A road recognition method and system based on intelligent lamp poles

Abstract

This application discloses a road recognition method and system based on intelligent lampposts. This system utilizes machine vision-based artificial intelligence monitoring technology to enhance the clarity of road monitoring images and extract the multi-scale implicit correlation feature distribution information in a high-dimensional space. This information is then used to determine road congestion levels. In other words, artificial intelligence technology is applied to smart transportation to construct a road recognition solution that can measure road congestion levels. This approach enables accurate and intelligent recognition of real-time road conditions, meeting travel needs and ensuring travel safety.

Description

Technical Field

[0001] The present application relates to the field of intelligent recognition technology, and more specifically, to a road recognition method and system based on intelligent lamp poles. Background Art

[0002] With the increasing intelligence of cities, people need accurate road information before and during travel to improve safety. Existing solutions rely on staff or users to input and update road information to ensure its timeliness. However, this approach, which relies on staff and users to input road information, suffers from incomplete and inaccurate real-time road information, as well as user inconvenience, failing to meet users' demand for road information.

[0003] Therefore, an optimized road information recognition solution is expected. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a road recognition method and system based on smart lamp poles. It adopts artificial intelligence monitoring technology based on machine vision to enhance the clarity of road monitoring images, extract the multi-scale implicit correlation feature distribution information in high-dimensional space, and use this to judge the degree of road congestion. That is, artificial intelligence technology is applied to smart transportation to construct a road recognition solution to detect the level of road congestion. In this way, the real-time situation of the road can be accurately and intelligently identified to meet people's travel needs and ensure travel safety.

[0005] According to one aspect of the present application, a road recognition method based on an intelligent lamp pole is provided, which includes:

[0006] Obtain road monitoring images collected by cameras deployed on smart light poles;

[0007] Passing the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image;

[0008] The enhanced road monitoring image is passed through a first convolutional neural network model using a dilated convolution kernel with a first dilation rate to obtain a first-scale road monitoring feature map;

[0009] The enhanced road monitoring image is passed through a second convolutional neural network model using a dilated convolution kernel with a second dilation rate to obtain a second-scale road monitoring feature map;

[0010] fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map;

[0011] Performing feature distribution adjustment on the road monitoring feature map to obtain an optimized road monitoring feature map;

[0012] The optimized road monitoring feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent a grade label of the road congestion degree.

[0013] In the above-mentioned road recognition method based on smart lamp poles, the step of passing the road monitoring image through an image definition enhancement module based on an automatic codec to obtain an enhanced road monitoring image includes:

[0014] performing explicit spatial encoding of the road monitoring image using a convolutional layer by an encoder of the autocodec to obtain image features;

[0015] The image features are deconvolved using a deconvolution layer by a decoder of the autocodec to obtain the enhanced road monitoring image.

[0016] In the above-mentioned road recognition method based on smart lamp poles, the enhanced road monitoring image is subjected to a first convolutional neural network model using a dilated convolution kernel having a first dilation rate to obtain a first-scale road monitoring feature map, including:

[0017] Each layer of the first convolutional neural network model with a dilated convolution kernel of a first dilation rate performs the following operations on the input data in the forward pass of the layer:

[0018] Performing convolution on the input data to obtain a first convolution feature map;

[0019] Performing mean pooling on the first convolutional feature map to obtain a first pooled feature map;

[0020] Performing nonlinear activation on the first pooled feature map to obtain a first activated feature map;

[0021] Among them, the output of the last layer of the first convolutional neural network model with a hole convolution kernel with a first hole rate is the first-scale road monitoring feature map, and the input of the first layer of the first convolutional neural network model with a hole convolution kernel with a first hole rate is the enhanced road monitoring image.

[0022] In the above-mentioned road recognition method based on smart lamp poles, the enhanced road monitoring image is subjected to a second convolutional neural network model using a dilated convolution kernel having a second dilation rate to obtain a second-scale road monitoring feature map, including:

[0023] Each layer of the second convolutional neural network model with the dilated convolution kernel of the second dilation rate performs the following operations on the input data in the forward pass of the layer:

[0024] Performing convolution on the input data to obtain a second convolution feature map;

[0025] Performing mean pooling on the second convolutional feature map to obtain a second pooled feature map;

[0026] Performing nonlinear activation on the second pooled feature map to obtain a second activated feature map;

[0027] Among them, the output of the last layer of the second convolutional neural network model with the hole convolution kernel of the second hole rate is the second-scale road monitoring feature map, and the input of the first layer of the second convolutional neural network model with the hole convolution kernel of the second hole rate is the enhanced road monitoring image.

[0028] In the above-mentioned road recognition method based on smart lamp poles, the feature distribution of the road monitoring feature map is adjusted to obtain an optimized road monitoring feature map, including:

[0029] Performing feature map expansion on the road monitoring feature map to obtain a road monitoring expansion feature vector;

[0030] performing geometric constraint reparameterization of a positive definite normed space on the road monitoring expansion feature vector to obtain an optimized road monitoring expansion feature vector;

[0031] The optimized road monitoring expanded feature vector is reconstructed into the optimized road monitoring feature map.

