Visibility determination method, device and medium combining semantic segmentation and frequency domain analysis

Abstract

The present invention discloses a visibility discrimination method, device and medium combining semantic segmentation and frequency domain analysis, the method comprising: for the same road section, respectively obtaining a first road section image with clear visibility and a second road section image with current visibility; segmenting the edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain each first characteristic part and the second characteristic part; filtering the first characteristic part and the second characteristic part respectively based on frequency domain analysis to obtain a first high-frequency information image and a second high-frequency information image; using an edge detector to respectively extract the first edge pixel in the first high-frequency information image and the second edge pixel in the second high-frequency information image; calculating the ratio of the edge pixel to the total pixels of the characteristic part; calculating the difference between the first ratio and the second ratio, and using the difference as the current visibility value of the road section. The present invention can improve the accuracy of identifying the visibility of the road section.

Description

Technical Field

[0001] The present invention relates to the field of meteorological detection technology, and in particular to a visibility determination method, device and medium combining semantic segmentation and frequency domain analysis. Background Art

[0002] Low visibility due to weather conditions such as haze, rain, snow or dust is an important cause of traffic accidents, especially on highways with high speeds and heavy traffic. The low visibility caused by haze and other weather conditions makes it difficult for drivers to see the situation ahead and around them, resulting in inaccurate estimates of vehicle distance and speed, making it difficult to identify other vehicles, traffic signs or road facilities, and easily causing major traffic accidents such as rear-end collisions. Therefore, it is necessary to develop a method to accurately determine the visibility of a road section, which is of great significance to traffic control and safety.

[0003] Common visibility determination methods include visual inspection by personnel and instrument inspection. However, visual inspection by personnel is not timely, covers a limited range of highway sections, and consumes high labor costs. Instrument inspection mainly uses infrared light and laser equipment, which are expensive and generally sparsely installed (for example, about 20km apart), making it difficult to meet the requirements of large-scale rain and fog weather inspection. Summary of the invention

[0004] The embodiments of the present invention provide a visibility determination method, device and medium combining semantic segmentation and frequency domain analysis, aiming to improve the accuracy of visibility recognition of road sections.

[0005] In a first aspect, an embodiment of the present invention provides a visibility determination method combining semantic segmentation and frequency domain analysis, including:

[0006] For the same road section, respectively obtaining a first road section image with clear visibility and a second road section image with current visibility;

[0007] Segmenting edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain first characteristic parts and second characteristic parts corresponding to each other;

[0008] Based on frequency domain analysis, the first characteristic part and the second characteristic part are filtered respectively to obtain the first high-frequency information image and the second high-frequency information image corresponding to each other;

[0009] Using an edge detector to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image;

[0010] Calculating a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and calculating a second ratio of the second edge pixels to the total pixels of the second characteristic portion;

[0011] The difference between the first ratio and the second ratio is calculated, and the difference is used as the current visibility value of the road segment.

[0012] In a second aspect, an embodiment of the present invention provides a visibility determination device combining semantic segmentation and frequency domain analysis, including:

[0013] An image acquisition unit, used for acquiring, for the same road section, a first road section image with clear visibility and a second road section image with current visibility;

[0014] A semantic segmentation unit, configured to segment edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain first characteristic parts and second characteristic parts corresponding to each other;

[0015] A frequency domain analysis unit, configured to filter the first characteristic portion and the second characteristic portion based on frequency domain analysis to obtain first high-frequency information images and second high-frequency information images corresponding to the first and second characteristics;

[0016] A first extraction unit, configured to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image using an edge detector;

[0017] a ratio calculation unit, configured to calculate a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and to calculate a second ratio of the second edge pixels to the total pixels of the second characteristic portion;

[0018] The difference calculation unit is used to calculate the difference between the first ratio and the second ratio, and use the difference as the current visibility value of the road section.

[0019] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the visibility determination method combining semantic segmentation and frequency domain analysis as described in the first aspect is implemented.

