Image segmentation method and device, vehicle-mounted equipment and special vehicle

Abstract

The invention relates to an image segmentation method and device, vehicle-mounted equipment and a special vehicle, and the method comprises the steps: obtaining a road scene image under a preset illumination condition, carrying out the preprocessing of the road scene image, carrying out the discrete transformation of the preprocessed scene image, and obtaining feature maps under different scales, performing segmentation coding on the preprocessed scene image and the feature maps under different scales, performing conditional coding on the road scene image, combining the feature map after conditional coding and the feature map after segmentation coding, and performing segmentation decoding on the combination result to obtain a decoded scene image; calculating the probability that different pixel points in the decoded scene image belong to the target area, and adjusting the initial segmentation result based on the probability that the different pixel points belong to the target area to obtain a final segmentation result of the target area; according to the invention, the segmentation precision of the target region and the smoothness of the edge of the segmented target region can be improved, and the robustness of the image segmentation method is improved.

Description

Technical Field

[0001] The present application relates to the field of testing technology, and in particular to an image segmentation method, device, vehicle-mounted equipment and special vehicle. Background Art

[0002] Image segmentation is the technology and process of dividing an image into several specific regions with unique properties and proposing objects of interest. Image segmentation is a basic problem in image processing and computer vision. Segmentation algorithms in related technologies, such as threshold segmentation, region growing, clustering, etc., are less effective in dealing with problems such as noise and uneven lighting. In recent years, image segmentation methods have gradually attracted attention. In particular, the segmentation of road scene images of special vehicles (including police cars, military vehicles, traffic enforcement vehicles, ambulances, etc.) under dark light conditions is a challenging problem. Road scene images under dark light conditions often cannot achieve ideal results due to insufficient lighting, low image contrast, and unclear edges (low accuracy of segmentation results, poor smoothness of segmentation edges, etc.).

[0003] Therefore, it is necessary to improve the image segmentation method in the related art. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present application provides an image segmentation method, device, vehicle-mounted equipment and special vehicle to solve the above-mentioned technical problems.

[0005] The present application provides an image segmentation method, which includes: acquiring a road scene image under preset lighting conditions; preprocessing the road scene image, and performing discrete transformation on the preprocessed scene image to obtain feature maps at different scales; the preprocessing includes noise reduction and histogram averaging; segmenting and encoding the preprocessed scene image and the feature maps at different scales; conditionally encoding the road scene image; and combining the conditionally encoded feature map and the segmented encoded feature map; segmenting and decoding the combined result to obtain a decoded scene image, the decoded scene image including an initial segmentation result of a target area; the target area includes a road area; calculating the probability that different pixels in the decoded scene image belong to the target area, and adjusting the initial segmentation result based on the probability that different pixels belong to the target area to obtain a final segmentation result of the target area.

[0006] In one embodiment of the present application, the process of performing discrete transformation on the preprocessed scene image includes: performing a one-dimensional discrete wavelet transform on the preprocessed scene image to obtain a one-dimensional feature map; performing a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map; and using the one-dimensional feature map and the two-dimensional feature map as feature maps at different scales.

[0007] In one embodiment of the present application, the process of performing a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map includes: selecting a target subband from the two-dimensional feature map, performing a two-dimensional discrete wavelet transform on the target subband, and recording the number of transformations; the target subband includes a low-frequency subband, and the frequency of the low-frequency subband is less than or equal to a preset frequency threshold; if the number of transformations is greater than or equal to the preset number threshold, the transformed subband is used as the two-dimensional feature map; if the number of transformations is less than the preset number threshold, the target subband continues to be selected from the transformed subband, the target subband continues to be selected. Discrete wavelet transform, and the number of transformations is updated until the updated number of transformations reaches the preset number threshold, and the subband continues to be transformed as the two-dimensional feature map.

[0008] In one embodiment of the present application, the process of segmenting and encoding the preprocessed scene image and the feature maps at different scales includes: inputting the preprocessed scene image into a preset image segmentation encoder to obtain a first segmented encoded image; element-wise multiplying the first segmented encoded image with the one-dimensional feature map to obtain a first multiplied image; inputting the first multiplied image into the preset image segmentation encoder to obtain a second segmented encoded image; element-wise multiplying the second segmented encoded image with the two-dimensional feature map to obtain a second multiplied image; and inputting the second multiplied image into the preset image segmentation encoder to obtain the segmented encoded feature map.

[0009] In one embodiment of the present application, the process of conditionally encoding the road scene image includes: embedding a preset noise signal into the road scene image to obtain a noise scene image; and reconstructing the noise scene image based on the propagation law of the preset noise signal to obtain a first denoised image; inputting the denoised scene image into a preset conditional encoder to obtain a first conditionally encoded image; inputting the first conditionally encoded image into a preset frequency analyzer to obtain a first frequency analysis result, and removing high-frequency noise from the first frequency analysis result to obtain a second denoised image; the frequency of the high-frequency noise is greater than a preset frequency threshold; inputting the second denoised image into the preset conditional encoder to obtain a second conditionally encoded image; inputting the second conditionally encoded image into the preset frequency analyzer to obtain a second frequency analysis result; and removing high-frequency noise from the second frequency analysis result to obtain a third denoised image; inputting the third denoised image into the preset conditional encoder to obtain the conditionally encoded feature map.

[0010] In one embodiment of the present application, if the decoded scene image also includes a non-target area, the process of adjusting the initial segmentation result based on the probability that different pixels belong to the target area includes: calculating the probability that different pixels transfer between the target area and the non-target area; constructing a joint probability distribution function between different pixels belonging to the target area based on the probability that different pixels belong to the target area and the probability that different pixels transfer between the target area and the non-target area; solving the joint probability distribution function between different pixels belonging to the target area to obtain a final allocation result of different pixels belonging to the target area; adjusting the initial segmentation result according to the final allocation result of different pixels belonging to the target area to obtain a final segmentation result of the target area.

