A Fault Prediction Method and Device Based on Vehicle Simulation Model

Abstract

This invention discloses a fault prediction method and apparatus based on a vehicle simulation model, relating to the technical field of image processing. It involves acquiring point cloud data and image data of a target region, projecting the point cloud data onto a two-dimensional plane to obtain a visualized image, enhancing the visualized image to obtain a target image, extracting features from the target image to obtain multi-dimensional features, enhancing the image data based on the multi-dimensional features to obtain target image data, and modeling a vehicle simulation model based on the target image data. Alternatively, it can project a three-dimensional scene onto a two-dimensional plane to obtain a visualized image, perform preprocessing, enhancement, and feature extraction on the image, enhance the image data, and construct a vehicle simulation model based on the enhanced image data, thereby improving the accuracy and reliability of the modeling.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a fault prediction method and apparatus based on a vehicle simulation model. Background Technology

[0002] With the continuous advancement of technology, 3D driving technology has made significant progress in recent years. From the initial construction of 3D scenes to the deep integration of vehicle simulation models and actual operating data, this technology has not only improved the efficiency and safety of driver training, but also provided strong support for the construction of intelligent transportation systems.

[0003] Patent No. CN119672207A discloses a method, apparatus, and device for constructing a three-dimensional scene; obtaining an original scene description and inputting the original scene description into a scene decomposition agent to obtain multiple hierarchical scene descriptions decomposed by the scene decomposition agent, the scene decomposition agent being implemented based on a first major language model; using a target scene construction agent, taking the target hierarchical scene description from the multiple hierarchical scene descriptions as input to generate a target hierarchical three-dimensional scene, the target scene construction agent being implemented based on a second major language model; integrating the target hierarchical three-dimensional scenes to determine the target three-dimensional scene.

[0004] While existing technologies have solved the problem of low efficiency, some issues still exist. For example, during the data acquisition process, environmental factors can reduce the quality of the collected point cloud data, thus degrading the quality of the constructed 3D scene. Summary of the Invention

[0005] The purpose of this invention is to solve the problem mentioned in the background art that the quality of point cloud data collected is reduced due to environmental problems during the data acquisition process, thus resulting in a decrease in the quality of the constructed three-dimensional scene. Therefore, this invention proposes a fault prediction method and device based on a vehicle simulation model.

[0006] A first aspect of this invention provides a fault prediction method based on a vehicle simulation model, the method comprising:

[0007] Acquire point cloud data and image data of the target area, and project the point cloud data onto a two-dimensional plane to obtain a visualized image;

[0008] The visualization image is enhanced to obtain a target image, and features are extracted from the target image to obtain multi-dimensional features;

[0009] The image data is enhanced based on the multi-dimensional features to obtain target image data, and a vehicle simulation model is obtained based on the target image data.

[0010] Optionally, projecting the point cloud data onto a two-dimensional plane to obtain a visualized image includes:

[0011] The point cloud data is preprocessed to obtain initial point cloud data, and the centroid of the initial point cloud data is transformed to the origin using the centroid calculation formula to obtain the target point cloud data.

[0012] The projection angle is calculated based on the target point cloud data, and the target point cloud data is mapped onto a two-dimensional plane based on the projection angle to obtain an initial image;

[0013] The blank points of the initial image are determined, and the blank points are interpolated using the centroid interpolation method to obtain a visualization image.

[0014] Optionally, enhancing the visualized image to obtain the target image includes:

[0015] The visualized image is binarized to obtain a grayscale image. The pixels of the grayscale image are clustered according to the grayscale level to obtain a pixel count set. The pixel count set contains the number of pixels corresponding to any grayscale level.

[0016] A grayscale probability set is calculated based on the set of pixel counts, and a grayscale value set is calculated based on the occurrence probability of grayscale levels in the grayscale probability set; the grayscale probability set contains multiple occurrence probabilities of grayscale levels; the grayscale value set contains multiple grayscale values.

