Method and device for determining road friction coefficient, electronic equipment and storage medium

Abstract

The invention relates to a method and a device for determining a road friction coefficient, electronic equipment and a storage medium. The method comprises the following steps: acquiring a road image of a target road section sent by a shooting device; the road image is processed through a target recognition model, at least one crack type contained in the road image is recognized, and different crack types have different degrees of influence on the road friction coefficient; counting the number of cracks of each crack type in the road image; and inputting the crack number of each crack type into a pre-trained road friction coefficient model to obtain the road friction coefficient model, and outputting the road friction coefficient of the target road section. The efficiency of determining the road friction coefficient can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, device, electronic device, and storage medium for determining road friction coefficient. Background Art

[0002] In the field of automotive engineering, especially intelligent driving technology, there is an important need for the identification of road friction coefficient. Road friction coefficient is one of the important factors affecting vehicle driving stability and safety. It determines key parameters such as braking distance and turning radius of vehicles on different road surfaces. Therefore, accurately identifying road friction coefficient is of great significance for the safe driving of vehicles.

[0003] Currently, the identification of road friction coefficient mainly relies on physical measurement methods. For example, technicians use a friction coefficient tester to directly measure the road friction coefficient. However, there are some problems with the above methods. Technicians taking the instrument to the road for real-time measurement can easily lead to very low efficiency. Summary of the Invention

[0004] This application provides a method, device, electronic device, and storage medium for determining road friction coefficient to solve the problem of low efficiency in determining road friction coefficient.

[0005] In a first aspect, this application provides a method for determining road friction coefficient, the method including:

[0006] Obtain a road image of a target section sent by a photographing device;

[0007] Process the road image through a target recognition model to identify at least one crack type included in the road image;

[0008] Count the number of cracks of each crack type in the road image, where different crack types and crack numbers have different degrees of influence on the road friction coefficient;

[0009] Input the number of cracks of each crack type into a pre-trained road friction coefficient model to obtain the road friction coefficient of the target section output by the road friction coefficient model.

[0010] Optionally, processing the road image through a target recognition model to identify at least one crack type included in the road image includes:

[0011] Filter out the noise in the road image by dynamically adjusting the standard deviation in the Gaussian filtering algorithm;

[0012] Process the road image after filtering out the noise through a target detection model to identify at least one crack type included in the road image.

[0013] Optionally, filtering the noise in the road image by dynamically adjusting the standard deviation in the Gaussian filtering algorithm includes:

[0014] Determine each pixel point in the road image and the neighborhood corresponding to each pixel point;

[0015] Determine the variance in the neighborhood corresponding to the pixel point, where the variance is used to indicate the degree of dispersion of the pixel point;

[0016] Dynamically adjust the standard deviation in the neighborhood by using a preset amplitude adjustment coefficient and the variance, where the standard deviation is used to adjust the smoothing intensity of the Gaussian filter;

[0017] After filtering the road image by using the dynamically adjusted standard deviation, construct a comparison function according to the product of the standard deviation and a preset two-dimensional Gaussian filtering function;

[0018] If the function value of the comparison function does not reach the set threshold, recalculate the variance according to the filtered road image until the function value of each pixel point reaches the set threshold, where the set threshold is used to indicate that the degree of dispersion of the pixel point is the same as the degree of dispersion of the two-dimensional Gaussian filtering function.

[0019] Optionally, the pre-training process of the road friction coefficient model includes:

[0020] Obtain training samples through a friction coefficient tester, where the training samples include the number of cracks corresponding to each crack type in the sample section and the road friction coefficient corresponding to each crack type;

[0021] Preliminarily train an initial road friction coefficient model by using the training samples;

[0022] Verify the preliminarily trained road friction coefficient model by using the set of crack numbers in the verification samples and the training samples to obtain an output result;

[0023] If the output result is different from the road friction coefficient in the verification samples, adjust the parameters in the preliminarily trained road friction coefficient model until a trained road friction coefficient model is obtained, where the output result of the trained road friction coefficient model is the same as the road friction coefficient in the verification samples.

[0024] Optionally, the initial road friction coefficient model is a generalized regression neural network, and the generalized regression neural network includes an input layer, a pattern layer, a summation layer, and an output layer. Verifying the preliminarily trained road friction coefficient model by using the set of crack numbers in the verification samples and the training samples to obtain an output result includes:

[0025] Use the set of crack numbers in the verification sample as input data and input it into the input layer;

[0026] Determine the weights of the pattern layer based on the input data and the training sample Ri, where the training sample Ri is used to indicate the number of cracks corresponding to the i-th neuron in the input layer, and the weights are used to indicate the similarity between the input data X and the training sample Ri;

[0027] Determine the sum of weights of each weight output by the pattern layer, and determine the output value of the summation layer based on the sum of the products of the j-th output value of all training samples and the corresponding weights. Each training sample Ri corresponds to an output sample, and the output value of the summation layer is used to indicate the value of the j-th element of the j-th output sample;

[0028] Determine the output result of the j-th element in the output layer according to the quotient of the output value of the summation layer and the sum of weights.

