Digital key positioning methods, devices, vehicles and storage media

Abstract

This application provides a digital key positioning method, device, vehicle, and storage medium. The method includes: acquiring multi-anchor point ranging values ​​corresponding to the digital key; wherein the multi-anchor point ranging values ​​include ranging values ​​from the digital key to different ranging modules on the vehicle; calling a first model to identify the multi-anchor point ranging values ​​and determining a first identification result; if the first identification result indicates that the digital key is located inside the vehicle, calling a second model and a third model to identify the multi-anchor point ranging values; wherein the second model is trained based on a first training set, and the third model is trained based on a second training set, the second training set including training data that resulted in identification errors during the training of the second model; and determining the position of the digital key based on the identification results output by the second model and the third model. This method can more accurately determine the position of the digital key in complex environments, effectively improving positioning accuracy and robustness.

Description

Technical Field

[0001] This application relates to digital key positioning technology, and more particularly to a digital key positioning method, device, vehicle, and storage medium. Background Technology

[0002] Digital key technology is an important component of the smart car industry, aiming to enable keyless entry (PE) and keyless start (PS) functions through wireless communication and identity authentication. This technology relies on the accurate identification of the digital key's location to determine whether it is inside or outside the vehicle, thereby deciding whether to execute the PE or PS function.

[0003] Currently, the common method for vehicles to identify the location of digital keys is to measure the distance between the digital key and the vehicle, calculate the spatial position of the digital key based on the distance, and then identify whether the digital key is inside or outside the vehicle based on the spatial position of the digital key.

[0004] However, in practical applications, signal obstruction and reflection caused by the vehicle body structure can easily lead to significant deviations in the spatial position of the digital key, resulting in errors in the vehicle's positioning of the digital key. Summary of the Invention

[0005] This application provides a digital key positioning method, apparatus, vehicle, and storage medium. It enables more accurate determination of the digital key's location in complex environments, effectively improving positioning accuracy and robustness.

[0006] The technical solution of this application embodiment is implemented as follows:

[0007] This application provides a digital key positioning method, the method comprising:

[0008] Obtain the multi-anchor point distance measurement value corresponding to the digital key; wherein, the multi-anchor point distance measurement value includes the distance measurement value from the digital key to different distance measurement modules on the vehicle;

[0009] The first model is invoked to identify the distance measurement values ​​of multiple anchor points, and the first identification result is determined;

[0010] If the first recognition result indicates that the digital key is inside the vehicle, the second and third models are invoked to identify the multi-anchor distance values. The second model is trained based on the first training set, and the third model is trained based on the second training set. The second training set includes training data that resulted in recognition errors during the training of the second model. The third model is used to verify the recognition results of the second model.

[0011] The location of the digital key is determined based on the recognition results output by the second and third models.

[0012] Based on the aforementioned technical methods, the method first acquires multi-anchor point ranging values ​​corresponding to the digital key to construct input features, improving the comprehensiveness of the positioning information. Secondly, it performs coarse positioning by calling a first model and executes a decision-making mechanism based on the coarse positioning results. When the first recognition result output by the first model indicates that the digital key is outside the vehicle, the location of the digital key is determined to be outside the vehicle. When the first recognition result output by the first model indicates that the digital key is inside the vehicle, a second and third model are further called for secondary recognition. The second model is trained based on the first training set, while the third model is optimized based on data misjudged by the second model, thereby improving the third model's ability to compensate for the second model's errors. This method can more accurately determine the location of the digital key in complex environments, effectively improving positioning accuracy and robustness.

[0013] In some embodiments, the method further includes:

[0014] Obtain historical multi-anchor ranging values ​​and construct the first training set based on the historical multi-anchor ranging values;

[0015] The decision tree model is trained based on the first training set to obtain the second model;

[0016] A second training set is constructed based on the training data from which recognition errors occurred during the training of the second model;

[0017] The support vector classification model is trained based on the second training set to obtain the third model.

[0018] Based on the aforementioned technical methods, a first training set is constructed using historical multi-anchor point ranging values, ensuring the authenticity and representativeness of the training data. A decision tree model is used as the second model and trained on it. Leveraging its fast computation speed and strong adaptability, it can quickly complete localization judgments in most situations. Simultaneously, error samples are collected during the training of the second model to construct a second training set, and a support vector classification model is trained as the third model to further optimize the ability to identify complex boundary regions. This approach not only improves the overall performance of the model but also enhances the system's adaptability to real-world application environments.

[0019] In some embodiments, before training the support vector classification model based on the second training set, the method further includes:

[0020] The second model was deployed in a real vehicle environment, and the first test data of the second model's recognition error in the real vehicle environment was recorded.

[0021] A first data sample is generated based on the first measured data;

[0022] Add the first data sample to the second training set.

[0023] Based on the aforementioned technical means, during the joint training of the second and third models, after the second model is trained, it is deployed to a real vehicle environment to identify the test data and continue to capture the first test data with identification errors. These first test data with identification errors can reflect typical problems in actual operation, thus making the second training set closer to the actual application scenario. By adding the first test data to the second training set, the third model trained subsequently has higher generalization ability and adaptability, and can better compensate for the second model, which helps to reduce the misjudgment rate of the model in actual use.

[0024] In some embodiments, after training the support vector classification model based on the second training set to obtain the third model, the method further includes:

[0025] The second and third models were deployed in a real vehicle environment, and the second test data was recorded in which both the second and third models made recognition errors in the real vehicle environment.

[0026] A second data sample is generated based on the second measured data;

[0027] The second training set is updated based on the second data sample, and the third model is trained iteratively using the updated second training set.

[0028] Based on the aforementioned technical means, by continuously monitoring the performance of the second and third models in a real vehicle environment and collecting data on the recognition errors that both models exhibit, the content of the second training set is further enriched, making it more targeted. Then, the third model is iteratively trained using the updated second training set, which can continuously optimize the model parameters, improve its recognition accuracy in complex scenarios, and thus enhance the overall stability and reliability of the system.

[0029] In some embodiments, the vehicle is equipped with an SLE master module and multiple SLE slave modules to obtain multi-anchor distance values ​​corresponding to the digital key, including:

[0030] The SLE main module establishes a communication connection with the digital key and detects the distance from the digital key to the SLE main module to obtain the first distance measurement value.

[0031] Multiple SLEs are controlled to detect the distance from the digital key to themselves from the module, and multiple second ranging values ​​are obtained;

[0032] Multi-anchor point distance values ​​are obtained based on the first distance value and multiple second distance values.

[0033] Based on the aforementioned technical methods, multiple anchor points refer to the SLE master module and multiple SLE slave modules deployed on the vehicle. By establishing a connection with the digital key through the SLE master module and measuring the distance, and by having multiple SLE slave modules independently measure their respective distances to the digital key, distance measurement information from multiple directions can be obtained. Combining these distance measurement values ​​to form multi-anchor point distance values ​​can more comprehensively reflect the spatial positional relationship of the digital key relative to the vehicle, thereby providing richer input data for subsequent model recognition and improving the accuracy of positioning judgment.

[0034] In some embodiments, determining the location of the digital key based on the recognition results output by the second and third models includes:

[0035] If either the second or third model identifies that the digital key is inside the vehicle, then the location of the digital key is determined to be inside the vehicle.

[0036] If both the second and third models identify that the digital key is outside the vehicle, then the location of the digital key is determined to be outside the vehicle.

[0037] According to the aforementioned technical means, when either the second or third model identifies the digital key as being inside the vehicle, it indicates that at least two of the three models have determined that the digital key is inside the vehicle, thus the reliability of the identification result is high. Similarly, if the first model identifies the digital key as being inside the vehicle, the reliability of the identification result can only be determined if the second and third modules simultaneously determine that the digital key is being outside the vehicle. In this embodiment, by determining the final identification result when at least two models have the same identification result, the accuracy and stability of the positioning result are improved.

[0038] In some embodiments, the first model is invoked to identify the multi-anchor-point ranging values, and a first identification result is determined, including:

[0039] The first model is invoked to compare the multi-anchor point ranging values ​​with the preset multi-anchor point calibration threshold, and the first recognition result is determined based on the comparison result.

[0040] Based on the above technical means, by comparing the distance measurement values ​​of multiple anchor points with the preset calibration threshold, a preliminary judgment of whether the vehicle is inside or outside can be achieved without relying on a complex model, simplifying the calculation process. This method is suitable for scenarios with obvious signal characteristics. For example, when the digital key is far away from the vehicle, the conclusion that it is located outside the vehicle can be quickly drawn, thereby reducing unnecessary model calls and improving overall computing efficiency.

[0041] In some embodiments, if the first identification result indicates that the digital key is located outside the vehicle, the location of the digital key is determined to be outside the vehicle.

[0042] Based on the aforementioned technical means, when the first identification result indicates that the digital key is located outside the vehicle, it typically refers to a location outside the vehicle with very obvious signal characteristics, typically more than 5 meters away. In this case, the location result can be quickly provided directly through the first model, without needing to call the second and third models, thereby improving computational efficiency.

[0043] Secondly, embodiments of this application provide a digital key positioning device, which includes:

[0044] The acquisition module is used to acquire the multi-anchor point distance measurement values ​​corresponding to the digital key; wherein, the multi-anchor point distance measurement values ​​include the distance measurement values ​​from the digital key to different distance measurement modules on the vehicle;

[0045] The first calling module is used to call the first model to identify the distance measurement values ​​of multiple anchor points and determine the first identification result;

[0046] The second calling module is used to call the second model and the third model to identify the multi-anchor distance values ​​when the first recognition result indicates that the digital key is inside the vehicle; wherein, the second model is trained based on the first training set, and the third model is trained based on the second training set, the second training set including the training data in which recognition errors occurred during the training of the second model; the third model is used to verify the recognition results of the second model;

[0047] The positioning module is used to determine the location of the digital key based on the recognition results output by the second and third models.

[0048] Thirdly, embodiments of this application provide a vehicle including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions being executed by the processor to implement the steps of the digital key positioning method as described in any of the first aspects above.

[0049] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the digital key positioning method as described in any one of the first aspects above.

[0050] Fifthly, embodiments of this application provide a computer program product, including a computer program or computer executable instructions, which, when executed by a processor, implement the steps of the digital key positioning method as described in any of the first aspects above. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating a digital key positioning method provided in an embodiment of this application;

[0052] Figure 2 This is a schematic diagram of the judgment process of a digital key positioning method provided in an embodiment of this application;

[0053] Figure 3 This is a flowchart illustrating a joint training method provided in an embodiment of this application;

[0054] Figure 4 This is a schematic diagram of a vehicle provided in an embodiment of this application;

[0055] Figure 5 This is a schematic diagram of in-vehicle measurement points provided in an embodiment of this application; wherein, (a) shows the measurement points in the front row area of ​​the vehicle, (b) shows the measurement points in the rear row area, (c) shows the measurement points at the rear seat armrest, and (d) shows the measurement points at the vehicle storage compartment.

[0056] Figure 6 This is a schematic diagram of the distribution of a fusion model provided in an embodiment of this application;

[0057] Figure 7 This is a logical schematic diagram of a digital key positioning method provided in an embodiment of this application;

[0058] Figure 8 A logic block diagram of a digital key positioning device provided in an embodiment of this application;

[0059] Figure 9 This is a schematic diagram of the hardware structure of a vehicle provided in an embodiment of this application.

[0060] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0061] In order to gain a more detailed understanding of the features and technical content of the embodiments of this application, the implementation of the embodiments of this application will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for reference and illustration only and are not intended to limit the embodiments of this application.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0063] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0064] It should also be noted that the terms "first, second, and third" used in the embodiments of this application are only used to distinguish similar objects and do not represent a specific order of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0065] Furthermore, the reference to "embodiment" herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0066] Passive Start (PS) and Passive Entry (PE) are the core functions of a car's digital key. Through authentication and communication positioning of the digital key, users can lock and unlock the vehicle and start it without carrying a physical key. The implementation of PS and PE is closely related to the digital key's positioning inside and outside the vehicle. When the vehicle detects that the digital key is located in the PS area inside the vehicle, it allows the user to press the brake or a button to start the vehicle; when the vehicle's infotainment system detects that the digital key is located in the PE area outside the vehicle, it allows the user to press a microswitch or capacitive switch on the door handle to lock or unlock the vehicle.

[0067] For vehicle protection, the PS (Power Switch) area inside the vehicle needs to cover every corner, while controlling the overflow distance outside the vehicle within a certain range; the PE (Power Provider) area outside the vehicle must ensure that it does not overflow into the vehicle to prevent the digital key from being accidentally locked inside when the PE is locked, allowing it to be unlocked by someone else using the PE, resulting in property damage. Therefore, high requirements are placed on the positioning of the digital key's internal and external boundaries.

[0068] The SparkLink digital key uses a multi-anchor SLE (SparkLink Low Energy) ranging signal to calculate two-dimensional coordinates for positioning. However, in practical applications, the process of measuring the distance between the digital key and the vehicle can be affected by many factors, such as the digital key being located in the boundary area between the vehicle's interior and exterior, obstacles between the digital key and the vehicle, signal obstruction and reflection caused by the vehicle's sheet metal structure, etc. These factors can easily lead to deviations in the positioning results of the digital key, which in turn can cause PS and PE function failures or even property security issues.

[0069] To address the issue of poor accuracy in existing digital key positioning methods, this application provides a digital key positioning method. This method first acquires multi-anchor point ranging values ​​corresponding to the digital key to construct input features, improving the comprehensiveness of positioning information. Second, it performs coarse positioning by calling a first model and executes a decision-making mechanism based on the coarse positioning result. When the first recognition result output by the first model indicates that the digital key is outside the vehicle, the location of the digital key is determined to be outside the vehicle. When the first recognition result output by the first model indicates that the digital key is inside the vehicle, a second and third model are further called for secondary recognition. The second model is trained based on a first training set, while the third model is optimized based on data misjudged by the second model, thereby improving the third model's ability to compensate for the second model's errors. This approach effectively compensates for the potential blind spots that may exist in a single model, enabling more accurate determination of the digital key's location in complex environments and effectively improving positioning accuracy and robustness.

[0070] This application provides a method for vehicle interior and exterior identification and positioning based on a fusion model for StarFlash digital keys. It combines a decision tree module and an SVC (Support Vector Classification) model with a traditional threshold calibration model to construct an efficient, flexible, and highly generalizable fusion positioning model. A decision mechanism is established based on the characteristics of the three models, solving the problems of poor threshold calibration accuracy and poor flexibility and weak generalization ability of machine learning models, thus improving the accuracy of vehicle interior and exterior identification and positioning. An efficient joint training method is proposed, allowing different models (the second and third models) to leverage their strengths and converge quickly, enhancing the feasibility of the identification and positioning method. The StarFlash digital key vehicle interior and exterior identification and positioning method provided in this application, under the same hardware solution and cost control, improves the accuracy of vehicle interior and exterior identification and positioning through data-driven software algorithm optimization, significantly enhancing the PE and PS performance experience for digital key users, reducing the probability of positioning-related function failures, and protecting users' property security.

[0071] In its implementation, this application introduces a collaborative working mechanism of multiple machine learning models and forms an adaptive fusion model through data collection, cleaning, training, and iterative optimization in a real-vehicle environment. This model can improve positioning accuracy and reduce the probability of misjudgment while ensuring computational efficiency, and it exhibits good robustness, especially in complex environments.

[0072] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0073] In this embodiment, the processing steps of the digital key positioning method can be executed by the vehicle's control system. This control system includes, but is not limited to, the vehicle's main module, a CAN bus system, and embedded processing chips.

[0074] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a digital key positioning method provided in an embodiment of this application. The following description uses a vehicle as the executing entity to illustrate this digital key positioning method. Figure 1 As shown, the method may include steps 101 to 104.

[0075] Step 101: Obtain the multi-anchor point distance measurement value corresponding to the digital key.

[0076] Among them, the multi-anchor point distance measurement values ​​include distance measurement values ​​from the digital key to different distance measurement modules on the vehicle.

[0077] In one implementation, the vehicle is equipped with an SLE master module and multiple SLE slave modules, where multiple anchor points refer to these master and slave modules deployed on the vehicle. Typically, one SLE master module is located in the center console or headliner inside the vehicle. It is responsible for establishing a communication connection with the digital key and detecting the distance from the digital key to the SLE master module, obtaining a first distance measurement value. Multiple SLE slave modules (e.g., four or more SLE slave modules) can be deployed on both sides of the exterior bumper. After the SLE master module establishes a communication connection with the digital key, the SLE slave modules can listen to the echo signals returned by the digital key to detect the distance from the digital key to themselves, obtaining multiple second distance measurement values. The SLE slave modules then transmit the second distance measurement values ​​to the SLE master module via the Controller Area Network (CAN) bus. The SLE master module then aggregates these values, obtains the multi-anchor point distance measurement value based on the first distance measurement value and the multiple second distance measurement values, and transmits the multi-anchor point distance measurement value to the vehicle's CAN bus, thus enabling the vehicle to obtain the multi-anchor point distance measurement value. For example, during the data acquisition process, the vehicle's SLE main module can send multi-anchor distance values ​​to the CAN bus at a period of 100ms, and the data is updated in real time as the digital key changes position.

[0078] In some embodiments, after establishing a communication connection with the digital key, the SLE master module and multiple SLE slave modules on the vehicle continuously collect multi-anchor point ranging values ​​corresponding to the digital key in real time and store the ranging values ​​locally. The vehicle can retrieve the multi-anchor point ranging values ​​from its local memory.

[0079] In this scenario, the location of the digital key is constantly changing. The SLE master module and multiple SLE slave modules continuously measure the distance to the digital key and update the multi-anchor distance values ​​in local memory. The vehicle uses the latest multi-anchor distance values ​​to locate the digital key each time.

[0080] In some embodiments, occasional measurement anomalies, duplications, and invalidities may occur during actual measurements. To ensure accurate positioning of the digital key, the ranging data collected by the SLE main module and multiple SLE slave modules can be cleaned to obtain multi-anchor-point ranging values. Data cleaning refers to removing outliers, invalid values, and duplicate values ​​from the ranging data to obtain a high-quality dataset, accelerate model training, and improve model performance.

[0081] Step 102: Call the first model to identify the distance measurement values ​​of multiple anchor points and determine the first identification result.

[0082] In the embodiments of this application, the first model is a binary classification model. For example, the first model may be a threshold calibration model.

[0083] In this embodiment, the threshold calibration model is used for coarse positioning of the digital key. Specifically, the threshold calibration model can first define a relatively large range of vehicle interior and exterior boundaries. The interior range should cover all corners of the vehicle and extend a certain distance outwards. Please refer to [reference needed]. Figure 4 As shown, Figure 4 The diagram illustrates eight dashed circles on the outer side of the vehicle, with each circle indicating the overflow distance outside the vehicle. This overflow distance needs to be controlled within a certain range, for example, defining a distance of 1-2 meters outside the vehicle as the overflow range. Therefore, when the digital key is located within 1-2 meters outside the vehicle, i.e., within the overflow range, the threshold calibration model may mistakenly identify it as being inside the vehicle, leading to a false positive. Conversely, when the threshold calibration model determines the digital key is outside the vehicle, its actual location is at least outside the overflow range, thus ensuring higher reliability. In this case, if the first recognition result indicates the digital key is outside the vehicle, this result can be used as the final determination, confirming the key's location as outside. This eliminates the need to call second and third models, allowing for rapid identification and improved computational efficiency.

[0084] When the initial identification result indicates that the digital key is inside the vehicle, the actual location of the digital key may be inside the vehicle or outside the vehicle's out-of-range area. In this case, the reliability of the initial identification result is reduced.

[0085] In some embodiments, the first model includes a calibration threshold, where the calibration threshold refers to a threshold corresponding to a certain anchor point (SLE master module or SLE slave module). Taking the SLE master module (master anchor point) as an example, if the first model includes a calibration threshold corresponding to the SLE master module, then if the distance measured by the SLE master module is greater than the calibration threshold, it is determined that the digital key is outside the vehicle. If the distance measured by the SLE master module is less than or equal to the calibration threshold, it is determined that the digital key is inside the vehicle.

[0086] In some embodiments, the first module includes multiple anchor point calibration thresholds, where each anchor point (SLE master module or SLE slave module) has a corresponding threshold. The vehicle can call the first model to compare each distance measurement value in the multiple anchor point distance measurement values ​​with the corresponding anchor point's calibration threshold. If all distance measurement values ​​are greater than their respective corresponding calibration thresholds, it is determined that the digital key is outside the vehicle. Otherwise, it is determined that the digital key is inside the vehicle. Taking one SLE master module and one SLE slave module as an example, the multiple anchor point distance measurement values ​​include distance measurement value a measured by the SLE master module, distance measurement value b measured by the SLE slave module, calibration threshold A corresponding to the SLE master module, and calibration threshold B corresponding to the SLE slave module. Then, distance measurement value a is compared with calibration threshold A, and distance measurement value b is compared with calibration threshold B. If distance measurement value a is greater than calibration threshold A and distance measurement value b is greater than calibration threshold B, it is determined that the digital key is outside the vehicle. If the distance measurement value a is less than the calibration threshold A, or the distance measurement value b is less than the calibration threshold B, then the digital key is determined to be inside the vehicle.

[0087] In this embodiment, since the SLE master module and each SLE slave module are deployed at different locations on the vehicle, the calibration thresholds corresponding to the SLE master module and each SLE slave module are not the same.

[0088] The actual threshold calibration model in this application embodiment will combine multi-anchor point threshold calibration, using "or" and "and" relationships to achieve a more detailed differentiation effect.

[0089] Step 103: If the first recognition result indicates that the digital key is inside the vehicle, the second and third models are invoked to identify the multi-anchor distance values.

[0090] The second model is trained on the first training set, and the third model is trained on the second training set. The second training set includes training data that resulted in recognition errors during the training of the second model.

[0091] In this embodiment of the application, a decision mechanism for whether to call the machine learning model is established through a threshold calibration model. That is, when the first positioning result of the threshold calibration model indicates that the digital key is inside the vehicle, it is necessary to further call the machine learning model for secondary positioning verification.

[0092] The machine learning model includes a second model and a third model. In this embodiment, the second model is, for example, a decision tree model, and the third model is, for example, a support vector classification model.

[0093] In this embodiment, the second and third models are obtained through joint training. The joint training process includes: constructing a first training set and training the second model using the first training set; then recording training data in the first training set that satisfy a first condition, where training data satisfying the first condition refers to training data that resulted in recognition errors during the training of the second model; finally, constructing a second training set based on the training data in the first training set that satisfy the first condition, and training the third model using the second training set.

[0094] In this embodiment, since the third model is trained based on training data that resulted in misjudgments during the training of the second model, the third model is better able to capture misjudgments in edge regions, thereby improving the overall positioning accuracy.

[0095] In one implementation, the process of calling the second model and the third model to identify the multi-anchor-point ranging values ​​includes: inputting the multi-anchor-point ranging values ​​into the second model and the third model respectively, having the second model and the third model independently execute the recognition logic, and outputting the recognition results.

[0096] In this embodiment, when the first positioning result indicates that the digital key is inside the vehicle, the vehicle will activate the second and third models for dual verification to prevent inaccurate positioning due to misjudgment by the first model. Especially in the area near the vehicle's interior and exterior boundaries, a joint judgment mechanism composed of the first, second, and third models can improve positioning accuracy.

[0097] Step 104: Determine the location of the digital key based on the positioning results of the second model and the third model.

[0098] In this embodiment of the application, if either the second model or the third model identifies that the digital key is inside the vehicle, the location of the digital key is determined to be inside the vehicle; if both the second model and the third model identify that the digital key is outside the vehicle, the location of the digital key is determined to be outside the vehicle.

[0099] In this embodiment, if either the second or third model identifies the digital key as being inside the vehicle, it means that at least two of the three models (the first and second models, or the first and third models) have determined that the digital key is inside the vehicle, thus indicating a high degree of reliability in the identification result. Similarly, if the first model identifies the digital key as being inside the vehicle, the second and third modules must simultaneously determine that the digital key is being outside the vehicle for the identification result to be considered reliable. In this embodiment, by determining the final identification result when at least two models have the same identification result, the accuracy and stability of the positioning result are improved.

[0100] Please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the judgment process of a digital key positioning method provided in an embodiment of this application. It includes the following steps:

[0101] In step 201, the distance measurement values ​​of multiple anchor points are obtained.

[0102] In step 202, the threshold calibration model is invoked to determine whether the digital key is inside the vehicle.

[0103] If it is determined that the digital key is inside the vehicle, proceed to step 203.

[0104] If it is determined that the digital key is outside the vehicle, proceed to step 206.

[0105] In step 203, the decision tree model is invoked to determine whether the digital key is inside the vehicle.

[0106] If it is determined that the digital key is inside the vehicle, proceed to step 204.

[0107] If it is determined that the digital key is outside the vehicle, proceed to step 205.

[0108] In step 204, it is determined that the location result of the digital key is inside the vehicle.

[0109] In step 205, the support vector classification model is invoked to determine whether the digital key is inside the vehicle.

[0110] If it is determined that the digital key is inside the vehicle, proceed to step 204.

[0111] If it is determined that the digital key is outside the vehicle, proceed to step 206.

[0112] In step 206, it is determined that the location result of the digital key is outside the vehicle.

[0113] from Figure 2As can be seen, only when the first positioning result indicates that the digital key is inside the vehicle, and the two models output the same recognition result, is the recognition result of the two models determined to be credible and determined as the final recognition result.

[0114] It should be noted that in step 203, step 205 is only executed when it is determined that the digital key is outside the vehicle. However, in practical applications, the process of calling the support vector classification model to identify multi-anchor distance values ​​is performed simultaneously with the process of calling the decision tree model to identify multi-anchor distance values.

[0115] In this embodiment, the third model is primarily responsible for correcting misjudgments by the second model. The third model is trained on a more refined dataset to further optimize the localization results. The recognition results of the second and third models are comprehensively judged through a fusion mechanism: if either the second or third model determines the location is inside the vehicle, the final output result is "inside the vehicle"; if both the second and third models determine the location is outside the vehicle, the final output result is "outside the vehicle". This joint judgment mechanism of the second and third models effectively captures blind spots inside the vehicle, improving localization reliability. Especially in complex environments such as vehicle body panel obstruction and signal interference, the joint judgment mechanism of the second and third models demonstrates good robustness and generalization ability.

[0116] The identification and positioning method provided in this application combines a decision tree model, a support vector classification model, and a threshold calibration method to construct an efficient, flexible, and highly generalizable fusion positioning model. In practical applications, the identification and positioning method provided in this application can significantly improve the accuracy of vehicle interior and exterior identification and positioning while ensuring computational efficiency, reducing the probability of false positives, and exhibiting good robustness, especially in complex environments. Furthermore, through an efficient joint training method, the identification and positioning method provided in this application enables different sub-models to leverage their respective advantages and achieve rapid convergence, thereby enhancing the feasibility of the identification and positioning method. In summary, this method not only improves the reliability and user experience of the digital key system but also reduces the probability of location-related function failures, protecting the user's property security.

[0117] In practical applications, decision trees offer fast computation speed and can handle large amounts of features and data, but they are prone to overfitting, resulting in poor generalization. Addressing overfitting with pruning and limiting decision tree depth can further decrease model accuracy. SVC (Self-Volume Classification) offers good classification performance and avoids overfitting, but its computation involves high-dimensional matrices with large sample sizes, consuming significant memory and computation time, making it unsuitable for embedded implementations. Threshold calibration is the simplest and most intuitive method, allowing for flexible adjustment of threshold values, but its coarse classification principles lead to suboptimal results. To address these issues, this application proposes a joint training method. For the decision tree model, offline validation with real-vehicle data is used to quickly adjust the offline model, avoiding repetitive embedded code writing. For the SVC model, online validation in a real-vehicle environment is used to more accurately capture blind spots and accelerate model convergence. The joint training process for the second and third models is described below.

[0118] Please refer to Figure 3 , Figure 3 This is a flowchart illustrating a joint training method provided in an embodiment of this application.

[0119] Step 301: Obtain historical multi-anchor ranging values ​​and construct the first training set based on the historical multi-anchor ranging values.

[0120] The first training set refers to the initial data set used to train the machine learning model. The first training set includes multiple training data sets, each of which includes a set of multi-anchor distance values.

[0121] In this embodiment, the data in the first training set originates from sampling tests in a real-vehicle environment. The digital key is placed in different locations inside and outside the vehicle, and data is collected at fixed step sizes. The data in the first training set is cleaned and formatted into a structured format to facilitate subsequent model training. In this embodiment, the process of constructing the first training set includes:

[0122] Step 1: Obtain historical multi-anchor point distance measurement values.

[0123] Please refer to Figure 4 , Figure 4 This is a schematic diagram of a vehicle provided in an embodiment of this application, wherein an SLE main module M is provided in the roof of the vehicle. SLE slave modules A1, A2, A3, and A4 are respectively installed on both sides of the front and rear bumpers of the vehicle.

[0124] Please refer to Figure 5 , Figure 5 This is a schematic diagram of an in-vehicle measurement point provided in an embodiment of this application. Figure 5 (a) shows the measurement points in the front row area of ​​the vehicle. Figure 5 (b) shows the measurement points in the rear area. Figure 5 (c) shows the measurement point at the rear seat armrest. Figure 5 (d) in the diagram shows the measurement point at the vehicle's storage compartment. Figure 5 The numbers in the table indicate the serial numbers of the measurement points. These measurement points refer to locations inside the vehicle where a digital key can be placed. It should be noted that... Figure 5 The points shown do not constitute a limitation on the measurement points inside the vehicle. In practice, a ratio can be set... Figure 5 More or fewer measurement points in the middle.

[0125] In this embodiment of the application, the digital key can be placed in, for example... Figure 5 Any measurement point in the middle, and then using Figure 4 The SLE master module M and SLE slave modules A1, A2, A3, and A4 shown in the figure measure the digital key to obtain historical multi-anchor distance values.

[0126] In addition, the external area, including the range where the digital key is enabled, typically refers to the area within 10 meters of the vehicle body. Data can be collected at fixed points in steps of a fixed length (e.g., 1 meter). During data acquisition, the vehicle's SLE master module M can send historical multi-anchor point ranging values ​​to the CAN bus at 100ms intervals. Furthermore, the digital key can be placed at a critical point both inside and outside the vehicle, and the SLE master module M and SLE slave modules A1, A2, A3, and A4 can be used to measure the digital key and obtain historical multi-anchor point ranging values.

[0127] Step 2, data cleaning.

[0128] The collected historical multi-anchor point distance measurements inside and outside the vehicle are cleaned. For example, outliers can be removed using the interquartile range (IQR) method, or Gaussian smoothing filtering can be used for data cleaning.

[0129] Step 3: Construct the first training set.

[0130] The historical multi-anchor point ranging values ​​obtained in Step 2 are labeled, with all in-vehicle data labeled as "0" and all out-of-vehicle data labeled as "1". Each historical multi-anchor point ranging value and its corresponding data label constitute a training dataset.

[0131] In this embodiment of the application, after obtaining multiple training data, they can be divided into a training set and a test set in a 7:3 ratio, wherein the training set is the first training set.

[0132] By acquiring a high-quality first training set, we can ensure the representativeness and accuracy of the data during the model training process, thereby improving the generalization ability and recognition accuracy of the final model.

[0133] Step 302: Train the decision tree model based on the first training set to obtain the second model.

[0134] In this embodiment, decision tree modeling is first performed. A decision tree model is a supervised learning algorithm that generates a tree structure for classification by conditionally judging the distance measurements of multiple anchor points. The input to the decision tree model is preprocessed historical distance measurements from multiple anchor points, and the output is a binary classification result: either inside or outside the vehicle.

[0135] In this model, the "decision tree attribute" is the distance measurement value corresponding to a certain anchor point in a data set, the "test on an attribute" is the judgment that the distance measurement value corresponding to the anchor point is greater than or less than a certain threshold, and the "leaf node" is the decision result for inside or outside the vehicle. In this embodiment, the decision tree model has two leaf nodes, representing inside and outside the vehicle, respectively.

[0136] Among them, the decision tree model is characterized by its fast computation speed and strong interpretability, making it suitable for processing large-scale datasets. The decision tree model uses the multi-anchor distance values ​​from each group in the first training set as input features, and outputs a binary classification result of whether the data is inside or outside the vehicle. During training, cross-validation is used to select optimal parameters, such as tree depth and split node thresholds, to prevent overfitting and improve the stability of the decision tree model.

[0137] In this embodiment of the application, during the training of the decision tree model based on the first training set, the F1 score can be calculated; when the F1 score reaches the maximum value, the decision tree model is solidified to obtain the second model.

[0138] During the training of decision tree models, the training device evaluates the model's performance in real time. One important evaluation metric is the F1 score. The F1 score measures the balance between precision and recall in a classification model, and is particularly suitable for imbalanced datasets. The F1 value is the harmonic mean of precision and recall. Precision: the proportion of samples predicted as "positive" by the model that are actually "positive" (focusing on prediction accuracy). Recall: the proportion of samples that are actually "positive" that are successfully predicted as "positive" by the model (focusing on coverage completeness). The F1 value ranges from [0,1]. The closer it is to 1, the better the balance between precision and recall, making it particularly suitable for solving imbalanced class problems.

[0139] A higher F1 score indicates greater accuracy and robustness of the decision tree model in identifying in-vehicle and out-of-vehicle states. When the F1 score reaches its peak during training, it signifies that the model parameters of the decision tree model have reached their optimal state on the first training set at that point. To ensure model stability and reusability, further iterations will cease, and the model parameters will be saved to form a final, usable second model. This second model can be deployed in embedded systems to achieve real-time digital key location detection.

[0140] Step 303: Construct the second training set based on the training data in which recognition errors occurred during the training of the second model.

[0141] In this embodiment of the application, during the training of the second model, the training data that the second model identifies incorrectly during the training process can be recorded.

[0142] For example, if a training data A has a data label of "1", but during the model training process, the recognition result output by the second model is "0", that is, the recognition result output by the second model is inconsistent with the data label, in this case, the training data A can be recorded and added to the second training set.

[0143] In one implementation, after recording training data A, the surrounding area can be determined based on the location of the digital key indicated by training data A. The surrounding area is the region centered on the location of training data A, with a first distance as its radius. Multiple candidate points can be determined within the surrounding area, and multi-anchor distance values ​​can be obtained for each candidate point. Data labels are determined based on the multi-anchor distance values ​​corresponding to the candidate points, and training data is constructed and added to a second training set. This method expands the sample range of the second training set.

[0144] In one implementation, during the training of the second model, training data with incorrect identification is not recorded at the beginning of training. Instead, training data with incorrect identification is recorded only when the F1 score of the second model is greater than a first preset value.

[0145] When the F1 score of the second model is greater than the first preset value, it indicates that the recognition accuracy of the second model has reached a high level. In this case, if the second model still makes a mistake, it means that the location point corresponding to the training data is the detection blind zone of the second model. In this case, the training data is added to the second training set to train the third model so that the third model can verify the recognition results of the second model.

[0146] In this embodiment, the training data included in the second training set typically corresponds to boundary regions or scenarios with strong signal interference, such as ranging deviations caused by body panel occlusion. By extracting data points that were misjudged by the decision tree model during training, the model can be further optimized, especially in blind spot detection.

[0147] In this embodiment, after training the second model using the first training set, it is necessary to test the second model using a test set. During testing, if the second model outputs an incorrect recognition result, the test data showing the incorrect recognition result is recorded and added to the second training set.

[0148] For example, when the second model tests test data B, it outputs a recognition result of "0", while the actual data label of test data B is "1". The recognition result output by the second model is inconsistent with the data label of the test data. In this case, test data B can be recorded and added to the second training set.

[0149] In some embodiments, a surrounding area can be determined based on the location of the digital key indicated by test data B. This surrounding area is a region centered on the location of test data B and with a radius equal to a first distance. Multiple candidate points can be determined within this surrounding area, and multi-anchor distance values ​​can be obtained for each candidate point. Data labels are determined based on the multi-anchor distance values ​​corresponding to the candidate points, and training data is constructed and added to a second training set. This method expands the sample range of the second training set.

[0150] In another implementation, after the second model is trained, it can be deployed to a real vehicle environment, and the first test data in which the second model makes a recognition error in the real vehicle environment is recorded; a first data sample is generated based on the first test data; and the first data sample is added to the second training set.

[0151] The real-vehicle environment refers to the physical scenario of actual vehicle operation and testing, as opposed to simulation or laboratory environments, and is used to simulate positioning challenges under real-world usage conditions. Multi-anchor distance measurements refer to the distance measurement results between the digital key and multiple SLE master and slave modules, typically including distance information from one SLE master module and four SLE slave modules, forming a five-dimensional feature vector used to determine whether the digital key is inside or outside the vehicle.

[0152] Recognition error refers to a situation where the output of the second model is inconsistent with the actual location. For example, the digital key may actually be located inside the vehicle, but the second model mistakenly identifies it as being outside. In this embodiment, the data in which the second model makes a recognition error in a real vehicle environment is defined as the first measured data, which includes multiple distance measurements of the same digital key.

[0153] In this embodiment, for each first measured data point, a data tag is determined based on the actual location of the digital key corresponding to the first measured data point. For example, all in-vehicle data tags are "0", and all out-of-vehicle data tags are "1". Thus, a first data sample is generated based on the first measured data and the data tags.

[0154] In this embodiment, the first data sample can be added to the second training set to enrich the sample sources in the second training set. The first data sample consists of identification error samples collected in the real vehicle environment, including specific timestamps, ranging values, and actual labels, and is used to supplement the real boundary situations that may be missing in the second training set, thereby improving the generalization ability of the second model.

[0155] The second training set is an enhanced training dataset composed of two parts: one part consists of misidentification samples from the real-world vehicle environment (the first set of data), and the other part consists of misidentification samples that occurred during the training process when testing the second model. This hybrid approach enables the second model to better learn boundary features in complex scenes, especially edge cases that are prone to misjudgment.

[0156] In this embodiment, by introducing recognition error data collected in a real vehicle environment, the diversity of training data is enriched, improving the robustness and accuracy of the second model in actual deployment. Furthermore, since the first data samples originate from a real-world application environment, they can reflect the performance deficiencies of the second model in real-world scenarios. Therefore, the parameters of the second model can be optimized in a targeted manner, reducing the misclassification rate of the second model in embedded systems.

[0157] In this embodiment of the application, by constructing a second training set, the recognition error of the decision tree model can be effectively captured and corrected, thereby improving the adaptability and robustness of the decision tree model in complex scenarios.

[0158] Step 304: Train the support vector classification model based on the second training set to obtain the third model.

[0159] In this embodiment, Support Vector Classification (SVC) is an efficient binary classification model suitable for classification problems in high-dimensional spaces. In this embodiment, the SVC model is trained using training data from a second training set, focusing on addressing edge regions that decision tree models cannot cover, thereby improving overall localization accuracy.

[0160] In this embodiment, the SVC model separates in-vehicle and out-of-vehicle data by finding the optimal hyperplane and maximizing the margin between categories. Because the SVC model is insensitive to noise and can handle small sample data well, it is particularly suitable as a supplement to decision tree models. In this embodiment, the main function of the third model is to compensate for the second model; that is, it identifies data that the second model cannot accurately recognize, thus verifying the recognition results of the second model.

[0161] Here, the "hyperplane" refers to an expression that uses the multi-anchor SLE distance values ​​as variables. Substituting the multi-anchor SLE distance values ​​of the unknown measurement point into the expression, a result > 0 indicates that the measurement point is located on one side of the hyperplane, corresponding to the location "inside the vehicle"; a result < 0 indicates that the measurement point is located on the other side of the hyperplane, corresponding to the location "outside the vehicle". Note that the SVC model is not suitable for handling large-scale data. Therefore, in this embodiment, the SVC model is used as compensation for the decision tree model to capture small blind spots inside the vehicle. The dataset size can be reduced through subsampling for initial modeling, and then the model can be updated through subsequent optimization iterations.

[0162] In this embodiment, the SVC model is trained using a second training set and implemented in an embedded manner. Real-vehicle verification is repeated until no false positives are detected, indicating that the SVC model has reached its optimal performance. The SVC model is then solidified to obtain the third model. The SVC model can compensate for the shortcomings of the decision tree model in specific scenarios and improve the overall recognition performance of the fusion model, exhibiting stronger stability and reliability when processing edge data.

[0163] By acquiring the first training set and training the decision tree model and SVC model, the positioning accuracy of the StarFlash digital key's vehicle interior and exterior recognition can be improved. This can reduce misjudgments caused by signal interference or obstruction, thereby enhancing the stability of the PS and PE functions, and ultimately improving the user experience and vehicle safety protection level.

[0164] Based on the above embodiments, in this embodiment, after the second model and the third model are trained, the second model and the third model can be deployed to a real vehicle environment to record second test data in which both the second model and the third model make recognition errors in the real vehicle environment; a second data sample is generated based on the second test data; the second training set is updated based on the second data sample, and the third model is trained iteratively using the updated second training set.

[0165] In this embodiment, the second model refers to a decision tree model that has been initially trained and solidified, while the third model refers to an SVC (Support Vector Classification) model formed through optimization iterations. These two models respectively handle the functions of rapid judgment and high-precision verification. The real-vehicle environment refers to actual test scenarios where the StarFlash digital key is in different positions under real vehicle operating conditions.

[0166] In this real-vehicle environment, the second and third models were run simultaneously to evaluate the accuracy of their collaborative operation. When both the second and third models made a recognition error at a specific location, the training data corresponding to that location was recorded to form the second set of test data, which served as the basis for subsequent model optimization.

[0167] The data tag is determined based on the actual location of the Star Flash Digital Key corresponding to the second measured data, and then a second data sample is generated based on the second measured data and the data tag.

[0168] In this embodiment, the second training set is updated using second data samples. Then, the updated second training set is used to repeatedly train the second and third models until no misjudged data points are found during real-vehicle verification, at which point iteration stops. The updated second training set is formed by fusing the original training data with the newly generated second data samples, creating a more representative training dataset. By continuously adding new second data samples, the SVC model can be continuously optimized and iterated. This iterative training mechanism allows the SVC model to continuously learn new edge cases in practical applications, thereby improving the overall positioning accuracy and generalization ability. For example, in multiple real-vehicle verifications, it was found that some car models, due to their unique body design, repeatedly misjudged the StarFlash digital key near the door. By introducing misjudgment data caused by these car models for iterative training, the SVC model can gradually correct relevant parameters and improve recognition performance.

[0169] In this embodiment, by deploying the second and third models in a real vehicle environment and collecting data from both models that are misclassified to form a second data sample, dynamic updates to the training set and continuous optimization of the model are achieved. This effectively improves the model's adaptability to complex environments, thereby reducing the probability of misclassification and significantly enhancing the stability and reliability of the StarFlash digital key in vehicle interior and exterior recognition.

[0170] After training the second and third models, this embodiment establishes a fusion model based on a decision tree module and a support vector classification model. When the threshold calibration model determines that the digital key is outside the vehicle, the output of the threshold calibration model is used as the final recognition result. When the threshold calibration model determines that the digital key is inside the vehicle, the decision tree model and the support vector classification model are invoked. This decision mechanism can reduce the number of times the machine learning model is invoked, thereby reducing the computational load on the algorithm.

[0171] Furthermore, in this embodiment, when the decision tree model and SVC model perform well, and the fusion model aims to rely on these two machine learning models to the greatest extent possible, the calibration threshold set in the threshold calibration model can be increased, thus expanding the in-vehicle calibration range. In this way, the vehicle primarily uses the recognition results of the decision tree model and SVC model as the basis for its judgment.

[0172] If the decision tree model and the SVC model deviate to some extent during the transfer and generalization process and cannot be fully relied upon, the calibration threshold set in the threshold calibration model can be reduced, that is, the in-vehicle calibration range can be reduced. In this way, the vehicle mainly uses the recognition result of the threshold calibration model as the basis for judgment.

[0173] In this embodiment, the threshold calibration model, decision tree model, and support vector classification model are complementary. Decision tree models excel at handling routine scenarios, while support vector classification models focus more on misclassification issues in complex environments. Therefore, combining decision tree models with support vector classification models can effectively improve the overall recognition accuracy and robustness of the system.

[0174] Please refer to Figure 6 , Figure 6 This is a schematic diagram illustrating the distribution of a fusion model provided in an embodiment of this application. A StarSpark main module chip is deployed in the vehicle, and the StarSpark main module chip deploys a fusion model, which includes a first model, a second model, and a third model. According to... Figure 6 The roles of the three models in the fusion model are evident. The first model (threshold calibration model) is used for initial judgment, quickly providing location results with simple logic and complete coverage of all areas inside the vehicle. The second model (decision tree model) covers the vast majority of samples, but may have blind spots in border areas inside and outside the vehicle. The third model (support vector classification model) effectively covers these blind spots, compensating for the shortcomings of the second model. Therefore, the fusion model improves the accuracy of the model's judgment while retaining its flexibility and generalization ability.

[0175] Please refer to Figure 7 , Figure 7This is a logical schematic diagram of a digital key positioning method provided in an embodiment of this application.

[0176] Step 1: Obtain historical SLE data inside and outside the vehicle.

[0177] Step 2: Data cleaning to form the first dataset.

[0178] Step 3: Build a fusion model based on the first dataset.

[0179] The process of establishing a fusion model includes decision tree modeling, SVC modeling, and threshold calibration.

[0180] Step 4: Joint training of sub-models.

[0181] Sub-models refer to decision trees, SVC, and threshold calibration.

[0182] The joint training of sub-models includes step 4.1: decision tree model training; step 4.2: decision tree model validation; step 4.3: SVC model training; and step 4.4: joint training and iterative optimization.

[0183] Step 4.1, decision tree model training, includes: training the decision tree model, calculating the F1 score, and determining if the F1 score is the highest. If it is the highest, the iteration stops, the model is solidified, and the decision tree model is obtained. If it is not the highest, the model parameters are adjusted and the next round of decision tree model training begins. This process is repeated until the model is solidified.

[0184] After the model is solidified, proceed to step 4.2: Decision Tree Model Validation. This includes embedding the decision tree model, inputting false positives from real vehicle verification, and forming a second dataset.

[0185] Proceed to step 4.3: SVC model training. SVC model training includes training the SVC model using the second dataset.

[0186] Then proceed to step 4.4: joint training and iterative optimization. The SVC model is embedded; is there any misclassification point during real-vehicle verification? If there is a misclassification point, it is added to the second dataset, the second dataset is updated, and the SVC model is trained again using the updated second dataset. Then, steps 4.3 and 4.4 are repeated until no misclassification points appear during real-vehicle verification.

[0187] When there are no misjudgments, the iteration stops, the model is solidified, and the SVC model is obtained.

[0188] Step 5: Fusion of binary sub-models based on application scenarios.

[0189] from Figure 7As can be seen, after the decision tree model is trained, it will proceed to step 5; after the SVC model is trained, it will proceed to step 5; after the threshold is calibrated, it will proceed to step 5.

[0190] At this point, the three models can be deployed in a real vehicle environment.

[0191] In a real-world vehicle environment, after detecting the multi-anchor point ranging value corresponding to the digital key, a threshold calibration is used to determine whether the user is inside the vehicle. If so, a decision tree model is used to further determine whether the user is inside the vehicle. If so, the positioning result is determined to be inside the vehicle.

[0192] If the threshold calibration determines that the location is outside the vehicle, then the positioning result is directly determined to be outside the vehicle.

[0193] If the decision tree model determines the location is outside the vehicle, then the SVC model is used to determine if the location is inside the vehicle. If yes, the location result is determined to be inside the vehicle. If not, the location result is determined to be outside the vehicle.

[0194] The embodiments of this application obtain a fusion model through joint training, which fully covers possible key locations inside and outside the vehicle, and improve the accuracy of the positioning results by compensating the decision tree model with an SVC model.

[0195] It should be understood that the steps in the aforementioned accompanying drawings are not necessarily performed in the order indicated in the drawings. Unless otherwise expressly stated herein, there is no strict order restriction on the execution of these steps, and they may be performed in other orders. Moreover, at least some of the steps in these drawings may include multiple sub-steps or multiple stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0196] In another embodiment of this application, a digital key positioning device is provided; please refer to [reference needed]. Figure 8 , Figure 8 A logic block diagram of a digital key positioning device provided in an embodiment of this application. The digital key positioning device 800 may include: an acquisition module 801, a first invocation module 802, a second invocation module 803, and a positioning module 804, wherein:

[0197] The acquisition module 801 is used to acquire the multi-anchor point distance measurement value corresponding to the digital key; wherein, the multi-anchor point distance measurement value includes the distance measurement value from the digital key to different distance measurement modules on the vehicle;

[0198] The first calling module 802 is used to call the first model to identify the distance measurement values ​​of multiple anchor points and determine the first identification result;

[0199] The second calling module 803 is used to call the second model and the third model to identify the multi-anchor distance values ​​when the first recognition result indicates that the digital key is inside the vehicle; wherein, the second model is trained based on the first training set, and the third model is trained based on the second training set, the second training set including the training data in which recognition errors occurred during the training of the second model; the third model is used to verify the recognition result of the second model;

[0200] The positioning module 804 is used to determine the location of the digital key based on the recognition results output by the second and third models.

[0201] In some embodiments, the acquisition module 801 is further configured to:

[0202] Obtain historical multi-anchor ranging values ​​and construct the first training set based on the historical multi-anchor ranging values;

[0203] The decision tree model is trained based on the first training set to obtain the second model;

[0204] A second training set is constructed based on the training data from which recognition errors occurred during the training of the second model;

[0205] The support vector classification model is trained based on the second training set to obtain the third model.

[0206] In some embodiments, the acquisition module 801 is further configured to:

[0207] The second model was deployed in a real vehicle environment, and the first test data of the second model's recognition error in the real vehicle environment was recorded.

[0208] A first data sample is generated based on the first measured data;

[0209] Add the first data sample to the second training set.

[0210] In some embodiments, the acquisition module 801 is further configured to:

[0211] The second and third models were deployed in a real vehicle environment, and the second test data was recorded in which both the second and third models made recognition errors in the real vehicle environment.

