CN113469912A - Fog visibility estimation method and system based on deep learning - Google Patents
Fog visibility estimation method and system based on deep learning Download PDFInfo
- Publication number
- CN113469912A CN113469912A CN202110737978.6A CN202110737978A CN113469912A CN 113469912 A CN113469912 A CN 113469912A CN 202110737978 A CN202110737978 A CN 202110737978A CN 113469912 A CN113469912 A CN 113469912A
- Authority
- CN
- China
- Prior art keywords
- estimation
- map
- image
- transmittance
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000013135 deep learning Methods 0.000 title claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 claims abstract description 77
- 238000002834 transmittance Methods 0.000 claims abstract description 77
- 230000008033 biological extinction Effects 0.000 claims abstract description 53
- 238000003384 imaging method Methods 0.000 claims abstract description 24
- 238000007781 pre-processing Methods 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 239000000443 aerosol Substances 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a fog visibility estimation method and system based on deep learning, which is characterized by comprising the following steps of: 1) for the input foggy image, estimating a transmittance map through a transmittance estimation neural network; 2) restoring to obtain a defogged image by using the transmittance graph and the fog imaging model obtained by estimation; 3) inputting the defogged image into a depth estimation neural network estimation depth map; 4) calculating an extinction coefficient map through the depth map and the transmittance map; 5) the visibility map is calculated by Koschmieder's law through an extinction coefficient map. By analyzing the internal relation between the fog imaging model and the Koschmieder law and combining the two to estimate the visibility, the visibility estimation method has better interpretability, and can be widely applied to the field of visibility estimation in severe weather such as foggy days.
Description
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to a fog visibility estimation method and system based on deep learning.
Background
Visibility is the maximum distance a person with normal vision can identify an object from the background. Visibility is mainly affected by two factors, namely brightness difference between a target object and the background of the target object and atmospheric transparency. The atmosphere transparency is greatly related to weather, and the atmosphere transparency is affected by rain, fog, snow, sand, haze and other weather, so that the visibility is reduced. Thus, in general, visibility depends mainly on weather conditions.
In the measurement of visibility, the traditional method is to measure the extinction coefficient by an instrument so as to deduce the visibility. The limitations of these conventional approaches are also very significant: the atmospheric transmission instrument calculates visibility by measuring the air column transmittance between two points, and the method has larger error in low visibility weather such as rain, fog and the like; the automatic laser visibility measuring instrument calculates the visibility by measuring the atmospheric extinction coefficient, and although the method is more accurate, the instrument has the defects of high manufacturing cost, complex operation, difficult popularization and the like.
With the development of computer science, some methods based on digital image processing technology and data driving are also in progress. However, the current non-deep learning method can only estimate one visibility value for one image, which is too coarse and insufficient in accuracy for the visibility estimation problem under uneven weather conditions. In the field of computer vision, due to the development of deep neural network theory and the improvement of computing power, visibility estimation based on a deep learning method becomes a direction of research, but the deep learning method suffers from a problem due to poor interpretability.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a method and a system for estimating visibility in foggy days based on deep learning, which can output visibility at a pixel level by inputting only one image, thereby solving the problem of dependence of the conventional method on measuring instruments and measuring conditions, and solving the problem of too rough visibility estimation of a non-deep learning method in the field of computers due to the visibility at the pixel level. In addition, the method captures the intrinsic connection between the fog imaging model and the Koschmieder law and combines the fog imaging model and the Koschmieder law to carry out visibility estimation, so that the problem of insufficient interpretability of the deep learning method is effectively solved.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a method for estimating the visibility in foggy days based on deep learning, which comprises the following steps:
1) inputting the foggy image into a pre-constructed transmittance estimation neural network for transmittance estimation to obtain a transmittance graph of the foggy image;
2) processing the defogged image by using the transmittance image and the fog imaging model obtained by estimation, and recovering to obtain a defogged image;
3) inputting the defogged image into a depth estimation neural network which is constructed in advance for depth estimation to obtain a depth map;
4) calculating to obtain an extinction coefficient map through the depth map and the transmittance map;
5) and calculating the visibility graph by using Koschmieder law through an extinction coefficient graph.