[0020] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the visibility determination method combining semantic segmentation and frequency domain analysis as described in the first aspect is implemented.

[0021] An embodiment of the present invention provides a visibility determination method, device, computer equipment and storage medium combining semantic segmentation and frequency domain analysis. The method includes: for the same road section, respectively obtaining a first road section image with clear visibility and a second road section image with current visibility; segmenting the edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain the first characteristic part and the second characteristic part corresponding to each other; filtering the first characteristic part and the second characteristic part respectively based on frequency domain analysis to obtain the first high-frequency information image and the second high-frequency information image corresponding to each other; using an edge detector to respectively extract the first edge pixels in the first high-frequency information image and the second edge pixels in the second high-frequency information image; calculating the first ratio of the first edge pixels to the total pixels of the first characteristic part, and calculating the second ratio of the second edge pixels to the total pixels of the second characteristic part; calculating the difference between the first ratio and the second ratio, and using the difference as the current visibility value of the road section. The embodiment of the present invention obtains an information-rich high-frequency information image in the road section image by performing semantic segmentation and frequency domain analysis on the road section image, and simultaneously performs pixel ratio on the high-frequency information image to accurately obtain the current road section visibility, thereby improving the accuracy of road section visibility recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0023] Figure 1 A flowchart of a visibility determination method combining semantic segmentation and frequency domain analysis provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of a sub-process of a visibility determination method combining semantic segmentation and frequency domain analysis provided by an embodiment of the present invention;

[0025] Figure 3 A schematic block diagram of a visibility determination device combining semantic segmentation and frequency domain analysis provided by an embodiment of the present invention;

[0026] Figure 4 A sub-schematic block diagram of a visibility determination device combining semantic segmentation and frequency domain analysis provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] It should be understood that when used in this specification and the appended claims, the terms "comprises" and "comprising" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0029] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in this specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0030] It should be further understood that the term " / and / " used in this specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0031] Please refer to the following Figure 1 , Figure 1 a visibility discrimination method combining semantic segmentation and frequency domain analysis provided for the embodiments of the present invention, specifically including: steps S101 to S106.

[0032] S101. For the same road section, respectively obtain a first road section image with clear visibility and a second road section image with current visibility;

[0033] S102. Respectively segment the edge information in the first road section image and the second road section image through a semantic segmentation network to obtain corresponding first feature parts and second feature parts;

[0034] S103. Based on frequency domain analysis, respectively filter the first feature part and the second feature part to obtain corresponding first high-frequency information images and second high-frequency information images;

[0035] S104. Use an edge detector to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image;

[0036] S105, calculating a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and calculating a second ratio of the second edge pixels to the total pixels of the second characteristic portion;

[0037] S106: Calculate the difference between the first ratio and the second ratio, and use the difference as the current visibility value of the road section.

[0038] In this embodiment, for two road section images with different visibility obtained on the same road section, the regions of interest (such as buildings, etc.) with rich edge information in each image are respectively obtained through semantic segmentation technology to obtain the first feature part and the second feature part. Then, the high-frequency information in the first feature part and the second feature is filtered out by frequency domain analysis to obtain the corresponding first high-frequency information image and the second high-frequency information image. Then, an edge detector is used to obtain the edge pixels therein, and the pixel ratio is further calculated. Then, by comparing the pixel ratios of the two (i.e., the first ratio and the second ratio), the current visibility of the road section can be obtained.

[0039] This embodiment can accurately and effectively identify the visibility of a large range of rainy and foggy weather sections, and timely feedback the identification data, providing important protection for the safe driving of vehicles. In addition, when acquiring the road section image, this embodiment can directly use the monitoring camera installed on the highway section to perform visibility identification, so that there is no need to rely on other auxiliary components such as manpower, equipment, and instruments. At the same time, it can effectively monitor the visibility of each section of the highway, reduce the installation cost, and make the coverage of visibility identification wider.

