CN110378865A - A kind of greasy weather visibility intelligence hierarchical identification method and system under complex background - Google Patents
A kind of greasy weather visibility intelligence hierarchical identification method and system under complex background Download PDFInfo
- Publication number
- CN110378865A CN110378865A CN201910351794.9A CN201910351794A CN110378865A CN 110378865 A CN110378865 A CN 110378865A CN 201910351794 A CN201910351794 A CN 201910351794A CN 110378865 A CN110378865 A CN 110378865A
- Authority
- CN
- China
- Prior art keywords
- greasy weather
- visibility
- image
- formula
- follows
- 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
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 28
- 238000002834 transmittance Methods 0.000 claims abstract description 27
- 238000013480 data collection Methods 0.000 claims abstract description 23
- 238000001914 filtration Methods 0.000 claims abstract description 16
- 238000002372 labelling Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 35
- 239000003595 mist Substances 0.000 claims description 22
- 238000007781 pre-processing Methods 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 4
- 239000003897 fog Substances 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000009738 saturating Methods 0.000 claims description 3
- 238000011160 research Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000013316 zoning Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- JXSJBGJIGXNWCI-UHFFFAOYSA-N diethyl 2-[(dimethoxyphosphorothioyl)thio]succinate Chemical compound CCOC(=O)CC(SP(=S)(OC)OC)C(=O)OCC JXSJBGJIGXNWCI-UHFFFAOYSA-N 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 239000000047 product Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30192—Weather; Meteorology
Abstract
The present invention discloses the greasy weather visibility recognition methods and system under a kind of complex background, this method comprises: obtaining true greasy weather image data collection；End-to-end convolutional neural networks model is established, input greasy weather image data collection is trained and region labeling, transmittance figure is exported；Transmittance figure is optimized with guiding filtering method, obtains fining transmittance figure；Atmospheric scattering coefficient is estimated according to transmittance figure；Region Zenith Distance scattering coefficient is sought, grade identification is carried out to greasy weather picture.This method is suitable for the greasy weather visibility identification under complex background, such as weather station, highway, tourist attraction, alpine region, river with complicated field, artificial judgement by hand is not needed to a certain extent, break the traditional artificial limitation observed and need the expensive hardwares equipment such as visibility meter, compared to expressway foggy-dog visibility discrimination, the present invention is applied widely.
Description
Technical field
The present invention relates to the greasy weather visibility intelligence hierarchical identification method and system under a kind of complex background, belonging to the greasy weather is taken the photograph
As identification technology field.
Background technique
Currently, image defogging problem achieves many research achievements in scientific research circle and industry, and it is applied to highway
Video monitoring, the related fieldss such as take photo by plane.And under the conditions of the true greasy weather, atrocious weather situation will cause various safety in life
Hidden danger is related to the security of the lives and property of the people, if relevant weather department can accurately issue corresponding greasy weather visibility
Situation can help all trades and professions to improve management quality.At this stage, meteorological department is mainly base for the observation of dense fog visibility
In artificial observation, greasy weather situation is judged by arranging special observation station in each website.Greasy weather etc. is judged using ocular estimate
Grade has stronger subjectivity；It is smaller using laser range finder measurement range, it is not suitable for various complex scenes.Therefore, how
It can accurately estimate that greasy weather visibility scale identifies under cost limited circumstances, be to improve effectively arranging for meteorological department's working efficiency
It applies.According to the standard of weather station, greasy weather grade can be divided into mist, mist, dense fog, thick fog, strong thick fog according to visibility
Five grades.In the prior art, there are two main classes for greasy weather visibility method of discrimination: first kind method is based on image space feature
Foggy day detection method, such method needs a large amount of threshold decision, carries out adjusting thresholds according to different sections of highway information and obtained
Preferable effect has poor robustness.Second class method is the foggy day detection method based on dark channel prior, such method
It needs to calculate image transmission rate, further obtains corresponding visibility value, the image transmission rate that this method obtains is not fine enough, calculates
Amount is big, will affect the accuracy of visibility value.
Above-mentioned greasy weather visibility method of discrimination is mainly used in highway problem, and the image background of acquisition is more single, greatly
Mostly studied by starting point of traffic route.And under full-scale condition, greasy weather scene location be all it is variable, it is existing in order to solve
Deficiency in technology, the present invention propose a kind of greasy weather visibility recognition methods for any greasy weather scene.In recent years, with depth
The fast development of learning network is spent, especially convolutional neural networks is increasingly mature, is substantially better than some biographies to the processing of image
System method.Convolutional neural networks in deep learning have very high accuracy and validity for image characteristics extraction, at present
The research determined both at home and abroad using the method for deep learning greasy weather grade is less, needs a large amount of greasy weather image data collection, and
And the prior art is all made of front-end collection, the technical solution of rear end identification.Meanwhile automatic greasy weather grade determines also to become meteorology portion
One research emphasis of door, herein under the premise of, in conjunction with existing research condition, under the complex background for proposing a kind of front-end collection identification
Greasy weather visibility recognition methods and system.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, to provide the complicated back of a kind of front-end collection and identification
Greasy weather visibility intelligence hierarchical identification method and system under scape.