Further, in the step 1), the method for inputting the foggy image into a pre-constructed transmittance estimation neural network for transmittance estimation to obtain a transmittance map includes the following steps:
1.1) constructing and initializing a transmittance estimation neural network;
1.2) loading the network parameters of the pre-trained transmittance estimation neural network and setting the network parameters into a test mode;
1.3) preprocessing the input fog image according to the requirement of the transmittance estimation neural network;
1.4) inputting the preprocessed fog-carrying image into a transmittance estimation neural network to obtain a transmittance map.
Further, in the step 2), the method for recovering the defogged image through the transmittance map and the fogging map comprises the following steps:
2.1) calculating to obtain an atmospheric light component A according to the input fogged image;
2.2) substituting the fog image, the transmittance image obtained in the step 1) and the atmospheric light component A into a fog imaging model to obtain a defogged image.
Further, the method for calculating and obtaining the atmospheric light component a according to the input foggy image in the step 2.1) comprises the following steps:
2.1.1) solving the minimum value of each pixel of the input fog-carrying image on three channels of RGB;
2.1.2) carrying out minimum value filtering on the minimum value image obtained in the step 2.1.1) to obtain a dark channel image;
2.1.3) taking the first 0.1% pixels according to the brightness of the dark channel map obtained in the step 2.1.2), finding the corresponding pixel values of the pixels in the input fog image, and taking the average value of the pixel values as the atmospheric light component A.
Further, in the step 3), the method for inputting the defogged image into a depth estimation neural network constructed in advance to perform depth estimation to obtain a depth map includes the following steps:
3.1) constructing and initializing a depth estimation neural network;
3.2) loading the pre-trained network parameters of the deep estimation neural network, and setting the network parameters as a test mode;
3.3) preprocessing the defogged image according to requirements;
and 3.4) inputting the preprocessed and defogged image into a depth estimation neural network for depth estimation to obtain a depth map.
Further, in the step 4), the method for obtaining the extinction coefficient map through calculation by using the depth map and the transmittance map includes:
firstly, calculating the extinction coefficient corresponding to each pixel according to the transmittance map and the depth map, wherein the calculation formula is as follows:
t(x)=e-β(x)d(x)wherein, t (x) is the transmittance of the transmittance graph at a pixel point; beta (x) is an extinction coefficient corresponding to each pixel(ii) a d (x) is the depth of the depth map at one pixel;
next, an extinction coefficient map is obtained from the extinction coefficient β (x) corresponding to each pixel.
Further, in the step 5), a method for calculating a visibility map by using Koschmieder's law through an extinction coefficient map includes:
firstly, calculating the visibility of each pixel point according to an extinction coefficient diagram and a Koschmieder law, wherein the calculation formula is as follows:
wherein, v (x) is the visibility of a pixel; ε is the eye contrast threshold; beta (x) is the extinction coefficient of the extinction coefficient graph at a pixel point;
and secondly, obtaining a visibility graph according to the visibility V (x) of each pixel point.
In a second aspect of the present invention, a fog visibility estimation system based on deep learning is provided, which includes:
the transmissivity graph determining module is used for inputting the fog images into a preset transmissivity estimation neural network for transmissivity estimation to obtain a transmissivity graph;
the defogged image determining module is used for recovering and obtaining a defogged image by utilizing the transmittance graph and the fog imaging model which are obtained by estimation;
the depth map determining module is used for inputting the defogged image into a pre-constructed depth estimation neural network for depth estimation to obtain a depth map;
the extinction coefficient map determining module is used for calculating to obtain an extinction coefficient map through the depth map and the transmissivity map;
and the visibility map determining module is used for calculating the visibility map by utilizing the Koschmieder law through the extinction coefficient map.