[0011] In one embodiment of the present application, if the initial segmentation result includes initial allocation results of different pixels belonging to the target area, then the process of adjusting the initial segmentation result according to the final allocation results of different pixels belonging to the target area includes: selecting a pixel as the target pixel; if the initial allocation result of the target pixel belonging to the target area is consistent with the final allocation result of the target pixel belonging to the target area, then the initial allocation result of the target pixel belonging to the target area is used as the allocation result of the target pixel; if the initial allocation result of the target pixel belonging to the target area is inconsistent with the final allocation result of the target pixel belonging to the target area, then the final allocation result of the target pixel belonging to the target area is used as the allocation result of the target pixel; continue to select another pixel, compare the initial allocation result of the other pixel belonging to the target area with the final allocation result of the other pixel belonging to the target area, and adjust the initial allocation result of the other pixel belonging to the target area according to the comparison result until the allocation results of all pixels are obtained; combine the allocation results of all pixels to obtain the final segmentation result of the target area.

[0012] According to one aspect of an embodiment of the present application, an image segmentation device is provided, the device comprising: an image acquisition module, for acquiring a road scene image under preset lighting conditions; an image transformation module, for preprocessing the road scene image, and performing discrete transformation on the preprocessed scene image to obtain feature maps at different scales; the preprocessing comprises noise reduction and histogram averaging; an image encoding module, for segmenting and encoding the preprocessed scene image and the feature maps at different scales; conditionally encoding the road scene image; and combining the conditionally encoded feature map and the segmented encoded feature map; segmenting and decoding the combined result to obtain a decoded scene image, the decoded scene image comprising an initial segmentation result of a target area; the target area comprises a road area; a result adjustment module, for calculating the probability that different pixels in the decoded scene image belong to the target area, and adjusting the initial segmentation result based on the probability that different pixels belong to the target area to obtain a final segmentation result of the target area.

[0013] According to one aspect of an embodiment of the present application, a vehicle-mounted device is provided, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the vehicle-mounted device implements the image segmentation method as described above.

[0014] According to one aspect of an embodiment of the present application, a special vehicle is provided, and the special vehicle includes the image segmentation device as described above or the vehicle-mounted equipment as described above.

[0015] The beneficial effects of the present application are as follows: the present application obtains a road scene image under preset lighting conditions, preprocesses the road scene image, and performs discrete transformation on the preprocessed scene image to obtain feature maps at different scales, segment and encodes the preprocessed scene image and the feature maps at different scales, conditionally encodes the road scene image, and combines the conditionally encoded feature map with the segmented encoded feature map, segment and decodes the combined result to obtain a decoded scene image, calculates the probability that different pixels in the decoded scene image belong to a target area, and adjusts the initial segmentation result based on the probability that different pixels belong to the target area to obtain a final segmentation result of the target area. In the above process, after obtaining image features at different scales, the road scene image is segmented based on the preprocessed scene image and the feature maps at different scales, and the initial segmentation result is adjusted based on the probability that different pixels belong to the target area, thereby improving the accuracy of segmenting the target area and the smoothness of the edge of the segmented target area, and at the same time, improving the robustness of the image segmentation method.

[0016] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0018] Figure 1 is a schematic diagram of an exemplary system architecture shown in an exemplary embodiment of the present application;

[0019] Figure 2 is a flowchart of an image segmentation method shown in an exemplary embodiment of the present application;

[0020] Figure 3 is a flowchart of performing discrete transformation on a preprocessed scene image shown in an exemplary embodiment of the present application;

[0021] Figure 4 is a schematic diagram showing segmentation of a road scene image by a diffusion model based on a frequency analyzer according to an exemplary embodiment of the present application;

[0022] Figure 5 is a schematic diagram showing a method of performing frequency analysis on an input image by a frequency analyzer according to an exemplary embodiment of the present application;

[0023] Figure 6 is a flowchart of adjusting the initial segmentation result based on a conditional random field model (Conditional Random Fields, CRF) shown in an exemplary embodiment of the present application;

[0024] Figure 7 is a block diagram of an image segmentation device shown in an exemplary embodiment of the present application;

[0025] Figure 8 A schematic diagram of the structure of a computer system of a vehicle-mounted device suitable for implementing an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] The following will describe the implementation methods of the present application with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for illustrating the present application, not for limiting the scope of protection of the present application.

[0027] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application, and thus the drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed at will, and the component layout may also be more complicated.

[0028] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application. However, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present application difficult to understand.

[0029] Figure 1 It is a schematic diagram of an exemplary system architecture shown in an exemplary embodiment of the present application.

[0030] Reference Figure 1 As shown, the system architecture may include an acquisition device 101 and a vehicle controller 102. The vehicle controller 102 may be at least one of a desktop graphics processing unit (GPU) computer, a GPU computing cluster, a neural network computer, etc. Relevant technicians may use the vehicle controller 102 to obtain a road scene image under a preset lighting condition, preprocess the road scene image, and perform a discrete transformation on the preprocessed scene image to obtain feature maps at different scales, segment and encode the preprocessed scene image and the feature maps at different scales, conditionally encode the road scene image, and combine the conditionally encoded feature map and the segmented encoded feature map, segment and decode the combined result to obtain a decoded scene image, calculate the probability that different pixels in the decoded scene image belong to the target area, and adjust the initial segmentation result based on the probability that different pixels belong to the target area to obtain the final segmentation result of the target area. The acquisition device 101 is used to acquire a road scene image under a preset lighting condition, and after acquiring the road scene image under the preset lighting condition, provide it to the vehicle controller 102 for processing.

