Integrated prediction method and system for vehicle point cloud drag coefficient based on feature recognition

Abstract

The present invention provides a method and system for integrated prediction of the drag coefficient of a whole vehicle point cloud based on feature recognition, which relates to the technical field of drag coefficient prediction. The method and system specifically include obtaining point cloud data of a whole vehicle model and preprocessing the point cloud data; constructing a classification network model, inputting the preprocessed point cloud data into the classification network model for vehicle type classification, and outputting vehicle type categories and probability values ​​corresponding to the vehicle type categories; constructing several specific models and a generalized model corresponding one-to-one to the vehicle type categories; judging whether the maximum probability value in the vehicle type classification label is higher than a preset threshold, and if so, selecting the specific model corresponding to the vehicle type category for drag coefficient prediction; if not, selecting the generalized model for drag coefficient prediction; the present invention greatly improves the prediction speed and the accuracy of the prediction results by classifying vehicle types and matching suitable drag coefficient prediction models.

Description

Technical Field

[0001] The present invention relates to the technical field of drag coefficient prediction, and in particular to a method and system for integrated prediction of the drag coefficient of a vehicle point cloud based on feature recognition. Background Art

[0002] With the development of the automobile industry, aerodynamic performance, especially the air resistance coefficient of the vehicle, has become an important indicator for measuring vehicle performance, improving fuel economy and reducing emissions.

[0003] However, existing drag coefficient prediction methods are mostly based on traditional computational fluid dynamics (CFD) simulations, wind tunnel experiments, or mathematical modeling. However, these methods often suffer from computational complexity, time-consuming processes, or reliance on extensive experimental data. This is particularly true during vehicle design, which requires extensive wind tunnel data or high-precision CFD calculations, making it difficult to optimize vehicle designs quickly and in real time. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for integrated prediction of the drag coefficient of a whole vehicle point cloud based on feature recognition, so as to solve the problem mentioned in the above background technology that the existing technology requires a large amount of wind tunnel test data or high-precision CFD calculations, making it difficult to optimize the vehicle design scheme quickly and in real time.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a method for integrated prediction of the drag coefficient of a whole vehicle point cloud based on feature recognition, comprising: obtaining point cloud data of a whole vehicle model and preprocessing the point cloud data; constructing a classification network model, inputting the preprocessed point cloud data into the classification network model for vehicle type classification, and outputting a vehicle type classification label, wherein the vehicle type classification label includes a vehicle type category and a probability value corresponding to the vehicle type category; constructing a drag coefficient prediction model library, wherein the drag coefficient prediction model library includes several specific models corresponding one-to-one to the vehicle type categories and a generalized model; judging whether the maximum probability value in the vehicle type classification label is higher than a preset threshold, and if so, selecting the specific model corresponding to the vehicle type category for drag coefficient prediction, and if not, selecting the generalized model for drag coefficient prediction.

[0006] Optionally, the preprocessing step specifically includes: performing regularization processing on the point cloud data; performing enhancement processing on the point cloud data after regularization processing; and performing maximum and minimum normalization processing on the point cloud data after enhancement processing.

[0007] Optionally, the steps of constructing a classification network model, inputting the preprocessed point cloud data into the classification network model for vehicle type classification, and outputting vehicle type classification labels specifically include: constructing a PointNet classification network model, processing the point cloud data through several MLP layers of the same structure, generating a global feature vector through a maximum pooling layer, and then performing feature extraction and classification through a fully connected layer, and finally outputting each vehicle type category and the corresponding probability value through a softmax layer.

[0008] Optionally, before the step of constructing the PointNet classification network model, the step also includes: constructing a spatial transformation network model to perform global spatial transformation on the point cloud data.

[0009] Optionally, the vehicle type classification labels include: fastback vehicle type and its corresponding probability value ; Travel backpack models and their corresponding probability values ; Foldback models and their corresponding probability values ;in, 、 、 ,and .

[0010] Optionally, the prediction threshold is ,and .

[0011] Optionally, the specific model is trained using point cloud data of the corresponding vehicle type category, and the generalized model is trained using point cloud data of all vehicle type categories.

[0012] Optionally, each model in the drag coefficient prediction model library is based on a dynamic graph edge convolutional network, which uses the KNN algorithm to calculate the distance between the input point cloud data and its nearest neighbor points and construct an adjacency graph to capture the local features of the point cloud data; the relative position vector of each point is calculated through the edge convolution EdgeConv operation, and the local features are transformed and updated in combination with the multi-layer perceptron MLP to extract the global features of the vehicle point cloud; the local features and the global features are converged, and the drag coefficient prediction value is output through the fully connected layer.

