CN115170437A - Fire scene low-quality image recovery method for rescue robot - Google Patents

Abstract

The invention provides a fire scene low-quality image recovery method for a rescue robot. According to the method, the flame region is segmented based on the region threshold segmentation algorithm, so that the influence of a flame light source on the global atmospheric light estimation is avoided; an atmospheric light detection operator is designed, accurate atmospheric light parameters are obtained based on the super-pixel block, and the problem of global atmospheric light estimation distortion is solved; a transmissivity estimation optimization module is constructed, the transmissivity is refined based on a bilateral weighting guide filtering method, and the problems of halation and the like caused by the traditional method are solved. The method provided by the invention is beneficial to improving the clarity of the fire scene image recovery, and provides a data basis for improving the scene detection, identification, environment mapping and path planning of the rescue robot.

Description

A low-quality image restoration method of fire scene for rescue robot

technical field

The invention belongs to the field of image processing of rescue robots, and in particular relates to a low-quality image restoration method of fire scenes for rescue robots.

Background technique

In the fire rescue environment, the combustion is usually accompanied by uneven fire and smoke. When the rescue robot performs tasks in such scenes, the collected images are usually affected by environmental factors such as light, flame, and fog.

Due to the uneven distribution of ambient light and the uneven fog density of images in fire scenes, the existing methods have the problem of estimation distortion in dealing with fire scenes. Through the research on the clearing method of low-quality images in the rescue fire environment, the low-quality degraded images are restored to high-quality images, which provides a data basis for on-site detection, recognition, environmental mapping and path planning of rescue robots.

SUMMARY OF THE INVENTION

In order to solve the problems in the background art, the present invention proposes a low-quality image restoration method of a fire scene for a rescue robot.

The technical scheme adopted in the present invention is as follows:

1. A low-quality image restoration system for fire scenes for rescue robots

It includes a regional atmospheric light estimation module, a transmittance estimation optimization module and an image reconstruction and restoration module, and the regional atmospheric light estimation module includes a fire regional atmospheric light estimation module and a global atmospheric light estimation module;

The atmospheric light estimation module in the flare area is used to segment the original image into a flare area image and a non-fire area image;

The global atmospheric light estimation module is used to perform superpixel segmentation on the non-fire area image and design the atmospheric light detection operator to obtain the estimated value of the global atmospheric light;

The transmittance estimation optimization module is used to estimate the rough transmittance, and use the bilateral weighted guided filtering method to calculate the precise transmittance;

Image reconstruction and restoration module, used to reconstruct and restore clear images.

2. Low-quality image restoration method of fire scene for rescue robot using the above system

Step 1: The original image I is divided into a fire area image _IF and a non-fire area image I _NF by the fire area atmospheric light estimation module;

Step 2: The global atmospheric light estimation module calculates the global atmospheric light estimation value A ₀ by constructing an atmospheric light detection operator;

Step 3: Input the dark channel map and the estimated value of global atmospheric light into the transmittance estimation optimization module to estimate the rough transmittance, and use the bilateral weighted filter transmittance optimization to obtain the precise transmittance t;

Step 4: Restoring the final clear image through the image reconstruction and restoration module.

Described step 1) is specifically:

1.1) combine the RGB color space criterion and the HLS color space criterion to divide the original image I to obtain the initial flare area image I ₁ ;

1.2) Perform the morphological processing of opening and closing the initial flare area image _I1 , delete the isolated points in the image and fill the holes inside the area image _I1 , and further perform Gaussian filtering on the morphologically processed image to obtain the final result. The flare area image _IF ; and the non-flare area image I _NF is obtained according to the original image I and the flare area image _IF .

Described step 2) is specifically:

2.1) minimum value filtering is performed on the R, G, B three channels of the original image I to obtain the dark channel map I _d of the image;

2.2) For the original image I, construct the atmospheric light detection operator score:

score=(1-S) _Id

Wherein, S is the saturation component of the original image I;

2.3) Perform superpixel segmentation on the non-fire area image I _NF to obtain a segmentation map Is, and calculate the atmospheric light detection operator score of each superpixel block _{s i} in the segmentation map Is ( _si ∈I _{s )}

为超像素块s_i中像素点的个数，x为超像素块s_i中的像素点；in,

is the number of pixels in the superpixel block _si , and x is the pixel in the superpixel block _si ;