[0029] Optionally, the calculation formula for the weights of the pattern layer is:

[0030]

[0031] where Pi is the weight of the pattern layer, X is the input data, Ri is the training sample of the i-th neuron in the input layer, T represents matrix transpose, σ is the smoothing factor, and n is the number of neurons in the input layer.

[0032] Optionally, the calculation formula for the output value of the summation layer is:

[0033]

[0034] where S Nj represents the output value of the summation layer, Yij represents the j-th output value of the i-th training sample, Pi represents the similarity between the input data X and the training sample Ri, n is the number of training samples, and w is the dimension of the output sample.

[0035] In a second aspect, the present application provides a device for determining the road friction coefficient, and the device includes:

[0036] An acquisition module, configured to acquire a road image of a target road section sent by a photographing device;

[0037] An identification module, configured to process the road image through a target recognition model to identify at least one type of crack included in the road image;

[0038] A statistics module, configured to count the number of cracks of each type of crack in the road image, where different types of cracks and the number of cracks have different degrees of influence on the road friction coefficient;

[0039] An input / output module, configured to input the number of cracks of each crack type into a pre-trained road friction coefficient model, and obtain the road friction coefficient of the target road section output by the road friction coefficient model.

[0040] In a third aspect, the present application provides an electronic device, including: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus.

[0041] In a fourth aspect, the present application further provides a computer storage medium storing computer-executable instructions for executing the method for determining the road friction coefficient according to any one of the above of the present application.

[0042] The above technical solutions provided by the embodiments of the present application have the following advantages compared with the prior art: By automatically identifying the crack type for each crack in the road image, the number of cracks of each crack type is statistically obtained, and then the road friction coefficient model is used to process and analyze the number of cracks of multiple cracks to determine the road friction coefficient in the road image. Compared with the manual determination of the coefficient in the related art, the present application can complete these operations in a very short time by using automated technology, improving the efficiency of determining the road friction coefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0045] One or more embodiments are exemplarily illustrated by the pictures in the corresponding accompanying drawings. These exemplary illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements, unless otherwise stated, and the drawings in the drawings do not constitute a proportional limitation.

[0046] Figure 1 It is a schematic diagram of a system for determining the road friction coefficient provided by the embodiments of the present application;

[0047] Figure 2 It is a flowchart of a method for determining the road friction coefficient provided by the embodiments of the present application;

[0048] Figure 3 It is a schematic structural diagram of the generalized regression neural network provided by the embodiment of the present application;

[0049] Figure 4 It is a schematic structural diagram of a device for determining road friction coefficient provided by the embodiment of the present application;

[0050] Figure 5 It is a schematic structural diagram of an electronic device provided by the embodiment of the present application. Detailed implementation manners

[0051] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0052] The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and settings of specific examples are described below. Of course, they are only examples and are not intended to limit the present invention. In addition, the present invention may repeat reference numerals and / or letters in different examples. Such repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0053] To solve the problems mentioned in the background art, according to one aspect of the embodiments of the present application, an embodiment for determining road friction coefficient is provided.

[0054] Optionally, in the embodiments of the present application, the above method for determining road friction coefficient may be applied to a hardware environment composed of a photographing device 101 and a processor 103 as shown in Figure 1 As shown in Figure 1 The processor 103 is connected to the photographing device 101 through a network, and can be used to process the image sent by the photographing device to determine the road friction coefficient in the image. A database 105 may be set on the processor or independently of the processor to provide data storage services for the processor 103. The above network includes but is not limited to: wide area network, metropolitan area network or local area network.

[0055] Next, a method for determining road friction coefficient provided by the embodiments of the present application will be described in detail in combination with the specific implementation manners. Applied to the processor, as shown in Figure 2 The specific steps are as follows:

[0056] Step 201: Obtain a road image of a target road section sent by a photographing device;

[0057] Step 202: Process the road image through a target recognition model to identify at least one type of crack contained in the road image;

[0058] Step 203: Count the number of cracks of each type of crack in the road image, where different types of cracks and the number of cracks have different degrees of influence on the road friction coefficient;

[0059] Step 204: Input the number of cracks of each type of crack into a pre-trained road friction coefficient model to obtain the road friction coefficient of the target road section output by the road friction coefficient model.

[0060] In the embodiment of the present application, a road image captured by a high-resolution camera is used as perception data input, and a processor processes these road images using a target recognition model to identify at least one type of crack contained therein, such as longitudinal cracks, transverse cracks, and reticular cracks. After identifying the crack type, the processor counts each type of crack, counts the number of each type of crack, and then collects the number of each type of crack and inputs it into a pre-trained road friction coefficient model uniformly, so as to obtain the road friction coefficient of the target road section. Since different types of cracks and the number of cracks have different degrees of influence on the road friction coefficient model, the road friction coefficient determined based on the crack type and the number of cracks is more accurate.

[0061] Optionally, the photographing device may be a high-resolution static camera or a dynamic camera. The static camera is, for example, a camera installed on a street lamp or an electronic capture device already installed on a road. The dynamic camera is, for example, a camera installed on a vehicle or a camera located on a drone, etc.

[0062] Exemplarily, a camera on a street lamp captures a road image within a range of three meters from the street lamp. Through the recognition model, it is recognized that the road image includes 5 longitudinal cracks, 2 transverse cracks, and 4 reticular cracks, generating a crack number set [5, 2, 4]. The road friction coefficient model processes this crack number set to identify that the road friction coefficient in the road image is 0.4.

[0063] In the present application, the crack type of each crack in the road image is automatically identified, so as to statistically obtain the number of cracks of each type of crack, and then the road friction coefficient model is used to process and analyze the number of cracks of multiple cracks to determine the road friction coefficient in the road image. Compared with the manual determination coefficient in the related art, the present application can complete these operations in a very short time using an automated technology, improving the efficiency of determining the road friction coefficient. In addition, the automated technology can also reduce human errors and improve accuracy and reliability.

[0064] As an alternative implementation, before image processing, it is generally necessary to use the Gaussian filtering algorithm to denoise the image. However, in the traditional Gaussian filtering algorithm, the standard deviation is set artificially, and the fixed standard deviation results in a poor image denoising effect. Therefore, the embodiments of the present application perform the following operations:

[0065] Step 2021: Filter the noise in the road image by dynamically adjusting the standard deviation in the Gaussian filtering algorithm;

[0066] Step 2022: Process the road image after filtering the noise through the object detection model to identify at least one type of crack contained in the road image.

[0067] The embodiments of the present application adopt an adaptive Gaussian filtering algorithm. The adaptive Gaussian filtering algorithm uses a dynamic standard deviation σ, which means that in a smooth area (such as a uniform background), the standard deviation can be larger, so as to more effectively filter the noise; while in an area with rich edges or textures, the standard deviation will be smaller to avoid blurring the edges and textures, achieving better filtering of the image noise, thereby improving the accuracy of subsequent crack type recognition based on the road image, and further improving the accuracy of the road friction coefficient.

[0068] As an alternative implementation, in step 2021, the process of filtering the noise in the road image by the adaptive Gaussian filtering algorithm includes the following steps:

[0069] Step S11: Determine each pixel point in the road image and the neighborhood corresponding to each pixel point.

[0070] The processor traverses each pixel point (x, y) in the road image and defines a neighborhood window for each pixel point (x, y), usually a square window, and the size of the neighborhood window is (2k + 1) * (2k + 1). The size of the neighborhood window can be adjusted according to actual needs. Exemplarily, the neighborhood size is a window of 3 * 3 or 5 * 5.

[0071] Step S12: Determine the variance in the neighborhood corresponding to the pixel point, where the variance is used to indicate the degree of dispersion of the pixel point.

[0072] The processor extracts all pixel values within the neighborhood window, calculates the mean value μ(x, y) of the pixel values within the neighborhood window, and then calculates the variance within the neighborhood window by combining multiple pixel values I(x + a, y + b) within the neighborhood window and the mean value μ(x, y). The calculation formula of the variance is as follows:

[0073]

[0074] Among them, D(x,y) represents the variance within the neighborhood, k represents the neighborhood radius, m and n represent the offset of the neighborhood window in the horizontal and vertical directions respectively, and their value range is from -k to k, a and b represent the relative position coordinates of the current pixel point in its neighborhood window respectively. For example, in a 3*3 neighborhood window, the position of the center pixel point is (0,0), and the position of the upper left corner pixel point is (-1,-1), and μ(x,y) represents the mean value within the neighborhood.

[0075] The formula D(x,y) represents the variance of the pixel point at coordinate (x,y) and its surrounding (2k+1)*(2k+1) neighborhood. The larger the variance D, the greater the degree of dispersion of the pixel matrix in this area, and the standard deviation σ of the Gaussian filter needs to be reduced to reduce the smoothing effect; conversely, when the variance D is small, σ should be increased to enhance the smoothing effect.

[0076] Step S13: dynamically adjusting the standard deviation in the neighborhood using a preset amplitude adjustment coefficient and variance, wherein the standard deviation is used to adjust the smoothing strength of the Gaussian filter.

[0077] The processor selects a preset amplitude adjustment coefficient α, and then dynamically adjusts the standard deviation based on the variance and the amplitude adjustment coefficient. The standard deviation is calculated as:

[0078]

[0079] Among them, σ(x,y) represents the standard deviation in the neighborhood, σbase is a basic standard deviation value, α is the amplitude adjustment coefficient, and D(x,y) is the variance in the neighborhood.

[0080] The specific value of α is determined manually based on experience, and is usually a value between 0 and 1; the value of σbase is also determined manually.

[0081] When D(x,y) increases, σ(x,y) decreases; when D(x,y) decreases, σ(x,y) increases. The larger the variance, the more dramatic the change in pixel values ​​in the area, which may contain more details or noise. Therefore, the adaptive Gaussian filter algorithm dynamically adjusts the standard deviation according to the size of the variance, so as to retain the important features of the image as much as possible while removing noise.

[0082] Step S14: After filtering the road image using the dynamically adjusted standard deviation, a contrast function is constructed based on the product of the standard deviation and a preset two-dimensional Gaussian filter function.

[0083] The processor generates a Gaussian kernel according to the dynamically adjusted standard deviation σ(x,y), and then uses the generated Gaussian kernel to perform a convolution operation on the pixel values in the neighborhood, thereby filtering the road image. Then, a contrast function is constructed based on the product of the standard deviation σ(x,y) and the preset two-dimensional Gaussian filtering function P(x,y;σ). The calculation formula of the contrast function is as follows:

[0084]

[0085] where F(x,y) represents the contrast function, σ(x,y) is the standard deviation, P(x,y;σ) is the preset two-dimensional Gaussian filtering function, σbase is a basic standard deviation value, and k is a preset parameter.

[0086] Among them, the preset two-dimensional Gaussian filtering function is a fixed Gaussian function, and its standard deviation is preset.

[0087] Step S15: If the function value of the contrast function does not reach the set threshold, recalculate the variance according to the filtered road image until the function value of each pixel point reaches the set threshold. Here, the set threshold is used to indicate that the dispersion degree of the pixel point is the same as that of the two-dimensional Gaussian filtering function.

[0088] In the embodiment of the present application, a set threshold T is defined to indicate that the dispersion degree of the pixel point is the same as that of the two-dimensional Gaussian filtering function. For each pixel point (x,y), check whether the function value of the contrast function reaches the set threshold T. If the contrast function value does not reach the set threshold, recalculate the variance D according to the filtered road image, and repeat steps S12 to S15 until the contrast function value of each pixel point reaches the set threshold.

[0089] When F = 1, that is, the dispersion degree of the pixel values in the image area is the same as that of the two-dimensional Gaussian filtering function, which means that the weight of the parameters in the Gaussian kernel is closest to the weight of the matrix pixel gray values in the area. For the convenience of actual processing, a Gaussian kernel size of 3*3 or 5*5 is usually selected. Therefore, the standard deviation σ is determined by the variance D of the pixel values in the image area. By analogy, through repeated iteration, the value of the standard deviation changes with the change of the variance D in each iteration, so as to obtain an adaptive Gaussian filtering based on the improvement of the traditional Gaussian filtering algorithm. The simplified formula of the adaptive Gaussian filtering is as follows:

[0090]

[0091] As can be seen from the above formula, when F is a fixed value, the larger the variance D, the smaller the standard deviation σ; vice versa. Adaptive Gaussian filtering dynamically adjusts the standard deviation of the Gaussian kernel according to the characteristics of the local area of the image, so a larger standard deviation is used in areas with more noise to more effectively filter out the noise; in areas with less noise, the standard deviation is smaller, avoiding over-smoothing and retaining more image details. By dynamically adjusting the standard deviation, the adaptive Gaussian filtering algorithm can retain the details of the image while filtering out the noise, and has better adaptability and flexibility.

[0092] As an alternative implementation, in step 2022, the object detection model can adopt the YOLOv5s algorithm. As an advanced single-stage object detection model, the YOLOv5s algorithm mainly consists of four core modules: the input end, the backbone network, the Neck network (feature fusion part), and the Head output layer, making it outstanding in dealing with detection tasks in complex scenarios such as the identification of different road surface cracks and having excellent performance.

[0093] The training process of the object detection model is as follows: First, a high-quality and diverse training dataset is constructed to train the YOLOv5s algorithm. The embodiments of the present application use a publicly available dataset, which contains multiple images covering various road conditions. These images detail the most common road crack types such as horizontal, vertical, and reticular. To ensure the effectiveness and reliability of the data, the images are strictly sorted and labeled according to the VOC2007 training dataset format, which helps improve the efficiency and accuracy of model training. In the division of the dataset, the golden ratio of 7:2:1 is followed, and the dataset is divided into a data training set, a result verification set, and an actual test set. Such a division strategy not only ensures that the object detection model can fully learn crack features during the training process but also provides enough samples for it to evaluate its generalization ability during the verification and testing phases.

[0094] Regarding the diversity of road cracks, there are mainly three main crack types: A1 (horizontal crack), A2 (vertical crack), and A3 (reticular crack). To accurately identify these cracks, a labeling tool is used to detail the cracks in each image and assign them corresponding category labels. This step is crucial for the accuracy during the algorithm recognition process and directly determines whether different types of cracks can be accurately distinguished, thus effectively reducing the false detection rate and the missed detection rate.

[0095] As an alternative implementation, before using the road friction coefficient model, it is necessary to construct and train the road friction coefficient model. The pre-training process of the road friction coefficient model includes:

[0096] Step S21: Obtain training samples through a friction coefficient tester. The training samples include the number of cracks corresponding to each crack type in the sample section and the road friction coefficient corresponding to each crack type.

[0097] Step S22: Use the training samples to preliminarily train the initial road friction coefficient model.

[0098] Step S23: Use the set of crack numbers in the validation samples and the training samples to verify the road friction coefficient model after preliminary training, and obtain the output result.

[0099] Step S24: If the output result is different from the road friction coefficient in the validation samples, adjust the parameters in the road friction coefficient model after preliminary training until a trained road friction coefficient model is obtained, where the output result of the trained road friction coefficient model is the same as the road friction coefficient in the validation samples.

[0100] In step S21, use a friction coefficient tester to test on multiple sample sections, identify the crack type and quantity of each section, count the friction coefficients at multiple measurement points of each section and take the average value as the road friction coefficient of that section. Then record the number of cracks of each crack type (such as longitudinal cracks, transverse cracks, reticular cracks, etc.) and the corresponding road friction coefficient in each sample section, and organize the collected data into a training sample set. Each sample contains the crack type and quantity and the corresponding friction coefficient.

[0101] In step S22, preprocess the training samples, including data cleaning, normalization, etc., to improve the data quality and the accuracy of model training. Then use the training samples to preliminarily train the initial road friction coefficient model. During the training process, the model will learn the relationship between the number of cracks and the road friction coefficient. Among them, the road friction coefficient model can be selected from linear regression, random forest, Generalized Regression Neural Network (GRNN), etc. The formula for normalization processing is:

[0102]

[0103] where, is the normalized data, S is the data before normalization, S max is the maximum value in the vector, S min is the minimum value in the vector elements.

[0104] In step S23, a set of independent verification samples that have not been used for preliminary training are prepared to verify the performance of the model. Then, the set of crack numbers in the verification samples is input into the preliminarily trained model to obtain the output result of the model, and the output result of the model, that is, the predicted road friction coefficient, is recorded.

[0105] In step S24, the processor calculates the error between the output result of the model and the actual road friction coefficient in the verification samples. Common error metrics include mean square error, mean absolute error, etc. Then, the parameters of the model are adjusted according to the error situation. Optimization algorithms such as gradient descent and stochastic gradient descent can be used for parameter adjustment. Steps S23 and S24 are repeated until the error between the output result of the model and the actual road friction coefficient in the verification samples is within an acceptable range, so as to use the trained model to evaluate the road friction coefficient.

[0106] Through the training of the road friction coefficient model, it can be found that there is a certain linear relationship between the crack number and the road friction coefficient. When the crack number is small, the road friction coefficient is relatively small; when the crack number is large, the road friction coefficient is relatively large. Then, road cracks will directly affect the road friction coefficient during vehicle driving. When the vehicle passes through a crack, the friction coefficient will increase significantly and the braking distance will be extended.

[0107] As an alternative implementation, the embodiment of the present application uses a generalized regression neural network as the road friction coefficient model. The generalized regression neural network includes an input layer, a pattern layer, a summation layer, and an output layer. As Figure 3 shown, the training process of verifying the road friction coefficient model includes the following steps:

[0108] Step S31: Take the set of crack numbers in the verification samples as input data and input it into the input layer.

[0109] The processor takes the verification sample X = [X1, X2,..., X z T as input data and inputs it into the input layer. X represents the set of crack numbers, X1 represents the number of cracks of the first crack type, X2 represents the number of cracks of the second crack type,..., and Xz represents the number of cracks of the zth crack type. T represents matrix transpose.

[0110] Step S32: Determine the weights of the pattern layer based on the input data and the training sample Ri. Among them, the training sample Ri is used to indicate the crack number corresponding to the ith neuron in the input layer, and the weight is used to indicate the similarity between the input data X and the training sample Ri.

[0111] The calculation formula for the weights of the pattern layer is:

[0112] ​

[0113] Among them, Pi is the weight of the pattern layer, representing the similarity degree between the input data X and the training sample Ri. X is the input data, Ri is the training sample of the i-th neuron in the input layer, T represents matrix transpose, σ is the smoothing factor, m is the number of neurons in the input layer, and i represents the i-th neuron.

[0114] Step S33: Determine the weight sum value of each weight output by the pattern layer, and determine the output value of the summation layer based on the sum of the products of the j-th output value of all training samples and the corresponding weights. Among them, each training sample Ri corresponds to an output sample, and the output value of the summation layer is used to indicate the value of the j-th element of the j-th output sample.

[0115] Among them, the calculation formula for the weight sum value is:

[0116] The calculation formula for the output value of the summation layer is:

[0117]

[0118] Among them, S Nj is the output value of the summation layer, representing the value of the j-th element of the j-th output sample. Yij represents the j-th output value of the i-th training sample, Pi represents the similarity degree between the input data X and the training sample Ri, n is the number of training samples, w is the dimension of the output sample, and j is the j-th output sample.

[0119] The meaning of this formula is that in order to obtain the value S Nj of the j-th element of the j-th output sample, it is necessary to traverse all training samples, multiply the j-th element Yij of each training sample by its corresponding weight Pi, and then add up these products.

[0120] Step S34: Determine the output result of the j-th element in the output layer according to the quotient of the output value of the summation layer and the weight sum value.

[0121] The calculation formula for the output result of the j-th element in the output layer is:

[0122] Among them, Yj represents the output result of the j-th element in the output layer.

[0123] In summary, the training process of the above road friction coefficient model is as follows:

[0124] 1. Input data X: Assume that X is a set containing the number of cracks, for example, X = [X1, X2,..., X z T .

[0125] ​2. Training samples: Assume that there are m neurons in the input layer, and each neuron corresponds to a training sample. For example, the number of cracks in the training sample corresponding to the i-th neuron is Ri = [R1, R2, …, Rm].

[0126] 3. Output sample Yi: Each training sample Ri corresponds to an output sample Yi. Assume that each output sample has w output values. For example, the output sample corresponding to the i-th training sample is Yi = [Yi1, Yi2, …, Yiw].

[0127] 4. Similarity degree weight Pi: Calculate the similarity degree between the input data X and the i-th training sample Ri to obtain the weight Pi.

[0128] 5. Weight sum value S D :

[0129] 6. Output S of the summation layer Nj : For the j-th element S of the j-th output sample Nj , calculate the sum of the products of the j-th output value Yij of all training samples and the corresponding weight Pi.

[0130] 7. Output result of the output layer:

[0131] Based on the same technical concept, this application provides a method for determining the road friction coefficient, including the following steps:

[0132] Step 1: Obtain training samples through a friction coefficient tester. The training samples include the number of cracks corresponding to each crack type in the sample section and the road friction coefficient corresponding to each crack type.

[0133] Step 2: Use the training samples to preliminarily train the initial road friction coefficient model.

[0134] Step 3: Input the validation sample X = [X1, X2,..., X z T as input data into the input layer.

[0135] Step 4: Calculate the similarity degree weight Pi:

[0136]

[0137] Step 5: Calculate the weight sum value S D : and calculate the output value S of the summation layer Nj :

[0138]

[0139] ​Step 6: The calculation formula for the output result of the j-th element in the output layer is as follows:

[0140]

[0141] Step 7: Adjust the parameters of the road friction coefficient prediction model to obtain a trained road friction coefficient model and optimize the prediction performance.

[0142] Step 8: Obtain the road image of the target section sent by the photographing device.

[0143] Step 9: Adjust the standard deviation using the adaptive Gaussian filtering algorithm to filter out the noise in the road image.

[0144] Step 10: Identify the crack types included in the road image through the target detection model and count the number of cracks of each crack type in the road image.

[0145] Step 11: Use the pre-trained road friction coefficient model to determine the road friction coefficient of the target section.

[0146] The embodiments of the present application can be applied in fields such as road engineering construction, traffic safety management, or intelligent driving. On the one hand, road images can be obtained through a camera, and then the road surface can be analyzed through image processing and computer vision technologies to estimate the road friction coefficient. For the road engineering construction field, this can more accurately evaluate the condition and safety of the road, thereby improving the quality and efficiency of road construction. On the other hand, for the traffic safety management field, real-time road friction coefficient information can be provided, which helps the traffic management department take corresponding measures in a timely manner to ensure traffic safety. On the further hand, for the intelligent driving field, real-time road friction coefficient information can be provided for autonomous driving vehicles, which helps autonomous driving vehicles make more accurate decisions and improve driving safety.

[0147] The beneficial effects of the embodiments of the present application are as follows: 1. Non-contact measurement: The road image is obtained through a camera and then analyzed through image processing and computer vision techniques. There is no need for actual driving tests on the road, which is a non-contact measurement method, saving time and effort and avoiding damage to the road. 2. Real-time and high efficiency: Compared with the existing driving test-based methods, the road surface can be quickly analyzed by obtaining road images in real time, thereby estimating the road friction coefficient, greatly improving the measurement efficiency. 3. Strong adaptability: Based on computer vision and image processing techniques, it can handle various complex road conditions, such as rainy days and snowy days, with strong adaptability. In related technologies, the performance of driving test-based methods may be affected under these complex road conditions. 4. High accuracy of prediction results: Through image processing and computer vision techniques, features that can reflect the road friction coefficient can be more accurately extracted, and then predicted through a machine learning model, thereby improving the accuracy of the prediction results.

[0148] Based on the same technical concept, the present application provides a schematic diagram of a device for determining the road friction coefficient, as Figure 4 shown, including the following:

[0149] An acquisition module 401, configured to acquire a road image of a target section sent by a photographing device;

[0150] An identification module 402, configured to process the road image through a target recognition model to identify at least one type of crack included in the road image, where different types of cracks have different degrees of influence on the road friction coefficient;

[0151] A statistics module 403, configured to count the number of cracks of each type of crack in the road image;

[0152] An input / output module 404, configured to input the number of cracks of each type of crack into a pre-trained road friction coefficient model to obtain the road friction coefficient of the target section output by the road friction coefficient model.

[0153] Optionally, the identification module 402 is configured to:

[0154] Filter out the noise in the road image by dynamically adjusting the standard deviation in the Gaussian filtering algorithm;

[0155] Process the road image after filtering out the noise through a target detection model to identify at least one type of crack included in the road image.

[0156] Optionally, the identification module 402 is configured to:

[0157] Determine each pixel point in the road image and the neighborhood corresponding to each pixel point;

[0158] Determine the variance in the neighborhood corresponding to the pixel, where the variance is used to indicate the degree of dispersion of the pixel;

[0159] Dynamically adjust the standard deviation within the neighborhood using a preset amplitude adjustment coefficient and the variance, where the standard deviation is used to adjust the smoothing intensity of the Gaussian filter;

[0160] After filtering the road image using the dynamically adjusted standard deviation, construct a comparison function based on the product of the standard deviation and a preset two-dimensional Gaussian filtering function;

[0161] If the function value of the comparison function does not reach the set threshold, recalculate the variance based on the filtered road image until the function value of each pixel reaches the set threshold, where the set threshold is used to indicate that the degree of dispersion of the pixel is the same as that of the two-dimensional Gaussian filtering function.

[0162] Optionally, the device is further configured to:

[0163] Obtain training samples through a friction coefficient tester, where the training samples include the number of cracks corresponding to each crack type in the sample section and the road friction coefficient corresponding to each crack type;

[0164] Preliminarily train an initial road friction coefficient model using the training samples;

[0165] Verify the preliminarily trained road friction coefficient model using the set of crack numbers in the validation samples and the training samples to obtain an output result;

[0166] If the output result is different from the road friction coefficient in the validation samples, adjust the parameters in the preliminarily trained road friction coefficient model until a trained road friction coefficient model is obtained, where the output result of the trained road friction coefficient model is the same as the road friction coefficient in the validation samples.

[0167] Optionally, the initial road friction coefficient model is a generalized regression neural network, and the generalized regression neural network includes an input layer, a pattern layer, a summation layer, and an output layer. The device is further configured to:

[0168] Use the set of crack numbers in the validation samples as input data and input it into the input layer;

[0169] Determine the weights of the pattern layer based on the input data and the training sample Ri, where the training sample Ri is used to indicate the number of cracks corresponding to the i-th neuron in the input layer, and the weights are used to indicate the similarity between the input data X and the training sample Ri;

[0170] Determine the weighted sum value of each weight output by the pattern layer, and determine the output value of the summation layer based on the sum of the products of the j-th output value of all training samples and the corresponding weights. Here, each training sample Ri corresponds to an output sample, and the output value of the summation layer is used to indicate the value of the j-th element of the j-th output sample;

[0171] Determine the output result of the j-th element in the output layer according to the quotient of the output value of the summation layer and the weighted sum value.

[0172] Optionally, the calculation formula for the weight of the pattern layer is:

[0173]

[0174] where Pi is the weight of the pattern layer, X is the input data, Ri is the training sample of the i-th neuron in the input layer, T represents matrix transpose, σ is the smoothing factor, and n is the number of neurons in the input layer.

[0175] Optionally, the calculation formula for the output value of the summation layer is:

[0176]

[0177] S Nj represents the output value of the summation layer, Yij represents the j-th output value of the i-th training sample, Pi represents the similarity degree between the input data X and the training sample Ri, n is the number of training samples, and w is the dimension of the output sample.

[0178] As Figure 5 shown, an embodiment of the present application provides an electronic device, including a processor 501, a communication interface 502, a memory 503, and a communication bus 504. Among them, the processor 501, the communication interface 502, and the memory 503 complete communication with each other through the communication bus 504.

[0179] The memory 503 is used to store a computer program.

[0180] In an embodiment of the present application, when the processor 501 is used to execute the program stored on the memory 503, it implements the method for determining the road friction coefficient provided by any one of the foregoing method embodiments.

[0181] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method for determining the road friction coefficient provided by any one of the foregoing method embodiments.

[0182] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0183] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0184] It should be understood that the terms used in this document are only for the purpose of describing specific example embodiments and are not intended to be restrictive. Unless otherwise clearly indicated in the context, the singular forms "a", "an", and "the" used in this document may also represent the plural form. The terms "include", "comprise", "contain", and "have" are inclusive and thus specify the presence of the stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or their combinations. The method steps, processes, and operations described in this document are not to be construed as necessarily requiring them to be executed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps can be used.

[0185] The above description is only the specific implementation manners of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in this document, but will conform to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for determining the road friction coefficient, characterized in that, The method includes: Obtaining a road image of a target road section sent by a photographing device; Processing the road image through a target recognition model to identify at least one type of crack contained in the road image; Counting the number of cracks of each type of crack in the road image, where different types of cracks and crack numbers have different degrees of influence on the road friction coefficient; Inputting the number of cracks of each type of crack into a pre-trained road friction coefficient model to obtain the road friction coefficient of the target road section output by the road friction coefficient model.

2. The method according to claim 1, characterized in that, Processing the road image through a target recognition model to identify at least one type of crack contained in the road image includes: Filtering the noise in the road image by dynamically adjusting the standard deviation in the Gaussian filtering algorithm; Processing the road image after filtering the noise through a target detection model to identify at least one type of crack contained in the road image.

3. The method according to claim 2, wherein Filtering the noise in the road image by dynamically adjusting the standard deviation in the Gaussian filtering algorithm includes: Determining each pixel point in the road image and the neighborhood corresponding to each pixel point; Determining the variance in the neighborhood corresponding to the pixel point, where the variance is used to indicate the degree of dispersion of the pixel point; Dynamically adjusting the standard deviation in the neighborhood by using a preset amplitude adjustment coefficient and the variance, where the standard deviation is used to adjust the smoothing intensity of the Gaussian filter; After filtering the road image by using the dynamically adjusted standard deviation, constructing a comparison function according to the product of the standard deviation and a preset two-dimensional Gaussian filtering function; If the function value of the comparison function does not reach a set threshold, recalculating the variance according to the road image after filtering until the function value of each pixel point reaches the set threshold, where the set threshold is used to indicate that the degree of dispersion of the pixel point is the same as the degree of dispersion of the two-dimensional Gaussian filtering function.

4. The method according to claim 1, wherein The pre-training process of the road friction coefficient model includes: Obtaining training samples through a friction coefficient tester, where the training samples include the number of cracks corresponding to each type of crack in a sample road section and the road friction coefficient corresponding to each type of crack; Performing preliminary training on an initial road friction coefficient model by using the training samples; Validating the preliminarily trained road friction coefficient model by using the set of crack numbers in the validation samples and the training samples to obtain an output result; If the output result is different from the road friction coefficient in the validation samples, adjusting the parameters in the preliminarily trained road friction coefficient model until a trained road friction coefficient model is obtained, where the output result of the trained road friction coefficient model is the same as the road friction coefficient in the validation samples.

5. The method according to claim 4, wherein The initial road friction coefficient model is a generalized regression neural network, and the generalized regression neural network includes an input layer, a pattern layer, a summation layer, and an output layer. Validating the preliminarily trained road friction coefficient model by using the set of crack numbers in the validation samples and the training samples to obtain an output result includes: Taking the set of crack numbers in the validation samples as input data and inputting it into the input layer; Determine the weights of the pattern layer based on the input data and the training sample Ri, where the training sample Ri is used to indicate the number of cracks corresponding to the i-th neuron in the input layer, and the weights are used to indicate the similarity between the input data and the training sample Ri; Determine the weight sum value of each weight output by the pattern layer, and determine the output value of the summation layer based on the sum of the products of the j-th output values of all training samples and the corresponding weights, where each training sample Ri corresponds to an output sample, and the output value of the summation layer is used to indicate the value of the j-th element of the j-th output sample; Determine the output result of the j-th element in the output layer according to the quotient of the output value of the summation layer and the weight sum value.

6. The method according to claim 5, characterized in that The calculation formula for the weights of the pattern layer is: where Pi is the weight of the pattern layer, X is the input data, Ri is the training sample of the i-th neuron in the input layer, T represents matrix transpose, σ is the smoothing factor, and n is the number of neurons in the input layer.

7. The method according to claim 5, wherein The calculation formula for the output value of the summation layer is: Among them, S Nj represents the output value of the summation layer, Yij represents the j-th output value of the i-th training sample, Pi represents the similarity between the input data X and the training sample Ri, n is the number of training samples, and w is the dimension of the output sample.

8. A device for determining the road friction coefficient, characterized in that, The device includes: An acquisition module, configured to acquire a road image of a target section sent by a photographing device; An identification module, configured to process the road image through a target identification model to identify at least one type of crack included in the road image; A statistics module, configured to count the number of cracks of each type of crack in the road image, where different types of cracks and the number of cracks have different degrees of influence on the road friction coefficient; An input-output module, configured to input the number of cracks of each type of crack into a pre-trained road friction coefficient model to obtain the road friction coefficient of the target section output by the road friction coefficient model.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus; The memory is used to store a computer program; The processor is configured to implement the method according to any one of claims 1-7 when executing the program stored on the memory.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1-7 is implemented.

Images