Further, the transmittance map determination module includes:
the transmissivity estimation neural network construction module is used for constructing and initializing a transmissivity estimation neural network;
the first network parameter loading module is used for loading the network parameters of the pre-trained transmittance estimation neural network and setting the network parameters into a test mode;
the first preprocessing module is used for preprocessing the input image with fog according to requirements;
and the transmissivity estimation module is used for inputting the preprocessed foggy image into a transmissivity estimation neural network for transmissivity estimation to obtain a transmissivity graph.
Further, the depth map determination module comprises:
the depth estimation neural network construction module is used for constructing and initializing a depth estimation neural network;
the second network parameter loading module is used for loading the network parameters of the pre-trained deep estimation neural network and setting the network parameters as a test mode;
the second preprocessing module is used for preprocessing the defogged image according to requirements;
and the depth estimation module is used for inputting the preprocessed and defogged image into a depth estimation neural network for depth estimation to obtain a depth map.
Due to the adoption of the technical scheme, the invention has the following advantages:
1. the invention analyzes the internal relation between the fog imaging model and the Koschmieder law and combines the two to estimate the visibility, so the invention has better interpretability;
2. the method is not only suitable for the foggy days, but also suitable for the weather conditions of other imaging models which are the same as those of the foggy days;
3. different from the traditional method in the field of computer vision, the method can obtain visibility estimation at a pixel level, and the traditional method can only obtain visibility estimation at a picture level, so that the method can better estimate uneven fog, has stronger robustness and has finer granularity;
4. the invention has the characteristics of modularization and low coupling, so that with the development of scientific technology, modules in the framework of the invention can be replaced by a method with better effect, thereby improving the total estimation effect.
Therefore, the method can be widely applied to the field of visibility estimation in severe weather such as foggy days.
Drawings
FIG. 1 is a flow chart of a fog visibility estimation method based on deep learning according to the present invention;
FIG. 2 is a flow chart of the present invention for transmittance estimation of a hazy image using a transmittance estimation neural network;
FIG. 4 is a flow chart of the present invention for solving for atmospheric light components;
FIG. 3 is a flow chart of inputting a defogged image into a depth estimation neural network for estimation to obtain a depth map thereof in the present invention;
FIG. 5 is a schematic diagram of a network structure of DehazeNet constructed in an embodiment of the present invention;
fig. 6 is a schematic diagram of a network structure of a dorn (deep atomic Regression network) constructed in the embodiment of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and examples.
The invention provides a fog visibility estimation method based on deep learning based on a fog imaging model, a Koschmieder law and related work in the field of image defogging and depth estimation at present, and the following is a brief introduction of the Koschmieder law and the fog imaging model which are theoretical foundations of the invention:
the Koschmieder law is the basis for determining visibility. When human eyes visually observe a far ground target facing the terrace sky along the horizontal direction, aerosol particles in the atmosphere generate an absorption scattering effect due to the influence of direct solar light, the sky, a cloud layer and ground diffuse scattering, so that an observer feels like a layer of shielding curtain in the atmosphere, the direct contrast between a target object and the background is reduced, and the distance that the observer can see the target object is reduced. The law of the visual contrast C as a function of the distance d, i.e. Koschmieder's law, is expressed as follows:
wherein C is the visual contrast, also called visual contrast, which is the relative brightness difference between the target and the background; beta is an extinction coefficient, which is a characteristic quantity for representing the transparency of the atmosphere in any case; d is the distance of the observer from the target object; i isobjectAnd IbackgroundThe brightness of the object and the background, respectively.