[0031] Schematically, after acquiring the road scene image under the preset lighting conditions in the acquisition device 101, the on-board controller 102 preprocesses the road scene image, and performs discrete transformation on the preprocessed scene image to obtain feature maps at different scales, segment and encode the preprocessed scene image and the feature maps at different scales, conditionally encode the road scene image, and combine the conditionally encoded feature map and the segmented encoded feature map, segment and decode the combined result to obtain a decoded scene image, calculate the probability that different pixels in the decoded scene image belong to the target area, and adjust the initial segmentation result based on the probability that different pixels belong to the target area to obtain the final segmentation result of the target area. In the above process, after obtaining image features at different scales, the road scene image is segmented based on the preprocessed scene image and the feature maps at different scales, and the initial segmentation result is adjusted based on the probability that different pixels belong to the target area, thereby improving the accuracy of segmenting the target area and the smoothness of the edge of the segmented target area, and at the same time, improving the robustness of the image segmentation method.

[0032] It should be noted that the image segmentation method provided in the embodiment of the present application is generally executed by the vehicle-mounted controller 102 , and accordingly, the image segmentation device is generally disposed in the vehicle-mounted controller 102 .

[0033] The implementation details of the technical solution of the embodiment of the present application are described in detail below:

[0034] Figure 2 is a flowchart of an image segmentation method shown in an exemplary embodiment of the present application. The image segmentation method can be executed by a computing processing device. The computing processing device can be Figure 1 The vehicle controller 102 shown in FIG. Figure 2 As shown, the image segmentation method at least includes steps S210 to S240, which are described in detail as follows:

[0035] In step S210, a road scene image under a preset lighting condition is acquired.

[0036] In one embodiment of the present application, the road scene images under preset lighting conditions include road scene images taken by a vehicle-mounted camera of a special vehicle when driving at night, and road scene images taken by the vehicle-mounted camera of the special vehicle when the light intensity is less than a preset light intensity threshold. The preset light intensity threshold can be set according to actual conditions and is not specifically limited here.

[0037] In step S220, the road scene image is preprocessed, and the preprocessed scene image is discretely transformed to obtain feature maps at different scales.

[0038] In this embodiment, the preprocessing includes noise reduction and histogram averaging. The process of performing noise reduction on the road scene image includes: performing a two-dimensional convolution operation on the road scene image and a Gaussian filter function to obtain a filtered scene image. The expression of the filtered scene image is as follows:

[0039]

[0040] Among them, G(x,y) represents the scene image after filtering, x and y represent the coordinates of the pixel points respectively, G(u,v) represents the Gaussian kernel in the Gaussian filter function, u and v represent the coordinates of the Gaussian kernel, 2K+1 represents the size of the Gaussian kernel, and I1 represents the road scene image.

[0041] In this embodiment, the expression of the Gaussian kernel is as follows:

[0042]

[0043] Among them, G(u,v) represents the Gaussian kernel in the Gaussian filter function, u and v represent the coordinates of the Gaussian kernel, and σ is the standard deviation of the Gaussian kernel, which is used to control the smoothness of the Gaussian filter.

[0044] In this embodiment, the filtered scene image is subjected to histogram averaging processing to obtain a preprocessed scene image. The expression of the preprocessed scene image is as follows:

[0045]

[0046] Among them, T(r) represents the preprocessed scene image, r represents the grayscale value of the image after denoising, L represents the number of grayscale levels (set to 256), and p r (w) represents the probability of each gray value appearing in the filtered scene image (ie, the probability density function of the gray value), and w represents the probability of the gray value appearing.

[0047] In this embodiment, the road scene image is subjected to Gaussian filtering to remove noise in the road scene image, and the quality, visibility and contrast of the road scene image are further improved after the filtered scene image is subjected to histogram averaging processing, so that the feature maps obtained at different scales contain the frequency information of the road scene image at different spatial resolutions, which is conducive to capturing the structure and texture features of the road scene image, and provides an important basis for the subsequent road scene image segmentation task.

[0048] In step S230, the preprocessed scene image and feature maps at different scales are segmented and encoded; the road scene image is conditionally encoded; the conditionally encoded feature map and the segmented encoded feature map are combined; the combined result is segmented and decoded to obtain a decoded scene image.

[0049] In this embodiment, the decoded scene image includes an initial segmentation result of a target area; the target area includes a road area. The process of segmenting and decoding the combined result is implemented by an image segmentation decoder.

[0050] In this embodiment, the process of combining the conditionally encoded feature map and the segmented encoded feature map includes: performing an addition operation on the conditionally encoded feature map and the segmented encoded feature map to obtain a combination result.

[0051] In step S240, the probabilities that different pixels in the decoded scene image belong to the target area are calculated, and based on the probabilities that different pixels belong to the target area, the initial segmentation result is adjusted to obtain a final segmentation result of the target area.

[0052] In this embodiment, after obtaining image features at different scales, the road scene image is segmented based on the preprocessed scene image and feature maps at different scales, and the initial segmentation result is adjusted based on the probability that different pixels belong to the target area, thereby improving the accuracy of segmenting the target area and the smoothness of the edge of the target area after segmentation, and at the same time, improving the robustness of the image segmentation method.

[0053] In one embodiment of the present application, the process of performing discrete transformation on the preprocessed scene image includes:

[0054] Perform one-dimensional discrete wavelet transform on the preprocessed scene image to obtain a one-dimensional feature map.

[0055] In this embodiment, the expression of one-dimensional discrete wavelet transform is as follows:

[0056]

[0057] Among them, W ψ (j, k) represents a one-dimensional feature map, T(r) represents the preprocessed scene image, r represents the grayscale value of the image after denoising, ψ j,k (r) represents the wavelet basis function used, and is obtained by scaling and translation transformation of the mother wavelet function ψ(t). t represents that the support domain is the time domain. j represents the scale, which is used to control the contraction of the wavelet. k represents the translation, which is used to control the position of the wavelet.

[0058] Perform a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map.

[0059] In this embodiment, the process of performing a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map includes: selecting a target subband from the two-dimensional feature map, performing a two-dimensional discrete wavelet transform on the target subband, and recording the number of transformations; the target subband includes a low-frequency subband, and the frequency of the low-frequency subband is less than or equal to a preset frequency threshold; if the number of transformations is greater than or equal to the preset number threshold, the transformed subband is used as the two-dimensional feature map; if the number of transformations is less than the preset number threshold, the target subband continues to be selected from the transformed subband, the two-dimensional discrete wavelet transform is performed on the continued selected target subband, and the number of transformations is updated until the updated number of transformations reaches the preset number threshold, and the transformed subband continues to be used as the two-dimensional feature map.

