Traffic sign recognition method and device and storage medium

Abstract

The invention provides a traffic sign recognition method and device and a storage medium, and belongs to the technical field of sign recognition, and the method comprises the steps: importing a plurality of original traffic sign pictures, carrying out the preprocessing of all original traffic sign pictures, and collecting the preprocessing results to obtain a preprocessed traffic sign picture set; dividing the preprocessed traffic sign picture set into a traffic sign training set and a traffic sign test set according to a preset proportion; and constructing a training model, and training the training model through the traffic sign training set to obtain a to-be-processed model. The problem of traffic sign recognition in severe weather can be solved, the reliability and safety of an automobile auxiliary driving system in a complex environment are improved, the life safety of drivers and passengers is guaranteed, high-quality development of the automobile industry and the intelligent network connection technology is promoted, and key technical support is provided for intelligent traffic system construction.

Description

Technical Field

[0001] This invention relates to the field of sign recognition technology, specifically to a traffic sign recognition method, device, and storage medium. Background Technology

[0002] Current traffic sign recognition algorithms struggle to effectively extract key features from sensor data in adverse environments, leading to errors in the recognition of traffic signs, lane lines, and objects. For example, in heavy rain, camera images are interfered with by rainwater, making accurate lane line recognition difficult; millimeter-wave radar accuracy decreases due to moisture, failing to effectively compensate for errors. In low-light conditions, limited image information from cameras hinders accurate traffic sign recognition. Frequent traffic accidents caused by assisted driving system malfunctions due to severe weather fully expose the vulnerability of current traffic sign recognition systems in complex environments. This not only poses a potential threat to the lives of drivers and passengers but also seriously hinders the further promotion and application of assisted driving functions, posing a significant challenge to the development of intelligent driving systems. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a traffic sign recognition method, device and storage medium to address the shortcomings of the prior art.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A traffic sign recognition method, comprising the following steps: Import multiple original traffic sign images, preprocess all the original traffic sign images, and combine the preprocessed results to obtain a preprocessed traffic sign image set. The preprocessed traffic sign image set is divided into a traffic sign training set and a traffic sign test set according to a preset ratio. A training model is constructed, and the training model is trained using the traffic sign training set to obtain the model to be processed; The traffic sign recognition model is obtained by testing and analyzing the model to be processed using the traffic sign test set. Import the traffic sign image to be recognized, and use the traffic sign recognition model to recognize the traffic sign image to obtain the traffic sign recognition result.

[0005] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: A traffic sign recognition device, comprising: The import module is used to import multiple original traffic sign images; The preprocessing module is used to preprocess all the original traffic sign images and combine the preprocessed results to obtain a preprocessed traffic sign image set. The segmentation module is used to divide the preprocessed traffic sign image set into a traffic sign training set and a traffic sign test set according to a preset ratio. The training module is used to build a training model and train the training model using the traffic sign training set to obtain the model to be processed. The test analysis module is used to test and analyze the model to be processed using the traffic sign test set to obtain a traffic sign recognition model. The import module is also used to import images of traffic signs to be recognized; The recognition result acquisition module is used to recognize the traffic sign image to be recognized through the traffic sign recognition model and obtain the traffic sign recognition result.

[0006] Based on the above-mentioned traffic sign recognition method, the present invention also provides a traffic sign recognition system.

[0007] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a traffic sign recognition system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the traffic sign recognition method as described above.

[0008] Based on the above-described traffic sign recognition method, the present invention also provides a computer-readable storage medium.

[0009] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the traffic sign recognition method as described above.

