Digital key positioning method and system and storage medium

Abstract

The invention provides a digital key positioning method and system and a storage medium. The digital key positioning method comprises the following steps: firstly, training a multi-task neural network; the multi-task neural network can calculate the distance, the area probability and the direction probability at the same time; inputting the Bluetooth RSSI values of the plurality of modules in the first device and the second device into a multi-task neural network to obtain a distance, an area probability and a direction probability; judging an area where the second equipment is located according to the area probability; when it is judged that the second equipment is located in the unlocking area, direction judgment is conducted according to the direction probability, and the direction where the second equipment is located is obtained; and finally, outputting the area and the direction of the second equipment. According to the invention, one neural network model can be utilized to predict the distance, the area and the direction at the same time, the positioning efficiency and accuracy are improved, the area and the direction of the user relative to the automobile are judged, and the area and the direction of the user are rapidly and accurately output.

Description

Technical Field

[0001] The present application relates to the field of smart car technology, and in particular to a digital key positioning method, system and storage medium. Background Art

[0002] The BLE digital key uses Bluetooth Low Energy (BLE) to enable functions such as seamless entry into the car, seamless locking of the car when walking away, and automatic welcome when approaching.

[0003] The BLE positioning algorithm needs to determine the area (inside the car, unlocking area, locking area, etc.) and direction (in front of the car, behind the car, left of the car, right of the car, etc.) of the mobile phone relative to the car in real time. The positioning principle of the BLE digital key is to use the relationship between the RSSI strength and distance between the car Bluetooth module and the mobile phone to calculate the distance between the mobile phone and the car, and then determine the area where the mobile phone is located, and determine the direction of the mobile phone relative to the car based on the strength relationship between each Bluetooth module. Deep learning technology, with its powerful feature extraction ability, can learn the nonlinear relationship between the RSSI of multiple modules on the car and the distance and direction of the mobile phone, and has been successfully applied to BLE positioning. However, the current BLE positioning based on deep learning generally only predicts the distance. The judgment of direction and area is obtained by post-processing after the real-time distance is obtained using the neural network.

[0004] Traditional BLE positioning algorithms are generally based on distance attenuation models and artificial rules to complete the above positioning tasks. The disadvantage is that the simple distance attenuation model cannot model the complex nonlinear relationship between the RSSI strength and distance of multiple vehicle-mounted Bluetooth modules, resulting in high positioning errors.

[0005] The traditional BLE positioning algorithm is affected by the low computing power of the MCU chip, the neural network model has fewer parameters, and the risk of overfitting in single-task learning is greater, resulting in poor model generalization ability.

[0006] The distance-based post-processing logic needs to be manually formulated, which is generally more complex, less efficient, and difficult to guarantee accuracy. In addition, the additional post-processing logic will increase the algorithm's calculation time and memory space usage. Summary of the invention

[0007] To solve the above technical problems, the present application proposes a digital key positioning method, system and storage medium.

[0008] In a first aspect, the present application proposes a digital key positioning method, and the digital key positioning method specifically includes:

[0009] S1: Input the Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network to obtain the distance, area probability and direction probability;

[0010] S2: Determine the area where the second device is located according to the area probability;

[0011] S3: When the second device is in the unlocking area, determine the direction according to the direction probability to obtain the direction of the second device;

[0012] S4: Obtaining digital key positioning information according to the area where the second device is located and the direction where the second device is located.

[0013] The present application can utilize a neural network model to simultaneously predict distance, area, and direction, thereby improving positioning efficiency and accuracy, determining the area and direction of the user relative to the car, and quickly and accurately outputting the area and direction of the user.

[0014] Furthermore, the multi-task neural network training process includes:

[0015] A multi-task learning strategy is used to train an end-to-end neural network model, and three training tasks including distance regression, region classification and direction classification are performed simultaneously.

[0016] Furthermore, the multi-task neural network training process includes:

[0017] S11: Obtain Bluetooth RSSI values ​​of multiple modules in the first device and the second device, obtain task materials and perform forward propagation.

[0018] S12: Extract features from the task materials to obtain underlying features of each training task, and share information and constrain different underlying features;

[0019] S13: Integrate the underlying features of each training task to obtain the high-level features of each training task;

[0020] S14: Output prediction results: distance, area probability and direction probability;

[0021] S15: Calculate the conditional joint loss function, automatically ignoring the directional cross entropy loss of the non-unlocked area;

[0022] S16: back-propagating the prediction results, using the gradient descent method to update the network parameters;

[0023] S17: iterative training until the multi-task neural network reaches a preset number of iterations or converges;

[0024] Among them, the sample direction labels in the non-unlocking area are assigned a null value, and the sample direction labels in the unlocking area are assigned a normal value.