Invention adopts the following technical scheme that realize: greasy weather image data collection is acquired using front end intelligent identification device, by
Backstage image identification system training establishes end-to-end convolutional neural networks model, then model is implanted into front end intelligent identification device,
Intelligent grade is directly carried out to greasy weather picture by front end intelligent identification device to identify and export；Specifically includes the following steps:
Step 1: set five white marking points in road surface region, from as far as person of modern times be choose four sections of line segmentation sections, as
Target area acquires a large amount of true greasy weather image data collection using digital camera, establish data set；
Step 2: establishing end-to-end convolutional neural networks model, obtained according to the picture that step 1 inputs using dark channel prior
Rate calculation formula is penetrated, data set input network is trained, while region labeling is carried out to greasy weather image data collection, output is saturating
Penetrate rate figure；
Step 3: filtering optimization is guided to the transmittance figure that step 2 trains；
Step 4: atmospheric scattering coefficient is estimated according to the transmittance figure that step 3 obtains；
Step 5: region Zenith Distance scattering coefficient being asked according to step 4, grade identification is carried out to greasy weather picture.
Further, in the step 1, the method for establishing greasy weather image data collection is as follows:
Step 1-1: setting five white marking points in road surface region, from being to choose four sections of line segmentation sections as far as person of modern times, makees
For target area, with digital camera shooting with the greasy weather picture of various concentration under scene, be made into the form of file training and
Test data set, and the picture in each file is marked into grade.
Step 1-2: using file as label, picture demarcates data set as input data.
Further, in the step 2, the method based on dark channel prior estimation transmissivity is as follows:
Input arbitrary image J, the value description of each pixel of dark channel image are as follows:
(1) in formula, JcFor the corresponding Color Channel of original image, Ω (x) is a partial x centered region, JD Lin rkTo help secretly
The corresponding pixel value of road image.
For fog free images, dark description are as follows:
Atmospherical scattering model is simplified shown as:
I (x)=J (x) t (x)+A (1-t (x)) (3)
(3) in formula, I (x) is foggy image, and J (x) is fog free images, and t (x) is transmissivity, reacts the depth letter of scene objects
Breath, A are atmosphere light intensity value, indicate the overall strength of various light in ambient enviroment.
(3) formula both sides are deformed simultaneously are as follows:
In conjunction with (1) formula, (4) formula is simplified are as follows:
Obtain the calculated transmissivity formula of dark channel prior of foggy image.
Further, in the step 2, end-to-end convolutional neural networks model is described as follows:
Network is mainly made of 5 hidden layers, respectively convolutional layer and pond layer, wherein 4 layers are passed through up-sampling and pondization series connection shape
It is exported at bypass, finally exports last prediction result by a convolutional layer.The design of end-to-end convolutional network is mainly used for
Training greasy weather picture, carries out image characteristics extraction.
It mainly include 3 convolutional layers and 2 pond layers in end-to-end convolutional neural networks.Convolutional layer uses 3 × 3 volume
Each layer of different sizes are connected Layer1 to Layer4 by product core by the unified jump of method of up-sampling and maximum pond
It connects, forms series connection feature finally by ReLu activation primitive and export characteristic pattern.
Further, in the step 2, the original greasy weather image data collection that step 1 is obtained is as input, by (5) formula to volume
The training of product neural network obtains transmittance figure as output result.
Further, in the step 2, the method for demarcating scene areas is as follows:
It mainly include road surface and sky two parts for any greasy weather picture, it is bigger that range is blocked on the denseer place road surface of fog, and
Room and time inefficiency is calculated to whole picture, therefore, for greasy weather scene, target area division is carried out, to detect mist
Gas concentration.
Step 2-1: estimation camera parameter.For the greasy weather picture of camera shooting, there are image coordinates and road surface coordinate
Mapping relations need to calculate the upright position v on camera horizontal linehWith camera parameter λ, meanwhile, need to be arranged pavement markers point.
Upright position v on camera horizontal linehCalculation formula are as follows:
Wherein, y1And y2The respectively vertical coordinate of pavement markers point, d1And d2Respectively actual distance value.
Camera parameter λ calculation formula are as follows:
Wherein, v1And v2For the ordinate of mark point in the picture.
Step 2-2: calibration scene target area.Divide scene objects region, it is necessary first to pavement markers point is set, according to
The Misty Image that step 1 is shot artificially chooses four sections of line segmentation sections, as target area, so as to effectively avoid all kinds of block
Object.
Further, as follows to transmittance figure optimization method using guiding filtering in the step 3:
The description of guiding filtering expression formula are as follows:
ti=∑jWij(I)*tj (8)
(8) in formula, ti is filtering output image, and I is navigational figure, tjFor input picture, WijFor weighting function, indicate are as follows:
(9) in formula, | ω | it is filter window ωkThe pixel quantity for including, μkAnd σkIt is navigational figure I in filter window ωkIt is equal
Value and variance size, ε is smoothing factor.
For the transmissivity of Misty Image, the transmissivity that step 2 is trained is as input picture t to be filteredj, will be former
Beginning foggy image is as navigational figure I, and simultaneous (8) (9) formula, acquiring filtering output image is the transmittance figure t after optimizingi。
Further, in the step 4, the method for calculating atmospheric scattering coefficient is as follows:
Transmissivity and atmospheric scattering coefficient meet following relationship:
T (x)=e-βd(x) (10)
(10) in formula, d is scene point to the distance of observation point, as scene depth；β is atmospheric scattering coefficient, indicates atmosphere pair
The scattering power of light；T is transmissivity, reacts the depth information of scene objects.
Using the digital camera of step 1, obtain target the distance between to observation point, i.e. depth of field distance, according to formula
(10) in the scene of zoning target point atmospheric scattering coefficient.