[0032] In the above-mentioned road recognition method based on smart lamp poles, the road monitoring feature map is expanded to obtain a road monitoring expanded feature vector, including: expanding the road monitoring feature map along a column vector to obtain the road monitoring expanded feature vector.

[0033] In the above-mentioned road recognition method based on smart lamp poles, the road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive definite normed space to obtain an optimized road monitoring expansion feature vector, including:

[0034] The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive normed space using the following formula to obtain an optimized road monitoring expansion feature vector;

[0035] Wherein, the formula is:

[0036]

[0037] Among them, μ and σ are the feature set v i The mean and variance of ∈V, represents the square of the vector's two-norm, ||·|| Frepresents the Frobenius norm of the matrix, V represents the road monitoring expansion eigenvector, V T represents the transpose of the road monitoring expansion feature vector, exp represents the exponential function value with the natural constant e as the base, v i is the eigenvalue of the i-th position of the road monitoring expansion feature vector V, v i ′ is the eigenvalue of the u-th position of the optimized road monitoring expansion feature vector.

[0038] In the above-mentioned road recognition method based on smart lamp poles, the optimized road monitoring feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent a grade label of road congestion level, including:

[0039] Expanding the optimized road monitoring feature map into a classification feature vector according to row vectors or column vectors;

[0040] Performing full-connection encoding on the classification feature vector using multiple fully-connected layers of the classifier to obtain an encoded classification feature vector;

[0041] The encoded classification feature vector is passed through the Softmax classification function of the classifier to obtain the classification result.

[0042] According to another aspect of the present application, a road recognition system based on an intelligent lamp pole is provided, comprising:

[0043] A monitoring image acquisition module is used to acquire road monitoring images collected by cameras deployed on smart light poles;

[0044] an automatic codec module, configured to pass the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image;

[0045] A first convolutional encoding module is configured to obtain a first-scale road monitoring feature map by encoding the enhanced road monitoring image through a first convolutional neural network model using a dilated convolution kernel having a first dilation rate;

[0046] A second convolutional coding module is configured to obtain a second-scale road monitoring feature map by encoding the enhanced road monitoring image through a second convolutional neural network model using a dilated convolution kernel having a second dilation rate;

[0047] a fusion module, configured to fuse the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map;

[0048] an optimization module, configured to adjust the characteristic distribution of the road monitoring characteristic map to obtain an optimized road monitoring characteristic map;

[0049] The classification result generating module is used to pass the optimized road monitoring feature map through a classifier to obtain a classification result, and the classification result is used to represent a grade label of the road congestion degree.

[0050] In the above-mentioned intelligent lamp pole-based road recognition system, the automatic encoding and decoding module includes:

[0051] an encoding unit for performing explicit spatial encoding on the road monitoring image using a convolutional layer through an encoder of the autocodec to obtain image features;

[0052] A decoding unit is configured to perform deconvolution processing on the image features using a deconvolution layer through a decoder of the autocodec to obtain the enhanced road monitoring image.

[0053] Compared to existing technologies, the intelligent lamppost-based road recognition method and system provided in this application utilizes machine vision-based artificial intelligence monitoring technology. By enhancing the clarity of road monitoring images, the system extracts the multi-scale implicit correlation feature distribution information in a high-dimensional space and uses this information to determine road congestion levels. In other words, artificial intelligence technology is applied to smart transportation to construct a road recognition solution, thereby detecting the level of road congestion. This approach enables accurate and intelligent recognition of real-time road conditions, meeting people's travel needs and ensuring travel safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0055] Figure 1 Schematic diagram of a scenario of a road recognition method based on smart lamp poles according to an embodiment of the present application.

[0056] Figure 2 Flowchart of a road recognition method based on smart lamp poles according to an embodiment of the present application.

[0057] Figure 3 Schematic diagram of the architecture of a road recognition method based on smart lamp poles according to an embodiment of the present application.

[0058] Figure 4 4 is a flowchart of the sub-steps of step S120 in the road recognition method based on smart lamp poles according to an embodiment of the present application.

[0059] Figure 54 is a flowchart of the sub-steps of step S170 in the road recognition method based on smart lamp poles according to an embodiment of the present application.

[0060] Figure 6 4 is a block diagram of a road recognition system based on smart lamp poles according to an embodiment of the present application. DETAILED DESCRIPTION

[0061] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0062] Scenario Overview

[0063] As mentioned above, due to the use of staff and users to input road information, there are problems such as incomplete and inaccurate real-time road information and inconvenience for users, which cannot meet the needs of users for obtaining road information during travel. Therefore, an optimized road information recognition solution is needed.

