Methods, devices and storage media for generating navigation satellite observation data

Abstract

A method, apparatus, and storage medium for generating navigation satellite observation data are disclosed. The method includes: acquiring visual data to be processed; inputting the visual data to be processed into a scene feature-error prior learning model to determine a first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error features corresponding to the scene features in the visual data to be processed; based on the visual data to be processed, retrieving scene entries matching the visual data to be processed from an association database, wherein the association database stores the mapping relationship between scene features and error prior parameters; correcting the first error prior parameter based on the scene entries to determine a second error prior parameter; and inputting the second error prior parameter and the visual data to be processed into a conditional prediction model to determine the navigation satellite observation data corresponding to the visual data to be processed. This application has the technical effect of improving the generation efficiency of navigation satellite observation data.

Description

Technical Field

[0001] This disclosure relates to the field of global navigation satellite system positioning technology, and in particular to a method, apparatus and storage medium for generating navigation satellite observation data. Background Technology

[0002] Global Navigation Satellite System (GNSS), as a crucial space information infrastructure, is widely used in surveying and mapping, transportation, and autonomous driving. Navigation satellite observation data is essential foundational data for research, algorithm verification, and equipment development related to navigation and positioning technologies. However, current methods for acquiring navigation satellite observation data primarily rely on actual field observations. This requires specialized GNSS receivers and data acquisition terminals, resulting in high observation costs. Furthermore, limitations imposed by geographical environment, meteorological conditions, and specific satellite transit times make it difficult to acquire navigation satellite observation data covering diverse scenarios and operating conditions. Moreover, data acquisition cycles are typically lengthy for special scenarios or extreme conditions. Therefore, there is an urgent need for a method to generate navigation satellite observation data that can reduce field acquisition costs, improve scenario coverage, and enhance data acquisition efficiency. Summary of the Invention

[0003] In view of this, the present disclosure provides a method, apparatus and storage medium for generating navigation satellite observation data, in order to improve the efficiency of generating navigation satellite observation data.

[0004] In a first aspect, a method for generating navigation satellite observation data is provided, comprising: acquiring visual data to be processed; inputting the visual data to be processed into a scene feature-error prior learning model to determine a first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error features corresponding to the scene features in the visual data to be processed; based on the visual data to be processed, retrieving scene entries matching the visual data to be processed from an association database, wherein the association database stores the mapping relationship between scene features and error prior parameters; correcting the first error prior parameter based on the scene entries to determine a second error prior parameter; and inputting the second error prior parameter and the visual data to be processed into a conditional prediction model to determine the navigation satellite observation data corresponding to the visual data to be processed.

[0005] The above method for generating navigation satellite observation data acquires visual data to be processed, determines the first error prior parameter using a scene feature-error prior learning model, corrects the first error prior parameter using an association database to obtain the second error prior parameter, and then outputs navigation satellite observation data corresponding to the visual data to be processed using a conditional prediction model. This method can establish an effective correspondence between the visual scene information acquired by the carrier and the navigation satellite observation data, enabling the generated navigation satellite observation data to better reflect the error characteristics and signal propagation laws under the target scene. At the same time, relying on the characteristics of easy acquisition of visual data and rich scene information, it can cover a variety of complex scenes such as cities, deep mountains, oceans, and indoors, and quickly generate corresponding navigation satellite observation data after inputting visual data of the corresponding scene. This reduces the manpower, material resources, and time costs brought about by complex field observations, and improves the data acquisition efficiency and scene adaptability in the research and development of navigation satellite related technologies and equipment testing.

[0006] Optionally, the scene feature-error prior learning model is constructed based on the following steps: acquiring sample visual data and sample navigation satellite observation data; preprocessing the sample visual data and sample navigation satellite observation data to determine the first model training samples; and training and establishing the scene feature-error prior learning model based on the first model training samples.

[0007] Optionally, the sample visual data and sample navigation satellite observation data are preprocessed, including one or more of the following preprocessing methods: processing based on the data type of the sample visual data, including: when the sample visual data is image data, performing one or more of the following processing methods on the image data: image denoising, image size normalization, image illumination correction, or image distortion correction; when the sample visual data is point cloud data, performing one or more of the following processing methods on the point cloud data: point cloud denoising, point cloud downsampling, or point cloud registration; performing one or more processing methods on the sample visual data: rotation, flipping, scaling, or adding noise; and performing one or more processing methods on the sample navigation satellite observation data: outlier removal, data smoothing, missing value completion, or normalization.

[0008] Optionally, it also includes: aligning the data based on the first timestamp of the preprocessed sample visual data and the second timestamp of the preprocessed sample navigation satellite observation data to form the first model training sample.

