A navigation scene perception method and device, computer equipment and medium

Abstract

Embodiments of the present application provide a navigation scene perception method and device, computer equipment and medium, wherein the method comprises the following steps: selecting a plurality of features from GNSS signals outdoors and audio signals indoors as signal features, respectively, filtering the signal features to generate preferred signal features; collecting feature values corresponding to the preferred signal features through sensors, respectively, labeling scene labels for the feature values according to actual scenes to generate labeled feature values, and taking the labeled feature values as a dataset of a scene perception model; dividing the processed dataset into a training set and a test set, training a navigation scene perception model using the training set, testing the final scene perception model using the test set, and generating a navigation scene perception result. The scheme combines GNSS signals and audio signals to perceive the navigation scene, thereby improving the accuracy of navigation scene perception.

Description

Technical Field

[0001] This invention relates to the field of navigation technology, and in particular to a method, apparatus, computer device, and medium for sensing navigation scenarios. Background Technology

[0002] Navigation scenarios are broadly defined as the physical space in which a navigation terminal is located and the sum of various factors affecting the reception, generation, transmission, and processing of sensor data. They are constituted by the observation environment and the behavior of the vehicle, and possess various attributes such as segmentation, merging, and connectivity. Under normal conditions, outdoor positioning directly uses GNSS signals; indoor scenarios, GNSS signals are blocked, and positioning is typically achieved using multi-source sensors such as radio signals, visual cameras, or LiDAR. When the vehicle moves between indoor and outdoor environments, or when the positioning area has complex obstruction, it is necessary to switch to an appropriate positioning method based on the scene perception results to ensure positioning performance, primarily the continuity and stability of positioning, and to avoid sudden jumps in positioning. Therefore, the accuracy of indoor and outdoor scene perception directly affects positioning performance and is crucial for achieving seamless indoor / outdoor navigation and positioning, falling under the category of ubiquitous location services. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method for perceiving navigation scenes to solve the technical problem of insufficient accuracy in navigation scene perception in the prior art. The method includes:

[0004] Based on the signal strength of different navigation scenarios, multiple features are selected from outdoor GNSS signals and indoor audio signals as signal features. A heat map reflecting the correlation between signal features is constructed, the importance of signal features is evaluated, and based on the evaluation results and the heat map, the signal features are filtered to generate the preferred signal features.

[0005] The sensor collects the feature values ​​corresponding to the optimized signal features, labels the feature values ​​with scene labels according to the actual scene, generates labeled feature values, and uses the labeled feature values ​​as the dataset of the scene perception model. The scene labels include indoor, outdoor and semi-indoor.

[0006] The dataset of the navigation scene perception model is subjected to feature description and normalization to generate a processed dataset, which is then divided into a training set and a test set.

[0007] A navigation scene perception model is trained using a training set. The fitness of the navigation scene perception model is evaluated based on the training results. If the evaluation results meet the preset conditions, the trained navigation scene perception model is used as the final scene perception model. The final scene perception model is then tested using a test set to generate the perception results of the navigation scene.

[0008] This invention also provides a navigation scene perception device to address the technical problem of insufficient accuracy in navigation scene perception in the prior art. The device includes:

[0009] The feature optimization module is used to select multiple features as signal features from outdoor GNSS signals and indoor audio signals according to the signal strength of different navigation scenarios, construct a heat map that reflects the correlation between signal features, evaluate the importance of signal features, and filter signal features based on the evaluation results and heat map to generate optimized signal features.

[0010] The feature annotation module is used to collect the feature values ​​corresponding to the optimized signal features through sensors, annotate the feature values ​​with scene labels according to the actual scene, generate annotated feature values, and use the annotated feature values ​​as the dataset of the scene perception model. The scene labels include indoor, outdoor and semi-indoor.

[0011] The dataset generation module is used to perform feature description and normalization on the dataset of the navigation scene perception model, generate a processed dataset, and divide the processed dataset into a training set and a test set.

[0012] The perception result acquisition module is used to train the navigation scene perception model using the training set, evaluate the fitness of the navigation scene perception model based on the training results, and if the evaluation results meet the preset conditions, the trained navigation scene perception model is used as the final scene perception model. The final scene perception model is then tested using the test set to generate the perception results of the navigation scene.

[0013] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for perceiving any navigation scenario, thereby solving the technical problem of insufficient accuracy in navigation scenario perception in the prior art.

[0014] This invention also provides a computer-readable storage medium storing a computer program that executes the above-described perception method for any of the navigation scenarios, in order to solve the technical problem of insufficient accuracy in navigation scenario perception in the prior art.