[0013] Optionally, the step of aggregating the local features and the global features includes: aggregating the local features and the global features through maximum pooling and / or average pooling.

[0014] On the other hand, the present invention also provides an integrated prediction system for the drag coefficient of a whole vehicle point cloud based on feature recognition, including: a preprocessing module, used to obtain point cloud data of the whole vehicle model and preprocess the point cloud data; a classification module, used to construct a classification network model, input the preprocessed point cloud data into the classification network model for vehicle type classification, and output a vehicle type classification label, wherein the vehicle type classification label includes a vehicle type category and a probability value corresponding to the vehicle type category; a prediction module, used to construct a drag coefficient prediction model library, wherein the drag coefficient prediction model library includes several specific models corresponding one-to-one to the vehicle type category and a generalized model; a matching module, used to determine whether the maximum probability value in the vehicle type classification label is higher than a preset threshold. If so, the specific model corresponding to the vehicle type category is selected for drag coefficient prediction; if not, the generalized model is selected for drag coefficient prediction.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] The present application first constructs a classification network model, inputs the preprocessed point cloud data into the classification network model for vehicle type classification, and outputs a vehicle type classification label; then, a drag coefficient prediction model library is constructed, which includes several specific models corresponding to the vehicle type categories and a generalized model; finally, the probability value in the classification label is selected to match with the specific model or the generalized model to predict the drag coefficient; the present invention classifies the vehicle type and matches the appropriate drag coefficient prediction model, which greatly improves the prediction speed and the accuracy of the prediction results compared to the traditional method of using a unified model for prediction, which requires a large amount of calculation and has a large calculation error. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the process steps of the present invention.

[0018] Figure 2 Schematic diagram of the overall framework of the system of the present invention.

[0019] Figure 3 Schematic diagram of the construction of a specific model database of the present invention.

[0020] Figure 4 This is a schematic diagram of the feature recognition workflow based on the PointNet network of the present invention.

[0021] Figure 5 Schematic diagram of the vehicle point cloud drag coefficient prediction model of the present invention.

[0022] Figure 6 Schematic diagram of the system structure of the present invention.

[0023] In the figure: 10-preprocessing module, 20-classification module, 30-prediction module, 40-matching module. DETAILED DESCRIPTION

[0024] The following will provide a clear and complete description of the solutions of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of this application refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0027] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0028] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not imply the order of execution. The order of execution of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of this application.

[0029] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0030] Please refer to Figure 1-Figure 2 The present invention provides a method for integrated prediction of vehicle point cloud drag coefficient based on feature recognition, comprising:

[0031] S100: Acquire point cloud data of the entire vehicle model and pre-process the point cloud data.

[0032] Specifically, after starting the program, the whole vehicle model that has been format-converted is input, and the whole vehicle model is converted into point cloud data through file conversion. Each point cloud is composed of multiple points, and each point contains spatial coordinates (x, y, z). These points present the three-dimensional geometric features of the vehicle in space; the preprocessing includes at least one of regularization, enhancement, maximum and minimum normalization, and denoising, which can improve data quality and facilitate subsequent network models to process point cloud data.

[0033] S200: Construct a classification network model, input the preprocessed point cloud data into the classification network model to perform vehicle classification, and output a vehicle classification label, wherein the vehicle classification label includes a vehicle category and a probability value corresponding to the vehicle category.

[0034] Specifically, a classification network model is constructed and the processed point cloud samples are used to train the classification network model. After training, the model can be used to accurately identify the vehicle type of the input point cloud data. The classification network model calculates the probability value corresponding to the point cloud data and the vehicle type through feature extraction and classification algorithms, and then uses this probability value to match the subsequent drag coefficient model. Point cloud data is typically represented by an unordered, sparse set of three-dimensional points. The geometric features of different vehicle models vary significantly, and these differences are crucial for predicting the drag coefficient. By classifying the input point cloud data and assigning it to the most appropriate vehicle type category, the corresponding specific prediction model can be selected for drag coefficient calculation, thereby improving prediction accuracy.

[0035] S300: Construct a drag coefficient prediction model library, wherein the drag coefficient prediction model library includes a plurality of specific models corresponding to the vehicle type categories and a generalized model.