[0009] Optionally, the association database is constructed based on the following steps: inputting the training samples of the first model into the trained scene feature-error prior learning model, determining the sample error prior parameters corresponding to the sample visual data; establishing the mapping relationship between the scene features of the sample visual data and the sample error prior parameters; and storing the mapping relationship in the association database.

[0010] Optionally, the conditional prediction model is constructed based on the following steps: extracting features from the sample visual data to determine the sample scene features of the sample visual data; forming a second model training sample based on the sample scene features, sample error prior parameters, and sample navigation satellite observation data; training and establishing the conditional prediction model based on the second model training sample; and further including: when a scene entry matching the visual data to be processed cannot be retrieved in the association database, updating the mapping relationship between the first error prior parameters and the scene features of the video data to be processed to the association database.

[0011] Optionally, the conditional prediction model includes: a visual feature extraction subnetwork, a prior fusion subnetwork, and an output prediction subnetwork; wherein, the visual feature extraction subnetwork is used to extract sample scene features, the prior fusion subnetwork is used to fuse sample scene features with sample error prior parameters to determine the fusion result, and the output prediction subnetwork is used to output navigation satellite observation data corresponding to the sample visual data based on the fusion result.

[0012] Optionally, the prior error parameters include at least one of the following parameters: occlusion degree, multipath intensity index, building density, illumination intensity, noise scale parameter of navigation satellite observations, and error decomposition weights; the navigation satellite observation data include at least one of the following parameters: pseudorange observations, carrier phase observations, Doppler shift observations, and satellite ephemeris.

[0013] In a second aspect, a device for generating navigation satellite observation data is provided, comprising: an acquisition unit for acquiring visual data to be processed; a first determining unit for inputting the visual data to be processed into a scene feature-error prior learning model to determine a first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error feature corresponding to the scene feature in the visual data to be processed; a matching unit for retrieving scene entries matching the visual data to be processed from an association database based on the visual data to be processed, wherein the association database stores the mapping relationship between scene features and error prior parameters; a correction unit for correcting the first error prior parameter based on the scene entries to determine a second error prior parameter; and a second determining unit for inputting the second error prior parameter and the visual data to be processed into a conditional prediction model to determine the navigation satellite observation data corresponding to the visual data to be processed.

[0014] Thirdly, a computer-readable storage medium is provided having instructions stored thereon, which, when executed by a processor, implement the method for generating navigation satellite observation data as provided in the first aspect. Attached Figure Description

[0015] The accompanying drawings used in the description of the embodiments of this disclosure are briefly introduced below: Figure 1A flowchart illustrating a method for generating navigation satellite observation data provided in some embodiments of this application is shown; Figure 2 The diagram illustrates a flowchart of a method for constructing a scene feature-error prior learning model provided in some embodiments of this application. Figure 3 The diagram shows a flowchart illustrating a method for constructing a conditional prediction model according to some embodiments of this application; Figure 4 A schematic diagram of the structure of a navigation satellite observation data generation device provided in some embodiments of this application is shown. Detailed Implementation

[0016] To more clearly illustrate the technical solutions in the embodiments of this disclosure, examples of implementation methods of this disclosure will be described below with reference to the accompanying drawings. The accompanying drawings described below are merely some embodiments of this disclosure. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without creative effort. Adjustments and improvements made without departing from the concept of this disclosure are all within the protection scope of this disclosure.

[0017] To keep the drawings simple, each figure only schematically shows the parts relevant to the embodiment, and they do not represent the actual structure of the product. In addition, for the sake of clarity and ease of understanding, some figures only schematically show parts of components with the same structure or function, and there may actually be more or fewer components with the same structure or function.

[0018] In this disclosure, unless otherwise expressly specified and limited, ordinal numbers, such as “first”, “second”, etc., are used only to distinguish and describe related objects, and should not be construed as indicating or implying the relative importance or order between related objects; furthermore, they do not represent the quantity of related objects. “Multiple” includes two or more, and other quantifiers are similar.