[0212] A second data sample is generated based on the second measured data;

[0213] The second training set is updated based on the second data sample, and the third model is trained iteratively using the updated second training set.

[0214] In some embodiments, an SLE master module and multiple SLE slave modules are deployed on the vehicle. The acquisition module 801 is specifically used to: control the SLE master module to establish a communication connection with the digital key, and detect the distance from the digital key to the SLE master module to obtain a first distance measurement value.

[0215] Multiple SLEs are controlled to detect the distance from the digital key to themselves from the module, and multiple second ranging values ​​are obtained;

[0216] Multi-anchor point distance values ​​are obtained based on the first distance value and multiple second distance values.

[0217] In some embodiments, the positioning module 804 is specifically used for:

[0218] If either the second or third model identifies that the digital key is inside the vehicle, then the location of the digital key is determined to be inside the vehicle.

[0219] If both the second and third models identify that the digital key is outside the vehicle, then the location of the digital key is determined to be outside the vehicle.

[0220] In some embodiments, the first calling module 802 is specifically used to: call the first model to compare the multi-anchor point ranging value with the preset multi-anchor point calibration threshold, and determine the first identification result based on the comparison result.

[0221] Each module in the aforementioned digital key positioning device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0222] Please refer to Figure 9 , Figure 9 This is a schematic diagram of the hardware structure of a vehicle provided in an embodiment of this application. The vehicle may include a communication interface 901, a memory 902, and a processor 903; the various components are coupled together through a bus system 904. It is understood that the bus system 904 is used to realize the connection and communication between these components. In addition to a data bus, the bus system 904 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 9 The general designated all buses as Bus System 904.

[0223] In this embodiment, the communication interface 901 is used to send and receive information with other external devices; the memory 902 is used to store a computer program that can run on the processor 903; the processor 903 is used to execute the steps of the digital key positioning method described in any of the foregoing embodiments when running the computer program.

[0224] It is understood that the memory 902 in this embodiment of the application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate Synchronous DRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 902 of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0225] The processor 903 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 903 or by software instructions. The processor 903 can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 902, and the processor 903 reads the information in memory 902 and, in conjunction with its hardware, completes the steps of the above method.

[0226] It is also understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0227] For software implementation, the techniques described herein can be implemented through modules (e.g., procedures, functions, etc.) that perform the functions described herein. Software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or externally. Wherein, if implemented as a software functional module and not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0228] In another embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the digital key positioning method described in the foregoing embodiments.

[0229] In another embodiment of this application, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the steps of the digital key positioning method as described in the foregoing embodiments.

[0230] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, devices, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage and optical storage) containing computer-usable program code.

[0231] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0232] The sequence numbers of the embodiments in this application are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The features disclosed in the several product embodiments provided in this application can be arbitrarily combined to obtain new product embodiments without conflict. Similarly, the features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method or device embodiments without conflict. The above descriptions are merely specific implementations of this application, but the protection scope of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the protection scope of this application.

Claims

1. A digital key positioning method, characterized in that, The method includes: Obtain the multi-anchor point ranging value corresponding to the digital key; wherein, the multi-anchor point ranging value includes the ranging value from the digital key to different ranging modules on the vehicle; The first model is invoked to identify the multi-anchor point distance measurement values, and a first identification result is determined; If the first recognition result indicates that the digital key is inside the vehicle, the second and third models are invoked to identify the multi-anchor distance values. The second and third models are obtained through joint training; the second model is a decision tree model, and the third model is a support vector classification model. The second model is trained based on a first training set, and the third model is trained based on a second training set, which includes training data from which recognition errors occurred during the training of the second model. The third model is used to verify the recognition results of the second model. The location of the digital key is determined based on the recognition results output by the second model and the third model; The process of jointly training the second model and the third model includes: The decision tree model is trained based on the first training set to obtain the second model; The second training set is constructed based on the training data from which recognition errors occurred during the training of the second model; The third model is obtained by training the support vector classification model based on the second training set.

2. The method according to claim 1, characterized in that, The method further includes: Obtain historical multi-anchor distance values, and construct the first training set based on the historical multi-anchor distance values.

3. The method according to claim 1, characterized in that, Before training the support vector classification model based on the second training set, the method further includes: The second model is deployed in a real vehicle environment, and the first measured data of the second model's recognition error in the real vehicle environment is recorded. A first data sample is generated based on the first measured data; Add the first data sample to the second training set.

4. The method according to claim 1, characterized in that, After training the support vector classification model based on the second training set to obtain the third model, the method further includes: The second model and the third model are deployed in a real vehicle environment, and second test data is recorded in the real vehicle environment in which both the second model and the third model make recognition errors. A second data sample is generated based on the second measured data; The second training set is updated based on the second data sample, and the third model is trained iteratively using the updated second training set.

5. The method according to any one of claims 1-4, characterized in that, The vehicle is equipped with an SLE master module and multiple SLE slave modules. The acquisition of the multi-anchor distance measurement value corresponding to the digital key includes: The system controls the SLE main module to establish a communication connection with the digital key and detects the distance from the digital key to the SLE main module to obtain a first distance measurement value. The multiple SLE modules are controlled to detect the distance from the digital key to themselves, and multiple second distance values ​​are obtained; The multi-anchor point distance measurement value is obtained based on the first distance measurement value and the plurality of second distance measurement values.

6. The method according to any one of claims 1-4, characterized in that, Determining the location of the digital key based on the recognition results output by the second model and the third model includes: If either the second model or the third model identifies that the digital key is inside the vehicle, then the location of the digital key is determined to be inside the vehicle. If both the second model and the third model identify that the digital key is located outside the vehicle, then the location of the digital key is determined to be outside the vehicle.

7. The method according to any one of claims 1-4, characterized in that, The step of calling the first model to identify the multi-anchor point ranging values ​​and determining the first identification result includes: The first model is invoked to compare the multi-anchor point ranging values ​​with a preset multi-anchor point calibration threshold, and the first identification result is determined based on the comparison result.

8. The method according to any one of claims 1-4, characterized in that, If the first identification result indicates that the digital key is located outside the vehicle, then the location of the digital key is determined to be outside the vehicle.

9. A digital key positioning device, characterized in that, The digital key positioning device includes: The acquisition module is used to acquire the multi-anchor point distance measurement value corresponding to the digital key; wherein, the multi-anchor point distance measurement value includes the distance measurement value from the digital key to different distance measurement modules on the vehicle; The first calling module is used to call the first model to identify the multi-anchor point ranging values ​​and determine the first identification result; The second invocation module is used to invoke the second model and the third model to identify the multi-anchor distance value when the first recognition result indicates that the digital key is located inside the vehicle; wherein the second model and the third model are obtained through joint training, the second model is a decision tree model, and the third model is a support vector classification model; the second model is trained based on a first training set, and the third model is trained based on a second training set, the second training set including training data in which recognition errors occurred during the training of the second model; the third model is used to verify the recognition result of the second model; A positioning module is used to determine the location of the digital key based on the recognition results output by the second model and the third model; The process of jointly training the second model and the third model includes: The decision tree model is trained based on the first training set to obtain the second model; The second training set is constructed based on the training data from which recognition errors occurred during the training of the second model; The third model is obtained by training the support vector classification model based on the second training set.

10. A vehicle, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that can run on the processor, the program or instructions being executed by the processor to implement the steps of the digital key positioning method as described in any one of claims 1 to 8.

Images