[0036] Specifically, in order to ensure efficient drag coefficient prediction on different types of point cloud data, the present invention constructs a drag coefficient prediction model library, which has multiple drag coefficient models. In this application, the drag coefficient prediction model library includes several specific models and a generalized model. The specific model is a drag coefficient prediction model obtained by training and testing specifically for a type of vehicle. The point cloud data of each vehicle type will be independently trained with a specific model. By specifically training samples of each category, each model can focus on a specific type of data. The specific model corresponds one-to-one with the vehicle type in the category label. The generalized model is a drag coefficient prediction model obtained by training and testing for multiple vehicle types. It can process data from all vehicle types and learn a more extensive drag coefficient prediction method, thereby improving the robustness and generalization ability of the model. These models work together to select the most appropriate model for drag coefficient prediction based on the classification results or generalization judgment of the point cloud data.

[0037] S400: Determine whether the maximum probability value in the vehicle type classification label is higher than a preset threshold. If so, select a specific model corresponding to the vehicle type category to predict the drag coefficient. If not, select a generalized model to predict the drag coefficient.

[0038] Specifically, the classification network model outputs a probability distribution, indicating the likelihood that the input point cloud data belongs to each vehicle category. If the maximum probability value is greater than a pre-set probability threshold, the classification result is used as the type of the point cloud. If the maximum probability value is less than the pre-set probability threshold, the features of the point cloud data are considered too complex to be clearly classified, and the point cloud data is assigned to the generalization model for processing.

[0039] The system uses a classification network model to classify input point cloud data. Once the vehicle type is identified, the system selects the corresponding dedicated prediction model to calculate the drag coefficient, minimizing prediction errors between different vehicle types. Furthermore, if certain input data cannot be confidently classified as a specific vehicle type, it is automatically assigned to a generalized model for prediction, ensuring the system's applicability and robustness.

[0040] As can be understood, the present invention first constructs a classification network model, inputs the preprocessed point cloud data into the classification network model for vehicle type classification, and outputs a vehicle type classification label. Next, a drag coefficient prediction model library is constructed, comprising several specific models corresponding to the vehicle type categories and a generalized model. Finally, the probability values ​​in the classification labels are used to select a specific model or a generalized model for matching, thereby predicting the drag coefficient. The present invention avoids the need to classify vehicle types and match appropriate drag coefficient prediction models. Compared to traditional methods that require large computational load and suffer from large computational errors due to predictions using a unified model, the present invention significantly improves prediction speed and accuracy. The present invention can be widely applied to drag coefficient evaluation and optimization in automotive design, shortening development cycles and reducing R&D costs. Furthermore, compared to existing drag coefficient prediction methods, the present invention effectively addresses the differences in drag coefficient predictions between different vehicle types by using specific models trained for different vehicle types, thereby improving prediction accuracy and reliability. By fully combining the high precision of specific models with the versatility of generalized models, the present invention not only improves prediction accuracy but also enhances the adaptability and robustness of the system for diverse vehicle types. Overall, this application not only improves the accuracy of drag coefficient prediction, but also improves the automation level of the system, which can better support aerodynamic performance optimization in vehicle design.

[0041] Please refer to Figure 2 In some embodiments, the preprocessing step specifically includes: performing regularization processing on the point cloud data; performing enhancement processing on the point cloud data after regularization processing; and performing maximum and minimum normalization processing on the point cloud data after enhancement processing.

[0042] Specifically, the quality and robustness of point cloud data are improved through the preprocessing. Point cloud regularization includes: normalizing point cloud data through rotation, scaling and displacement operations; data enhancement includes: random sampling, interpolation and adding noise points to improve the model's adaptability to different point cloud distributions; normalization includes: unifying the scale of point cloud data to avoid the impact of scale differences on the classification model.

[0043] More specifically, the input point cloud data is a three-dimensional matrix. Assume that the input point cloud has N points, and the feature of each point is a 3D vector (x, y, z); therefore, the input data can be represented as a tensor of shape (N, 3), where N is the number of points and 3 represents the coordinates of each point (x, y, z). Next, data enhancement techniques are applied, including translation (moving the point cloud by random factors), adding Gaussian noise (jittering), and randomly discarding some points in the point cloud. These enhancement operations help the model learn more spatial transformations and noise robustness. Finally, the code normalizes the point cloud data and maps the coordinate values ​​of each point cloud to the range of [0, 1] to facilitate subsequent deep learning training. Specifically, the maximum and minimum normalization method is used to normalize the 3D point cloud data. By traversing each three-dimensional point in the data, setting max and min to normalize the data, the calculation formula is as follows: , where is the value after maximum and minimum normalization, x is the value of the original data, is the minimum value in the original data, is the maximum value in the original data.

[0044] A series of standard processing transformations are performed on the image to transform it into a fixed standard form. The entire data processing process ensures the diversity and standardization of the point cloud data, providing more representative and generalizable data for model training.

[0045] Please refer to Figure 4 In some embodiments, the steps of constructing a classification network model, inputting the preprocessed point cloud data into the classification network model for vehicle type classification, and outputting vehicle type classification labels specifically include: constructing a PointNet classification network model, processing the point cloud data through several MLP layers of the same structure, generating a global feature vector through a maximum pooling layer, and then performing feature extraction and classification through a fully connected layer, and finally outputting each vehicle type category and the corresponding probability value through a softmax layer.

[0046] Specifically, the present application processes point cloud data, improves the robustness of the network model, and ensures the stability of the system under complex data distribution. Since point cloud data does not have a clear structured order, the disordered input point cloud data is processed by a PointNet classification network model with several MLP layers of the same structure; in the process of training the PointNet classification grid, a labeled point cloud dataset is constructed, which contains three-dimensional point cloud samples of fastback vehicle sample data, station wagon vehicle sample data, and folding vehicle sample data; a supervised learning method is used to optimize the model through the cross-entropy loss function; through multiple rounds of iterative training, the classification model has the ability to generalize to new point cloud data.

[0047] Furthermore, the PointNet classification network model extracts the local features of each point through several multi-layer perceptron (MLP) networks with the same structure. The features of each point are processed by MLP to obtain a new feature representation, which is calculated as follows: , where is the result obtained after processing by the multi-layer perceptron. MLP is a combination of a series of linear transformations and activation functions, and x is the input data.

[0048] After each point passes through the MLP, a high-dimensional feature vector is generated, reflecting the local geometric information of the point, which is the local feature. To perform the same processing on each point, the PointNet classification network model adopts a shared weight strategy, that is, using the same MLP structure for all points, avoiding the influence of point order on the network.

[0049] Next, the Max Pooling operation is used to aggregate the local features of each point into a global feature vector. This operation can retain the most significant feature information in the point cloud and ensure that the network outputs a stable global feature representation even if the order of the point cloud is disordered. The formula for the Max Pooling operation is: , where is the feature vector of each point, and max is the maximum pooling operation performed on each feature dimension. is the global eigenvector.

[0050] Global Features This is fed into the fully connected layer for further processing, ultimately outputting a probability distribution for each category. Specifically, the softmax function converts all category scores into a normalized probability distribution, ensuring that the probability of each category lies in the range [0, 1] and that the sum of all category probabilities is 1. The category with the highest probability is the network's prediction. If the highest probability does not exceed 0.8, the data is considered to belong to the generalized model.

[0051] In some embodiments, the step of constructing the PointNet classification network model further includes: constructing a spatial transformation network model to perform global spatial transformation on the point cloud data.

[0052] Specifically, the spatial transformer network STN3d will perform global spatial transformation on the entire point cloud data; the spatial transformer network gradually extracts the features of the point cloud through three 1D convolutional layers and performs maximum pooling to extract global features. After that, the three fully connected layers will further process the features and output a 9-dimensional transformation matrix (3x3), which is used to align the point cloud data so that the input point cloud data can be processed in the same reference frame.

[0053] In some embodiments, the vehicle type classification labels include: fastback vehicle type and its corresponding probability value ; Travel backpack models and their corresponding probability values ; Foldback models and their corresponding probability values ;in, 、 、 ,and .

[0054] It should be understood that in order to achieve the above-mentioned purpose, the present invention adopts the following solution: Unlike using a single generalized model to predict the drag coefficient of multiple vehicle models, the present invention classifies the input point cloud data based on the vehicle's appearance characteristics and selects the most suitable specific prediction model for different body designs, thereby improving the prediction accuracy.

[0055] Specifically, this paper categorizes vehicles into three main categories based on their roofline and rear end structure: fastback, station wagon, and convertible. These three designs encompass the majority of passenger car structures on the market and effectively represent the aerodynamic characteristics of various sedans and some sports vehicles.

[0056] Fastback vehicles are characterized by a roofline that smoothly transitions from the front windshield to the rear, creating an overall streamlined shape, such as some sports cars and sports sedans. These vehicles typically have a low drag coefficient due to the shallow rear inclination angle and late airflow separation.

[0057] A wagon-style vehicle, like a station wagon or hatchback, has a longer roof that tapers sharply toward the rear, creating a larger trunk. Because the rear is relatively vertical, airflow separation zones are more likely to form, resulting in different drag characteristics than fastback or convertible models.