[0025] A multi-task learning strategy is adopted. An end-to-end neural network is used to simultaneously learn multiple labels of distance, area, and direction. Due to the correlation between the three tasks of distance, area, and direction, the features of different tasks can be shared, complemented, and constrained, which can effectively alleviate the problem of model overfitting, improve the model's generalization ability for different tasks, and effectively improve the model performance.

[0026] An incompletely annotated data set is used to reduce the difficulty of data collection. Since general BLE positioning only requires the judgment of the direction of the unlocked area, the data in the non-unlocked area in the training data set cannot be labeled with a direction label. When establishing a training data set, the present invention assigns a null value to the direction label of the non-unlocked area sample and assigns a normal value to the direction label of the unlocked area sample. A conditional joint loss function is constructed to adaptively ignore the direction classification loss of the non-unlocked area during training, thereby realizing multi-task learning on an incompletely annotated data set.

[0027] Furthermore, the step S2 further includes:

[0028] The distance is used to correct and restrict the area where the second device is located.

[0029] Furthermore, the step S2 further includes:

[0030] S21: obtaining a preliminary region through the regional probability;

[0031] S22: Based on the distance, determine whether the preliminary area exceeds a predetermined range. If so, correct the preliminary area. After obtaining the corrected area, proceed to step S23. If not, the preliminary area is the area where the second device is located.

[0032] S23: Determine whether the correction area exceeds a predetermined range. If so, correct the correction area and then proceed to step S23. If not, the correction area is the area where the second device is located.

[0033] In actual scenarios, the 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 the area.

[0034] Furthermore, the step S3 includes:

[0035] The direction of the second device relative to the first device is determined by combining the angle information between the multiple modules in the first device and the second device with the direction probability.

[0036] Furthermore, the step S3 also includes: when the second device is in a locked area, or when the second device is in the vehicle, ending the current digital key positioning process.

[0037] Direction probability is a statistic used to represent the likelihood of the direction of the second device. By combining angle information and direction probability, the direction of the second device relative to the first device can be determined more accurately. By collecting angle information from multiple modules and combining direction probability technology, the direction of the second device relative to the first device can be inferred more accurately. This method usually combines angle difference with statistical reasoning to achieve high-precision direction estimation.

[0038] In a second aspect, the present application proposes a system based on the digital key positioning method described in the first aspect, the system comprising:

[0039] An input module, used to input Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network;

[0040] Multi-task neural network to obtain distance, area probability and direction probability;

[0041] A region determination module, used for determining the region where the second device is located according to the region probability;

[0042] a direction determination module, configured to determine the direction of the second device according to the direction probability when the second device is in the unlocking area, and obtain the direction of the second device;

[0043] An output module is used to output the area and direction of the second device.

[0044] Furthermore, the multi-task neural network also includes:

[0045] The input layer is used to obtain the Bluetooth RSSI values ​​of multiple modules in the first device and the second device, obtain the task material and perform forward propagation;

[0046] The backbone network is used to extract features from the task materials to obtain the underlying features of each training task;

[0047] The Head layer is used to integrate the underlying features of each training task to obtain the high-level features of each training task;

[0048] The output layer is used to output the prediction results: distance, area probability and direction probability, and back-propagate the prediction results. The output layer also includes a calculation unit for calculating the conditional joint loss function, automatically ignoring the directional cross entropy loss of the non-unlocked area.

[0049] In a third aspect, the present application proposes a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a control processor, the digital key positioning method as described in the first aspect is implemented.

[0050] In summary, the present application proposes a digital key positioning method, system and storage medium, wherein the digital key positioning method first trains a multi-task neural network. The multi-task neural network can simultaneously calculate the distance, area probability and direction probability; input the Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network to obtain the distance, area probability and direction probability; then determine the area where the second device is located according to the area probability; determine that when the second device is in the unlocking area, perform direction judgment according to the direction probability to obtain the direction of the second device; finally output the area and direction of the second device.