Further, in the step 5, the method for calculating Zenith Distance scattering coefficient is as follows:
It averages to all atmospheric scattering coefficients that step 4 acquires, obtains the Zenith Distance scattering coefficient in region scene, with
Guarantee that the accuracy of identification greasy weather grade, calculation formula are as follows:
Further, as follows to greasy weather photo grade method of discrimination in the step 5:
According to the atmospheric scattering coefficient of step 5, definition in conjunction with meteorology to visibility has following formula:
5 mark points are provided for any Misty Image according to formula (12), compared between judge distance and formula distance
Error.
The present invention provides the greasy weather visibility identifying system under a kind of complex background, mainly by digital camera, embedded
Equipment, server apparatus and display equipment composition.Including front end intelligent identification device and backstage image identification system；Front end intelligence
Identification device is made of digital camera, analysis hardware and embedded device, is mainly used for obtaining greasy weather image data；Scheme on backstage
As identifying system be mainly used for identify image gradation, by server apparatus and display equipment form, server apparatus training convolutional
Neural network exports photo grade to identify various greasy weather pictures on the display device；Digital camera is successively hard with analysis
Part is connected with embedded device, and embedded device is connect with display equipment respectively with server apparatus；
The embedded device mainly includes data acquisition module and preprocessing module, data acquisition module and preprocessing module
Connection, preprocessing module are connect with display equipment；Data acquisition module is for acquiring a large amount of greasy weather pictures；Preprocessing module is used for
The greasy weather picture that digital camera acquires is made into trained and test data set, and grade mark and label are carried out to every picture
Processing；
The analysis hardware is connect with digital camera by cable, for analyzing the photo of acquisition；
The server apparatus mainly includes training module, computing module and categorization module；Computing module respectively with training mould
Block is connected with categorization module, and training module is connect with preprocessing module, and categorization module is connect with display equipment；Training module is used for
Training convolutional neural networks model inputs greasy weather image data collection, exports transmittance figure；Computing module is for optimizing training module
The transmittance figure of output, and calculate atmospheric scattering coefficient；Categorization module according to computing module as a result, identify strong thick fog, thick fog,
Five dense fog, mist, mist grades；
The display equipment is mainly used for exporting that preprocessing module and categorization module obtain as a result, for monitoring number in real time
According to.
Beneficial effects of the present invention are as follows: 1, the method for the present invention is handling greasy weather image data collection with deep learning method
On the basis of, it proposes a kind of greasy weather visibility recognition methods and system based under complex background, passes through convolutional neural networks logarithm
According to the training of collection, picture feature information can be effectively extracted, to improve greasy weather grade identification precision.2, it is suitable for complicated back
Greasy weather visibility identification under scape, such as weather station, highway, tourist attraction, alpine region, river with complicated field,
Artificial judgement by hand is not needed to a certain extent, breaks the traditional artificial office observed and need the expensive hardwares equipment such as visibility meter
Sex-limited, compared to expressway foggy-dog visibility discrimination, the method for the present invention is applied widely.3, system hardware and software combines,
Relevant departments' mapping out the work in observation point is considered, using remote server real-time monitoring data, guarantees the permanent of data.
Detailed description of the invention
Fig. 1 is overall flow figure.
Fig. 2 is convolutional neural networks structure chart.
Fig. 3 is system module frame diagram.
Specific embodiment
The present invention acquires greasy weather image data collection using front end intelligent identification device, trained by backstage image identification system,
End-to-end convolutional neural networks model is established, then model is implanted into front end intelligent identification device, by front end intelligent identification device pair
Greasy weather picture directly carries out intelligent grade and identifies and export；After backstage image identification system trains model, model is placed in
Among the intelligent identification device of front end, backstage image identification system is no longer participate in identification process, only front end intelligent identification device
Just having identification function belongs to smart machine, in order to make the object, technical solutions and advantages of the present invention definitely, in conjunction with attached drawing
And invention is further described in detail for specific embodiment, but the present invention is not limited to this.
The present invention includes the following steps, as shown in Figure 1:
Step 1 sets five white marking points in road surface region, from as far as person of modern times be choose four sections of line segmentation sections, as
Target area acquires greasy weather image data collection using digital camera, establishes data set；
Step 2 establishes end-to-end convolutional neural networks model, is obtained according to the picture that step 1 inputs using dark channel prior
Rate calculation formula is penetrated, data set input network is trained, while region labeling is carried out to greasy weather image data collection, output is saturating
Penetrate rate figure；
Step 3 guides filtering optimization to the transmittance figure that step 2 trains, and obtains fining transmittance figure；
Step 4 estimates atmospheric scattering coefficient according to the transmittance figure of step 3；
Step 5 seeks region Zenith Distance scattering coefficient according to step 4, carries out grade identification to greasy weather picture.
The method that image data collection is established in the step 1 is as follows:
Step 1-1, five white marking points are set in road surface region, from being to choose four sections of line segmentation sections as far as person of modern times, made
For target area, with digital camera shooting with the greasy weather picture of various concentration under scene, be made into the form of file training and
Test data set, and the picture in each file is marked into grade；By greasy weather grade be divided into strong thick fog, thick fog, dense fog, mist,
Five grades of mist choose the visibility of calibration point and divide greasy weather grade by calculating, visibility second mark point with
Between first mark point, distance is strong thick fog less than 50 meters；Visibility third mark point and second mark point it
Between, distance is thick fog between 50 to 200 meters；At the 4th between mark point and third mark point, distance exists visibility
It is dense fog between 200 to 500 meters；Visibility between the 5th mark point and the 4th mark point, distance less than 1000 meters,
For mist；Visibility is in the 5th mark point, and distance is in 1 km between 10 kms, being mist；
Step 1-2, using file as label, picture demarcates data set as input data.
