully, make IsLicencePlate=true, otherwise car plate is located unsuccessful IsLicencePlate=false.
At this, the value of D1 and D2 is that the height that obtains car plate after determining by the up-and-down boundary of car plate determines.The border, left and right of car plate is to determine according to the spacing distance between character, but owing to having a stain between second character and the 3rd character, the phenomenon that can disappear after threshold process when this stain has, in addition the random character arabic numeral " 1 " that occur in position thus the width of the width after vertical projection and other characters is different causes the spacing distance between characters on license plate to have diversity and uncertainty.For the spacing distance between unified character, adopted character pitch in step (4) apart from the method for adjusting, so since by more accurately determining that to the adjustment at interval between character the border, left and right of car plate finally can improve the accuracy that car plate is located like this.
3, identification colors abnormal image
The color exception characteristic of image is mainly reflected in the half-tone information in statistical law and the gray level image of R, G in former RGB coloured image, tri-passages of B.
1) figure kine bias indigo plant, the partially blue method step of recognition image is described below:
Step (1), in original color image, add up blue pixel value in pixel value interval [96,160] situation, definition RatioBlue[96,160] for the ratio of the sum of all pixels of the number of pixels sum of blue pixel value in this interval and entire image and try to achieve this value.
At this, explain lower selection [96, 160] reason: we analyze by the pixel value to its blue channel in RGB coloured image, main by the average pixel value [0 of dividing, 255] interval method, add up respectively the ratio of the sum of all pixels of its blue pixel value in the interior number of pixels sum in division back zone and whole image, interval and the integration of dividing before then progressively dwindling according to the feature of this ratio is interval, finally obtain the ratio that a pixel value interval makes the number of pixels sum of blue pixel value in this interval and sum of all pixels and reach maximum, by great many of experiments with repeatedly test us to obtain this pixel value interval be [96, 160], namely in the time that figure kine bias is blue, the pixel value of its blue channel mainly concentrates on [96, 160] in interval.
Step (2), according to the method for step (1), same definition RatioGreen[96,160] and RatioRed[96,160] be respectively the ratio of the sum of all pixels of number of pixels sum in this interval of green pixel values and red pixel value and entire image, meanwhile, try to achieve this two values.
Since feature when figure kine bias is blue is mainly that a large amount of blue pixel values is in pixel value interval [96,160], and the pixel value of remaining red channel and green channel only has a small part to concentrate in [96,160] interval.Meanwhile, we also test and show that the red and ratio of green channel in this interval is a very little number.
Step (3), three ratio: RatioBlue[96 that calculate according to step (1) and step (2), 160], RatioGreen[96,160] and RatioRed[96,160], judge whether to meet condition below:
3.1:RatioBlue[96,160]>Max
3.2:Ratio6reen[96,160]<Min?and?RatioRed[96,160]<Min;
3.3:IsLicencePlate=false
What satisfy condition this image of explanation belongs to figure kine bias indigo plant extremely, otherwise is not.
In experiment, we obtain respectively Max=0.7 and the successfully identification to figure kine bias indigo plant of Min=0.05.
2) figure kine bias is dark
In the time of the dark feature of analysis chart kine bias, be to analyze grey-level statistics in its gray level image, thereby reduced the directly calculated amount of three passages in RGB coloured image.The method step that recognition image is partially dark is described below:
Step (1), original color image is carried out to gray processing processing, adopts weighted average method:
GrayValue=(RValue*299+GValue*587+BValue*114)/1000
RValue, GValue, Bvalue respectively value are the numerical value of three passages of RGB in original color image, and GrayValue is the gray-scale value of each pixel after gray processing.The average gray value that calculates gray level image is AverGray.
Step (2), in gray level image, statistics half-tone information the regularity of distribution.Calculate GrayValueLess (nAverGray) value, and make this value trend maximum.GrayValueLess (nAverGray) is less than the sum of the number of pixels of nAverGray for gray-scale value.
Step (3), the result drawing according to step (2), calculate Ratio[GrayValueLess (nAverGray)] value, this value is the ratio of the sum of all pixels of GrayValueLess (nAverGray) and entire image.
Step (4), the result of calculating according to step (3), judge whether to meet condition below:
Ratio[GrayValueLess(nAverGray)]>Y
Satisfy condition this image of explanation extremely to belong to figure kine bias dark, otherwise be not.
In the time analyzing the abnormal characteristic of partially dark image, we analyze its half-tone information, find in the time that figure kine bias is dark, and the absolutely large part of the gray-scale value of its pixel is all less than a threshold value, and this threshold value is relevant with the average gray value of its entire image.At this, get n=2, Y=0.8 can well identify partially dark image from whole image data base.