[0064] Currently, deep learning and neural networks are widely used in fields such as computer vision, natural language processing, and speech signal processing. Furthermore, deep learning and neural networks have demonstrated capabilities approaching or even surpassing those of humans in areas such as image classification, object detection, semantic segmentation, and text translation.

[0065] In recent years, the development of deep learning and neural networks has provided new solutions and plans for the intelligent identification of road information in smart transportation.

[0066] Accordingly, considering that when identifying road information, the road congestion situation can be obtained through the road monitoring images collected by the camera. However, considering that there will be a lot of useless interference information in the road monitoring images collected by the camera, this will affect the judgment of the degree of road congestion, and in the process of collecting monitoring images, the collected images may be unclear due to environmental factors such as weather, which brings difficulties to the recognition of road information. Based on this, in the technical solution of this application, artificial intelligence monitoring technology based on machine vision is adopted to enhance the clarity of road monitoring images, and extract the multi-scale implicit correlation feature distribution information in high-dimensional space, and use it to judge the degree of road congestion. That is, artificial intelligence technology is applied to smart transportation to construct a road recognition solution to perform road congestion level detection. In this way, the real-time situation of the road can be accurately and intelligently identified to meet people's travel needs and ensure travel safety.

[0067] Specifically, in the technical solution of the present application, first, a road monitoring image is collected by a camera deployed on an intelligent lamp pole. Next, considering that when identifying road information, environmental factors such as weather and dust will affect the clarity of the road monitoring image, this will cause the features of the road information in the road monitoring image to become blurred, thereby reducing the accuracy of judging road congestion. Therefore, in the technical solution of the present application, it is necessary to further enhance the clarity of the road monitoring image. Specifically, the road monitoring image is passed through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image. In particular, here, the encoder uses a convolutional layer to perform explicit spatial encoding on the road monitoring image to obtain image features; and the decoder uses a deconvolution layer to perform deconvolution processing on the image features to obtain the enhanced road monitoring image.

[0068] Furthermore, a convolutional neural network model with excellent performance in extracting local implicit features of an image is used to perform feature mining of the enhanced road monitoring image. In particular, considering that the enhanced road monitoring image contains interference feature information that is unrelated to the degree of road congestion, and the feature information about the road congestion situation has different degrees of correlation at different positions in the image. That is, the road information feature has a multi-scale associated feature distribution representation in the enhanced road monitoring image. Therefore, in the technical solution of the present application, a convolutional neural network model with dilated convolution kernels having different dilation rates is used to perform feature mining of the enhanced road monitoring image, so as to extract the multi-scale implicit associated feature distribution information about the road congestion situation in the enhanced road monitoring image. Specifically, the enhanced road monitoring image is subjected to a first convolutional neural network model using a dilated convolution kernel having a first dilation rate to obtain a first-scale road monitoring feature map, and the enhanced road monitoring image is subjected to a second convolutional neural network model using a dilated convolution kernel having a second dilation rate to obtain a second-scale road monitoring feature map.

[0069] The first-scale road monitoring feature map and the second-scale road monitoring feature map are then fused to generate a road monitoring feature map. This map represents the characteristic differences in road congestion between road monitoring images at different scales. This road monitoring feature map is then used as a classification feature map for classification processing in a classifier to produce a classification result with a grade label representing the degree of road congestion. This allows for intelligent identification of real-time road conditions to meet people's travel needs.

[0070] In particular, in the technical solution of the present application, here, when fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map, in order to fully utilize the image features of different scales of road vehicle distribution expressed by the first-scale road monitoring feature map and the second-scale road monitoring feature map, it is preferred to directly cascade the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain the road monitoring feature map. However, this will lead to the discretization of the overall feature distribution of the road monitoring feature map, so that the road monitoring feature map has a problem of poor convergence relative to the predetermined multiple class labels of the classifier when classified by the classifier, which will affect the training speed of the classifier and the accuracy of the classification results.

[0071] Therefore, in the technical solution of the present application, the road monitoring expanded feature vector obtained by expanding the road monitoring feature map, for example, represented as V, is subjected to geometric constraint reparameterization in a positive definite normed space, specifically:

[0072]

[0073] μ and σ are the feature sets v i The mean and variance of ∈V, represents the square of the vector's two-norm, ||·|| F represents the Frobenius norm of the matrix, v i and v i ′ are the eigenvalues ​​of the i-th position of the road monitoring expanded feature vector V before and after correction, respectively, and V is in row vector form.

[0074] Here, the geometrically constrained reparameterization of the positive-definite normed space of the road monitoring expanded feature vector V can be based on the projection modulus-length metric relationship of the Bessel inequality. By projecting the square of the vector norm, expressed as an inner product, onto the vector's own associated set space, the distribution set of the vector is made modulus-length constrained within the geometric metric subspace of the normed space with positive definiteness. Based on the geometric constraints of the feature distribution, the distribution space is reparameterized to a bounded positive-definite normed space with a closed subspace. This improves the convergence of the overall feature distribution of the road monitoring expanded feature vector V under multiple predetermined class labels, thereby increasing the training speed of the classifier and the accuracy of the classification results. The optimized road monitoring expanded feature vector is then reconstructed into the optimized road monitoring feature map. This enables accurate and intelligent identification of real-time road conditions to meet people's travel needs and ensure travel safety.