[0040] In one embodiment, the step S102 includes:

[0041] Collect images of different road sections in advance to build a data set, and annotate the characteristic parts of each road section image in the data set;

[0042] Using the data set to train the HRNet network;

[0043] The edge information in the first road section image and the second road section image is segmented by using the trained HRNet network.

[0044] In this embodiment, firstly, images of each road section are collected, and characteristic parts such as roads, vegetation, buildings and sky in the images are annotated, so as to construct a data set including training set images, test set images and validation set images, and then the HRNet network is used to train the distinguished data set, so that the HRNet network can identify the pixels belonging to characteristic parts such as roads, vegetation, buildings, sky, etc. in the road section images, thereby obtaining the segmented images of each characteristic part, so as to complete the network training. Of course, in other embodiments, it can also be combined with public data sets, such as Cityscapes data sets for training, and the trained semantic segmentation model can be used to accurately segment characteristic parts such as roads, vegetation, buildings and sky in the video screen, so as to extract the parts with rich edge information (i.e., the first characteristic part and the second characteristic part), such as buildings containing rich edges, and the extracted partial images are subjected to subsequent processing.

[0045] In one embodiment, if Figure 2 As shown, the step S103 includes: steps S201 to S203.

[0046] S201, transforming the first characteristic part and the second characteristic part from the spatial domain to the frequency domain respectively by Fourier transform, to obtain a first frequency domain image and a second frequency domain image corresponding to each other;

[0047] S202, using a high-pass filter to filter out low-frequency information in the first frequency domain image and the second frequency domain image respectively;

[0048] S203 . Convert the first frequency domain image and the second frequency domain image from which low-frequency information is filtered out from the frequency domain to the spatial domain through inverse Fourier transform to obtain the first high-frequency information image and the second high-frequency information image.

[0049] In this embodiment, the first characteristic part and the second characteristic part are processed using a Fourier transform function, that is, the first characteristic part and the second characteristic part are converted from the spatial domain to the frequency domain through Fourier transform, and then a high-pass filter is used to filter the frequency domain image to filter out the low-frequency information and retain the high-frequency information. Then, the frequency domain image that only retains the high-frequency information is converted from the frequency domain to the spatial domain image through an inverse Fourier transform to obtain the first high-frequency information image and the second high-frequency information image.

[0050] Here, after the first characteristic part and the second characteristic part are transformed into the first frequency domain image and the second frequency domain image by Fourier transform, since the central part of the image is a low-frequency area and the surrounding part is a high-frequency area, a high-pass filter is used to filter the middle low-frequency area and retain the high-frequency area.

[0051] Furthermore, in one embodiment, step S201 includes:

[0052] According to the following formula, the first characteristic part is transformed from the spatial domain to the frequency domain using a two-dimensional discrete Fourier transform:

[0053]

[0054] Wherein, (M, N) represents the two-dimensional image dimension of the first feature part, (u, v) represents the position of a point in the first feature part, j represents the imaginary unit, and j 2 =-1.

[0055] In one embodiment, the step S104 includes:

[0056] The first edge pixels in the first high-frequency information image and the second edge pixels in the second high-frequency information image are respectively extracted using a Canny edge detection operator.

[0057] In this embodiment, the Canny edge detection operator is used to extract the first edge pixel and the second edge pixel by calculating the speed of change of the grayscale value of the image pixel. The steps of extracting pixels using the Canny edge detection operator specifically include: Gaussian smoothing the input image to reduce the error rate; calculating the gradient amplitude and direction to estimate the edge strength and direction at each point; and performing non-maximum suppression on the gradient amplitude according to the gradient direction. In essence, it is a further refinement of the results of operators such as Sobel and Prewitt; using double threshold processing and connecting edges.