Mentioned above, in the time that figure kine bias is dark, be divided into two kinds of situations: the visible and partially dark car plate of partially dark car plate was invisible.Which kind of at this moment need the result IsLicencePlate=false or ture of combining image car plate location to identify specifically to belong to figure kine bias dark.
4, identification abnormal behavior image
Abnormal behavior image is because video camera candid photograph abnormal behavior causes, the target vehicle comprising in the present image of property list of this image abnormal, comprises lacking in image in target vehicle, image, having comprised the three kinds of situations of target vehicle malposition in abnormal target vehicle, image.Analyze thisly when abnormal, mainly find and best embody this abnormal rule according to zoning method.
1) empty bat
Empty bat refers in the image of being captured and comprises various vehicles and pedestrian without any target object.Empty to clap the content overwhelming majority in image be all road surface, is subject to greenbelt and road mark line etc. and has the impact of the factor of vertical directivity feature, causes its empty feature of clapping image and have in the horizontal direction homogeneity.This homogeneity corresponds to has the trickle property of row changing features in its gray level image, the namely variation of the average gray value of every row pixel does not present jumping characteristic, but centered by the average gray value of entire image, a very stable variation up and down in the scope of fractional value X.So, in the horizontal direction image equalization being divided into N little rectangular area, the difference of the average gray value in N rectangular area and the average gray value of entire image is all very little.The empty image method step of clapping of identification is described as follows:
Step (1), original color image is carried out to gray processing processing, adopt weighted average method.
Step (2), in gray level image, the height that obtains dividing N little rectangular area is LHeight=image.Height ()/N, and can to obtain the individual little rectangular area of N with the width value of this height and original image be LRECT i=[i*LHeight, (i+1) * LHeight, 0, image.Width ()-1];
Step (3), calculate respectively the average gray value AverGray (LRECT of N little rectangular area i) and the average gray value AverGray of view picture gray level image judge whether to meet condition below:
3.1:|AverGray(LRECT i)-AverGray(Image)|<X
32:IsLicencePlate=false
Satisfy condition and represent that the abnormal of this image is empty bat, otherwise be not.
At this, i=1,2 ... N-1, image.Height (), image.Width () is respectively height and the width of image.Work as N=20, when X=5, can effectively identify the empty image of clapping.
Design a function IsNullPicture (int nHeight, int nLeft, int nRight, int nPara) represent that this image claps for empty when=true, the height that wherein parameter nHeight is image to be identified, nLeft is the left margin columns of image, the right margin columns that nRight is image, and nPara is threshold value.
2) line ball
Line ball refers to the situation of being captured pass through driveway separatrix in the time of replacing vehicle track time, this situation causes by capturing slowly, in line ball image, we cannot obtain complete license board information conventionally, and the position of incomplete target vehicle is all generally the lower left corner and the lower right corner place in image
With reference to accompanying drawing 8, the method step of identification line ball is described below:
Step (1), original color image through gray scale process after, gray level image equalization is divided into four regions: upper left, upper right, lower-left, bottom right, respectively note do: Alu, Aru, Alb, Arb.
Step (2), respectively its average gray value is calculated in Alu, Aru, Alb, tetra-regions of Arb, note is done: AverGray (Alu), AverGray (Aru), AverGray (Alb), AverGray (Arb) and try to achieve maximal value Max and the minimum M in of these four average gray values the insides.
There is larger difference in its average gray value of the region at target vehicle place and other trizonal average gray value, its average gray value of region at target vehicle place shows as the maximal value (max) of the inside, four regions or the feature of minimum value (min) conventionally.Meanwhile, other trizonal average gray values have similarity (similarity) and very little with the difference of the average gray value absolute value of entire image.
Step (3), step (1) gray scale process after, further carry out binary conversion treatment.Add up respectively the total nSumWhitePxl of white pixel and the total nSumBlackPxl of black picture element in bianry image, if nSumWhitePxl is greater than nSumBlackPxl, represents that the background of this bianry image is white, otherwise be black.
In the time that background is white, the sum of adding up respectively the black picture element in lower-left (Alb) and bottom right (Arb) two regions is designated as nSumBlackAlb and nSumBlackArb, and remembering that RatioBlack1 is the ratio of nSumBlackAlb and nSumBlackPxl, RatioBlack2 is the ratio of nSumBlackArb and nSumBlackPxl.If meet RatioBlack1 > X or this condition of RatioBlack2 > X, make IsPressLineThreAux=true, otherwise IsPressLineThreAux=false.