[0075] Based on this, the present application provides a road recognition method based on smart lamp poles, which includes: obtaining a road monitoring image collected by a camera deployed on a smart lamp pole; passing the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image; passing the enhanced road monitoring image through a first convolutional neural network model using a dilated convolution kernel with a first dilation rate to obtain a first-scale road monitoring feature map; passing the enhanced road monitoring image through a second convolutional neural network model using a dilated convolution kernel with a second dilation rate to obtain a second-scale road monitoring feature map; fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map; adjusting the feature distribution of the road monitoring feature map to obtain an optimized road monitoring feature map; and passing the optimized road monitoring feature map through a classifier to obtain a classification result, which is used to represent a level label of road congestion.

[0076] Figure 1 FIG is an application scenario diagram of the road recognition method based on the smart lamp pole according to the embodiment of the present application. Figure 1 As shown, in this application scenario, the data is obtained from the smart light pole (for example, Figure 1 L) of the camera (e.g., Figure 1 C) collected road monitoring images (e.g., Figure 1 Then, the road monitoring image is input to a server (e.g., Figure 1 In the S) shown in , the server is capable of generating a classification result of a grade label for indicating the degree of road congestion based on the smart lamp pole-based road recognition algorithm.

[0077] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0078] Exemplary Methods

[0079] Figure 2 FIG is a flow chart of a road recognition method based on an intelligent lamp pole according to an embodiment of the present application. Figure 2As shown, the road recognition method based on the smart lamp pole according to the embodiment of the present application includes the following steps: S110, obtaining a road monitoring image collected by a camera deployed on the smart lamp pole; S120, passing the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image; S130, passing the enhanced road monitoring image through a first convolutional neural network model using a hole convolution kernel with a first hole rate to obtain a first-scale road monitoring feature map; S140, passing the enhanced road monitoring image through a second convolutional neural network model using a hole convolution kernel with a second hole rate to obtain a second-scale road monitoring feature map; S150, fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map; S160, performing feature distribution adjustment on the road monitoring feature map to obtain an optimized road monitoring feature map; and, S170, passing the optimized road monitoring feature map through a classifier to obtain a classification result, which is used to represent a level label of road congestion.

[0080] Figure 3 FIG is a schematic diagram of the architecture of a road recognition method based on smart lamp poles according to an embodiment of the present application. Figure 3 As shown, in this network architecture, first, a road monitoring image collected by a camera deployed on a smart lamp pole is obtained; then, the road monitoring image is passed through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image; then, the enhanced road monitoring image is passed through a first convolutional neural network model using a dilated convolution kernel with a first dilation rate to obtain a first-scale road monitoring feature map; then, the enhanced road monitoring image is passed through a second convolutional neural network model using a dilated convolution kernel with a second dilation rate to obtain a second-scale road monitoring feature map; then, the first-scale road monitoring feature map and the second-scale road monitoring feature map are fused to obtain a road monitoring feature map; then, the feature distribution of the road monitoring feature map is adjusted to obtain an optimized road monitoring feature map; finally, the optimized road monitoring feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent a level label of road congestion.

[0081] More specifically, in step S110, road surveillance images captured by cameras deployed on smart light poles are obtained. Since road information is input by staff and users, real-time road information is incomplete, inaccurate, and inconvenient for users, failing to meet users' needs for road information during travel. Therefore, machine vision-based artificial intelligence monitoring technology is employed to enhance the clarity of road surveillance images and extract the multi-scale implicit correlation feature distribution information in a high-dimensional space. This information is then used to determine road congestion levels. In other words, artificial intelligence technology is applied to smart transportation to develop a road recognition solution that can be used to detect road congestion levels.

[0082] More specifically, in step S120, the road monitoring image is passed through an automatic codec-based image clarity enhancement module to obtain an enhanced road monitoring image. When identifying road information, environmental factors such as weather and dust can affect the clarity of the road monitoring image, blurring road features in the image and reducing the accuracy of road congestion assessments. Therefore, the technical solution of this application requires further clarity enhancement of the road monitoring image.

[0083] Accordingly, in a specific example, Figure 4 As shown, the road monitoring image is passed through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image, including: S121, performing explicit spatial encoding on the road monitoring image using a convolution layer through an encoder of the automatic codec to obtain image features; and, S122, performing deconvolution processing on the image features using a deconvolution layer through a decoder of the automatic codec to obtain the enhanced road monitoring image.

