Super-resolution high-precision map construction method and device, storage medium and program product

Abstract

The invention relates to the technical field of high-definition map construction, in particular to a super-resolution high-precision map construction method and device, a storage medium and a program product. The super-resolution high-precision map construction method comprises the following steps: (1) inputting a low-resolution image, extracting multi-scale features and calculating an entropy value, dividing the image into regions with different complexities according to the entropy value, optimizing the number of iterations by using a dynamic time step sampling strategy, iteratively refining the image to a target resolution, balancing the fidelity and the sense of reality, and constructing a super-resolution high-precision map; and outputting a high-resolution image. And (2) synthesizing a series of map prior scenes for training a model, introducing a shielding enhancement technology to improve the robustness of the model, fusing prior information and real-time perception data, and generating a super-resolution high-precision map. According to the method, the image super-resolution reconstruction quality and efficiency, the naturalness and the fidelity are remarkably improved, and the performance and the robustness of the model under the condition of crossing multiple prior scenes and shielding are improved.

Description

Technical Field

[0001] The present application relates to the technical field of high-definition map construction, and in particular to a method, device, storage medium and program product for constructing a super-resolution high-precision map using a change map prior. Background Art

[0002] With the rapid development of technologies such as autonomous driving and intelligent navigation, high-definition maps, as one of the core supporting technologies, are facing increasing demands for their accuracy and real-time performance. HD maps not only provide autonomous vehicles with critical data such as precise road information, traffic signs, and obstacle locations, but also serve as the foundation for intelligent navigation systems to achieve precise path planning and real-time navigation. However, in real-world application scenarios, the construction of HD maps faces a series of complex and severe challenges.

[0003] Maps in the real world change frequently and complexly, including road widening, building construction or demolition, and traffic sign updates. These changes require high-definition maps to reflect the latest status of the surface in real time. However, high-precision map annotation is an extremely time-consuming and labor-intensive task, requiring significant human and material investment. Consequently, the scarcity of high-quality training data has become a major constraint on the accuracy and generalization capabilities of HD map construction.

[0004] During the map-building process, environmental factors such as occlusion have a significant impact on the accuracy and real-time performance of HD map construction. For example, due to obstructions such as trees and buildings, autonomous vehicles may not be able to directly obtain map information in the obscured areas.

[0005] Therefore, how to enrich data resources, adapt to complex and changing environmental conditions, and enhance the generalization ability of the model have become key issues that need to be urgently addressed in the current field. Summary of the Invention

[0006] In order to solve the above problems, the present application provides a method, device, storage medium and program product for constructing super-resolution high-precision maps using changing map priors. The high-precision map construction method can achieve efficient super-resolution reconstruction, accurately identify and optimize the number of iterations to improve image quality. At the same time, an innovative synthesis method is used to generate map priors to enrich the training set, improve the model generalization ability, and introduce occlusion enhancement technology to enhance the robustness of the model. By improving the map prior query design, integrating prior information with real-time perception data, high-precision, real-time updated high-definition maps are generated to meet the real-time requirements of technologies such as autonomous driving for high-definition maps.

[0007] A method for constructing a super-resolution high-precision map comprises the following steps: (1) Super-resolution reconstruction of low-resolution images: Input a low-resolution image, extract multi-scale features and calculate the entropy value, divide the image into regions of different complexity according to the entropy value, optimize the number of iterations using a dynamic time step sampling strategy, iteratively refine the image to the target resolution, and at the same time, balance fidelity and realism, adjust the model parameters, and output a high-resolution image.

[0008] (2) Constructing real-time high-definition maps based on variable map priors: Generate a series of map prior scenarios based on the synthesis method for model training, and introduce occlusion enhancement technology to improve the robustness of the model; finally, improve the map prior query design, integrate prior information with real-time perception data, and generate super-resolution high-precision maps.

[0009] By adopting the above technical solution, in the field of image super-resolution reconstruction, this application uses a multi-index latent entropy module to accurately identify the complexity and reconstruction difficulty of different regions in the image. Combined with a dynamic time-step sampling strategy to optimize the number of iterations, the reconstruction process is more efficient and targeted. At the same time, the trade-off between fidelity and realism ensures that the output image presents a more natural and realistic visual effect while preserving the original information. The output high-resolution image provides high-resolution image input for map construction.

[0010] In the construction of real-time HD maps, this application utilizes a synthetic method to generate rich prior map data, effectively addressing the scarcity of real-world map change annotations. Furthermore, map occlusion enhancement technology further enhances the model's map construction capabilities in the presence of occlusions. By integrating prior information with real-time perception data, high-precision, real-time HD maps are generated, meeting the real-time requirements of HD maps for technologies such as autonomous driving.

[0011] A preferred solution of the super-resolution high-precision map construction method is that in step (1), the low-resolution image is input, multi-scale features are extracted and entropy values ​​are calculated, the image is divided into regions of different complexity according to the entropy values, the number of iterations is optimized using a dynamic time step sampling strategy, and the iterative refinement of the image to the target resolution specifically includes: A low-resolution image of a real-world scene is input, and a deep learning network is used to extract the multi-scale features of the image. The weighted average entropy of the multi-scale features is calculated; the weighted average entropy is converted into a complexity index; based on the complexity index, the image is divided into regions of different complexity; a dynamic time step sampling strategy is used to assign an appropriate number of iterations to each region. The model iteratively refines each region and generates a target resolution image as output; the number of iterations and complexity of each region are linearly positively correlated.

[0012] By employing the aforementioned technical solutions, this approach quantifies regional complexity by calculating the entropy of image data in the latent space, thereby identifying areas that require more iterations for detailed reconstruction and those that can achieve satisfactory reconstruction results with fewer steps. A dynamic time-step sampling strategy is employed to adaptively adjust the number of iterations for different image regions based on the output of the multi-metric latent entropy module, thereby optimizing the allocation of computing resources, reducing unnecessary iterations, and accelerating the overall processing flow.

[0013] A preferred solution of the super-resolution high-precision map construction method is that in step (1), balancing fidelity and realism, adjusting model parameters, and outputting high-resolution images specifically include: Design a trade-off function that takes the fidelity score and the realism score as input and outputs a weighed score. Use a linear relationship for weighing to obtain the formula: S=λ·P+(1-λ)·R. In this formula, S represents the weighed score, P represents the fidelity score of the target resolution image, R represents the realism score of the target resolution image, and λ is a weight coefficient between 0 and 1.

[0014] Adjust the model parameters so that the realism score reaches the maximum when the fidelity score is fixed at a value; repeatedly adjust the model parameters so that the fidelity score and the realism score are adjusted synchronously, so that the weighed score reaches the maximum value and the final high-resolution image is output.

[0015] By adopting the above technical solution and introducing a trade-off mechanism between fidelity (i.e., the similarity between the restored image and the original high-resolution image) and realism (i.e., the naturalness and credibility of the restored image), we can ensure that the naturalness and credibility of the image are maintained while improving the image resolution.

[0016] A preferred solution of the super-resolution high-precision map construction method is that in step (2), the synthesis method is used to generate a series of map prior scenes for training the model, which specifically includes: A synthesis function is defined that receives diverse scene parameters and synthesizes realistic map data based on these scene parameters. The set of realistic map data is combined with the set of real data to obtain an enhanced training set. The enhanced training set contains a series of prior map scenes. The series of map prior scenes in the enhanced training set are input into the model to train the model.

[0017] By adopting the above technical solutions, this method uses a synthesis method to generate a series of map prior knowledge, simulates various possible map change scenarios, and enriches the content of the training set.

[0018] A preferred solution of the super-resolution high-precision map construction method is that in step (2), the introduction of occlusion enhancement technology to improve the robustness of the model specifically includes: During training, an occlusion mask is first applied to the input data to block out parts of the map, simulating real-world occlusions. The model is then trained with occlusion-augmented data, allowing it to leverage prior information for map construction even in the presence of occlusions.

[0019] By adopting the above technical solution, adding occlusion enhancement data during the training process can enable the model to still use prior information to build maps in the case of occlusion. Autonomous driving cars can also maintain high navigation accuracy when facing obstructions such as trees and buildings.

[0020] A preferred solution of the super-resolution high-precision map construction method is that in step (2), the improved map query design, the fusion of prior information and real-time perception data, and the generation of the super-resolution high-precision map specifically include: Prior information is added to the original query to optimize the query. The key areas of the map are located through the optimized query. The system prioritizes collecting real-time perception data of the key areas and fuses this real-time perception data with the map prior scene to generate a fused super-resolution high-definition map.

[0021] By employing the aforementioned technical solution, this method fuses prior information with real-time perception data to generate highly accurate, constantly updated HD maps, providing reliable navigation support for autonomous vehicles. This not only improves driving safety and comfort for autonomous vehicles, but also provides a strong foundation for autonomous navigation in complex and changing road environments.

[0022] An electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for constructing a super-resolution high-precision map as described above is implemented.

[0023] A computer-readable storage medium having a computer program stored thereon, which, when executed, implements the super-resolution high-precision map construction method as described above.

[0024] A computer program product, when the computer program product is run on a computer, the computer executes the super-resolution high-precision map construction method as described above.

[0025] In summary, the super-resolution high-precision map construction method, device, storage medium, and program product of this application have the following beneficial effects: A. Significantly improve the reconstruction quality of images and maps In image super-resolution reconstruction, this application uses a multi-index latent entropy module to accurately identify the complexity and reconstruction difficulty of different regions in the image. This, combined with a dynamic time-step sampling strategy to optimize the number of iterations, makes the reconstruction process more efficient and targeted. Furthermore, a trade-off between fidelity and realism ensures that the output image retains the original information while presenting a more natural and realistic visual effect.

[0026] In the area of ​​real-time HD map construction, this application utilizes a synthetic approach to generate rich map prior data, effectively addressing the scarcity of real-world map change annotations. Through comparative experiments and optimized training, the generalized model demonstrates comparable or even superior performance to expert models across a variety of prior scenarios. Furthermore, map occlusion enhancement technology further enhances the model's map construction capabilities in the presence of occlusion.

[0027] B. Improve processing efficiency and robustness During the image super-resolution reconstruction process, the dynamic time step sampling strategy adaptively adjusts the number of iterations according to the complexity of the image area, effectively reducing the number of unnecessary iterations and thus speeding up the overall processing flow.

[0028] In terms of real-time HD map construction, an improved map prior query design more fully incorporates prior map information and, combined with real-time perception data, enables high-precision, real-time HD map generation. Furthermore, map occlusion enhancement technology ensures the model remains stable even in complex situations such as occlusion, improving the system's robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of the super-resolution high-precision map construction method of this application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments are described clearly and completely below. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the following embodiments, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0031] A super-resolution high-precision map construction method using a change map prior, reference Figure 1 , the method comprises the following steps in order: (1) Super-resolution reconstruction of low-resolution images: Input a low-resolution image, extract multi-scale features and calculate entropy, identify areas requiring fine reconstruction, and then use a dynamic strategy to optimize the number of iterations, iteratively refining the image to a resolution close to the target high resolution. At the same time, a trade-off function is used to balance fidelity and realism, adjusting parameters until optimal. Ultimately, a natural, realistic high-resolution image is output, achieving a balance between quality and efficiency.

[0032] (2) Constructing a real-time HD map based on variable map priors: We use an innovative synthesis method to generate map priors. By defining a synthesis function S that accepts diverse parameters to generate realistic map data, we enrich the training set and then define HD map tasks. We use a performance evaluation function E to guide model learning and demonstrate the effectiveness of the synthetic data augmentation method. We also introduce occlusion enhancement technology to improve model robustness. Finally, we improve the map prior query design, integrating prior information with real-time perception data to generate high-precision, real-time updated HD maps.

[0033] Step (1) specifically includes the following AC steps.

[0034] A. Multi-Indicator Latent Entropy Module: This module quantifies the complexity of regions by calculating the entropy of image data in the latent space. This allows us to identify regions that require more iterations for detailed reconstruction and regions that can achieve satisfactory reconstruction results with fewer iterations. This step is implemented in the following steps (ad).

[0035] a. Input image: Input a low-resolution image I of a real-world scene LR ,The quality of the input image (such as noise level, blur degree) will affect the subsequent processing effect and computational complexity.

[0036] b. Feature extraction: Using deep learning network CNN to extract low-resolution image I LR The multi-scale features include texture, edge, color and other information, which have different distributions and entropy values ​​in the latent space. The extracted feature set is expressed as: F ext =F(I LR ) Among them, F( ) is the feature extraction function, F ext Represents the extracted feature set, including LR The multi-scale features extracted from

[0037] c. Entropy calculation: Calculate the entropy value of these features in the latent space to obtain the complexity index and reconstruction difficulty of each image region. Define the feature set F ext A certain eigenvector in is f i , whose probability distribution is P(f i ), then the entropy value of this feature can be expressed as:

[0038] Where j represents the feature vector f i In practical applications, it is necessary to discretize or quantize it in order to calculate the probability distribution.

[0039] Taking into account the entropy values ​​of multiple features, the average entropy or weighted average entropy can be calculated. Define the feature set F ext There are N features in , then the average entropy It can be expressed as:

[0040] If each feature weight is different, the weighted average entropy is used :

[0041] Among them, w i Represents feature f i The weight of H(f i ) represents the feature f i The entropy value of .

[0042] Based on the entropy value, a complexity index is defined to measure the complexity and reconstruction difficulty of different regions of the image. The complexity index C can be expressed as:

[0043] If the contribution of multiple features is considered, the complexity index C can be expressed as:

[0044] Where f is a mapping function used to convert entropy value into complexity index.

[0045] d. Region Segmentation: Based on the complexity index calculated by entropy, the image is divided into regions of varying complexity. These regions include smooth areas (such as the sky and water), textured areas (such as leaves and grass), and edge areas (such as building outlines).

[0046] B. Dynamic Time-Step Sampling Strategy: Based on the output of the multi-metric latent entropy module, the number of iterations (i.e., time steps) for different image regions is adaptively adjusted to optimize the allocation of computing resources, reduce unnecessary iterations, and accelerate the overall processing flow. This step is implemented in the following sub-steps (ac).

[0047] a. Receive complexity metrics: First, receive complexity metrics from the multi-metric latent entropy module. These metrics are usually expressed as numerical values ​​or vectors and are used to measure the complexity and reconstruction difficulty of different regions of the image.

[0048] b. Determine the number of iterations: Based on the received complexity index C, the dynamic time step sampling strategy allocates an appropriate number of iterations to each image region. Here, a linear relationship is used for allocation, namely:

[0049] Among them, C i is the complexity index of the i-th region, T i represents the number of iterations in the i-th region, α and β are model parameters, α is a positive number used to adjust the relationship between the number of iterations and complexity.

[0050] c. Iterative refinement: After determining the number of iterations for each region, the model begins iterative refinement of each region.

[0051] The iterative refinement process usually involves a series of mathematical operations and image processing techniques, such as gradient descent, convolution operations, nonlinear activation, etc. These operations are designed to gradually optimize the image quality, making it gradually closer to the target high-resolution image.

[0052] In each iteration, the model adjusts the optimization strategy according to the current image quality and complexity indicators to ensure the best reconstruction effect within a limited number of iterations. The image after the kth iteration is defined as I k , then the iterative refinement process can be expressed as:

[0053] Where f represents the iterative refinement function, θ represents the model parameters, and T i represents the number of iterations for the i-th region.

[0054] Through multiple iterations of refinement, the model generates an image close to the target high resolution as output.

[0055] C. Trade-off between Fidelity and Realism: By introducing a trade-off between fidelity (i.e., the similarity between the restored image and the original high-resolution image) and realism (i.e., the naturalness and credibility of the restored image), we ensure that the naturalness and credibility of the image are maintained while increasing the image resolution. This step is implemented in the following steps (e).

[0056] a. Fidelity Assessment and Realism Optimization: Peak Signal-to-Noise Ratio (PSNR) or Structural Similarity Index (SSIM) is used to measure the similarity between the restored image and the original high-resolution image. While maintaining a certain level of fidelity, image realism is optimized by adjusting model parameters or introducing additional regularization terms (such as the Perceptual Quality Index (PIQ) and Natural Image Quality Evaluation (NIQE). This typically involves adjusting details such as image texture, color, and contrast.

[0057] b. Trade-off mechanism design: Design a trade-off function that takes fidelity and realism as input and outputs a weighed score. Here, a linear relationship is used for the trade-off: S=λ·P+(1-λ)·R Among them, S represents the weighted score, P represents the fidelity score, R represents the realism score, and λ is a weight coefficient between 0 and 1, which is used to adjust the relative importance between fidelity and realism.

[0058] c. Parameter adjustment and optimization: Based on the results of the trade-off function, adjust model parameters (such as the number of iterations, regularization term weight, etc.) to improve the realism of the image while maintaining a certain level of fidelity.

[0059] d. Iterative optimization: Repeat the above process until the best trade-off point is found, that is, the score after trade-off reaches the highest.

[0060] e. Final output: After a trade-off between fidelity and realism, the final high-resolution image is output. This image achieves a relatively balanced state between fidelity and realism, preserving important information from the original image while presenting a natural and realistic visual effect.

[0061] Step (2) specifically includes the following DH steps.

[0062] D. Generate map priors using synthetic methods: Map changes in the real world are not fully annotated in the dataset, and their occurrence frequency is extremely low, approaching 0.

[0063] P(map change)≈0(training set) This scarcity severely restricts the model’s ability to learn map change patterns (such as road expansion, building height changes, etc.).

[0064] Therefore, this application uses a synthetic method to generate map priors. The core of this method is to simulate various possible map change scenarios to enrich the content of the training dataset. Define a synthetic function S that can accept a series of changing parameters (such as road width, building height, etc.) and generate realistic synthetic map data based on these parameters:

[0065] Call a single synthesis function to generate a single synthetic map data M synth A large number of calls to the synthesis function S (parameters) generate a synthetic map data set D synth , which contains a series of M synth .

[0066] The general training set is a collection of real data. By adding more synthetic map data to the general training set, an enhanced training set is obtained. This method is used as an enhancement method for general HD map construction. The enhanced training set is expressed as:

[0067] Among them, D real represents the set of real data, D synth Represents a collection of synthetic map data. The enhanced training set contains a series of map prior scenes.

[0068] E. HD map completion task definition: Input a series of map prior scenes into the model to learn to handle different map priors. The set of prior scenes P scenes Expressed as:

[0069] Among them, S n represents the nth prior scene.

[0070] Define a performance evaluation function E that accepts the model output M output and the true label M true Takes as input, and returns a performance score:

[0071] F. Comparison between general model and expert model: Build expert model and general model respectively:

[0072]

[0073] Among them, M expert is the expert model, M generalist is a general model, F represents the model function, and P represents the prior. specific It's P scenes A specific scenario example in P specific Represents the expert model M expert The single scene prior that is relied upon. diverse Represents the general model M generalist The diversified priors relied on, P scenes Is to build P diverse The basic unit set, namely P diverse It is from P scenes The joint priors of multiple scenes are extracted to train a universal model to adapt to cross-scene tasks.

[0074] Through comparative experiments, we found that the general model performs comparable to the expert model across a variety of prior scenarios, and even outperforms it in some cases, namely:

[0075] S perf,expert is the performance score of the expert model. S perf,generalist is the performance score of the general model, which is obtained by M generalist The performance of the general model is evaluated in various scenarios. The performance demonstrates the effectiveness of the synthetic data augmentation method.

[0076] G. Map occlusion enhancement: During training, an occlusion mask can be applied to the input data to simulate real-world occlusion by occluding parts of the map. augmented Contains the original data X and its occluded version X occluded . Input data X after occlusion occluded Expressed as:

[0077] Among them, X represents the original input data, M occluded represents an occlusion mask and ⨀ represents element-wise multiplication.

[0078] The process of training the model is expressed as:

[0079] Among them, D augmented represents the enhanced training set, M represents the model, θ trained Represents the trained model parameters.

[0080]

[0081] M output For the model to occlude the test data X occluded,test prediction results.

[0082] Including occlusion enhancement data during training allows the model to use prior information to build maps even in the presence of occlusion.

[0083] H. Map Prior Query Design: Improve query design to more fully incorporate prior map information. The query function is expressed as:

[0084] Among them, Q optimized Indicates the query result, Q original represents the original query, P represents the prior information, and f optimize Represents an optimization function.

[0085] By optimizing the query Q optimized After locating the key area, the system will give priority to collecting real-time data of the area. real-time, through advanced algorithms and technologies, real-time data D real-time HDM with prior maps prior Fusion to generate high-precision, real-time updated HDM maps fused The fused HDM map fused Expressed as:

[0086] HDM prior Denotes the prior high-definition map, D real-time represents real-time perception data, f fuse represents the fusion function, Q optimized As a weight matrix, the real-time data D real-time Perform weighted fusion.

[0087] A computer program generated based on the above-mentioned super-resolution high-precision map construction method is stored on an electronic device. The electronic device can be a computer, a mobile phone, a single-chip microcomputer, etc., which includes a memory and a processor. The computer program is stored in the memory and can be run by the processor. When the processor executes the computer program, the super-resolution high-precision map construction method is implemented.

[0088] This embodiment also introduces a computer-readable storage medium on which the aforementioned computer program is stored. The computer-readable storage medium may be a USB flash drive, a hard drive, an optical disk, or the like. When executed, the computer program implements the aforementioned method for constructing a super-resolution, high-precision map.

[0089] This embodiment also introduces a computer program product, which, when running on a computer, executes the super-resolution high-precision map construction method as described above.

[0090] The super-resolution high-precision map construction method of this application significantly improves the quality and efficiency, naturalness and realism of image super-resolution reconstruction by integrating multiple advanced technologies such as multi-index entropy modules, dynamic sampling, fidelity trade-offs, synthetic data and occlusion enhancement, and improves the performance and robustness of the model across multiple prior scenarios and occlusion conditions. By improving the map prior query design and integrating prior information with real-time perception data, high-precision, real-time updated high-definition maps are generated to meet the real-time requirements of technologies such as autonomous driving for high-definition maps.

[0091] Taking an actual application scenario as an example, in a complex and changeable road environment, including road widening, new building construction, traffic sign updates, etc., the super-resolution high-precision map construction method based on the change map prior proposed in this application is adopted. Autonomous driving vehicles can obtain low-resolution images in real time through sensors such as on-board cameras, and perform super-resolution reconstruction to generate high-resolution road images. At the same time, the map prior data generated by innovative synthesis methods enriches the training set of autonomous driving vehicles and improves their adaptability to different road environments. In addition, after the introduction of occlusion enhancement technology, autonomous driving vehicles can maintain a high level of navigation accuracy when facing obstructions such as trees and buildings. This application provides reliable navigation support for autonomous driving vehicles, which not only improves the driving safety and comfort of autonomous driving vehicles, but also provides a strong guarantee for their autonomous navigation in complex and changing road environments.

[0092] Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for constructing a super-resolution high-precision map, characterized in that: The following steps are involved: (1) Super-resolution reconstruction of low-resolution images: Input a low-resolution image, extract multi-scale features and calculate the entropy value, divide the image into regions of different complexity according to the entropy value, optimize the number of iterations using a dynamic time step sampling strategy, iteratively refine the image to the target resolution, and at the same time, balance fidelity and realism, adjust the model parameters, and output a high-resolution image; (2) Constructing real-time high-definition maps based on variable map priors: Generate a series of map prior scenarios based on the synthesis method for model training, and introduce occlusion enhancement technology to improve the robustness of the model; finally, improve the map prior query design, integrate prior information with real-time perception data, and generate super-resolution high-precision maps.

2. The method for constructing a super-resolution high-precision map according to claim 1, characterized in that: In step (1), the input low-resolution image, the extraction of multi-scale features and the calculation of entropy values, the division of the image into regions of different complexity according to the entropy values, the optimization of the number of iterations using a dynamic time step sampling strategy, and the iterative refinement of the image to the target resolution specifically include: A low-resolution image of a real-world scene is input, and a deep learning network is used to extract the multi-scale features of the image. The weighted average entropy of the multi-scale features is calculated; the weighted average entropy is converted into a complexity index; based on the complexity index, the image is divided into regions of different complexity; a dynamic time step sampling strategy is used to assign an appropriate number of iterations to each region. The model iteratively refines each region and generates a target resolution image as output; the number of iterations and complexity of each region are linearly positively correlated.

3. The method for constructing a super-resolution high-precision map according to claim 1 or 2, characterized in that: In step (1), balancing fidelity and realism, adjusting model parameters, and outputting high-resolution images specifically include: Design a trade-off function that takes the fidelity score and the realism score as input and outputs a weighed score. The linear relationship is used to perform the trade-off, resulting in the formula: S = λ·P + (1-λ)·R. In this formula, S represents the weighed score, P represents the fidelity score of the target resolution image, R represents the realism score of the target resolution image, and λ is a weight coefficient between 0 and 1. Adjust the model parameters so that the realism score reaches the maximum when the fidelity score is fixed at a value; repeatedly adjust the model parameters so that the fidelity score and the realism score are adjusted synchronously, so that the weighed score reaches the maximum value and the final high-resolution image is output.

4. The method for constructing a super-resolution high-precision map according to claim 1, wherein: In step (2), the generation of a series of map prior scenes based on the synthesis method for training the model specifically includes: A synthesis function is defined that receives diverse scene parameters and synthesizes realistic map data based on these scene parameters. The set of realistic map data is combined with the set of real data to obtain an enhanced training set. The enhanced training set contains a series of prior map scenes. The series of map prior scenes in the enhanced training set are input into the model to train the model.

5. The method for constructing a super-resolution high-precision map according to claim 1, wherein: In step (2), the introduction of occlusion enhancement technology to improve model robustness specifically includes: During training, an occlusion mask is first applied to the input data to block out parts of the map, simulating real-world occlusions. The model is then trained with occlusion-augmented data, allowing it to leverage prior information for map construction even in the presence of occlusions.

6. The method for constructing a super-resolution high-precision map according to claim 1, wherein: In step (2), the improved map query design, the integration of prior information and real-time perception data, and the generation of super-resolution high-precision maps specifically include: Prior information is added to the original query to optimize the query. The key areas of the map are located through the optimized query. The system prioritizes collecting real-time perception data of the key areas and fuses this real-time perception data with the map prior scene to generate a fused super-resolution high-definition map.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for constructing a super-resolution high-precision map as described in any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for constructing a super-resolution high-precision map according to any one of claims 1 to 6 is implemented.

9. A computer program product, characterized in that When the computer program product runs on a computer, the computer executes the super-resolution high-precision map construction method according to any one of claims 1 to 6.

Images