When background is that black is, the sum of adding up equally respectively the white pixel in lower-left (Alb) and bottom right (Arb) two regions is designated as nSumWhiteAlb and nSumWhiteArb, and remembering that RatioWhite1 is the ratio of nSumWhiteAlb and nSumWhitePxl, RatioWhite2 is the ratio of nSumWhiteArb and nSumWhitePxl.If meet RatioBlack1 > X or this condition of RatioBlack2 > X, make IsPressLineThreAux=true, otherwise IsPressLineThreAux=false.Be 0.65 in this X value.
Step (4), according to the result of step (2) and step (3), carry out last judgement, whether meet following condition:
(4.1):AverGray(Alb)=Max?or?AverGray(Alb)=Min
(4.2):IsNullPicture(image.Height(),image.Width()/2,image.Width(),5)=true
(4.3):IsPressLineThreAux=true
(4.4):IsLicencePlate=false
Satisfy condition and represent that the abnormal of this image is line ball, otherwise be not.
At this, explain three conditions in lower step (4): in the time that vehicle line ball travels by candid photograph, the position of part target vehicle is in lower-left (Alb) or region, bottom right (Arb).There is larger difference in its average gray value of the region at target vehicle place and other trizonal average gray value, conventionally its average gray value of region at target vehicle place shows as the maximal value (max) of the inside, four regions or the feature of minimum value (min), so respective conditions 4.1; Meanwhile, other trizonal average gray values have similarity (similarity) and very little with the difference of the average gray value absolute value of entire image.Namely indicate, in the time that line ball situation belongs to lower left corner line ball, image in the vertical direction is equally divided into two parts in left and right, and the right half part of image meets feature when image is empty to be clapped, and when the line ball of the lower right corner, in like manner can obtain, thus respective conditions 4.2; Because line ball situation is a kind of approaching and the situation of normal picture, only sometimes easily cause erroneous judgement by the provincial characteristics of its gray level image, thus, in order to strengthen the reliability of identifying, in identification line ball image process, also need to utilize the aid identification effect of its bianry image, therefore respective conditions 4.3.
5, identify other abnormal images
The unique point that other off-notes of image mainly comprise from size and the image of image shows, and analyzes this abnormal characteristic and mainly adopt the method for feature extraction.
1) image repeats
In the middle of experiment, the resolution of the view data getting from candid photograph equipment of using is 1408*1088.Conventionally,, for best visual effect, the ratio of width to height of image is generally 4 to 3.But in the time of abnormal attribute that image duplicates, the width that its resolution reaches its image of 2408*1088 is the twice of normal picture width, causes its width and the ratio of height to be greater than 2.According to statistics, the ratio of width to height that most candid cameras are captured normal picture is all less than 2, so the method step that recognition image repeats is described below:
Step (1), obtain respectively the width image.Width () of original color image and the height image.Height () of image.
Step (2), calculating the ratio of width to height are image.Width ()/image.Height () and judge whether this ratio is greater than 2.Ratio is greater than 2 and shows that the abnormal of this image is that image repeats, otherwise is not.
2) image is reflective
In the time that image is reflective, there is a branch of light tone light beam in the vertical direction to cross over entire image at original color image center section, correspond in its bianry image, light tone light beam becomes white light beam.At this, we are defined as this white light beam in bianry image a little rectangular area of continuous row composition, and the number sum of the white pixel point of each row is greater than 95% with the ratio of the sum of this row pixel.So in bianry image, we mainly find the white light beam with certain width and ad-hoc location, and there is larger difference in average gray value in the neighborhood of average gray value in this white light beam and its right and left certain width.Found, this image table reveal reflective abnormal, otherwise, do not there is reflective abnormal characteristic.
With reference to accompanying drawing 9, the step of the reflective method of recognition image is described below:
Step (1), original color image are processed and are carried out binary conversion treatment again through gray scale, carry out column scan in its bianry image, calculate the statistical conditions of each row monochrome pixels.
Step (2), in the time that the ratio of the number sum of a certain row white pixel point and the sum of this row pixel is greater than 0.95, these row are defined as to white light beam row, and make BackLight[i]=1, meanwhile, BackLight[i]=-1 represent that i classifies non-white light beam row as.
Step (3), travel through whole BackLight[i] array, if meet BackLight[i+1]=1 and BackLight[i-49 ..., i]=-1, BackLight[i-49 ..., i] represent to be listed as each the row BackLight[i i row from i-49] and value.50, i+1 row left side columns is all non-white light beam row simultaneously, represent that i+1 classifies non-white light beam as and is listed as the saltus step row that white light beam is listed as, so i+1 row are defined as to primary election white light beam left margin row, and make blLeftArray[blaI]=i+1, blLeftArray[blaI] be the set of white light beam left margin row.