值按由大到小的顺序进行排序，选取大气光检测算子得分最大的超像素块，记为s_max，计算超像素块s_max中所有像素点的像素值平均值得到全局大气光的估计值A₀：2.4) Each superpixel block corresponds to

The values are sorted in descending order, and the superpixel block with the largest atmospheric light detection operator score is selected, denoted as s _max , and the average value of all pixels in the superpixel block s _max is calculated to obtain the estimation of the global atmospheric light Value A ₀ :

为超像素块s_max中像素点的个数，I(x)为像素块s_max中像素点x的像素值。in,

is the number of pixels in the _superpixel block smax, and I(x) is the pixel value of the pixel x in the pixel block _smax .

Described step 3) is specifically:

3.1) Through the L channel of the original image I ₁ , calculate the adaptive confidence t ^* (x):

Among them, Ω is the minimum filter interval, p is the confidence adjustment parameter, and L(y) is the L channel value of the pixel point y in the region Ω;

2) Calculate the rough transmittance characteristic map t ₀ of the image according to the dark channel map I _d of the original image I ₁ and the estimated global atmospheric light value A ₀ , and the calculation formula is as follows:

3) Use bilateral weighted guided filtering to optimize t ₀ to obtain the image transmittance t. The calculation formula is as follows:

t=a* _Ig +b

Among them, I _g is the grayscale image of the original image I, a and b are the guided filter coefficients, and the calculation formula is as follows:

Among them, ε is the tolerance factor, d is the filter interval, ω(i,j,k,l) is the filter weighting coefficient, (k,l) represents the center coordinate of the filter window, (i,j) represents the other coordinates of the window, ω _m is the kernel function, and the value of m is {1, 2, 3, 4}. The calculation formula is as follows:

Among them, σ _d is the air domain weight, σ _r is the value domain weight;

Among them, I _m (i,j), I _m (k,l), m∈{1,2,3,4} represents the pixel value of the corresponding image at points (i,j), (k,l), calculate The formula is as follows:

I ₂ =t ₀

I ₃ =I _g

表示矩阵哈达马积。in,

represents the matrix Hadamard product.

In the described step 4), the image restoration module outputs the final clear image J, which is expressed as:

Beneficial effects of the present invention:

A low-quality image restoration method for a fire scene for rescue robots proposed by the present invention divides the flame region through a region threshold segmentation algorithm, so as to avoid the distorted influence of the flame light source on the global atmospheric light estimation, and design an atmospheric light detection operator based on superpixels. The accurate atmospheric light parameters can be obtained from the block, which solves the problem of global atmospheric light estimation distortion; an optimization module for transmittance estimation is constructed, and the refinement of transmittance is realized based on the bilateral weighted guided filtering method, and the problems such as halo caused by the general method are solved. The method proposed by the invention helps to improve the clarity of fire scene image recovery, and provides a data basis for improving the scene detection, identification, environment mapping and path planning of rescue robots.

Description of drawings

Figure 1 is a flow chart of image clarity at the rescue fire scene;

Fig. 2 is the original image in the embodiment of the present invention;

Fig. 3 is a fire light area segmentation module and a fire light area diagram after binarization;

Fig. 4 is a rough transmittance estimation diagram;

FIG. 5 is a fine transmittance estimation diagram after bilateral weighted guided filtering processing.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments.

1. The present invention includes an image clearing system for rescue fire scene. Including regional atmospheric light estimation module, transmittance estimation optimization module and image reconstruction recovery module. The regional atmospheric light estimation module includes a fire regional atmospheric light estimation module and a global atmospheric light estimation module.

2. As shown in Figure 1, the input original image I estimates the atmospheric light through the regional atmospheric light estimation module, designs the regional threshold segmentation algorithm to segment the flame area, further divides the image superpixels and designs the atmospheric light detection operator to obtain the global atmospheric light. ; Build a transmittance estimation and optimization module, estimate the rough transmittance, and further calculate the refined transmittance based on the bilateral weighted guided filtering method; The input image restoration module restores high-quality pictures.

3. The atmospheric light estimation module in the flare area divides the area I of the original image shown in FIG. 2 into a flare area image _IF and a non-fire area image I _NF .