If the target object is just visible, it means that the visual contrast C reaches the eye contrast threshold e, which is 0.02 or 0.05 for a normal person. At this time, the distance d between the observer and the target object represents the visibility V, and the above formula (1) can be expressed as:
ε=e-βV (2)
namely:
further, in computer vision and computer graphics, a fog imaging model described by the following equation is widely used:
I(x)=J(x)t(x)+A(1-t(x)) (4)
t(x)=e-β(x)d(x) (5)
wherein, I (x) is the value of the image to be defogged at a pixel point; j (x) is the value of the image after defogging at a pixel point; a is an atmospheric light component, which is constant for one image; t (x) is the transmittance of the image at a pixel; beta (x) is an extinction coefficient, the size of the extinction coefficient reflects fog concentration, generally, when the fog concentration is uniform, the fog concentration can be regarded as a constant for a specific image, and when the fog concentration is nonuniform, the extinction coefficient refers to the extinction coefficient of a pixel point in the image; and d (x) is the actual physical distance of the target object relative to the shooting source on one pixel point of the image.
According to the theoretical analysis, the visibility can be calculated by utilizing the Koschmieder law as long as the extinction coefficient is calculated by the fog imaging model. As shown in fig. 1, the invention provides a method for estimating visibility in foggy weather based on deep learning, which comprises the following steps:
1) inputting the foggy image into a pre-constructed transmittance estimation neural network for transmittance estimation to obtain a transmittance graph of the foggy image;
2) processing the defogged image by using the transmittance image and the fog imaging model obtained by estimation, and recovering to obtain a defogged image;
3) inputting the defogged image into a depth estimation neural network which is constructed in advance for depth estimation to obtain a depth map of the image;
4) calculating an extinction coefficient map through the obtained depth map and the obtained transmittance map;
5) and calculating the visibility graph by using Koschmieder law through an extinction coefficient graph.
As shown in fig. 2, the method for obtaining the transmittance map of the foggy image by inputting the foggy image into the transmittance estimation neural network constructed in advance in the step 1) and performing transmittance estimation includes the following steps:
1.1) constructing and initializing a transmissivity estimation neural Network, wherein the neural Network can adopt DehazeNet, DCPDN (Densely Connected Pyramid Dehazing Network) and the like;
1.2) loading the network parameters of the pre-trained transmittance estimation neural network and setting the network parameters into a test mode;
1.3) preprocessing input fog images according to the requirements of a transmittance estimation neural network, such as clipping, scaling, normalizing mean value and variance and the like;
1.4) inputting the preprocessed fog-carrying image into a transmittance estimation neural network to obtain a transmittance graph of the fog-carrying image.
In the step 2), the method for obtaining the defogged image comprises the following steps:
2.1) calculating to obtain an atmospheric light component A based on the foggy image;
2.2) substituting the fog image, the transmittance image obtained in the step 1) and the atmospheric light component A into a fog imaging model (namely formula (4)) to obtain a defogged image.
As shown in fig. 3, there are many algorithms for obtaining the atmospheric light component a in the step 2.1), and a method of using a pixel which is 0.1% of the pixel in the dark channel map according to the brightness as a is widely used at present, and the method includes the following steps:
2.1.1) solving the minimum value of each pixel of the fogged image on three channels of RGB;
2.1.2) carrying out minimum value filtering on the obtained minimum value graph to obtain a dark channel graph;
2.1.3) taking the first 0.1% of pixels according to the brightness of the obtained dark channel map, finding the corresponding pixel values of the pixels in the input image, and taking the average value of the pixel values as the atmospheric light component A.
As shown in fig. 4, the method for constructing the depth estimation neural network, initializing network parameters of the depth estimation neural network, and estimating the transmittance using the depth estimation neural network in step 3) includes the following steps:
3.1) constructing and initializing a depth estimation neural network, such as FCRN (full capacitive recovery networks), DORN (deep atomic recovery networks), and the like;
3.2) loading the pre-trained deep estimation neural network parameters and setting the parameters as a test mode;
3.3) preprocessing the defogged image according to the requirement of the depth estimation application network, such as cutting, zooming, standardizing the mean value and the variance and the like;
and 3.4) inputting the preprocessed and defogged image into a depth estimation neural network to obtain a depth estimation image.
In the step 4), the transmittance map is obtained according to the step 1), and the depth map is obtained according to the step 3), so that t (x) and d (x) in the fog imaging model are known, and the extinction coefficient β (x) corresponding to each pixel can be calculated by using the formula (5), i.e., the extinction coefficient map is obtained.