Step (4), travel through whole BackLight[i] array, if meet BackLight[i]=1 and BackLight[i+1, ... i+50]=-1, BackLight[i+1 ... i+50] represent to be listed as each the row BackLight[i i+50 row from i+1] and value.50, the right of i row columns is all non-white light beam row simultaneously, represent that i classifies white light beam as and is listed as the saltus step row that non-white light beam is listed as, so i row are defined as to primary election white light beam right margin row, and make blRightArray[braJ]=i, blRightArray[braJ] be the set of white light beam left margin row.
Step (5), at blLeftArray[blaI] and blRightArray[braJ] set Rigen be less than braJ and BackLight[blaI according to blaI ..., braJ]=1 find paired row, this paired row is combined into the white light beam region of primary election.
Step (6), in its corresponding gray level image, calculate the average gray value AverGrayNext in 5 pixel regions of average gray value AverGrayBL and left and right, this white light beam region in white light beam region of primary election.Be greater than 20 if meet the difference of AverGrayBL and AverGrayNext, can represent that this white light beam region is the characteristic white light beam in the reflective image that we need to look for, and it is that image is reflective that IsLicencePlate=false can judge the abnormal of this image, otherwise is not.
The method of the various image abnormities of identification of introducing separately in sum, comprehensively obtains computing machine and identifies on the whole the method step of abnormal image and be described below with reference to accompanying drawing 11:
Step (1), computing machine read image data base to be identified continuously, read after an original color image, respectively it are carried out to gray scale processing and binary conversion treatment, and preserve corresponding view data after treatment.
Step (2), obtain the result of car plate location in conjunction with coloured image and bianry image, the unsuccessful N that is masked as in car plate location.
Step (3), directly utilize coloured image successively recognition image whether repeat, whether the result N recognition image that adds car plate location is partially blue, to finish the identification that this takes turns after the information of this abnormal image is saved in database by the words of (Y), read next image, the words of no (N) are carried out next step.
Step (4), result N whether partially dark and empty bat of recognition image successively of utilizing gray level image and car plate location, the same step of processing (3) of recognition result.
Step (5), result N whether line ball and empty bat of recognition image successively of utilizing gray level image and bianry image and car plate location, the same step of processing (3) of recognition result.
Step (6), the result N identification off-note of locating according to car plate are not other abnormal images clearly.
Above-mentioned is only to illustrate with the order of this kind of identification computing machine is how to identify abnormal image, in fact, along with the operation of system, will in time adjust the order of identification according to the database of abnormal image information.
6, the realization of the adaptivity of recognizer
The adaptivity of recognizer refers to that algorithm can adjust the order of the various abnormal images of identification in time, guarantees algorithm lasting high efficiency.At this, we are divided into internal cause and external cause the factor that affects efficiency of algorithm.Internal cause refers to the efficiency of every kind of image abnormity of algorithm self identification, and our unit of definition of this efficiency is: second/kind, identify every kind of time that image abnormity algorithm is required.Internal cause is relatively-stationary, unless just we have further improvement to increase to algorithm.External cause refers to the operation along with system, and the data of system the inside are more and more, and for huge system database, the ratio of the shared whole data of any abnormal image is larger on earth, and we are unpredictable arriving.The sequence of the ratio of the shared whole data of various abnormal images is the results along with a stochastic process of time variation.At this, we are defined as external cause: the ratio of this kind of shared whole data of abnormal image.And to internal cause and external cause, according to it, the size to whole algorithm affects calculates rational weights respectively, tries to achieve their weighted sum value.This value is exactly to identify the priority of various image abnormities.Determine recognition sequence according to the sequence of priority.
7, the Visual Implementation of resultant fault information
1) regional information of resultant fault information is visual
Regional information is visual to be referred in certain territorial scope, and the failure message of capturing equipment for all intelligent transportation in this region does detailed and comprehensive statistical work, draws a visual result as a reference.