[0060] In this embodiment, feature information of different frequency sub-bands is obtained by performing two-dimensional discrete wavelet transform on the preprocessed scene image for multiple times.

[0061] The one-dimensional feature map and the two-dimensional feature map are used as feature maps at different scales.

[0062] In this embodiment, the feature maps at different scales include a one-dimensional feature map and a two-dimensional feature map, which further enriches the frequency information at different spatial resolutions.

[0063] In one embodiment of the present application, the process of performing a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map includes:

[0064] Select the target subband from the two-dimensional feature map, perform two-dimensional discrete wavelet transform on the target subband, and record the number of transformations.

[0065] In this embodiment, the target subband includes a low-frequency subband, which is an approximate subband obtained by transforming the target subband through a two-dimensional discrete wavelet transform. The frequency of the low-frequency subband is less than or equal to a preset frequency threshold, and the preset frequency threshold is determined according to actual conditions.

[0066] If the number of transformations is greater than or equal to a preset number threshold, the transformed sub-band is used as a two-dimensional feature map.

[0067] In this embodiment, the preset number threshold can be set according to actual conditions.

[0068] If the number of transformations is less than the preset threshold, the target subband will continue to be selected from the transformed subband, and the two-dimensional discrete wavelet transform will be performed on the continued selected target subband, and the number of transformations will be updated until the updated number of transformations reaches the preset threshold, and the transformed subband will continue to be used as the two-dimensional feature map.

[0069] In this embodiment, the frequency of the target sub-band that continues to be selected is lower than the frequency of the target sub-band selected last time. The preprocessed scene image is subjected to multiple two-dimensional discrete wavelet transforms through a recursive method to achieve multi-level decomposition of the preprocessed scene image, thereby further enriching the frequency information at different spatial resolutions.

[0070] In one embodiment of the present application, the process of segmenting and encoding the preprocessed scene image and feature maps at different scales includes:

[0071] The preprocessed scene image is input into a preset image segmentation encoder to obtain a first segmented encoded image.

[0072] In this embodiment, the preset image segmentation encoder can be a convolutional neural network model or a deep neural network model, etc., and the preprocessed scene image is input into the preset image segmentation encoder to extract features in the preprocessed scene image to obtain semantic features in the preprocessed scene image.

[0073] The first segmented coded image is element-wise multiplied with the one-dimensional feature map to obtain a first multiplied image.

[0074] In this embodiment, the expression of the first multiplied image is as follows:

[0075]

[0076] Among them, A1(m I1 ,m x1 ) represents the first multiplication image, m x1 represents the first segmented coded image, m I1 represents a one-dimensional feature map, represents element-wise multiplication, and LN(·) represents layer normalization.

[0077] The first multiplied image is input into a preset image segmentation encoder to obtain a second segmentation encoded image.

[0078] In this embodiment, the first multiplied image is input into a preset image segmentation encoder, thereby performing feature extraction on the first multiplied image and obtaining semantic features in the first multiplied image.

[0079] The second segmented encoded image is element-wise multiplied with the two-dimensional feature map to obtain a second multiplied image.

[0080] In this embodiment, the expression of the second multiplied image is as follows:

[0081]

[0082] Among them, A2(m I2 ,m x2) represents the second multiplied image, m x2 represents the second segmented coded image, m I2 represents a two-dimensional feature map, represents element-wise multiplication, and LN(·) represents layer normalization.

[0083] The second multiplied image is input into a preset image segmentation encoder to obtain a segmentation-encoded feature map.

[0084] In this embodiment, the second multiplied image is input into a preset image segmentation encoder, thereby performing feature extraction on the second multiplied image and obtaining semantic features in the second multiplied image.

[0085] In one embodiment of the present application, the process of conditionally encoding a road scene image includes:

[0086] A preset noise signal is embedded into a road scene image to obtain a noise scene image; and based on a propagation law of the preset noise signal, the noise scene image is reconstructed to obtain a first denoised image.

[0087] In this embodiment, the process expression of reconstructing the noisy scene image is as follows:

[0088]

[0089] Among them, p θ (f T-1 |f T ) represents the distribution from variable p θ (f T ) is transformed to the variable distribution p θ (f T-1 ), θ represents the reverse processing parameter, t represents the step value introduced by the noise, T represents the number of iterations, and p θ (f t-1 |f t ) represents the distribution from variable p θ (f t ) is transformed to the variable distribution p θ (f t-1 ), f t represents the noise when the step value is t, f T represents the noise when the number of iterations is T.

[0090] In this embodiment, the variable distribution p θ (f T ) is as follows:

[0091] p θ (f T )=N(f T ;0,I x,y ) Formula (8)

[0092] Among them, p θ (f T ) represents the noise f in the road scene image when the number of iterations is T T , I x,y represents the road scene image, and x and y represent the coordinates of the pixel points in the noise scene image.

[0093] In this embodiment, based on the propagation law of the preset noise signal, the noise scene image is reconstructed, the preset noise signal added to the road scene image and the original noise in the road scene image are removed, and the image quality and clarity are improved.

[0094] The denoised scene image is input into a preset conditional encoder to obtain a first conditional encoded image.

[0095] In this embodiment, the preset condition encoder can be a convolutional neural network model or a deep neural network model, etc. Inputting the denoised scene image into the preset condition encoder can generate feature data related to the input conditions (i.e., pixel labels), thereby increasing the diversity and controllability of the feature data.

[0096] The first conditionally encoded image is input into a preset frequency analyzer to obtain a first frequency analysis result, and high-frequency noise is removed from the first frequency analysis result to obtain a second denoised image.

[0097] In this embodiment, the frequency of the high-frequency noise is greater than a preset frequency threshold, and the preset frequency threshold is set according to actual needs.