In the step 5), since the extinction coefficient map is obtained according to the step 4), the extinction coefficient β in Koschmieder's law is known, and the visibility V of each pixel point can be calculated by substituting the equation (3) with ∈ ═ 0.02 or ∈ ═ 0.05, so that the visibility map is obtained.
According to the fog visibility estimation method based on deep learning, the invention also provides a fog visibility estimation system based on deep learning, which comprises the following steps: the transmissivity graph determining module is used for inputting the fog-carrying images into a preset transmissivity estimation neural network for transmissivity estimation to obtain a transmissivity graph of the fog-carrying images; the defogged image determining module is used for recovering and obtaining a defogged image by utilizing the transmittance graph and the fog imaging model which are obtained by estimation; the depth map determining module is used for inputting the defogged image into a pre-constructed depth estimation neural network for depth estimation to obtain a depth map; the extinction coefficient map determining module is used for calculating to obtain an extinction coefficient map through the depth map and the transmissivity map; and the visibility map determining module is used for calculating the visibility map by utilizing the Koschmieder law through the extinction coefficient map.
Further, the transmittance map determination module includes: the transmissivity estimation neural network construction module is used for constructing and initializing a transmissivity estimation neural network; the first network parameter loading module is used for loading the network parameters of the pre-trained transmittance estimation neural network and setting the network parameters into a test mode; the first preprocessing module is used for preprocessing the input image with fog according to requirements; and the transmissivity estimation module is used for inputting the preprocessed foggy image into a transmissivity estimation neural network for transmissivity estimation to obtain a transmissivity graph.
Further, the depth map determination module comprises: the depth estimation neural network construction module is used for constructing and initializing a depth estimation neural network; the second network parameter loading module is used for loading the network parameters of the pre-trained deep estimation neural network and setting the network parameters as a test mode; the second preprocessing module is used for preprocessing the defogged image according to requirements; and the depth estimation module is used for inputting the preprocessed and defogged image into a depth estimation neural network for depth estimation to obtain a depth map.
The first embodiment is as follows:
the embodiment is used for estimating the visibility in foggy days on a certain road, and comprises the following steps:
1) for the input fogged image, a transmittance map is estimated by a transmittance estimation neural network.
As shown in fig. 5, DehazeNet is selected as a transmittance estimation neural network, a network is built using a Pytorch programming environment (version 1.6) and parameters are initialized, network parameters downloaded from the internet are loaded, the transmittance estimation neural network is set to a test mode, an input image is subjected to preprocessing for normalization of mean and variance, and then input into DehazeNet, so that a transmittance map is obtained.
2) And recovering and obtaining the defogged image by utilizing the transmittance graph and the fog imaging model which are obtained by estimation.
Specifically, the method comprises the following steps:
2.1) obtaining an atmospheric light component A;
2.1.1) solving the minimum value of each pixel of the input image on three channels of RGB;
2.1.2) carrying out minimum value filtering on the minimum value graph obtained in the step 2.1.1), wherein the size of a filter is selected to be 5 x 5, and obtaining a dark channel graph;
2.1.3) taking the first 0.1% of pixels according to the brightness of the dark channel map obtained in the step 2.1.2), finding the corresponding pixel values of the pixels in the input image, and taking the average value of the pixel values as the atmospheric light component A.
And 2.2) substituting t (x), I (x) and A into a formula of the fog imaging model to obtain a defogged image.
3) And inputting the defogged image into a depth estimation neural network estimation depth map.
As shown in fig. 6, select a DORN (deep atomic Regression network) as a depth estimation neural network, build a network using a Pytorch programming environment (version 1.6) and initialize parameters, load network parameters downloaded from the internet, set the depth estimation neural network as a test mode, perform preprocessing of normalization of mean and variance on an input image, and input the input image into the DORN to obtain a depth estimation image.
4) And calculating an extinction coefficient map through the depth map and the transmittance map.
