Synthetic data set construction method of asphalt pavement texture three-dimensional reconstruction model

Abstract

The invention provides a synthetic data set construction method of an asphalt pavement texture three-dimensional reconstruction model, and particularly relates to the technical field of road engineering. The synthetic data set construction method specifically comprises the following steps: S1, acquiring a bituminous pavement texture three-dimensional point cloud data set by adopting a laser texture instrument or a texture generation model; s2, performing rendering processing on one piece of asphalt pavement texture three-dimensional point cloud data by using a D5 rendering workflow or a Corona rendering workflow to obtain a monocular image-depth image pair of the asphalt pavement texture three-dimensional point cloud data; and S3, repeatedly executing the step S2, sequentially performing automatic batch rendering processing on the remaining three-dimensional point cloud data, obtaining the remaining monocular image-depth image pairs, and further obtaining an asphalt pavement texture synthetic data set for three-dimensional reconstruction model training. The obtained asphalt pavement texture synthetic data set is specially used for three-dimensional model training based on deep learning, and daily detection and maintenance of the pavement and management and control of the road construction quality can be rapidly and efficiently achieved.

Description

Technical Field

[0001] This invention patent relates to the field of road engineering technology, specifically to a method for constructing a synthetic dataset of a three-dimensional reconstruction model of asphalt pavement texture. Background Technology

[0002] In recent years, with the rapid development of computer vision and deep learning technologies, 3D reconstruction technology based on monocular images has become a research hotspot. In such tasks, the quality of dataset construction and annotation directly determines the model training effect and reconstruction accuracy. Generally, datasets used for 3D reconstruction model training are mainly divided into two categories: real-label datasets and synthetic datasets. Real-label datasets are constructed from monocular images and real 3D depth information acquired through high-precision measurement equipment (such as laser scanners), and are often used as standards or validation benchmarks for model training. Synthetic datasets refer to images constructed using computer-generated techniques such as 3D modeling, rendering, or simulation, along with their corresponding depth information, normal vectors, and reflection features, with precise annotations. They offer advantages such as controllable cost, scalability, and error-free annotation. Therefore, in scenarios where obtaining real data is difficult or annotation costs are too high, synthetic data has become an important means of model pre-training and data augmentation, and is widely used in fields such as depth estimation, semantic segmentation, and object detection.

[0003] Currently, the field of computer vision has relatively abundant publicly available datasets, such as NYU Depth V2, KITTI, Make3D, and MiDaS for general scene depth estimation. These datasets provide a foundation for algorithm research in indoor and outdoor environments, traffic scenes, and natural objects. However, in the field of road surface texture reconstruction, especially for the reconstruction of the microscopic and macroscopic 3D texture features of asphalt pavements, there is currently a lack of large-scale, standardized, and publicly available datasets. This deficiency severely restricts the development and application of deep learning-based 3D reconstruction models for asphalt pavements. The main reasons for this situation include: Firstly, real-world data acquisition requires high-precision equipment such as laser scanners or structured light systems, which are costly, complex to operate, and have high requirements for the acquisition environment and stability. Furthermore, during large-scale or multi-scale texture data acquisition, extensive pose calibration and geometric registration are required, making post-processing cumbersome and time-consuming. Secondly, there is currently no synthetic dataset specifically designed for training 3D reconstruction models of asphalt pavement textures. In response to the practical needs in this field, there is an urgent need for a method that can efficiently and cost-effectively generate large-scale road surface texture image data with real depth labels, so as to provide data support for the training and validation of monocular image depth estimation models. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this invention provides a method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture, the specific technical solution of which is as follows: A method for constructing a synthetic dataset for a 3D reconstruction model of asphalt pavement texture, specifically including the following steps: S1. Use a laser texture analyzer or texture generation model to obtain a 3D point cloud dataset of asphalt pavement texture; S2. Use the D5 rendering workflow or the Corona rendering workflow to render one of the asphalt pavement texture 3D point cloud data in the S1 dataset to generate a monocular texture image, and pair it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair. S3. Repeat S2 to automatically batch render the remaining asphalt pavement texture 3D point cloud data, obtain the remaining monocular image-depth image pairs, and construct the asphalt pavement texture synthesis dataset for training the 3D reconstruction model from all monocular image-depth image pairs.