This region can refer to that a city can be also the whole province, so since, just can facilitate to such an extent that draw in the public transport links in a city, mainly refer to the candid photograph device fault information on crossroad, citywide and Important Sections.Such as, the equipment failure in some crossing or section occurs frequent.Also can conveniently obtain capturing in the whole province's freeway traffic net the distribution of device fault information, as which bar highway, the failure message of which section highway and which charge station accounts for many, this visual result, normally using the figure of the whole province's freeway traffic net as main body, use respectively the different highway of different color descriptions, color is redder represents that this highway equipment failure is more serious, green can represent that equipment failure entirety is also relatively good, by show the freeway traffic net figure of such different colours to user, can there is a comprehensively cognition to the candid photograph equipment failure situation on whole highway at a glance.Had the visual result of this regional information, user just can make the breakdown maintenance of candid photograph equipment and the decision-making of daily servicing easily rapidly.
2) temporal information of resultant fault information is visual
Visual the referring to according to temporal characteristics of temporal information expressed package failure message.The visual fully consersion unit fault of temporal characteristics sometime section in occurrence frequency, such as: can research equipment in which of 1 year in the month equipment failure of entirety occur more frequently, daily like this plant maintenance work just can be strengthened its maintenance work targetedly according to this result in specific several months of easily breaking down.Meanwhile, also can count respectively single equipment within a shorter time period, corresponding occurrence frequency, particularly for light compensating lamp, this result can predict the situation that flashlamp breaks down completely, to prevent trouble before it happens.There is the situation of the frequent exposure deficiency in the short time in ordinary flash lantern festival, for this situation, provides separately a visualization result about flashlamp fault and can provide extremely important reference information to user.Visual histogram or the similar stock trend graph of adopting of temporal information represents.
3) the failure rate information visualization of resultant fault information
Failure rate information visualization refers within a certain period of time, investigates the fault occupation rate situation of video camera under a set of candid photograph equipment, flashlamp, wagon detector.To express the larger information of likelihood ratio that under a set of equipment, which equipment breaks down conventionally.And this information conventionally can feed back to and captures device hardware equipment purchase department the most authoritative reference is provided, guarantee that the quality of a whole set of candid photograph equipment of next group is higher.At this, failure rate information visualization can whole candid photograph equipment principle of work schematic diagram (with reference to accompanying drawing 1) be main body, can dwindle respectively or amplify according to the incidence of equipment failure the figure of corresponding equipment representative.Such as, in its principle of work schematic diagram, video camera is consistent with flashlamp and the initial size of wagon detector, in the time that the failure rate of flashlamp is all large than the failure rate of other two equipment, at this moment, can amplify in proportion the figure of corresponding flashlamp.Obviously just can distinguish clearly greatly three failure rate situations between equipment than the figure of video camera and wagon detector according to the figure of flashlamp like this.
8, Fault Diagnosis of Mechanical Equipment is captured in the intelligent transportation based on image abnormity feature
With reference to accompanying drawing 10.
This system is mainly made up of configuration module, identification module, data analysis module.
Configuration module mainly completes: system convention administration configuration, comprises three grades of customer administrators' management; File configuration, refers to the original input path of view data and the outgoing route of diagnostic result; And the work of algorithm initialization, refer to before system commencement of commercial operation, the order of recognizer is adjusted accordingly, guarantee that optimum recognizer supports the high efficiency of whole identifying.
Identification module mainly completes: identify various abnormal images according to the method for above-mentioned introduction.
Data analysis module mainly completes: the result obtaining according to identification module, the data volume situation of every kind of abnormal image of statistical study, this statistic analysis result will directly feed back to configuration module and complete the initialization of algorithm, need in addition the printing of finishing equipment fault list, this list mainly comprises filename and this abnormal corresponding two contents of equipment failure of abnormal image, at this, according to the naming rule of image file, its corresponding filename can navigate to candid photograph equipment.Finally, provide a comprehensive failure message visualization result according to whole failure messages.