[0098] In this embodiment, the expression of the first frequency analysis result is as follows:

[0099]

[0100] Among them, Q k It represents the first frequency analysis result, that is, the spectrum amplitude of the sampling signal q(n) at the discrete frequency ω, n represents the sequence number of the sampling point, N represents the total number of sampling points, ω=0,1,2,...,N-1 represents the discrete frequency index, and the sampling signal q(n) is sampled from the first conditional coded image.

[0101] In this embodiment, the process of removing high-frequency noise from the first frequency analysis result to obtain a second denoised image includes: removing high-frequency noise from the first frequency analysis result to obtain the first frequency analysis result after noise removal, and performing an inverse fast Fourier transform on the first frequency analysis result after noise removal to obtain a second denoised image.

[0102] The second denoised image is input into a preset conditional encoder to obtain a second conditionally encoded image.

[0103] In this embodiment, inputting the second denoised image into the preset condition encoder can further generate feature data related to the input conditions (ie, pixel labels), thereby increasing the diversity and controllability of the feature data.

[0104] The second conditionally encoded image is input into a preset frequency analyzer to obtain a second frequency analysis result; and high-frequency noise is removed from the second frequency analysis result to obtain a third denoised image.

[0105] In this embodiment, the process of removing high-frequency noise from the second frequency analysis result to obtain a third denoised image includes: removing the high-frequency noise from the second frequency analysis result to obtain a second frequency analysis result after noise removal, and performing an inverse fast Fourier transform on the second frequency analysis result after noise removal to obtain a third denoised image, thereby further eliminating the negative impact of the high-frequency noise component, achieving differentiated processing of different image areas, and effectively suppressing noise and detail information in the image.

[0106] The third denoised image is input into a preset conditional encoder to obtain a conditionally encoded feature map.

[0107] In this embodiment, inputting the third denoised image into the preset condition encoder can further generate feature data related to the input conditions (ie, pixel labels), thereby increasing the diversity and controllability of the feature data.

[0108] In one embodiment of the present application, if the decoded scene image also includes a non-target area, the process of adjusting the initial segmentation result based on the probability that different pixels belong to the target area includes:

[0109] Calculate the probability of different pixels transferring between the target area and the non-target area.

[0110] In this embodiment, taking a certain pixel point as an example, the expression of the probability of the pixel point transferring between the target area and the non-target area is as follows:

[0111] V ab (z a ,z b |I2)=μ(z a ,z b )w(I2 a ,I2 b ) Formula (10)

[0112] Among them, V ab (z a ,z b |I2) represents the probability of the ath pixel in the decoded scene image transferring between the target area and the non-target area, z a represents the image feature at the ath pixel, zb represents the image feature at the b-th pixel, I2 represents the decoded scene image, and I2 a Indicates the label at the a-th pixel, and the a-th pixel is located in the target area, a≠b, I2 b represents the label at the b-th pixel, and the b-th pixel is located in the non-target area, μ(z a ,z b ) represents the compatibility function, i.e., the image feature z a With image feature z b The compatibility between a ,I2 b ) represents the weight function of pixel similarity, that is, the similarity weight between the a-th pixel and the b-th pixel.

[0113] Based on the probability that different pixels belong to the target area and the probability that different pixels transfer between the target area and the non-target area, a joint probability distribution function between different pixels belonging to the target area is constructed.

[0114] In this embodiment, taking a certain pixel point as an example, the expression of the probability that the pixel point belongs to the target area is as follows:

[0115] U c (z c |I21)=-logP(z c |I21) Formula (11)

[0116] Among them, U c (z c |I21) represents the probability that the cth pixel belongs to the target area, z c It represents the image feature at the c-th pixel point. The c-th pixel point is located in the target area or the non-target area. I21 represents the label of the target area in the decoded scene image.

[0117] In this embodiment, taking a certain pixel point as an example, the expression of the joint probability distribution function between the pixel point and the target area is as follows:

[0118]

[0119] Among them, P(Z|I2) represents the joint probability distribution function between the c-th pixel point belonging to the target area, C(I2) represents the normalization constant, the probability that the c-th pixel point belongs to the target area, z c represents the image feature at the cth pixel, the cth pixel is located in the target area or the non-target area, I21 represents the label of the target area in the decoded scene image, V ab (z a ,z b|I2) represents the probability of the ath pixel in the decoded scene image transferring between the target area and the non-target area, z a represents the image feature at the ath pixel, z b represents the image feature at the b-th pixel, I2 represents the decoded scene image, the a-th pixel is located in the target area, a≠b, and the b-th pixel is located in the non-target area.

[0120] The joint probability distribution function between different pixels belonging to the target area is solved to obtain the final allocation result of different pixels belonging to the target area.

[0121] In this embodiment, the process of solving the joint probability distribution function between different pixels belonging to the target area can be implemented by referring to the solution method in the relevant technical literature, and is not specifically limited here.

[0122] In this embodiment, taking a certain pixel point as an example, the expression of the final allocation result of the pixel point belonging to the target area is as follows:

[0123]

[0124] in, Indicates the final assignment result of the c-th pixel belonging to the target area, U c (z c |I21) represents the probability that the cth pixel belongs to the target area, z c represents the image feature at the cth pixel, the cth pixel is located in the target area or the non-target area, I21 represents the label of the target area in the decoded scene image, V ab (z a ,z b |I2) represents the probability of the ath pixel in the decoded scene image transferring between the target area and the non-target area, z a represents the image feature at the ath pixel, z b represents the image feature at the b-th pixel, I2 represents the decoded scene image, the a-th pixel is located in the target area, a≠b, the b-th pixel is located in the non-target area, D a Represents the neighborhood of the a-th pixel.

[0125] In this embodiment, the final assignment result of each pixel point belonging to the target area is the region label to which each pixel point belongs when the probability value is the maximum.

[0126] According to the final assignment results of different pixels belonging to the target area, the initial segmentation result is adjusted to obtain the final segmentation result of the target area.