Obtaining a depth map according to the step 3), so that when t (x) and d (x) in the fog imaging model are known, the extinction coefficient beta corresponding to each pixel can be obtained through calculation by a formula of the fog imaging model, and the extinction coefficient map is obtained.
5) The visibility map is calculated by Koschmieder's law through an extinction coefficient map.
And 4) obtaining an extinction coefficient graph according to the step 4), wherein the extinction coefficient beta in the Koschmieder law is known, epsilon is taken to be 0.02, and the epsilon is substituted into a formula to calculate the visibility V of each pixel point, so that the visibility graph is obtained.
A specific embodiment is given above, but the invention is not limited to the described embodiment. The basic idea of the present invention lies in the above solution, and it is obvious to those skilled in the art that it is not necessary to spend creative efforts to design various modified models, formulas and parameters according to the teaching of the present invention. Variations, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the invention, and still fall within the scope of the invention.
Claims (10)
1. A fog visibility estimation method based on deep learning is characterized by comprising the following steps:
1) inputting the foggy image into a pre-constructed transmittance estimation neural network for transmittance estimation to obtain a transmittance graph of the foggy image;
2) processing the defogged image by using the transmittance image and the fog imaging model obtained by estimation, and recovering to obtain a defogged image;
3) inputting the defogged image into a depth estimation neural network which is constructed in advance for depth estimation to obtain a depth map;
4) calculating to obtain an extinction coefficient map through the depth map and the transmittance map;
5) and calculating the visibility graph by using Koschmieder law through an extinction coefficient graph.
2. The fog visibility estimation method based on deep learning as claimed in claim 1, wherein: in the step 1), the method for obtaining the transmittance map by inputting the foggy image into a pre-constructed transmittance estimation neural network for transmittance estimation comprises the following steps:
1.1) constructing and initializing a transmittance estimation neural network;
1.2) loading the network parameters of the pre-trained transmittance estimation neural network and setting the network parameters into a test mode;
1.3) preprocessing the input fog image according to the requirement of the transmittance estimation neural network;
1.4) inputting the preprocessed fog-carrying image into a transmittance estimation neural network to obtain a transmittance map.
3. The fog visibility estimation method based on deep learning as claimed in claim 1, wherein: in the step 2), the method for recovering the defogged image through the transmittance map and the fogging map comprises the following steps:
2.1) calculating to obtain an atmospheric light component A according to the input fogged image;
2.2) substituting the fog image, the transmittance image obtained in the step 1) and the atmospheric light component A into a fog imaging model to obtain a defogged image.
4. The fog visibility estimation method based on deep learning as claimed in claim 3, wherein: in the step 2.1), the method for calculating and obtaining the atmospheric light component a according to the input fogged image comprises the following steps:
2.1.1) solving the minimum value of each pixel of the input fog-carrying image on three channels of RGB;
2.1.2) carrying out minimum value filtering on the minimum value image obtained in the step 2.1.1) to obtain a dark channel image;
2.1.3) taking the first 0.1% pixels according to the brightness of the dark channel map obtained in the step 2.1.2), finding the corresponding pixel values of the pixels in the input fog image, and taking the average value of the pixel values as the atmospheric light component A.
5. The fog visibility estimation method based on deep learning as claimed in claim 1, wherein: in the step 3), the method for inputting the defogged image into the pre-constructed depth estimation neural network for depth estimation to obtain the depth map comprises the following steps:
3.1) constructing and initializing a depth estimation neural network;
3.2) loading the pre-trained network parameters of the deep estimation neural network, and setting the network parameters as a test mode;
3.3) preprocessing the defogged image according to requirements;
and 3.4) inputting the preprocessed and defogged image into a depth estimation neural network for depth estimation to obtain a depth map.