[0015] Compared with the prior art, the beneficial effects that at least one technical solution adopted in the embodiments of this specification can achieve include at least:

[0016] Compared to using only audio signal features or only GNSS signal features to perceive navigation scenes, using a combination of GNSS signal features and audio signal features improves the accuracy of scene perception. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a navigation scene perception method provided in an embodiment of the present invention;

[0019] Figure 2 This is the distribution of GNSS signal feature values ​​in outdoor / semi-indoor / indoor scenarios in the navigation scenario perception method provided in this embodiment of the invention;

[0020] Figure 3 This is the average distribution of audio signal intensity in outdoor / semi-indoor / indoor navigation scenarios in the navigation scenario perception method provided in this embodiment of the invention;

[0021] Figure 4 This is a schematic diagram of feature importance evaluation in the perception method for navigation scenarios provided in this embodiment of the invention;

[0022] Figure 5 This is a heatmap of feature correlation in the perception method for navigation scenarios provided in this embodiment of the invention;

[0023] Figure 6 This is a flowchart of a perception model for implementing the above-described navigation scenario perception method according to an embodiment of the present invention;

[0024] Figure 7 This is a structural block diagram of a computer device provided in an embodiment of the present invention;

[0025] Figure 8 This is a structural block diagram of a navigation scene perception device provided in an embodiment of the present invention. Detailed Implementation

[0026] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0027] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] In this embodiment of the invention, a method for perceiving navigation scenarios is provided, such as... Figure 1 As shown, the method includes:

[0029] Step S101: Based on the signal strength of different navigation scenarios, select multiple features from outdoor GNSS signals and indoor audio signals as signal features, construct a heat map that reflects the correlation between signal features, evaluate the importance of signal features, and filter signal features based on the evaluation results and heat map to generate optimized signal features;

[0030] Step S102: Collect feature values ​​corresponding to the optimized signal features through sensors, label the feature values ​​with scene labels according to the actual scene, generate labeled feature values, and use the labeled feature values ​​as the dataset of the scene perception model. The scene labels include indoor, outdoor and semi-indoor.

[0031] Step S103: Perform feature description and normalization on the dataset of the navigation scene perception model to generate a processed dataset, and divide the processed dataset into a training set and a test set.

[0032] Step S104: Train the navigation scene perception model using the training set, evaluate the fitness of the navigation scene perception model based on the training results, and if the evaluation results meet the preset conditions, use the trained navigation scene perception model as the final scene perception model, test the final scene perception model using the test set, and generate the perception results of the navigation scene.

[0033] In practice, to initially select some features as signal characteristics based on signal strength, the following steps are taken to select multiple features as signal characteristics from outdoor GNSS signals and indoor audio signals, respectively, according to the signal strength of different navigation scenarios:

[0034] During movement in indoor and / or outdoor scenarios, feature observations corresponding to multiple features of GNSS and audio signals are collected. Based on the navigation scenario in which the feature observations are located, scenario labels are assigned to the feature observations, generating labeled feature observations. Based on the frequency of the labeled feature observations of the GNSS signals, a frequency histogram is generated corresponding to the labeled feature observations. Based on the frequency histogram, features corresponding to feature observations with frequency differences in different navigation scenarios are selected as signal features. Based on the mean intensity of the labeled feature observations of the audio signals, features corresponding to feature observations with mean intensity differences in different navigation scenarios are selected as signal features.

[0035] Specifically, the method for selecting GNSS signal features is as follows:

[0036] The GNSS module in a smartphone can provide users with real-time location information. However, during indoor / outdoor movement, signal obstruction and reflection can alter indicators such as the signal-to-noise ratio (SNR), dilution of precision (DOP), and elevation angle. To verify the correlation between GNSS chip-collected data and navigation scenarios, multiple sets of GNSS observation data from smartphones during indoor / outdoor movement were collected. The data were then labeled according to the actual navigation scenario, and frequency histograms were plotted for the number of available satellites, mean SNR, DOP, and mean satellite elevation angle. The results are shown below. Figure 2 As shown. From Figure 2 As can be seen, there are significant differences in the number of available satellites, the average signal-to-noise ratio, PDOP, and the average elevation angle of all satellites at the current epoch across the three navigation scenarios: outdoor, semi-indoor, and indoor.

[0037] Apart from Figure 2 In addition to the four types of features, statistical quantities such as mean, maximum, minimum, and standard deviation can be obtained from the original GNSS observations. Table 1 shows the 12 selected GNSS signal features, where n=5 represents the sliding window size of 5 when calculating the variance.

[0038]

[0039]

[0040] Table 1

[0041] Specifically, the method for selecting audio signal feature values ​​is as follows:

[0042] In indoor environments with deployed audio base stations, smartphones' built-in microphones receive signals from these stations, providing users with real-time location information. When moving between indoor and outdoor scenes, the distance to the deployed base stations changes, causing variations in signal strength. To further verify the correlation between audio signals and navigation scenarios, multiple sets of audio observation data of smartphones moving indoors and outdoors were collected. These data were assigned corresponding scene labels based on the smartphone's actual location, and frequency classification histograms were plotted. Figure 3 The distribution of the mean audio signal intensity is given. From Figure 3 As can be seen, the average audio signal strength of the three base stations differs significantly in the three navigation scenarios: outdoor, semi-indoor, and indoor.