[0084] A convolutional neural network model with excellent performance in extracting local implicit features of images is used to perform feature mining of the enhanced road monitoring image. In particular, considering that the enhanced road monitoring image contains interference feature information that is unrelated to the degree of road congestion, and the feature information about the road congestion situation has different degrees of correlation at different positions in the image. In other words, the road information features have a multi-scale associated feature distribution representation in the enhanced road monitoring image. Therefore, in the technical solution of the present application, a convolutional neural network model with dilated convolution kernels with different dilation rates is used to perform feature mining of the enhanced road monitoring image, so as to extract the multi-scale implicit associated feature distribution information about the road congestion situation in the enhanced road monitoring image.

[0085] More specifically, in step S130, the enhanced road monitoring image is subjected to a first convolutional neural network model using a dilated convolution kernel having a first dilation rate to obtain a first-scale road monitoring feature map.

[0086] Accordingly, in a specific example, the enhanced road monitoring image is obtained by using a first convolutional neural network model with a dilated convolution kernel having a first dilation rate to obtain a first-scale road monitoring feature map, including: each layer of the first convolutional neural network model with a dilated convolution kernel having a first dilation rate performs the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a first convolutional feature map; mean pooling processing on the first convolutional feature map to obtain a first pooled feature map; and nonlinear activation on the first pooled feature map to obtain a first activation feature map; wherein the output of the last layer of the first convolutional neural network model with a dilated convolution kernel having a first dilation rate is the first-scale road monitoring feature map, and the input of the first layer of the first convolutional neural network model with a dilated convolution kernel having a first dilation rate is the enhanced road monitoring image.

[0087] More specifically, in step S140, the enhanced road monitoring image is subjected to a second convolutional neural network model using a dilated convolution kernel with a second dilation rate to obtain a second-scale road monitoring feature map.

[0088] Accordingly, in a specific example, the enhanced road monitoring image is obtained by using a second convolutional neural network model with a hole convolution kernel having a second hole rate to obtain a second-scale road monitoring feature map, including: each layer of the second convolutional neural network model with a hole convolution kernel having a second hole rate performs the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a second convolution feature map; mean pooling processing on the second convolution feature map to obtain a second pooled feature map; and nonlinear activation on the second pooled feature map to obtain a second activation feature map; wherein the output of the last layer of the second convolutional neural network model with a hole convolution kernel having a second hole rate is the second-scale road monitoring feature map, and the input of the first layer of the second convolutional neural network model with a hole convolution kernel having a second hole rate is the enhanced road monitoring image.

[0089] More specifically, in step S150 , the first-scale road monitoring feature map and the second-scale road monitoring feature map are fused to obtain a road monitoring feature map.

[0090] Accordingly, in a specific example, fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map includes: fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map in a cascade manner to obtain the road monitoring feature map.

[0091] In particular, in the technical solution of the present application, when fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map, in order to fully utilize the image features of different scales of road vehicle distribution expressed by the first-scale road monitoring feature map and the second-scale road monitoring feature map, it is preferred to directly cascade the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain the road monitoring feature map. However, this will lead to the discretization of the overall feature distribution of the road monitoring feature map, so that the road monitoring feature map has a problem of poor convergence relative to the predetermined multiple class labels of the classifier when classified by the classifier, which will affect the training speed of the classifier and the accuracy of the classification results.

[0092] More specifically, in step S160, the feature distribution of the road monitoring feature map is adjusted to obtain an optimized road monitoring feature map. That is, based on the above reasons, in the technical solution of the present application, the road monitoring feature map is expanded to obtain a road monitoring expanded feature vector, for example, represented as V, which is subjected to geometric constraint reparameterization in a positive definite normed space.

[0093] Accordingly, in a specific example, the feature distribution of the road monitoring feature map is adjusted to obtain an optimized road monitoring feature map, including: performing feature map expansion on the road monitoring feature map to obtain a road monitoring expansion feature vector; performing geometric constraint reparameterization of the road monitoring expansion feature vector in a positive definite normed space to obtain an optimized road monitoring expansion feature vector; and reconstructing the optimized road monitoring expansion feature vector into the optimized road monitoring feature map.

[0094] Accordingly, in a specific example, performing feature map expansion on the road monitoring feature map to obtain a road monitoring expanded feature vector includes: expanding the road monitoring feature map along a column vector to obtain the road monitoring expanded feature vector.

[0095] Accordingly, in a specific example, performing geometric constraint reparameterization of a positive normed space on the road monitoring expansion feature vector to obtain an optimized road monitoring expansion feature vector includes:

[0096] The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive normed space using the following formula to obtain an optimized road monitoring expansion feature vector;

[0097] Wherein, the formula is:

[0098]

[0099] Among them, μ and σ are the feature set v i The mean and variance of ∈V, represents the square of the vector's two-norm, ||·|| F represents the Frobenius norm of the matrix, V represents the road monitoring expansion eigenvector, V T represents the transpose of the road monitoring expansion feature vector, exp represents the exponential function value with the natural constant e as the base, v i is the eigenvalue of the i-th position of the road monitoring expansion feature vector V, v i ′ is the eigenvalue of the i-th position of the optimized road monitoring expansion feature vector.