[0017] If the target grayscale value is greater than the maximum grayscale value, then the target grayscale value is updated to the maximum grayscale value;

[0018] If the target grayscale value is less than the minimum grayscale value, then update the target grayscale value to the minimum grayscale value;

[0019] If the minimum grayscale value ≤ the target grayscale value ≤ the maximum grayscale value, then the target grayscale value is retained.

[0020] Optionally, feature extraction is performed on the target image to obtain multi-dimensional features, including:

[0021] Multi-dimensional features are obtained by extracting features from the target image using the target model;

[0022] The working principle of the target model includes:

[0023] The target image is sequentially input into a 1×1 convolutional layer and a depthwise separable convolutional layer to obtain a first feature map. The first feature map is sequentially input into a global average pooling layer, a first fully connected layer, and a second fully connected layer to obtain a second feature map. The second feature map and the first feature map are multiplied together to obtain a third feature map.

[0024] The third feature map is sequentially input into a 1×1 convolutional layer and a Dropout layer to obtain a fourth feature map. The fourth feature map and the target image are added together to obtain a fifth feature map, which is then used as a multi-dimensional feature.

[0025] Optionally, after obtaining a vehicle simulation model based on the target image data, the process includes:

[0026] Obtain the operating data of the target vehicle, and input the vehicle simulation model and the operating data into a preset model to obtain a fault assessment score;

[0027] If the fault assessment score is greater than the risk threshold, the target driver's operation is determined to be a high-risk operation.

[0028] A second aspect of this invention provides a fault prediction device based on a vehicle simulation model, the device comprising:

[0029] The image projection module is used to acquire point cloud data and image data of the target area, and project the point cloud data onto a two-dimensional plane to obtain a visualized image;

[0030] The feature extraction module is used to enhance the visualization image to obtain a target image, and to extract multi-dimensional features from the target image.

[0031] The 3D modeling module is used to enhance the image data based on the multi-dimensional features to obtain target image data, and to model the vehicle simulation model based on the target image data.

[0032] Optionally, the image projection module includes: the centroid transformation module, the angle projection module, and the blank interpolation module.

[0033] The centroid transformation module is used to preprocess the point cloud data to obtain initial point cloud data, and then transform the centroid of the initial point cloud data to the origin using the centroid calculation formula to obtain the target point cloud data.

[0034] The angle projection module is used to calculate the projection angle based on the target point cloud data, and to map the target point cloud data onto a two-dimensional plane based on the projection angle to obtain an initial image;

[0035] The blank interpolation module is used to determine the blank points of the initial image and interpolate the blank points using the centroid interpolation method to obtain a visualized image.

[0036] Optionally, the feature extraction module includes: the pixel clustering module, the grayscale value generation module, and the grayscale value selection module.

[0037] The pixel clustering module is used to binarize the visualized image to obtain a grayscale image, and to cluster the pixels of the grayscale image according to the grayscale level to obtain a pixel count set; the pixel count set contains the number of pixels corresponding to any grayscale level.

[0038] The grayscale value generation module is used to calculate a grayscale probability set based on the pixel quantity set, and to calculate a grayscale value set based on the grayscale level occurrence probability in the grayscale probability set; the grayscale probability set contains multiple grayscale level occurrence probabilities; the grayscale value set contains multiple grayscale values.

[0039] The grayscale value selection module is used to perform the following judgment:

[0040] If the target grayscale value is greater than the maximum grayscale value, then the target grayscale value is updated to the maximum grayscale value;

[0041] If the target grayscale value is less than the minimum grayscale value, then update the target grayscale value to the minimum grayscale value;

[0042] If the minimum grayscale value ≤ the target grayscale value ≤ the maximum grayscale value, then the target grayscale value is retained.

[0043] Optionally, the feature extraction module is further configured to extract multi-dimensional features from the target image using a target model;

[0044] The working principle of the target model includes:

[0045] The target image is sequentially input into a 1×1 convolutional layer and a depthwise separable convolutional layer to obtain a first feature map. The first feature map is sequentially input into a global average pooling layer, a first fully connected layer, and a second fully connected layer to obtain a second feature map. The second feature map and the first feature map are multiplied together to obtain a third feature map.

[0046] The third feature map is sequentially input into a 1×1 convolutional layer and a Dropout layer to obtain a fourth feature map. The fourth feature map and the target image are added together to obtain a fifth feature map, which is then used as a multi-dimensional feature.

[0047] Optionally, the device further includes:

[0048] The fault assessment module is used to acquire the operating data of the target vehicle, and input the vehicle simulation model and the operating data into a preset model to obtain a fault assessment score;

[0049] The operation judgment module is used to determine that the target driver's operation is a high-risk operation if the fault assessment score is greater than the risk threshold.

[0050] The beneficial effects of this invention are:

[0051] This invention proposes a fault prediction method based on a vehicle simulation model. It acquires point cloud data and image data of the target area, projects the point cloud data onto a two-dimensional plane to obtain a visualized image, enhances the visualized image to obtain a target image, extracts features from the target image to obtain multi-dimensional features, enhances the image data based on the multi-dimensional features to obtain target image data, and models a vehicle simulation model based on the target image data. By projecting a three-dimensional scene onto a two-dimensional plane to obtain a visualized image, and performing preprocessing, enhancement, and feature extraction on the image data, a vehicle simulation model is constructed based on the enhanced image data, improving the accuracy and reliability of the modeling. Attached Figure Description

[0052] Figure 1 A flowchart of a fault prediction method based on a vehicle simulation model is provided for an embodiment of the present invention;

[0053] Figure 2 This invention provides a schematic diagram of the model structure for a fault prediction method based on a vehicle simulation model, as shown in this embodiment.

[0054] Figure 3 A schematic diagram of a fault prediction device based on a vehicle simulation model is provided for an embodiment of the present invention. Detailed Implementation

[0055] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0056] This invention provides a fault prediction method based on a vehicle simulation model. See also... Figure 1 , Figure 1 A flowchart illustrating a fault prediction method based on a vehicle simulation model, provided as an embodiment of the present invention. The method includes the following steps:

[0057] S101, acquire point cloud data and image data of the target area, and project the point cloud data onto a two-dimensional plane to obtain a visualized image;

[0058] S102, enhance the visualized image to obtain the target image, and extract features from the target image to obtain multi-dimensional features;

[0059] S103, enhance the image data based on multi-dimensional features to obtain target image data, and model the vehicle simulation model based on the target image data.

[0060] The fault prediction method based on a vehicle simulation model provided by this invention projects a three-dimensional scene onto a two-dimensional plane to obtain a visualized image. The image is then preprocessed, enhanced, and its features are extracted to enhance the image data. Based on the enhanced image data, a vehicle simulation model is constructed, thereby improving the accuracy and reliability of the modeling.

[0061] In one implementation, a vehicle simulation model, or virtual driving platform (driving the vehicle in a three-dimensional environment), can be constructed by acquiring point cloud data and image data. The three-dimensional scene is obtained by modeling using point cloud data. Although there is a lack of accuracy in the point cloud data acquisition process, it is still possible to construct the basic framework of the scene and present the spatial structure of the target area in a three-dimensional form. This provides a model foundation with certain spatial information for subsequent processing and lays the spatial model foundation for subsequent processes.

[0062] In one implementation, a 3D scene is projected onto a 2D plane to obtain a visualized image, and an initial image is obtained through preprocessing. This achieves the conversion from 3D to 2D, allowing complex spatial information to be presented in the form of a 2D image, which facilitates subsequent image processing operations. The preprocessing operation can remove noise and interference from the image, improve the image quality, and make the image clearer and more regular. This provides better data conditions for subsequent image enhancement and feature extraction operations, ensuring the quality and usability of the image data.

[0063] In one implementation, the initial image is enhanced to obtain the target image, and features are extracted from the target image to obtain multi-dimensional features. Image enhancement can further improve the quality and contrast of the image, making important details in the image more prominent and obvious. Feature extraction transforms the key information in the image into the form of multi-dimensional features. Multi-dimensional features can effectively represent important features in the image, enabling more effective extraction and expression of key information in point cloud data, and providing rich feature information for subsequent feature vector-based operations.