In the step 2, the method based on dark channel prior estimation transmissivity is as follows:
Input arbitrary image J, the value description of each pixel of dark channel image are as follows:
In formula, JcFor the corresponding Color Channel of original image, Ω (x) is a partial x centered region, JdarkFor dark channel diagram
As corresponding pixel value；
For fog free images, dark description are as follows:
Atmospherical scattering model is simplified shown as:
I (x)=J (x) t (x)+A (1-t (x)) (3)
In formula, I (x) is foggy image, and J (x) is fog free images, and t (x) is transmissivity, reacts the depth information of scene objects, A
For atmosphere light intensity value, the overall strength of various light in ambient enviroment is indicated；
(3) formula both sides are deformed simultaneously are as follows:
In conjunction with (1) formula, (4) formula is simplified are as follows:
Obtain the calculated transmissivity formula of dark channel prior of foggy image.
In the step 2, end-to-end convolutional neural networks are described as follows:
Convolutional neural networks model is mainly made of convolutional layer and pond layer, wherein 4 layers form by up-sampling and pondization series connection
Bypass output, finally exports last prediction result by a convolutional layer.The design of end-to-end convolutional network is mainly used for instructing
Practice greasy weather picture, carries out image characteristics extraction；
In the step 2, the method for training convolutional neural networks is as follows:
As inputting, the training by (5) formula to convolutional neural networks obtains the original greasy weather image data collection that step 1 is obtained
To transmittance figure as output result.
In the step 2, the method for demarcating scene areas is as follows:
Step 2-1, estimate camera parameter.For the greasy weather picture of camera shooting, there are the mappings of image coordinate and road surface coordinate
Relationship needs to calculate the upright position v on camera horizontal linehWith camera parameter λ, meanwhile, need to be arranged pavement markers point.
Upright position v on camera horizontal linehCalculation formula are as follows:
Wherein, y1And y2The respectively vertical coordinate of pavement markers point, d1And d2Respectively actual distance value.
Camera parameter λ calculation formula are as follows:
Wherein, v1And v2For the ordinate of mark point in the picture.
Step 2-2, scene target area is demarcated.Divide scene objects region, it is necessary first to pavement markers point is set, according to
The Misty Image that step 1 is shot artificially chooses four sections of line segmentation sections, as target area, so as to effectively avoid all kinds of block
Object.
It is as follows to transmittance figure optimization method using guiding filtering in the step 3:
The description of guiding filtering expression formula are as follows:
ti=∑jWij(I)*tj (8)
(6) in formula, tiImage is exported for filtering, I is navigational figure, tjFor input picture, WijFor weighting function, indicate are as follows:
(7) in formula, | ω | it is filter window ωkThe pixel quantity for including, μkAnd σkIt is navigational figure I in filter window ωkIt is equal
Value and variance size, ε is smoothing factor.
For the transmissivity of Misty Image, the transmissivity that step 2 is trained is as input picture t to be filteredj, will be former
Beginning foggy image is as navigational figure I, and simultaneous (8) (9) formula, acquiring filtering output image is the transmittance figure t after optimizingi。
In the step 4, the method for calculating atmospheric scattering coefficient is as follows:
Transmissivity and atmospheric scattering coefficient meet following relationship:
T (x)=e-βd(x) (10)
(10) in formula, d is scene point to the distance of observation point, as scene depth；β is atmospheric scattering coefficient, indicates atmosphere pair
The scattering power of light；T is transmissivity, reacts the depth information of scene objects.
Using the digital camera of step 1, obtain target the distance between to observation point, i.e. depth of field distance, according to formula
(10) in the scene of zoning target point atmospheric scattering coefficient.