[0058] In one embodiment, the step S105 includes:

[0059] Set the pixel value range to (0,1);

[0060] Calculate a first pixel sum and a second pixel sum corresponding to the first edge pixel and the second edge pixel respectively;

[0061] The first pixel sum and the second pixel sum are divided by the total pixels of the first characteristic part and the total pixels of the second characteristic part respectively, to obtain the first ratio and the second ratio respectively.

[0062] In this embodiment, the pixel value range is set to be between 0 and 1, wherein the closer the pixel value is to 1, the higher the degree of high-frequency information is. Therefore, the pixel values ​​of the first edge pixel and the second edge pixel are summed respectively, and the obtained first pixel sum and second pixel sum can each represent the corresponding amount of high-frequency information. Then, they are divided by the total number of pixels respectively to obtain the ratio of the high-frequency information edge image pixels to the pixels of the original image, that is, the first ratio and the second ratio.

[0063] In one embodiment, the step of training the HRNet network using the data set includes:

[0064] For each road section image in the data set, the road section image is input into a high-resolution module of the HRNet network, and the resolution of the road section image is reduced by using multiple convolutional layers in the high-resolution module;

[0065] The road section image is upsampled through the low-resolution module in the HRNet network to improve the resolution of the road section image in the low-resolution module;

[0066] The road section image output by the high-resolution module and the road section image output by the low-resolution module are subjected to feature fusion, and the feature fusion result is output as the segmentation result, thereby completing the training.

[0067] In the HRNet network of this embodiment, starting from the high-resolution subnetwork as the first stage, the high-resolution to low-resolution subnetworks are gradually added to form more stages, and the multi-resolution subnetworks are connected in parallel. In the whole process, multi-scale repeated fusion is performed by repeatedly exchanging information on the parallel multi-resolution subnetworks, so that each high-resolution to low-resolution representation repeatedly receives information from other parallel representations, thereby obtaining rich high-resolution representations. Specifically, the high-resolution module needs to be reduced to the same resolution as the low-resolution module by one or several consecutive 3x3 convolutions with stride=2 (2 consecutive 3x3 convolutions with stride=2 are 4 times downsampling), and then element wise sum is used to sum different resolutions. The low-resolution module is first upgraded to the same resolution as the high-resolution module by an upsampling (Upsample, using nearest neighbor interpolation, using 2 times or 4 times the upsampling rate), and then a 1x1 convolution is used to make the number of channels consistent with the high resolution, and then a sum operation is performed, and then the segmentation result of the HRNet network is obtained by feature fusion, thereby completing the network training.

[0068] Figure 3 A schematic block diagram of a visibility determination device 300 combining semantic segmentation and frequency domain analysis provided in an embodiment of the present invention, the device 300 includes:

[0069] An image acquisition unit 301 is used to acquire, for the same road section, a first road section image with clear visibility and a second road section image with current visibility;

[0070] A semantic segmentation unit 302, configured to segment edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain first characteristic parts and second characteristic parts corresponding to each other;

[0071] A frequency domain analysis unit 303 is used to filter the first characteristic part and the second characteristic part respectively based on frequency domain analysis to obtain the first high-frequency information image and the second high-frequency information image corresponding to each other;

[0072] A first extraction unit 304 is used to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image using an edge detector;

[0073] A ratio calculation unit 305, configured to calculate a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and to calculate a second ratio of the second edge pixels to the total pixels of the second characteristic portion;

[0074] The difference calculation unit 306 is used to calculate the difference between the first ratio and the second ratio, and use the difference as the current visibility value of the road section.

[0075] In one embodiment, the semantic segmentation unit 302 includes:

[0076] A feature annotation unit is used to collect images of different road sections in advance to construct a data set, and to annotate the feature parts of each road section image in the data set;

[0077] A network training unit, used for training the HRNet network using the data set;

[0078] An information segmentation unit is used to segment edge information in the first road section image and the second road section image through a trained HRNet network.