[0019] As a crucial space information infrastructure, the Global Navigation Satellite System (GNSS) is widely used in surveying and mapping, transportation, and autonomous driving, enabling the acquisition of navigation satellite observation data for positioning. Currently, the acquisition of navigation satellite observation data primarily relies on actual field observations, typically requiring the use of high-precision GNSS receivers and related acquisition terminals to obtain navigation satellite observation data within the context of the environment in which the GNSS-equipped system is located. However, this method has significant limitations. On one hand, field observations require specialized equipment and consume considerable manpower, resources, and time, resulting in high overall data acquisition costs. On the other hand, the actual observation process is easily affected by factors such as geographical environment, weather conditions, and specific time windows, making data acquisition difficult. Furthermore, due to limitations imposed by actual acquisition conditions, existing methods struggle to meet the data coverage requirements of different scenarios and operating conditions. For example, in environments such as deep mountains, oceans, and densely populated urban areas with high-rise buildings, or under conditions such as heavy rain, heavy snow, or strong electromagnetic interference, it is often difficult to efficiently acquire navigation satellite observation data for the corresponding scenarios. For some special scenarios or extreme operating conditions, it may also require a long waiting period before specific acquisition conditions can be met before data acquisition can be completed, leading to a prolonged overall data acquisition cycle. Therefore, it is worthwhile to focus on how to reduce reliance on complex field observations, lower data acquisition costs, shorten data acquisition cycles, and improve the coverage of navigation satellite observation data under different scenarios and operating conditions, so as to better meet the basic observation data requirements for navigation and positioning. The following description, in conjunction with the accompanying figures, illustrates this point: Figure 1 The diagram illustrates a flowchart of a method for generating navigation satellite observation data according to some embodiments of this application. This method for generating navigation satellite observation data includes at least the following steps: S110: Acquire visual data to be processed; S120: Input the visual data to be processed into the scene feature-error prior learning model, and determine the first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error feature corresponding to the scene feature in the visual data to be processed; S130: Based on the visual data to be processed, retrieve scene entries that match the visual data to be processed from the relational database, wherein the relational database stores the mapping relationship between scene features and error prior parameters; S140: Based on the scene entries, correct the first error prior parameters and determine the second error prior parameters; S150: Input the second error prior parameter and the visual data to be processed into the conditional prediction model to determine the navigation satellite observation data corresponding to the visual data to be processed.

[0020] The visual data to be processed can be input data containing environmental information of the scene space where the carrier currently carrying the global navigation satellite system is located. For example, the visual data to be processed can be scene images acquired by a camera or point cloud data acquired by a lidar. During the acquisition of visual data, auxiliary information such as the position, attitude, or shooting parameters of the current acquisition device can also be recorded. In practical applications, the visual data to be processed can be input into the system as external environmental representation information of the current target scene. Due to differences in occlusion, reflection, and spatial environment distribution in different scenes, navigation satellite observation data often exhibits different error correlation behaviors in different scenes. This application can process the visual data using a scene feature-error prior learning model to determine the first error prior parameter corresponding to the visual data, thereby linking the scene features in the visual data to be processed with the error correlation attributes that may appear in the navigation satellite observation data. Furthermore, the first error prior parameter can be constrained and corrected using a relational database. The relational database in this application pre-stores the mapping relationship between scene features and error prior parameters. Therefore, after the system retrieves a matching scene entry from the relational database based on the visual data to be processed, it can use the existing mapping relationship reflected by the scene entry to further correct the first error prior parameter, thereby determining the second error prior parameter. The first error prior parameter output by the scene feature-error prior learning model reflects the model's inference result on the current input data, while the scene entries in the relational database reflect the established relationship between historical scene features and error prior parameters. Combining the two makes the obtained second error prior parameter more consistent with the actual error characteristics of the scene corresponding to the current visual data to be processed. Compared with the first error prior parameter, the second error prior parameter better reflects the error characteristics of the scene corresponding to the visual data to be processed, which is beneficial to improving the scene adaptability in the subsequent navigation satellite observation data determination process. After obtaining the second error prior parameter, it can be input into the conditional prediction model along with the visual data to be processed, and the conditional prediction model determines the navigation satellite observation data corresponding to the visual data to be processed. In this process, the conditional prediction model uses the scene environment information represented by the visual data to be processed and the error feature information represented by the second error prior parameter to determine the navigation satellite observation data. The visual data to be processed can provide the scene basis, and the second error prior parameter can provide error constraints. Under the combined effect of the scene basis and error constraints, the conditional prediction model outputs the navigation satellite observation data corresponding to the current scene.

[0021] This application acquires visual data to be processed, determines a first error prior parameter using a scene feature-error prior learning model, corrects the first error prior parameter using an association database to obtain a second error prior parameter, and then outputs navigation satellite observation data corresponding to the visual data to be processed using a conditional prediction model. This establishes an effective correspondence between visual scene information and navigation satellite observation data, enabling the generated navigation satellite observation data to better reflect the error characteristics and signal propagation patterns of the target scene. Furthermore, leveraging the ease of acquisition and rich scene information of visual data, it can cover various complex scenes such as cities, mountains, oceans, and indoor environments. It can also quickly generate corresponding navigation satellite observation data after inputting visual data for the corresponding scene, thereby reducing the manpower, material resources, and time costs associated with complex field observations and improving the data acquisition efficiency and scene adaptability in navigation satellite-related technology research and development and equipment testing.