Specifically include:

Step 1: Combine the original image I with the RGB color space criterion and the HLS color space criterion to divide the fire area to obtain an area image I ₁ .

The RGB color space criterion is:

R> _RT

R≥G≥B

The HLS color space criterion is:

L _min ≤L≤L _max

R, G, B are the red, green and blue components of the image RGB color space, R _T is the red threshold, S, L are the saturation and luminance components of the image HLS color space, L _min is the minimum luminance threshold, L _max is the maximum brightness threshold.

Step 2: Perform the morphological processing of opening first and closing later on the region image I ₁ , delete the isolated points in the image and fill the holes inside the target area, and further perform Gauss filtering on the result of the image morphological processing to obtain the image shown in Figure 3. The flare area image _IF and the non-flare area image I _NF .

4. The global atmospheric light estimation module calculates the global atmospheric light value A ₀ by constructing an atmospheric light detection operator.

Specifically include:

Step 1: the R, G, B three channels of the input image I carry out minimum filtering to obtain the dark channel map I _d of the image;

Step 2: Input image I and design the atmospheric light detection operator score:

score=(1-S) _Id

where S is the saturation component of image I.

Step 3: Segment the non-fire area image I _NF superpixels, obtain the segmentation map I _s , and calculate the atmospheric light detection operator score of each _si ∈ I _s superpixel block

为超像素块s_i像素点个数。in,

is the number of pixels in the superpixel block _si .

值按由大到小的顺序进行排序，选取score值最大的超像素块s_max将该超像素中所有像素点的平均值计算大气光最大值的估计值A₀。Step 4: Put the

The values are sorted in descending order, and the superpixel block s _max with the largest score value is selected to calculate the estimated value A ₀ of the maximum atmospheric light value from the average value of all pixels in the superpixel.

为超像素块s_max像素点个数。in,

is the number of pixels in the superpixel block s _max .

5. The transmittance estimation optimization module, input the dark channel map and the global atmospheric light estimation value to estimate the rough transmittance, and optimize the precise transmittance t by the bilateral weighted guided filter transmittance.

Specifically include:

Step 1: Calculate the adaptive confidence t ^* (x) through the L channel of the image:

Among them, Ω is the minimum filter interval, and p is the confidence adjustment parameter:

Step 2: As shown in Figure 4, the rough transmittance characteristic map t ₀ of the image is calculated, and the calculation formula is as follows:

Step 3: As shown in Figure 5, use bilateral weighted guided filtering to optimize t ₀ to estimate the accurate transmittance t of the image. Calculated as follows:

t=a* _Ig +b

a, b are bilateral weighted guidance coefficients, and the calculation formula is as follows:

Among them, ε is the tolerance factor, d is the filter interval, ω(i,j,k,l) is the filter weighting coefficient, (k,l) represents the center coordinate of the window, (i,j) represents the other coordinates of the window, ω _m is the kernel function, and the value of m is {1, 2, 3, 4}. The calculation formula is as follows:

Among them, σ _d is the air domain weight, and σ _r is the value domain weight.

Among them, _Im (i, j), _Im (k, l) represent the pixel value of the corresponding image point, and the calculation formula is as follows:

I ₂ =t ₀

I ₃ =Ig

表示矩阵哈达马积。

represents the matrix Hadamard product.

6. The image restoration module outputs the final clear image J, which is expressed as:

Compared with other methods, the present invention has the advantages of high recovery quality for low-quality images of fire scenes, while only occupying very few computing resources, and is effectively applied to image preprocessing work such as fire scene detection and recognition of rescue robots.

Claims (6)

1. A fire scene low-quality image recovery system for a rescue robot is characterized by comprising a regional atmosphere light estimation module, a transmissivity estimation optimization module and an image reconstruction recovery module, wherein the regional atmosphere light estimation module comprises a fire region atmosphere light estimation module and a global atmosphere light estimation module;

the device comprises a flare region atmospheric light estimation module, a flare region atmospheric light estimation module and a flare region atmospheric light estimation module, wherein the flare region atmospheric light estimation module is used for dividing an original image into a flare region image and a non-flare region image;

the global atmospheric light estimation module is used for carrying out superpixel segmentation on the image and designing an atmospheric light detection operator to obtain an estimated value of global atmospheric light;

the transmissivity estimation optimization module is used for estimating rough transmissivity and calculating accurate transmissivity based on a bilateral weighting guide filtering method;

and the image reconstruction and recovery module is used for reconstructing and recovering a clear image.