[0010] The beneficial effects of this invention are as follows: By preprocessing the original traffic sign images to obtain a preprocessed traffic sign image set, the preprocessed traffic sign image set is divided into a traffic sign training set and a traffic sign test set according to a preset ratio. The training model is trained using the traffic sign training set to obtain the model to be processed. The traffic sign recognition model is obtained by testing and analyzing the model to be processed using the traffic sign test set. The traffic sign recognition result is obtained by recognizing the traffic sign images to be recognized using the traffic sign recognition model. This invention can overcome the problem of traffic sign recognition in severe weather, improve the reliability and safety of automotive assisted driving systems in complex environments, protect the lives of drivers and passengers, promote the high-quality development of the automotive industry and intelligent connected vehicle technology, and provide key technical support for the construction of intelligent transportation systems. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the traffic sign recognition method provided in an embodiment of the present invention. Figure 2 A schematic diagram of the training model structure for the traffic sign recognition method provided in an embodiment of the present invention; Figure 3 This is a block diagram of a traffic sign recognition device provided in an embodiment of the present invention. Detailed Implementation

[0012] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0013] Figure 1 This is a flowchart illustrating a traffic sign recognition method provided in an embodiment of the present invention.

[0014] like Figure 1 As shown, a traffic sign recognition method includes the following steps: S1: Import multiple original traffic sign images, preprocess all the original traffic sign images, and combine the preprocessed results to obtain a set of preprocessed traffic sign images; S2: Divide the preprocessed traffic sign image set into a traffic sign training set and a traffic sign test set according to a preset ratio; S3: Construct a training model by training the training model using the traffic sign training set to obtain the model to be processed; S4: Test and analyze the model to be processed using the traffic sign test set to obtain a traffic sign recognition model; S5: Import the traffic sign image to be recognized, and use the traffic sign recognition model to recognize the traffic sign image to obtain the traffic sign recognition result.

[0015] It should be understood that the processed sample data (i.e., the preprocessed set of traffic sign images) is divided into training samples (i.e., the traffic sign training set) and test samples (i.e., the traffic sign test set).

[0016] In the above embodiments, a preprocessed traffic sign image set is obtained by preprocessing the original traffic sign images. The preprocessed traffic sign image set is then divided into a traffic sign training set and a traffic sign test set according to a preset ratio. The training model is trained using the traffic sign training set to obtain the model to be processed. The traffic sign recognition model is obtained by testing and analyzing the model to be processed using the traffic sign test set. The traffic sign recognition result is obtained by recognizing the traffic sign images to be recognized using the traffic sign recognition model. This approach can overcome the challenge of traffic sign recognition in adverse weather conditions, improve the reliability and safety of automotive driver assistance systems in complex environments, protect the lives of drivers and passengers, promote the high-quality development of the automotive industry and intelligent connected vehicle technology, and provide key technical support for the construction of an intelligent transportation system.

[0017] Optionally, as an embodiment of the present invention, the process of preprocessing all the original traffic sign images and combining the preprocessed results to obtain a preprocessed traffic sign image set includes: A preset first noise is added to each of the original traffic sign images to obtain multiple low-light traffic sign images; Each of the original traffic sign images is blurred according to a preset first blur rule to obtain multiple blurred traffic sign images; A preset second noise is added to each of the blurred traffic sign images to obtain multiple foggy traffic sign images; Each of the original traffic sign images was blurred according to the second blurring rule to obtain multiple rainy day traffic sign images; The brightness of each of the original traffic sign images is adjusted to obtain multiple adjusted traffic sign images; Preset artificial light sources are added to each of the adjusted traffic sign images to obtain multiple traffic sign images with light source interference. Each of the original traffic sign images is blurred according to the third blurring rule to obtain multiple blurred traffic sign images; A preprocessed set of traffic sign images is obtained by combining all the images of low-light traffic signs, all the images of foggy traffic signs, all the images of rainy traffic signs, all the images of traffic signs with light source interference, and all the images of blurred traffic signs.

[0018] It should be understood that sample image preprocessing (because it is necessary to simulate traffic sign recognition efficiently in severe weather, but the number of images under severe weather conditions is limited, so corresponding operations are required): obtain sample data of sample images (i.e., original traffic sign images), and perform a series of operations on the sample data (i.e., original traffic sign images) to simulate severe weather.

[0019] It should be understood that the preset first noise can be Gaussian noise, salt and pepper noise, or speckle noise; the preset first fuzzy rule can be Gaussian fuzziness; the preset second noise can be white noise; the second fuzzy rule can be dynamic fuzziness; and the third fuzzy rule can be motion fuzziness.