[0022] Figure 2 The diagram illustrates a flowchart of a method for constructing a scene feature-error prior learning model according to some embodiments of this application. The construction steps include: S210: Acquire sample visual data and sample navigation satellite observation data; S220: Preprocess the sample visual data and sample navigation satellite observation data to determine the first model training samples; S230: Based on the training samples of the first model, train and establish a scene feature-error prior learning model.

[0023] In the above embodiments, to enable the constructed scene feature-error prior learning model to learn the changing patterns of navigation satellite observation error features under different scenarios, sample visual data and corresponding sample navigation satellite observation data can be acquired for learning. Sample visual data can characterize the environmental state under different target scenarios, while sample navigation satellite observation data can be used to characterize the navigation satellite observation results under the corresponding scenarios. By simultaneously introducing sample visual data and sample navigation satellite observation data, the scene feature-error prior learning model can establish a foundation for the correlation between scene information and observation error features during the training phase. Sample visual data can be acquired through image acquisition devices or devices such as LiDAR. For example, high-definition cameras, panoramic cameras, or LiDAR can be used to acquire scene image data or point cloud data in different scenarios such as urban roads, deep mountains, oceans, indoor spaces, or densely populated high-rise areas, and can acquire corresponding scene information under conditions such as sunny days, cloudy days, rainy days, foggy days, daytime, nighttime, or different seasons. To improve the ability of the model to represent scene changes during subsequent training, auxiliary information such as the position, attitude, and shooting parameters of the acquisition device can also be recorded simultaneously when acquiring sample visual data. Correspondingly, the sample navigation satellite observation data can be acquired at the same time and location as the sample visual data through a high-precision navigation satellite receiver, so that the sample visual data and the sample navigation satellite observation data have a corresponding relationship in the scene.

[0024] After acquiring the sample visual data and sample navigation satellite observation data, preprocessing is performed to determine the first model training samples. The purpose of preprocessing is to improve the quality of the original sample data, reduce the impact of noise and outliers on subsequent model training, and make the resulting first model training samples more suitable for training the scene feature-error prior learning model. Alternatively, to ensure a more accurate correspondence between the sample visual data and the sample navigation satellite observation data, temporal alignment can be performed after preprocessing. Thus, when training the scene feature-error prior learning model, the model learns the correspondence between two types of data within the same scene, which helps improve the accuracy of the model in learning scene error prior rules. During training, the scene feature-error prior learning model receives the sample visual data from the first model training samples and combines it with its corresponding sample navigation satellite observation data to learn the changing patterns of navigation satellite observation error characteristics under different scenes, establishing a mapping capability from visual scene information to error prior parameters. The trained scene feature-error prior learning model can output error prior parameters related to the scene corresponding to the input visual data, so as to reflect the error-related attributes that the navigation satellite observation data may present in the current scene.

[0025] In some implementations, the sample visual data and sample navigation satellite observation data are preprocessed, including one or more of the following preprocessing methods: processing based on the data type of the sample visual data, including: when the sample visual data is image data, performing one or more of the following processing methods on the image data: image denoising, image size normalization, image illumination correction, or image distortion correction; when the sample visual data is point cloud data, performing one or more of the following processing methods on the point cloud data: point cloud denoising, point cloud downsampling, or point cloud registration; performing one or more processing methods on the sample visual data: rotation, flipping, scaling, or adding noise; and performing one or more processing methods on the sample navigation satellite observation data: outlier removal, data smoothing, missing value completion, or normalization.

[0026] The purpose of the above preprocessing is to improve the quality of sample visual data and sample navigation satellite observation data, reduce the impact of noise and abnormal data on the model training process, and thus make the resulting first model training samples more suitable for subsequent model training. Since the data types of sample visual data can vary, targeted processing can be performed based on the data type. For example, for image data, image denoising can reduce the interference of acquisition noise on image content, image size normalization can unify the scale of input data, image illumination correction can reduce the impact of different acquisition lighting conditions, and image distortion correction can correct geometric deviations formed during imaging. After the above processing, scene information in the image data can participate in subsequent training in a more stable form. When the sample visual data is point cloud data, one or more of the following processing methods can be performed: point cloud denoising, point cloud downsampling, or point cloud registration. Point cloud denoising can reduce the interference caused by discrete or invalid points on scene representation; point cloud downsampling can reduce data redundancy while retaining the main scene structural information; and point cloud registration can improve the consistency between point cloud data from different viewpoints or at different times. The above processing allows the spatial distribution and scene structure information in point cloud data to participate in training in a form more suitable for model processing. Furthermore, one or more of the following processing methods can be applied to the sample visual data: rotation, flipping, scaling, or adding noise. This essentially constitutes sample augmentation, expanding the variations in the sample visual data and enhancing the model's adaptability to visual scenes under different acquisition angles, scales, and perturbations. This, in turn, improves the model's robustness when processing visual data in subsequent application stages.