6. The fog visibility estimation method based on deep learning as claimed in claim 1, wherein: in the step 4), the method for obtaining the extinction coefficient map through calculation by using the depth map and the transmittance map comprises the following steps:
firstly, calculating the extinction coefficient corresponding to each pixel according to the transmittance map and the depth map, wherein the calculation formula is as follows:
t(x)=e-β(x)d(x)
wherein, t (x) is the transmittance of the transmittance graph at a pixel point; β (x) is an extinction coefficient corresponding to each pixel; d (x) is the depth of the depth map at one pixel;
next, an extinction coefficient map is obtained from the extinction coefficient β (x) corresponding to each pixel.
7. The fog visibility estimation method based on deep learning as claimed in claim 1, wherein: in the step 5), the method for calculating the visibility map by using the Koschmieder law through the extinction coefficient map comprises the following steps:
firstly, calculating the visibility of each pixel point according to an extinction coefficient diagram and a Koschmieder law, wherein the calculation formula is as follows:
wherein, v (x) is the visibility of a pixel; ε is the eye contrast threshold; beta (x) is the extinction coefficient of the extinction coefficient graph at a pixel point;
and secondly, obtaining a visibility graph according to the visibility V (x) of each pixel point.
8. A fog visibility estimation system based on deep learning and adopting the method as claimed in any one of claims 1 to 7, which is characterized by comprising:
the transmissivity graph determining module is used for inputting the fog images into a preset transmissivity estimation neural network for transmissivity estimation to obtain a transmissivity graph;
the defogged image determining module is used for recovering and obtaining a defogged image by utilizing the transmittance graph and the fog imaging model which are obtained by estimation;
the depth map determining module is used for inputting the defogged image into a pre-constructed depth estimation neural network for depth estimation to obtain a depth map;
the extinction coefficient map determining module is used for calculating to obtain an extinction coefficient map through the depth map and the transmissivity map;
and the visibility map determining module is used for calculating the visibility map by utilizing the Koschmieder law through the extinction coefficient map.
9. The fog visibility estimation system based on deep learning as claimed in claim 8, wherein: the transmittance map determination module includes:
the transmissivity estimation neural network construction module is used for constructing and initializing a transmissivity estimation neural network;
the first network parameter loading module is used for loading the network parameters of the pre-trained transmittance estimation neural network and setting the network parameters into a test mode;
the first preprocessing module is used for preprocessing the input image with fog according to requirements;
and the transmissivity estimation module is used for inputting the preprocessed foggy image into a transmissivity estimation neural network for transmissivity estimation to obtain a transmissivity graph.
10. The fog visibility estimation system based on deep learning as claimed in claim 8, wherein: the depth map determination module comprises:
the depth estimation neural network construction module is used for constructing and initializing a depth estimation neural network;
the second network parameter loading module is used for loading the network parameters of the pre-trained deep estimation neural network and setting the network parameters as a test mode;
the second preprocessing module is used for preprocessing the defogged image according to requirements;
and the depth estimation module is used for inputting the preprocessed and defogged image into a depth estimation neural network for depth estimation to obtain a depth map.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110737978.6A CN113469912B (en) | 2021-06-30 | 2021-06-30 | Deep learning-based foggy day visibility estimation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110737978.6A CN113469912B (en) | 2021-06-30 | 2021-06-30 | Deep learning-based foggy day visibility estimation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113469912A true CN113469912A (en) | 2021-10-01 |
CN113469912B CN113469912B (en) | 2024-06-14 |
Family
ID=77876590