[0127] In this embodiment, the initial segmentation result is adjusted so that the edge of the target area obtained in the initial segmentation result is smoother and the edge division of the target area is more accurate.

[0128] In one embodiment of the present application, if the initial segmentation result includes initial allocation results of different pixels belonging to the target area, then according to the final allocation results of different pixels belonging to the target area, the process of adjusting the initial segmentation result includes:

[0129] Select a pixel as the target pixel.

[0130] In this embodiment, the method for selecting pixel points includes: sorting the pixel points according to rows or columns to obtain sorting results, and sequentially selecting similar points from the sorting results as target similar points.

[0131] If the initial allocation result of the target pixel belonging to the target area is consistent with the final allocation result of the target pixel belonging to the target area, the initial allocation result of the target pixel belonging to the target area is used as the allocation result of the target pixel; if the initial allocation result of the target pixel belonging to the target area is inconsistent with the final allocation result of the target pixel belonging to the target area, the final allocation result of the target pixel belonging to the target area is used as the allocation result of the target pixel.

[0132] In this embodiment, if the initial allocation result of the target pixel point belonging to the target area is consistent with the final allocation result of the target pixel point belonging to the target area, there is no need to adjust the initial allocation result of the target pixel point; if the initial allocation result of the target pixel point belonging to the target area is inconsistent with the final allocation result of the target pixel point belonging to the target area, the final allocation result of the target pixel point belonging to the target area is used as the allocation result of the target pixel point, thereby improving the accuracy of the allocation result of the target pixel point.

[0133] Continue to select another pixel point, compare the initial allocation result of another pixel point belonging to the target area with the final allocation result of another pixel point belonging to the target area, and adjust the initial allocation result of another pixel point belonging to the target area according to the comparison result until the allocation results of all pixels are obtained.

[0134] In this embodiment, the initial allocation result of another pixel point belonging to the target area is compared with the final allocation result of another pixel point belonging to the target area, and the initial allocation results of all pixels belonging to the target area are adjusted according to the comparison result to improve the accuracy of the allocation result of each pixel point.

[0135] The allocation results of all pixels are combined to obtain the final segmentation result of the target area.

[0136] In this embodiment, after the allocation results of all the pixel points are combined, the final segmentation result of the target area obtained has higher segmentation accuracy, higher edge smoothness and higher image clarity than the initial segmentation result of the target area.

[0137] Figure 3 is a flowchart of performing discrete transformation on a preprocessed scene image as shown in an exemplary embodiment of the present application. Figure 3 As shown, the process of performing discrete transformation on the preprocessed scene image includes: (1) the special vehicle obtains a road scene image under dark light conditions (light intensity is less than a preset light intensity threshold) by photographing through an on-board mobile device; (2) a Gaussian filter is used to denoise the road scene image to obtain a filtered scene image; (3) the filtered scene image is subjected to histogram averaging processing to obtain a preprocessed scene image; (4) a one-dimensional discrete wavelet transform is performed on the preprocessed scene image to obtain a one-dimensional feature map; (5) a two-dimensional discrete wavelet transform is performed on the preprocessed scene image to obtain a two-dimensional feature map; (6) the one-dimensional feature map and the two-dimensional feature map are used as feature maps at different scales.

[0138] Figure 4 is a schematic diagram showing an exemplary embodiment of the present application, which shows segmentation of a road scene image by a diffusion model based on a frequency analyzer. Figure 4 As shown, the diffusion model based on the frequency analyzer includes an image segmentation encoder, a conditional encoder, a frequency analyzer and an image segmentation decoder. The image segmentation encoder is used to segment and encode the processed scene image, the first multiplied image and the third multiplied image; the conditional encoder is used to conditionally encode the first denoised image, the second denoised image and the second denoised image; the frequency analyzer is used to analyze the frequency components in the first conditionally encoded image and the second conditionally encoded image; the image segmentation decoder is used to decode the combination result of the conditionally encoded feature map and the segmented encoded feature map.

[0139] Figure 5 is a schematic diagram showing an exemplary embodiment of the present application, in which a frequency analyzer is used to perform frequency analysis on an input image. Figure 5 As shown, the process of performing frequency analysis on an input image through a frequency analyzer includes: inputting a first conditionally coded image into a preset frequency analyzer to obtain a first frequency analysis result, and removing high-frequency noise from the first frequency analysis result to obtain a first frequency analysis result after noise removal; performing an inverse fast Fourier transform on the first frequency analysis result after noise removal to obtain a second denoised image.

[0140] Figure 6 FIG. 1 is a flowchart of adjusting the initial segmentation result based on the conditional random field model shown in an exemplary embodiment of the present application. Figure 6 As shown in FIG. 1 , the process of adjusting the initial segmentation result based on the conditional random field model includes: (1) calculating the probability that different pixels in the decoded scene image belong to the target area; (2) calculating the probability that different pixels transfer between the target area and the non-target area; (3) constructing a joint probability distribution function between different pixels belonging to the target area; (4) solving the joint probability distribution function between different pixels belonging to the target area; (5) adjusting the initial segmentation result according to the final assignment result of different pixels belonging to the target area.

[0141] In this embodiment, after the final segmentation result of the target area is obtained, the conditional random field model parameters are adjusted by calculating the segmentation accuracy and IoU (Intersection-over-Union) of the target area.

[0142] In this embodiment, a road scene image under dark light conditions is obtained by using a special vehicle-mounted mobile device to shoot, and the road scene image is subjected to Gaussian filtering and histogram averaging to achieve denoising of the road scene image and improve the contrast and visibility of the road scene image; and more detailed multi-scale features are obtained by performing one-dimensional discrete wavelet transform and two-dimensional discrete wavelet transform on the preprocessed scene image respectively.

[0143] In this embodiment, the diffusion model based on the frequency analyzer can improve the generalization ability in the dark road scene image segmentation task, and can make full use of the process of inverse noise reconstruction and multi-scale feature fusion to effectively suppress noise-related features while maintaining the segmentation performance.