[0100] Here, the geometrically constrained reparameterization of the positive-definite normed space of the road monitoring feature vector V can be based on the projected norm-length metric relationship of the Bessel inequality. By projecting the square of the vector norm, expressed as an inner product, onto the vector's associated set space, the distribution set of the vector is constrained to have a norm-length constraint within the geometric metric subspace of the positive-definite normed space. This reparameterization of the distribution space to a bounded positive-definite normed space with a closed subspace is based on the geometric constraints of the feature distribution. This improves the convergence of the overall feature distribution of the road monitoring feature vector V under multiple predetermined class labels, thereby increasing the training speed of the classifier and the accuracy of the classification results.

[0101] More specifically, in step S170, the optimized road monitoring feature map is passed through a classifier to obtain a classification result, which is used to provide a classification label indicating the degree of road congestion. The optimized road monitoring feature map is then processed by the classifier to obtain a classification result indicating the degree of road congestion. This allows for intelligent identification of real-time road conditions to meet people's travel needs.

[0102] Accordingly, in a specific example, Figure 5As shown, the optimized road monitoring feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent a grade label of the road congestion degree, including: S171, expanding the optimized road monitoring feature map into a classification feature vector according to a row vector or a column vector; S172, using multiple fully connected layers of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and, S173, passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0103] In summary, the smart lamppost-based road recognition method of the present application utilizes machine vision-based artificial intelligence monitoring technology to enhance the clarity of road monitoring images and extract the multi-scale implicit correlation feature distribution information in high-dimensional space. This information is then used to determine road congestion levels. In other words, artificial intelligence technology is applied to smart transportation to construct a road recognition solution, thereby detecting the level of road congestion. This approach enables accurate and intelligent recognition of real-time road conditions, meeting people's travel needs and ensuring travel safety.

[0104] Exemplary Systems

[0105] Figure 6 FIG is a block diagram of a road recognition system 100 based on an intelligent lamp pole according to an embodiment of the present application. Figure 6 As shown, the road recognition system 100 based on the smart lamp pole according to the embodiment of the present application includes: a monitoring image acquisition module 110, which is used to acquire a road monitoring image collected by a camera deployed on the smart lamp pole; an automatic encoding and decoding module 120, which is used to pass the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image; a first convolutional coding module 130, which is used to pass the enhanced road monitoring image through a first convolutional neural network model using a first hole convolution kernel with a first hole rate to obtain a first-scale road monitoring feature map; a second convolutional coding module 140, which is used to convert the road monitoring image into a road monitoring feature map of a first scale; The enhanced road monitoring image is obtained by using a second convolutional neural network model with a dilated convolution kernel having a second dilation rate to obtain a second-scale road monitoring feature map; a fusion module 150 is used to fuse the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map; an optimization module 160 is used to adjust the feature distribution of the road monitoring feature map to obtain an optimized road monitoring feature map; and a classification result generation module 170 is used to pass the optimized road monitoring feature map through a classifier to obtain a classification result, and the classification result is used to represent a grade label of road congestion degree.

[0106] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the automatic encoding and decoding module 120 includes: an encoding unit for explicitly spatially encoding the road monitoring image using a convolutional layer through the encoder of the automatic codec to obtain image features; and a decoding unit for deconvolution processing the image features using a deconvolution layer through the decoder of the automatic codec to obtain the enhanced road monitoring image.

[0107] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the first convolutional encoding module 130 is used to: each layer of the first convolutional neural network model with a hole convolution kernel of a first hole rate performs the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a first convolution feature map; mean pooling processing on the first convolution feature map to obtain a first pooled feature map; and nonlinear activation on the first pooled feature map to obtain a first activation feature map; wherein the output of the last layer of the first convolutional neural network model with a hole convolution kernel of a first hole rate is the first-scale road monitoring feature map, and the input of the first layer of the first convolutional neural network model with a hole convolution kernel of a first hole rate is the enhanced road monitoring image.

[0108] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the second convolutional encoding module 140 is used to: each layer of the second convolutional neural network model with a second hole convolution kernel performs the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a second convolution feature map; mean pooling processing on the second convolution feature map to obtain a second pooled feature map; and nonlinear activation on the second pooled feature map to obtain a second activation feature map; wherein the output of the last layer of the second convolutional neural network model with a second hole convolution kernel is the second-scale road monitoring feature map, and the input of the first layer of the second convolutional neural network model with a second hole convolution kernel is the enhanced road monitoring image.

[0109] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the optimization module 160 includes: a feature map expansion unit, used to perform feature map expansion on the road monitoring feature map to obtain a road monitoring expansion feature vector; a reparameterization unit, used to perform geometric constraint reparameterization of the road monitoring expansion feature vector in a positive definite normed space to obtain an optimized road monitoring expansion feature vector; and a reconstruction unit, used to reconstruct the optimized road monitoring expansion feature vector into the optimized road monitoring feature map.