[0079] In one embodiment, if Figure 4 As shown, the frequency domain analysis unit 303 includes:

[0080] A transform unit 401 is used to transform the first characteristic part and the second characteristic part from the spatial domain to the frequency domain by Fourier transform, so as to obtain a first frequency domain image and a second frequency domain image corresponding to each other;

[0081] A filtering unit 402, configured to use a high-pass filter to filter out low-frequency information in the first frequency domain image and the second frequency domain image respectively;

[0082] The inverse transform unit 403 is used to transform the first frequency domain image and the second frequency domain image after filtering out the low frequency information from the frequency domain to the spatial domain through inverse Fourier transform to obtain the first high frequency information image and the second high frequency information image.

[0083] In one embodiment, the first extraction unit 304 includes:

[0084] The second extraction unit is used to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image by using a Canny edge detection operator.

[0085] In one embodiment, the ratio calculation unit 305 includes:

[0086] Pixel setting unit, used to set the pixel value range (0,1);

[0087] A pixel sum calculation unit, configured to calculate a first pixel sum and a second pixel sum corresponding to the first edge pixel and the second edge pixel respectively;

[0088] The pixel dividing unit is used to divide the first pixel sum and the second pixel sum by the total pixels of the first characteristic part and the total pixels of the second characteristic part respectively, so as to obtain the first ratio and the second ratio respectively.

[0089] In one embodiment, the transform unit 401 includes:

[0090] The two-dimensional discrete transformation unit is used to transform the first feature part from the spatial domain to the frequency domain by using a two-dimensional discrete Fourier transform according to the following formula:

[0091]

[0092] Wherein, (M, N) represents the two-dimensional image dimension of the first feature part, (u, v) represents the position of a point in the first feature part, j represents the imaginary unit, and j 2 =-1.

[0093] In one embodiment, the network training unit includes:

[0094] A resolution reduction unit, for each road section image in the data set, inputting the road section image into a high-resolution module of the HRNet network, and reducing the resolution of the road section image by using a plurality of convolutional layers in the high-resolution module;

[0095] A resolution improvement unit, used for performing up-sampling processing on the road section image through the low-resolution module in the HRNet network, so as to improve the resolution of the road section image in the low-resolution module;

[0096] The feature fusion unit is used to perform feature fusion on the road section image output by the high-resolution module and the road section image output by the low-resolution module, and output the feature fusion result as the segmentation result, so as to complete the training.

[0097] Since the embodiments of the apparatus part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the apparatus part, which will not be repeated here.

[0098] The embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed, the steps provided in the above embodiment can be implemented. The storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0099] The embodiment of the present invention also provides a computer device, which may include a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps provided in the above embodiment may be implemented. Of course, the computer device may also include various network interfaces, power supplies and other components.

[0100] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of this application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of this application.

[0101] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. A visibility discrimination method combining semantic segmentation and frequency domain analysis, It is characterized in that include: For the same road section, respectively obtaining a first road section image with clear visibility and a second road section image with current visibility; Segmenting edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain first characteristic parts and second characteristic parts corresponding to each other; Based on frequency domain analysis, the first characteristic part and the second characteristic part are filtered respectively to obtain the first high-frequency information image and the second high-frequency information image corresponding to each other; Using an edge detector to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image; Calculating a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and calculating a second ratio of the second edge pixels to the total pixels of the second characteristic portion; The difference between the first ratio and the second ratio is calculated, and the difference is used as the current visibility value of the road segment.

2. The visibility determination method combining semantic segmentation and frequency domain analysis according to claim 1, It is characterized in that The segmenting of the edge information in the first road section image and the second road section image by the semantic segmentation network to obtain the first characteristic part and the second characteristic part corresponding to each other comprises: Collect images of different road sections in advance to build a data set, and annotate the characteristic parts of each road section image in the data set; Using the data set to train the HRNet network; The edge information in the first road section image and the second road section image is segmented by using the trained HRNet network.