[0005] Preferably, in S1, the two methods for obtaining the 3D point cloud dataset of asphalt pavement texture are as follows: S1.1 uses a laser texture analyzer to acquire a 3D point cloud dataset of asphalt pavement texture, which includes the following sub-steps: S1.1.1 Plan and clean the scanned road surface to avoid the presence of dust and gravel; S1.1.2 Check and calibrate the laser texture analyzer, set the laser texture scanning parameters, and ensure that the laser texture analyzer is properly connected to each auxiliary scanning sensor; S1.1.3 Complete the three-dimensional scanning of each measuring point in sequence. To reduce random errors, repeat the scanning at the same measuring point at least three times and take the average value to obtain the three-dimensional point cloud dataset of asphalt pavement texture. Throughout the scanning process, the coordinates of each measuring point, scanning time, weather, road segment number, road grade, lane type, basic instrument parameters, and calibration records must be recorded to facilitate subsequent data traceability. S1.2 uses a texture generation model to obtain a 3D point cloud dataset of asphalt pavement texture, specifically including the following sub-steps: S1.2.1 Generate a two-dimensional random matrix that conforms to a standard normal distribution using a random function; S1.2.2 Normalizes the two-dimensional random matrix and maps it to the interval (−1, 1); S1.2.3 Input the normalized two-dimensional random matrix into the texture generation model to obtain the three-dimensional point cloud data of the asphalt pavement texture; S1.2.4 By repeatedly sampling a two-dimensional random matrix, multiple sets of virtual road surface textures can be generated in batches for constructing a three-dimensional point cloud dataset of synthesized asphalt road surface textures.

[0006] Preferably, the two methods of image rendering processing in S2 are as follows: S2.1 uses the D5 rendering workflow for image rendering processing, which includes the following sub-steps: S2.1.1 Import the 3D point cloud dataset of asphalt pavement texture obtained in S1 into 3ds Max and convert it into a format that can be recognized by the D5 renderer; S2.1.2 Import the asphalt pavement texture 3D point cloud dataset from S2.1.1 (after format conversion) into the D5 renderer; S2.1.3 sets the environmental parameters, road surface materials, and camera configuration for the D5 renderer. The environmental parameters mainly include configuring HDRI lighting, importing a cloudy environment texture, and simulating realistic natural lighting and shadows. The road surface material setting selects "asphalt." The camera configuration includes setting the virtual light source, camera imaging parameters, and position. The virtual light source uses natural lighting conditions. The camera imaging parameters include focal length, focus, filter, exposure, chromatic aberration, and vignetting. The camera position is set directly above the road surface model. S2.1.4 Renders one of the asphalt pavement texture 3D point cloud data in the dataset imported in S2.1.2, generates a monocular image, and pairs it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair. S2.2 uses the Corona rendering workflow for image rendering, which includes the following sub-steps: S2.2.1 Import the 3D point cloud dataset of asphalt pavement texture obtained in S1 into the 3ds Max scene and set the renderer to Corona. At the same time, perform geometry optimization, scene configuration, and material and lighting settings. S2.2.2 Renders one of the asphalt pavement texture 3D point cloud data from the dataset imported in S2.2.1, generates a monocular image, and pairs it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair.

[0007] Preferably, in S2.2.1, the scene configuration specifically includes: setting the noise level limit to 1.5% to 2.2%, selecting the "Corona High Quality" mode in the noise reduction module, locking the rendering viewport and adding a Corona camera, placing the camera in the right or left view, and adjusting the camera position to be directly above the road surface model to ensure that the viewpoint covers the entire texture area; the material and lighting settings specifically include: adding a light source and setting the light in the right view, selecting the "Metal Sandblasted Black" material in the Corona material library, using the eyedropper tool to sample the model material in the "Simplified Material Editor" mode, and adjusting the refractive index, bump intensity, roughness, and anisotropy.

[0008] Preferably, the batch rendering process described in S3 employs either the Python method or the MaxScript macro recording method.

[0009] Preferably, the Python method specifically includes the following sub-steps: (1) Use a Python script to traverse the directory containing a large number of 3D road surface model files; (2) For each model file in the directory, the Python script calls the 3ds Max command line through the subprocess module or directly connects to 3ds Max using the pymxs library and passes the automatically generated MaxScript script to it; (3) Use MaxScript scripts to automatically perform the following operations within 3ds Max: import the current model, apply preset uniform material parameters, set fixed camera and lighting, and configure the renderer; (4) Rendering or exporting instructions are executed: If using the D5 renderer, first export the models in batches as .d5a format files, then perform batch rendering and save the high-resolution images to the specified path; If the Corona renderer is used, high-resolution images will be rendered in batches and saved to the specified path directly; (5) The MaxScript script automatically handles the naming logic to ensure that the name of the output file corresponds one-to-one with the name of the input model, and integrates an exception handling mechanism to ensure that the failure of processing a single file does not affect the overall batch task, and obtains the remaining monocular image-depth image pairs.

[0010] More preferably, the MaxScript macro recording method specifically includes the following sub-steps: (1) Using the macro recorder, automatically record the entire process of executing S2.1 with D5 renderer or executing S2.2 with Corona renderer and generate the corresponding MaxScript code; (2) The recorded MaxScript code is encapsulated into a function that can accept the model file path as an input parameter. The key modification is to replace the part of the code that manually opens a specific file with a function parameter. (3) Add a loop control structure outside the encapsulated function so that it can traverse all 3D road surface model files in the specified folder and automatically generate MaxScript script using 3ds Max; (4) Run the MaxScript script. 3ds Max will automatically open each model file in sequence, apply the same rendering or export settings and automatically output the results to complete the batch processing. Ensure that the name of the output file corresponds one-to-one with the name of the input model and obtain the remaining monocular image-depth image pairs.

[0011] The beneficial effects of this invention are: 1. This invention can quickly construct an asphalt pavement texture synthesis dataset composed of 3D point cloud data and its corresponding monocular rendered images, replacing physical equipment such as high-definition cameras and laser scanners with digital generation and automatic rendering, which greatly reduces hardware costs; 2. This invention utilizes a fully automated batch processing workflow to generate hundreds of thousands to millions of samples with real depth labels within hours to days. By replacing manual collection with digital generation, it significantly reduces the cost of dataset construction and makes it possible to build ultra-large-scale datasets. 3. The synthetic dataset constructed in this invention is specifically designed for training deep learning-based 3D reconstruction models of asphalt pavement textures, and can provide a reliable data foundation for the convergence and performance improvement of monocular, binocular and multi-view 3D reconstruction algorithms. Attached Figure Description

[0012] The accompanying drawings constituting this invention are provided to further understand this application and do not constitute an undue limitation of this application.

[0013] Figure 1 This is a flowchart of the present invention; Figure 2 This is a flowchart of Example 1, which is based on an asphalt pavement texture generation model and a D5 rendering method. Figure 3 This is a flowchart of Example 2 based on a laser texture scanner and the Corona rendering method; Figure 4 This is a rendering (monocular image) of one set of 3D point clouds (depth maps) from the synthetic dataset constructed in Example 1; Figure 5 This is a rendering (monocular image) of one of the three-dimensional point clouds (depth maps) from the synthetic dataset constructed in Example 2. Detailed Implementation

[0014] The specific implementation of the method for constructing a synthetic dataset of a three-dimensional reconstruction model of asphalt pavement texture provided by the present invention will be further described in conjunction with the accompanying drawings and embodiments.

[0015] like Figure 1 As shown, a method for constructing a synthetic dataset for a 3D reconstruction model of asphalt pavement texture specifically includes the following steps: S1. Use a laser texture analyzer or texture generation model to obtain a 3D point cloud dataset of asphalt pavement texture; S2. Use the D5 rendering workflow or the Corona rendering workflow to render one of the asphalt pavement texture 3D point cloud data in the S1 dataset, generate a corresponding monocular texture image, and pair it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair. S3. Repeat S2 to automatically batch render the remaining asphalt pavement texture 3D point cloud data, obtain the remaining monocular image-depth image pairs, and merge all monocular image-depth image pairs to construct an asphalt pavement texture synthesis dataset for training the 3D reconstruction model.

[0016] Preferably, in step S1, a high-precision laser texture analyzer is used to scan the real road surface to obtain a 3D point cloud dataset of asphalt road surface texture, specifically including the following sub-steps: S1.1.1 Plan the measurement points and clean the scanned road surface with a dry towel to avoid the presence of dust and gravel; S1.1.2 Check and calibrate the laser texture analyzer, set the laser texture scanning parameters, and ensure that the laser texture analyzer is properly connected to each auxiliary scanning sensor; S1.1.3 Complete the three-dimensional scanning of each measuring point in sequence. To reduce random errors, repeat the scanning at the same measuring point at least three times and take the average value to obtain the three-dimensional point cloud dataset of asphalt pavement texture. Throughout the scanning process, the coordinates of each measuring point, scanning time, weather, road segment number, road grade, lane type, basic instrument parameters, and calibration records must be recorded to facilitate subsequent data traceability. Preferably, in S1, the texture generation model is mainly based on the material parameters of asphalt pavement such as gradation and porosity. The model obtains a 3D point cloud dataset of asphalt pavement texture, specifically including the following sub-steps: S1.2.1 Generate a two-dimensional random matrix that conforms to a standard normal distribution using a random function; S1.2.2 Normalizes the two-dimensional random matrix and maps it to the interval (−1, 1); S1.2.3 Input the normalized two-dimensional random matrix into the texture generation model to obtain the three-dimensional point cloud data of the asphalt pavement texture; S1.2.4 By repeatedly sampling a two-dimensional random matrix, multiple sets of virtual road surface textures can be generated in batches for constructing a three-dimensional point cloud dataset of synthesized asphalt road surface textures.

[0017] Import the obtained 3D point cloud data of asphalt pavement texture into 3D modeling software (3ds Max). It is worth noting that the imported model needs to be preprocessed before rendering: accurately set parameters related to the optical properties of real asphalt pavement, including but not limited to: refractive index, roughness, anisotropy, and metallicity.

[0018] Preferably, the image rendering process can be performed using either of the following two methods: S2.1 uses the D5 rendering workflow for image rendering processing, which includes the following sub-steps: S2.1.1 Import the 3D point cloud dataset of asphalt pavement texture obtained in S1 into 3ds Max and convert it into a format that can be recognized by the D5 renderer; S2.1.2 Import the asphalt pavement texture 3D point cloud dataset from S2.1.1 (after format conversion) into the D5 renderer; S2.1.3 sets the environmental parameters, road surface materials, and camera configuration for the D5 renderer. The environmental parameters primarily include configuring HDRI lighting, importing a cloudy environment texture, and simulating realistic natural lighting and shadows. The road surface material settings select "asphalt" to enhance surface roughness and graininess, simulating the microscopic optical response of a real road surface. The camera configuration includes setting the virtual light source, camera imaging parameters, and position. The virtual light source uses natural lighting conditions to simulate natural lighting and generate a rendered image with realistic lighting and shadow layers. The camera imaging parameters include focal length, focus, filter, exposure, chromatic aberration, and vignetting. The camera position is set directly above the road surface model. S2.1.4 Renders one of the asphalt pavement texture 3D point clouds in the dataset imported from S2.1.2, generates a monocular image, and pairs it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair. S2.2 uses the Corona rendering workflow for image rendering, which includes the following sub-steps: S2.2.1 Import the 3D point cloud dataset of asphalt pavement texture obtained in S1 into the 3ds Max scene and set the renderer to Corona. At the same time, perform geometry optimization, scene configuration, and material and lighting settings. S2.2.2 Renders one of the asphalt pavement texture 3D point cloud data from the dataset imported in S2.2.1, generates a monocular image, and pairs it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair.

[0019] Preferably, in S2.2.1, the scene configuration specifically includes: setting the noise level limit to 1.5% to 2.2%, selecting the "Corona High Quality" mode in the noise reduction module, locking the rendering viewport and adding a Corona camera, placing the camera in the right or left view, and adjusting the camera position to be directly above the road surface model to ensure that the viewpoint covers the entire texture area; the material and lighting settings specifically include: adding a light source and setting the light in the right view, selecting the "Metal Sandblasted Black" material in the Corona material library, using the eyedropper tool to sample the model material in the "Simplified Material Editor" mode, and adjusting the refractive index, bump intensity, roughness, and anisotropy.

[0020] Preferably, the batch rendering process described in S3 employs either the Python method or the MaxScript macro recording method.

[0021] Preferably, the Python method specifically includes the following sub-steps: (1) Use a Python script to traverse the directory containing a large number of 3D road surface model files; (2) For each model file in the directory, the Python script calls the 3ds Max command line through the subprocess module or directly connects to 3ds Max using the pymxs library and passes the automatically generated MaxScript script to it; (3) Use MaxScript scripts to automatically perform the following operations within 3ds Max: import the current model, apply preset uniform material parameters, set fixed camera and lighting, and configure the renderer; (4) Rendering or exporting instructions are executed: If using the D5 renderer, first export the models in batches as .d5a format files, then perform batch rendering and save the high-resolution images to the specified path; If the Corona renderer is used, high-resolution images will be rendered in batches and saved to the specified path directly; (5) The MaxScript script automatically handles the naming logic to ensure that the name of the output file corresponds one-to-one with the name of the input model, and integrates an exception handling mechanism to ensure that the failure of processing a single file does not affect the overall batch task, and obtains the remaining monocular image-depth image pairs.

[0022] More preferably, the MaxScript macro recording method specifically includes the following sub-steps: (1) Using the macro recorder, automatically record the entire process of executing S2.1 with D5 renderer or executing S2.2 with Corona renderer and generate the corresponding MaxScript code; (2) The recorded MaxScript code is encapsulated into a function that can accept the model file path as an input parameter. The key modification is to replace the part of the code that manually opens a specific file with a function parameter. (3) Add a loop control structure outside the encapsulated function so that it can traverse all 3D road surface model files in the specified folder and automatically generate MaxScript script using 3ds Max; (4) Run the MaxScript script. 3ds Max will automatically open each model file in sequence, apply the same rendering or export settings and automatically output the results to complete the batch processing. Ensure that the name of the output file corresponds one-to-one with the name of the input model and obtain the remaining monocular image-depth image pairs.

[0023] Example 1: Taking SMA13 pavement texture as an example, based on the design parameters of SMA13 asphalt mixture material, the 3D point cloud data of SMA13 asphalt pavement texture is generated using an asphalt pavement texture generation model, and batch rendering is achieved using the D5 renderer combined with Python. Figure 2 As shown: First, the model obtains the design parameters (asphalt-aggregate ratio, void ratio, aggregate void ratio, asphalt saturation, etc.) of SMA13 asphalt mixture and generates a 3D point cloud dataset of asphalt pavement texture. Then, MATLAB is used to preprocess and triangulate the 3D point cloud data of asphalt pavement texture to generate a .stl file. Secondly, import the .stl format file into 3ds Max software and convert it to a .d5a format file that the D5 renderer can recognize; Next, import the .d5a format file into the D5 renderer and set the environment parameters, road surface materials, and camera configuration for the D5 renderer: (1) Configure HDRI lighting to simulate realistic natural light and shadow; (2) Scene and environment settings: Select the "HDRI" option and import the cloud environment map (this type of map can enhance the sense of light and shadow on the road surface); Record the current environment parameters and post-processing parameters in "Add New Scene" for easy reference during subsequent batch rendering; (3) Set the imaging parameters for the camera configuration to ensure a fixed viewpoint and realistic image: Use the default camera parameters (focal length, focus, etc. remain at default). Value); Select “None” in LUT filter (if you need to adjust the image style, you can select black and white filter and set the intensity to 0.2), turn off “auto exposure”, set the dispersion to 0.2 and the vignette to 0.4 to enhance the realism of the rendered image; (4) Material assignment, add real asphalt material to the model and optimize the texture performance: select “asphalt” as the “road surface” material and drag it to the road surface model; in the material parameter panel, metallicity, UV and three-way mapping repair are often used to adjust the display effect. Open the “three-way mapping repair” option, adjust the UV parameters to keep the texture shape and optimize the texture display effect.

[0024] Next, a large-scale standardized image dataset is automatically output: the image ratio is set to 16:9 and the preset size is 2K; all SMA13 road surface model files that need to be rendered are selected in the rendering queue for rendering, and the rendering results are batch processed and exported using Python to complete the SMA13 road surface texture rendering. Finally, all monocular image-depth image pairs are used to construct a synthetic asphalt pavement texture dataset for training the 3D reconstruction model. For example... Figure 4 As shown, this is an example of one set of image pairs from the synthetic dataset obtained in this embodiment.

[0025] Example 2: Taking AC5 pavement texture as an example, 3D point cloud data of AC5 asphalt pavement texture is generated using a high-precision laser texture scanner, and batch rendering is achieved using the Corona renderer combined with MaxScript macro recording method. Figure 3 As shown: First, the real road surface scan data is converted into an editable 3D model while retaining the original roughness. The 3D point cloud data of the asphalt pavement texture of AC5 road surface is obtained. Then, MATLAB is used to preprocess and triangulate the 3D point cloud data of the asphalt pavement texture to generate a .stl format file. Next, launch 3ds Max software and import the 3D point cloud data of the asphalt pavement texture in .stl format into the 3ds Max scene. When importing, you can leave the automatic smoothing option unchecked to preserve the original roughness of the texture as much as possible. Next, perform geometry optimization, scene configuration, and material and lighting settings for the Corona renderer: (1) Smooth the sharp corners of the model to achieve a balance between noise reduction and maintaining texture details: Select the imported asphalt pavement texture 3D point cloud data model, select the "Turbo Smoothing" function to perform preliminary smoothing on the model to optimize the sharp corners of the texture; In addition, according to actual needs, the turbo smoothing operation can be repeated to achieve different degrees of optimization, but in order to maintain the original texture shape, it is recommended to control the number of smoothing times; (2) Set a high-quality rendering environment and a fixed viewing angle to ensure image standardization: Set the noise level limit to 2% and select the "Corona High Quality" mode in the noise reduction module; (3) Lock the rendering viewport, add a Corona camera, place the camera in the right view or left view, and adjust the phase Position the camera directly above the road surface model to ensure that the view covers the entire texture area. Select the camera as the filter to avoid selecting texture objects. (4) Set the materials and lighting to give the model realistic optical performance and light and shadow effects: Select the Corona sun as the light source, add a light source in the right view and zoom in, and adjust the position and angle of the light source. Select the "Metal Sandblasting Black" material in the material library and switch to the "Simplified Material Editor" mode. Use the eyedropper tool to sample the model material and adjust the following parameters: IOR (refractive index) is set to 1.6, bump intensity is set to 0.8, roughness is set to 0.2, anisotropy is set to 0.2, rotation angle is set to 90°, uncheck metallicity, and drag the "basic bump" map to the "displacement" channel to enhance the surface roughness and graininess. Then, switch the perspective, start the Corona real-time rendering window in the camera view, observe the rendering effect, and fine-tune the exposure and color through functions such as "color mapping" and "ACES output conversion". Adjust the parameters appropriately according to the actual situation, use the MaxScript macro recording method to batch process and export the rendering results, and complete the AC5 road surface texture rendering. Finally, all monocular image-depth image pairs are used to construct a synthetic asphalt pavement texture dataset for training the 3D reconstruction model. For example... Figure 5 As shown, this is an example of one set of image pairs from the synthetic dataset obtained in this embodiment.

[0026] The asphalt pavement texture synthesis dataset obtained by this invention is specifically designed for training deep learning-based 3D models, enabling rapid and efficient daily pavement inspection and maintenance, as well as quality control of road construction.

[0027] In this invention, terms such as "upper," "lower," "bottom," and "top" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are merely used to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any particular component or element in this invention, nor should they be construed as limiting the invention. Terms such as "connected" and "linked" should be interpreted broadly, indicating a fixed connection, an integral connection, or a detachable connection; a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can determine the specific meaning of the above terms in this invention based on the specific circumstances, and they should not be construed as limiting the invention.

[0028] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture, characterized in that, Specifically, the following steps are included: S1. Use a laser texture analyzer or texture generation model to obtain a 3D point cloud dataset of asphalt pavement texture; S2. Use the D5 rendering workflow or the Corona rendering workflow to render one of the asphalt pavement texture 3D point cloud data in the S1 dataset to generate a monocular texture image, and pair it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair. S3. Repeat S2 to automatically batch render the remaining asphalt pavement texture 3D point cloud data, obtain the remaining monocular image-depth image pairs, and construct the asphalt pavement texture synthesis dataset for training the 3D reconstruction model from all monocular image-depth image pairs.

2. The method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture according to claim 1, characterized in that, In S1, there are two methods for obtaining the 3D point cloud dataset of asphalt pavement texture, specifically: S1.1 uses a laser texture analyzer to acquire a 3D point cloud dataset of asphalt pavement texture, which includes the following sub-steps: S1.1.1 Plan the measurement points and clean the scanned road surface to avoid the presence of dust and gravel; S1.1.2 Check and calibrate the laser texture analyzer, set the laser texture scanning parameters, and ensure that the laser texture analyzer is properly connected to each auxiliary scanning sensor; S1.1.3 Complete the three-dimensional scanning of each measuring point in sequence. To reduce random errors, repeat the scanning at the same measuring point at least three times and take the average value to obtain the three-dimensional point cloud dataset of asphalt pavement texture. Throughout the scanning process, the coordinates of each measuring point, scanning time, weather, road segment number, road grade, lane type, basic instrument parameters, and calibration records must be recorded to facilitate subsequent data traceability. S1.2 uses a texture generation model to obtain a 3D point cloud dataset of asphalt pavement texture, which specifically includes the following sub-steps: S1.2.1 Generate a two-dimensional random matrix that conforms to a standard normal distribution using a random function; S1.2.2 Normalizes the two-dimensional random matrix and maps it to the interval (−1, 1); S1.2.3 Input the normalized two-dimensional random matrix into the texture generation model to obtain the three-dimensional point cloud data of the asphalt pavement texture; S1.2.4 By repeatedly sampling a two-dimensional random matrix, multiple sets of virtual road surface textures can be generated in batches for constructing a three-dimensional point cloud dataset of synthesized asphalt road surface textures.

3. The method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture according to claim 2, characterized in that, In S2, there are two methods for image rendering processing, specifically: S2.1 uses the D5 rendering workflow for image rendering processing, which includes the following sub-steps: S2.1.1 Import the 3D point cloud dataset of asphalt pavement texture obtained in S1 into 3ds Max and convert it into a format that can be recognized by the D5 renderer; S2.1.2 Import the asphalt pavement texture 3D point cloud dataset from S2.1.1, which has undergone format conversion, into the D5 renderer; S2.1.3 sets the environmental parameters, road surface materials, and camera configuration for the D5 renderer. The environmental parameters mainly include configuring HDRI lighting, importing a cloudy environment texture, and simulating realistic natural lighting and shadows. The road surface material setting selects "asphalt." The camera configuration includes setting the virtual light source, camera imaging parameters, and position. The virtual light source uses natural lighting conditions. The camera imaging parameters include focal length, focus, filter, exposure, chromatic aberration, and vignetting. The camera position is set directly above the road surface model. S2.1.4 Renders one of the asphalt pavement texture 3D point cloud data in the dataset imported in S2.1.2, generates a monocular image, and pairs it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair. S2.2 uses the Corona rendering workflow for image rendering, which includes the following sub-steps: S2.2.1 Import the 3D point cloud dataset of asphalt pavement texture obtained in S1 into the 3ds Max scene and set the renderer to Corona. At the same time, perform geometry optimization, scene configuration, and material and lighting settings. S2.2.2 Renders one of the asphalt pavement texture 3D point cloud data from the dataset imported in S2.2.1, generates a monocular image, and pairs it with the original depth map of the asphalt pavement texture 3D point cloud data to form a monocular image-depth image pair.

4. The method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture according to claim 3, characterized in that, In S2.2.1, the scene configuration specifically includes: setting the noise level limit to 1.5% to 2.2%, selecting the "Corona High Quality" mode in the noise reduction module, locking the rendering viewport and adding a Corona camera, placing the camera in the right or left view, adjusting the camera position to be directly above the road surface model, and ensuring that the viewpoint covers the entire texture area. The specific material and lighting settings include: adding a light source and setting the light in the right view, selecting the "Metal Sandblasted Black" material in the Corona material library, using the eyedropper tool to sample the model material in the "Simplified Material Editor" mode, and adjusting the refractive index, bump intensity, roughness, and anisotropy.

5. The method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture according to claim 3, characterized in that, The batch rendering process described in S3 can be performed using either the Python method or the MaxScript macro recording method.

6. The method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture according to claim 5, characterized in that, The Python method specifically includes the following sub-steps: (1) Use a Python script to traverse the directory containing a large number of 3D road surface model files; (2) For each model file in the directory, the Python script calls the 3ds Max command line through the subprocess module or directly connects to 3ds Max using the pymxs library and passes the automatically generated MaxScript script to it; (3) Use MaxScript scripts to automatically perform the following operations within 3ds Max: import the current model, apply preset uniform material parameters, set fixed camera and lighting, and configure the renderer; (4) Rendering or exporting instructions are executed: If using the D5 renderer, first export the models in batches as .d5a format files, then perform batch rendering and save the high-resolution images to the specified path; If the Corona renderer is used, high-resolution images will be rendered in batches and saved to the specified path directly; (5) MaxScript script automatically handles the naming logic to ensure that the name of the output file corresponds one-to-one with the name of the input model, and integrates an exception handling mechanism to ensure that the failure of processing a single file does not affect the overall batch task, and obtains the remaining monocular image-depth image pairs.

7. The method for constructing a synthetic dataset for a three-dimensional reconstruction model of asphalt pavement texture according to claim 5, characterized in that, The MaxScript macro recording method specifically includes the following sub-steps: (1) Enable macro recorder to automatically record the entire process of executing S2.1 using D5 renderer or executing S2.2 using Corona renderer and generate the corresponding MaxScript code; (2) The recorded MaxScript code is encapsulated into a function that can accept the model file path as an input parameter. The key modification is to replace the part of the code that manually opens a specific file with a function parameter. (3) Add a loop control structure outside the encapsulated function so that it can traverse all 3D road surface model files in the specified folder and automatically generate MaxScript script using 3ds Max; (4) Run the MaxScript script. 3ds Max will automatically open each model file in sequence, apply the same rendering or export settings and automatically output the results to complete the batch processing. Ensure that the name of the output file corresponds one-to-one with the name of the input model and obtain the remaining monocular image-depth image pairs.

Images