[0064] In one implementation, target image data is obtained by enhancing image data based on multi-dimensional features. A vehicle simulation model is then constructed based on this target image data. The extracted multi-dimensional features are used to perform targeted enhancement processing on the image data (real-world images, i.e., actual scenes, not point cloud-converted images), making the image data more suitable for the construction requirements of the vehicle simulation model. The enhanced image data can more accurately reflect the key elements in the target scene, thereby constructing a vehicle simulation model that is closer to actual needs. This improves the accuracy and reliability of the model, enabling it to better simulate vehicle behavior and environmental interactions in the target scene.

[0065] In one embodiment, projecting point cloud data onto a two-dimensional plane to obtain a visualized image includes:

[0066] Preprocessing the point cloud data yields initial point cloud data, and the centroid of the initial point cloud data is transformed to the origin using the centroid calculation formula to obtain the target point cloud data;

[0067] The projection angle is calculated based on the target point cloud data, and the target point cloud data is mapped onto a two-dimensional plane based on the projection angle to obtain the initial image;

[0068] The blank points of the initial image are identified, and the blank points are interpolated using the centroid interpolation method to obtain the visualization image.

[0069] In one implementation, the centroid is calculated using the formula... Translating the initial point cloud centroid to the origin can eliminate spatial positional biases in the point cloud data. Here, N is the total number of points in the point cloud, and P... i Represents the three-dimensional coordinates (x, y) of the i-th point. i ,y i ,z i This unifies the point clouds of vehicles with different postures into the same coordinate system, so that the subsequent projection angle calculation does not depend on the absolute position of the point cloud.

[0070] In one implementation, the projection angle formula θ is used. y =arctan2(y,x) and Mapping a 3D point cloud onto a 2D plane. Here, (x, y, z) are the point cloud coordinates, and θ... y θz and θz represent the projection angles in the horizontal and vertical directions, respectively. While preserving the spatial distribution characteristics of the point cloud, the data volume is compressed, and the perspective distortion of traditional orthographic projection is avoided through angle mapping, so that the two-dimensional image can accurately reflect the proportion of the three-dimensional structure.

[0071] In one implementation, the centroid coordinate interpolation formula is:

[0072] The barycentric coordinate interpolation formula is used to fill the blank points in the projection. Here, (v0,v1,v2) are the coordinates of the triangle vertices, and (u,v,w) are the interpolation weights, which improves the pixel integrity of the two-dimensional image and avoids feature loss caused by sparse point clouds.

[0073] In one embodiment, enhancing a visualized image to obtain a target image includes:

[0074] The visualized image is binarized to obtain a grayscale image. The pixels of the grayscale image are clustered according to the grayscale level to obtain a pixel count set. The pixel count set contains the number of pixels corresponding to any grayscale level.

[0075] A grayscale probability set is calculated based on the set of pixel counts, and a grayscale value set is calculated based on the occurrence probability of grayscale levels in the grayscale probability set; the grayscale probability set contains the occurrence probability of multiple grayscale levels; the grayscale value set contains multiple grayscale values.

[0076] If the target grayscale value is greater than the maximum grayscale value, then the target grayscale value is updated to the maximum grayscale value;

[0077] If the target grayscale value is less than the minimum grayscale value, then update the target grayscale value to the minimum grayscale value;

[0078] If the minimum grayscale value ≤ the target grayscale value ≤ the maximum grayscale value, then the target grayscale value is retained.

[0079] If the target gray value is greater than the maximum gray value and / or the target gray value is less than the minimum gray value, then the target gray value is updated to the maximum gray value and / or the minimum gray value; otherwise, the target gray value is retained. The target gray value is any one of the gray values ​​in the set.

[0080] In one implementation, the visualized image is binarized to obtain a grayscale image, and pixels are clustered according to grayscale levels to obtain a pixel count set. Classifying and statistically analyzing the pixels in the image according to grayscale levels presents the information in the image in a more concise and easier-to-analyze form. Through binarization and pixel clustering, the distribution of pixels at different grayscale levels in the image can be quickly understood, providing basic data for subsequent image analysis and processing, and improving the efficiency and targeting of image processing.

[0081] In one implementation, a grayscale probability set (P) is calculated based on the set of pixel counts. k =N k / N, P k Let N be the probability of gray level k occurring. k Let k be the number of pixels corresponding to gray level k, N be the total number of pixels in the image, and k be the gray level index, ranging from 0 to L-1, where L is the total number of gray levels (e.g., in an 8-bit grayscale image, L = 256). Then, the gray value set is calculated based on the gray level probability (using the formula...). Among them, F k P represents the cumulative distribution function value of gray level k, where k is the gray level index, N is the total number of pixels in the image, and P is the cumulative distribution function value of the gray level k. k Let k be the probability of gray level k; calculate the cumulative probability of each gray level (reflecting the cumulative distribution of pixel values), using the formula... Where G is the target gray value after equalization (obtained by mapping the cumulative probability of the original gray level k to a new gray range), and F k Let F be the cumulative distribution function value of gray level k. min The cumulative distribution function F kThe minimum value of the grayscale probability set (usually the smallest non-zero cumulative probability value in the image, to avoid a zero denominator, used for normalization), L is the number of grayscale levels in the image, i.e., the number of possible grayscale values ​​(e.g., L = 256 in an 8-bit grayscale image), and (L-1) represents the maximum value of the target grayscale range (e.g., 0 to 255). Quantizing the grayscale information in the image allows us to understand the probability of each grayscale level appearing in the image through the grayscale probability set, while the grayscale value set provides representative grayscale values ​​in the image. This quantization process makes the grayscale information of the image clearer and more explicit, facilitating subsequent in-depth analysis and processing of the image, such as image enhancement and feature extraction, thus improving the accuracy and reliability of image processing.

[0082] In one implementation, the target gray value in the gray value set is judged and adjusted. If the target gray value exceeds the maximum gray value or falls below the minimum gray value, it is updated to the corresponding boundary value; otherwise, the original value is retained. This adjustment process can effectively avoid abnormal fluctuations in gray values, ensuring that the gray values ​​in the image are within a reasonable range, thereby improving the image quality and stability, optimizing the visual effect of the image, and reducing image distortion or misjudgment caused by abnormal gray values.

[0083] In one embodiment, feature extraction of the target image yields multi-dimensional features, including:

[0084] Multi-dimensional features are obtained by extracting features from the target image using a target model;

[0085] The working principle of the target model includes:

[0086] The target image is sequentially input into a 1×1 convolutional layer and a depthwise separable convolutional layer to obtain the first feature map. The first feature map is sequentially input into a global average pooling layer, a first fully connected layer, and a second fully connected layer to obtain the second feature map. The second feature map and the first feature map are multiplied to obtain the third feature map.

[0087] The third feature map is sequentially input into a 1×1 convolutional layer and a Dropout layer to obtain the fourth feature map. The fourth feature map is added to the target image to obtain the fifth feature map, which is then used as a multi-dimensional feature.

[0088] In one implementation, see [link to implementation details]. Figure 1 , Figure 2 This invention provides a schematic diagram of the model structure for a fault prediction method based on a vehicle simulation model, as shown in this embodiment. Represents addition. It represents multiplication.

[0089] In one implementation, multi-dimensional features are obtained by extracting features from the target image through a target model. This captures rich information about the target image from different angles and levels. Compared with single-dimensional feature extraction, it can more comprehensively represent the features of the image, providing richer and more detailed feature descriptions for subsequent image analysis and applications, and helping to improve the accuracy and robustness of image processing.

[0090] In one implementation, the first feature map is multiplied by the second feature map to obtain the third feature map, and the fourth feature map is added to the target image to obtain the fifth feature map. Through feature fusion and enhancement operations, feature information extracted at different stages can be effectively integrated, making the features complementary and enhanced to each other. For example, the multiplication of the first and second feature maps can highlight important feature information, while the addition of the fourth feature map to the target image can retain the key information of the original image, avoiding the loss of information during feature extraction, thereby obtaining more representative and discriminative multi-dimensional features, and improving the quality and effectiveness of the features.

[0091] In one implementation, the target model utilizes operations such as 1×1 convolutional layers, depthwise separable convolutional layers, and global average pooling layers. The 1×1 convolutional layers can recombine and weight the feature maps across channels, facilitating the extraction of more abstract features. Depthwise separable convolutional layers reduce computational cost while maintaining good feature extraction capabilities, improving model efficiency. Global average pooling layers reduce the spatial dimensionality of the feature maps, preserving the most important feature information. This combination of operations optimizes the feature extraction process, enabling the model to improve computational efficiency while maintaining feature extraction quality, making it suitable for processing complex image data.

[0092] In one embodiment, after obtaining a vehicle simulation model based on the target image data, the process includes:

[0093] Obtain the target vehicle's operating data, input the vehicle simulation model and operating data into the preset model to obtain a fault assessment score;

[0094] If the fault assessment score is greater than the risk threshold, the target driver's operation is determined to be a high-risk operation.

[0095] In one implementation, a fault assessment score is obtained by inputting a vehicle simulation model and operational data into a preset model, thereby achieving a quantitative assessment of the degree of risk in driver operations. This makes risk assessment more intuitive and operable.

[0096] In one implementation, if the fault assessment score exceeds the risk threshold, the target driver's operation is determined to be a high-risk operation, and high-risk driving behavior is identified in a timely manner to provide the target driver with real-time early warning information.

[0097] In one implementation, operational data includes, for example: Vehicle speed: including instantaneous and average speed, reflecting the vehicle's speed. Acceleration: including longitudinal and lateral acceleration, reflecting the vehicle's acceleration and deceleration. Steering angle: reflecting the vehicle's steering operation, including the steering wheel's rotation angle and frequency. Braking information: including the brake pedal's depressance and frequency, reflecting the vehicle's braking operation. Throttle information: including the accelerator pedal's depressance and frequency, reflecting the vehicle's acceleration operation. Vehicle position and trajectory: obtaining the vehicle's driving trajectory and position information through GPS or other positioning systems. Vehicle attitude: including the vehicle's pitch angle, roll angle, etc., reflecting the vehicle's driving attitude. Vehicle fault information: such as engine fault codes, abnormal tire pressure, etc., reflecting the vehicle's mechanical condition.

[0098] In one implementation, the pre-set model can be a deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN). The model's training data mainly includes a large amount of labeled vehicle simulation model data and vehicle operation data. This data covers various driving scenarios and operational behaviors, as well as corresponding fault assessment scores. The model is trained by learning the relationship between the features in this data and the fault assessment scores. Updated parameters include the model's weights and biases, which are continuously adjusted during training using optimization algorithms (such as gradient descent) to minimize the difference between the fault assessment scores predicted by the model and the actual labeled scores, thereby improving the model's prediction accuracy and reliability.

[0099] Based on the same inventive concept, this invention also provides a fault prediction device based on a vehicle simulation model. See also Figure 3 , Figure 3 A schematic diagram of a fault prediction device based on a vehicle simulation model is provided for an embodiment of the present invention, comprising:

[0100] The image projection module is used to acquire point cloud data and image data of the target area, and project the point cloud data onto a two-dimensional plane to obtain a visualized image.

[0101] The feature extraction module is used to enhance the visualized image to obtain the target image, and to extract multi-dimensional features from the target image.

[0102] The 3D modeling module is used to enhance image data based on multi-dimensional features to obtain target image data, and then to model the vehicle simulation model based on the target image data.

[0103] The fault prediction device based on a vehicle simulation model provided by this invention projects a three-dimensional scene onto a two-dimensional plane to obtain a visualized image, and performs preprocessing, enhancement, and feature extraction on the image to enhance the image data. Based on the enhanced image data, a vehicle simulation model is constructed, thereby improving the accuracy and reliability of modeling.

[0104] In one embodiment, the image projection module includes: a centroid transformation module, an angle projection module, and a blank interpolation module.

[0105] The centroid transformation module is used to preprocess point cloud data to obtain initial point cloud data, and then transform the centroid of the initial point cloud data to the origin using the centroid calculation formula to obtain the target point cloud data.

[0106] The angle projection module is used to calculate the projection angle based on the target point cloud data, and then map the target point cloud data onto a two-dimensional plane to obtain an initial image based on the projection angle.

[0107] The blank interpolation module is used to determine the blank points of the initial image and interpolate the blank points using the centroid interpolation method to obtain a visualized image.

[0108] In one embodiment, the feature extraction module includes: a pixel clustering module, a grayscale value generation module, and a grayscale value selection module.

[0109] The pixel clustering module is used to binarize the visualized image to obtain a grayscale image, and then cluster the pixels of the grayscale image according to the grayscale level to obtain a pixel count set; the pixel count set contains the number of pixels corresponding to any grayscale level.

[0110] The grayscale value generation module is used to calculate a grayscale probability set based on the pixel quantity set, and to calculate a grayscale value set based on the occurrence probability of grayscale levels in the grayscale probability set; the grayscale probability set contains the occurrence probability of multiple grayscale levels; the grayscale value set contains multiple grayscale values.

[0111] The grayscale value selection module is used to perform the following judgments:

[0112] If the target grayscale value is greater than the maximum grayscale value, then the target grayscale value is updated to the maximum grayscale value;

[0113] If the target grayscale value is less than the minimum grayscale value, then update the target grayscale value to the minimum grayscale value;

[0114] If the minimum grayscale value ≤ the target grayscale value ≤ the maximum grayscale value, then the target grayscale value is retained.

[0115] In one embodiment, the feature extraction module is further configured to extract multi-dimensional features from the target image using the target model;

[0116] The working principle of the target model includes:

[0117] The target image is sequentially input into a 1×1 convolutional layer and a depthwise separable convolutional layer to obtain the first feature map. The first feature map is sequentially input into a global average pooling layer, a first fully connected layer, and a second fully connected layer to obtain the second feature map. The second feature map and the first feature map are multiplied to obtain the third feature map.

[0118] The third feature map is sequentially input into a 1×1 convolutional layer and a Dropout layer to obtain the fourth feature map. The fourth feature map is added to the target image to obtain the fifth feature map, which is then used as a multi-dimensional feature.

[0119] In one embodiment, the apparatus further includes:

[0120] The fault assessment module is used to acquire the operating data of the target vehicle, input the vehicle simulation model and operating data into the preset model to obtain the fault assessment score;

[0121] The operation judgment module is used to determine that the target driver's operation is a high-risk operation if the fault assessment score is greater than the risk threshold.

[0122] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A fault prediction method based on a vehicle simulation model, characterized in that, The method includes: Acquire point cloud data and image data of the target area, and project the point cloud data onto a two-dimensional plane to obtain a visualized image; The visualization image is enhanced to obtain a target image, and features are extracted from the target image to obtain multi-dimensional features; The image data is enhanced based on the multi-dimensional features to obtain target image data, and a vehicle simulation model is obtained by modeling based on the target image data. The process of projecting the point cloud data onto a two-dimensional plane to obtain a visualization image includes: preprocessing the point cloud data to obtain initial point cloud data; transforming the centroid of the initial point cloud data to the origin using a centroid calculation formula to obtain target point cloud data; calculating the projection angle based on the target point cloud data; mapping the target point cloud data onto a two-dimensional plane based on the projection angle to obtain an initial image; determining blank points in the initial image; and interpolating the blank points using a centroid interpolation method to obtain a visualization image. Enhancing the visualized image to obtain a target image includes: binarizing the visualized image to obtain a grayscale image; clustering the pixels of the grayscale image according to grayscale levels to obtain a pixel count set; the pixel count set contains the number of pixels corresponding to any grayscale level; calculating a grayscale probability set based on the pixel count set; calculating a grayscale value set based on the occurrence probability of grayscale levels in the grayscale probability set; the grayscale probability set contains multiple occurrence probabilities of grayscale levels; the grayscale value set contains multiple grayscale values; if the target grayscale value > the maximum grayscale value, then update the target grayscale value to the maximum grayscale value; if the target grayscale value < the minimum grayscale value, then update the target grayscale value to the minimum grayscale value; if the minimum grayscale value ≤ the target grayscale value ≤ the maximum grayscale value, then retain the target grayscale value. The method involves extracting multi-dimensional features from the target image, including: extracting multi-dimensional features from the target image using a target model; the working principle of the target model includes: sequentially inputting the target image into a 1×1 convolutional layer and a depthwise separable convolutional layer to obtain a first feature map; sequentially inputting the first feature map into a global average pooling layer, a first fully connected layer, and a second fully connected layer to obtain a second feature map; multiplying the second feature map and the first feature map to obtain a third feature map; sequentially inputting the third feature map into a 1×1 convolutional layer and a Dropout layer to obtain a fourth feature map; adding the fourth feature map and the target image to obtain a fifth feature map; and using the fifth feature map as a multi-dimensional feature. After obtaining a vehicle simulation model based on the target image data, the method further includes: acquiring the target vehicle's operating data, inputting the vehicle simulation model and the operating data into a preset model to obtain a fault assessment score; if the fault assessment score is greater than a risk threshold, then the target driver's operation is determined to be a high-risk operation.

2. A fault prediction device based on a vehicle simulation model, characterized in that, The apparatus includes a module for performing the method of claim 1, specifically comprising: The image projection module is used to acquire point cloud data and image data of the target area, and project the point cloud data onto a two-dimensional plane to obtain a visualized image; The feature extraction module is used to enhance the visualization image to obtain a target image, and to extract multi-dimensional features from the target image. The three-dimensional modeling module is used to enhance the image data according to the multi-dimensional features to obtain target image data, and to model the vehicle simulation model according to the target image data. The image projection module includes: a centroid transformation module, used to preprocess the point cloud data to obtain initial point cloud data, and transform the centroid of the initial point cloud data to the origin using a centroid calculation formula to obtain target point cloud data; an angle projection module, used to calculate the projection angle based on the target point cloud data, and map the target point cloud data onto a two-dimensional plane based on the projection angle to obtain an initial image; and a blank interpolation module, used to determine the blank points of the initial image, and interpolate the blank points using a centroid interpolation method to obtain a visualization image. The feature extraction module includes: a pixel clustering module, used to binarize the visualized image to obtain a grayscale image, and cluster the pixels of the grayscale image according to the grayscale level to obtain a pixel quantity set; a grayscale value generation module, used to calculate a grayscale probability set according to the pixel quantity set, and calculate a grayscale value set according to the occurrence probability of grayscale levels in the grayscale probability set; and a grayscale value selection module, used to perform the following judgments: if the target grayscale value > the maximum grayscale value, then update the target grayscale value to the maximum grayscale value; if the target grayscale value < the minimum grayscale value, then update the target grayscale value to the minimum grayscale value; if the minimum grayscale value ≤ the target grayscale value ≤ the maximum grayscale value, then retain the target grayscale value. The feature extraction module is also used to extract multi-dimensional features from the target image using the target model; The working principle of the target model includes: inputting the target image sequentially into a 1×1 convolutional layer and a depthwise separable convolutional layer to obtain a first feature map; inputting the first feature map sequentially into a global average pooling layer, a first fully connected layer, and a second fully connected layer to obtain a second feature map; multiplying the second feature map and the first feature map to obtain a third feature map; inputting the third feature map sequentially into a 1×1 convolutional layer and a Dropout layer to obtain a fourth feature map; adding the fourth feature map and the target image to obtain a fifth feature map; and using the fifth feature map as a multi-dimensional feature. The device further includes: a fault assessment module, used to acquire the operating data of the target vehicle, input the vehicle simulation model and the operating data into a preset model to obtain a fault assessment score; and an operation judgment module, used to determine that the target driver's operation is a high-risk operation if the fault assessment score is greater than a risk threshold.

Images