[0110] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the feature map expansion unit includes: expanding the road monitoring feature map along the column vector to obtain the road monitoring expanded feature vector.

[0111] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the re-parameterization unit includes:

[0112] The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive normed space using the following formula to obtain an optimized road monitoring expansion feature vector;

[0113] Wherein, the formula is:

[0114]

[0115] Among them, μ and σ are the feature set v i The mean and variance of ∈V, represents the square of the vector's two-norm, ||·|| F represents the Frobenius norm of the matrix, V represents the road monitoring expansion eigenvector, V T represents the transpose of the road monitoring expansion feature vector, exp represents the exponential function value with the natural constant e as the base, v i is the eigenvalue of the i-th position of the road monitoring expansion feature vector V, v i ′ is the eigenvalue of the i-th position of the optimized road monitoring expansion feature vector.

[0116] In one example, in the above-mentioned smart lamp pole-based road recognition system 100, the classification result generation module 170 is further used to: expand the optimized road monitoring feature map into a classification feature vector according to a row vector or a column vector; use multiple fully connected layers of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and pass the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0117] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned intelligent lamp pole-based road recognition system 100 have been described in detail above. Figures 1 to 5 The method for road recognition based on intelligent lamp poles has been described in detail, and therefore, its repeated description will be omitted.

[0118] As described above, the smart light pole-based road recognition system 100 according to the embodiments of the present application can be implemented in various wireless terminals, such as a server that implements a smart light pole-based road recognition algorithm. In one example, the smart light pole-based road recognition system 100 according to the embodiments of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the smart light pole-based road recognition system 100 can be a software module within the operating system of the wireless terminal, or an application developed specifically for the wireless terminal. Of course, the smart light pole-based road recognition system 100 can also be one of the many hardware modules of the wireless terminal.

[0119] Alternatively, in another example, the smart lamp pole-based road recognition system 100 and the wireless terminal may also be separate devices, and the smart lamp pole-based road recognition system 100 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0120] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this application are merely illustrative and not restrictive, and it should not be assumed that these advantages, strengths, and effects are required of each embodiment of this application. In addition, the specific details disclosed above are merely illustrative and facilitating understanding, and are not restrictive. The above details do not limit this application to necessarily being implemented using the above specific details.

[0121] The block diagrams of the devices, devices, equipment, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0122] It should also be noted that in the apparatus, device, and method of the present application, each component or each step can be decomposed and / or recombined, and such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0123] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0124] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A road recognition method based on intelligent lamp poles, characterized in that: include: Obtain road monitoring images collected by cameras deployed on smart light poles; Passing the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image; The enhanced road monitoring image is passed through a first convolutional neural network model using a dilated convolution kernel with a first dilation rate to obtain a first-scale road monitoring feature map; The enhanced road monitoring image is passed through a second convolutional neural network model using a dilated convolution kernel with a second dilation rate to obtain a second-scale road monitoring feature map; fusing the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map; Performing feature distribution adjustment on the road monitoring feature map to obtain an optimized road monitoring feature map; Passing the optimized road monitoring feature map through a classifier to obtain a classification result, wherein the classification result is used to represent a grade label of road congestion degree; Adjusting the characteristic distribution of the road monitoring characteristic map to obtain an optimized road monitoring characteristic map includes: Expanding the road monitoring feature map to obtain a road monitoring expanded feature vector; performing geometric constraint reparameterization of a positive definite normed space on the road monitoring expansion feature vector to obtain an optimized road monitoring expansion feature vector; Reconstructing the optimized road monitoring expanded feature vector into the optimized road monitoring feature map; Expanding the road monitoring feature map to obtain a road monitoring expanded feature vector includes: expanding the road monitoring feature map along a column vector to obtain the road monitoring expanded feature vector; The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive definite normed space to obtain an optimized road monitoring expansion feature vector, comprising: The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive normed space using the following formula to obtain the optimized road monitoring expansion feature vector; Wherein, the formula is: in, and is a feature set The mean and variance of represents the square of the vector's two-norm, represents the Frobenius norm of the matrix, represents the road monitoring expansion feature vector, represents the transpose of the road monitoring expansion feature vector, represents the exponential function value with the natural constant e as the base, is the road monitoring expansion feature vector No. The eigenvalues ​​at the positions, is the first feature vector of the optimized road monitoring expansion The eigenvalues ​​at each position.

2. The road recognition method based on intelligent lamp pole according to claim 1 is characterized in that: The step of passing the road monitoring image through an image definition enhancement module based on an automatic codec to obtain an enhanced road monitoring image comprises: performing explicit spatial encoding of the road monitoring image using a convolutional layer by an encoder of the autocodec to obtain image features; The image features are deconvolved using a deconvolution layer by a decoder of the autocodec to obtain the enhanced road monitoring image.

