CN107545216A - A kind of vehicle identification method available for power-line patrolling - Google Patents
A kind of vehicle identification method available for power-line patrolling Download PDFInfo
- Publication number
- CN107545216A CN107545216A CN201610490262.XA CN201610490262A CN107545216A CN 107545216 A CN107545216 A CN 107545216A CN 201610490262 A CN201610490262 A CN 201610490262A CN 107545216 A CN107545216 A CN 107545216A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- image
- power
- vehicle image
- identification method
- 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.)
- Pending
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of vehicle identification method available for power-line patrolling, is related to field of intelligent video surveillance.The vehicle identification method is by obtaining the video image for including vehicle image, then the vehicle image in video image is detected, the vehicle image detected is alignd according to standard vehicle information, then the vehicle image of alignment is subjected to pyramid change, feature point extraction is carried out to vehicle image simultaneously, finally the characteristic point of extraction is classified with Linear SVM, so as to identify the type of vehicle.The present invention can effectively solve the problem that the problems such as such as noise, illumination variation, dimensional variation, translation, rotation, enables adaptation to the vehicle vehicle identification of complex environment, realizes for power-line patrolling.
Description
Technical field
The present invention relates to field of intelligent video surveillance, more particularly to one kind to prevent that transmission line equipment tripping operation, the tower etc. that falls are outer
Power breaks the vehicle identification method based on video image available for power-line patrolling that accident occurs.
Background technology
With the high speed development of national economy, city is big to be built exhaling and goes out, and high ferro, overhead, the size such as repair the roads construction is every
Nian Doucheng geometry multiple increases, and urban development gradually spreads to suburb.The various constructions of the thing followed, squatter building, trees etc. trigger
Transmission line equipment tripping operation, fall the external force malicious event such as tower also in ascendant trend year by year.Particularly used in work progress
Large-scale, superelevation equipment, vehicle(Such as crane, excavator, dump truck, concrete mixer)The transmission line equipment tripping operation of initiation,
The external force malicious event such as tower is most.This not only brings great economic loss to electric power enterprise, at the same also to electric power netting safe running,
People's lives and properties constitute great threat.Therefore, in order to prevent transmission line equipment tripping operation, the external force malicious event such as tower that falls
Generation, need large-scale, the superelevation equipment to being used in work progress, vehicle badly(Such as crane, excavator, dump truck, concrete pump
Car etc.)It is monitored, is easy to take action in time.And in monitoring process, first technical problems to be solved are exactly that vehicle is entered
Row identification.
At present, existing vehicle recongnition technique, image preprocessing, image segmentation, feature extraction, vehicle classification are generally comprised
The steps such as identification.Wherein, there is numerous feature extracting methods in characteristic extraction step, such as Fourier descriptor, moment characteristics, conversion
Characteristic of field, edge contour feature, intersection point feature etc..But can solve such as to make an uproar well without a kind of feature extracting method
The problems such as sound, illumination variation, dimensional variation, translation, rotation.And mainly have based on template matches in vehicle classification identification step
Recognition methods, the recognition methods based on statistical model, recognition methods based on neutral net etc..However, because vehicle class is numerous
More and difference is little, without obvious distinguishing characteristics;Influenceed by visible change, the automobile characteristic difference change of different angle
Greatly;Being influenceed by natural environment particularly illumination condition, serious illumination, which is reflected, causes vehicle wheel profile fuzzy, color deviation,
A series of reasons such as it is difficult to recognize, causes current vehicle recongnition technique can not adapt to the vehicle vehicle identification of complex environment, nothing
Method is realized for power-line patrolling.
The content of the invention
To solve the above problems, the present invention provides a kind of vehicle identification method available for power-line patrolling.
The technical solution adopted for the present invention to solve the technical problems is:A kind of vehicle identification side available for power-line patrolling
Method, including step:
S101. the video image for including vehicle image is obtained;
S102. the vehicle image in video image is detected;
S103. the vehicle image detected is alignd according to standard vehicle information;
S104. the vehicle image of alignment is subjected to pyramid change, while feature point extraction is carried out to vehicle image;
S105. the characteristic point of extraction is classified with Linear SVM, so as to identify the type of vehicle.
Further, the step S102 is specially:From the vehicle in hog+svm detection mode detection video image
Image.
Further, the step S103 includes:
A. the local binary feature of vehicle image is chosen;
B. vehicle image is alignd according to the local binary feature of vehicle image.
Further, the step S104 includes:
A. feature point detection is carried out with SIFT;
B. the dictionary of M word is included according to BOF thought structure;
C. vehicle image is divided into using image pyramid the characteristic bag of multiple yardsticks, then calculates and fall into each characteristic bag
The number of the word to belong to a different category, then the relation between vehicle image X and final matching degree Y be:
Histogram feature point under all levels, which is connected one dimension of composition, is:
Feature, the characteristic vector as classification.
Further, the step S105 includes:
A. by characteristic vector compared with the corresponding characteristic vector of vehicle sample in database, with acquisition and vehicle image
Vehicle sample in the immediate database;
B. the type of the vehicle sample in the database of acquisition determines the type of the vehicle.
Include after the step S105:According to the type of vehicle identified to causing transmission line equipment tripping operation, the tower that falls
The risk factor that accident generation is broken etc. external force carries out grade classification.
The present invention can be used for the advantages of vehicle identification method of power-line patrolling to be:Effectively solve such as noise, illumination to become
The problems such as change, dimensional variation, translation, rotation, the vehicle vehicle identification of complex environment is enabled adaptation to, realizes and is patrolled for electric power
Line.
Brief description of the drawings
Fig. 1 is the step flow chart for the vehicle identification method that the embodiment of the present invention one can be used for power-line patrolling.
Embodiment
The embodiment of the present invention is described in further detail below in conjunction with the accompanying drawings.
As shown in figure 1, a kind of vehicle identification method available for power-line patrolling, including:
Step S101. obtains the video image for including vehicle image.
In the present embodiment, the video image for including vehicle image is obtained by monitoring device, monitoring device is installed on electric power
Patrol UAV, monitoring device include camera and image pick-up card.The present embodiment is obtained by camera and image pick-up card
Video image, then the Computer Vision of specific region is analyzed by using computer, completes vehicle detection and vehicle
Classification and Identification.This method can accurately and effectively identify the vehicle of corresponding species, and accuracy rate is high, and rate of false alarm is low, greatly improves defeated
The video surveillance of electric line surrounding enviroment is horizontal.
Vehicle image in step S102. detection video images.
In this step, from the vehicle image in hog+svm detection mode detection video image.Including vehicle figure
In the video image of picture, vehicle is first found, then vehicle is positioned, pluck out by vehicle image after positioning.
Step S103. is alignd the vehicle image detected according to standard vehicle information.
This step includes:
A. the local binary feature of vehicle image is chosen.
B. vehicle image is alignd according to the local binary feature of vehicle image.
The present embodiment carries out vehicle detection to video image, generates a width bianry image(0 represents background, and 1 represents vehicle picture
Vegetarian refreshments), while morphology processing is carried out to bianry image, make vehicle movement region more complete.
After vehicle image is plucked out from video image, the local binary feature of vehicle image, Ran Hougen are first chosen
Vehicle image is alignd according to the local binary feature of vehicle image.
The vehicle image of alignment is carried out pyramid change by step S104., while carries out feature point extraction to vehicle image.
Pyramid is a kind of graphical representation mode for combining down-sampled operation and smooth operation.An its very big benefit
It is that each layer of pixel count is all constantly reduced from bottom to top, this can greatly reduce amount of calculation.
This step includes:
A. feature point detection is carried out with SIFT.
B. the dictionary of M word is included according to BOF thought structure.
C. vehicle image is divided into using image pyramid the characteristic bag of multiple yardsticks, then calculates and fall into each feature
The number of word to be belonged to a different category in bag, then the relation between vehicle image X and final matching degree Y be:
Histogram feature point under all levels, which is connected one dimension of composition, is:
Feature, the characteristic vector as classification.
The characteristic point of moving image in video image is determined by following steps:
A frame video image is read, the image is I(X, y), with Gaussian function G(X, y, σ)Convolution, obtain the image of metric space
L(X, y, σ);
L(X, y, σ)=G(X, y, σ)*I(X, y)
Wherein, G (x, y, σ)=12 π σ 2e- (x2+y2)/2 σ 2;
Between σ span is 0~20, the smoothness of its size decision image, the general picture feature of large scale correspondence image,
The minutia of small yardstick correspondence image.Big σ values correspond to coarse scale (low resolution), conversely, corresponding fine dimension (high score
Resolution).In order to effectively detect stable characteristic point in metric space, using Gaussian difference scale space, pass through following public affairs
Formula obtains the D in different scale space(X, y, σ);
D (x, y, σ)=(G (x, y, σ) * I (x, y))=L (x, y, k σ)-L (x, y, σ)
Wherein, k=1.414, σ=0~20 is divided according to 10 grades point, the image of 10 difference Gaussian scale-spaces can be obtained.Can
Regard 10 I of 10 stackings as(x、y)Image.
In the image of the yardstick of current layer, if a pixel in 8 neighborhoods of this layer and two adjacent layer,
The maximum or minimum value of the response of Dog operators, then the point is a characteristic point under the yardstick.Wherein, in 8 neighborhoods
Pixel, is divided into upper and lower two adjacent layers, every layer totally 9 pixels, the pixel of this layer do not include itself, common 9+9+8=26
Individual pixel.
This step also includes, and removes unsuitable characteristic point.
By being fitted three-dimensional quadratic function accurately to determine the position of characteristic point and yardstick(Reach sub-pixel precision), simultaneously
The characteristic point and unstable skirt response point for removing low contrast (can produce stronger edge to ring because of difference of Gaussian
Should), with enhancing matching stability, improve noise resisting ability.
This step also includes, and establishes description for each characteristic point, is matched by the information for describing son.Establish
The process for describing son is as follows:According to the gradient of identified this feature point under each yardstick and its pixel in surrounding neighbors,
Establish histogram of gradients;Wherein, surrounding neighbors are rectangle, and 16*16 pixel may be selected in pixel, and characteristic point is in rectangular pixels
The center of point.The histogram of gradients includes multiple different angle sections, and each angular interval is multiple gradients in the section
The sum of the mould length of interior pixel.
The each pixel L long m of mould and angle, θ is determined by equation below:
m(x,y)=(Lx+1,y-L(x-1,y))2+(L(x,y+1)-L(x,y-1))2]]>
θ(x,y)=arctan2L(x,y+1)-L(x,y-1)L(x+1,y)-L(x-1,y)]]>
Using an angle in angular range corresponding to the maximum norm of gradient in the histogram of gradients as principal direction;Such as:
Mould in the range of 30~40 degree and to be maximum in all angles scope, end points or intermediate point may be selected as principal direction,
Such as 30 degree or 35 degree be used as principal direction.With the Gaussian function that a center is entreated in this region to each in the surrounding neighbors
The mould weighting of the gradient of individual pixel;In weighting procedure, σ=1.5 are taken;
G(x,y,σ)=12πσ2e-(x2+y2)/2σ2]]>
Each pixel in the surrounding neighbors is divided into multiple blocks, after the weighting of the pixel in each block
Mould is long, relative to the angle difference of the principal direction, establishes histogram of gradients, determines the vector of the block;By 16*16 pixel
Point, using 4*4 pixel as a block, it is divided into 16 blocks, statistical gradient histogram, histogram are pressed in this block of cells
45 degree of divisions of irradiation angle value, altogether including 360 degree/45 degree=8 Direction intervals;Contain in the region that so whole description is covered
Information be exactly 16 × 8=128.
The vector information that whole blocks are recorded with the form of multi-C vector forms characteristic vector, the description as this feature point
Son.Whole description can be regarded as the vector of one 128 dimension, i.e. characteristic vector.
This step also includes, and remove influences caused by illumination variation.If illumination variation is contrast change, equivalent to
It is that a constant has been multiplied by the gradient of each point, then this constant is just eliminated after standardization;If illumination variation is
The change of brightness, then add a constant relative to the pixel value to each point, the no any influence of change on gradient.
But because some nonlinear illumination variations can make the gradient modulus value of some pixels produce large change, while to gradient direction
Do not influence, therefore all gradient modulus value more than some threshold value are all set to this threshold by us in statistical gradient histogram
Value, it is possible to reduce the influence of illumination variation.
Step S105. classifies with Linear SVM to the characteristic point of extraction, so as to identify the type of vehicle.
This step includes:
A. by characteristic vector compared with the corresponding characteristic vector of vehicle sample in database, with acquisition and vehicle image
Vehicle sample in the immediate database;
B. the type of the vehicle sample in the database of acquisition determines the type of the vehicle.
The present invention has found that nonlinear SVM has found mesh when classifying to extensive target in experimentation
Scalar functions convergence is too slow, and overall operation speed is excessively slow, therefore the present embodiment chooses linear liblinear svm as the present invention
Grader.SVM methods are by a Nonlinear Mapping p, and sample space is mapped to a higher-dimension or even infinite dimensional spy
Levy in space(Hilbert spaces)So that it is converted into original sample space the problem of Nonlinear separability in feature space
In linear separability the problem of.Traditional svm is primarily used to carry out the generation of nonlinear s vm graders, and liblinear
SVM mainly tackles large-scale data sorting task, because the training of linear classifier is than the training meter of Nonlinear Classifier
It is much lower to calculate complexity, the time is also few a lot, and the performance on large-scale data and nonlinear classifier performance phase
When the large-scale data training faced so as to solve us is difficult to convergent problem.
Include after the step S105:According to the type of vehicle identified to causing transmission line equipment tripping operation, the tower that falls
The risk factor that accident generation is broken etc. external force carries out grade classification.
Then the present embodiment detects the vehicle image in video image by obtaining the video image for including vehicle image,
The vehicle image detected is alignd according to standard vehicle information again, the vehicle image of alignment is subjected to pyramid change,
Feature point extraction is carried out to vehicle image simultaneously, finally the characteristic point of extraction classified with Linear SVM, so as to identify
The type of vehicle.The present embodiment can effectively solve the problem that the problems such as such as noise, illumination variation, dimensional variation, translation, rotation, from
And the vehicle vehicle identification of complex environment is can adapt to, realize for power-line patrolling.
Claims (6)
1. a kind of vehicle identification method available for power-line patrolling, it is characterised in that including step:
S101. the video image for including vehicle image is obtained;
S102. the vehicle image in video image is detected;
S103. the vehicle image detected is alignd according to standard vehicle information;
S104. the vehicle image of alignment is subjected to pyramid change, while feature point extraction is carried out to vehicle image;
S105. the characteristic point of extraction is classified with Linear SVM, so as to identify the type of vehicle.
2. the vehicle identification method according to claim 1 available for power-line patrolling, it is characterised in that the step S102
Specially:From the vehicle image in hog+svm detection mode detection video image.
3. the vehicle identification method according to claim 1 available for power-line patrolling, it is characterised in that the step S103
Including:
A. the local binary feature of vehicle image is chosen;
B. vehicle image is alignd according to the local binary feature of vehicle image.
4. the vehicle identification method according to claim 1 available for power-line patrolling, it is characterised in that the step S104
Including:
A. feature point detection is carried out with SIFT;
B. the dictionary of M word is included according to BOF thought structure;
C. vehicle image is divided into using image pyramid the characteristic bag of multiple yardsticks, then calculates and fall into each characteristic bag
The number of the word to belong to a different category, then the relation between vehicle image X and final matching degree Y be:
Histogram feature point under all levels, which is connected one dimension of composition, is:
Feature, the characteristic vector as classification.
5. the vehicle identification method according to claim 1 available for power-line patrolling, it is characterised in that the step S105
Including:
A. by characteristic vector compared with the corresponding characteristic vector of vehicle sample in database, with acquisition and vehicle image
Vehicle sample in the immediate database;
B. the type of the vehicle sample in the database of acquisition determines the type of the vehicle.
6. the vehicle identification method according to claim 1 available for power-line patrolling, it is characterised in that the step S105
Include afterwards:According to the type of vehicle identified to causing transmission line equipment tripping operation, the external force such as tower of falling to break the danger that accident occurs
Dangerous degree carries out grade classification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610490262.XA CN107545216A (en) | 2016-06-29 | 2016-06-29 | A kind of vehicle identification method available for power-line patrolling |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610490262.XA CN107545216A (en) | 2016-06-29 | 2016-06-29 | A kind of vehicle identification method available for power-line patrolling |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107545216A true CN107545216A (en) | 2018-01-05 |
Family
ID=60963053
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610490262.XA Pending CN107545216A (en) | 2016-06-29 | 2016-06-29 | A kind of vehicle identification method available for power-line patrolling |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107545216A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108648187A (en) * | 2018-05-14 | 2018-10-12 | 南方医科大学 | A kind of sorting technique based on depth characteristic bag |
CN110672950A (en) * | 2019-10-08 | 2020-01-10 | 深圳海岸语音技术有限公司 | Power equipment fault sound image detection system and method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103488973A (en) * | 2013-09-12 | 2014-01-01 | 上海依图网络科技有限公司 | Method and system for recognizing vehicle brand based on image |
CN103839074A (en) * | 2014-02-24 | 2014-06-04 | 西安电子科技大学 | Image classification method based on matching of sketch line segment information and space pyramid |
CN105184286A (en) * | 2015-10-20 | 2015-12-23 | 深圳市华尊科技股份有限公司 | Vehicle detection method and detection device |
-
2016
- 2016-06-29 CN CN201610490262.XA patent/CN107545216A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103488973A (en) * | 2013-09-12 | 2014-01-01 | 上海依图网络科技有限公司 | Method and system for recognizing vehicle brand based on image |
CN103839074A (en) * | 2014-02-24 | 2014-06-04 | 西安电子科技大学 | Image classification method based on matching of sketch line segment information and space pyramid |
CN105184286A (en) * | 2015-10-20 | 2015-12-23 | 深圳市华尊科技股份有限公司 | Vehicle detection method and detection device |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108648187A (en) * | 2018-05-14 | 2018-10-12 | 南方医科大学 | A kind of sorting technique based on depth characteristic bag |
CN108648187B (en) * | 2018-05-14 | 2021-06-15 | 南方医科大学 | Depth feature bag based classification method |
CN110672950A (en) * | 2019-10-08 | 2020-01-10 | 深圳海岸语音技术有限公司 | Power equipment fault sound image detection system and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106056086B (en) | Vehicle brand type identifier method based on Fast Learning frame | |
CN103699905B (en) | Method and device for positioning license plate | |
CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
CN103258432A (en) | Traffic accident automatic identification processing method and system based on videos | |
CN102622607A (en) | Remote sensing image classification method based on multi-feature fusion | |
CN104268528A (en) | Method and device for detecting crowd gathered region | |
CN109635733B (en) | Parking lot and vehicle target detection method based on visual saliency and queue correction | |
CN103279738B (en) | Automatic identification method and system for vehicle logo | |
CN106778633B (en) | Pedestrian identification method based on region segmentation | |
CN108960055A (en) | A kind of method for detecting lane lines based on local line's stage mode feature | |
CN105868734A (en) | Power transmission line large-scale construction vehicle recognition method based on BOW image representation model | |
CN109670449B (en) | Vehicle illegal judgment method based on vertical snapshot mode | |
CN111915583A (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
CN106778540A (en) | Parking detection is accurately based on the parking event detecting method of background double layer | |
Mammeri et al. | North-American speed limit sign detection and recognition for smart cars | |
Zheng et al. | A review of remote sensing image object detection algorithms based on deep learning | |
CN110321855A (en) | A kind of greasy weather detection prior-warning device | |
Baluyan et al. | Novel approach for rooftop detection using support vector machine | |
CN110348307B (en) | Path edge identification method and system for crane metal structure climbing robot | |
CN113158954B (en) | Automatic detection method for zebra crossing region based on AI technology in traffic offsite | |
CN117197700B (en) | Intelligent unmanned inspection contact net defect identification system | |
CN107545216A (en) | A kind of vehicle identification method available for power-line patrolling | |
CN110188607A (en) | A kind of the traffic video object detection method and device of multithreads computing | |
CN106407904A (en) | Bang zone determining method and device | |
CN115937761A (en) | Subway flood-proof door state monitoring method based on image recognition |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180105 |