[0051] Compared with the prior art, this application has at least the following beneficial effects:

[0052] This application uses a multi-task training strategy to obtain a multi-task neural network that can simultaneously predict distance, area, and direction. By combining distance information and location information, the model can quickly and accurately output the area and direction of the user, thereby providing more accurate 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 caused by environmental interference or sensor errors, ensuring the high reliability of positioning information. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a digital key positioning method according to an embodiment of the present invention.

[0054] Figure 2 This is a reasoning diagram of a digital key positioning method shown in an embodiment of the present invention.

[0055] Figure 3 This is a flow chart of a multi-task neural network training process according to an embodiment of the present invention.

[0056] Figure 4 This is a structural diagram of a character key positioning system according to an embodiment of the present invention.

[0057] Figure 5 A multi-task neural network structure diagram shown in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0059] Embodiment 1:

[0060] like Figure 1 As shown, the present application proposes a digital key positioning method, which specifically includes:

[0061] S1: Input the Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network to obtain the distance, area probability and direction probability;

[0062] S2: Determine the area where the second device is located according to the area probability;

[0063] S3: When the second device is in the unlocking area, determine the direction according to the direction probability to obtain the direction of the second device;

[0064] S4: Obtaining digital key positioning information according to the area where the second device is located and the direction where the second device is located.

[0065] In this embodiment, a neural network model is used to simultaneously predict the distance, area and direction, improve the positioning efficiency and accuracy, determine the area and direction of the user relative to the car, and quickly and accurately output the area and direction of the user.

[0066] In this embodiment, the first device may be a vehicle, and the second device may be a user's mobile phone, but the present invention is not limited thereto. Figure 2 This is a specific digital key positioning reasoning flow chart of an embodiment of the present invention.

[0067] In an embodiment of the present invention, optionally, the multi-task neural network training process includes:

[0068] A multi-task learning strategy is used to train an end-to-end neural network model, and three training tasks including distance regression, region classification and direction classification are performed simultaneously.

[0069] In the embodiment of the present invention, optionally, Figure 3 As shown, the multi-task neural network training process includes:

[0070] S11: Obtain Bluetooth RSSI values ​​of multiple modules in the first device and the second device, obtain task materials and perform forward propagation.

[0071] S12: Extract features from the task materials to obtain underlying features of each training task, and share information and constrain different underlying features;

[0072] S13: Integrate the underlying features of each training task to obtain the high-level features of each training task;

[0073] S14: Output prediction results: distance, area probability and direction probability;

[0074] S15: Calculate the conditional joint loss function, automatically ignoring the directional cross entropy loss of the non-unlocked area;

[0075] S16: back-propagating the prediction results, using the gradient descent method to update the network parameters;

[0076] S17: iterative training until the multi-task neural network reaches a preset number of iterations or converges;

[0077] Among them, the sample direction labels in the non-unlocking area are assigned a null value, and the sample direction labels in the unlocking area are assigned a normal value.

[0078] A multi-task learning strategy is adopted. An end-to-end neural network is used to simultaneously learn multiple labels of distance, area, and direction. Due to the correlation between the three tasks of distance, area, and direction, the features of different tasks can be shared, complemented, and constrained, which can effectively alleviate the problem of model overfitting, improve the model's generalization ability for different tasks, and effectively improve the model performance.

[0079] An incompletely annotated data set is used to reduce the difficulty of data collection. Since general BLE positioning only requires the judgment of the direction of the unlocked area, the data in the non-unlocked area in the training data set cannot be labeled with a direction label. When establishing a training data set, the present invention assigns a null value to the direction label of the non-unlocked area sample and assigns a normal value to the direction label of the unlocked area sample. A conditional joint loss function is constructed to adaptively ignore the direction classification loss of the non-unlocked area during training, thereby realizing multi-task learning on an incompletely annotated data set.

[0080] In this embodiment, deep learning is used to train a neural network to learn the nonlinear relationship between RSSI intensity and distance, and the positioning accuracy is higher. Deep learning has a powerful feature extraction capability and can learn very complex nonlinear relationships. Compared with the traditional distance attenuation model, the positioning accuracy is higher.

[0081] Deep learning is a subfield of machine learning that places particular emphasis on using deep neural networks to learn features and patterns in data.

[0082] Designing a neural network model is an important step in the deep learning process. Depending on the task (such as classification, regression, generation, etc.), choose the appropriate network structure.

[0083] Common deep learning network architectures include:

[0084] Convolutional Neural Network (CNN): Mainly used for image processing, using convolutional layers to extract local features.

[0085] Recurrent Neural Network (RNN) and its variants (LSTM, GRU): used to process sequence data, such as time series prediction, language model, etc.

[0086] Fully connected neural network (FNN): used for general regression and classification tasks.

[0087] Generative Adversarial Network (GAN): used for generating models, especially image generation, text generation, etc.

[0088] Number of layers and units: Determine the depth of the network (number of layers) and the number of neurons in each layer (width). Deep networks are usually able to capture more complex features, but are also more prone to overfitting, so they need to be chosen carefully.

[0089] Commonly used activation functions are:

[0090] ReLU (Rectified Linear Unit): Commonly used in hidden layers to help introduce nonlinearity into the model.

[0091] Sigmoid, Tanh: used in the output layer (especially for binary classification tasks).

[0092] Softmax: Output layer for multi-classification tasks.

[0093] Choose an appropriate loss function based on the task:

[0094] For classification tasks, cross-entropy loss is usually used.

[0095] For regression tasks, the mean squared error loss (MSE) is usually used.

[0096] In this embodiment, the multi-task neural network is mainly used to calculate distance, area probability and direction probability, which involves classification tasks, so a fully connected neural network (FNN) and a cross entropy loss function are selected. The cross entropy loss function is a conditional joint loss function, which is as follows:

[0097]

[0098] The specific training process is as follows:

[0099] Forward propagation: The input data is calculated through each layer of the neural network and the prediction result is finally obtained. Through the weighted summation and activation function between layers, the neural network gradually transforms the input data.

[0100] Calculate loss: According to the difference between the network output and the true label, the loss value is calculated using the loss function. The smaller the loss value, the better the model performance.

[0101] Backpropagation: Using the chain rule, we calculate the gradient of the loss function with respect to each parameter in the network. These gradients represent the degree of influence of each parameter on the total loss.

[0102] Gradient Descent: Update weights by back-propagating the gradient. Common optimization algorithms include:

[0103] SGD (Stochastic Gradient Descent): Stochastic gradient descent, simple but sometimes slow to converge.

[0104] Adam: An adaptive optimization algorithm that generally performs well, especially on large datasets.

[0105] Training Iteration: The training process is a multi-epoch process. In each iteration, the model performs forward propagation and backward propagation on the training set, continuously optimizing the parameters until the loss function converges.

[0106] Among them, back propagation and gradient descent are key steps in the training process. When training a neural network, back propagation is used to calculate the gradient of each parameter, that is, the partial derivative of the loss function with respect to each weight and bias. Gradient descent uses the gradient calculated by back propagation to update the parameters, thereby minimizing the loss function.

[0107] Specifically, use forward propagation to get the output and calculate the loss function; use the backpropagation algorithm to calculate the gradient of each layer according to the loss function; use the gradient descent method to update the model parameters (weights and biases).

[0108] Backpropagation calculates 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). This gradient information is then passed back layer by layer. The gradient of each layer is propagated to the previous layer through the chain rule until the input layer. The gradient information of each layer helps calculate the updated values ​​of the weights and biases of each layer.

[0109] The back propagation algorithm uses the chain rule, that is, if there is a composite function composed of multiple functions, the chain rule can help calculate the derivative of the composite function relative to the input. In a neural network, the output of each layer is obtained by transforming the input of the previous layer through the activation function. Back propagation propagates the error from the output layer back to each layer through the chain rule and calculates the gradient of each layer.

[0110] Gradient descent is an optimization algorithm that aims to minimize the value of the loss function by iteratively adjusting the parameters of the model (such as weights and biases in a neural network). The basic idea is to gradually approach the minimum value of the loss function by calculating the gradient of the loss function for each parameter (i.e., the derivative of the loss function) and updating the parameters in the direction of gradient descent.

[0111] At each iteration, the gradient descent algorithm calculates the gradient of the loss function with respect to the model parameters. This gradient indicates the direction in which the loss function changes most rapidly at a certain point.

[0112] The direction of the gradient indicates how the model parameters should be adjusted to reduce the loss as quickly as possible.

[0113] Once the gradient is obtained, the model parameters are updated in the negative gradient direction. That is, if the gradient is positive, the parameter value is decreased; if the gradient is negative, the parameter value is increased.

[0114] The updated formula is: [\theta_{new}=\theta_{old}-\eta\cdot\nabla_\theta J(\theta)] where:

[0115] (\theta) are model parameters (such as weights and biases),

[0116] (\eta) is the learning rate (step size), which controls the magnitude of each update.

[0117] (\nabla_\theta J(\theta)) is the gradient of the loss function (J(\theta)) with respect to the parameters (\theta).

[0118] 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.

[0119] In the process of deep learning, low-level features refer to the more basic and simple features extracted from the original input data at the initial stage of the data. They are usually more specific and direct, and contain the basic components of the data. These features tend to be less abstract and are mainly used to describe simple properties of the data.

[0120] In the process of deep learning, high-level features are more abstract and semantically rich features in the data, which are usually obtained through the training of multi-layer networks in the process of deep learning and machine learning. They can help the model better understand and process complex data.

[0121] In the embodiment of the present invention, optionally, the step S2 further includes:

[0122] The distance is used to correct and restrict the area where the second device is located.

[0123] In the embodiment of the present invention, optionally, the step S2 further includes:

[0124] S21: obtaining a preliminary region through the regional probability;

[0125] S22: Based on the distance, determine whether the preliminary area exceeds a predetermined range. If so, correct the preliminary area. After obtaining the corrected area, proceed to step S23. If not, the preliminary area is the area where the second device is located.

[0126] S23: Determine whether the correction area exceeds a predetermined range. If so, correct the correction area and then proceed to step S23. If not, the correction area is the area where the second device is located.

[0127] In actual scenarios, the 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 the area.

[0128] In the embodiment of the present invention, optionally, step S3 includes:

[0129] The direction of the second device relative to the first device is determined by combining the angle information between the multiple modules in the first device and the second device with the direction probability.

[0130] In the embodiment of the present invention, optionally, the step S3 further includes: when the second device is in a locked area, or when the second device is in the vehicle, ending the current digital key positioning process.

[0131] Direction probability is a statistic used to represent the likelihood of the direction of the second device. By combining angle information and direction probability, the direction of the second device relative to the first device can be determined more accurately. By collecting angle information from multiple modules and combining direction probability technology, the direction of the second device relative to the first device can be inferred more accurately. This method usually combines angle difference with statistical reasoning to achieve high-precision direction estimation.

[0132] In this embodiment, by combining the angle information and the direction probability, the direction of the second device relative to the first device can be determined more accurately. The specific steps may be:

[0133] The two devices collect relevant angle information (such as the angle relative to the ground or a fixed reference object) through sensors. For example, the first device may know that its orientation is 10° and the second device's orientation is 50°.

[0134] Combined with the direction probability, the direction of the second device relative to the first device can be inferred. For example, if device 1 knows that it is facing 10° and device 2 is facing 50°, based on the known angle difference and combined with the direction probability, the precise direction of the second device relative to the first device can be inferred.

[0135] In practical applications, direction estimation may be affected by noise and uncertainty. The angle information and direction probability can be used for optimization to reduce the error and obtain a more accurate relative direction.

[0136] Embodiment 2:

[0137] like Figure 4 As shown, the present application proposes a system based on the digital key positioning method described in Embodiment 1, the system comprising:

[0138] An input module, used to input Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network;

[0139] Multi-task neural network to obtain distance, area probability and direction probability;

[0140] A region determination module, used for determining the region where the second device is located according to the region probability;

[0141] a direction determination module, configured to determine the direction of the second device according to the direction probability when the second device is in the unlocking area, and obtain the direction of the second device;

[0142] An output module is used to output the area and direction of the second device.

[0143] In the embodiment of the present invention, optionally, Figure 5 As shown, the multi-task neural network also includes:

[0144] The input layer is used to obtain the Bluetooth RSSI values ​​of multiple modules in the first device and the second device, obtain the task material and perform forward propagation;

[0145] The backbone network is used to extract features from the task materials to obtain the underlying features of each training task;

[0146] The Head layer is used to integrate the underlying features of each training task to obtain the high-level features of each training task;

[0147] The output layer is used to output the prediction results: distance, area probability and direction probability, and back-propagate the prediction results. The output layer also includes a calculation unit for calculating the conditional joint loss function, automatically ignoring the directional cross entropy loss of the non-unlocked area.

[0148] Embodiment three:

[0149] Based on the same inventive concept, the present application proposes a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a control processor, the digital key positioning method as described in Example 1 is implemented.

[0150] In the computer-readable storage medium, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk SolidState Disk (SSD)), etc.

[0151] In summary, the present application proposes a digital key positioning method, system and storage medium, wherein the digital key positioning method first trains a multi-task neural network. The multi-task neural network can simultaneously calculate the distance, area probability and direction probability; input the Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network to obtain the distance, area probability and direction probability; then determine the area where the second device is located according to the area probability; determine that when the second device is in the unlocking area, perform direction judgment according to the direction probability to obtain the direction of the second device; finally output the area and direction of the second device.

[0152] The present application can utilize a neural network model to simultaneously predict distance, area, and direction, thereby improving positioning efficiency and accuracy, determining the area and direction of the user relative to the car, and quickly and accurately outputting the area and direction of the user.

[0153] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.

[0154] If the functions are implemented in the form of software function 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 the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling an electronic device to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0155] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A digital key positioning method, characterized in that: The digital key positioning method specifically includes: S1: Input the Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network to obtain the distance, area probability and direction probability; S2: Determine the area where the second device is located according to the area probability; S3: When the second device is in the unlocking area, determine the direction according to the direction probability to obtain the direction of the second device; S4: Obtaining digital key positioning information according to the area where the second device is located and the direction where the second device is located.

2. The digital key positioning method according to claim 1, characterized in that: The multi-task neural network comprises: A multi-task learning strategy is used to train an end-to-end neural network model; wherein the multi-task learning strategy includes three training tasks: distance regression, region classification and direction classification.

3. The digital key positioning method according to claim 2, characterized in that: The multi-task neural network comprises: S11: Obtain Bluetooth RSSI values ​​of multiple modules in the first device and the second device, perform forward propagation and obtain task materials; S12: Extract features from the task materials to obtain underlying features of each training task, and share information and constrain different underlying features; S13: Integrate the underlying features of each training task to obtain the high-level features of each training task; S14: Output prediction results: distance, area probability and direction probability; S15: Calculate the conditional joint loss function, automatically ignoring the directional cross entropy loss of the non-unlocked area; S16: Back-propagating the Bluetooth RSSI value, using a gradient descent method to update network parameters; S17: iterative training until the multi-task neural network reaches a preset number of iterations or converges; The step S14 further includes: assigning a null value to the direction tag of the non-unlocking area, and assigning a normal value to the direction tag of the unlocking area.

4. The digital key positioning method according to claim 1, characterized in that: The step S2 further comprises: The distance is used to correct and restrict the area where the second device is located.

5. The digital key positioning method according to claim 4, characterized in that: The step S2 further comprises: S21: obtaining a preliminary region through the regional probability; S22: Based on the distance, determine whether the preliminary area exceeds a predetermined range. If so, correct the preliminary area. After obtaining the corrected area, proceed to step S23. If not, the preliminary area is the area where the second device is located. S23: Determine whether the correction area exceeds a predetermined range. If so, correct the correction area and then proceed to step S23. If not, the correction area is the area where the second device is located.

6. The digital key positioning method according to claim 1, characterized in that: The step S3 comprises: The direction of the second device relative to the first device is determined by combining the angle information between the multiple modules in the first device and the second device with the direction probability.

7. The digital key positioning method according to claim 6, characterized in that: The step S3 further includes: when the second device is in the locked area, or When the second device is in the car, the digital key positioning process is ended.

8. A system based on the digital key positioning method according to any one of claims 1 to 7, the system comprising: An input module, used to input Bluetooth RSSI values ​​of multiple modules in the first device and the second device into the multi-task neural network; Multi-task neural network to obtain distance, area probability and direction probability; A region determination module, used for determining the region where the second device is located according to the region probability; a direction determination module, configured to determine the direction of the second device according to the direction probability when the second device is in the unlocking area, and obtain the direction of the second device; An output module is used to output the area and direction of the second device.

9. The system of claim 8, wherein the multi-task neural network further comprises: The input layer is used to obtain the RSSI values ​​of multiple modules and perform forward propagation or backward propagation; The backbone network is used to extract features from the task materials to obtain the underlying features of each training task; The Head layer is used to integrate the underlying features of each training task to obtain the high-level features of each training task; The output layer is used to output the prediction results: distance, area probability and direction probability, and back-propagate the prediction results. The output layer also includes a calculation unit for calculating the conditional joint loss function, automatically ignoring the directional cross entropy loss of the non-unlocked area.

10. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by the control processor, the digital key positioning method as described in any one of claims 1-7 is implemented.

Images