[0027] For sample navigation satellite observation data, outlier removal can reduce the interference of obvious abnormal observations on model training results; data smoothing can reduce random fluctuations in the observation data; missing value completion can improve the completeness of sample data; and normalization can weaken the impact of different data volumes on the model training process. By performing one or more of the above processing on the sample navigation satellite observation data, it can be made more suitable to form the first model training samples together with the sample visual data, thereby improving the stability and effectiveness of the scene feature-error prior learning model training process. In this embodiment, one or more of the above preprocessing methods can be selected and combined according to actual application needs to form the first model training samples that meet the training requirements. That is to say, image data processing, point cloud data processing, visual sample enhancement processing, and sample navigation satellite observation data processing can be used individually or in combination, as long as they can optimize the quality of sample visual data and sample navigation satellite observation data and improve the applicability of subsequent model training, they are all optional implementation methods of this application.

[0028] In some implementations, the method further includes: aligning the data based on the first timestamp of the preprocessed sample visual data and the second timestamp of the preprocessed sample navigation satellite observation data to form a first model training sample.

[0029] Since sample visual data and sample navigation satellite observation data typically originate from different data acquisition devices, these devices may differ in sampling time, sampling frequency, and data recording methods. Therefore, data alignment can ensure a temporal correspondence between sample visual data and sample navigation satellite observation data, enabling the subsequent first model training samples to accurately reflect visual and navigation satellite observation information within the same spatiotemporal scene. A first timestamp can be used to characterize the acquisition time of the sample visual data, and a second timestamp can be used to characterize the acquisition time of the sample navigation satellite observation data. By comparing the first and second timestamps, temporally corresponding data items can be selected from the sample visual data and sample navigation satellite observation data, thus forming a temporally consistent sample data set. This ensures that each set of first model training samples corresponds to both visual and navigation satellite observation representations within the same scene. Data alignment can employ timestamp synchronization technology, based on the respective timestamps of the sample visual data and sample navigation satellite observation data. After performing time consistency matching on the two, it can be ensured that each set of samples contains visual information and navigation satellite observation information in the same spatiotemporal scene. This provides a time-consistent training sample basis for the subsequent training of the scene feature-error prior learning model, and improves the accuracy of the scene feature-error prior learning model in learning the correspondence between scene information and navigation satellite observation error features.

[0030] In some implementations, the association database is constructed based on the following steps: inputting the first model training samples into the trained scene feature-error prior learning model, determining the sample error prior parameters corresponding to the sample visual data; establishing a mapping relationship between the scene features of the sample visual data and the sample error prior parameters; storing the mapping relationship in the association database; and further including: when no scene entry matching the visual data to be processed can be retrieved in the association database, updating the mapping relationship between the first error prior parameters and the scene features of the video data to be processed to the association database.

[0031] The association database can be used to store the correspondence between scene features and error prior parameters, thus providing a mapping basis for subsequent retrieval of matching scene entries based on the visual data to be processed and for correcting the error prior parameters. This application can input the training samples of the first model into the trained scene feature-error prior learning model, which then outputs the corresponding sample error prior parameters for the sample visual data. These sample error prior parameters can be parameterized representations of navigation satellite observation error features in the scene corresponding to the sample visual data. By first determining the sample error prior parameters and then establishing their correspondence with the scene features of the sample visual data, the information stored in the association database can be scene-oriented error prior information. After obtaining the sample error prior parameters, a mapping relationship between the scene features of the sample visual data and the sample error prior parameters can be established based on the sample visual data. Scene features can be used to characterize the environmental features of the scene corresponding to the sample visual data, and sample error prior parameters can be used to characterize the navigation satellite observation error features in that scene. The established mapping relationship can reflect the correspondence between different scene features and corresponding error prior parameters. By establishing this mapping relationship, the system can retrieve matching scene entries from the association database based on the scene information represented by the visual data to be processed when it subsequently receives such data. This provides support for correcting the prior error parameters. After establishing the mapping relationship, it is stored in the association database, thus completing the construction of the database. In some implementations, scene features can be represented as scene feature vectors, or the scene feature vectors can be further indexed, and the corresponding sample error prior parameters can be written into the association database as storage content corresponding to the scene features. In this way, in subsequent application stages, the system can retrieve the corresponding scene entries based on the input visual data to be processed and correct the prior error parameters accordingly.

[0032] In addition to storing the prior parameters of sample errors themselves, the relational database can also store statistical information related to these parameters to enhance its reference capability for subsequent scene matching and parameter correction. For example, the relational database can store scene feature vectors or their hash indices as keys and the prior parameters of sample errors and their statistical distribution information as values. In this way, when faced with new visual data to be processed, the system can not only obtain the prior parameters of errors corresponding to the matching scene, but also use the statistical information related to these prior parameters to provide richer evidence for the parameter correction process. Furthermore, the relational database can be updated according to newly added scene data to adapt to the usage requirements under new scene or equipment conditions. Thus, the relational database can not only support scene mapping, ensuring that the prior parameters of errors not only come from model inference results, but can also be further constrained and supplemented by historical scene mapping relationships, further improving the scene correspondence capability in the process of generating navigation satellite observation data.

[0033] In some implementations, the association database can be continuously updated during the generation of navigation satellite observation data to adapt to a wider range of scenarios. When no scene entry matching the visual data to be processed can be found in the association database, the mapping relationship between the first error prior parameter and the scene features of the video data to be processed is updated in the association database. This increases the richness of the various error prior parameters corresponding to multiple scene features in the association database. Furthermore, the updated content of the association database can be reviewed to improve the accuracy of the data processing.

[0034] Figure 3 A flowchart illustrating a method for constructing a conditional prediction model according to some embodiments of this application is shown. The construction method includes: S310: Extract features from the sample visual data to determine the sample scene features of the sample visual data; S320: Based on sample scene features, sample error prior parameters, and sample navigation satellite observation data, a second model training sample is formed; S330: Based on the training samples of the second model, train and establish a conditional prediction model.

[0035] Conditional prediction models can establish a correspondence between scene information, prior error information, and navigation satellite observation data. Construction can begin with sample visual data, first extracting features to determine the sample scene features. These scene features characterize the environmental information within the scene corresponding to the sample visual data, extracting relevant scene information from the original visual data to allow subsequent model training to revolve around this scene representation. After determining the scene features, the scene features, prior error parameters, and corresponding navigation satellite observation data can be organized to form a second model training sample. The prior error parameters reflect the navigation satellite observation error characteristics in that scene, while the navigation satellite observation data characterizes the observation results corresponding to that scene and the prior error parameters. By linking these three elements to form the second model training sample, the conditional prediction model learns the correspondence between scene information, prior error information, and navigation satellite observation data during training. This conditional prediction model can include a visual feature extraction subnetwork, a prior fusion subnetwork, and an output prediction subnetwork. The visual feature extraction subnetwork processes the sample visual data and extracts scene features, transforming scene-related information into feature representations suitable for subsequent computation. The prior fusion subnetwork receives the sample scene features and prior error parameters, fusing them to determine the fusion result. This result can simultaneously contain scene environment information and prior error information, serving as the basis for subsequent navigation satellite observation data output. The output prediction subnetwork outputs navigation satellite observation data corresponding to the sample visual data based on the fusion result. Thus, in the entire conditional prediction model, the visual feature extraction subnetwork extracts scene information, the prior fusion subnetwork jointly models the scene information and prior error information, and the output prediction subnetwork maps the fusion result to navigation satellite observation data. These subnetworks are functionally interconnected, working together to construct and train the conditional prediction model.

[0036] When training a conditional prediction model based on training samples from the second model, sample scene features and prior error parameters can be used as the conditional input, with corresponding sample navigation satellite observation data as the model output. This allows the conditional prediction model to gradually learn the changing patterns of navigation satellite observation data under different scenarios and prior error conditions. After training, the conditional prediction model can receive the visual data to be processed and the second prior error parameters in the application phase, and determine the navigation satellite observation data along the processing link corresponding to the training phase. Specifically, the conditional prediction model can first extract feature information related to the current scene based on the visual data to be processed, then fuse this feature information with the second prior error parameters, and finally output the navigation satellite observation data corresponding to the visual data to be processed. In this way, the navigation satellite observation data output by the conditional prediction model can simultaneously reflect the scene environment features reflected by the visual data to be processed and the navigation satellite observation error features related to that scene.

[0037] The visual feature extraction subnetwork described above can be implemented using convolutional neural networks, Transformers, or their fusion architectures to improve the ability to extract deep scene information from sample visual data. The output prediction subnetwork can be implemented using fully connected layers or recurrent neural networks to improve the output capability from the fusion result to navigation satellite observation data. Through the above network structure settings, the conditional prediction model can have good scene representation capabilities and navigation satellite observation data generation capabilities under scenarios of varying complexity.

[0038] In some implementations, the prior error parameters include at least one of the following parameters: occlusion degree, multipath intensity index, building density, illumination intensity, noise scale parameter of navigation satellite observations, and error decomposition weights; the navigation satellite observation data includes at least one of the following parameters: pseudorange observations, carrier phase observations, Doppler shift observations, and satellite ephemeris.

[0039] Due to variations in occlusion conditions, reflection conditions, spatial distribution, and ambient brightness across different scenarios, the error characteristics exhibited by navigation satellite signals during propagation and reception also change accordingly. The degree of occlusion characterizes the extent to which buildings, trees, mountains, or other obstructions in the target scene block the propagation path of navigation satellite signals; the multipath intensity index characterizes the strength of the multipath effect caused by scene reflective surfaces; building density characterizes the density of the built environment in the target scene; and illumination intensity reflects the ambient brightness conditions during visual scene acquisition, thus aiding in characterizing environmentally relevant factors. By introducing these parameters, prior error parameters can reflect the composition and changing trends of navigation satellite observation error characteristics in the target scene from multiple perspectives.

[0040] Noise scale parameters of navigation satellite observations can be further used to characterize the degree of fluctuation of different observations in the current scenario. For example, noise scale parameters of navigation satellite observations can include pseudorange noise parameters, carrier phase noise parameters, and Doppler noise parameters to reflect the noise level of different observations in the target scenario, respectively. Error decomposition weights can be used to characterize the contribution of different error sources to the navigation satellite observation error in the current scenario. For example, error decomposition weights can characterize the relative contributions of multipath factors, tropospheric factors, and ionospheric factors to the overall error. By introducing noise scale parameters and error decomposition weights of navigation satellite observations, the expression of navigation satellite observation error characteristics by prior error parameters is made more refined, thereby improving the error constraint capability in the subsequent navigation satellite observation data generation process.

[0041] The navigation satellite observation data output by the conditional prediction model can include at least one of the following: pseudorange observations, carrier phase observations, Doppler shift observations, and satellite ephemeris. Pseudorange observations characterize measurements related to the propagation distance of the navigation satellite signal; carrier phase observations characterize measurements related to changes in the carrier phase; Doppler shift observations characterize measurements related to changes in the frequency of the navigation satellite signal; and satellite ephemeris characterizes information related to the satellite's orbit and state. By setting the navigation satellite observation data to at least one of these parameters, the generated navigation satellite observation data can more closely resemble the actual data composition during the navigation satellite observation process, thus meeting the usage requirements of different navigation and positioning algorithms, error analysis methods, and equipment testing processes for different observation quantities. The conditional prediction model can output all parameters from the above-mentioned navigation satellite observation data types, thus forming a relatively complete navigation satellite observation data result; alternatively, it can output only some parameters from pseudorange, carrier phase, Doppler shift, and satellite ephemeris, depending on the actual application requirements. In other words, in this embodiment, the output content of navigation satellite observation data can be flexibly determined according to different usage scenarios. As long as the output results can characterize the navigation satellite observation information required in the target scenario, they can be applied to the technical solution of this application.

[0042] Figure 4A schematic diagram of a navigation satellite observation data generation apparatus provided in some embodiments of this application is shown. The generation apparatus 400 includes: an acquisition unit 410 for acquiring visual data to be processed; a first determination unit 420 for inputting the visual data to be processed into a scene feature-error prior learning model to determine a first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error features corresponding to the scene features in the visual data to be processed; a matching unit 430 for retrieving scene entries matching the visual data to be processed from an association database based on the visual data to be processed, wherein the association database stores the mapping relationship between scene features and error prior parameters; a correction unit 440 for correcting the first error prior parameter based on the scene entries to determine a second error prior parameter; and a second determination unit 450 for inputting the second error prior parameter and the visual data to be processed into a conditional prediction model to determine the navigation satellite observation data corresponding to the visual data to be processed.

[0043] The specific implementation description and determination of the beneficial effects of the above embodiments can be referred to the embodiments of the above-described method for generating navigation satellite observation data, and will not be repeated here.

[0044] Based on the same technical concept, this application also provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, implement the method for generating navigation satellite observation data as provided in the first aspect.

[0045] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not described in detail or in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Furthermore, the above embodiments can be freely combined as needed.

Claims

1. A method for generating navigation satellite observation data, characterized in that, include: Acquire the visual data to be processed; The visual data to be processed is input into the scene feature-error prior learning model to determine the first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error feature corresponding to the scene feature in the visual data to be processed. The scene feature-error prior learning model is constructed based on the following steps: acquiring sample visual data and sample navigation satellite observation data; preprocessing the sample visual data and the sample navigation satellite observation data to determine the first model training samples; and training and establishing the scene feature-error prior learning model based on the first model training samples. Based on the visual data to be processed, scene entries that match the visual data to be processed are retrieved in the association database, wherein the association database stores the mapping relationship between scene features and error prior parameters. The association database is constructed based on the following steps: inputting the training samples of the first model into the trained scene feature-error prior learning model, determining the sample error prior parameters corresponding to the sample visual data; establishing a mapping relationship between the scene features of the sample visual data and the sample error prior parameters; and storing the mapping relationship in the association database. Based on the scenario entries, the first error prior parameter is corrected to determine the second error prior parameter; The second prior error parameter is used in the input condition prediction model of the visual data to be processed to determine the navigation satellite observation data corresponding to the visual data to be processed. The conditional prediction model is constructed based on the following steps: extracting features from the sample visual data to determine the sample scene features of the sample visual data; forming a second model training sample based on the sample scene features, the sample error prior parameters, and the sample navigation satellite observation data; and training and establishing the conditional prediction model based on the second model training sample.

2. The method for generating navigation satellite observation data according to claim 1, characterized in that, The preprocessing of the sample visual data and the sample navigation satellite observation data includes one or more of the following preprocessing methods: Based on the data type of the sample visual data, corresponding processing is performed, including: when the sample visual data is image data, one or more of the following processing is performed on the image data: image denoising, image size normalization, image illumination correction, or image distortion correction; when the sample visual data is point cloud data, one or more of the following processing is performed on the point cloud data: point cloud denoising, point cloud downsampling, or point cloud registration. The sample visual data is subjected to one or more of the following processing methods: rotation, flipping, scaling, or adding noise. The sample navigation satellite observation data is subjected to one or more of the following processing methods: outlier removal, data smoothing, missing value completion, or normalization.

3. The method for generating navigation satellite observation data according to claim 2, characterized in that, Also includes: Data alignment is performed based on the first timestamp of the preprocessed sample visual data and the second timestamp of the preprocessed sample navigation satellite observation data to form the first model training sample.

4. The method for generating navigation satellite observation data according to claim 1, characterized in that, It also includes: when no scene entry matching the visual data to be processed can be found in the association database, updating the mapping relationship between the first error prior parameter and the scene features of the video data to be processed to the association database.

5. The method for generating navigation satellite observation data according to claim 1, characterized in that, The conditional prediction model includes: a visual feature extraction subnetwork, a prior fusion subnetwork, and an output prediction subnetwork; The visual feature extraction subnetwork is used to extract the sample scene features, the prior fusion subnetwork is used to fuse the sample scene features and the sample error prior parameters to determine the fusion result, and the output prediction subnetwork is used to output navigation satellite observation data corresponding to the sample visual data based on the fusion result.

6. The method for generating navigation satellite observation data according to any one of claims 1 to 5, characterized in that, The prior error parameters include at least one of the following parameters: occlusion degree, multipath intensity index, building density, illumination intensity, noise scale parameter of navigation satellite observations, and error decomposition weights; The navigation satellite observation data includes at least one of the following parameters: pseudorange observations, carrier phase observations, Doppler shift observations, and satellite ephemeris.

7. A device for generating navigation satellite observation data, characterized in that, include: The acquisition unit is used to acquire the visual data to be processed. The first determining unit is configured to input the visual data to be processed into a scene feature-error prior learning model, and determine a first error prior parameter corresponding to the visual data to be processed, wherein the first error prior parameter is used to characterize the navigation satellite observation error feature corresponding to the scene feature in the visual data to be processed; wherein the scene feature-error prior learning model is constructed based on the following steps: acquiring sample visual data and sample navigation satellite observation data; preprocessing the sample visual data and the sample navigation satellite observation data to determine a first model training sample; and training and establishing the scene feature-error prior learning model based on the first model training sample. A matching unit is configured to retrieve scene entries that match the visual data to be processed from an association database, wherein the association database stores a mapping relationship between scene features and error prior parameters; wherein the association database is constructed based on the following steps: inputting the first model training samples into the trained scene feature-error prior learning model to determine the sample error prior parameters corresponding to the sample visual data; establishing a mapping relationship between the scene features of the sample visual data and the sample error prior parameters; and storing the mapping relationship in the association database. The correction unit is used to correct the first error prior parameter based on the scene entry and determine the second error prior parameter; The second determining unit inputs the second prior error parameter into the conditional prediction model of the visual data to be processed to determine the navigation satellite observation data corresponding to the visual data to be processed; wherein, the conditional prediction model is constructed based on the following steps: extracting features from the sample visual data to determine the sample scene features of the sample visual data; forming a second model training sample based on the sample scene features, the sample prior error parameter, and the sample navigation satellite observation data; and training and establishing the conditional prediction model based on the second model training sample.

8. A computer-readable storage medium, characterized in that, It stores instructions that, when executed by a processor, implement the method for generating navigation satellite observation data as described in any one of claims 1 to 6.

Images