[0020] Specifically, (1) Gaussian noise, salt-and-pepper noise, or speckle noise are added to the image (i.e., the original traffic sign image) to simulate signal interference or low-light conditions. (2) Gaussian blur is used and white noise is added to simulate a foggy effect. (3) Motion blur is used to simulate the effect of flowing rain. (4) The brightness of the image (i.e., the original traffic sign image) is reduced, and halos and reflections of artificial light sources (such as streetlights) are added. (5) Motion blur is added to the image (i.e., the original traffic sign image) to simulate the dynamic blur effect when vehicles are moving quickly.

[0021] In the above embodiments, all original traffic sign images are preprocessed, and the preprocessed results are combined to obtain a preprocessed traffic sign image set, which simulates traffic signs under various severe conditions and improves the recognition of traffic signs in severe weather.

[0022] Optionally, as an embodiment of the present invention, the process of constructing a training model and training the training model using the traffic sign training set to obtain the model to be processed includes: S31: Construct and train the model; S32: Update the sparse coefficients of the training model using the traffic sign training set to obtain the first updated training model; S33: Update the dictionary of the first updated training model using the traffic sign training set to obtain the second updated training model; S34: The traffic sign training set is predicted using the second updated training model to obtain multiple first prediction results; S35: Import the real training labels of traffic signs corresponding to each traffic sign training image in the traffic sign training set, and calculate the loss value for all the first prediction results and all the real training labels of traffic signs to obtain the loss value. S36: Determine whether the difference between the loss value and the loss value of the previous iteration is equal to 0. If not, use the second updated training model as the training model for the next iteration and return to S32; if yes, use the second updated training model as the model to be processed.

[0023] To understand this, the training set (i.e., the traffic sign training set) is input into a deep nonlinear nonnegative convolutional dictionary learning model (i.e., the training model). The LPOM-optimized model (i.e., the training model) is solved using the block coordinate descent (BCD) method to obtain sparse features and a dictionary (weights). Traffic sign images from the training set (i.e., the traffic sign training set) are then applied to each dictionary learning layer, with weights and activation functions applied to update the data. Finally, it calculates the loss (i.e., the loss value) between the output and the target label, as well as the accuracy, and returns the loss value and accuracy.

[0024] Specifically, keeping the dictionary parameters of each layer unchanged, the sparse coefficients of each layer are updated by optimizing the objective function through LPOM based on the input features of the training samples (i.e., the traffic sign training set), ensuring that the sparse coefficients can accurately represent the key features of the input image (such as the shape and color of the traffic signs).

[0025] Specifically, while keeping the sparsity coefficients unchanged, the dictionary is re-optimized so that it can better "fit" the current sparsity coefficients and improve the accuracy of feature reconstruction (i.e., make the dictionary atoms fit the real features of traffic signs better).

[0026] In the above embodiments, the training model is trained using a traffic sign training set to obtain the model to be processed. This can overcome the challenge of traffic sign recognition in adverse weather conditions, improve the reliability and safety of the vehicle's driver assistance system in complex environments, protect the lives of drivers and passengers, promote the high-quality development of the automotive industry and intelligent connected technology, and provide key technical support for the construction of an intelligent transportation system.

[0027] Optionally, as an embodiment of the present invention, such as Figure 1 and 2 As shown, the process of S31 includes: The objective function is obtained through the first equation, and a training model is constructed using the objective function. The first equation is: , in, , , in, For traffic sign training set, It is a nonlinear mapping function. For the nth level dictionary, Let be the sparse coefficient matrix of the nth layer, and ∗ represent the convolution process. Let be the regularization coefficient of the i-th layer. The square of the Frobenius norm. Let be the sparse coefficient matrix of the i-th layer. For the (i-1)th level dictionary, Let i be the sparse coefficient matrix of the (i-1)th layer. Let be the regularization constraint function for the sparse coefficient matrix of the i-th layer. Let be the first constraint function. This is the second constraint function. The elements are in the sparse coefficient matrix. represents the elements in the matrix resulting from the convolution of the sparse coefficient matrix and the dictionary.

[0028] Specifically, such as Figure 2 As shown, the steps for constructing a deep nonlinear nonnegative convolutional dictionary learning model (i.e., training the model) are as follows: The data is decomposed into nonlinear mappings across multiple dictionary layers, with coefficients in deeper layers learned through dictionary learning from the previous layer. Local regularization Sp(·) is introduced at each sparse coding layer of the model to enhance feature extraction and obtain a sparse representation of the data. Nonlinear convolutional mappings combining dictionaries and coefficients are progressively propagated through a multi-layered sparse coding structure. Therefore, the richer, composite nonlinear convolutional dictionary representations learned in earlier layers are used to activate the coefficients in the final layer.

[0029] To optimize the DNNCDL model and prevent accuracy loss due to inverse operations of nonlinear functions, the Boosting Neighbor Operator Machine (LPOM) technique is employed to transform the nonlinear function into an equivalent neighbor operator, which is then integrated into the model as a penalty term. This method transforms the nonconvex optimization problem into a series of convex subproblems, enabling the acquisition of the optimal dictionary and coefficients through alternating updates.

[0030] The following equation can be approximated using the LPOM method. The constrained optimization problem in [the context of the problem]. Here, Y is the input signal, ψ represents a nonlinear function, and * denotes the convolution operation. and The sparse coefficients (i.e., the sparse coefficient matrix) and dictionary of the i-th layer, Sp( The expression is a unified representation of applying sparsity constraints to the sparse coefficients (i.e., the sparse coefficient matrix) of the i-th layer, where Sp() is the regularization constraint on the sparse coefficients (i.e., the sparse coefficient matrix). This represents the error limit. It's important to note that Sp() is a regularization constraint on the sparse coefficients (i.e., the sparse coefficient matrix), using adaptive sparse coefficient constraints instead of traditional fixed sparsity constraints (such as the L1 norm). Adaptive sparse coefficient constraints can dynamically adjust sparsity based on data characteristics, thus better adapting to the sparsity requirements of different data and improving the robustness and flexibility of the model. The Sparse Representation Updating (SRU) unit acts as a sparse coefficient constraint.

[0031] Through a series of operations, the original constrained optimization problem was transformed into a new reconstruction problem using the LPOM method. Therefore, the model (i.e., the trained model) of this invention was obtained, as follows: , in, , .

[0032] In the above embodiments, the constructed training model can dynamically adjust the sparsity according to the characteristics of the data, better adapt to the sparsity requirements of different data, and improve the robustness and flexibility of the model.

[0033] Optionally, as an embodiment of the present invention, the process of testing and analyzing the model to be processed using the traffic sign test set to obtain a traffic sign recognition model includes: S41: Test the model to be processed using the traffic sign test set to obtain multiple second prediction results; S42: Import the actual test labels of traffic signs corresponding to each traffic sign test image in the traffic sign test set, and calculate the accuracy of all the second prediction results and all the actual test labels of traffic signs to obtain the accuracy. S43: Determine whether the accuracy is greater than or equal to a preset threshold. If not, use the model to be processed as the training model for the next iteration and return to S32. If yes, use the model to be processed as a traffic sign recognition model.

[0034] As you can understand, the test set (i.e., the traffic sign test set) is input into the deep nonlinear nonnegative convolutional dictionary learning model (i.e., the model to be processed). The traffic sign images in the test set (i.e., the traffic sign test set) are updated by applying the trained weights and activation functions to each dictionary learning layer. Finally, it calculates the loss and accuracy between the output and the target label, and returns the loss value and accuracy.

[0035] In the above embodiments, a traffic sign recognition model is obtained by testing and analyzing the model to be processed through a traffic sign test set. This improves the reliability and safety of the vehicle's assisted driving system in complex environments, protects the lives of drivers and passengers, promotes the high-quality development of the automotive industry and intelligent connected technology, and provides key technical support for the construction of an intelligent transportation system.

[0036] Optionally, as another embodiment of the present invention, the present invention aims to develop a recognition system driven by deep nonlinear non-negative convolution dictionary learning, to overcome the problem of traffic sign recognition in severe weather, improve the reliability and safety of new energy vehicle assisted driving systems in complex environments, promote the high-quality development of the new energy vehicle industry and intelligent connected vehicle technology, help achieve relevant national planning goals, and provide key technical support for the construction of intelligent transportation systems.

[0037] Alternatively, as another embodiment of the present invention, in the fields of signal processing and machine learning, traditional dictionary learning (DL) methods represent signals through linear combinations of global dictionary atoms. While performing well in many applications, they have some limitations. These methods typically assume that signals can be represented by linear combinations of global dictionary atoms, ignoring the local structure and spatial correlation of the signal, which is particularly evident when processing image data with complex spatial structures. Furthermore, traditional dictionary learning suffers from high computational complexity and storage costs when processing high-dimensional data, and struggles to effectively capture local features in the signal. In recent years, convolutional dictionary learning (CDL), by introducing convolution operations and utilizing the sliding of convolution kernels on the signal, can naturally capture local features of the signal. Simultaneously, it significantly reduces the number of parameters that need to be learned by utilizing parameter sharing, thereby lowering computational complexity and storage costs while preserving the spatial structure of the signal.

[0038] However, most existing convolutional dictionary learning methods are limited to single-layer linear models, making it difficult to effectively handle nonlinear information in data and capture deep features. To overcome these limitations, this invention proposes a Deep Nonlinear Nonnnegative Convolutional Dictionary Learning (DNNCDL) method. By introducing multi-layer nonlinear nonnegative structures, it can extract deep features of signals layer by layer, further enhancing its ability to represent complex data. Furthermore, this invention uses adaptive sparse coefficient constraints to replace traditional fixed sparsity constraints (such as the L1 norm) to improve the model's adaptability to different data sparsity requirements and its robustness to noise. To optimize the DNCDL model, this invention employs a Lifted Proximal Operator Machine (LPOM) technique, transforming nonlinear functions into equivalent proximal operators and incorporating them as penalty terms into the model. This method transforms the nonconvex optimization problem into a series of convex subproblems, obtaining the optimal dictionary and coefficients through alternating updates, significantly improving the model's optimization efficiency and convergence speed. Through these improvements, the DNNCDL method proposed in this invention performs excellently when processing complex data, providing a new and effective tool for signal representation and analysis.

[0039] Alternatively, as another embodiment of the present invention, the main innovative points of the present invention are as follows: 1. Deep Nonlinear Feature Extraction: Traditional traffic sign recognition often employs shallow or linear models, which have limited feature extraction capabilities in complex scenarios. This invention utilizes a deep nonlinear algorithm to extract deep features of the signal layer by layer. The output of each layer serves as the input for the next, enabling the learning of more abstract and discriminative deep nonlinear features of traffic signs. For example, under complex lighting and occlusion conditions, it can effectively capture subtle textures and shapes, overcoming the limitations of traditional methods and improving recognition accuracy.

[0040] 2. Non-negative Convolutional Dictionary Learning: Introducing non-negative convolutional dictionary learning forces the sparse coefficient matrix to be non-negative, giving the learned sparse coefficients (dictionary) a clear physical meaning. This better aligns with the feature representation needs of real-world scenarios and maintains sparsity. In convolutional dictionary learning, through sparse encoding, the model automatically selects the most effective dictionary atoms for the input data. This feature selection process is interpretable. In traffic sign recognition, this ensures the stability and interpretability of feature representation, focuses on key features, removes redundant information, reduces computational complexity, and improves model generalization ability while reducing the risk of overfitting. For example, when modeling changes in lighting in a vehicle's driving environment, non-negative encoding avoids unreasonable negative features, thus more accurately reflecting the impact of light intensity and color changes on image features. In traffic sign recognition, non-negativity constraints enable the model to focus on learning the positive features of signs, such as color brightness and shape contours, eliminating negative interference and improving recognition accuracy.

[0041] 3. Improve computational efficiency and reduce memory consumption: (1) Computation and storage optimization brought about by nonnegativity constraints: Since only nonnegative values ​​need to be stored, smaller data types can be used to represent these values ​​in memory, thereby reducing memory usage. At the same time, nonnegativity can also make some sparse representation methods more efficient, further compressing data storage space. For example, in multiplication and addition operations, there is no need to consider the sign, reducing the amount of computation.

[0042] (2) Feature selection using nonlinear mapping: The input data undergoes complex transformations, mapping it to a more discriminative feature space. During this process, the model can automatically select the most useful features for the recognition task, ignoring irrelevant information. Thus, in subsequent calculations, only these key features need to be processed, reducing unnecessary computation. For example, when processing image data of traffic signs, nonlinear mapping can directly capture key features such as the shape, color, and lane line texture of the signs, without requiring point-by-point calculations for every pixel of the image.

[0043] (3) Locality and parameter sharing in convolutional dictionary learning: Local receptive field: Convolutional operations are localized, meaning each convolutional kernel processes only a local region of the input data. In traffic sign and lane line recognition, this locality is ideal for capturing local features in images, such as the edges of signs and the breaks in lane lines. Compared to fully connected layers, convolutional operations significantly reduce the number of parameters that need to be learned, thereby reducing computational complexity.

[0044] Parameter sharing: During convolution, the same convolution kernel slides across the entire input data to perform the convolution operation. This means that the parameters of the convolution kernel are shared at different locations. Parameter sharing not only reduces the number of parameters in the model and lowers memory consumption, but also makes the model translation invariant, enabling it to better adapt to features at different locations.

[0045] (4) To optimize the DNCDL model, this invention employs the Lifted Proximal Operator Machine (LPOM) technique. This technique transforms the nonlinear function into an equivalent proximal operator and incorporates it as a penalty term into the model. This method transforms the nonconvex optimization problem into a series of convex subproblems, and the optimal dictionary and coefficients are obtained through alternating updates, significantly improving the optimization efficiency and convergence speed of the model.

[0046] Figure 3 This is a block diagram of a traffic sign recognition device provided in an embodiment of the present invention.

[0047] Alternatively, as another embodiment of the present invention, such as Figure 3 As shown, a traffic sign recognition device includes: The import module is used to import multiple original traffic sign images; The preprocessing module is used to preprocess all the original traffic sign images and combine the preprocessed results to obtain a preprocessed traffic sign image set. The segmentation module is used to divide the preprocessed traffic sign image set into a traffic sign training set and a traffic sign test set according to a preset ratio. The training module is used to build a training model and train the training model using the traffic sign training set to obtain the model to be processed. The test analysis module is used to test and analyze the model to be processed using the traffic sign test set to obtain a traffic sign recognition model. The import module is also used to import images of traffic signs to be recognized; The recognition result acquisition module is used to recognize the traffic sign image to be recognized through the traffic sign recognition model and obtain the traffic sign recognition result.

[0048] Optionally, as an embodiment of the present invention, the preprocessing module is specifically used for: A preset first noise is added to each of the original traffic sign images to obtain multiple low-light traffic sign images; Each of the original traffic sign images is blurred according to a preset first blur rule to obtain multiple blurred traffic sign images; A preset second noise is added to each of the blurred traffic sign images to obtain multiple foggy traffic sign images; Each of the original traffic sign images was blurred according to the second blurring rule to obtain multiple rainy day traffic sign images; The brightness of each of the original traffic sign images is adjusted to obtain multiple adjusted traffic sign images; Preset artificial light sources are added to each of the adjusted traffic sign images to obtain multiple traffic sign images with light source interference. Each of the original traffic sign images is blurred according to the third blurring rule to obtain multiple blurred traffic sign images; A preprocessed set of traffic sign images is obtained by combining all the images of low-light traffic signs, all the images of foggy traffic signs, all the images of rainy traffic signs, all the images of traffic signs with light source interference, and all the images of blurred traffic signs.

[0049] Optionally, as an embodiment of the present invention, the training module is specifically used for: S31: Construct and train the model; S32: Update the sparse coefficients of the training model using the traffic sign training set to obtain the first updated training model; S33: Update the dictionary of the first updated training model using the traffic sign training set to obtain the second updated training model; S34: The traffic sign training set is predicted using the second updated training model to obtain multiple first prediction results; S35: Import the real training labels of traffic signs corresponding to each traffic sign training image in the traffic sign training set, and calculate the loss value for all the first prediction results and all the real training labels of traffic signs to obtain the loss value. S36: Determine whether the difference between the loss value and the loss value of the previous iteration is equal to 0. If not, use the second updated training model as the training model for the next iteration and return to S32; if yes, use the second updated training model as the model to be processed.

[0050] Optionally, another embodiment of the present invention provides a traffic sign recognition system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the traffic sign recognition method as described above. This system can be a computer or similar system.

[0051] Optionally, another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the traffic sign recognition method as described above.

[0052] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0053] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0055] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0056] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0057] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A traffic sign recognition method characterized by, The method comprises the following steps: Importing a plurality of original traffic sign pictures, preprocessing all the original traffic sign pictures, and collecting the preprocessed results to obtain a preprocessed traffic sign picture set; According to a preset ratio, the preprocessed traffic sign picture set is divided into a traffic sign training set and a traffic sign test set; A training model is constructed, the training model is trained by the traffic sign training set, and a to-be-processed model is obtained; The to-be-processed model is tested and analyzed by the traffic sign test set, and a traffic sign recognition model is obtained; Importing a to-be-recognized traffic sign picture, recognizing the to-be-recognized traffic sign picture by the traffic sign recognition model, and obtaining a traffic sign recognition result.

2. The traffic sign recognition method according to claim 1, characterized in that, The process of preprocessing all the original traffic sign pictures and collecting the preprocessed results to obtain a preprocessed traffic sign picture set comprises: A preset first noise is added to each of the original traffic sign pictures to obtain a plurality of low-light traffic sign pictures; According to a preset first blur rule, each of the original traffic sign pictures is blurred to obtain a plurality of blurred traffic sign pictures; A preset second noise is added to each of the blurred traffic sign pictures to obtain a plurality of foggy traffic sign pictures; According to a second blur rule, each of the original traffic sign pictures is blurred to obtain a plurality of rainy traffic sign pictures; Each of the original traffic sign pictures is adjusted in brightness to obtain a plurality of adjusted traffic sign pictures; A preset artificial light source is added to each of the adjusted traffic sign pictures to obtain a plurality of light source interference traffic sign pictures; According to a third blur rule, each of the original traffic sign pictures is blurred to obtain a plurality of blurred traffic sign pictures; All the low-light traffic sign pictures, all the foggy traffic sign pictures, all the rainy traffic sign pictures, all the light source interference traffic sign pictures, and all the blurred traffic sign pictures are collected to obtain a preprocessed traffic sign picture set.

3. The traffic sign recognition method according to claim 1, characterized in that, The process of constructing a training model, training the training model by the traffic sign training set, and obtaining a to-be-processed model comprises: S31: Constructing a training model; S32: Updating the sparse coefficients of the training model by the traffic sign training set to obtain a first updated training model; S33: Updating the dictionary of the first updated training model by the traffic sign training set to obtain a second updated training model; S34: Predicting the traffic sign training set by the second updated training model to obtain a plurality of first prediction results; S35: Importing traffic sign real training labels corresponding to each traffic sign training picture in the traffic sign training set, and calculating loss values for all the first prediction results and all the traffic sign real training labels to obtain a loss value; S36: determining whether the difference between the loss value and the loss value of the previous iteration number is equal to 0, if not, taking the second updated training model as the training model of the next iteration number, and returning to S32; if yes, taking the second updated training model as the to-be-processed model.

4. The traffic sign recognition method according to claim 3, characterized in that, The process of S31 comprises: An objective function is obtained by a first formula, and a training model is constructed by the objective function, the first formula being: ， wherein , , wherein, is a traffic sign training set, is a nonlinear mapping function, is an nth layer dictionary, is an nth layer sparse coefficient matrix, * is a convolution processing, is an ith layer regularization coefficient, is a square of a Frobenius norm, is an ith layer sparse coefficient matrix, is an i-1th layer dictionary, is an i-1th layer sparse coefficient matrix, is a regularization constraint function of an ith layer sparse coefficient matrix, is a first constraint function, is a second constraint function, is an element in a sparse coefficient matrix, is an element in a matrix after convolution of a sparse coefficient matrix and a dictionary.

5. The traffic sign recognition method according to claim 3, characterized in that, The process of testing and analyzing the to-be-processed model by the traffic sign test set comprises: S41: testing the to-be-processed model by the traffic sign test set to obtain a plurality of second prediction results; S42: importing traffic sign real test labels corresponding to each traffic sign test picture in the traffic sign test set, and calculating the accuracy of all the second prediction results and all the traffic sign real test labels to obtain an accuracy; S43: determining whether the accuracy is greater than or equal to a preset threshold, if not, taking the to-be-processed model as the training model of the next iteration number, and returning to S32; if yes, taking the to-be-processed model as a traffic sign recognition model.

6. A traffic sign recognition apparatus characterized by comprising: Comprise: An importing module is configured to import a plurality of original traffic sign pictures; A preprocessing module is configured to preprocess all the original traffic sign pictures, and obtain a set of preprocessed traffic sign pictures by collecting the preprocessed results; A division module is configured to divide the set of preprocessed traffic sign pictures into a traffic sign training set and a traffic sign test set according to a preset proportion; A training module is configured to construct a training model, train the training model by the traffic sign training set, and obtain a to-be-processed model; A testing and analyzing module is configured to test and analyze the to-be-processed model by the traffic sign test set, and obtain a traffic sign recognition model; The importing module is further configured to import a to-be-recognized traffic sign picture; An identification result obtaining module is configured to recognize the to-be-recognized traffic sign picture by the traffic sign recognition model, and obtain a traffic sign recognition result.

7. The traffic sign recognition apparatus according to claim 6, characterized in that, The preprocessing module is specifically configured to: add a preset first noise to each of the original traffic sign pictures to obtain a plurality of low-light traffic sign pictures; perform blur processing on each of the original traffic sign pictures according to a preset first blur rule to obtain a plurality of blurred traffic sign pictures; add a preset second noise to each of the blurred traffic sign pictures to obtain a plurality of foggy traffic sign pictures; perform blur processing on each of the original traffic sign pictures according to a second blur rule to obtain a plurality of rainy traffic sign pictures; perform brightness adjustment on each of the original traffic sign pictures to obtain a plurality of adjusted traffic sign pictures; add a preset artificial light source to each of the adjusted traffic sign pictures to obtain a plurality of light source interference traffic sign pictures; perform blur processing on each of the original traffic sign pictures according to a third blur rule to obtain a plurality of blurred traffic sign pictures; The low-light traffic sign picture set, the foggy weather traffic sign picture set, the rainy weather traffic sign picture set, the light source interference traffic sign picture set and the blurred traffic sign picture set are combined to obtain a preprocessed traffic sign picture set.

8. The traffic sign recognition apparatus according to claim 6, characterized by The training module is specifically configured to: S31: constructing a training model; S32: updating sparse coefficients of the training model by using the traffic sign training set to obtain a first updated training model; S33: updating a dictionary of the first updated training model by using the traffic sign training set to obtain a second updated training model; S34: predicting the traffic sign training set by using the second updated training model to obtain a plurality of first prediction results; S35: importing traffic sign real training labels corresponding to each traffic sign training picture in the traffic sign training set, and calculating loss values of all the first prediction results and all the traffic sign real training labels to obtain loss values; S36: determining whether a difference between the loss value and a loss value of a previous iteration number is equal to 0, if not, taking the second updated training model as a training model of a next iteration number, and returning to S32; if yes, taking the second updated training model as a to-be-processed model.

9. A traffic sign recognition apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, the traffic sign recognition method of any one of claims 1 to 5 is implemented.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. When the computer program is executed by the processor, the traffic sign recognition method of any one of claims 1 to 5 is implemented.

Images