[0144] In this embodiment, based on the probability that different pixels belong to the target area, the initial segmentation result is adjusted to obtain the final segmentation result of the target area, and then the initial segmentation result is optimized using methods such as conditional random field (CRF) to eliminate noise and smooth the segmentation boundary of the target area.

[0145] The following describes an apparatus embodiment of the present application, which can be used to execute the image segmentation method in the above-mentioned embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, please refer to the above-mentioned embodiment of the image segmentation method of the present application.

[0146] Figure 7 is a block diagram of an image segmentation device shown in an exemplary embodiment of the present application. The device can be applied to Figure 1 The implementation environment shown is specifically configured in the vehicle controller 102. The device may also be applicable to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which the device is applicable.

[0147] like Figure 7 As shown, the exemplary image segmentation device includes:

[0148] The image acquisition module 701 is used to acquire a road scene image under a preset lighting condition.

[0149] The image transformation module 702 is used to preprocess the road scene image and perform discrete transformation on the preprocessed scene image to obtain feature maps at different scales.

[0150] The image encoding module 703 is used to segment and encode the preprocessed scene image and feature maps at different scales; conditionally encode the road scene image; and combine the conditionally encoded feature map and the segmented encoded feature map; segment and decode the combined result to obtain a decoded scene image.

[0151] The result adjustment module 704 is used to calculate the probability that different pixels in the decoded scene image belong to the target area, and adjust the initial segmentation result based on the probability that different pixels belong to the target area to obtain the final segmentation result of the target area.

[0152] In one embodiment of the present application, the road scene images under preset lighting conditions include road scene images taken by a vehicle-mounted camera of a special vehicle when driving at night, and road scene images taken by the vehicle-mounted camera of the special vehicle when the light intensity is less than a preset light intensity threshold. The preset light intensity threshold can be set according to actual conditions and is not specifically limited here.

[0153] In this embodiment, the preprocessing includes noise reduction and histogram averaging. The process of performing noise reduction on the road scene image includes: performing a two-dimensional convolution operation on the road scene image and a Gaussian filter function to obtain a filtered scene image. The expression of the filtered scene image is shown in formula (1).

[0154] In this embodiment, the filtered scene image is subjected to histogram averaging processing to obtain a preprocessed scene image. The expression of the preprocessed scene image is shown in formula (3).

[0155] In this embodiment, the road scene image is subjected to Gaussian filtering to remove noise in the road scene image, and the quality, visibility and contrast of the road scene image are further improved after the filtered scene image is subjected to histogram averaging processing, so that the feature maps obtained at different scales contain the frequency information of the road scene image at different spatial resolutions, which is conducive to capturing the structure and texture features of the road scene image, and provides an important basis for the subsequent road scene image segmentation task.

[0156] In this embodiment, the decoded scene image includes an initial segmentation result of a target area; the target area includes a road area. The process of segmenting and decoding the combined result is implemented by an image segmentation decoder.

[0157] In this embodiment, the process of combining the conditionally encoded feature map and the segmented encoded feature map includes performing an addition operation on the conditionally encoded feature map and the segmented encoded feature map to obtain a combination result.

[0158] In this embodiment, after obtaining image features at different scales, the road scene image is segmented based on the preprocessed scene image and feature maps at different scales, and the initial segmentation result is adjusted based on the probability that different pixels belong to the target area, thereby improving the accuracy of segmenting the target area and the smoothness of the edge of the target area after segmentation, and at the same time, improving the robustness of the image segmentation method.

[0159] It should be noted that the image segmentation device provided in the above embodiment and the image segmentation method provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment, and will not be repeated here. In practical applications, the image segmentation device provided in the above embodiment can distribute the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0160] An embodiment of the present application also provides a vehicle-mounted device, including: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by one or more processors, the vehicle-mounted device implements the image segmentation method provided in the above-mentioned embodiments.

[0161] An embodiment of the present application further provides a special vehicle, which includes an image segmentation device as provided in each of the above embodiments or a vehicle-mounted device as provided in each of the above embodiments.

[0162] Figure 8 The structure diagram of the computer system of the vehicle-mounted device suitable for implementing the embodiment of the present application is shown. It should be noted that: Figure 8 The computer system 800 of the vehicle-mounted device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0163] like Figure 8As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 802 or the program loaded from the storage part 808 to the random access memory (RAM) 803, such as executing the method in the above embodiment. In the RAM 803, various programs and data required for system operation are also stored. The CPU 801, ROM 802 and RAM 803 are connected to each other through a bus 804. The input / output (I / O) interface 805 is also connected to the bus 804.

[0164] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read therefrom is installed into the storage section 808 as needed.

[0165] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication section 809, and / or installed from a removable medium 811. When the computer program is executed by a central processing unit (CPU) 801, various functions defined in the system of the present application are executed.

[0166] It should be noted that the computer-readable medium shown in the embodiment of the present application can be a computer-readable signal medium or a computer-readable medium or any combination of the above two. Computer-readable media can be, for example, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The computer program contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0167] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0168] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.

[0169] Another aspect of the present application provides a computer readable medium having a computer program stored thereon, which, when executed by a processor of a computer, causes the computer to execute the image segmentation method as described above. The computer readable medium may be included in the vehicle-mounted device described in the above embodiment, or may exist independently without being assembled into the vehicle-mounted device.

[0170] Another aspect of the present application also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable medium. A processor of a computer device reads the computer instructions from the computer-readable medium, and the processor executes the computer instructions, so that the computer device executes the image segmentation method provided in each of the above embodiments.

[0171] The above embodiments are merely illustrative of the principles and effects of the present application, and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. An image segmentation method, characterized in that: The method comprises: Acquire a road scene image under preset lighting conditions; Preprocessing the road scene image, and performing discrete transformation on the preprocessed scene image to obtain feature maps at different scales; the preprocessing includes noise reduction and histogram averaging; Segment and encode the preprocessed scene image and the feature maps at different scales; conditionally encode the road scene image; and combine the conditionally encoded feature map and the segmented encoded feature map; segment and decode the combined result to obtain a decoded scene image, wherein the decoded scene image includes an initial segmentation result of a target area; the target area includes a road area; The probabilities that different pixels in the decoded scene image belong to the target area are calculated, and based on the probabilities that different pixels belong to the target area, the initial segmentation result is adjusted to obtain a final segmentation result of the target area.