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110737978.6A Active CN113469912B (en) | 2021-06-30 | 2021-06-30 | Deep learning-based foggy day visibility estimation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113469912B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116664448A (en) * | 2023-07-24 | 2023-08-29 | 南京邮电大学 | Medium-high visibility calculation method and system based on image defogging |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104680494A (en) * | 2015-03-14 | 2015-06-03 | 西安电子科技大学 | Optimal fog image recovery method based on artificial fog addition |
CN104809707A (en) * | 2015-04-28 | 2015-07-29 | 西南科技大学 | Method for estimating visibility of single fog-degraded image |
US9288458B1 (en) * | 2015-01-31 | 2016-03-15 | Hrl Laboratories, Llc | Fast digital image de-hazing methods for real-time video processing |
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 |
CN107749052A (en) * | 2017-10-24 | 2018-03-02 | 中国科学院长春光学精密机械与物理研究所 | Image defogging method and system based on deep learning neutral net |
CN112365467A (en) * | 2020-11-11 | 2021-02-12 | 武汉长江通信智联技术有限公司 | Foggy image visibility estimation method based on single image depth estimation |
-
2021
- 2021-06-30 CN CN202110737978.6A patent/CN113469912B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9288458B1 (en) * | 2015-01-31 | 2016-03-15 | Hrl Laboratories, Llc | Fast digital image de-hazing methods for real-time video processing |
CN104680494A (en) * | 2015-03-14 | 2015-06-03 | 西安电子科技大学 | Optimal fog image recovery method based on artificial fog addition |
CN104809707A (en) * | 2015-04-28 | 2015-07-29 | 西南科技大学 | Method for estimating visibility of single fog-degraded image |
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 |
CN107749052A (en) * | 2017-10-24 | 2018-03-02 | 中国科学院长春光学精密机械与物理研究所 | Image defogging method and system based on deep learning neutral net |
CN112365467A (en) * | 2020-11-11 | 2021-02-12 | 武汉长江通信智联技术有限公司 | Foggy image visibility estimation method based on single image depth estimation |
Non-Patent Citations (1)
Title |
---|
于佳松,韩夏清: "基于图像识别技术白天能见度的算法分析", 电子测量技术, 31 October 2020 (2020-10-31) * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116664448A (en) * | 2023-07-24 | 2023-08-29 | 南京邮电大学 | Medium-high visibility calculation method and system based on image defogging |
CN116664448B (en) * | 2023-07-24 | 2023-10-03 | 南京邮电大学 | Medium-high visibility calculation method and system based on image defogging |
Also Published As
Publication number | Publication date |
---|---|
CN113469912B (en) | 2024-06-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Haze visibility enhancement: A survey and quantitative benchmarking | |
Li et al. | Image dehazing using residual-based deep CNN | |
Ma et al. | Perceptual quality assessment for multi-exposure image fusion | |
Wu et al. | Accurate transmission estimation for removing haze and noise from a single image | |
CN107424133B (en) | Image defogging method and device, computer storage medium and mobile terminal | |
CN104809707A (en) | Method for estimating visibility of single fog-degraded image | |
Hautière et al. | Enhanced fog detection and free-space segmentation for car navigation | |
CN110135434B (en) | Underwater image quality improvement method based on color line model | |
EP2846306B1 (en) | Image processing apparatus for removing haze contained in still image and method thereof | |
CN103020914A (en) | Rapid image defogging method based on spatial continuity principle | |
CN116309781B (en) | Cross-modal fusion-based underwater visual target ranging method and device | |
CN113313702A (en) | Aerial image defogging method based on boundary constraint and color correction | |
TW201435806A (en) | Image recovery method | |
CN113469912B (en) | Deep learning-based foggy day visibility estimation method and system | |
Yuan et al. | A confidence prior for image dehazing | |
Kumari et al. | Fast and efficient visibility restoration technique for single image dehazing and defogging | |
CN112017130A (en) | Novel image restoration method based on self-adaptive anisotropic total variation regularization | |
Singh et al. | Visibility enhancement and dehazing: Research contribution challenges and direction | |
Husain et al. | VRHAZE: The simulation of synthetic haze based on visibility range for dehazing method in single image | |
CN117152016A (en) | Image defogging method and system based on improved dark channel prior | |
Gao et al. | Variable exponent regularization approach for blur kernel estimation of remote sensing image blind restoration | |
Mittal et al. | FEMT: a computational approach for fog elimination using multiple thresholds | |
CN113763261B (en) | Real-time detection method for far small target under sea fog weather condition | |
CN113888420A (en) | Underwater image restoration method and device based on correction model and storage medium | |
CN112598777B (en) | Haze fusion method based on dark channel prior |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant |