Adams pavement modeling method and system based on computer vision and deep reinforcement learning algorithm

Abstract

The invention provides an Adams road surface modeling method and system based on computer vision and a deep reinforcement learning algorithm, and the method comprises the steps: carrying out the image segmentation and preprocessing of an original road image, and obtaining a preprocessed road image; performing road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data, and performing sorting processing on road coordinates through a deep reinforcement learning algorithm based on an attention mechanism to generate coordinate information; and importing the coordinate information into Adams software to construct a road surface model so as to obtain a corresponding road surface model, and placing the whole vehicle model on the road surface model to carry out multi-body dynamics simulation. Various simulations are completed based on real road images, road key information extraction and path planning are automatically carried out, physical sample vehicles and field sampling are not needed, and the whole vehicle test cost and period are reduced. Road key information can be automatically extracted, directed sorting is performed by using a deep reinforcement learning algorithm, and the accuracy of a pavement model and the reliability of a simulation environment are improved.

Description

Technical Field

[0001] The present invention relates to the field of automobile test modeling, and in particular to an Adams road surface modeling method and system based on computer vision and deep reinforcement learning algorithms. Background Art

[0002] In the automotive field, the actual vehicle test after the prototype is produced is an important basis for improving the vehicle structure and achieving lightweight. In the whole vehicle development process based on virtual prototype technology, digital prototypes are used instead of physical prototype tests, which not only reduces the physical iteration cost, but also helps to shorten the development cycle and reduce development risks. Adams software is a whole vehicle dynamics simulation software with a wide range of uses, but the road modeling method that comes with Adams software has the following problems: First, although Adams software comes with road files, the roads established by these road files are very simple and regular, and cannot objectively reflect the real road conditions, thus failing to provide a high-quality and high-reliability simulation environment for vehicle performance simulation; Second, if a road with complex graphics is constructed through Adams software, a large amount of data needs to be manually input, which is extremely cumbersome and requires a lot of manual work. Summary of the invention

[0003] The purpose of the present invention is to provide an Adams road surface modeling method and system based on computer vision and deep reinforcement learning algorithm, which performs image segmentation and preprocessing on the original road image to obtain a preprocessed road image; performs road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data, and sorts the road coordinates through a deep reinforcement learning algorithm based on an attention mechanism to generate coordinate information; then imports the coordinate information into Adams software to construct a road surface model to obtain a corresponding road surface model, and places a whole vehicle model on the road surface model for multi-body dynamics simulation, which extracts data and completes modeling through real road images, cooperates with Adams software to complete various simulations, automatically extracts key road information and plans paths, does not require physical prototypes and on-site sampling, reduces the cost and cycle of whole vehicle testing, accurately restores the real track based on the analysis and processing of real road images, and expands the scope of software application; automatically extracts key road information and uses a deep reinforcement learning algorithm based on an attention mechanism for directed sorting, improves the accuracy of the road surface model and the reliability of the simulation environment, can more accurately evaluate and verify vehicle performance, and improves the reliability of the entire simulation experiment.

[0004] The present invention is achieved through the following technical solutions:

[0005] Adams pavement modeling method based on computer vision and deep reinforcement learning algorithm, including:

[0006] The target road image is preprocessed to obtain a preprocessed road image;

[0007] Performing road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data; performing a sorting process on the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information;

[0008] The coordinate information is imported into the Adams software to construct a road surface model to obtain a corresponding road surface model; a whole vehicle model is also constructed through the Adams software, and the whole vehicle model is placed on the road surface model for multi-body dynamics simulation.

[0009] Optionally, performing image segmentation on the original road image to obtain a target road image; and preprocessing the target road image to obtain a preprocessed road image includes:

[0010] Taking panoramic photos of a real road to obtain an original road image of the real road; performing annotation recognition on the original road image to obtain distribution position information of all annotations on the original road image, and performing image segmentation on the original road image based on the distribution position information to obtain an unannotated target road image;

[0011] The target road image is subjected to machine learning processing for reducing invalid data to obtain a deep learning road image; wherein the machine learning processing includes grayscale processing, Gaussian pyramid processing or binarization processing; and the deep learning road image is subjected to picture filling preprocessing to obtain a preprocessed road image; wherein the preprocessed road image includes a single road contour.

[0012] Optionally, before performing machine learning processing on the target road image, contrast adjustment is performed on the target road image, including:

[0013] extracting the target road image;

[0014] Scanning each pixel point of the target road image, extracting edge pixel points of a corresponding area of ​​the target road image, and obtaining grayscale values ​​of the edge pixel points of the corresponding area of ​​the target road image;

[0015] Extracting the grayscale values ​​corresponding to the non-edge pixels in the target road image except the edge pixels;

[0016] Obtaining a first contrast adjustment coefficient using grayscale values ​​corresponding to non-edge pixels in the target road image;

[0017] The first contrast adjustment coefficient is obtained by the following formula:

[0018]

[0019] Among them, J 01 represents the first contrast adjustment coefficient; n represents the number of non-edge pixels; H i Represents the gray value corresponding to the i-th non-edge pixel; H z Represents the central gray value of the target road image; H bmax and H bmin Indicates the maximum and minimum grayscale values ​​of edge pixels;

[0020] Extracting pixel points inside the non-target road image area connected to the edge pixel points as reference pixel points, and obtaining grayscale values ​​of the reference pixel points;

[0021] Obtaining a second contrast adjustment coefficient using the grayscale value of the reference pixel point combined with the grayscale value of the edge pixel point;

[0022] The second contrast adjustment coefficient is obtained by the following formula:

[0023]

[0024] Among them, J 02 represents the second contrast adjustment coefficient; k represents the number of reference pixels; m represents the number of edge pixels; H ki Represents the gray value corresponding to the i-th reference pixel; H z Represents the central gray value of the target road image; H bmax Indicates the maximum grayscale value of edge pixels; H cmax Indicates the maximum grayscale value in the reference pixel; H bi Represents the gray value corresponding to the i-th edge pixel;

[0025] The contrast in the target road image is adjusted using the first contrast adjustment coefficient and the second contrast adjustment coefficient. The adjusted contrast is obtained by the following formula:

[0026]

[0027] Among them, L t represents the contrast after adjustment; L0 represents the contrast before adjustment; J 02 represents the second contrast adjustment coefficient; J 01 Represents the first contrast adjustment coefficient.

[0028] Optionally, road contour recognition is performed on the preprocessed road image to obtain road contour pixel coordinate data; and the road contour pixel coordinate data is sorted by a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information, including:

[0029] Performing trajectory recognition on a single road contour contained in the preprocessed road image to obtain trajectory information of the single road contour on the preprocessed road image screen; performing road contour pixel point extraction processing on the preprocessed road image based on the trajectory information to obtain position data of all road contour pixel coordinate points;

[0030] The position data of all road contour pixel coordinate points are sorted using a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are sorted in a directed manner to obtain the corresponding coordinate information.

[0031] Optionally, the coordinate information is imported into Adams software to construct a road surface model to obtain a corresponding road surface model; a whole vehicle model is also constructed by the Adams software, and the whole vehicle model is placed on the road surface model for multi-body dynamics simulation, including:

[0032] Importing the coordinate information into the road builder module of the Adams software to construct a road model, obtaining a corresponding road model, and uploading the road model to a road model library after marking it;

[0033] The whole vehicle model is constructed through the car module of the Adams software. Based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation.

[0034] Adams pavement modeling system based on computer vision and deep reinforcement learning algorithms, including:

[0035] An image segmentation module is used to segment the original road image to obtain a target road image;

[0036] An image preprocessing module, used for preprocessing the target road image to obtain a preprocessed road image;

[0037] A road contour recognition module, used to perform road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data;

[0038] A road coordinate sorting module is used to sort the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information;

[0039] A pavement model building module is used to import the coordinate information into Adams software to build a pavement model and obtain a corresponding pavement model;

[0040] The dynamics simulation module is used to construct a whole vehicle model through the Adams software, and place the whole vehicle model on the road surface model to perform multi-body dynamics simulation.

[0041] Optionally, before performing machine learning processing on the target road image, contrast adjustment is performed on the target road image, including:

[0042] extracting the target road image;

[0043] Scanning each pixel point of the target road image, extracting edge pixel points of a corresponding area of ​​the target road image, and obtaining grayscale values ​​of the edge pixel points of the corresponding area of ​​the target road image;

[0044] Extracting the grayscale values ​​corresponding to the non-edge pixels in the target road image except the edge pixels;

[0045] Obtaining a first contrast adjustment coefficient using grayscale values ​​corresponding to non-edge pixels in the target road image;

[0046] The first contrast adjustment coefficient is obtained by the following formula:

[0047]

[0048] Among them, J 01 represents the first contrast adjustment coefficient; n represents the number of non-edge pixels; H i Represents the gray value corresponding to the i-th non-edge pixel; H z Represents the central gray value of the target road image; H bmax and H bmin Indicates the maximum and minimum grayscale values ​​of edge pixels;

[0049] Extracting pixel points inside the non-target road image area connected to the edge pixel points as reference pixel points, and obtaining grayscale values ​​of the reference pixel points;

[0050] Obtaining a second contrast adjustment coefficient using the grayscale value of the reference pixel point combined with the grayscale value of the edge pixel point;

[0051] The second contrast adjustment coefficient is obtained by the following formula:

[0052]

[0053] Among them, J 02 represents the second contrast adjustment coefficient; k represents the number of reference pixels; m represents the number of edge pixels; H ki Represents the gray value corresponding to the i-th reference pixel; H zRepresents the central gray value of the target road image; H bmax Indicates the maximum grayscale value of edge pixels; H cmax Indicates the maximum grayscale value in the reference pixel; H bi Represents the gray value corresponding to the i-th edge pixel;

[0054] The contrast in the target road image is adjusted using the first contrast adjustment coefficient and the second contrast adjustment coefficient. The adjusted contrast is obtained by the following formula:

[0055]

[0056] Among them, L t represents the contrast after adjustment; L0 represents the contrast before adjustment; J 02 represents the second contrast adjustment coefficient; J 01 Represents the first contrast adjustment coefficient.

[0057] Optionally, the image segmentation module is used to perform image segmentation on the original road image to obtain the target road image, including:

[0058] Taking panoramic photos of a real road to obtain an original road image of the real road; performing annotation recognition on the original road image to obtain distribution position information of all annotations on the original road image, and performing image segmentation on the original road image based on the distribution position information to obtain an unannotated target road image;

[0059] The image preprocessing module is used to preprocess the target road image to obtain a preprocessed road image, including:

[0060] The target road image is subjected to machine learning processing for reducing invalid data to obtain a deep learning road image; wherein the machine learning processing includes grayscale processing, Gaussian pyramid processing or binarization processing; and the deep learning road image is subjected to picture filling preprocessing to obtain a preprocessed road image; wherein the preprocessed road image includes a single road contour.

[0061] Optionally, the road contour recognition module is used to perform road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data, including:

[0062] Performing trajectory recognition on a single road contour contained in the preprocessed road image to obtain trajectory information of the single road contour on the preprocessed road image screen; performing road contour pixel point extraction processing on the preprocessed road image based on the trajectory information to obtain position data of all road contour pixel coordinate points;

[0063] The road coordinate sorting module is used to perform a sorting process on the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information, including:

[0064] The position data of all road contour pixel coordinate points are sorted using a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are sorted in a directed manner to obtain the corresponding coordinate information.

[0065] Optionally, the pavement model building module is used to import the coordinate information into Adams software to build a pavement model to obtain a corresponding pavement model, including:

[0066] Importing the coordinate information into the road builder module of the Adams software to construct a road model, obtaining a corresponding road model, and uploading the road model to a road model library after marking it;

[0067] The dynamics simulation module is used to construct a whole vehicle model through the Adams software, and place the whole vehicle model on the road surface model to perform multi-body dynamics simulation, including:

[0068] The whole vehicle model is constructed through the car module of the Adams software. Based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] The Adams road modeling method and system based on computer vision and deep reinforcement learning algorithm provided in the present application performs image segmentation and preprocessing on the original road image to obtain the preprocessed road image; performs road contour recognition on the preprocessed road image to obtain the road contour pixel coordinate data, and sorts the road coordinates through the deep reinforcement learning algorithm based on the attention mechanism to generate coordinate information; then imports the coordinate information into the Adams software to construct the road model to obtain the corresponding road model, and places the whole vehicle model on the road model for multi-body dynamics simulation, which extracts data and completes modeling through the real road image, cooperates with the Adams software to complete various simulations, automatically extracts key road information and plans the path, does not require physical sample vehicles and on-site sampling, reduces the cost and cycle of vehicle testing, accurately restores the real track based on the analysis and processing of real road images, and expands the scope of software application; automatically extracts key road information and uses the deep reinforcement learning algorithm based on the attention mechanism for directed sorting, improves the accuracy of the road model and the reliability of the simulation environment, can more accurately evaluate and verify vehicle performance, and improve the reliability of the entire simulation experiment. . BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0072] Figure 1 A schematic flow chart of the Adams pavement modeling method based on computer vision and deep reinforcement learning algorithm provided by the present invention.

[0073] Figure 2 A schematic diagram of the structure of the Adams pavement modeling system based on computer vision and deep reinforcement learning algorithm provided by the present invention. DETAILED DESCRIPTION

[0074] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. It is to be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some structures related to the present application are shown in the accompanying drawings, rather than all structures. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0075] The terms "include" and "have" and any variations thereof in this application are intended to cover non-exclusive inclusions. For example, a process, method, method, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0076] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0077] See also Figure 1 As shown, an Adams pavement modeling method based on computer vision and deep reinforcement learning algorithm is provided in one embodiment of the present application. The Adams pavement modeling method based on computer vision and deep reinforcement learning algorithm includes:

[0078] Performing image segmentation on the original road image to obtain a target road image; performing preprocessing on the target road image to obtain a preprocessed road image;

[0079] Performing road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data; performing sorting processing on the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information;

[0080] The coordinate information is imported into the Adams software to construct a road surface model to obtain a corresponding road surface model; a whole vehicle model is also constructed through the Adams software, and the whole vehicle model is placed on the road surface model for multi-body dynamics simulation.

[0081] The beneficial effects of the above embodiments are as follows: the Adams road modeling method based on computer vision and deep reinforcement learning algorithm performs image segmentation and preprocessing on the original road image to obtain a preprocessed road image; performs road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data, and sorts the road coordinates through a deep reinforcement learning algorithm based on an attention mechanism to generate coordinate information; then the coordinate information is imported into the Adams software to construct a road model to obtain a corresponding road model, and the whole vehicle model is placed on the road model for multi-body dynamics simulation, which extracts data and completes modeling through real road images, cooperates with Adams software to complete various simulations, automatically extracts key road information and plans paths, does not require physical prototypes and on-site sampling, reduces the cost and cycle of vehicle testing, accurately restores the real track based on the analysis and processing of real road images, and expands the scope of software application; automatically extracts key road information and uses a deep reinforcement learning algorithm based on an attention mechanism for directed sorting, improves the accuracy of the road model and the reliability of the simulation environment, can more accurately evaluate and verify vehicle performance, and improves the reliability of the entire simulation experiment.

[0082] In another embodiment, performing image segmentation on the original road image to obtain a target road image; and preprocessing the target road image to obtain a preprocessed road image includes:

[0083] Taking panoramic photos of a real road to obtain an original road image of the real road; performing annotation recognition on the original road image to obtain distribution position information of all annotations on the original road image; performing image segmentation on the original road image based on the distribution position information to obtain an unannotated target road image;

[0084] The target road image is subjected to machine learning processing for reducing invalid data to obtain a deep learning road image; wherein the machine learning processing includes grayscale processing, Gaussian pyramid processing or binarization processing; and the deep learning road image is subjected to picture filling preprocessing to obtain a preprocessed road image; wherein the preprocessed road image includes a single road contour.

[0085] The beneficial effects of the above embodiments are as follows: in order to make the road surface model constructed by the Adams software more compatible with the actual road surface conditions and have a rich and diverse road surface structure, the actual road is first photographed in a panoramic manner to obtain the corresponding original road image, which comprehensively and truly records the road surface contour shape of the actual road, and can provide good original data for the Adams software to construct the road surface model. The original road image obtained by shooting will have annotation information (such as road traffic signs, etc.) that is irrelevant to the road surface contour shape, and these annotation information will interfere with the construction of the road surface model by the Adams software. For this reason, the original road image is annotated and identified to obtain the distribution position information of all annotations existing on the global screen of the original road image, so that the original road image is segmented to remove the annotation information on the screen, so as to obtain the target road image without annotations, which can reduce the image interference information component of the road image. In addition, the original road image obtained by shooting has a large amount of data, and all the data information contained therein is not necessarily the information required by the Adams software to build a road model. If the original road image is directly input into the Adams software for processing, the data processing workload of the Adams software will increase. For this reason, the target road image is subjected to machine learning processing such as grayscale processing, Gaussian pyramid processing or binarization processing to reduce invalid data, and obtain a deep learning road image. In this way, while retaining important image information, the storage space occupied by invalid data that does not contribute to the construction of the road model by the Adams software is reduced and the interference items therein are reduced, so as to facilitate the subsequent rapid acquisition of the road contour under the premise of ensuring the accuracy of the road. In addition, the deep learning road image is preprocessed with picture filling to obtain a preprocessed road image, so that the preprocessed road image includes a single road contour, so that the road contour of the image screen can be uniquely identified to avoid the subsequent road contour recognition errors. The picture filling preprocessing belongs to the conventional technical means in this field and is not described in detail here.

[0086] In another embodiment, before performing machine learning processing on the target road image, contrast adjustment is performed on the target road image, including:

[0087] extracting the target road image;

[0088] Scanning each pixel point of the target road image, extracting edge pixel points of a corresponding area of ​​the target road image, and obtaining grayscale values ​​of the edge pixel points of the corresponding area of ​​the target road image;

[0089] Extracting the grayscale values ​​corresponding to the non-edge pixels in the target road image except the edge pixels;

[0090] Obtaining a first contrast adjustment coefficient using grayscale values ​​corresponding to non-edge pixels in the target road image;

[0091] The first contrast adjustment coefficient is obtained by the following formula:

[0092]

[0093] Among them, J 01 represents the first contrast adjustment coefficient; n represents the number of non-edge pixels; H i Represents the gray value corresponding to the i-th non-edge pixel; H z Represents the central gray value of the target road image; H bmax and H bmin Indicates the maximum and minimum grayscale values ​​of edge pixels;

[0094] Extracting pixel points inside the non-target road image area connected to the edge pixel points as reference pixel points, and obtaining grayscale values ​​of the reference pixel points;

[0095] Obtaining a second contrast adjustment coefficient using the grayscale value of the reference pixel point combined with the grayscale value of the edge pixel point;

[0096] The second contrast adjustment coefficient is obtained by the following formula:

[0097]

[0098] Among them, J 02 represents the second contrast adjustment coefficient; k represents the number of reference pixels; m represents the number of edge pixels; H ki Represents the gray value corresponding to the i-th reference pixel; H z Represents the central gray value of the target road image; H bmax Indicates the maximum grayscale value of edge pixels; H cmax Indicates the maximum grayscale value in the reference pixel; H bi Represents the gray value corresponding to the i-th edge pixel;

[0099] The contrast in the target road image is adjusted using the first contrast adjustment coefficient and the second contrast adjustment coefficient. The adjusted contrast is obtained by the following formula:

[0100]

[0101] Among them, L t represents the contrast after adjustment; L0 represents the contrast before adjustment; J 02 represents the second contrast adjustment coefficient; J 01 Represents the first contrast adjustment coefficient.

[0102] The beneficial effect of the above embodiment is that by extracting edge pixels and non-edge pixels of the target road image and calculating their grayscale values ​​respectively, the contrast of the image can be adjusted more specifically. This method not only considers the overall grayscale distribution of the image, but also pays special attention to the grayscale difference between the edge area and the non-edge area, thereby more effectively enhancing the contrast of the image. 01 ), which is calculated based on the grayscale value of non-edge pixels, the central grayscale value of the target road image, and the maximum and minimum grayscale values ​​of edge pixels. This helps to optimize the detail performance of non-edge areas in the image, making road textures, sign lines, etc. clearer. The second contrast adjustment coefficient (J 02 ) combines the grayscale values ​​of edge pixels and reference pixels, and by comparing the grayscale differences of these pixels, the edge sharpness of the image can be further adjusted. This helps to improve the recognition accuracy of key information such as road edges and vehicle outlines in the image. This technical solution can automatically adjust the contrast according to different road image features, so it has strong adaptability. Regardless of whether it is a bright or dim road environment, this solution can provide a relatively consistent contrast enhancement effect. Contrast adjustment is an important step in image preprocessing, and its effect directly affects the subsequent machine learning processing results. Contrast adjustment through this technical solution can significantly improve the accuracy and efficiency of subsequent machine learning tasks such as target detection and road recognition.

[0103] In summary, the technical effects of this technical solution in terms of performance indicators are mainly reflected in enhancing image contrast, optimizing detail performance, improving edge sharpness, strong adaptability, and improving subsequent processing effects. These effects jointly improve the quality of the target road image and provide a more reliable data foundation for subsequent machine learning processing.

[0104] In another embodiment, the preprocessed road image is subjected to road contour recognition to obtain road contour pixel coordinate data; the road contour pixel coordinate data is subjected to a sorting process using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information, including:

[0105] Performing trajectory recognition on a single road contour included in the preprocessed road image to obtain trajectory information of the single road contour on the preprocessed road image screen; performing road contour pixel point extraction processing on the preprocessed road image based on the trajectory information to obtain position data of all road contour pixel coordinate points;

[0106] The position data of all road contour pixel coordinate points are sorted using a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are sorted in a directed manner to obtain the corresponding coordinate information.

[0107] The beneficial effects of the above embodiments are that the road contour on the road image directly reflects the direction of the road and the road surface structure state. In order to provide comprehensive data for building a road surface model, the trajectory of the single road contour contained in the pre-processed road image is identified to obtain the trajectory information of the single road contour on the pre-processed road image screen, and the road contour pixel point extraction processing is performed on the pre-processed road image to obtain the position data of all road contour pixel coordinate points, so that the position data can fully characterize the physical structure state of the road. The position data of all road contour pixel coordinate points are also sorted by a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are directed sorted to obtain the corresponding coordinate information, which can ensure that the discrete position points in the coordinate information can present a regular arrangement, improve the orderliness of the coordinate information, and facilitate the subsequent Adams software to directly use the coordinate information for model construction.

[0108] In another embodiment, the coordinate information is imported into Adams software to construct a road surface model to obtain a corresponding road surface model; a vehicle model is also constructed by the Adams software, and the vehicle model is placed on the road surface model for multi-body dynamics simulation, including:

[0109] The coordinate information is imported into the road builder module of the Adams software to construct a road model, and the corresponding road model is obtained. The road model is marked and uploaded to the road model library;

[0110] The whole vehicle model is constructed through the car module of the Adams software. Based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation.

[0111] The beneficial effect of the above embodiment is that the coordinate information is imported into the road builder module of the Adams software to build a road model, so that the Adams software can directly fit the coordinate information through its own road builder module to generate a road model that is highly matched with the real road, and the road model is also marked and uploaded to the road model library, which can enrich the model types of the road model library and provide a variety of road model options for subsequent vehicle test simulations. In addition, the whole vehicle model is built through the car module of the Adams software, and based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation, which can more accurately evaluate and verify vehicle performance and improve the reliability of the entire simulation experiment.

[0112] See also Figure 2 As shown, an Adams road surface modeling system based on computer vision and deep reinforcement learning algorithm is provided in one embodiment of the present application. The Adams road surface modeling system based on computer vision and deep reinforcement learning algorithm includes:

[0113] An image segmentation module is used to segment the original road image to obtain a target road image;

[0114] An image preprocessing module is used to preprocess the target road image to obtain a preprocessed road image;

[0115] A road contour recognition module is used to perform road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data;

[0116] A road coordinate sorting module is used to sort the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information;

[0117] The pavement model building module is used to import the coordinate information into the Adams software to build the pavement model and obtain the corresponding pavement model;

[0118] The dynamics simulation module is used to construct a whole vehicle model through the Adams software, and place the whole vehicle model on the road surface model for multi-body dynamics simulation.

[0119] The beneficial effects of the above embodiments are as follows: the Adams road modeling system based on computer vision and deep reinforcement learning algorithm performs image segmentation and preprocessing on the original road image to obtain a preprocessed road image; performs road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data, and sorts the road coordinates through a deep reinforcement learning algorithm based on an attention mechanism to generate coordinate information; then the coordinate information is imported into the Adams software to construct a road model to obtain a corresponding road model, and the whole vehicle model is placed on the road model for multi-body dynamics simulation, which extracts data and completes modeling through real road images, cooperates with Adams software to complete various simulations, automatically extracts key road information and plans paths, does not require physical prototypes and on-site sampling, reduces the cost and cycle of vehicle testing, accurately restores the real track based on the analysis and processing of real road images, and expands the scope of software application; automatically extracts key road information and uses a deep reinforcement learning algorithm based on an attention mechanism for directed sorting, improves the accuracy of the road model and the reliability of the simulation environment, can more accurately evaluate and verify vehicle performance, and improves the reliability of the entire simulation experiment.

[0120] In another embodiment, a real road is photographed in a panoramic manner to obtain an original road image of the real road; annotations are identified on the original road image to obtain distribution position information of all annotations on the original road image; based on the distribution position information, the original road image is segmented to obtain an unannotated target road image;

[0121] The image preprocessing module is used to preprocess the target road image to obtain a preprocessed road image, including:

[0122] The target road image is subjected to machine learning processing for reducing invalid data to obtain a deep learning road image; wherein the machine learning processing includes grayscale processing, Gaussian pyramid processing or binarization processing; and the deep learning road image is subjected to picture filling preprocessing to obtain a preprocessed road image; wherein the preprocessed road image includes a single road contour.

[0123] The beneficial effects of the above embodiments are as follows: in order to make the road surface model constructed by the Adams software more compatible with the actual road surface conditions and have a rich and diverse road surface structure, the actual road is first photographed in a panoramic manner to obtain the corresponding original road image, which comprehensively and truly records the road surface contour shape of the actual road, and can provide good original data for the Adams software to construct the road surface model. The original road image obtained by shooting will have annotation information (such as road traffic signs, etc.) that is irrelevant to the road surface contour shape, and these annotation information will interfere with the construction of the road surface model by the Adams software. For this reason, the original road image is annotated and identified to obtain the distribution position information of all annotations existing on the global screen of the original road image, so that the original road image is segmented to remove the annotation information on the screen, so as to obtain the target road image without annotations, which can reduce the image interference information component of the road image. In addition, the original road image obtained by shooting has a large amount of data, and all the data information contained therein is not necessarily the information required by the Adams software to build a road model. If the original road image is directly input into the Adams software for processing, the data processing workload of the Adams software will increase. For this reason, the target road image is subjected to machine learning processing such as grayscale processing, Gaussian pyramid processing or binarization processing to reduce invalid data, and obtain a deep learning road image. In this way, while retaining important image information, the storage space occupied by invalid data that does not contribute to the construction of the road model by the Adams software is reduced and the interference items therein are reduced, so as to facilitate the subsequent rapid acquisition of the road contour under the premise of ensuring the accuracy of the road. In addition, the deep learning road image is preprocessed with picture filling to obtain a preprocessed road image, so that the preprocessed road image includes a single road contour, so that the road contour of the image screen can be uniquely identified to avoid the subsequent road contour recognition errors. The picture filling preprocessing belongs to the conventional technical means in this field and is not described in detail here.

[0124] In another embodiment, before performing machine learning processing on the target road image, contrast adjustment is performed on the target road image, including:

[0125] extracting the target road image;

[0126] Scanning each pixel point of the target road image, extracting edge pixel points of a corresponding area of ​​the target road image, and obtaining grayscale values ​​of the edge pixel points of the corresponding area of ​​the target road image;

[0127] Extracting the grayscale values ​​corresponding to the non-edge pixels in the target road image except the edge pixels;

[0128] Obtaining a first contrast adjustment coefficient using grayscale values ​​corresponding to non-edge pixels in the target road image;

[0129] The first contrast adjustment coefficient is obtained by the following formula:

[0130]

[0131] Among them, J 01 represents the first contrast adjustment coefficient; n represents the number of non-edge pixels; H i Represents the gray value corresponding to the i-th non-edge pixel; H z Represents the central gray value of the target road image; H bmax and H bmin Indicates the maximum and minimum grayscale values ​​of edge pixels;

[0132] Extracting pixel points inside the non-target road image area connected to the edge pixel points as reference pixel points, and obtaining grayscale values ​​of the reference pixel points;

[0133] Obtaining a second contrast adjustment coefficient using the grayscale value of the reference pixel point combined with the grayscale value of the edge pixel point;

[0134] The second contrast adjustment coefficient is obtained by the following formula:

[0135]

[0136] Among them, J 02 represents the second contrast adjustment coefficient; k represents the number of reference pixels; m represents the number of edge pixels; H ki Represents the gray value corresponding to the i-th reference pixel; H z Represents the central gray value of the target road image; H bmax Indicates the maximum grayscale value of edge pixels; H cmax Indicates the maximum grayscale value in the reference pixel; H bi Represents the gray value corresponding to the i-th edge pixel;

[0137] The contrast in the target road image is adjusted using the first contrast adjustment coefficient and the second contrast adjustment coefficient. The adjusted contrast is obtained by the following formula:

[0138]

[0139] Among them, L t represents the contrast after adjustment; L0 represents the contrast before adjustment; J 02 represents the second contrast adjustment coefficient; J 01 Represents the first contrast adjustment coefficient.

[0140] The beneficial effect of the above embodiment is that by extracting edge pixels and non-edge pixels of the target road image and calculating their grayscale values ​​respectively, the contrast of the image can be adjusted more specifically. This method not only considers the overall grayscale distribution of the image, but also pays special attention to the grayscale difference between the edge area and the non-edge area, thereby more effectively enhancing the contrast of the image. 01 ), which is calculated based on the grayscale value of non-edge pixels, the central grayscale value of the target road image, and the maximum and minimum grayscale values ​​of edge pixels. This helps to optimize the detail performance of non-edge areas in the image, making road textures, sign lines, etc. clearer. The second contrast adjustment coefficient (J 02 ) combines the grayscale values ​​of edge pixels and reference pixels, and by comparing the grayscale differences of these pixels, the edge sharpness of the image can be further adjusted. This helps to improve the recognition accuracy of key information such as road edges and vehicle outlines in the image. This technical solution can automatically adjust the contrast according to different road image features, so it has strong adaptability. Regardless of whether it is a bright or dim road environment, this solution can provide a relatively consistent contrast enhancement effect. Contrast adjustment is an important step in image preprocessing, and its effect directly affects the subsequent machine learning processing results. Contrast adjustment through this technical solution can significantly improve the accuracy and efficiency of subsequent machine learning tasks such as target detection and road recognition.

[0141] In summary, the technical effects of this technical solution in terms of performance indicators are mainly reflected in enhancing image contrast, optimizing detail performance, improving edge sharpness, strong adaptability, and improving subsequent processing effects. These effects jointly improve the quality of the target road image and provide a more reliable data foundation for subsequent machine learning processing.

[0142] In another embodiment, the road contour recognition module is used to perform road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data, including:

[0143] Performing trajectory recognition on a single road contour included in the preprocessed road image to obtain trajectory information of the single road contour on the preprocessed road image screen; performing road contour pixel point extraction processing on the preprocessed road image based on the trajectory information to obtain position data of all road contour pixel coordinate points;

[0144] The road coordinate sorting module is used to sort the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information, including:

[0145] The position data of all road contour pixel coordinate points are sorted using a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are sorted in a directed manner to obtain the corresponding coordinate information.

[0146] The beneficial effects of the above embodiments are that the road contour on the road image directly reflects the direction of the road and the road surface structure state. In order to provide comprehensive data for building a road surface model, the trajectory of the single road contour contained in the pre-processed road image is identified to obtain the trajectory information of the single road contour on the pre-processed road image screen, and the road contour pixel point extraction processing is performed on the pre-processed road image to obtain the position data of all road contour pixel coordinate points, so that the position data can fully characterize the physical structure state of the road. The position data of all road contour pixel coordinate points are also sorted by a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are directed sorted to obtain the corresponding coordinate information, which can ensure that the discrete position points in the coordinate information can present a regular arrangement, improve the orderliness of the coordinate information, and facilitate the subsequent Adams software to directly use the coordinate information for model construction.

[0147] In another embodiment, the pavement model building module is used to import the coordinate information into Adams software to build a pavement model, and obtain a corresponding pavement model, including:

[0148] The coordinate information is imported into the road builder module of the Adams software to construct a road model, and the corresponding road model is obtained. The road model is marked and uploaded to the road model library;

[0149] The dynamics simulation module is used to construct a vehicle model through the Adams software, and place the vehicle model on the road model for multi-body dynamics simulation, including:

[0150] The whole vehicle model is constructed through the car module of the Adams software. Based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation.

[0151] The beneficial effect of the above embodiment is that the coordinate information is imported into the road builder module of the Adams software to build a road model, so that the Adams software can directly fit the coordinate information through its own road builder module to generate a road model that is highly matched with the real road, and the road model is also marked and uploaded to the road model library, which can enrich the model types of the road model library and provide a variety of road model options for subsequent vehicle test simulations. In addition, the whole vehicle model is built through the car module of the Adams software, and based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation, which can more accurately evaluate and verify vehicle performance and improve the reliability of the entire simulation experiment.

[0152] In general, the Adams road modeling method and system based on computer vision and deep reinforcement learning algorithm performs image segmentation and preprocessing on the original road image to obtain a preprocessed road image; performs road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data, and sorts the road coordinates through a deep reinforcement learning algorithm based on an attention mechanism to generate coordinate information; then the coordinate information is imported into the Adams software to construct a road model to obtain a corresponding road model, and the whole vehicle model is placed on the road model for multi-body dynamics simulation. It extracts data and completes modeling through real road images, cooperates with Adams software to complete various simulations, and automatically extracts key road information and plans paths. There is no need for physical prototypes and on-site sampling, which reduces the cost and cycle of vehicle testing. Based on the analysis and processing of real road images, the real track is accurately restored, and the scope of software application is expanded. Key road information is automatically extracted and directed sorting is performed using a deep reinforcement learning algorithm based on an attention mechanism to improve the accuracy of the road model and the reliability of the simulation environment, which can more accurately evaluate and verify vehicle performance and improve the reliability of the entire simulation experiment.

[0153] The above is only a specific implementation of the present invention, and any other improvements made based on the concept of the present invention are considered to be within the protection scope of the present invention.

Claims

1. The Adams pavement modeling method based on computer vision and deep reinforcement learning algorithm is characterized by: include: Perform image segmentation on the original road image to obtain the target road image; Preprocessing the target road image to obtain a preprocessed road image; Performing road contour recognition on the preprocessed road image to obtain road contour pixel coordinate data; performing a sorting process on the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information; The coordinate information is imported into the Adams software to construct a road surface model to obtain a corresponding road surface model; a whole vehicle model is also constructed through the Adams software, and the whole vehicle model is placed on the road surface model for multi-body dynamics simulation.

2. The Adams road surface modeling method based on computer vision and deep reinforcement learning algorithm as claimed in claim 1, characterized in that: Perform image segmentation on the original road image to obtain the target road image; Preprocessing the target road image to obtain a preprocessed road image includes: Taking panoramic photos of a real road to obtain an original road image of the real road; performing annotation recognition on the original road image to obtain distribution position information of all annotations on the original road image, and performing image segmentation on the original road image based on the distribution position information to obtain an unannotated target road image; The target road image is subjected to machine learning processing for reducing invalid data to obtain a deep learning road image; wherein the machine learning processing includes grayscale processing, Gaussian pyramid processing or binarization processing; and the deep learning road image is subjected to picture filling preprocessing to obtain a preprocessed road image; wherein the preprocessed road image includes a single road contour.

3. The Adams road surface modeling method based on computer vision and deep reinforcement learning algorithm as claimed in claim 2, characterized in that: Before performing machine learning processing on the target road image, the contrast of the target road image is adjusted, including: extracting the target road image; Scanning each pixel point of the target road image, extracting edge pixel points of a corresponding area of ​​the target road image, and obtaining grayscale values ​​of the edge pixel points of the corresponding area of ​​the target road image; Extract the grayscale values ​​corresponding to the non-edge pixels in the target road image except the edge pixels; Obtaining a first contrast adjustment coefficient using grayscale values ​​corresponding to non-edge pixels in the target road image; The first contrast adjustment coefficient is obtained by the following formula: Among them, J 01 represents the first contrast adjustment coefficient; n represents the number of non-edge pixels; H i Represents the gray value corresponding to the i-th non-edge pixel; H z Represents the central gray value of the target road image; H bmax and H bmin Indicates the maximum and minimum grayscale values ​​of edge pixels; Extracting pixel points inside the non-target road image area connected to the edge pixel points as reference pixel points, and obtaining grayscale values ​​of the reference pixel points; Obtaining a second contrast adjustment coefficient using the grayscale value of the reference pixel point combined with the grayscale value of the edge pixel point; The second contrast adjustment coefficient is obtained by the following formula: Among them, J 02 represents the second contrast adjustment coefficient; k represents the number of reference pixels; m represents the number of edge pixels; H ki Represents the gray value corresponding to the i-th reference pixel; H z Represents the central gray value of the target road image; H bmax Indicates the maximum grayscale value among edge pixels; H cmax Indicates the maximum grayscale value in the reference pixel; H bi Represents the gray value corresponding to the i-th edge pixel; The contrast in the target road image is adjusted using the first contrast adjustment coefficient and the second contrast adjustment coefficient. The adjusted contrast is obtained by the following formula: Among them, L t represents the contrast after adjustment; L0 represents the contrast before adjustment; J 02 represents the second contrast adjustment coefficient; J 01 Represents the first contrast adjustment coefficient.

4. The Adams road surface modeling method based on computer vision and deep reinforcement learning algorithm as claimed in claim 1, characterized in that: Performing road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data; The road contour pixel coordinate data is sorted using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information, including: Performing trajectory recognition on a single road profile included in the pre-processed road image to obtain trajectory information of the single road profile on the pre-processed road image screen; Based on the trajectory information, performing road contour pixel point extraction processing on the pre-processed road image to obtain position data of all road contour pixel coordinate points; The position data of all road contour pixel coordinate points are sorted using a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are sorted in a directed manner to obtain the corresponding coordinate information.

5. The Adams pavement modeling method based on computer vision and deep reinforcement learning algorithm as claimed in claim 1, characterized in that: Importing the coordinate information into Adams software to construct a road surface model to obtain a corresponding road surface model; A whole vehicle model is also constructed by using the Adams software, and the whole vehicle model is placed on the road surface model for multi-body dynamics simulation, including: Importing the coordinate information into the road builder module of the Adams software to construct a road model, obtaining a corresponding road model, and uploading the road model to a road model library after marking it; The whole vehicle model is constructed through the car module of the Adams software. Based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation.

6. Adams road modeling system based on computer vision and deep reinforcement learning algorithm, characterized by: An image segmentation module is used to perform image segmentation on the original road image to obtain a target road image; an image preprocessing module is used to preprocess the target road image to obtain a preprocessed road image; A road contour recognition module, used to perform road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data; A road coordinate sorting module is used to sort the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information; A pavement model building module is used to import the coordinate information into Adams software to build a pavement model and obtain a corresponding pavement model; The dynamics simulation module is used to construct a whole vehicle model through the Adams software, and place the whole vehicle model on the road surface model to perform multi-body dynamics simulation.

7. The Adams road surface modeling system based on computer vision and deep reinforcement learning algorithm as claimed in claim 6, characterized in that: The image segmentation module is used to perform image segmentation on the original road image to obtain a target road image, including: Taking panoramic photos of a real road to obtain an original road image of the real road; performing annotation recognition on the original road image to obtain distribution position information of all annotations on the original road image, and performing image segmentation on the original road image based on the distribution position information to obtain an unannotated target road image; The image preprocessing module is used to preprocess the target road image to obtain a preprocessed road image, including: The target road image is subjected to machine learning processing for reducing invalid data to obtain a deep learning road image; wherein the machine learning processing includes grayscale processing, Gaussian pyramid processing or binarization processing; and the deep learning road image is subjected to picture filling preprocessing to obtain a preprocessed road image; wherein the preprocessed road image includes a single road contour.

8. The Adams road surface modeling system based on computer vision and deep reinforcement learning algorithm as claimed in claim 7, characterized in that: Before performing machine learning processing on the target road image, the contrast of the target road image is adjusted, including: extracting the target road image; Scanning each pixel point of the target road image, extracting edge pixel points of a corresponding area of ​​the target road image, and obtaining grayscale values ​​of the edge pixel points of the corresponding area of ​​the target road image; Extract the grayscale values ​​corresponding to the non-edge pixels in the target road image except the edge pixels; Obtaining a first contrast adjustment coefficient using grayscale values ​​corresponding to non-edge pixels in the target road image; The first contrast adjustment coefficient is obtained by the following formula: Among them, J 01 represents the first contrast adjustment coefficient; n represents the number of non-edge pixels; H i Represents the gray value corresponding to the i-th non-edge pixel; H z Represents the central gray value of the target road image; H bmax and H bmin Indicates the maximum and minimum grayscale values ​​of edge pixels; Extracting pixel points inside the non-target road image area connected to the edge pixel points as reference pixel points, and obtaining grayscale values ​​of the reference pixel points; Obtaining a second contrast adjustment coefficient using the grayscale value of the reference pixel point combined with the grayscale value of the edge pixel point; The second contrast adjustment coefficient is obtained by the following formula: Among them, J 02 represents the second contrast adjustment coefficient; k represents the number of reference pixels; m represents the number of edge pixels; H ki Represents the gray value corresponding to the i-th reference pixel; H z Represents the central gray value of the target road image; H bmax Indicates the maximum grayscale value of edge pixels; H cmax Indicates the maximum grayscale value in the reference pixel; H bi Represents the gray value corresponding to the i-th edge pixel; The contrast in the target road image is adjusted using the first contrast adjustment coefficient and the second contrast adjustment coefficient. The adjusted contrast is obtained by the following formula: Among them, L t represents the contrast after adjustment; L0 represents the contrast before adjustment; J 02 represents the second contrast adjustment coefficient; J 01 Represents the first contrast adjustment coefficient.

9. The Adams road modeling system based on computer vision and deep reinforcement learning algorithm as claimed in claim 6, characterized in that: The road contour recognition module is used to perform road contour recognition on the pre-processed road image to obtain road contour pixel coordinate data, including: Performing trajectory recognition on a single road contour contained in the preprocessed road image to obtain trajectory information of the single road contour on the preprocessed road image screen; performing road contour pixel point extraction processing on the preprocessed road image based on the trajectory information to obtain position data of all road contour pixel coordinate points; The road coordinate sorting module is used to sort the road contour pixel coordinate data using a deep reinforcement learning algorithm based on an attention mechanism to obtain corresponding coordinate information, including: The position data of all road contour pixel coordinate points are sorted using a deep reinforcement learning algorithm based on the attention mechanism, and the scattered points in all road contour pixel coordinate points are sorted in a directed manner to obtain the corresponding coordinate information.

10. The Adams road surface modeling system based on computer vision and deep reinforcement learning algorithm as claimed in claim 6, characterized in that: The pavement model building module is used to import the coordinate information into the Adams software to build a pavement model, and obtain a corresponding pavement model, including: The coordinate information is imported into the road builder module of the Adams software to construct a road model, and the corresponding road model is obtained. The road model is marked and uploaded to the road model library; the dynamics simulation module is used to construct a whole vehicle model through the Adams software, and the whole vehicle model is placed on the road model for multi-body dynamics simulation, including: The whole vehicle model is constructed through the car module of the Adams software. Based on the current vehicle dynamics simulation requirements, the matching road model is retrieved from the road model library, and the whole vehicle model is placed on the simulation track corresponding to the retrieved model through the car module for multi-body dynamics simulation.

Images