Claims (3)

1. a method of capturing equipment fault diagnosis based on the intelligent transportation of image abnormity feature, is characterized in that: comprise the following steps:
1), set up the mapping relations of capturing between equipment failure and abnormal image;
Capture equipment failure and comprise video camera most common failure, flashlamp most common failure and inductive coil most common failure; Image abnormity characteristic comprises color exception, abnormal behavior and other are abnormal; Set up their mapping relations between the two according to capturing equipment failure and image abnormity feature;
2), the car plate position fixing process based on color characteristic and textural characteristics:
Judge in image whether have effective license board information according to whether can successfully navigate to car plate in image, can navigate in car plate key diagram picture and have license board information, otherwise there is no license board information;
The feature of car plate mainly concentrates on color and above character, the wrongly written or mispronounced character of the blue end of four kinds of color characteristics of car plate, yellow end surplus, white gravoply, with black engraved characters and black matrix wrongly written or mispronounced character, character is that the additional cut-point symbol of seven character group being made up of Chinese character, English alphabet and arabic numeral respectively forms, and the interval between character has unified standard;
Car plate location based on color characteristic and textural characteristics is mainly the candidate region that utilizes these four kinds of color characteristics of car plate to carry out Primary Location to obtain car plate, thereby then in its binary image, carrying out secondary location according to the fixed intervals feature between gray scale jumping characteristic and character between characters on license plate obtains the pinpoint result of car plate;
3), the abnormal image recognition methods that combines based on many features, process is:
3.1) identification colors abnormal image
The color exception characteristic of image is mainly reflected in the half-tone information in statistical law characteristic and the gray level image of R, G in former RGB coloured image, tri-passages of B, and the color exception image of identifying refers to: figure kine bias is blue and figure kine bias is dark;
The process of identification colors abnormal image is:
3.1.1) figure kine bias indigo plant, the partially blue method step of recognition image is described below:
3.1.1.1), in original color image, add up blue pixel value in pixel value interval [96; 160] situation; definition RatioBlue[96,160] for the ratio of the sum of all pixels of the number of pixels sum of blue pixel value in this interval and entire image and try to achieve this value;
3.1.1.2), according to step 3.1.1.1) method, same definition RatioGreen[96,160] and RatioRed[96,160] be respectively the ratio of the sum of all pixels of number of pixels sum in this interval of green pixel values and red pixel value and entire image, meanwhile, try to achieve this two values;
3.1.1.3), three ratio: RatioBlue[96 calculating according to step (3.1.1.1) and step (3.1.1.2), 160], RatioGreen[96,160] and RatioRed[96,160], judge whether to meet condition below:
RatioBlue[96,160]>Max???(3-1)
RatioGreen[96,160]<Min?and?RatioRed[96,160]<Min???(3-2)
IsLicencePlate=false???(3-3)
What satisfy condition this image of explanation belongs to figure kine bias indigo plant extremely, otherwise is not;
3.1.2) figure kine bias is dark, and process is as follows:
3.1.2.1), original color image is carried out to gray processing processing, employing weighted average method:
GrayValue=(RValue*299+GValue*587+BValue*114)/1000???(3-4)
RValue, GValue, Bvalue respectively value are the numerical value of three passages of RGB in original color image, and GrayValue is the gray-scale value of each pixel after gray processing; The average gray value that calculates gray level image is AverGray;
3.1.2.2), in gray level image, the regularity of distribution of statistics half-tone information, calculate GrayValueLess (nAverGray) value, and make this value trend maximum, GrayValueLess (nAverGray) is less than the sum of the number of pixels of nAverGray for gray-scale value;
3.1.2.3), according to step 3.1.2.2) result that draws, calculate Ratio[GrayValueLess (nAverGray)] value, this value is GrayValueLess (nAverGray) and the ratio of the sum of all pixels of entire image;
3.1.2.4), according to step 3.1.2.3) result calculated, judge whether to meet condition below:
Ratio[GrayValueLess(nAverGray)]>Y???(3-5)
Wherein, get Y=0.8;
Satisfy condition this image of explanation extremely to belong to figure kine bias dark, otherwise be not;
3.2) identification abnormal behavior image
Judge whether to exist abnormal behavior image according to zoning method, if exist to lack in image and comprised the target vehicle malposition in abnormal target vehicle and image in target vehicle, image, judge and exist extremely, identification abnormal behavior image specifically refers to: empty image and the identification line ball image clapped of identification;
The process of identification abnormal behavior image is as follows:
3.2.1) the empty image method step of clapping of identification is described as follows:
3.2.1.1), original color image is carried out to gray processing processing, employing weighted average method;
3.2.1.2), in gray level image, the height that obtains dividing N little rectangular area is LHeight=image.Height ()/N, and can to obtain the individual little rectangular area of N with the width value of this height and original image be LRECT i=[i*LHeight, (i+1) * LHeight, 0, image.Width ()-1];
3.2.1.3), calculate respectively the average gray value AverGray (LRECT of N little rectangular area i) and the average gray value AverGray of view picture gray level image judge whether to meet condition below:
|AverGray(LRECT i)-AverGray(Image)|<X???(3-6)
IsLicencePlate=false???(3-7)
Wherein, get N=20, X=5;
Satisfy condition and represent that the abnormal of this image is empty bat, otherwise be not;
At this, i=1,2 ... N-1, image.Height (), image.Width () is respectively height and the width of image;
Design a function IsNullPicture (int nHeight, int nLeft, int nRight, int nPara) represent that this image claps for empty when=true, the height that wherein parameter nHeight is image to be identified, nLeft is the left margin columns of image, the right margin columns that nRight is image, and nPara is threshold value;