3. The visibility determination method combining semantic segmentation and frequency domain analysis according to claim 1, It is characterized in that The filtering of the first characteristic part and the second characteristic part based on the frequency domain analysis to obtain the first high-frequency information image and the second high-frequency information image corresponding to each other comprises: The first characteristic part and the second characteristic part are respectively transformed from the spatial domain to the frequency domain by Fourier transform to obtain the first frequency domain image and the second frequency domain image corresponding to each other; Using a high-pass filter to filter out low-frequency information in the first frequency domain image and the second frequency domain image respectively; The first frequency domain image and the second frequency domain image from which low-frequency information is filtered are converted from the frequency domain to the spatial domain by inverse Fourier transform to obtain the first high-frequency information image and the second high-frequency information image.

4. The visibility determination method combining semantic segmentation and frequency domain analysis according to claim 1, It is characterized in that The step of respectively extracting first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image by using an edge detector comprises: The first edge pixels in the first high-frequency information image and the second edge pixels in the second high-frequency information image are respectively extracted using a Canny edge detection operator.

5. The visibility determination method combining semantic segmentation and frequency domain analysis according to claim 1, It is characterized in that The calculating a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and the calculating a second ratio of the second edge pixels to the total pixels of the second characteristic portion, include: Set the pixel value range to (0,1); Calculate a first pixel sum and a second pixel sum corresponding to the first edge pixel and the second edge pixel respectively; The first pixel sum and the second pixel sum are divided by the total pixels of the first characteristic part and the total pixels of the second characteristic part respectively, to obtain the first ratio and the second ratio respectively.

6. The visibility determination method combining semantic segmentation and frequency domain analysis according to claim 3, It is characterized in that The first characteristic part and the second characteristic part are respectively transformed from the spatial domain to the frequency domain by Fourier transform to obtain the first frequency domain image and the second frequency domain image corresponding to each other, including: According to the following formula, the first characteristic part is transformed from the spatial domain to the frequency domain using a two-dimensional discrete Fourier transform: Wherein, (M, N) represents the two-dimensional image dimension of the first feature part, (u, v) represents the position of a point in the first feature part, j represents the imaginary unit, and j 2 =-1.

7. The visibility determination method combining semantic segmentation and frequency domain analysis according to claim 2, It is characterized in that The step of training the HRNet network using the data set includes: For each road section image in the data set, the road section image is input into a high-resolution module of the HRNet network, and the resolution of the road section image is reduced by using multiple convolutional layers in the high-resolution module; The road section image is upsampled through the low-resolution module in the HRNet network to improve the resolution of the road section image in the low-resolution module; The road section image output by the high-resolution module and the road section image output by the low-resolution module are subjected to feature fusion, and the feature fusion result is output as the segmentation result, thereby completing the training.

8. A visibility determination device combining semantic segmentation and frequency domain analysis, It is characterized in that include: An image acquisition unit, used for acquiring, for the same road section, a first road section image with clear visibility and a second road section image with current visibility; A semantic segmentation unit, configured to segment edge information in the first road section image and the second road section image respectively through a semantic segmentation network to obtain first characteristic parts and second characteristic parts corresponding to each other; A frequency domain analysis unit, configured to filter the first characteristic portion and the second characteristic portion based on frequency domain analysis to obtain first high-frequency information images and second high-frequency information images corresponding to the first and second characteristics; A first extraction unit, configured to respectively extract first edge pixels in the first high-frequency information image and second edge pixels in the second high-frequency information image using an edge detector; a ratio calculation unit, configured to calculate a first ratio of the first edge pixels to the total pixels of the first characteristic portion, and to calculate a second ratio of the second edge pixels to the total pixels of the second characteristic portion; The difference calculation unit is used to calculate the difference between the first ratio and the second ratio, and use the difference as the current visibility value of the road section.

9. A computer device, It is characterized in that The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the visibility determination method combining semantic segmentation and frequency domain analysis as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the visibility determination method combining semantic segmentation and frequency domain analysis as described in any one of claims 1 to 7 is implemented.

Images