Road surface riding comfort evaluation method and device and electronic equipment

Abstract

The invention provides a road surface riding comfort evaluation method and device and electronic equipment, and relates to the technical field of urban planning, and the method comprises the steps: obtaining a continuous vibration signal, all road surface image information and current illumination intensity in a current time period; converting the continuous vibration signal into a riding vibration value, and determining a first riding comfort corresponding to the riding vibration value from a preset relationship between a preset vibration value and a preset comfort value; determining road surface texture features according to all the road surface image information, inputting image shooting angles corresponding to all the road surface image information and the road surface texture features to a preset comfort level prediction model, and obtaining a second riding comfort level output by the preset comfort level prediction model; and determining the riding comfort in the current time period according to the first riding comfort, the second riding comfort and the current illumination intensity. According to the method, the riding comfort can be evaluated more comprehensively and accurately by comprehensively considering the characteristics of the two dimensions of the continuous vibration signal and the road surface image information.

Description

Technical Field

[0001] The present invention relates to the technical field of urban planning, and particularly to an evaluation method, device and electronic device for the riding comfort of a road surface. Background Art

[0002] Chinese invention patent CN 107087026 A discloses a bicycle intelligent auxiliary device with voice interaction and its data processing method. It can display the current environmental quality, the rider's physical condition, and the condition of the bicycle in real time through a mobile phone and upload them to the server. At the same time, the server processes these monitoring data to obtain the intuitive real-time data collected by various sensors, the comprehensive riding comfort, and the safety factor, and recommends and feedbacks them to the rider for adjustment to achieve safety warning. Chinese invention patent CN107290156 A discloses a method and device for evaluating the riding comfort of a road surface by bicycle. It obtains the vibration data felt by the cyclist at the handlebar during a riding process, the vehicle speed data of the bicycle during a riding process, and the driving trajectory data during a riding process through a self-made test device. After smoothing the obtained vibration acceleration data by the Savitzky-Golay method, it calculates the bicycle riding comfort index RC of a certain road surface and the bicycle riding dynamic comfort index DRC of a certain road surface.

[0003] Although the above-mentioned prior art discloses a method for evaluating riding comfort using vibration data, riding comfort cannot be simply considered from only one dimension of the road surface bumpiness. This will make the final evaluation result relatively one-sided and unable to comprehensively evaluate the comfort from multiple dimensions. For example, whether the width of the road surface, the road surface direction, the friction force and smoothness of the road surface, etc. are considered, all of which will affect the riding comfort.

[0004] How to design a method that can comprehensively evaluate the riding comfort of a road surface from multiple dimensions has become a technical problem to be solved urgently at present. At present, there is no technical solution that can solve the above-mentioned technical problems, and there is no evaluation method, device and electronic device for the riding comfort of a road surface. Summary of the Invention

[0005] The present invention provides an evaluation method, device and electronic device for the riding comfort of a road surface, which analyzes the riding comfort of a road surface from two dimensions of vibration and image, and improves the accuracy of the riding comfort of a road surface.

[0006] In a first aspect, the present invention provides an evaluation method for the riding comfort of a road surface, including:

[0007] Obtain the continuous vibration signal, all road surface image information, and the current light intensity within the current time period, where the current time period is a time period intercepted with a preset duration as the time window during a continuous cycling process;

[0008] Convert the continuous vibration signal into a cycling vibration value, and determine the first cycling comfort corresponding to the cycling vibration value from the preset relationship between the preset vibration value and the preset comfort value;

[0009] Determine the road surface texture feature according to all the road surface image information, input the image shooting angle corresponding to all the road surface image information and the road surface texture feature into a preset comfort prediction model, and obtain the second cycling comfort output by the preset comfort prediction model;

[0010] Determine the cycling comfort within the current time period according to the first cycling comfort, the second cycling comfort, and the current light intensity;

[0011] The preset comfort prediction model is determined after training according to the sample texture features corresponding to each sample image information at all sample shooting angles within the preset duration, and the sample cycling comfort corresponding to each sample image information.

[0012] According to the evaluation method for road surface cycling comfort provided by the present invention, the obtaining of the continuous vibration signal, all road surface image information, and the current light intensity within the current time period includes:

[0013] Use an in-vehicle vibration sensor to obtain the continuous vibration signal within the current time period;

[0014] Use an in-vehicle camera to obtain all the road surface image information within the current time period;

[0015] Use a photosensitive sensor to obtain the current light intensity within the current time period.

[0016] According to the evaluation method for road surface cycling comfort provided by the present invention, the road surface texture feature includes road surface textures under different road widths and different road orientations, where different road surface textures are used to characterize different asphalt and aggregate particles;

[0017] The determining of the road surface texture feature according to all the road surface image information includes:

[0018] Use the convolution kernels of the convolutional layer in a preset D-ConvNet model to extract the road surface texture features corresponding to all the road surface image information.

[0019] According to the evaluation method for road surface cycling comfort provided by the present invention, the preset D-ConvNet model further includes a pooling layer, an activation layer, a Drop-out layer, and a Softmax layer;

[0020] The pooling layer is used to downsample the output of the convolutional layer, reduce the spatial size and the matrix spatial dimension, and obtain an image feature matrix;

[0021] The activation layer and the Drop-out layer use the ReLU function to improve the recognition speed of the preset D-ConvNet model and reduce overfitting;

[0022] The Softmax layer is used to classify the comfort level of the image feature matrix to obtain the second riding comfort level.

[0023] According to the method for evaluating the riding comfort on the road surface provided by the present invention, the preset comfort prediction model includes sub-prediction models corresponding to different image shooting angles;

[0024] Inputting the image shooting angle corresponding to all the road surface image information and the road surface texture feature into the preset comfort prediction model to obtain the second riding comfort level output by the preset comfort prediction model includes:

[0025] Determining a target sub-prediction model corresponding to the image shooting angle corresponding to all the road surface image information;

[0026] Inputting the road surface texture feature into the target sub-prediction model to obtain the second riding comfort level output by the target sub-prediction model.

[0027] According to the method for evaluating the riding comfort on the road surface provided by the present invention, determining the riding comfort level during the current period according to the first riding comfort level, the second riding comfort level, and the current light intensity includes:

[0028] When the current light intensity is greater than or equal to the preset light threshold, determining the second riding comfort level as the riding comfort level during the current period;

[0029] When the current light intensity is less than the preset light threshold, determining the first riding comfort level as the riding comfort level during the current period.

[0030] According to the method for evaluating the riding comfort on the road surface provided by the present invention, determining the riding comfort level during the current period according to the first riding comfort level, the second riding comfort level, and the current light intensity includes:

[0031]

[0032] Wherein, C is the riding comfort level during the current period, C1 is the first riding comfort level, C2 is the second riding comfort level, I is the current light intensity, and I max is the maximum light intensity.

[0033] According to the evaluation method for road surface riding comfort provided by the present invention, after determining the riding comfort during the current period, the method further includes:

[0034] Repeatedly execute the following steps:

[0035] Obtain the next vibration signal, the next road surface image information, and the next light intensity during the next period in the continuous riding process;

[0036] Determine the riding comfort during the next period according to the next vibration signal, the next road surface image information, and the next light intensity during the next period;

[0037] Until the continuous riding is completed, use different preset colors to mark and display the different riding comforts during each period in the preset display interface;

[0038] The riding comfort includes very comfortable, comfortable, uncomfortable, and very uncomfortable.

[0039] In a second aspect, an evaluation device for road surface riding comfort is provided, including:

[0040] An acquisition unit, which is used to acquire the continuous vibration signal, all road surface image information, and the current light intensity during the current period, and the current period is a period intercepted with a preset duration as a time window in the continuous riding process;

[0041] A conversion unit, which is used to convert the continuous vibration signal into a riding vibration value, and determine the first riding comfort corresponding to the riding vibration value from the preset relationship between the preset vibration value and the preset comfort value;

[0042] An input unit, which is used to determine the road surface texture feature according to all the road surface image information, input the image shooting angle corresponding to all the road surface image information and the road surface texture feature into a preset comfort prediction model, and obtain the second riding comfort output by the preset comfort prediction model;

[0043] A determination unit, which is used to determine the riding comfort during the current period according to the first riding comfort, the second riding comfort, and the current light intensity;

[0044] The preset comfort prediction model is determined after being trained according to the sample texture features corresponding to each sample image information at all sample shooting angles within the preset duration, and the sample riding comfort corresponding to each sample image information.

[0045] In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the evaluation method for the riding comfort on the road surface is implemented.

[0046] By comprehensively considering the characteristics of continuous vibration signals and road surface image information in these two dimensions, the present invention can evaluate the riding comfort more comprehensively and accurately. The vibration signal directly reflects the impact of road surface unevenness on riding, while the road surface image information provides road surface texture characteristics with different widths and orientations, characterizing the road surface conditions of different asphalt and aggregate particles, thereby analyzing the riding resistance of the current road section.

[0047] Dynamically adjust the weights of the first riding comfort and the second riding comfort according to the current light intensity. When the light intensity is relatively high, it is more dependent on judging the riding comfort through image information; when the light intensity is relatively low, it is more dependent on judging the riding comfort through vibration signals. This adaptive mechanism enables the evaluation result to maintain a high level of accuracy under different light conditions, ensuring the reliability of the evaluation result in weak light or strong light environments; the continuous and visual comfort data can be used to optimize the riding route planning or provide a basis for road maintenance departments to evaluate the road surface quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following-described drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0049] Figure 1 is a schematic flowchart of the evaluation method for the riding comfort on the road surface provided by the present invention;

[0050] Figure 2 is a schematic diagram of unitizing the collected data using the time history provided by the present invention;

[0051] Figure 3 is a schematic diagram of the visual receptive field provided by the present invention;

[0052] Figure 4 is a schematic diagram of the accuracy and prediction results of different asphalt road surfaces provided by the present invention;

[0053] Figure 5 is a schematic diagram of the continuous real-time detection of the riding comfort on the asphalt road surface provided by the present invention;

[0054] Figure 6It is a schematic diagram of the test results of the continuous real-time riding comfort of asphalt pavement provided by the present invention;

[0055] Figure 7 It is a schematic structural diagram of an evaluation device for the riding comfort of the road surface provided by the present invention;

[0056] Figure 8 It is a schematic structural diagram of an electronic device provided by the present invention. Detailed implementation manners

[0057] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0058] The detection of the riding comfort of asphalt pavement requires a more automated, standardized and simplified test method for urban road management and maintenance departments. The present invention provides a detection method with higher environmental inclusiveness, lower technical barriers for users and suitable for large-scale promotion. The deep convolutional neural network (D-Convnet) is an unsupervised deep learning tool, which has unique advantages and prospects for image recognition and processing. A well-trained D-Convnet can accurately classify it by inputting samples (images). With this feature, the present invention uses the deep convolutional neural network as a technical means to construct a D-Convnet for determining the riding comfort of asphalt pavement based on road surface video files. Among the acquisition objects of on-site vibration data, the measured sections are all asphalt pavements. The environment and road surface conditions of the measured sections can include the road surface being covered with sundries (such as leaves, garbage, water stains, etc.). To measure the riding vibration signal while recording the surface topography of the measured asphalt pavement, the present invention uses a test vehicle for system installation, including an acceleration recorder, a GPS recorder, and a motion camera. The results measured by the GPS and the acceleration recorder; the digital video of the road surface topography recorded by the motion camera.

[0059] Figure 1 It is a schematic flow diagram of an evaluation method for the riding comfort of the road surface provided by the present invention. The evaluation method for the riding comfort of the road surface includes:

[0060] Step 101, obtain continuous vibration signals, all road surface image information and the current light intensity within the current time period, where the current time period is a time period intercepted with a preset duration as a time window during a continuous riding process;

[0061] Step 102: Convert the continuous vibration signal into a riding vibration value, and determine a first riding comfort corresponding to the riding vibration value from a preset relationship between the preset vibration value and the preset comfort value;

[0062] Step 103: Determine the road surface texture feature according to all the road surface image information, input the image shooting angle corresponding to all the road surface image information and the road surface texture feature into a preset comfort prediction model, and obtain a second riding comfort output by the preset comfort prediction model;

[0063] Step 104: Determine the riding comfort during the current period according to the first riding comfort, the second riding comfort, and the current light intensity;

[0064] The preset comfort prediction model is determined after being trained according to the sample texture feature corresponding to each sample image information at all sample shooting angles within the preset duration, and the sample riding comfort corresponding to each sample image information.

[0065] In step 101, the obtaining of the continuous vibration signal, all the road surface image information, and the current light intensity during the current period includes: obtaining the continuous vibration signal during the current period by using an in-vehicle vibration sensor; obtaining all the road surface image information during the current period by using an in-vehicle camera; obtaining the current light intensity during the current period by using a photosensitive sensor.

[0066] Optionally, the current period is a period intercepted with a preset duration as a time window during a continuous riding process, which realizes the real-time acquisition of various data during the riding process and provides comprehensive data support for subsequent evaluation.

[0067] Optionally, the present invention uses a deep convolutional neural network as a technical means, takes multiple urban roads as research objects, and uses a dynamic cycling profile sampling (DCPS) system as a data acquisition means to establish a D-Convnet for determining the riding comfort of asphalt pavements based on road surface video files. The test results of the DCPS system used include riding trajectories, riding speeds, mileage, and three-axis riding vibration signals. The DCPS system adds a motion camera for continuously shooting videos of the road surface topography during the riding process on the basis of the DCC system. This camera is specially designed for shooting sports scenes, and its main performance parameters can be: 12 million effective pixels, 30 - 120 frames (fps) / second, and picture quality 720 - 4000 dpi. The shooting parameters used in the present invention are picture quality 1080 dpi, 50 frames / second, and shutter speed 1 / 40. For typical shooting results, the motion camera is firmly installed on the edge of the front frame of the bicycle.

[0068] Figure 2 This is a schematic diagram of the unitization of the collected data using the time history provided by the present invention. The processing of the original test data includes two parts: the unitization of the test data and the preprocessing of the data units. In the unitization of the test data, to meet the basic requirements of the D-ConvNet for the number of input samples, the vibration signals measured on each road section and the digital videos of the road surface are divided into input units with a duration of 5 s according to the time history.

[0069] In step 102, the vibration signal is converted into a riding vibration value that is easy to understand and process, providing a quantitative index for subsequent evaluation. The first riding comfort can be quickly and accurately determined through a preset relationship, improving the efficiency and accuracy of the evaluation. The purpose of the preprocessing of the data units is to convert the unitized vibration signal into a vibration intensity, and then compare it with the vibration-comfort threshold, so as to provide comfort data in the dimension of vibration for calculating the riding comfort at the current time period.

[0070] In step 103, the preprocessing of the data units further includes extracting the image of the road surface from the digital video of the road surface topography, and determining the road surface texture features according to all the road surface image information. The road surface texture features include road surface textures under different road widths and different road orientations. Among them, different road surface textures are used to characterize different asphalt and aggregate particles;

[0071] Determining the road surface texture features according to all the road surface image information includes:

[0072] Using the convolution kernels of the convolutional layer in the preset D-ConvNet model to extract the road surface texture features corresponding to all the road surface image information.

[0073] Optionally, when extracting the road surface image, to avoid the subjective deviation of the samples caused by the extracted image not covering all the roads ridden in 5 s, 50 consecutive digital images are jointly extracted from each 5 s digital video unit according to the time history by using the interest point detection method (IPD) and the adaptive non-maximum suppression method (ANMS). The above processing is all carried out on the Python platform, and the same operation is performed on all road sections. After processing, the digital videos of the road surface topography of the measured road sections are converted into 190 input sample units, each unit contains 50 digital photos of the road surface (less than 50 in some road sections), and a total of 9470 digital photos are involved in the establishment of the deep ConvNet. The photo pixels are 1024×2048 pixel.

[0074] Optionally, the preset D-ConvNet model further includes a pooling layer, an activation layer, a Drop-out layer, and a Softmax layer;

[0075] The pooling layer is used to downsample the output of the convolutional layer, reducing the spatial size and the matrix spatial dimension to obtain an image feature matrix;

[0076] The activation layer and the Drop-out layer use the ReLU function to improve the recognition speed of the preset D-ConvNet model and reduce overfitting;

[0077] The Softmax layer is used to classify the comfort level of the image feature matrix to obtain the second riding comfort level.

[0078] During the training process of the preset D-ConvNet model, the present invention randomly selects 7,104 road surface images from 9,470 road surface images as the training sample database of the convolutional neural network. The vibration data units where these 7,104 images are located are corresponded to the riding comfort levels of their data units, and the comfort levels corresponding to these 7,104 road surface images are manually annotated, represented by "VC" (very comfortable), "C" (comfortable), "U" (uncomfortable), and "VU" (very uncomfortable). The remaining 2,366 images among the 9,470 images are used to establish the test sample database. Similarly, the images in the test sample database are manually annotated. After the data samples are established, the implementation method for evaluating the riding comfort of asphalt pavement based on D-ConvNet includes the following steps: (1) Use stochastic gradient descent to calculate the gradient loss of D-ConvNet once; (2) Use the backpropagation algorithm to correct the weights of D-ConvNet; (3) Repeat (1) and (2) until the accuracy meets the requirements, and save the trained D-ConvNet; (4) Input the test images into the trained D-ConvNet, and the output result is the predicted riding comfort.

[0079] Optionally, in the preset D-ConvNet model, the first layer is the input layer, with a size of 1024×1024×3×2 pixel resolution. The first three dimensions respectively represent the height, width, and channels of the image. The channel is defined as the color space dimension. The last dimension indicates that a continuous image with a resolution of 1024×2048 pixels is cropped into two images of 1024×1024 pixels, and these two images are imported into the input layer together. The purpose of cropping is to simplify the convolutional layer. The input data is converted into a vector of 1×1×384 after convolution and pooling. Then, the vector with 384 elements is input into the fully connected layer and the softmax layer to evaluate the riding comfort.

[0080] Optionally, the convolutional layer extracts features from the input data through three steps. First, an element-wise dot product is performed between the sub-matrix of the input data and the receptive field. The convolutional layer obtains different image features through the convolution operation of each convolutional kernel with the input matrix. The initial weights of the convolutional kernel matrix are randomly generated. The weights are adjusted through the forward feedback algorithm and stochastic gradient descent. The size of the sub-matrix is the same as that of the convolutional kernel. The dot product values are added together to form a new matrix. Finally, each element of the new matrix is added to a value to obtain the output result.

[0081] The pooling layer is another core structural layer of the deep ConvNet model. The pooling layer (also known as downsampling) is used to reduce the spatial size of the input data, and the process of reducing the matrix spatial size is called downsampling. There are 2 types of pooling layers commonly used in the deep ConvNet model. The output result of max pooling is the maximum value of the sub-array of the input data, while the output result of average pooling is the average value of the sub-array. Existing research shows that the max pooling method is more conducive to the fitting of the deep ConvNet model in the object recognition process. Therefore, the deep ConvNet model designed in the present invention adopts max pooling.

[0082] In the neural network model, the most commonly used non-linear activation functions in the activation layer are the Sigmoidal function and the Tanh function. However, research shows that using the Sigmoidal function and the Tanh function as activation functions will significantly reduce the recognition speed of the deep CONVNET. Therefore, the present invention adopts the ReLU function as the activation function of the deep ConvNet model. Compared with the Sigmoidal function and the Tanh function, when the ReLU function is used for positive activation, its output result is not limited by the numerical value, and when it is used for negative activation, its output result is 0. This feature is beneficial to improving the recognition speed of the deep ConvNet. Overfitting has always been a key problem to be overcome in machine learning. In the deep ConvNet model for recognizing road surface features, overfitting is mainly manifested as that the deep ConvNet model can accurately classify the features of the training set, but the recognition results of the feature classification of the validation set and the test set are not ideal. To reduce the occurrence of this situation, the present invention adds a Drop-out layer between the activation layer and the fully connected layer. In the training of the deep ConvNet model, too many neurons will lead to overfitting. The Drop-out layer reduces the influence of overfitting neurons in each layer on subsequent calculations by randomly disconnecting the connections between some neurons. In the present invention, the Drop-out layer reduces the overfitting situation by randomly disconnecting the connection between some convolutional features of the last convolutional layer and the fully connected layer. The Drop-out rate adopted in the present invention is 0.5.

[0083] In order to use the image feature matrix generated by the convolutional layer and the pooling layer to achieve comfort classification, a dedicated classification layer needs to be set in the neural network model. The present invention uses the Softmax function to achieve this goal. In the deep ConvNet model of the present invention, there is a deviation between the initial output result and the actual result of the deep ConvNet model. In order to evaluate this deviation and use it for weight adjustment, the present invention can define a loss function of Softmax. To reduce the calculated deviation, it is necessary to adjust the convolutional kernel matrix and the Softmax matrix through the training of the deep ConvNet model. The present invention uses the stochastic gradient descent and the forward feedback algorithm to train the deep ConvNet. Compared with the traditional overall gradient descent, the stochastic gradient descent only uses part of the training samples to calculate the deviation L in one iteration. In the present invention, 100 images are used as training samples in one iteration.

[0084] Compared with the unsupervised algorithm, the discrimination of cycling comfort based on the deep ConvNet can automatically obtain the features in the image by adjusting the weights and biases of the convolutional kernel. During the training process of the deep ConvNet, there are 7104 training images and 2366 validation images, and the ratio of training images to validation images is about 3:1. The present invention respectively uses the GPU mode and the CPU mode to train, validate and test the deep ConvNet model. After comparison, it takes 80 minutes to iterate 70 times in the GPU mode, while it takes 38 hours to iterate to 70 times in the CPU mode. Considering the computing efficiency, the GPU mode is used for the subsequent testing of the deep ConvNet model. The results show that the highest accuracy rates in the training and validation of the deep ConvNet are 93.62% (the 63rd iteration) and 93.13% (the 48th iteration) respectively. It can be seen that the accuracy of the final deep ConvNet model is very impressive.

[0085] To further analyze the training results of the D-ConvNet model and its recognition ability of cycling comfort, the present invention conducts an analysis of the morphological features of the L1 convolutional kernel in the convolutional layer. Figure 3It is a schematic diagram of the visualized receptive field provided by the present invention. The convolutional kernel is used to extract image features, and the image features are the data basis for the Softmax function to perform image classification. Therefore, the visualized convolutional kernel can be used to analyze whether the training of the D-ConvNet model is sufficient. The ideal morphological features of the L1 convolutional kernel should reflect the features related to riding comfort, such as road surface texture. The morphological features of the I1-II6 convolutional kernels are stripes with different widths and different directions. According to the characteristics of the convolutional operation, these convolutional kernels can extract the macroscopic road surface textures with different widths and different orientations. The morphological features of the II7-III8 convolutional kernels are gray-scale changes of different degrees, and these convolutional kernels can extract the different boundary features of asphalt and aggregate particles on the asphalt pavement. The morphological features of the IV1-IV8 convolutional kernels are gray-scales of different degrees, which are mainly used to stably extract road surface features under different natural conditions such as different lighting conditions. It can be seen that after the training of the present invention is completed, the morphological features of the L1 convolutional kernel of the D-ConvNet model are clear and can be used to extract different road surface features to characterize riding comfort.

[0086] Optionally, to quantitatively determine the minimum number of images required for the training dataset in the D-ConvNet training stage, the output accuracy of the D-ConvNet was compared and studied when the number of images in the dataset was 10%, 40%, 70%, 80%, 90% and 100% of the original dataset. The results show that as the number of images in the training dataset increases, the test accuracy and training accuracy of the D-ConvNet both increase to varying degrees; when the number of training dataset images is 70-90% of the number of original dataset images, the training and test accuracies of the D-ConvNet are relatively close. Therefore, considering the computational efficiency and the accuracy of the D-ConvNet, it is recommended that the capacity of the training dataset should not be less than 80% of the original dataset. To test the performance of the trained D-ConvNet, the test dataset was used to test the performance of the D-ConvNet, and the sample size of the test dataset was 2,366 road surface images. In the test dataset, the riding comfort of the road surface was defined as "very comfortable (VC)", "comfortable (C)", "uncomfortable (U)" and "very uncomfortable (VU)", and the image ratio was close to 1:1:1:1. The images in the test dataset were taken from different shooting angles and lighting conditions, and the gradation types of the asphalt mixtures were different. Therefore, the diversity of the samples in the test dataset is relatively high. Using the above test dataset to test the performance of the D-ConvNet, its accuracy reached 90.11%. The maximum error degree of the established D-ConvNet for judging the riding comfort of the road surface remains in the comfort category adjacent to it.

[0087] Further, the preset comfort prediction model is determined by training based on the sample texture features corresponding to each sample image information at all sample shooting angles within the preset duration and the sample riding comfort corresponding to each sample image information. The present invention can further quantitatively represent the sample riding comfort and characterize it as a numerical value, so that the second riding comfort output by it and the first riding comfort are on the same quantization scale.

[0088] Optionally, the preset comfort prediction model includes sub-prediction models corresponding to different image shooting angles;

[0089] Inputting the image shooting angles corresponding to all the road surface image information and the road surface texture features into the preset comfort prediction model to obtain the second riding comfort output by the preset comfort prediction model includes:

[0090] Determining the target sub-prediction model corresponding to the image shooting angle corresponding to all the road surface image information;

[0091] Inputting the road surface texture features into the target sub-prediction model to obtain the second riding comfort output by the target sub-prediction model.

[0092] Optionally, when evaluating riding comfort, the road surface condition is a key factor. Road surface texture, potholes, cracks, etc. will all affect the smoothness and comfort of riding. And the image shooting angle may affect the performance of road surface features in the image, thus affecting the accuracy of the analysis results. To overcome this technical difficulty, the preset comfort prediction model provided by the present invention is not a single model, but consists of multiple sub-prediction models, each sub-prediction model corresponding to a different image shooting angle. Considering the influence of the image shooting angle on the recognition of road surface features, when all the road surface image information is received, first determine the shooting angles corresponding to these images, and according to the shooting angles, select the corresponding sub-prediction model from the preset comfort prediction model as the target sub-prediction model. After determining the target sub-prediction model, extract the road surface texture features from the road surface image. These features may include the width and narrowness of the road surface, the bending direction of the road surface, roughness, the number and length of cracks, the depth and area of potholes, etc. The target sub-prediction model calculates according to the input road surface texture features and outputs a numerical value representing riding comfort. By training sub-prediction models for different image shooting angles respectively, the road surface features can be recognized and analyzed more accurately, thus improving the prediction accuracy of riding comfort.

[0093] Figure 4It is a schematic diagram of the accuracy of different asphalt pavements provided by the present invention and the prediction results. The surface morphology of the asphalt pavement has significant differences according to the different types of asphalt mixture gradations used. D-ConvNet should have sufficient ability to analyze the comfort of asphalt pavements paved with different materials and gradations during actual use. Therefore, in order to verify whether the classification results of D-ConvNet are sensitive to the gradation types of asphalt mixtures, the present invention selects 4 sections with significantly different surface morphologies as the research objects. Among them, 40 consecutive images are selected for each road surface as training samples, and the shooting angle is 90°±5°. The actual values and predicted values of D-ConvNet for these four road surface morphologies are as Figure 4 shown. Groundtruth represents the accuracy of the training set for classifying supervised learning techniques, also known as calibrated real data; Prediction represents the prediction results of D-ConvNet. By comparing the differences between Ground truth and Prediction, the accuracy of the output results can be judged. It can be seen that the judgment errors of D-ConvNet for the riding comfort of four different asphalt pavements are similar, and the errors are within an acceptable range. The reason is that a large number of convolutional kernels in the convolutional layer can automatically extract a large number of low-level, intermediate-level, and high-level features from the image, such as the surface texture features of different asphalt pavements. For example, the convolutional kernel morphologies of I1 to II6 are stripes with different widths and different directions. According to the characteristics of the convolutional operation, these convolutional kernels can extract the macroscopic textures of the road surface with different widths and different orientations. Therefore, the road surface material stability of D-ConvNet actually indicates that the trained D-ConvNet can distinguish consecutive images of different asphalt pavements and evaluate their riding comfort through their surface features (texture features). To sum up, if the input samples during the establishment of D-ConvNet contain sufficient gradation types of asphalt pavements, the established D-ConvNet can accurately distinguish them, and the discriminant results of riding comfort will have better stability.

[0094] In step 104, the determining the riding comfort during the current period according to the first riding comfort, the second riding comfort, and the current light intensity includes:

[0095] When the current light intensity is greater than or equal to a preset light threshold, determining that the second riding comfort is the riding comfort during the current period;

[0096] When the current light intensity is less than the preset light threshold, determining that the first riding comfort is the riding comfort during the current period.

[0097] Optionally, the riding comfort is affected by various factors, including but not limited to road conditions, weather conditions, traffic conditions, and lighting conditions. In this embodiment, the impact of light intensity on the evaluation of riding comfort is particularly emphasized. Light intensity not only affects the visual comfort of riders but may also affect the accuracy of image recognition technology, thereby affecting the evaluation of riding comfort based on image analysis. When the current light intensity is greater than or equal to the preset light threshold, it is considered that the lighting conditions are good and the image recognition technology can work accurately. Therefore, at this time, the second riding comfort (obtained based on image analysis) is used as the riding comfort during the current period. When the current light intensity is less than the preset light threshold, it is considered that the lighting conditions are poor and may affect the accuracy of image recognition technology. Therefore, at this time, the first riding comfort is used as the riding comfort during the current period. The present invention selects the corresponding riding comfort as the riding comfort during the current period according to the comparison result between the current light intensity and the preset light threshold. This decision-making process ensures that as accurate as possible riding comfort evaluation results can be obtained under different lighting conditions.

[0098] Optionally, when a first riding comfort with a preset duration of 5 s, and / or, a second riding comfort is 0 to 1.72 m / s 2 then the riding comfort is determined as "Very comfortable (VC)"; when it is 1.72 to 2.12 m / s 2 then it is determined as "Comfortable (C)"; when it is 2.12 to 3.19 m / s 2 then it is determined as "Uncomfortable (UC)"; when it is 3.19 m / s 2 or above, it is determined as "Very Uncomfortable (VU)". Through the above method, the riding comfort during the current period can be finally obtained.

[0099] Optionally, determining the riding comfort during the current period according to the first riding comfort, the second riding comfort, and the current light intensity includes:

[0100]

[0101] where C is the riding comfort during the current period, C1 is the first riding comfort, C2 is the second riding comfort, I is the current light intensity, and I max is the maximum light intensity.

[0102] Optionally, in this embodiment, the weights of the first riding comfort and the second riding comfort in the final riding comfort evaluation are determined by calculating the ratio of the current light intensity to the maximum light intensity. When the light intensity is low, the weight of the first riding comfort is large; conversely, when the light intensity is high, the weight of the second riding comfort is large. The present invention comprehensively considers the first riding comfort, the second riding comfort, and the current light intensity to determine the riding comfort during the current period. This method not only improves the accuracy of the evaluation but also enhances the flexibility and practicality of the evaluation.

[0103] Optionally, after determining the riding comfort during the current period, the method further includes:

[0104] Repeatedly execute the following steps:

[0105] Obtain the next vibration signal, the next road surface image information, and the next light intensity in the next period during the continuous riding process;

[0106] Determine the riding comfort in the next period according to the next vibration signal, the next road surface image information, and the next light intensity in the next period;

[0107] Until the continuous riding is completed, use different preset color marks to display the different riding comforts in each period on the preset display interface;

[0108] The riding comfort includes very comfortable, comfortable, uncomfortable, and very uncomfortable.

[0109] Optionally, during the riding process, the system continuously obtains vibration signals, road surface image information, and light intensity. These data are collected in real time and reflect the road surface conditions and light conditions experienced by the rider at different times. The continuous riding process is divided into multiple periods, and each period can be a fixed time interval or can be dynamically divided according to the activities or position changes of the rider. For each period, according to the vibration signal, road surface image information, and light intensity in that period, use the foregoing method to determine the riding comfort, and use different preset color marks to display the riding comfort of each period on the preset display interface (such as a mobile phone APP, a vehicle-mounted display screen, etc.). For example, very comfortable can be represented by green, comfortable by blue, uncomfortable by yellow, and very uncomfortable by red.

[0110] Optionally, after inputting one or more photos reflecting the basic topography of the asphalt pavement, the present invention can make a preliminary and approximate evaluation of the cycling comfort of this section of road. However, as is well known, the surface characteristics of the same section of asphalt pavement may be uneven due to factors such as materials, construction, and maintenance. Even the surface topographies of areas relatively close to each other on the same section of road are also different. Therefore, the discrimination results of the cycling comfort of the measured road section obtained only through a small number of road surface images are difficult to comprehensively and precisely reflect the actual situation of the measured road section. To obtain a comprehensive evaluation result, avoid overgeneralization, and improve the work efficiency of urban road managers, the present invention has developed a Video-based Continuous Real-time Cycling Comfort Evaluation System (VCR) for asphalt pavements based on road surface video files. Continuous real-time cycling comfort evaluation means that by using a continuously moving motion camera (which can be installed on a bicycle or a motor vehicle) to continuously photograph the road surface, the road surface images collected are automatically and real-time input into the D-ConvNet, and then the continuous cycling comfort level of the measured road surface is output in real-time. As can be seen from the above definition, the key points of this technology are continuity and real-time. Whether these two key points can be properly solved depends on the sample input strategy.

[0111] Generally, in the GPU mode, a well-trained D-ConvNet can process continuous road surface images with a pixel size of 1024×2048 pixels at a speed of 0.1 s / image. However, to ensure that road surface details can be clearly recorded, the shooting frame rate adopted is 50 FPS, that is, each second of video consists of 50 continuous images. On the other hand, the measured road surface area within a single image is approximately 1.5 m (width) × 1 m (length). At this shooting area, the time required for a bicycle to pass over a shooting area at the most commonly used cycling speed of 12 - 20 km / h is 0.18 s - 0.30 s. During this time period, the number of images collected by the motion camera is approximately 9 - 15, and the processing time is approximately 0.9 - 1.5 s. Obviously, according to the image processing efficiency of the D-ConvNet (0.1 s / image), continuous real-time comfort evaluation cannot be achieved. In addition, due to the high shooting frame rate of the motion camera, the images obtained during the process of passing over a shooting area are often highly overlapping. These highly overlapping images not only hinder continuous real-time detection and reduce the detection efficiency, but the results obtained are often the same. Therefore, to achieve continuous real-time comfort measurement, it is necessary to reasonably reduce the number of images captured in the video.

[0112] Figure 5It is a schematic diagram of the continuous real-time detection of the riding comfort of asphalt pavement provided by the present invention. Taking Figure 5 the situation shown in (a) as an example: when the driving speed is 20 km / h, the time required for the sports camera to pass over a shooting area (1.5 m (width) × 1 m (length)) is 0.3 s. In this case, within 0.3 s, the sports camera captured a total of 15 images, and 4 of them were selected for display. As Figure 5 shown in (b), the content contained in the 4 images is actually highly overlapping, that is, the latter image contains most of the content (area) of the previous image. Therefore, when conducting continuous real-time comfort testing, the first image at the start of the video file is counted as the extraction starting point, and the time required for the sports camera to pass over a shooting area (1.5 m (width) × 1 m (length)) (0.3 s in this example) is used as an image acquisition interval. One image is automatically extracted from the video file as a sample for input to the D-ConvNet for its comfort evaluation, as Figure 5 shown in c. It should be noted that the sample input strategy for continuous real-time detection requires the sports camera to maintain a stable driving speed, otherwise the input samples may have problems such as insufficient road surface coverage or excessive sample overlap. When the driving speed of the motor vehicle is different from the example Figure 5 , the image sample acquisition time interval should be adjusted according to the number of video frames and the driving speed.

[0113] Figure 6 It is a schematic diagram of the test results of the continuous real-time riding comfort of asphalt pavement provided by the present invention. To verify the accuracy of the continuous real-time riding comfort test method, the road surface video of a certain section is used as the source of input samples for the D-ConvNet. According to the riding speed (GPS record) and the number of video frames (50 FPS) of this section, the sample input strategy is to extract a road surface image from the video every 0.3 s as a representative image of a 1.5 m (width) × 1 m (length) shooting area.

[0114] As Figure 6 shown, Figure 6 (a) shows the on-site comfort test results using the continuous real-time comfort test method and the DCPS system; Figure 6 (b) shows the GPS positioning of the riding trajectory and the uncomfortable area; Figure 6 (c) shows the road surface topography of the uncomfortable area. As Figure 6(a) shows the test results of Arnott Crescent using the continuous real-time comfort test method and the on-site comfort test results using the DCPS system. Obviously, since the detection results of DCPS are calculated in units of 5 seconds, while the continuous real-time cycling comfort test is calculated in units of 0.3 seconds. Therefore, the comfort evaluation results obtained by the continuous real-time cycling comfort test have a higher resolution in the time history.

[0115] In Figure 6 (a), it can be observed at multiple moments that within the interval where DCPS is calculated in units of 5 seconds, the results of the continuous real-time test are different from it. This indicates that the test results of the continuous real-time test are more refined. In addition, another feature of the continuous real-time test is that with the help of GPS, it can accurately locate the uncomfortable section intervals on the road section, and its resolution can reach 1 m. This has important reference value for urban road management and maintenance departments for them to carry out maintenance work. To confirm this feature, based on the cycling trajectory recorded by GPS corresponding to the cycling time, the longitude and latitude of the uncomfortable road surface area can be traced back, such as Figure 6 (b) shows. The length of the uncomfortable road section area can also be calculated through the start and end points of the time history of the uncomfortable area, providing accurate data for later road surface maintenance. Finally, through the start and end points of the time history of the uncomfortable area, the road surface morphology of this area can be traced back from the road surface video file recorded by the test, giving managers an intuitive feeling and impression, such as Figure 6 (c) shows.

[0116] By comprehensively considering the characteristics of the two dimensions of continuous vibration signals and road surface image information, the present invention can evaluate the cycling comfort more comprehensively and accurately. The vibration signal directly reflects the impact of road surface unevenness on cycling, while the road surface image information provides road surface texture characteristics of different widths and orientations, characterizing the road surface conditions of different asphaltenes and aggregate particles, so as to analyze the cycling resistance of the current road section;

[0117] Dynamically adjust the weights of the first cycling comfort and the second cycling comfort according to the current light intensity. When the light intensity is relatively high, it is more dependent on judging the cycling comfort through image information; when the light intensity is relatively low, it is more dependent on judging the cycling comfort through vibration signals. This adaptive mechanism enables the evaluation results to maintain high accuracy under different light conditions, ensuring the reliability of the evaluation results in weak light or strong light environments; The continuous and visual comfort data can be used to optimize the cycling route planning or provide a basis for the road maintenance department to evaluate the road surface quality.

[0118] Figure 7It is a schematic structural diagram of an evaluation device for road surface riding comfort provided by the present invention. The evaluation device for road surface riding comfort includes an acquisition unit 1. The acquisition unit is used to acquire continuous vibration signals, all road surface image information, and the current light intensity within the current period. The current period is a period intercepted with a preset duration as a time window during a continuous riding process. The working principle of the acquisition unit 1 can refer to the foregoing step 101 and will not be elaborated here.

[0119] The evaluation device for road surface riding comfort further includes a conversion unit 2. The conversion unit is used to convert the continuous vibration signal into a riding vibration value, and determine a first riding comfort corresponding to the riding vibration value from a preset relationship between a preset vibration value and a preset comfort value. The working principle of the conversion unit 2 can refer to the foregoing step 102 and will not be elaborated here.

[0120] The evaluation device for road surface riding comfort further includes an input unit 3. The input unit is used to determine the road surface texture feature according to all the road surface image information, input the image shooting angle corresponding to all the road surface image information and the road surface texture feature into a preset comfort prediction model, and obtain a second riding comfort output by the preset comfort prediction model. The working principle of the input unit 3 can refer to the foregoing step 103 and will not be elaborated here.

[0121] The evaluation device for road surface riding comfort further includes a determination unit 4. The determination unit is used to determine the riding comfort within the current period according to the first riding comfort, the second riding comfort, and the current light intensity. The working principle of the determination unit 4 can refer to the foregoing step 104 and will not be elaborated here.

[0122] The preset comfort prediction model is determined after training according to the sample texture feature corresponding to each sample image information at all sample shooting angles within the preset duration, and the sample riding comfort corresponding to each sample image information.

[0123] By comprehensively considering the characteristics of the continuous vibration signal and the road surface image information in these two dimensions, the present invention can evaluate the riding comfort more comprehensively and accurately. The vibration signal directly reflects the impact of road surface unevenness on riding, while the road surface image information provides road surface texture features with different widths and orientations, characterizing the road surface conditions of different asphalt and aggregate particles, so as to analyze the riding resistance of the current section.

[0124] Dynamically adjust the weights of the first riding comfort and the second riding comfort according to the current light intensity. When the light intensity is high, it relies more on judging the riding comfort through image information; when the light intensity is low, it relies more on judging the riding comfort through vibration signals. This adaptive mechanism enables the evaluation results to maintain high accuracy under different light conditions, ensuring the reliability of the evaluation results in low-light or strong-light environments; the continuous and visual comfort data can be used to optimize the riding route planning or provide a basis for the road maintenance department to evaluate the road surface quality.

[0125] Figure 8 is a schematic structural diagram of the electronic device provided by the present invention. As Figure 8 shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the method for evaluating the riding comfort on the road surface. The method includes: obtaining continuous vibration signals, all road surface image information, and the current light intensity within the current time period, where the current time period is a time period intercepted with a preset time length as a time window during a continuous riding process; converting the continuous vibration signals into riding vibration values, and determining the first riding comfort corresponding to the riding vibration values from the preset relationship between the preset vibration values and the preset comfort values; determining the road surface texture features according to all the road surface image information, inputting the image shooting angles corresponding to all the road surface image information and the road surface texture features into a preset comfort prediction model, and obtaining the second riding comfort output by the preset comfort prediction model; determining the riding comfort within the current time period according to the first riding comfort, the second riding comfort, and the current light intensity; the preset comfort prediction model is determined after being trained according to the sample texture features corresponding to each sample image information at all sample shooting angles within the preset time length, and the sample riding comfort corresponding to each sample image information.

[0126] In addition, when the logical instructions in the above-mentioned memory 830 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0127] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute an evaluation method for the comfort of riding on a road provided by the above-mentioned various methods. The method includes: obtaining continuous vibration signals, all road surface image information, and the current light intensity within a current time period, where the current time period is a time period intercepted with a preset time length as a time window during a continuous riding process; converting the continuous vibration signals into riding vibration values, and determining a first riding comfort corresponding to the riding vibration values from a preset relationship between preset vibration values and preset comfort values; determining road surface texture features according to all the road surface image information, inputting the image shooting angles corresponding to all the road surface image information and the road surface texture features into a preset comfort prediction model, and obtaining a second riding comfort output by the preset comfort prediction model; determining the riding comfort within the current time period according to the first riding comfort, the second riding comfort, and the current light intensity; the preset comfort prediction model is determined after being trained according to the sample texture features corresponding to each sample image information at all sample shooting angles within the preset time length, and the sample riding comfort corresponding to each sample image information.

[0128] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements an evaluation method for the riding comfort of a road surface provided by the above-mentioned various methods. The method includes: obtaining continuous vibration signals, all road surface image information, and the current light intensity within a current time period, where the current time period is a time period intercepted with a preset time length as a time window during a continuous riding process; converting the continuous vibration signals into riding vibration values, and determining a first riding comfort corresponding to the riding vibration values from a preset relationship between preset vibration values and preset comfort values; determining road surface texture features according to all the road surface image information, inputting the image shooting angles corresponding to all the road surface image information and the road surface texture features into a preset comfort prediction model, and obtaining a second riding comfort output by the preset comfort prediction model; determining the riding comfort within the current time period according to the first riding comfort, the second riding comfort, and the current light intensity; the preset comfort prediction model is determined after being trained according to the sample texture features corresponding to each sample image information at all sample shooting angles within the preset time length, and the sample riding comfort corresponding to each sample image information.

[0129] 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. Those of ordinary skill in the art can understand and implement it without creative labor.

[0130] 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 necessary 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 prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing 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.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for evaluating the riding comfort of a road surface, characterized in that, Including: Obtaining continuous vibration signals, all road surface image information, and the current light intensity within the current time period, where the current time period is a time period intercepted with a preset duration as a time window during a continuous cycling process; Converting the continuous vibration signals into cycling vibration values, and determining a first cycling comfort level corresponding to the cycling vibration values from a preset relationship between preset vibration values and preset comfort levels; Determining road surface texture features based on all the road surface image information, inputting the image shooting angles corresponding to all the road surface image information and the road surface texture features into a preset comfort prediction model, and obtaining a second cycling comfort level output by the preset comfort prediction model; Determining the cycling comfort level within the current time period based on the first cycling comfort level, the second cycling comfort level, and the current light intensity; The preset comfort prediction model is determined after being trained based on the sample texture features corresponding to each sample image information at all sample shooting angles within the preset duration, and the sample cycling comfort levels corresponding to each sample image information.

2. The evaluation method for the riding comfort of a road surface according to claim 1, wherein The obtaining of the continuous vibration signals, all road surface image information, and the current light intensity within the current time period includes: Using an in-vehicle vibration sensor to obtain continuous vibration signals within the current time period; Using an in-vehicle camera to obtain all road surface image information within the current time period; Using a photosensitive sensor to obtain the current light intensity within the current time period.

3. The evaluation method for pavement riding comfort according to claim 1, characterized in that, The road surface texture features include road surface textures under different road widths and different road orientations, where different road surface textures are used to represent different asphalt and aggregate particles; The determining of the road surface texture features based on all the road surface image information includes: Using the convolution kernels of the convolutional layer in a preset D-ConvNet model to extract the road surface texture features corresponding to all the road surface image information.

4. The evaluation method for the riding comfort of a road surface according to claim 3, characterized in that, The preset D-ConvNet model further includes a pooling layer, an activation layer, a Drop-out layer, and a Softmax layer; The pooling layer is used to perform downsampling on the output of the convolutional layer, reducing the spatial size and the matrix spatial dimensions to obtain an image feature matrix; The activation layer and the Drop-out layer use the ReLU function to improve the recognition speed of the preset D-ConvNet model and reduce overfitting; The Softmax layer is used to perform comfort classification on the image feature matrix to obtain the second cycling comfort level.

5. The evaluation method for the riding comfort of a road surface according to claim 1, characterized in that, The preset comfort prediction model includes sub-prediction models corresponding to different image shooting angles; The inputting of the image shooting angles corresponding to all the road surface image information and the road surface texture features into the preset comfort prediction model to obtain the second cycling comfort level output by the preset comfort prediction model includes: Determining a target sub-prediction model corresponding to the image shooting angle corresponding to all the road surface image information; Inputting the road surface texture features into the target sub-prediction model to obtain the second cycling comfort level output by the target sub-prediction model.

6. The evaluation method for pavement riding comfort according to claim 1, characterized in that, The determining of the cycling comfort level within the current time period based on the first cycling comfort level, the second cycling comfort level, and the current light intensity includes: When the current light intensity is greater than or equal to the preset light threshold, determine that the second riding comfort level is the riding comfort level during the current period; When the current light intensity is less than the preset light threshold, determine that the first riding comfort level is the riding comfort level during the current period.

7. The evaluation method for pavement riding comfort according to claim 1, characterized in that, The determining the riding comfort level during the current period according to the first riding comfort level, the second riding comfort level, and the current light intensity includes: Among them, C is the riding comfort during the current period, C1 is the first riding comfort, C2 is the second riding comfort, I is the current light intensity, and I max is the maximum light intensity.

8. The evaluation method for the riding comfort of the road surface according to claim 1, characterized in that, After determining the riding comfort level during the current period, the method further includes: Repeatedly execute the following steps: Obtain the next vibration signal, the next road surface image information, and the next light intensity during the next period in the continuous riding process; Determine the riding comfort level during the next period according to the next vibration signal, the next road surface image information, and the next light intensity during the next period; Until the continuous riding is completed, use different preset colors to mark and display the different riding comfort levels during each period on the preset display interface; The riding comfort levels include very comfortable, comfortable, uncomfortable, and very uncomfortable.

9. An evaluation device for the riding comfort of a road surface, characterized in that, including: An acquisition unit, the acquisition unit is used to acquire the continuous vibration signal, all road surface image information, and the current light intensity during the current period, and the current period is a period intercepted with a preset duration as a time window in the continuous riding process; A conversion unit, the conversion unit is used to convert the continuous vibration signal into a riding vibration value, and determine the first riding comfort level corresponding to the riding vibration value from the preset relationship between the preset vibration value and the preset comfort value; An input unit, the input unit is used to determine the road surface texture feature according to all the road surface image information, input the image shooting angle corresponding to all the road surface image information and the road surface texture feature into the preset comfort prediction model, and obtain the second riding comfort level output by the preset comfort prediction model; A determination unit, the determination unit is used to determine the riding comfort level during the current period according to the first riding comfort level, the second riding comfort level, and the current light intensity; The preset comfort prediction model is determined after being trained according to the sample texture feature corresponding to each sample image information at all sample shooting angles within the preset duration, and the sample riding comfort level corresponding to each sample image information.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the evaluation method for road surface riding comfort as described in any one of claims 1 to 8.

Images