3.2.2) method step of identification line ball is described below:
3.2.2.1), original color image through gray scale process after, gray level image equalization is divided into four regions: upper left, upper right, lower-left, bottom right, respectively note do: Alu, Aru, Alb, Arb;
3.2.2.2), respectively its average gray value is calculated in Alu, Aru, Alb, tetra-regions of Arb, note is done: AverGray (Alu), AverGray (Aru), AverGray (Alb), AverGray (Arb) and try to achieve maximal value Max and the minimum M in of these four average gray values the insides;
3.2.2.3), at step 3.2.2.1) gray scale process after, further carry out binary conversion treatment; Add up respectively the total nSumWhitePxl of white pixel and the total nSumBlackPxl of black picture element in bianry image, if nSumWhitePxl is greater than nSumBlackPxl, represents that the background of this bianry image is white, otherwise be black;
In the time that background is white, the sum of adding up respectively the black picture element in lower-left Alb and ArbLiang Ge region, bottom right is designated as nSumBlackAlb and nSumBlackArb, and remembering that RatioBlack1 is the ratio of nSumBlackAlb and nSumBlackPxl, RatioBlack2 is the ratio of nSumBlackArb and nSumBlackPxl; If meet RatioBlack1>0.65 or RatioBlack2>0.65, make IsPressLineThreAux=true, otherwise IsPressLineThreAux=false;
In the time that background is black, the sum of adding up equally respectively the white pixel in lower-left Alb and ArbLiang Ge region, bottom right is designated as nSumWhiteAlb and nSumWhiteArb, and remembering that RatioWhite1 is the ratio of nSumWhiteAlb and nSumWhitePxl, RatioWhite2 is the ratio of nSumWhiteArb and nSumWhitePxl; If meet this condition of RatioBlack1>0.65 or RatioBlack2>0.65, make IsPressLineThreAux=true, otherwise IsPressLineThreAux=false;
3.2.2.4), according to step 3.2.2.2) and step 3.2.2.3) result, carry out last judgement, whether meet following condition:
AverGray(Alb)=Max?or?AverGray(Alb)=Min???(3-8)
IsNullPicture(image.Height(),image.Width()/2,image.Width(),5)=true???(3-9)
IsPressLineThreAux=true???(3-10)
IsLicencePlate=false???(3-11)
Satisfy condition and represent that the abnormal of this image is line ball, otherwise be not;
3.3) identify other abnormal images
Adopt the method for feature extraction to obtain the size of image and the unique point that image comprises, identify other abnormal images and specifically refer to: recognition image repetition and recognition image are reflective; The step of method of identifying other abnormal images is as follows:
3.3.1) method step that recognition image repeats is described below:
3.3.1.1), obtain respectively the width image.Width () of original color image and the height image.Height () of image;
3.3.1.2), calculating the ratio of width to height is image.Width ()/image.Height () and judges whether this ratio is greater than 2; Ratio is greater than 2 and shows that the abnormal of this image is that image repeats, otherwise is not;
3.3.2) step of the reflective method of recognition image is described below:
3.3.2.1), original color image through gray scale process carry out again binary conversion treatment, in its bianry image, carry out column scan, calculate the statistical conditions of each row monochrome pixels;
3.3.2.2), in the time that the ratio of the number sum of a certain row white pixel point and the sum of this row pixel is greater than 0.95, these row are defined as to white light beam row, and make BackLight[i]=1, meanwhile, BackLight[i]=-1 represent that i classifies non-white light beam row as;
3.3.2.3), travel through whole BackLight[i] array, if meet BackLight[i+1]=1 and BackLight[i-49 ..., i]=-1, BackLight[i-49 ..., i] represent to be listed as each the row BackLight[i i row from i-49] value; 50, i+1 row left side columns is all non-white light beam row simultaneously, represent that i+1 classifies non-white light beam as and is listed as the saltus step row that white light beam is listed as, so i+1 row are defined as to primary election white light beam left margin row, and make blLeftArray[blaI]=i+1, blLeftArray[blaI] be the set of white light beam left margin row;
3.3.2.4), travel through whole BackLight[i] array, if meet BackLight[i]=1 and BackLight[i+1 ... i+50]=-1, BackLight[i+1 ... i+50] represent to be listed as each the row BackLight[i i+50 row from i+1] value; 50, the right of i row columns is all non-white light beam row simultaneously, represent that i classifies white light beam as and is listed as the saltus step row that non-white light beam is listed as, so i row are defined as to primary election white light beam right margin row, and make blRightArray[braJ]=i, blRightArray[braJ] be the set of white light beam left margin row;
3.3.2.5), at blLeftArray[blaI] and blRightArray[braJ] set Rigen be less than braJ and BackLight[blaI according to blaI ..., braJ] and=1 searching paired row, this paired row is combined into the white light beam region of primary election;
3.3.2.6), in its corresponding gray level image, calculate the average gray value AverGrayNext in 5 pixel regions of average gray value AverGrayBL and left and right, this white light beam region in white light beam region of primary election; Be greater than 20 if meet the difference of AverGrayBL and AverGrayNext, can represent that this white light beam region is the characteristic white light beam in the reflective image that we need to look for, and it is that image is reflective that IsLicencePlate=false can judge the abnormal of this image, otherwise is not.
2. the method for equipment fault diagnosis is captured in a kind of intelligent transportation based on image abnormity feature as claimed in claim 1, it is characterized in that: in described step 3), adjust the order of the various abnormal images of identification.
3. the method for equipment fault diagnosis is captured in a kind of intelligent transportation based on image abnormity feature as claimed in claim 1 or 2, it is characterized in that: the method for described fault diagnosis also comprises: 4) resultant fault information is visual: comprehensive device fault information is done to a visualization processing, provide the visualization result of a human to be used for carrying out decision-making to user.
CN201210133557.3A 2012-04-28 2012-04-28 Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic Active CN102708378B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210133557.3A CN102708378B (en) 2012-04-28 2012-04-28 Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201210133557.3A CN102708378B (en) 2012-04-28 2012-04-28 Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic

Publications (2)

Publication Number Publication Date
CN102708378A CN102708378A (en) 2012-10-03
CN102708378B true CN102708378B (en) 2014-06-11

Family

ID=46901121

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210133557.3A Active CN102708378B (en) 2012-04-28 2012-04-28 Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic

Country Status (1)

Country Link
CN (1) CN102708378B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130208121A1 (en) * 2012-02-10 2013-08-15 Xerox Corporation Traffic camera diagnostics via test targets
CN104582209B (en) * 2013-10-29 2017-03-15 杭州海康威视数字技术股份有限公司 Light compensating lamp fault detection method and its device
CN104237808B (en) * 2014-07-28 2017-01-18 四川大学 Electronic police system light supplementary lamp fault detecting method based on image abnormities
CN105502115A (en) * 2014-09-26 2016-04-20 刘一 Maintenance quality online assessment method, device and system based on elevator
DE102015220884A1 (en) * 2014-10-27 2016-04-28 Denso Corporation Image processing device
CN108205675B (en) * 2016-12-20 2020-06-19 浙江宇视科技有限公司 License plate image processing method and device
CN107393309A (en) * 2017-08-31 2017-11-24 江苏开放大学 A kind of electronic police system
CN107909822B (en) * 2017-11-29 2019-11-22 银江股份有限公司 SCATS coil checker automatic diagnosis method based on flow and saturation analysis
CN108881896A (en) * 2017-12-21 2018-11-23 北京旷视科技有限公司 Test method, system and the storage medium of capture machine
CN109409191A (en) * 2018-08-24 2019-03-01 广东智媒云图科技股份有限公司 A kind of zebra stripes vehicle evacuation detection method and system based on machine learning

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1848370A (en) * 2005-12-05 2006-10-18 北京北方微电子基地设备工艺研究中心有限责任公司 Fault detecting method for semiconductor etching equipment
US20110157373A1 (en) * 2009-12-24 2011-06-30 Cognex Corporation System and method for runtime determination of camera miscalibration
CN101819411B (en) * 2010-03-17 2011-06-15 燕山大学 GPU-based equipment fault early-warning and diagnosis method for improving weighted association rules
CN102236783A (en) * 2010-04-29 2011-11-09 索尼公司 Method and equipment for detecting abnormal actions and method and equipment for generating detector

Also Published As

Publication number Publication date
CN102708378A (en) 2012-10-03

Similar Documents

Publication Publication Date Title
CN102708378B (en) Method for diagnosing fault of intelligent traffic capturing equipment based on image abnormal characteristic
US9704060B2 (en) Method for detecting traffic violation
Saha et al. License Plate localization from vehicle images: An edge based multi-stage approach
CN104766086B (en) The monitoring and managing method and system of a kind of way mark
CN106373426B (en) Parking stall based on computer vision and violation road occupation for parking monitoring method
CN101344988B (en) Image acquisition and processing equipment and method, vehicle monitoring and recording system
CN105354530A (en) Vehicle body color identification method and apparatus
CN102867417B (en) Taxi anti-forgery system and taxi anti-forgery method
CN102509090B (en) A kind of vehicle feature recognition device based on public safety video image in sky net engineering
CN1725266A (en) Vehicle intelligent monitoring recording syste mand method based on video trigger and speed measuring
CN102722704A (en) Method and system for recognizing vehicle license plate by integrating video dynamic tracking
CN103186982B (en) Intelligent access system for public security
CN102867411B (en) Taxi dispatching method and taxi dispatching system on basis of video monitoring system
CN102915433B (en) Character combination-based license plate positioning and identifying method
CN102156862B (en) License plate recognition system and license plate recognition method preventing blocking and altering
CN101377811B (en) Method and system for recognizing license plate
CN2718699Y (en) Intelligent monitoring and recording system based on video trigger and velocity measurement for vehicle
CN105741324A (en) Moving object detection identification and tracking method on moving platform
CN106448184B (en) Vehicle identification method and vehicle appearance recognition methods
CN105160297A (en) Masked man event automatic detection method based on skin color characteristics
CN104504744B (en) A kind of true method and device of plan for synthesizing license plate image
CN204884166U (en) Regional violating regulations parking monitoring devices is stopped to traffic taboo
CN108109385A (en) A kind of vehicle identification of power transmission line external force damage prevention and hazardous act judgement system and method
CN105046948A (en) System and method of monitoring illegal traffic parking in yellow grid line area
CN110660222A (en) Intelligent environment-friendly electronic snapshot system for black smoke vehicle on road

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant