Digital key positioning method and vehicle unlocking system

Abstract

A digital key positioning method, comprising: inputting, into a multi-task neural network, Bluetooth RSSI values of a plurality of modules in a first device with respect to a second device, so as to obtain a distance, a region probability and a direction probability; then on the basis of the distance and the region probability, obtaining a region where the second device is located, and when the region where the second device is located is an unlocking region, on the basis of the direction probability, obtaining a direction where the second device is located; in addition, on the basis of the region where the second device is located and the direction where the second device is located, obtaining digital key positioning information. The present application can improve the positioning efficiency and accuracy, thereby quickly and accurately outputting the region and direction where a user is located.

Description

A digital key positioning method and vehicle unlocking system Technical Field

[0001] This application relates to the field of intelligent vehicle technology, specifically to a digital key positioning method and a vehicle unlocking system. Background Technology

[0002] BLE digital keys utilize Bluetooth Low Energy (BLE) to enable car owners to enter their vehicles seamlessly, lock their cars without being noticed when they leave, and receive automatic welcome greetings upon approach.

[0003] BLE positioning algorithms need to determine the phone's relative area (inside the car, unlocked area, locked area, etc.) and direction (front, rear, left, right, etc.) relative to the car in real time. The positioning principle of BLE digital keys utilizes the relationship between the RSSI strength and distance between the vehicle's Bluetooth module and the phone to calculate the distance between the phone and the car, thereby determining the phone's location and, based on the strength relationships between various Bluetooth modules, determining the phone's direction relative to the car. Deep learning technology, with its powerful feature extraction capabilities, can learn the non-linear relationship between the RSSI of multiple vehicle modules and the phone's distance and direction, and has been successfully applied to BLE positioning. However, current deep learning-based BLE positioning generally only predicts distance. The determination of direction and area is obtained through post-processing after using a neural network to obtain the real-time distance.

[0004] Traditional BLE positioning algorithms typically rely on distance attenuation models and manually defined rules to accomplish the aforementioned positioning tasks. Their drawback lies in the fact that the simple distance attenuation model cannot model the complex nonlinear relationship between the RSSI strength of multiple vehicle Bluetooth modules and distance, leading to higher positioning errors.

[0005] Traditional BLE localization algorithms are limited by the low computing power of MCU chips, resulting in fewer parameters in the neural network model and a greater risk of overfitting in single-task learning, leading to poor model generalization ability.

[0006] Distance-based post-processing logic needs to be manually defined, is generally quite complex, has lower efficiency, and is difficult to guarantee accuracy. Furthermore, the additional post-processing logic increases the algorithm's computation time and memory usage. Summary of the Invention

[0007] To address the above technical issues, this application proposes a digital key positioning method and a vehicle unlocking system.

[0008] In a first aspect, this application proposes a digital key positioning method, which specifically includes: inputting the Bluetooth RSSI values ​​of multiple modules in a first device and a second device into a multi-task neural network to obtain initial positioning information; wherein the initial positioning information includes at least distance, area probability, and direction probability; obtaining the area where the second device is located based on the distance and area probability, and when the area where the second device is located is an unlocked area, determining the direction based on the direction probability to obtain the direction of the second device; and obtaining digital key positioning information based on the area where the second device is located and the direction of the second device.

[0009] This application can utilize a neural network model to simultaneously predict distance, region, and direction, thereby improving positioning efficiency and accuracy, determining the user's region and direction relative to the vehicle, and quickly and accurately outputting the user's location and direction.

[0010] Furthermore, before obtaining the initial positioning information, the process includes: training an end-to-end neural network model as the multi-task neural network using a multi-task learning strategy; wherein the multi-task learning strategy includes at least three training tasks: distance regression, region classification, and orientation classification.

[0011] Furthermore, training an end-to-end neural network model includes: acquiring the Bluetooth RSSI values ​​of multiple modules in the first device and the second device, performing forward propagation to obtain task materials; extracting features from the task materials to obtain the low-level features of each training task; wherein different low-level features share information and are constrained; integrating the low-level features of each training task to obtain the high-level features of each training task; and outputting prediction results: distance, region probability, and direction probability.

[0012] Furthermore, training an end-to-end neural network model further includes: calculating the conditional joint loss function, automatically ignoring the directional cross-entropy loss in the non-unlocked area; backpropagating the Bluetooth RSSI value, using gradient descent to update the network parameters; and iteratively training until the multi-task neural network reaches a preset number of iterations or converges.

[0013] Furthermore, the output prediction result also includes: assigning empty values ​​to the direction labels of non-unlocked areas and assigning normal values ​​to the direction labels of unlocked areas; wherein, the non-unlocked areas include at least the locked areas and the vehicle interior.

[0014] The above employs a multi-task learning strategy. It utilizes an end-to-end neural network to simultaneously learn multiple labels, including distance, region, and orientation. Because of the correlation between the three tasks (distance, region, and orientation), features from different tasks can be shared, complementing and constraining each other. This effectively alleviates the model's overfitting problem, improves its generalization ability across different tasks, and significantly enhances model performance.

[0015] This invention employs an incompletely labeled dataset to reduce the difficulty of data collection. Since BLE localization typically only requires direction determination for the unlocked region, direction labels cannot be added to data outside the unlocked region in the training dataset. To address this, the present invention assigns null values ​​to direction labels for samples outside the unlocked region and normal values ​​to those in the unlocked region when building the training dataset. A conditional joint loss function is constructed, and during training, the direction classification loss for the unlocked region is adaptively ignored, enabling multi-task learning on an incompletely labeled dataset.

[0016] Furthermore, obtaining the area where the second device is located includes: obtaining a preliminary area based on the area probability, and determining whether the preliminary area exceeds a predetermined range based on the distance; if so, correcting the preliminary area to obtain a corrected area; otherwise, taking the preliminary area as the area where the second device is located.

[0017] After obtaining the correction area, the method further includes: determining whether the correction area exceeds a predetermined range; if so, performing a second correction on the correction area until the correction area does not exceed the predetermined range, and then using the correction area as the area where the second device is located.

[0018] In real-world scenarios, area information cannot be completely accurate and may be affected by factors such as signal attenuation and environmental interference. Therefore, distance is needed to estimate and correct for the area.

[0019] Furthermore, obtaining the orientation of the second device includes: determining the orientation of the second device relative to the first device by using the angle information between multiple modules in the first device and the second device, and combining it with the orientation probability, so as to use the orientation of the second device as the orientation of the second device.

[0020] Furthermore, after obtaining the area where the second device is located, the process further includes: ending the current digital key positioning process when the area where the second device is located is a locked area, or when the area where the second device is located is inside a vehicle.

[0021] Here, direction probability is a statistic used to represent the likelihood of the second device's orientation. By combining angle information and direction probability, the orientation of the second device relative to the first device can be determined more accurately. By collecting angle information from multiple modules and combining it with direction probability techniques, the orientation of the second device relative to the first device can be inferred with relatively high precision. This method typically combines angle difference with statistical inference to achieve high-precision orientation estimation.

[0022] Secondly, this application proposes a vehicle unlocking system, the system including at least a first device and a second device, the first device being configured with a cockpit domain controller; the cockpit domain controller is used to receive Bluetooth signals emitted by the second device to obtain digital key positioning information based on the Bluetooth signals, and to unlock the first device according to the digital key positioning information.

[0023] The cockpit domain controller includes at least a memory and a processor; the memory stores computer programs for multiple functional layers, each functional layer including one or more neural networks associated with operating elements for controlling vehicle unlocking; the processor communicates with the memory to execute the computer programs for each of the multiple functional layers stored in the memory.

[0024] The memory includes at least a signal conversion function layer and a digital key positioning function layer; one or more neural networks in the signal conversion function layer are used to receive Bluetooth signals emitted by the second device and convert the Bluetooth signals into Bluetooth RSSI values; the digital key positioning function layer includes a multi-task neural network used to receive Bluetooth RSSI values ​​of multiple modules in the first device and the second device, so as to output digital key positioning information based on the Bluetooth RSSI values.

[0025] The multi-task neural network includes at least an input layer, a hidden layer, and an output layer; the input layer is used to receive Bluetooth RSSI values ​​between multiple modules in the first device and the second device; the hidden layer is used to obtain initial positioning information based on the Bluetooth RSSI values, and to obtain the area where the second device is located and the direction where the second device is located based on the initial positioning information; the output layer is used to output the area where the second device is located and the direction where the second device is located, as digital key positioning information.

[0026] The hidden layer includes at least an initial positioning information generation module, an area acquisition module, and a direction determination module. The initial positioning information generation module is used to obtain initial positioning information based on the Bluetooth RSSI value. The initial positioning information includes at least distance, area probability, and direction probability. The area acquisition module is used to obtain the area where the second device is located based on the distance and area probability. The direction determination module is used to determine the direction of the second device based on the direction probability when the area where the second device is located is an unlocked area.

[0027] The memory further includes an unlocking function layer; one or more neural networks in the unlocking function layer are used to receive the digital key positioning information and unlock the first device based on the digital key positioning information.

[0028] In summary, this application proposes a digital key positioning method and a vehicle unlocking system. The digital key positioning method first trains a multi-task neural network. The multi-task neural network can simultaneously calculate distance, area probability, and direction probability. The Bluetooth RSSI values ​​of multiple modules in the first device and the second device are input into the multi-task neural network to obtain distance, area probability, and direction probability. Then, the area where the second device is located is determined based on the area probability. When the second device is in the unlocking zone, the direction is determined based on the direction probability to obtain the direction of the second device. Finally, the area and direction of the second device are output.

[0029] Compared with the prior art, this application has at least the following beneficial effects:

[0030] This application employs a multi-task training strategy to obtain a multi-task neural network capable of simultaneously predicting distance, region, and direction. By combining distance and location information, the model can quickly and accurately output the user's location and direction, thus providing more precise real-time data support for the positioning system. This method not only improves the response speed of the positioning system but also effectively reduces the impact of environmental interference or sensor errors, ensuring high reliability of the positioning information. Attached Figure Description

[0031] Figure 1 is a flowchart of the digital key positioning method according to an embodiment of the present invention.

[0032] Figure 2 is a reasoning diagram of the digital key positioning method shown in an embodiment of the present invention.

[0033] Figure 3 is a flowchart illustrating the multi-task neural network training process according to an embodiment of the present invention.

[0034] Figure 4 is a framework diagram of a vehicle unlocking system according to an embodiment of the present invention.

[0035] Figure 5 is a framework diagram of the cockpit domain controller shown in an embodiment of the present invention.

[0036] Figure 6 is a memory framework diagram illustrating an embodiment of the present invention.

[0037] Figure 7 is a diagram of a multi-task neural network framework shown in an embodiment of the present invention.

[0038] Figure 8 is a hidden layer framework diagram shown in an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0040] Example 1:

[0041] As shown in Figure 1, this application proposes a digital key positioning method, which specifically includes:

[0042] S100: Input the Bluetooth RSSI values ​​of multiple modules in the first device and the second device into a multi-task neural network to obtain initial positioning information; wherein, the initial positioning information includes at least distance, area probability and direction probability.

[0043] S200: Obtain the area where the second device is located based on the distance and area probability, and when the area where the second device is located is an unlocked area, determine the direction based on the direction probability to obtain the direction where the second device is located.

[0044] S300: Obtain digital key positioning information based on the area where the second device is located and the direction of the second device.

[0045] In this embodiment, a neural network model is used to predict distance, region, and direction simultaneously, thereby improving positioning efficiency and accuracy, determining the user's region and direction relative to the car, and quickly and accurately outputting the user's location and direction.

[0046] In this embodiment, the first device can be a vehicle, and the second device can be a user's mobile phone, but neither is limited to these. Figure 2 is a flowchart of a specific digital key location reasoning process according to an embodiment of the present invention.

[0047] In an embodiment of the present invention, optionally, the multi-task neural network training process includes: training an end-to-end neural network model as the multi-task neural network using a multi-task learning strategy; wherein the multi-task learning strategy includes at least three training tasks: distance regression, region classification, and orientation classification.

[0048] In an embodiment of the present invention, optionally, as shown in Figure 3, the multi-task neural network training process includes: acquiring Bluetooth RSSI values ​​between multiple modules in the first device and the second device, performing forward propagation to obtain task materials; extracting features from the task materials to obtain the low-level features of each training task; wherein, different low-level features share information and are constrained; integrating the low-level features of each training task to obtain the high-level features of each training task; and outputting prediction results: distance, region probability, and direction probability. Further, a conditional joint loss function is calculated, automatically ignoring the directional cross-entropy loss in the non-unlocked area; the Bluetooth RSSI values ​​are backpropagated, and gradient descent is used to update the network parameters; and iterative training is performed until the multi-task neural network reaches a preset number of iterations or converges.

[0049] Specifically, the directional labels for non-unlocked areas are assigned empty values, while the directional labels for unlocked areas are assigned normal values; wherein, the non-unlocked areas include at least the locked areas and the vehicle interior.

[0050] A multi-task learning strategy is employed. An end-to-end neural network is used to simultaneously learn multiple labels, including distance, region, and orientation. Due to the correlation between the three tasks (distance, region, and orientation), features from different tasks can be shared, complementing and constraining each other. This effectively alleviates overfitting, improves the model's generalization ability across different tasks, and significantly enhances model performance.

[0051] This invention employs an incompletely labeled dataset to reduce the difficulty of data collection. Since BLE localization typically only requires direction determination for the unlocked region, direction labels cannot be added to data outside the unlocked region in the training dataset. To address this, the present invention assigns null values ​​to direction labels for samples outside the unlocked region and normal values ​​to those in the unlocked region when building the training dataset. A conditional joint loss function is constructed, and during training, the direction classification loss for the unlocked region is adaptively ignored, enabling multi-task learning on an incompletely labeled dataset.

[0052] In this embodiment, a neural network is trained using deep learning to learn the nonlinear relationship between RSSI intensity and distance, resulting in higher positioning accuracy. Deep learning has powerful feature extraction capabilities and can learn very complex nonlinear relationships, leading to higher positioning accuracy compared to traditional distance decay models.

[0053] Deep learning is a subfield of machine learning that places particular emphasis on using deep neural networks to learn the features and patterns of data.

[0054] Designing a neural network model is an important step in the deep learning process. The appropriate network structure is selected based on the specific task (e.g., classification, regression, generation).

[0055] Common deep learning network architectures include:

[0056] Convolutional Neural Networks (CNNs): Primarily used for image processing, they utilize convolutional layers to extract local features.

[0057] Recurrent Neural Networks (RNNs) and their variants (LSTM, GRU): used to process sequential data, such as time series prediction and language models.

[0058] Fully connected neural networks (FNNs): used for general regression and classification tasks.

[0059] Generative Adversarial Networks (GANs): Used for generative models, especially image generation, text generation, etc.

[0060] Number of layers and units: These determine the depth (number of layers) and the number of neurons (width) per layer of the network. Deep networks are generally able to capture more complex features, but they are also more prone to overfitting, so they need to be chosen carefully.

[0061] Commonly used activation functions include:

[0062] ReLU (Rectified Linear Unit): Commonly used in hidden layers to help introduce non-linearity into the model.

[0063] Sigmoid and Tanh: Used in the output layer (especially for binary classification tasks).

[0064] Softmax: The output layer used for multi-class classification tasks.

[0065] Choose an appropriate loss function based on the task:

[0066] For classification tasks, cross-entropy loss is typically used.

[0067] For regression tasks, mean squared error (MSE) loss is typically used.

[0068] In this embodiment, the multi-task neural network is mainly used to calculate distance, region probability, and orientation probability. Since this involves a classification task, a fully connected neural network (FNN) and a cross-entropy loss function are selected. The cross-entropy loss function is also known as the conditional joint loss function, and is as follows:

[0069]

[0070] The specific training process is as follows:

[0071] Forward propagation: The input data is processed through each layer of the neural network to obtain the final prediction result. Through weighted summation and activation functions between layers, the neural network progressively transforms the input data.

[0072] Loss calculation: Based on the difference between the network output and the true label, the loss function is used to calculate the loss value. The smaller the loss value, the better the model performance.

[0073] Backpropagation: Using the chain rule, the gradient of the loss function is calculated with respect to each parameter in the network. These gradients represent the degree of influence of each parameter on the total loss.

[0074] Gradient descent: Updates weights using the gradient obtained through backpropagation. Commonly used optimization algorithms include:

[0075] SGD (Stochastic Gradient Descent): Stochastic gradient descent is simple but sometimes converges slowly.

[0076] Adam: An adaptive optimization algorithm that typically performs well, especially on large datasets.

[0077] Training iteration: The training process is a multi-epoch process. In each iteration, the model performs forward and backward propagation on the training set, continuously optimizing the parameters until the loss function converges.

[0078] Backpropagation and gradient descent are key steps in the training process. When training a neural network, backpropagation is used to calculate the gradient of each parameter, which is the partial derivative of the loss function with respect to each weight and bias. Gradient descent uses the gradients calculated by backpropagation to update the parameters, thereby minimizing the loss function.

[0079] Specifically, forward propagation is used to obtain the output and calculate the loss function; backpropagation is used to calculate the gradient of each layer based on the loss function; and gradient descent is used to update the model's parameters (weights and biases).

[0080] Backpropagation can calculate the gradient of the loss function with respect to the output layer (i.e., the partial derivative of the loss function with respect to the output of each neuron).

[0081] Then, this gradient information is propagated backward layer by layer. The gradient of each layer propagates to the previous layer using the chain rule, until it reaches the input layer. The gradient information of each layer helps to calculate the updated values ​​of the weights and biases of each layer.

[0082] Backpropagation utilizes the chain rule, which states that if a function is a composite function composed of multiple functions, the chain rule helps calculate the derivative of the composite function with respect to the input. In a neural network, the output of each layer is obtained by transforming the input of the previous layer through an activation function. Backpropagation uses the chain rule to propagate the error from the output layer back to each layer and calculate the gradient of each layer.

[0083] Gradient descent is an optimization algorithm that aims to minimize the loss function by iteratively adjusting the parameters of a model (such as weights and biases in a neural network). Its basic idea is to calculate the gradient of the loss function with respect to each parameter (i.e., the derivative of the loss function) and update the parameters along the direction of gradient descent, gradually approaching the minimum value of the loss function.

[0084] In each iteration, the gradient descent algorithm calculates the gradient of the loss function with respect to the model parameters. This gradient represents the direction in which the loss function changes most rapidly at a given point.

[0085] The direction of the gradient indicates how the model parameters should be adjusted to minimize the loss.

[0086] Once the gradient is obtained, the model parameters are updated using the negative gradient direction. That is, if the gradient is positive, the parameter values ​​are decreased; if the gradient is negative, the parameter values ​​are increased.

[0087] The updated formula is: [ \theta_{new} = \theta_{old} - \eta \cdot \nabla_\theta J(\theta) ] ; where ( \theta ) are model parameters (e.g., weights and biases), ( \eta ) is the learning rate (step size), which controls the magnitude of each update, and ( \nabla_\theta J(\theta) ) is the gradient of the loss function ( J(\theta) ) with respect to the parameters ( \theta ).

[0088] The gradient descent algorithm repeats the above process until the loss function converges to a minimum point, or the number of iterations reaches a preset upper limit.

[0089] In deep learning, low-level features refer to the basic and simple features extracted from the raw input data in the initial stages. They are usually concrete and direct, containing the basic components of the data. These features tend to be less abstract and are mainly used to describe the simple properties of the data.

[0090] In deep learning, high-level features are more abstract and semantically rich features in the data, typically obtained through training multi-layer networks during deep learning and machine learning. They help models better understand and process complex data.

[0091] In an embodiment of the present invention, optionally, obtaining the area where the second device is located includes: obtaining a preliminary area based on the area probability, and determining whether the preliminary area exceeds a predetermined range based on the distance; if so, correcting the preliminary area to obtain a corrected area; otherwise, taking the preliminary area as the area where the second device is located.

[0092] After obtaining the correction area, the method further includes: determining whether the correction area exceeds a predetermined range; if so, performing a second correction on the correction area until the correction area does not exceed the predetermined range, and then using the correction area as the area where the second device is located.

[0093] In real-world scenarios, area information cannot be completely accurate and may be affected by factors such as signal attenuation and environmental interference. Therefore, distance is needed to estimate and correct for the area.

[0094] In an embodiment of the present invention, optionally, obtaining the orientation of the second device includes: determining the orientation of the second device relative to the first device by using the angle information between the multiple modules in the first device and the second device, and combining the orientation probability, so as to use the orientation of the second device as the orientation of the second device.

[0095] In an embodiment of the present invention, optionally, after obtaining the area where the second device is located, the process further includes: ending the current digital key positioning process when the area where the second device is located is a locked area, or when the area where the second device is located is inside a vehicle.

[0096] Directional probability is a statistical measure used to represent the likelihood of the second device's orientation. By combining angle information and directional probability, the orientation of the second device relative to the first device can be determined more accurately. By collecting angle information from multiple modules and combining it with directional probability techniques, the orientation of the second device relative to the first device can be inferred with relatively high precision. This method typically combines angle difference analysis with statistical inference to achieve high-precision orientation estimation.

[0097] In this embodiment, by combining angle information and direction probability, the direction of the second device relative to the first device can be determined more accurately. Specific steps may include:

[0098] Two devices acquire relevant angular information (such as the angle relative to the ground or a fixed reference object) through sensors. For example, the first device might know that its orientation is 10°, and the second device's orientation is 50°.

[0099] By combining directional probabilities, the orientation of the second device relative to the first device can be inferred. For example, if device 1 knows it is facing 10° and device 2 is facing 50°, based on the known angle difference and the directional probabilities, the precise orientation of the second device relative to the first device can be calculated.

[0100] In practical applications, direction estimation may be affected by noise and uncertainty. It can be optimized by using angle information and direction probability to reduce errors and obtain a more accurate relative direction.

[0101] Implementation 2:

[0102] As shown in Figure 4, this application proposes a vehicle unlocking system, which includes at least a first device 10 and a second device 20. The first device 10 is equipped with a cockpit domain controller 30. The cockpit domain controller 30 is used to receive Bluetooth signals sent by the second device 20, obtain digital key positioning information based on the Bluetooth signals, and unlock the first device 10 according to the digital key positioning information.

[0103] In an embodiment of the present invention, optionally, when the second device is a mobile phone and the first device is a vehicle, the output digital key positioning information can be used to determine whether the user is approaching the vehicle and from which direction. For example, when it is detected that the user is located in a specific area near the vehicle and is facing the door, the vehicle can automatically unlock the door to realize the keyless entry function.

[0104] As shown in Figure 5, the cockpit domain controller 30 includes at least a memory 301 and a processor 302; the memory 301 is used to store computer programs for multiple functional layers, each functional layer including one or more neural networks, the one or more neural networks being associated with operating elements for controlling vehicle unlocking; the processor 302 communicates with the memory 301 and is used to execute the computer program for each of the multiple functional layers stored in the memory 301.

[0105] As shown in Figure 6, the memory 301 includes at least a signal conversion function layer 303 and a digital key positioning function layer 304. One or more neural networks in the signal conversion function layer 303 are used to receive Bluetooth signals emitted by the second device 20 and convert the Bluetooth signals into Bluetooth RSSI values. The digital key positioning function layer 304 includes a multi-task neural network used to receive Bluetooth RSSI values ​​between multiple modules within the first device 10 and the second device, and to output digital key positioning information based on the Bluetooth RSSI values. The memory 301 also includes an unlocking function layer 305; one or more neural networks in the unlocking function layer 305 are used to receive the digital key positioning information and unlock the first device 10 based on the digital key positioning information.

[0106] As shown in Figure 7, the multi-task neural network includes at least an input layer 3041, a hidden layer 3042, and an output layer 3043. The input layer 3041 is used to receive Bluetooth RSSI values ​​between multiple modules in the first device 10 and the second device. The hidden layer 3042 is used to obtain initial positioning information based on the Bluetooth RSSI values, and to obtain the area and direction of the second device based on the initial positioning information. The output layer 3043 is used to output the area and direction of the second device as digital key positioning information.

[0107] As shown in Figure 8, the hidden layer 3042 includes at least an initial positioning information generation module 3044, an area acquisition module 3045, and a direction determination module 3046. The initial positioning information generation module 3044 is used to obtain initial positioning information based on the Bluetooth RSSI value. The initial positioning information includes at least distance, area probability, and direction probability. The area acquisition module 3045 is used to obtain the area where the second device is located based on the distance and area probability. The direction determination module 3046 is used to determine the direction of the second device based on the direction probability when the area where the second device is located is an unlocked area.

[0108] In an embodiment of the present invention, optionally, the initial positioning information generation module 3044 obtains the distance, area probability and direction probability based on the input Bluetooth RSSI value. This module will use the learning ability of neural networks to establish a mapping relationship between the distance, area probability and direction probability and the Bluetooth RSSI value by learning from a large amount of historical data. This can be achieved by using a fully connected layer or a multilayer perceptron.

[0109] The area acquisition module 3045 obtains a preliminary area based on the area probability in the initial positioning information, and then determines whether the preliminary area exceeds the predetermined range by combining the distance information. If it exceeds the predetermined range, the preliminary area will be corrected, possibly once or multiple times, until the corrected area does not exceed the predetermined range, and finally the area where the second device is located is determined.

[0110] The direction determination module 3046 determines the direction of the second device relative to the first device by using the angle information between multiple modules in the first device and the second device, and by combining the direction probability in the initial positioning information. The angle information can provide a more accurate direction reference, and the combination with the direction probability can improve the accuracy of the direction judgment.

[0111] In summary, this application proposes a digital key positioning method and a vehicle unlocking system. The digital key positioning method first trains a multi-task neural network. The multi-task neural network can simultaneously calculate distance, area probability, and direction probability. The Bluetooth RSSI values ​​of multiple modules in the first device and the second device are input into the multi-task neural network to obtain distance, area probability, and direction probability. Then, the area where the second device is located is determined based on the area probability. When the second device is in the unlocking zone, the direction is determined based on the direction probability to obtain the direction of the second device. Finally, the area and direction of the second device are output.

[0112] This application can utilize a neural network model to simultaneously predict distance, region, and direction, thereby improving positioning efficiency and accuracy, determining the user's region and direction relative to the vehicle, and quickly and accurately outputting the user's location and direction.

[0113] In the several embodiments provided in this application, it will be understood that each block in the flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the figures. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved.

[0114] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 an electronic device to execute all or part of the steps of the methods described in the various 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.

[0115] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application for those skilled in the art.

Claims

1. A digital key positioning method, the digital key positioning method specifically comprising: The Bluetooth RSSI values ​​of multiple modules in the first device and the second device are input into a multi-task neural network to obtain initial positioning information; wherein, the initial positioning information includes at least distance, area probability and direction probability (S100). The region where the second device is located is obtained based on the distance and region probability. When the region where the second device is located is an unlocked area, the direction is determined based on the direction probability to obtain the direction of the second device (S200). And, digital key positioning information is obtained based on the area where the second device is located and the direction where the second device is located (S300).

2. The digital key positioning method according to claim 1, before obtaining the initial positioning information, includes: An end-to-end neural network model is trained using a multi-task learning strategy as the multi-task neural network; wherein the multi-task learning strategy includes at least three training tasks: distance regression, region classification, and orientation classification.

3. The digital key positioning method according to claim 2, wherein training an end-to-end neural network model comprises: Obtain the Bluetooth RSSI values ​​of multiple modules in the first device and the second device, perform forward propagation, and obtain task materials; Feature extraction is performed on the task materials to obtain the underlying features of each training task; among them, information sharing and constraints are performed between different underlying features. The low-level features of each training task are integrated to obtain the high-level features of each training task. In addition, the output prediction results are: distance, region probability, and direction probability.

4. The digital key positioning method according to claim 3, wherein training an end-to-end neural network model further includes: Calculate the conditional joint loss function, automatically ignoring the directional cross-entropy loss in the non-unlocked area; The Bluetooth RSSI value is backpropagated, and the network parameters are updated using the gradient descent method. Furthermore, iterative training continues until the multi-task neural network reaches the preset number of iterations or converges.

5. The digital key positioning method according to claim 3, wherein the output prediction result further includes: The directional labels for non-unlocked areas are assigned empty values, while the directional labels for unlocked areas are assigned normal values; wherein, the non-unlocked areas include at least the locked areas and the interior of the vehicle.

6. The digital key positioning method according to claim 1, wherein obtaining the area where the second device is located includes: A preliminary region is obtained based on the region probability, and based on the distance, it is determined whether the preliminary region exceeds a predetermined range. If so, the preliminary region is corrected to obtain a corrected region. Otherwise, the preliminary area shall be taken as the area where the second device is located.

7. The digital key positioning method according to claim 6, further comprising, after obtaining the correction area: If the correction area exceeds a predetermined range, then the correction area is corrected a second time until it does not exceed the predetermined range, and then the correction area is taken as the area where the second device is located.

8. The digital key positioning method according to claim 1, wherein obtaining the orientation of the second device includes: The orientation of the second device relative to the first device is determined by using the angle information between multiple modules in the first device and the second device, and by combining the orientation probability, so as to determine the orientation of the second device.

9. The digital key positioning method according to claim 7, further comprising, after obtaining the area where the second device is located: The digital key location process ends when the area where the second device is located is a locked area, or when the area where the second device is located is inside the vehicle.

10. A vehicle unlocking system, the system comprising at least a first device (10) and a second device (20), wherein the first device (10) is configured with a cockpit domain controller (30). The cockpit domain controller (30) is used to receive the Bluetooth signal sent by the second device (20) to obtain digital key positioning information based on the Bluetooth signal, and to unlock the first device (10) according to the digital key positioning information.

11. The vehicle unlocking system according to claim 10, wherein the cockpit domain controller (30) includes at least a memory (301) and a processor (302). The memory (301) is used to store computer programs for multiple functional layers, each functional layer including one or more neural networks, the one or more neural networks being associated with operating elements for controlling vehicle unlocking; The processor (302) communicates with the memory (301) to execute computer programs for each of the multiple functional layers stored in the memory (301).

12. The vehicle unlocking system according to claim 11, wherein the memory (301) includes at least a signal conversion function layer (303) and a digital key positioning function layer (304). One or more neural networks of the signal conversion function layer (303) are used to receive Bluetooth signals emitted by the second device (20) and convert the Bluetooth signals into Bluetooth RSSI values; The digital key positioning function layer (304) includes a multi-task neural network for receiving Bluetooth RSSI values ​​from multiple modules within the first device (10) and the second device, and outputting digital key positioning information based on the Bluetooth RSSI values.

13. The vehicle unlocking system according to claim 12, wherein the multi-task neural network includes at least an input layer (3041), a hidden layer (3042), and an output layer (3043). The input layer (3041) is used to receive the Bluetooth RSSI values ​​of multiple modules in the first device (10) and the second device; The hidden layer (3042) is used to obtain initial positioning information based on the Bluetooth RSSI value, and to obtain the area where the second device is located and the direction of the second device based on the initial positioning information; The output layer (3043) is used to output the area where the second device is located and the direction where the second device is located, as digital key positioning information.

14. The vehicle unlocking system according to claim 12, wherein the hidden layer (3042) includes at least an initial positioning information generation module (3044), an area acquisition module (3045), and a direction determination module (3046). The initial positioning information generation module (3044) is used to obtain initial positioning information based on the Bluetooth RSSI value; wherein... The initial positioning information includes at least distance, region probability, and direction probability; The region acquisition module (3045) is used to obtain the region where the second device is located based on the distance and the region probability. The direction determination module (3046) is used to determine the direction of the second device based on the direction probability when the area where the second device is located is an unlocked area.

15. The vehicle unlocking system according to claim 12, wherein the memory (301) further comprises an unlocking function layer (305). One or more neural networks of the unlocking function layer (305) are used to receive the digital key positioning information and unlock the first device (10) based on the digital key positioning information.

Images