[0043] Table 2 shows three audio feature values ​​selected in one embodiment of the present invention. In one embodiment, four audio base stations were used, three audio signal strengths were received, and the average audio signal strength was calculated. The serial numbers in Table 2 are from Table 1, and a total of 16 GNSS and audio signal feature values ​​were selected. In other embodiments, there may be multiple audio base stations. The number of audio signal features is determined based on the number of audio base stations capable of receiving audio signals.

[0044]

[0045]

[0046] Table 2

[0047] In practice, to further narrow down the range based on the initial feature selection to obtain optimized signal features, the following steps are taken to construct a heatmap reflecting the correlation between signal features, evaluate the importance of signal features, and, based on the evaluation results and the heatmap, filter the signal features to generate optimized signal features:

[0048] Signal features are paired to generate signal feature groups. Based on the correlation between signal features in a signal feature group, the heat value of one signal feature relative to another is determined, and a heat map is constructed using these heat values. A heat value threshold is set, and the heat values ​​corresponding to the signal features in the signal feature group are obtained from the heat map. If the heat value is greater than or equal to the heat value threshold, the signal feature is assigned to the first feature group; if the heat value is less than the heat value threshold, the signal feature is assigned to the second feature group. The average reduction accuracy and average Gini index corresponding to the feature value are used as feature importance evaluation indicators, and an evaluation indicator threshold is set for the feature importance evaluation indicators. The comprehensiveness of each signal feature in the first feature group is calculated, and the feature value corresponding to the signal feature in the first feature group is compared with the evaluation indicator threshold. The signal feature with the highest comprehensiveness and a signal feature value greater than the evaluation indicator threshold is selected as the preferred signal feature, where comprehensiveness is the degree to which the signal feature covers other signal features. The feature values ​​of the signal features in the second feature group are compared with the evaluation indicator threshold, and the signal feature with a feature value greater than the evaluation indicator threshold is selected as the preferred signal feature.

[0049] In specific implementation, in order to further narrow down the range of feature selection based on the first feature selection to obtain optimized signal features, the optimized signal features are achieved through the following steps:

[0050] The optimized GNSS signal characteristics and optimized audio signal characteristics are as follows: the optimized GNSS signal characteristics include: mean signal-to-noise ratio (SNR), maximum SNR, minimum SNR, sum of SNR, moving variance of SNR, number of visible satellites, geometric accuracy attenuation factor, position accuracy attenuation factor, horizontal accuracy attenuation factor, vertical accuracy attenuation factor, moving variance of position accuracy attenuation factor, and mean elevation angle of all satellites; the optimized audio signal characteristics include: signal strength of audio base stations and mean audio signal strength, wherein there may be multiple audio base stations.

[0051] Specifically, a total of 16 feature values ​​were selected by combining the 12 GNSS signal features in Table 1 and the 4 audio signal features in Table 2. A rich set of feature values ​​is helpful for accurate perception in indoor and outdoor navigation scenarios; however, for mobile applications, too many feature values ​​can lead to information redundancy, reduced computational efficiency, and increased power consumption. Therefore, optimized selection of feature values ​​is particularly important.

[0052] As an indicator for assessing the importance of features, Figure 4 The distributions of two indicators, average reduction accuracy and average Gini index decline, are presented. Figure 4 The features represented by 1 to 16 are shown in Tables 1 and 2. From... Figure 4As can be seen, the values ​​of both are positively correlated with the importance of their corresponding features; compared with other features, SNR_mean, SNR_sum, Angle_mean, and signal strength_mean are more important to the model, while SNR_min and HDOP are relatively less important to the model.

[0053] Furthermore, Figure 5 The correlations between features are displayed using a heatmap. Figure 5 It can be seen that there is a high degree of correlation between the five features SNR_mean, SNR_max, SNR_sum, SNR_var, and NAST; among the four features GDOP, PDOP, HDOP, and VDOP; and between signal strength 1 and signal strength 2. There is no strong correlation between PDOP_var, Angle_mean, and the other features.

[0054] Combination Figure 4 and Figure 5 The result shows that:

[0055] ①The five features SNR_mean, SNR_max, SNR_sum, SNR_var, and NAST are similar. The first three features belong to the same category, and SNR_sum has the lowest importance among them, so it is removed.

[0056] ②The four features GDOP, PDOP, HDOP, and VDOP are highly correlated. GDOP is a combination of the other three features, and PDOP is a combination of HDOP and VDOP. Since PDOP, HDOP, and VDOP are less important, they are removed.

[0057] ③Angle_mean has no strong correlation with the other features, but its importance is relatively high. It is an important feature that describes indoor / outdoor conditions from different dimensions, so it is retained.