2. The image segmentation method according to claim 1, characterized in that: The process of performing discrete transformation on the preprocessed scene image includes: Performing a one-dimensional discrete wavelet transform on the preprocessed scene image to obtain a one-dimensional feature map; Performing a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map; The one-dimensional feature map and the two-dimensional feature map are used as feature maps at different scales.

3. The image segmentation method according to claim 2, characterized in that: The process of performing a two-dimensional discrete wavelet transform on the preprocessed scene image to obtain a two-dimensional feature map includes: Selecting a target subband from the two-dimensional feature map, performing a two-dimensional discrete wavelet transform on the target subband, and recording the number of transformations; the target subband includes a low-frequency subband, and the frequency of the low-frequency subband is less than or equal to a preset frequency threshold; If the number of transformations is greater than or equal to a preset number threshold, the transformed sub-band is used as the two-dimensional feature map; If the number of transformations is less than a preset threshold, the target subband continues to be selected from the transformed subband, a two-dimensional discrete wavelet transform is performed on the continuously selected target subband, and the number of transformations is updated until the updated number of transformations reaches the preset threshold, and the transformed subband continues to be used as the two-dimensional feature map.

4. The image segmentation method according to claim 2, characterized in that: The process of segmenting and encoding the preprocessed scene image and the feature maps at different scales includes: Inputting the preprocessed scene image into a preset image segmentation encoder to obtain a first segmented encoded image; Performing element-wise multiplication of the first segmented coded image and the one-dimensional feature map to obtain a first multiplied image; Inputting the first multiplied image into the preset image segmentation encoder to obtain a second segmented encoded image; Performing element-wise multiplication of the second segmented coded image and the two-dimensional feature map to obtain a second multiplied image; The second multiplied image is input into the preset image segmentation encoder to obtain the segmentation-encoded feature map.

5. The image segmentation method according to any one of claims 1 to 4, characterized in that: The process of conditionally encoding the road scene image includes: Embedding a preset noise signal into the road scene image to obtain a noise scene image; and reconstructing the noise scene image based on a propagation law of the preset noise signal to obtain a first denoised image; Inputting the denoised scene image into a preset conditional encoder to obtain a first conditional encoded image; Inputting the first conditionally encoded image into a preset frequency analyzer to obtain a first frequency analysis result, and removing high-frequency noise from the first frequency analysis result to obtain a second denoised image; the frequency of the high-frequency noise is greater than a preset frequency threshold; Inputting the second denoised image into the preset conditional encoder to obtain a second conditionally encoded image; Inputting the second conditionally encoded image into the preset frequency analyzer to obtain a second frequency analysis result; and removing high-frequency noise from the second frequency analysis result to obtain a third denoised image; The third denoised image is input into the preset conditional encoder to obtain the conditionally encoded feature map.

6. The image segmentation method according to any one of claims 1 to 4, characterized in that: If the decoded scene image also includes a non-target area, the process of adjusting the initial segmentation result based on the probability that different pixels belong to the target area includes: Calculating the probability of different pixel points transferring between the target area and the non-target area; Based on the probabilities that different pixels belong to the target area and the probabilities that different pixels transfer between the target area and the non-target area, constructing a joint probability distribution function between different pixels belonging to the target area; Solving the joint probability distribution function between different pixel points belonging to the target area, and obtaining the final distribution result that different pixel points belong to the target area; According to the final assignment results of the different pixels belonging to the target area, the initial segmentation result is adjusted to obtain the final segmentation result of the target area.

7. The image segmentation method according to claim 6, characterized in that: If the initial segmentation result includes initial allocation results of different pixels belonging to the target area, then according to the final allocation results of different pixels belonging to the target area, the process of adjusting the initial segmentation result includes: Select a pixel as the target pixel; If the initial allocation result that the target pixel point belongs to the target area is consistent with the final allocation result that the target pixel point belongs to the target area, the initial allocation result that the target pixel point belongs to the target area is used as the allocation result of the target pixel point; if the initial allocation result that the target pixel point belongs to the target area is inconsistent with the final allocation result that the target pixel point belongs to the target area, the final allocation result that the target pixel point belongs to the target area is used as the allocation result of the target pixel point; Continue to select another pixel point, compare the initial allocation result of the another pixel point belonging to the target area with the final allocation result of the another pixel point belonging to the target area, and adjust the initial allocation result of the another pixel point belonging to the target area according to the comparison result, until the allocation results of all the pixels are obtained; The allocation results of all pixels are combined to obtain the final segmentation result of the target area.

8. An image segmentation device, characterized in that: The device comprises: An image acquisition module, used to acquire road scene images under preset lighting conditions; An image transformation module, used for preprocessing the road scene image and performing discrete transformation on the preprocessed scene image to obtain feature maps at different scales; the preprocessing includes noise reduction and histogram averaging; An image encoding module is used to segment and encode the preprocessed scene image and the feature maps at different scales; conditionally encode the road scene image; and combine the conditionally encoded feature map and the segmented encoded feature map; segment and decode the combined result to obtain a decoded scene image, wherein the decoded scene image includes an initial segmentation result of a target area; the target area includes a road area; The result adjustment module is used to calculate the probability that different pixels in the decoded scene image belong to the target area, and adjust the initial segmentation result based on the probability that different pixels belong to the target area to obtain the final segmentation result of the target area.

9. A vehicle-mounted device, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the vehicle-mounted device to implement the image segmentation method as described in any one of claims 1 to 7.

10. A special vehicle, characterized in that: The special vehicle includes the image segmentation device as claimed in claim 8 or the vehicle-mounted equipment as claimed in claim 9.

Images