3. The road recognition method based on intelligent lamp pole according to claim 2 is characterized in that: The method of obtaining a first-scale road monitoring feature map by using a first convolutional neural network model with a dilated convolution kernel having a first dilation rate on the enhanced road monitoring image comprises: Each layer of the first convolutional neural network model with a dilated convolution kernel of a first dilation rate performs the following operations on the input data in the forward pass of the layer: Performing convolution on the input data to obtain a first convolution feature map; Performing mean pooling on the first convolutional feature map to obtain a first pooled feature map; Performing nonlinear activation on the first pooled feature map to obtain a first activated feature map; Among them, the output of the last layer of the first convolutional neural network model with a hole convolution kernel with a first hole rate is the first-scale road monitoring feature map, and the input of the first layer of the first convolutional neural network model with a hole convolution kernel with a first hole rate is the enhanced road monitoring image.

4. The road recognition method based on the intelligent lamp pole according to claim 3 is characterized in that: The method of obtaining a second-scale road monitoring feature map by using a second convolutional neural network model with a dilated convolution kernel having a second dilation rate on the enhanced road monitoring image comprises: Each layer of the second convolutional neural network model with the dilated convolution kernel of the second dilation rate performs the following operations on the input data in the forward pass of the layer: Performing convolution on the input data to obtain a second convolution feature map; Performing mean pooling on the second convolutional feature map to obtain a second pooled feature map; Performing nonlinear activation on the second pooled feature map to obtain a second activated feature map; Among them, the output of the last layer of the second convolutional neural network model with the hole convolution kernel of the second hole rate is the second-scale road monitoring feature map, and the input of the first layer of the second convolutional neural network model with the hole convolution kernel of the second hole rate is the enhanced road monitoring image.

5. The road recognition method based on smart lamp pole according to claim 1, characterized in that: The optimized road monitoring feature map is passed through a classifier to obtain a classification result, and the classification result is used to represent a grade label of road congestion level, including: Expanding the optimized road monitoring feature map into a classification feature vector according to row vectors or column vectors; Performing full-connection encoding on the classification feature vector using multiple fully-connected layers of the classifier to obtain an encoded classification feature vector; The encoded classification feature vector is passed through the Softmax classification function of the classifier to obtain the classification result.

6. A road recognition system based on intelligent lamp poles, characterized in that: include: A monitoring image acquisition module is used to acquire road monitoring images collected by cameras deployed on smart light poles; an automatic codec module, configured to pass the road monitoring image through an image clarity enhancement module based on an automatic codec to obtain an enhanced road monitoring image; A first convolutional encoding module is configured to obtain a first-scale road monitoring feature map by encoding the enhanced road monitoring image through a first convolutional neural network model using a dilated convolution kernel having a first dilation rate; A second convolutional coding module is configured to obtain a second-scale road monitoring feature map by encoding the enhanced road monitoring image through a second convolutional neural network model using a dilated convolution kernel having a second dilation rate; a fusion module, configured to fuse the first-scale road monitoring feature map and the second-scale road monitoring feature map to obtain a road monitoring feature map; an optimization module, configured to adjust the characteristic distribution of the road monitoring characteristic map to obtain an optimized road monitoring characteristic map; a classification result generating module, configured to pass the optimized road monitoring feature map through a classifier to obtain a classification result, wherein the classification result is used to indicate a grade label of road congestion degree; Adjusting the characteristic distribution of the road monitoring characteristic map to obtain an optimized road monitoring characteristic map includes: Expanding the road monitoring feature map to obtain a road monitoring expanded feature vector; performing geometric constraint reparameterization of a positive definite normed space on the road monitoring expansion feature vector to obtain an optimized road monitoring expansion feature vector; Reconstructing the optimized road monitoring expanded feature vector into the optimized road monitoring feature map; Expanding the road monitoring feature map to obtain a road monitoring expanded feature vector includes: expanding the road monitoring feature map along a column vector to obtain the road monitoring expanded feature vector; The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive definite normed space to obtain an optimized road monitoring expansion feature vector, comprising: The road monitoring expansion feature vector is subjected to geometric constraint reparameterization in a positive normed space using the following formula to obtain the optimized road monitoring expansion feature vector; Wherein, the formula is: in, and is a feature set The mean and variance of represents the square of the vector's two-norm, represents the Frobenius norm of the matrix, represents the road monitoring expansion feature vector, represents the transpose of the road monitoring expansion feature vector, represents the exponential function value with the natural constant e as the base, is the road monitoring expansion feature vector No. The eigenvalues ​​at the positions, is the first feature vector of the optimized road monitoring expansion The eigenvalues ​​at each position.

7. The road recognition system based on the intelligent lamp pole according to claim 6 is characterized in that: The automatic encoding and decoding module includes: an encoding unit for performing explicit spatial encoding on the road monitoring image using a convolutional layer through an encoder of the autocodec to obtain image features; A decoding unit is configured to perform deconvolution processing on the image features using a deconvolution layer through a decoder of the autocodec to obtain the enhanced road monitoring image.

Images