[0058] Through eigenvalue optimization, one embodiment of the present invention selects a total of 12 eigenvalues ​​(using 3 usable audio base stations), namely, mean signal-to-noise ratio (SNR), maximum SNR, minimum SNR, sum of SNR, moving variance of SNR (n=5), number of visible satellites, geometric accuracy attenuation factor, position accuracy attenuation factor, horizontal accuracy attenuation factor, vertical accuracy attenuation factor, moving variance of position accuracy attenuation factor (n=5), mean elevation angle of all satellites, signal strength of audio base stations 1, 2, and 3, and mean audio signal strength. In other embodiments, if multiple audio base stations are available, the signal strengths of multiple audio base stations are acquired and the mean audio signal strength of multiple audio base stations is calculated.

[0059] Specifically, the navigation scene classification model is an indoor / outdoor scene classification technology, which is used to identify whether the environment in which the navigation terminal is located is indoor, outdoor, or semi-indoor.

[0060] In practice, to adjust the ratio of training and test sets and improve the accuracy and precision of the navigation scene perception model, the processed dataset is divided into training and test sets through the following steps:

[0061] Set a fixed ratio between the test set and the training set; randomly select a fixed ratio of data from the dataset to generate the test set; randomly select a fixed ratio of data from the dataset to generate the training set.

[0062] In specific implementation, in order to predict the classification of navigation scenes as the perception result through the navigation scene perception model (classification model), the following steps are taken: The navigation scene perception model is trained using a training set; the fitness of the navigation scene perception model is evaluated based on the training results; if the evaluation results meet preset conditions, the trained navigation scene perception model is used as the final scene perception model; the final scene perception model is tested using a test set to generate the perception result of the navigation scene:

[0063] An optimal solution search algorithm is constructed to generate the optimal parameter combination. This optimal parameter combination is used as the input parameters for the navigation scene perception model. The input parameters include the number of decision trees and the forest depth. Multiple subsets are extracted from the training set using a resampling method. Each subset corresponds to a scene label, and the number of samples in each subset is the same. The navigation scene perception model is trained using these subsets, and its fitness is evaluated based on prediction error and correlation coefficient. Preset conditions are established. If the fitness evaluation results meet the preset conditions, the navigation scene perception model is adopted as the final scene perception model. If the fitness evaluation results do not meet the preset conditions, the optimal solution search algorithm is used to regenerate the input parameter combination, and the navigation scene perception model is retrained until the evaluation results meet the preset conditions. The final scene perception model is tested using a test set to generate the perception results of the navigation scene, where the perception results indicate that the terminal's environment is indoor and / or outdoor and / or semi-indoor.

[0064] Specifically, each input parameter combination is applied to the navigation scene perception model, and the fitness is calculated based on the prediction results. The fitness can be evaluated based on indicators such as prediction error and correlation coefficient.

[0065] In practice, to iteratively optimize the parameters of the navigation scene perception model using the optimal solution search algorithm, reduce the risk of overfitting, and further improve the robustness and generalization ability of the navigation scene perception model, the following steps are taken to construct the optimal solution search algorithm and generate parameter combinations using the optimal solution search algorithm:

[0066] A set of candidate solutions is randomly generated, where the candidate solutions include the input parameters of the navigation scene perception model; the average accuracy of the validated navigation scene perception model is used as the fitness function, and the fitness function value of each candidate solution is calculated through the fitness function. The candidate solution with the highest fitness function value is obtained as the global optimal solution; a spiral search model is constructed, and the global optimal solution and fitness function value are iterated and updated through the spiral search model until the maximum number of iterations or the stopping condition is met. The final global optimal solution is then used as the parameter combination of the navigation scene perception model.

[0067] In one embodiment, the process of generating prediction results through a navigation scene perception model is as follows: Figure 6 As shown, it mainly consists of the following 5 steps:

[0068] 1) Assign scene labels to the collected navigation scene features according to the actual situation to create a dataset. After feature description and normalization, use the hold-out method to divide the test set and training set into 30% and 70% (the ratio can be set).

[0069] 2) Initialize the parameters of the search optimal solution algorithm and the navigation scene perception model respectively.

[0070] 3) Calculate the fitness function value of each individual using the average accuracy of the validated model as the fitness function, confirm the best search agent of the navigation scene perception model, iteratively update the position and evaluate the fitness, and record the fitness function value of the best search agent of each generation until the preset conditions are met.

[0071] 4) Build a navigation scene perception model by outputting the optimal parameter combination, and test it using a test set.

[0072] 5) Evaluate the model's performance and output the prediction results.

[0073] Specifically, because the decision trees in navigation scene perception models (random forest algorithm) adopt a single classifier decision-making mode, they are susceptible to local extrema during training, have difficulty adjusting the parameters of the optimal solution search algorithm, and are prone to overfitting. The Whale Optimization Algorithm (optimal solution search algorithm) is a heuristic algorithm based on the behavior of whale groups in nature, searching for the optimal solution through cooperation and competition. Using the optimal solution search algorithm, the two parameters n_trees and n_layers (number of decision trees and forest depth) can be iteratively optimized, reducing the risk of overfitting and further improving the model's robustness and generalization ability.

[0074] The search-for-optimal-solution algorithm is optimized using the whale optimization algorithm (search-for-optimal-solution algorithm) to build a navigation scene-aware model, including:

[0075] (1) Encirclement and prey

[0076] Each whale represents a possible optimal solution for n_trees and n_layers, and its position represents a value. The prey position is the parameter combination with the best fitness value in the current iteration. Therefore, the position of the i-th whale in a population of N in 2D space is represented as X. i =(X i1 ,X i2 The mathematical model can be expressed as:

[0077] X(t+1)=X p (t)-A×D;

[0078] D = |C × X p (t)-X(t)|.

[0079] Where D is the distance, t represents the number of iterations, and X(t+1) represents the optimal solution for n_trees and n_layers after iterations. p (t) represents the current optimal solution, and A and C are coefficient vectors.

[0080] (2) Spiral attack

[0081] The mathematical model simulating spiral motion can be expressed as:

[0082] X(t+1)=e bl ×cos(2πl)×D+X p (t);

[0083] Where b is a constant and l is a random number between -1 and 1.

[0084] (3) Hunting

[0085] When the coefficient vector |A|>1, the prey search phase begins, and the search model is established as follows:

[0086] X(t + 1)=X rand (t)-A×D;

[0087] D = |C×X rand (t)-X(t)|。

[0088] Where, X rand represents any value of n_trees and n_layers in the search space.

[0089] The accuracy of the navigation scene perception model is affected by two parameters, n_trees and n_layers. To obtain the best parameter combination, the objective function is set as:

[0090]

[0091] Where, t ∈ [t min , t max , l ∈ [l min , l max . t and l represent n_trees and n_layers (the number of decision trees and the depth of the forest) respectively. [t min , t max and [l min , l max are the upper and lower limits of t and l respectively. n represents the total number of samples in the training set. x(i) and represent the true value and the predicted value of the i-th scene feature data collected respectively.

[0092] (4) Bootstrap resampling

[0093] Each sample in the training set is represented as F i [ =(F 1i , F 2i , F[[ID=5,4]] 3i , …, F 12i ), i ∈ [1, N], N is the number of samples, and F is the optimized GNSS and audio feature values. Using the Bootstrap method, randomly and with replacement, l (l < T) sample subsets are drawn from the training sample set T = {(F1, y1), (F2, y2), …, (F N , y N )} (y i = 1, 2, 3) for training decision trees, and ensure that the number of samples in each subset is the same, represented as: {T1, T2, … T l}. Where y i = 1 represents an outdoor scene, y i = 2 represents a semi-indoor scene, and y i = 3 represents an indoor scene.

[0094] (5) Decision tree construction

[0095] Based on the optimal parameters obtained in step (3), t decision trees with s layers are built to form a random forest.

[0096] (6) Soft voting

[0097] For the input test sample set, each decision tree makes a "soft" prediction, calculating the probability of belonging to each navigation scenario. Then, the predicted probabilities y for all decision trees are calculated. n (F) Take the average value to obtain the probability Y(F) of the random forest output:

[0098]

[0099] Where t is the number of decision trees, and n∈[1,t].

[0100] In other embodiments, other classification models may also be used to perceive the navigation scene. For example, (Naive) Bayes, decision trees, logistic regression, KNN, SVM, neural networks, etc.

[0101] In this embodiment, a computer device is provided, such as... Figure 7 As shown, it includes a memory 701, a processor 702, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the perception method for any of the above-described navigation scenarios.

[0102] Specifically, the computer device can be a computer terminal, a server, or a similar computing device.

[0103] In this embodiment, a computer-readable storage medium is provided, which stores a computer program that performs the perception method for any of the above-described navigation scenarios.

[0104] Specifically, computer-readable storage media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media does not include transient media, such as modulated data signals and carrier waves.

[0105] Based on the same inventive concept, this invention also provides a navigation scene sensing device, as described in the following embodiments. Since the principle of the navigation scene sensing device in solving the problem is similar to that of the navigation scene sensing method, the implementation of the navigation scene sensing device can refer to the implementation of the navigation scene sensing method, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0106] Figure 8 This is a structural block diagram of a navigation scene perception device according to an embodiment of the present invention, such as... Figure 8 As shown, it includes: a feature optimization module 801, a feature annotation module 802, a dataset generation module 803, and a perception result acquisition module 804. The structure is described below.

[0107] The feature selection module 801 is used to select multiple features as signal features from outdoor GNSS signals and indoor audio signals according to the signal strength of different navigation scenarios, construct a heat map that reflects the correlation between signal features, evaluate the feature importance of signal features, and filter signal features based on the evaluation results and heat map to generate optimized signal features.

[0108] The feature annotation module 802 is used to collect the feature values ​​corresponding to the optimized signal features through sensors, annotate the feature values ​​with scene labels according to the actual scene, generate annotated feature values, and use the annotated feature values ​​as the dataset of the scene perception model. The scene labels include indoor, outdoor and semi-indoor.

[0109] The dataset generation module 803 is used to perform feature description and normalization processing on the dataset of the navigation scene perception model, generate a processed dataset, and divide the processed dataset into a training set and a test set.

[0110] The perception result acquisition module 804 is used to train the navigation scene perception model using the training set, evaluate the fitness of the navigation scene perception model based on the training results, and if the evaluation results meet the preset conditions, use the trained navigation scene perception model as the final scene perception model, test the final scene perception model using the test set, and generate the perception results of the navigation scene.

[0111] In one embodiment, the feature selection module includes:

[0112] The observation acquisition unit is used to acquire feature observations corresponding to multiple features of GNSS signals and audio signals during movement in indoor and / or outdoor scenarios.

[0113] The labeled feature observation unit is used to label the feature observation with scene tags based on the navigation scene in which the feature observation is located, and generate labeled feature observations;

[0114] The GNSS signal feature generation unit is used to generate a frequency histogram corresponding to the labeled feature observations based on the frequency of the labeled feature observations of the GNSS signal, and select the features corresponding to the feature observations with different frequencies in different navigation scenarios as signal features based on the frequency histogram.

[0115] The audio signal feature generation unit is used to select features corresponding to feature observations with different intensity mean values ​​in different navigation scenarios as signal features based on the intensity mean of the labeled feature observations of the audio signal.

[0116] In one embodiment, the feature selection module further includes:

[0117] The heatmap construction unit is used to combine signal features in pairs to generate signal feature groups. Based on the correlation between signal features in the signal feature group, the heat value of one signal feature relative to another in the signal feature group is determined, and a heatmap is constructed based on the heat value.

[0118] The feature grouping unit is used to set the thermal value threshold, obtain the thermal value corresponding to the signal feature in the signal feature group through the thermal map, and if the thermal value is greater than or equal to the thermal value threshold, the signal feature is assigned to the first feature group; if the thermal value is less than the thermal value threshold, the signal feature is assigned to the second feature group.

[0119] The indicator threshold setting unit is used to set the threshold of the indicator corresponding to the feature importance assessment indicator by taking the average reduction accuracy and average decline Gini index corresponding to the feature value as the feature importance assessment indicator.

[0120] The feature optimization unit is used to calculate the comprehensiveness of each signal feature in the first feature group, compare the feature value corresponding to the signal feature in the first feature group with the evaluation index threshold, and select the signal feature with the highest comprehensiveness and the signal feature value greater than the evaluation index threshold as the optimized signal feature. Here, the comprehensiveness is the degree to which the signal feature covers other signal features.

[0121] The preferred feature selection unit is used to compare the feature values ​​of signal features in the second feature group with the evaluation index threshold, and select the signal features corresponding to feature values ​​greater than the evaluation index threshold as preferred signal features.

[0122] In one embodiment, a preferred feature selection unit is used to select the preferred signal features, including: selected GNSS signal features and selected audio signal features; the selected GNSS signal features include: mean signal-to-noise ratio (SNR), maximum SNR, minimum SNR, sum of SNR, moving variance of SNR, number of visible satellites, geometric accuracy attenuation factor, position accuracy attenuation factor, horizontal accuracy attenuation factor, vertical accuracy attenuation factor, moving variance of position accuracy attenuation factor, and mean elevation angle of all satellites; the selected audio signal features include: signal strength of audio base stations and mean audio signal strength, wherein there may be multiple audio base stations.

[0123] In one embodiment, the dataset generation module further includes:

[0124] The fixed ratio setting unit is used to set a fixed ratio between the test set and the training set;

[0125] The test set generation unit is used to randomly select a fixed proportion of data from the dataset to generate a test set;

[0126] The training set generation unit is used to randomly select a fixed proportion of data from the dataset to generate a training set.

[0127] In one embodiment, the module for obtaining perception results includes:

[0128] The model building unit is used to build an algorithm for searching for the optimal solution. The algorithm generates the optimal parameter combination and uses the optimal parameter combination as the input parameters of the navigation scene perception model. The navigation scene perception model is built using the input parameters, which include the number of decision trees and the forest depth.

[0129] The subsample generation unit is used to extract multiple subsample sets from the training set through a resampling method. Each subsample set corresponds to a scene label, and the number of samples in multiple subsample sets is the same.

[0130] The fitness evaluation unit is used to train the navigation scene perception model using a subset of samples and to evaluate the fitness of the navigation scene perception model using prediction error and correlation coefficient.

[0131] The model unit is determined to set preset conditions. If the fitness evaluation result meets the preset conditions, the navigation scene perception model is used as the final scene perception model.

[0132] The model reconstruction unit is used to regenerate the input parameter combination through the search for the optimal solution algorithm if the fitness evaluation result does not meet the preset conditions, and retrain the navigation scene perception model until the evaluation result meets the preset conditions.

[0133] The perception result generation unit is used to test the final scene perception model using a test set and generate perception results for the navigation scene, wherein the perception results indicate that the environment in which the terminal is located is indoor and / or outdoor and / or semi-indoor.

[0134] In one embodiment, the model building unit is used to randomly generate a set of candidate solutions, wherein the candidate solutions include input parameters in the navigation scene perception model; using the average accuracy of the validated navigation scene perception model as the fitness function, the fitness function value of each candidate solution is calculated through the fitness function, and the candidate solution with the highest fitness function value is obtained as the global optimal solution; a spiral search model is constructed, and the global optimal solution and fitness function value are iterated and updated through the spiral search model until the maximum number of iterations or the stopping condition is met, and the final global optimal solution is used as the parameter combination of the navigation scene perception model.

[0135] The embodiments of the present invention achieve the following technical effects:

[0136] Compared to using only audio signal features or only GNSS signal features for navigation scene perception, the combination of GNSS and audio signal features improves the accuracy of scene perception. Using a classification model combined with an optimal solution search algorithm, the model's accuracy, precision, recall, and F1 score all exceeded 96%, with differences between the four metrics not exceeding 1%. A difference analysis was conducted on GNSS and audio signal features, selecting 12 GNSS signal features and 4 audio signal features. The importance of each feature in model construction and the correlation between features were analyzed using two indicators: average reduction in precision and average decrease in Gini index, as well as a feature correlation heatmap. Features with low importance to model construction and excessively high correlations were eliminated. Ultimately, 12 signal features were selected. Reducing the number of features shortened the computation time, resulting in a 25% speed improvement compared to before feature optimization.

[0137] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular combination of hardware and software.

[0138] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for perceiving a navigation scene, characterized in that, include: Based on the signal strength of different navigation scenarios, multiple features are selected from outdoor GNSS signals and indoor audio signals as signal features. A heat map reflecting the correlation between the signal features is constructed. The importance of the signal features is evaluated. Based on the evaluation results and the heat map, the signal features are filtered to generate the preferred signal features. The feature values ​​corresponding to the optimized signal features are collected by sensors respectively, and the feature values ​​are labeled with scene tags according to the actual scene to generate labeled feature values. The labeled feature values ​​are used as the dataset of the navigation scene perception model. The scene tags include indoor, outdoor and semi-indoor. The dataset of the navigation scene perception model is subjected to feature description and normalization to generate a processed dataset, which is then divided into a training set and a test set. The navigation scene perception model is trained using the training set. The fitness of the navigation scene perception model is evaluated based on the training results. If the evaluation results meet the preset conditions, the trained navigation scene perception model is used as the final scene perception model. The final scene perception model is tested using the test set to generate the perception results of the navigation scene.

2. The navigation scene perception method as described in claim 1, characterized in that, Based on the signal strength in different navigation scenarios, multiple features are selected as signal features from both outdoor GNSS signals and indoor audio signals, including: During movement in indoor and / or outdoor scenarios, feature observations corresponding to multiple features of the GNSS signal and the audio signal are collected; Based on the navigation scene in which the feature observation is located, a scene label is assigned to the feature observation, and labeled feature observations are generated. Based on the frequency of the labeled feature observations of the GNSS signal, a frequency histogram corresponding to the labeled feature observations is generated. Based on the frequency histogram, features corresponding to feature observations with different frequencies in different navigation scenarios are selected as signal features. Based on the mean intensity of the labeled feature observations of the audio signal, the features corresponding to the feature observations whose mean intensity differs in different navigation scenarios are selected as signal features.

3. The navigation scene perception method as described in claim 1, characterized in that, A heatmap reflecting the correlation between the signal features is constructed; the feature importance of the signal features is evaluated; and based on the evaluation results and the heatmap, the signal features are filtered to generate preferred signal features, including: The signal features are combined in pairs to generate signal feature groups. Based on the correlation between the signal features in the signal feature groups, the thermal value of one signal feature relative to another in the signal feature groups is determined, and a heat map is constructed using the thermal value. A heat value threshold is set, and the heat value corresponding to the signal feature in the signal feature group is obtained through the heat map. If the heat value is greater than or equal to the heat value threshold, the signal feature is assigned to the first feature group; if the heat value is less than the heat value threshold, the signal feature is assigned to the second feature group. The average reduction accuracy and average Gini index corresponding to the feature value are used as feature importance evaluation indicators, and the evaluation indicator threshold corresponding to the feature importance evaluation indicators is set. Calculate the integration degree of each signal feature in the first feature group, compare the feature value corresponding to the signal feature in the first feature group with the evaluation index threshold, and select the signal feature with the highest integration degree and the signal feature value greater than the evaluation index threshold as the preferred signal feature, wherein the integration degree is the degree to which the signal feature covers other signal features; Compare the feature values ​​of the signal features in the second feature group with the evaluation index threshold, and select the signal features corresponding to the feature values ​​that are greater than the evaluation index threshold as the preferred signal features.

4. The navigation scene perception method as described in claim 3, characterized in that, The optimized signal features include: Optimized GNSS signal characteristics and optimized audio signal characteristics; The optimized GNSS signal features include: mean signal-to-noise ratio (SNR), maximum SNR, minimum SNR, sum of SNR, moving variance of SNR, number of visible satellites, geometric accuracy attenuation factor, position accuracy attenuation factor, horizontal accuracy attenuation factor, vertical accuracy attenuation factor, moving variance of position accuracy attenuation factor, and mean elevation angle of all satellites. The preferred audio signal characteristics include: the signal strength of the audio base station and the average audio signal strength, wherein there may be multiple audio base stations.

5. The navigation scene perception method as described in claim 1, characterized in that, The processed dataset is divided into a training set and a test set, including: Set a fixed ratio between the test set and the training set; The test set is generated by randomly selecting a fixed proportion of data from the dataset. The training set is generated by randomly selecting a fixed proportion of data from the dataset.

6. The method for perceiving a navigation scene as described in any one of claims 1 to 5, characterized in that, The navigation scene perception model is trained using the training set. The fitness of the navigation scene perception model is evaluated based on the training results. If the evaluation results meet preset conditions, the trained navigation scene perception model is used as the final scene perception model. The final scene perception model is then tested using a test set to generate navigation scene perception results, including: An optimal solution search algorithm is constructed, and an optimal parameter combination is generated through the optimal solution search algorithm. The optimal parameter combination is used as the input parameters of the navigation scene perception model, and the navigation scene perception model is constructed using the input parameters. The input parameters include the number of decision trees and the forest depth. Multiple sub-sample sets are extracted from the training set by a resampling method, wherein each sub-sample set corresponds to a scene label, and the number of samples in the multiple sub-sample sets is the same; The navigation scene perception model is trained using the subsample set, and the fitness of the navigation scene perception model is evaluated using prediction error and correlation coefficient. If the fitness evaluation result meets the preset conditions, the navigation scene perception model will be used as the final scene perception model. If the fitness evaluation result does not meet the preset conditions, the input parameter combination is regenerated by the search for the optimal solution algorithm, and the navigation scene perception model is retrained until the evaluation result meets the preset conditions. The final scene perception model is tested using the test set to generate a perception result for the navigation scene, wherein the perception result indicates that the environment in which the terminal is located is indoor and / or outdoor and / or semi-indoor.

7. The navigation scene perception method as described in claim 6, characterized in that, Constructing an algorithm to search for the optimal solution, and generating parameter combinations through the algorithm, including: A set of candidate solutions is randomly generated, wherein the candidate solutions include the input parameters in the navigation scene perception model; The average accuracy of the verified navigation scene perception model is used as the fitness function. The fitness function value of each candidate solution is calculated using the fitness function, and the candidate solution with the highest fitness function value is obtained as the global optimal solution. A spiral search model is constructed, and the global optimal solution and the fitness function value are iterated and updated through the spiral search model until the maximum number of iterations or the stopping condition is met. The final global optimal solution is then used as the parameter combination of the navigation scene perception model.

8. A sensing device for a navigation scene, characterized in that, include: The feature optimization module is used to select multiple features as signal features from outdoor GNSS signals and indoor audio signals according to the signal strength of different navigation scenarios, construct a heat map reflecting the correlation between the signal features, evaluate the feature importance of the signal features, and filter the signal features based on the evaluation results and the heat map to generate optimized signal features. The feature annotation module is used to collect feature values ​​corresponding to the optimized signal features through sensors, annotate the feature values ​​with scene labels according to the actual scene, generate annotated feature values, and use the annotated feature values ​​as the dataset of the navigation scene perception model. The scene labels include indoor, outdoor and semi-indoor. The dataset generation module is used to perform feature description and normalization processing on the dataset of the navigation scene perception model to generate a processed dataset, and divide the processed dataset into a training set and a test set. The perception result acquisition module is used to train the navigation scene perception model using the training set, evaluate the fitness of the navigation scene perception model based on the training results, and if the evaluation results meet the preset conditions, use the trained navigation scene perception model as the final scene perception model, test the final scene perception model using the test set, and generate the perception results of the navigation scene.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the navigation scene perception method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that performs the perception method of the navigation scene according to any one of claims 1 to 7.

Images