[0058] The folding back type refers to a traditional sedan with a clear three-box structure. The trunk part forms a clear angle with the rear windshield, which makes the air flow pattern at the rear significantly different from the fastback and station wagon types, and usually forms a strong vortex structure in the trunk area.

[0059] Specifically, point cloud data is classified into three categories: fastback, station wagon, and convertible. The system is highly adaptable and supports expansion to multiple vehicle types. Each category model is trained based on the geometric characteristics of each vehicle type to better capture these features and improve prediction accuracy. The classification module outputs a probability distribution representing the likelihood that the input point cloud data belongs to each category. The softmax function converts the scores for all categories into a normalized probability distribution, ensuring that the probability of each category is in the range [0, 1] and that the sum of all category probabilities is 1.

[0060] In some embodiments, the prediction threshold is ,and .

[0061] Specifically, the prediction threshold is set to ,and It can ensure the accuracy of the classification model, thereby reducing the computing resources and computing time of the drag coefficient prediction model and improving the accuracy of the drag coefficient prediction. If the threshold is too low, it will affect the accuracy of the specific model. For example, this application presets the threshold to 0.8, and the classification module will output a probability distribution, indicating the probability that the input point cloud data belongs to each category. If the maximum probability value is greater than 0.8, the classification result is used as the type of the point cloud; if the maximum probability value is less than 0.8, it is considered that the features of the point cloud data are too complex to be clearly classified, and it is assigned to the generalization model for processing.

[0062] It is understandable that this application will achieve a balance between classification accuracy and generalization ability by introducing a threshold of 0.8: when the classification result is relatively clear, the system will use a specific model to improve prediction accuracy; when the classification confidence is low, the system will use a generalized model to ensure prediction stability. This mechanism effectively avoids the impact of classification errors on the accuracy of drag coefficient prediction, while ensuring that the system can handle a wider range of vehicle designs, making it more reliable and adaptable in practical applications. By classifying point cloud data and matching it with a specific regression model, drag coefficient prediction is performed based on the characteristics of different vehicle models, solving the problems of poor generalization ability of traditional single models and insufficient prediction accuracy of generalized models.

[0063] Specifically, if the highest class probability for a point cloud is greater than a threshold of 0.8, the classification result is considered reliable, and the class-specific model corresponding to that class is used directly for drag coefficient prediction. A class-specific model, trained on the geometric features of that class, more accurately captures the aerodynamic characteristics of that vehicle type, thereby reducing prediction error and improving accuracy. However, in real-world applications, some vehicles may have exterior shapes that fall between different classes. For example, certain models with unique roof designs may have geometric features that exhibit both fastback and station wagon characteristics. In such cases, the classification network may struggle to make a clear judgment, resulting in similar class probabilities for all classes, with no class exceeding 0.8. If the maximum class probability is less than 0.8, the point cloud data is deemed too complex or ambiguous to be accurately classified, and the system automatically assigns it to the generalized model for processing. The generalized model exhibits strong adaptability and can make predictions for point cloud data with unclear classifications, ensuring the robustness and applicability of the entire system.

[0064] Please refer to Figure 3In some embodiments, the specific model is trained using point cloud data corresponding to the vehicle type category, and the generalized model is trained using point cloud data of all vehicle type categories.

[0065] Specifically, the sample data is randomly divided into samples for each category, with 80% of the data used for training and 20% for testing. Model 1, Model 2, and Model 3 are trained using the sample data for convertible, folding, and convertible models, respectively, resulting in three specific models. In addition to the specific models, a generalized model is required to handle point cloud data that cannot be clearly classified. When the classification module determines that the maximum classification probability of point cloud data is less than 0.8, the data is fed into the generalized model. To train the generalized model, Model 4, all samples (convertible, folding, and convertible sample data) are combined and randomly divided into 80% training data and 20% testing data. In this way, Model 4 can process data from all three models and learn a more extensive drag coefficient prediction method, thereby improving the robustness and generalization ability of the model; ultimately, the model library will contain four models: Model 1, Model 2, Model 3 and Model 4, three of which are specific models and one is a generalized model. These models work together to select the most appropriate model for drag coefficient prediction based on the classification results or generalization judgment of the point cloud data. The system uses feature recognition and can dynamically expand the specific model library according to the newly added vehicle model categories, with good adaptability and scalability.

[0066] Please refer to Figure 5 In some embodiments, each model in the drag coefficient prediction model library is based on a dynamic graph edge convolutional network, which uses the KNN algorithm to calculate the distance between the input point cloud data and its nearest neighbor points and construct an adjacency graph to capture the local features of the point cloud data; the relative position vector of each point is calculated through the edge convolution EdgeConv operation, and the local features are transformed and updated in combination with the multi-layer perceptron MLP to extract the global features of the vehicle point cloud; the local features and the global features are converged, and the drag coefficient prediction value is output through the fully connected layer.

[0067] Specifically, after classification is completed, the point cloud data will be further sent to the corresponding dedicated model in the model library for processing. These dedicated models use the graph convolutional neural network method to extract local and global features of point cloud data. In the graph convolutional network, the KNN (K-Nearest Neighbors) algorithm is first used to find the k nearest neighbors for each point. This process first involves calculating the distance between points, usually using Euclidean distance as the metric. The feature of each point is set to , where N is the number of points in the point cloud. The features of a point can be its spatial coordinates or other additional descriptive information. The Euclidean distance is calculated as follows: , where for point and The Euclidean distance between two points is often used in neighbor search for point cloud data. For each pair of points, the model calculates the inner product between the points through matrix multiplication and combines the square sum of the points to obtain the pairwise Euclidean distance. Specifically, assuming is the matrix of point cloud data, N is the number of points, and D is the dimension (three-dimensional coordinates) of each point. The pairwise distance matrix D can be calculated as follows: , where is the Euclidean distance between the i-th and j-th points, and are the eigenvectors of point i and point j respectively.

[0068] For the calculated distance matrix D, we need to find the k nearest neighbors for each point:

[0069] , where , To store the k nearest neighbor index of point i, the constraint condition Make sure that it does not select itself; the optimization goal ensures that the selected The distance between the points is the smallest. This method calculates the pairwise distances of all points and selects each point from the distance matrix. nearest neighbor points, get the neighbor's index .

[0070] Once the k nearest neighbors of each point are found, the next step is to calculate the relative position vector and construct the graph feature matrix. It is a pointer Relative to point The offset is calculated as follows:

[0071] , where is the relative position vector of each point, is the i-th point, is the jth point.

[0072] In this way, the relative position features between each point and its neighbors are extracted, and the graph feature matrix G is formed by concatenating the features of each point and the relative position features of its k neighbors. Assuming that the feature dimension of each point is D, the shape of the graph feature matrix of each point and its neighbors is ,in Includes the characteristics of each point and its relative position to its neighboring points.

[0073] Graph Convolution (GCN) is a convolution operation based on graph-structured data. It extracts local and global features by aggregating node neighbor information. Edge Convolution (EdgeConv) is used to update the features of each node using the information of its neighbors. Each layer of graph convolution uses the relative position vector and the original point features to calculate the convolution. Specifically, the EdgeConv operation can be expressed as:

[0074] First, the above KNN algorithm is used to find the Nearest neighbor, and establish local adjacency; for each point i and its neighbor j, calculate the relative position vector .

[0075] Afterwards, the density weight mechanism is introduced, combining the point's own characteristics , construct edge features:

[0076] , where is the density weight coefficient, for point and point The Euclidean distance calculated by KNN is used as the weight to reduce the waste of computing power. The density weight coefficient is introduced to improve the model's attention to edge-sensitive parts such as rearview mirrors, door handles, and wheels:

[0077] Use MLP for feature transformation:

[0078] Where, is the convolution feature of point i, is the neighbor set of point i, A point-based and neighbors The feature transformation function is typically implemented by a multi-layer perceptron (MLP). Within each convolution layer, multiple EdgeConv operations continuously combine local geometric information with global information, extracting more abstract and rich features. Following the convolution layer, the model employs pooling to enhance the representation of global information. Common pooling operations include max pooling and average pooling. Pooling allows the model to aggregate features at a global level, achieving a more comprehensive representation.

[0079] The features after max and average pooling are fed into multiple fully connected layers for processing, gradually reducing the feature dimensions to ultimately produce the model output. In regression tasks, the final output is typically a scalar, such as a predicted value for the drag coefficient. Through end-to-end training, the model continuously optimizes parameters during training, ensuring that the final prediction is as close to the true value as possible.

[0080] For regression tasks, the network's final output is the drag coefficient, a scalar quantity. The network optimizes the output using a regression loss function, such as the mean squared error (MSE), ultimately achieving the best prediction result through training.

[0081] Specifically, the regression loss function can be expressed as:

[0082] , where is the predicted value of the i-th sample, is the true value, and N is the number of samples.

[0083] In summary, the PointNet network, after classification, combines graph convolution operations to predict the drag coefficient of point cloud data. The KNN algorithm calculates the nearest neighbors of each point in the point cloud data, and graph convolution operations extract local and global features. The model gradually learns the spatial structure of the point cloud and ultimately outputs the prediction results through the regression layer. This method effectively captures the geometric characteristics of the point cloud when processing point cloud data, providing accurate predictions for regression tasks.

[0084] As you can understand, the prediction model is based on a dynamic graph edge convolutional network (DGCNN) (EdgeConv). It uses the KNN (K-nearest neighbor) algorithm to calculate the distance between the input point cloud data and its nearest neighbors, further constructing an adjacency graph. Next, graph edge convolutions propagate information between neighboring points, updating the feature representation of each point. The dynamic graph edge convolution layer effectively learns the local geometric structure of the point cloud data, enabling the model to better capture the relationships between different points in the point cloud. Next, features are aggregated using max pooling and average pooling. These pooling operations help the model extract global information and reduce data dimensionality to a certain extent, improving computational efficiency and stability. Max pooling typically retains the most significant features in the input data, while average pooling helps capture overall features. Finally, the pooled features are processed by a fully connected layer to generate the final drag coefficient prediction. By learning the global features of the point cloud data, the fully connected layer fuses multiple layers of information and outputs a regression value, which is the predicted drag coefficient. The entire process is trained through back-propagation and gradient descent optimization algorithms to minimize the error between the predicted value and the true drag coefficient.

[0085] In some embodiments, the step of aggregating the local features and the global features includes: aggregating the local features and the global features by maximum pooling and / or average pooling.

[0086] Please refer to Figure 6 On the other hand, the present invention also provides an integrated prediction system for the drag coefficient of a whole vehicle point cloud based on feature recognition, comprising: a preprocessing module for acquiring point cloud data of a whole vehicle model and preprocessing the point cloud data; a classification module for constructing a classification network model, inputting the preprocessed point cloud data into the classification network model for vehicle type classification, and outputting a vehicle type classification label, wherein the vehicle type classification label includes a vehicle type category and a probability value corresponding to the vehicle type category; a prediction module for constructing a drag coefficient prediction model library, wherein the drag coefficient prediction model library includes several specific models corresponding one-to-one to the vehicle type categories and a generalized model; a matching module for judging whether the maximum probability value in the vehicle type classification label is higher than a preset threshold; if so, selecting the specific model corresponding to the vehicle type category for drag coefficient prediction; if not, selecting the generalized model for drag coefficient prediction.

[0087] The 4,000 data samples used in the present invention include 34% fastback vehicle data samples, 36% station wagon vehicle data samples, and 30% convertible vehicle data samples. In the process of constructing the model library, first, the data of each category are randomly divided into 80% training data and 20% test data, and the sample data of fastback, station wagon, and convertible are used to train Model 1, Model 2, and Model 3 respectively; then all 4,000 data samples are randomly divided into 80% training data and 20% test data to train the generalized model Model 4. The model library contains a total of 3 specific models, Model 1, Model 2, Model 3, and Model 4, and one generalized model.

[0088] Furthermore, given a point cloud data input, the Spatial Transformer Network (STN3d) performs a global spatial transformation on the entire point cloud data. The Spatial Transformer Network progressively extracts point cloud features through three 1D convolutional layers and performs maximum pooling to extract global features. Three fully connected layers further process these features and output a 9-dimensional transformation matrix (3x3). This matrix is ​​used to align the point cloud data so that the input point cloud data can be processed in the same reference frame. The point cloud data processed by the transformation matrix enters the PointNet encoder, which first extracts local features through a convolutional layer (conv1, conv2, conv3). It then uses maximum pooling to extract global features, ultimately outputting a feature vector. For each point cloud data input, the PointNet network outputs a feature vector representing the probability distribution of the data in each category. The category corresponding to the highest probability is the network's classification result. If the highest probability does not exceed 0.8, the data is classified as a generalized model category.

[0089] After point cloud data is classified by the PointNet network, it is input into the corresponding model in the model library for further processing. First, the model uses the KNN algorithm to calculate the k (k = 20 in the figure) nearest neighbors of each point. To find the nearest neighbors for each point, the model calculates the pairwise distances between points using transposition and matrix multiplication. Specifically, the Euclidean distance between each pair of points is calculated using the inner product and the sum of the squares of the points, ultimately generating a pairwise distance matrix. Based on this distance matrix, the model selects the k points with the smallest distances and returns the indices of each point's k nearest neighbors. Using these nearest neighbors, the model calculates relative position vectors and constructs a graph feature matrix, forming a local graph feature. Next, the graph convolution (EdgeConv) operation extracts local and global features through multiple convolutional layers, continuously learning the spatial structure of the point cloud. In each convolutional layer, the original point features and the relative position features of neighboring points are combined to calculate the convolution. In this way, the network effectively captures the geometric characteristics of the point cloud data and gradually refines higher-order features through multiple convolutional layers. After the graph convolution layer, the model uses max and average pooling to globally aggregate features at each layer, further enhancing the representation of global information. These features are then processed through multiple fully connected layers, gradually reducing the feature dimension, and finally outputting the prediction results through the regression layer. In regression tasks, the final output is typically a scalar, such as the predicted value of the drag coefficient. The entire model uses an end-to-end training method to optimize network parameters, enabling accurate prediction of target values ​​given point cloud data.

[0090] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0091] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synch li nk) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory 10 bus dynamic RAM (RDRAM), etc.

[0092] The above are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's description and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for integrated prediction of vehicle point cloud drag coefficient based on feature recognition, characterized in that: include: Obtaining point cloud data of the entire vehicle model and preprocessing the point cloud data; Constructing a classification network model, inputting the preprocessed point cloud data into the classification network model to perform vehicle classification, and outputting a vehicle classification label, wherein the vehicle classification label includes a vehicle category and a probability value corresponding to the vehicle category; Constructing a drag coefficient prediction model library, wherein the drag coefficient prediction model library includes a plurality of specific models corresponding to the vehicle type categories and a generalized model; Determine whether the maximum probability value in the vehicle type classification label is higher than a preset threshold. If so, select a specific model corresponding to the vehicle type category to predict the drag coefficient. If not, select a generalized model to predict the drag coefficient.

2. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 1 is characterized in that: The pre-processing steps specifically include: performing regularization processing on the point cloud data; Performing enhancement processing on the point cloud data after regularization processing; The enhanced point cloud data is subjected to maximum and minimum normalization processing.

3. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 1 is characterized in that: The steps of constructing a classification network model, inputting the pre-processed point cloud data into the classification network model for vehicle type classification, and outputting a vehicle type classification label specifically include: A PointNet classification network model is constructed. The point cloud data is processed through several MLP layers with the same structure, and a global feature vector is generated through a maximum pooling layer. Feature extraction and classification are then performed through a fully connected layer. Finally, each vehicle type category and the corresponding probability value are output through a softmax layer.

4. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 3 is characterized in that: The step of constructing the PointNet classification network model also includes: A spatial transformation network model is constructed to perform global spatial transformation on the point cloud data.

5. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 1 is characterized in that: The vehicle type classification labels include: Fastback models and their corresponding probability values ; Travel backpack models and their corresponding probability values ; Folding back models and their corresponding probability values ; in, ,and .

6. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 1 is characterized in that: The prediction threshold is ,and .

7. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 1 is characterized in that: The specific model is trained using point cloud data corresponding to the vehicle type category, and the generalized model is trained using point cloud data of all vehicle type categories.

8. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 1 is characterized in that: Each model in the drag coefficient prediction model library is based on a dynamic graph edge convolutional network, which uses the KNN algorithm to calculate the distance between the input point cloud data and its nearest neighbor points and construct an adjacency graph to capture the local features of the point cloud data; The relative position vector of each point is calculated through the edge convolution operation, and the local features are transformed and updated in combination with the multi-layer perceptron MLP to extract the global features of the vehicle point cloud; The local features and the global features are aggregated, and a drag coefficient prediction value is output through a fully connected layer.

9. The integrated prediction method for vehicle point cloud drag coefficient based on feature recognition according to claim 8 is characterized in that: The step of aggregating the local features and the global features comprises: The local features and the global features are aggregated through maximum pooling and / or average pooling.

10. A vehicle point cloud drag coefficient integrated prediction system based on feature recognition, characterized in that: include: A preprocessing module, used to obtain point cloud data of the entire vehicle model and preprocess the point cloud data; A classification module is used to construct a classification network model, input the preprocessed point cloud data into the classification network model to perform vehicle classification, and output a vehicle classification label, wherein the vehicle classification label includes a vehicle category and a probability value corresponding to the vehicle category; A prediction module, configured to construct a drag coefficient prediction model library, wherein the drag coefficient prediction model library includes a plurality of specific models corresponding to the vehicle type categories and a generalized model; The matching module is used to determine whether the maximum probability value in the vehicle type classification label is higher than a preset threshold. If so, a specific model corresponding to the vehicle type category is selected to predict the drag coefficient. If not, a generalized model is selected to predict the drag coefficient.

Images