2. The fire scene low-quality image restoration method for the rescue robot using the system of claim 1,

step 1: dividing an original image I into a flare region image I by a flare region atmospheric light estimation module _F And image of non-flare area I _NF ；

Step 2: the global atmospheric light estimation module calculates a global atmospheric light estimation value A by constructing an atmospheric light detection operator ₀ ；

And 3, step 3: inputting the dark channel map and the global atmospheric light estimated value into a transmissivity estimation optimization module to estimate rough transmissivity, and applying bilateral weighting to guide filtering transmissivity optimization to obtain accurate transmissivity t;

and 4, step 4: and recovering the image through an image reconstruction and recovery module to obtain a final clear image.

3. The method for recovering the low-quality image of the fire scene of the rescue robot as recited in claim 2, wherein the step 1) is specifically as follows:

1.1 ) combining RGB color space criterion and HLS color space criterion to divide original image I to obtain initial flare area image I ₁ ；

1.2 Image I of the initial flare area ₁ Performing a first-to-open and then-to-close morphological processing, deleting isolated points in the image and filling in the region image I ₁ Further carrying out Gaussian filtering on the morphologically processed image by using the internal holes to obtain a final image I of the flare region _F (ii) a And according to the original image I and the image I of the flare area _F Obtaining images of non-bright areas I _NF 。

4. The method for recovering the low-quality image of the fire scene of the rescue robot as recited in claim 2, wherein the step 2) is specifically as follows:

2.1 Carrying out minimum value filtering on three channels R, G and B of an original image I to obtain a dark channel image I of the image _d ；

2.2 For the original image I), the atmospheric light detection operator score is constructed:

score＝(1-S)I _d

wherein S is a saturation component of the original image I;

2.3 For image I of non-flare area _NF Performing superpixel segmentation to obtain a segmentation map I _s Calculating a segmentation map I _s Each super pixel block s _i Atmospheric light detection operator score of

Wherein,

is a super pixel block s _i The number of the middle pixel points, x is a superpixel block s _i The pixel point in (1);

2.4 Each superpixel block is corresponded with

Sorting the values in the order from big to small, selecting the superpixel block with the largest atmospheric light detection operator score, and recording as s _max Calculating a superpixel block s _max Obtaining an estimated value A of the global atmospheric light by averaging the pixel values of all the pixel points ₀ ：

Wherein,

is a super pixel block s _max The number of middle pixel points, I (x) is the pixel block s _max The pixel value of the middle pixel point x.

5. The method for recovering the low-quality image of the fire scene of the rescue robot as recited in claim 2, wherein the step 3) is specifically as follows:

3.1 Through the original image I) ₁ L channel of (d), calculating an adaptive confidence t ^* (x)：

Wherein Ω is a minimum value filtering interval, p is a confidence coefficient adjusting parameter, and L (y) is an L channel value of a pixel point y in the region Ω;

2) From the original image I ₁ Dark channel diagram I _d And global atmospheric light estimate A ₀ Calculating a rough transmittance profile t of the image ₀ The calculation formula is as follows:

3) Applying a bilateral weighted guided filtering pair t ₀ Optimizing to obtain the image transmissivity t, wherein the calculation formula is as follows：

t＝a*I _g +b

Wherein, I _g The gray scale map of the original image I, a and b are guide filter coefficients, and the calculation formula is as follows:

where ε is a tolerance factor, d is a filter interval, ω (i, j, k, l) is a filter weighting coefficient, (k, l) represents the center coordinates of the filter window, (i, j) represents the other coordinates of the window, ω _m For the kernel function, m takes the value {1,2,3,4}, and the calculation formula is as follows:

wherein σ _d Is a spatial weight, σ _r Is a value range weight;

wherein, I _m (i,j)，I _m (k, l), ε ∈ {1,2,3,4} represents the pixel value of the corresponding image at point (i, j), (k, l), and the calculation formula is as follows:

I ₂ ＝t ₀

I ₃ ＝I _g

wherein,

representing the matrix hadamard product.

6. The method for recovering the low-quality image of the fire scene of the rescue robot as recited in claim 2, wherein in the step 4), the image recovery module outputs a final clear image J represented as:

Images