In the step 5, the method for calculating Zenith Distance scattering coefficient is as follows:
It averages to atmospheric scattering coefficients all in step 4, obtains the Zenith Distance scattering coefficient of region scene, to guarantee to know
The accuracy of other greasy weather grade, calculation formula are as follows:
It is as follows to greasy weather photo grade method of discrimination in the step 5:
According to Zenith Distance scattering coefficient, definition in conjunction with meteorology to visibility has following formula:
5 mark points are provided for any Misty Image according to formula (12), compared between judge distance and formula distance
Error.Calibrated error between sector is as shown in table 1:
| Mark point | Judge distance (m) | Formula distance (m) | Relative error |
| 1 | 36 | 36.14 | 0.29% |
| 2 | 200 | 198.64 | 1.21% |
| 3 | 380 | 367.28 | 2.41% |
| 4 | 520 | 530.12 | 5.83% |
| 5 | 700 | 686.16 | 6.62% |
Greasy weather grade is divided into five strong thick fog, thick fog, dense fog, mist, mist grades by the present invention, by calculating the calibration point chosen
Visibility divide greasy weather grade, greasy weather grade classification table is as shown in table 2:
| Label | Target area | Visibility value | Greasy weather grade |
| 1 | [second mark point, first mark point] | Less than 50 meters | Strong thick fog |
| 2 | [second mark point of third mark point] | Between 50-200 meters | Thick fog |
| 3 | [the 4th mark point third mark point] | Between 200-500 meters | Dense fog |
| 4 | [the 4th mark point of the 5th mark point] | Less than 1000 meters | Mist |
| 5 | [the 5th mark point] | Between 1km-10km | Mist |
As shown in figure 3, the present invention includes front end intelligent identification device and backstage image identification system；Intelligently known using front end
Other device acquires greasy weather image data collection, by the training of backstage image identification system, establishes end-to-end convolutional neural networks model, then
Model is implanted into front end intelligent identification device, directly carries out intelligent grade identification to greasy weather picture by front end intelligent identification device simultaneously
Output；
Front end intelligent identification device is made of digital camera 4, analysis hardware 5 and embedded device 1, is mainly used for obtaining the greasy weather
Image data；Backstage image identification system is mainly used for identifying image gradation, is made of, takes server apparatus 2 and display equipment 3
Business 2 training convolutional neural networks of device equipment export photo grade in display equipment 3 to identify various greasy weather pictures；Number
Camera 4 is successively connect with analysis hardware 5 and embedded device 1, and embedded device 1 is set with display respectively with server apparatus 2
Standby 3 connection；Embedded device 1 uses NVIDIA embedded board Jetson serial, is mainly used for obtaining greasy weather picture
Data；
The embedded device 1 mainly includes data acquisition module 11 and preprocessing module 12, data acquisition module 11 and pre-
Processing module 12 connects, and preprocessing module 12 is connect with display equipment 3；Data acquisition module 11 is for acquiring a large amount of greasy weather figures
Piece；Preprocessing module 12 is used to for the greasy weather picture that digital camera acquires being made into trained and test data set, and schemes to every
Piece carries out grade mark and tag processes；
The analysis hardware 5 is connect with digital camera 4 by cable, for analyzing the photo of acquisition；
The server apparatus 2 mainly includes training module 21, computing module 22 and categorization module 23；Computing module 22 is distinguished
It is connect with training module 21 and categorization module 23, training module 21 is connect with preprocessing module 12, and categorization module 23 is set with display
Standby 3 connection；Training module 21 is used for training convolutional neural networks model, inputs greasy weather image data collection, exports transmittance figure；Meter
The transmittance figure that module 22 is used to optimize training module output is calculated, and calculates atmospheric scattering coefficient；Categorization module 23 is according to calculating
Module as a result, identifying five strong thick fog, thick fog, dense fog, mist, mist grades；
The display equipment 3 is mainly used for exporting that preprocessing module 12 and categorization module 23 obtain as a result, for supervising in real time
Control data.
Claims (10)
1. a kind of greasy weather visibility intelligence hierarchical identification method under complex background, it is characterised in that: utilize front end intelligent recognition
Device acquires greasy weather image data collection, by the training of backstage image identification system, establishes end-to-end convolutional neural networks model, then will
Model is implanted into front end intelligent identification device, directly carries out intelligent grade to greasy weather picture by front end intelligent identification device and identifies and defeated
Out；
Specifically includes the following steps:
Step 1 sets five white marking points in road surface region, from as far as person of modern times be choose four sections of line segmentation sections, as
Target area acquires greasy weather image data collection using digital camera, establishes data set；
Step 2 establishes end-to-end convolutional neural networks model, is obtained according to the picture that step 1 inputs using dark channel prior
Rate calculation formula is penetrated, data set input network is trained, while region labeling is carried out to greasy weather image data collection, output is saturating
Penetrate rate figure；
Step 3 guides filtering optimization to the transmittance figure that step 2 trains, and obtains fining transmittance figure；
Step 4 estimates atmospheric scattering coefficient according to the transmittance figure of step 3；
Step 5 seeks region Zenith Distance scattering coefficient according to step 4, carries out grade identification to greasy weather picture.
2. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In the method for establishing image data collection in the step 1 is as follows:
Step 1-1, five white marking points are set in road surface region, from being to choose four sections of line segmentation sections as far as person of modern times, made
For target area, with digital camera shooting with the greasy weather picture of various concentration under scene, be made into the form of file training and
Test data set, and the picture in each file is marked into grade；By greasy weather grade be divided into strong thick fog, thick fog, dense fog, mist,
Five grades of mist choose the visibility of calibration point and divide greasy weather grade by calculating, visibility second mark point with
Between first mark point, distance is strong thick fog less than 50 meters；Visibility third mark point and second mark point it
Between, distance is thick fog between 50 to 200 meters；At the 4th between mark point and third mark point, distance exists visibility
It is dense fog between 200 to 500 meters；Visibility between the 5th mark point and the 4th mark point, distance less than 1000 meters,
For mist；Visibility is in the 5th mark point, and distance is in 1 km between 10 kms, being mist；
Step 1-2, using file as label, picture demarcates data set as input data.
3. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In in the step 2, the method based on dark channel prior estimation transmissivity is as follows:
Arbitrary image J is inputted, the value of each pixel of dark channel image is described as
In formula, JcFor the corresponding Color Channel of original image, Ω (x) is a partial x centered region, JdarkFor dark channel diagram
As corresponding pixel value；
For fog free images, dark is described as
Atmospherical scattering model is simplified shown as:
I (x)=J (x) t (x)+A (1-t (x)) (3)
In formula, I (x) is foggy image, and J (x) is fog free images, and t (x) is transmissivity, reacts the depth information of scene objects, A
For atmosphere light intensity value, the overall strength of various light in ambient enviroment is indicated；
(3) formula both sides are deformed simultaneously are as follows:
In conjunction with (1) formula, (4) formula is simplified are as follows:
Obtain the calculated transmissivity formula of dark channel prior of foggy image.
4. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In in the step 2, end-to-end convolutional neural networks are described as follows:
Convolutional neural networks model is mainly made of convolutional layer and pond layer, wherein 4 layers form by up-sampling and pondization series connection
Bypass output, finally exports last prediction result by a convolutional layer；The design of end-to-end convolutional network is mainly used for instructing
Practice greasy weather picture, carries out image characteristics extraction；
In the step 2, the method for training convolutional neural networks is as follows:
As inputting, the training by (5) formula to convolutional neural networks obtains the original greasy weather image data collection that step 1 is obtained
To transmittance figure as output result.
5. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In in the step 2, the method for demarcating scene areas is as follows:
Step 2-1, estimate camera parameter, for the greasy weather picture of camera shooting, there are the mappings of image coordinate and road surface coordinate
Relationship needs to calculate the upright position v on camera horizontal linehWith camera parameter λ, meanwhile, need to be arranged pavement markers point；
Upright position v on camera horizontal linehCalculation formula are as follows:
Wherein, y1And y2The respectively vertical coordinate of pavement markers point, d1And d2Respectively actual distance value；
Camera parameter λ calculation formula are as follows:
Wherein, v1And v2For the ordinate of mark point in the picture；
Step 2-2, scene target area is demarcated；Divide scene objects region, it is necessary first to which pavement markers point is set, according to step
The Misty Image of 1 shooting artificially chooses four sections of line segmentation sections, as target area, so as to effectively avoid all kinds of shelters.
6. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In as follows to transmittance figure optimization method using guiding filtering in the step 3:
The description of guiding filtering expression formula are as follows:
ti=∑jWij(I)*tj (8)
(6) in formula, tiImage is exported for filtering, I is navigational figure, tjFor input picture, WijFor weighting function, indicate are as follows:
(7) in formula, | ω | it is filter window ωkThe pixel quantity for including, μkAnd σkIt is navigational figure I in filter window ωkIt is equal
Value and variance size, ε is smoothing factor；
For the transmissivity of Misty Image, the transmissivity that step 2 is trained is as input picture t to be filteredj, have original
Mist image is as navigational figure I, and simultaneous (8) (9) formula, acquiring filtering output image is the transmittance figure t after optimizingi。
7. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In in the step 4, the method for calculating atmospheric scattering coefficient is as follows:
Transmissivity and atmospheric scattering coefficient meet following relationship
T (x)=e-βd(x) (10)
(10) in formula, d is scene point to the distance of observation point, as scene depth；β is atmospheric scattering coefficient, indicates atmosphere pair
The scattering power of light；T is transmissivity, reacts the depth information of scene objects；
Using the digital camera of step 1, target is obtained the distance between to observation point, i.e. depth of field distance is counted according to formula (10)
Calculate the atmospheric scattering coefficient of target point in region scene.
8. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In in the step 5, the method for calculating Zenith Distance scattering coefficient is as follows:
It averages to atmospheric scattering coefficients all in step 4, obtains the Zenith Distance scattering coefficient of region scene, to guarantee to know
The accuracy of other greasy weather grade, calculation formula are as follows:
9. the greasy weather visibility intelligence hierarchical identification method under a kind of complex background according to claim 1, feature exist
In as follows to greasy weather photo grade method of discrimination in the step 5:
According to Zenith Distance scattering coefficient, definition in conjunction with meteorology to visibility has following formula:
5 mark points are provided for any Misty Image according to formula (12), compared between judge distance and formula distance
Error.
10. the greasy weather visibility intelligent grading system under a kind of complex background, it is characterised in that:
Including front end intelligent identification device and backstage image identification system；
Front end intelligent identification device is made of digital camera (4), analysis hardware (5) and embedded device (1), is mainly used for obtaining
Take greasy weather image data；
Backstage image identification system is mainly used for identifying image gradation, is made of server apparatus (2) and display equipment (3), takes
Business device equipment (2) training convolutional neural networks export photo grade in display equipment (3) to identify various greasy weather pictures；
Digital camera (4) is successively connect with analysis hardware (5) and embedded device (1), and embedded device (1) is set with server
Standby (2) are connect with display equipment (3) respectively；
The embedded device (1) mainly includes data acquisition module (11) and preprocessing module (12), data acquisition module
(11) it is connect with preprocessing module (12), preprocessing module (12) is connect with display equipment (3)；Data acquisition module (11) is used for
Acquire a large amount of greasy weather pictures；Preprocessing module (12) is used to for the greasy weather picture that digital camera acquires to be made into training and test number
Grade mark and tag processes are carried out according to collection, and to every picture；
The analysis hardware (5) is connect with digital camera (4) by cable, for analyzing the photo of acquisition；The clothes
Device equipment (2) are engaged in mainly including training module (21), computing module (22) and categorization module (23)；Computing module (22) respectively with
Training module (21) and categorization module (23) connection, training module (21) are connect with preprocessing module (12), categorization module (23)
It is connect with display equipment (3)；Training module (21) is used for training convolutional neural networks model, inputs greasy weather image data collection, defeated
Transmittance figure out；Computing module (22) is used to optimize the transmittance figure of training module output, and calculates atmospheric scattering coefficient；Classification
Module (23) is according to computing module as a result, identifying five strong thick fog, thick fog, dense fog, mist, mist grades；
The display equipment (3) is mainly used for exporting that preprocessing module (12) and categorization module (23) obtain as a result, being used for
Real-time monitoring data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910351794.9A CN110378865A (en) | 2019-04-28 | 2019-04-28 | A kind of greasy weather visibility intelligence hierarchical identification method and system under complex background |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910351794.9A CN110378865A (en) | 2019-04-28 | 2019-04-28 | A kind of greasy weather visibility intelligence hierarchical identification method and system under complex background |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110378865A true CN110378865A (en) | 2019-10-25 |
Family
ID=68248532
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910351794.9A Pending CN110378865A (en) | 2019-04-28 | 2019-04-28 | A kind of greasy weather visibility intelligence hierarchical identification method and system under complex background |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378865A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866593A (en) * | 2019-11-05 | 2020-03-06 | 西南交通大学 | Highway severe weather identification method based on artificial intelligence |
| CN111192275A (en) * | 2019-12-30 | 2020-05-22 | 西安金路交通工程科技发展有限责任公司 | Highway fog visibility identification method based on dark channel prior theory |
| CN111259957A (en) * | 2020-01-15 | 2020-06-09 | 上海眼控科技股份有限公司 | Visibility monitoring and model training method, device, terminal and medium based on deep learning |
| CN111736237A (en) * | 2020-07-31 | 2020-10-02 | 上海眼控科技股份有限公司 | Radiation fog detection method and device, computer equipment and readable storage medium |
| CN112287757A (en) * | 2020-09-25 | 2021-01-29 | 北京百度网讯科技有限公司 | Water body identification method and device, electronic equipment and storage medium |
| CN112419272A (en) * | 2020-11-24 | 2021-02-26 | 湖北工业大学 | Method and system for quickly estimating visibility of expressway in foggy weather |
| CN112465822A (en) * | 2021-01-26 | 2021-03-09 | 长沙海信智能系统研究院有限公司 | Method, device and equipment for detecting cluster fog and computer readable storage medium |
| CN112818886A (en) * | 2021-02-09 | 2021-05-18 | 广州富港万嘉智能科技有限公司 | Flying dust detection method, readable storage medium, flying dust detection machine and intelligent food machine |
| CN113012077A (en) * | 2020-10-20 | 2021-06-22 | 杭州微帧信息科技有限公司 | Denoising method based on convolution guide graph filtering |
| WO2022012149A1 (en) * | 2020-07-17 | 2022-01-20 | 上海商汤智能科技有限公司 | Agglomerate fog recognition method and apparatus, electronic device, storage medium, and computer program product |
| CN114720425A (en) * | 2022-04-24 | 2022-07-08 | 安徽气象信息有限公司 | Visibility monitoring system and method based on image recognition |
| CN112287757B (en) * | 2020-09-25 | 2024-04-26 | 北京百度网讯科技有限公司 | Water body identification method and device, electronic equipment and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012100522A1 (en) * | 2011-01-26 | 2012-08-02 | 南京大学 | Ptz video visibility detection method based on luminance characteristic |
| CN103021177A (en) * | 2012-11-05 | 2013-04-03 | 北京理工大学 | Method and system for processing traffic monitoring video image in foggy day |
| KR101364727B1 (en) * | 2012-09-28 | 2014-02-20 | (주)한일에스티엠 | Method and apparatus for detecting fog using the processing of pictured image |
| CN104535356A (en) * | 2015-01-19 | 2015-04-22 | 中南大学 | Method and system for monitoring rope arrangement faults of drum steel wire rope on basis of machine vision |
| CN107194924A (en) * | 2017-05-23 | 2017-09-22 | 重庆大学 | Expressway foggy-dog visibility detecting method based on dark channel prior and deep learning |
| CN108363962A (en) * | 2018-01-25 | 2018-08-03 | 南京邮电大学 | A kind of method for detecting human face and system based on multi-level features deep learning |
| CN108734189A (en) * | 2017-04-20 | 2018-11-02 | 天津工业大学 | Vehicle License Plate Recognition System based on atmospherical scattering model and deep learning under thick fog weather |
| CN109034210A (en) * | 2018-07-04 | 2018-12-18 | 国家新闻出版广电总局广播科学研究院 | Object detection method based on super Fusion Features Yu multi-Scale Pyramid network |
| CN109493347A (en) * | 2017-09-12 | 2019-03-19 | 深圳科亚医疗科技有限公司 | The method and system that the object of sparse distribution is split in the picture |
-
2019
- 2019-04-28 CN CN201910351794.9A patent/CN110378865A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012100522A1 (en) * | 2011-01-26 | 2012-08-02 | 南京大学 | Ptz video visibility detection method based on luminance characteristic |
| KR101364727B1 (en) * | 2012-09-28 | 2014-02-20 | (주)한일에스티엠 | Method and apparatus for detecting fog using the processing of pictured image |
| CN103021177A (en) * | 2012-11-05 | 2013-04-03 | 北京理工大学 | Method and system for processing traffic monitoring video image in foggy day |
| CN104535356A (en) * | 2015-01-19 | 2015-04-22 | 中南大学 | Method and system for monitoring rope arrangement faults of drum steel wire rope on basis of machine vision |
| CN108734189A (en) * | 2017-04-20 | 2018-11-02 | 天津工业大学 | Vehicle License Plate Recognition System based on atmospherical scattering model and deep learning under thick fog weather |
| CN107194924A (en) * | 2017-05-23 | 2017-09-22 | 重庆大学 | Expressway foggy-dog visibility detecting method based on dark channel prior and deep learning |
| CN109493347A (en) * | 2017-09-12 | 2019-03-19 | 深圳科亚医疗科技有限公司 | The method and system that the object of sparse distribution is split in the picture |
| CN108363962A (en) * | 2018-01-25 | 2018-08-03 | 南京邮电大学 | A kind of method for detecting human face and system based on multi-level features deep learning |
| CN109034210A (en) * | 2018-07-04 | 2018-12-18 | 国家新闻出版广电总局广播科学研究院 | Object detection method based on super Fusion Features Yu multi-Scale Pyramid network |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866593A (en) * | 2019-11-05 | 2020-03-06 | 西南交通大学 | Highway severe weather identification method based on artificial intelligence |
| CN111192275A (en) * | 2019-12-30 | 2020-05-22 | 西安金路交通工程科技发展有限责任公司 | Highway fog visibility identification method based on dark channel prior theory |
| CN111259957A (en) * | 2020-01-15 | 2020-06-09 | 上海眼控科技股份有限公司 | Visibility monitoring and model training method, device, terminal and medium based on deep learning |
| WO2022012149A1 (en) * | 2020-07-17 | 2022-01-20 | 上海商汤智能科技有限公司 | Agglomerate fog recognition method and apparatus, electronic device, storage medium, and computer program product |
| CN111736237A (en) * | 2020-07-31 | 2020-10-02 | 上海眼控科技股份有限公司 | Radiation fog detection method and device, computer equipment and readable storage medium |
| CN112287757A (en) * | 2020-09-25 | 2021-01-29 | 北京百度网讯科技有限公司 | Water body identification method and device, electronic equipment and storage medium |
| CN112287757B (en) * | 2020-09-25 | 2024-04-26 | 北京百度网讯科技有限公司 | Water body identification method and device, electronic equipment and storage medium |
| CN113012077A (en) * | 2020-10-20 | 2021-06-22 | 杭州微帧信息科技有限公司 | Denoising method based on convolution guide graph filtering |
| CN112419272A (en) * | 2020-11-24 | 2021-02-26 | 湖北工业大学 | Method and system for quickly estimating visibility of expressway in foggy weather |
| CN112465822B (en) * | 2021-01-26 | 2021-05-28 | 长沙海信智能系统研究院有限公司 | Method, device and equipment for detecting cluster fog and computer readable storage medium |
| CN112465822A (en) * | 2021-01-26 | 2021-03-09 | 长沙海信智能系统研究院有限公司 | Method, device and equipment for detecting cluster fog and computer readable storage medium |
| CN112818886A (en) * | 2021-02-09 | 2021-05-18 | 广州富港万嘉智能科技有限公司 | Flying dust detection method, readable storage medium, flying dust detection machine and intelligent food machine |
| CN114720425A (en) * | 2022-04-24 | 2022-07-08 | 安徽气象信息有限公司 | Visibility monitoring system and method based on image recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378865A (en) | A kind of greasy weather visibility intelligence hierarchical identification method and system under complex background | |
| CN101430195B (en) | Method for computing electric power line ice-covering thickness by using video image processing technology | |
| CN108765404A (en) | A kind of road damage testing method and device based on deep learning image classification | |
| CN105279772B (en) | A kind of trackability method of discrimination of infrared sequence image | |
| CN107145851A (en) | Constructions work area dangerous matter sources intelligent identifying system | |
| CN107229929A (en) | A kind of license plate locating method based on R CNN | |
| CN111458721B (en) | Exposed garbage identification and positioning method, device and system | |
| CN113903081A (en) | Visual identification artificial intelligence alarm method and device for images of hydraulic power plant | |
| CN109145708A (en) | A kind of people flow rate statistical method based on the fusion of RGB and D information | |
| CN110660222A (en) | Intelligent environment-friendly electronic snapshot system for black smoke vehicle on road | |
| CN109086803B (en) | Deep learning and personalized factor-based haze visibility detection system and method | |
| CN107463931A (en) | A kind of real-time pointer instrument reading method and device based on ARM platforms | |
| CN111340951A (en) | Ocean environment automatic identification method based on deep learning | |
| CN109935080A (en) | The monitoring system and method that a kind of vehicle flowrate on traffic route calculates in real time | |
| CN112183472A (en) | Method for detecting whether test field personnel wear work clothes or not based on improved RetinaNet | |
| CN107016362A (en) | Vehicle based on vehicle front windshield sticking sign recognition methods and system again | |
| CN111145222A (en) | Fire detection method combining smoke movement trend and textural features | |
| CN106408526A (en) | Visibility detection method based on multilayer vectogram | |
| CN107264570A (en) | steel rail light band distribution detecting device and method | |
| CN112668375B (en) | Tourist distribution analysis system and method in scenic spot | |
| CN111476314B (en) | Fuzzy video detection method integrating optical flow algorithm and deep learning | |
| CN110765900B (en) | Automatic detection illegal building method and system based on DSSD | |
| Liu et al. | STCN-Net: A novel multi-feature stream fusion visibility estimation approach | |
| CN112686105B (en) | Fog concentration grade identification method based on video image multi-feature fusion | |
| CN112819817A (en) | River flow velocity estimation method based on graph calculation |
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 |