Living body detection model training method and system, detection method and system, medium and terminal

Abstract

The invention provides a living body detection model training method and system, a detection method and system, a medium and a terminal. The living body detection model training method comprises the steps of obtaining initial radar data in a preset area range, and obtaining living body input data based on the initial radar data and corresponding time sequence data; then constructing an initial living body detection model based on a one-dimensional convolutional network and an FIRE module; and further training the initial living body detection model through the living body input data to obtain a target living body detection model. According to the method, more abundant and more characteristic data can be provided by integrating the time sequence data, and misjudgment caused by environmental interference is reduced, so that the distinguishing capability of the model on a living body and a non-living body is improved; and the one-dimensional convolutional network and the FIRE module are adopted, so that the prediction accuracy can be improved, the calculation complexity is reduced, and the model prediction efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of liveness detection technology, and in particular to a liveness detection model training method, detection method, system, medium and terminal. Background Art

[0002] In modern society, with increasing attention paid to vehicle safety and intelligent driving, the need for liveness detection after a locked vehicle has become increasingly prominent. Radar-based liveness detection technology has become a research hotspot, aiming to promptly detect children, pets, and other individuals left behind in vehicles to prevent tragedies.

[0003] However, existing technologies have obvious flaws. In terms of accuracy, radar signals are easily affected by the complex environment inside the car, such as reflections from metal parts and electromagnetic interference from electronic equipment. This makes it difficult to extract signal features and accurately distinguish between living and non-living targets, resulting in large errors in detection results. In terms of efficiency, existing algorithms have high computational complexity, cumbersome signal processing and analysis processes, and a long time is required from data collection to obtaining detection results, which cannot meet the needs of rapid response. This not only affects the user experience, but can also have serious consequences in emergency situations. Therefore, there is an urgent need for a liveness detection model training method that can improve accuracy and efficiency. Summary of the Invention

[0004] In order to solve the above technical problems, the present application provides a liveness detection model training method, detection method, system, medium and terminal that can improve the efficiency and accuracy of liveness detection.

[0005] Specifically, the present application provides a method for training a liveness detection model, comprising the following steps: Acquire initial radar data within a preset area, and obtain liveness input data based on the initial radar data and corresponding time series data; construct an initial liveness detection model based on a one-dimensional convolutional network and a FIRE module; and train the initial liveness detection model using the liveness input data to obtain a target liveness detection model.

[0006] In the above technical solution, the integration of time series information can capture the temporal changes in living body movements, breathing, etc. Compared with the existing technology that relies only on a single radar signal, it can provide richer and more characteristic data, reduce misjudgments caused by environmental interference, and thus improve the model's ability to distinguish between living and non-living objects; the one-dimensional convolutional network has low computing power and fast time in deep learning, and the introduction of the FIRE module further improves the prediction accuracy while reducing the computational complexity, thereby improving the model's prediction efficiency.

[0007] Furthermore, obtaining the living input data includes: performing a downsampling operation on the initial radar data to obtain a first radar data sample; and obtaining time series data corresponding to the first radar data sample, and embedding the time series data into the first radar data sample to obtain a second radar data sample.

[0008] In the above technical solution, the amount of data is reduced through downsampling operations, which greatly reduces the computing resources required for subsequent model training and data processing, and improves overall computing efficiency. This is especially true when processing large-scale radar data. Time series data reflects the changes in radar signals over time. Physiological activities such as movement and breathing of living organisms have unique patterns and regularities in the time dimension. Integrating time series information into radar data provides the model with richer feature information, enabling the model to better capture the dynamic characteristics of living organisms, enhance the feature expression ability of the data, and help improve the accuracy and robustness of the liveness detection model.

[0009] Furthermore, the obtaining of live input data also includes: converting the second radar data sample into a multi-channel radar data sample through dimensional reconstruction to serve as the live input data; wherein, it also includes: dividing the live input data into a training set, a validation set, and a test set.

[0010] In the above technical solution, dimensional reconstruction converts the second radar data sample into a multi-channel radar data sample. Each channel can represent different types of feature information. The multi-channel setting enables the model to analyze and learn the data from multiple angles, greatly enriching the feature expression of the data, allowing the model to more comprehensively and accurately capture the characteristic patterns of living organisms, thereby improving the accuracy of living organism detection. The reasonable division of the training set, validation set, and test set helps the model learn the inherent laws of the data, rather than just memorizing specific samples in the training data.

[0011] Furthermore, the initial liveness detection model includes at least a convolutional layer, a squeeze layer, an expand layer and a fully connected layer; the training of the initial liveness detection model includes: initializing the model parameters of the initial liveness detection model; inputting the training set into the initial liveness detection model to perform feature extraction through the convolutional layer, squeeze layer and expand layer in sequence, and performing classification operations through the fully connected layer to obtain a liveness prediction output result; and updating the model parameters based on the liveness prediction output result and a preset true label to iteratively train the initial liveness detection model until the model converges.

[0012] In the above technical solution, the convolution layer can automatically extract local features from the in vivo input data, and perform convolution operations by sliding the convolution kernel on the data, which can capture feature patterns at different scales; the Squeeze layer performs global average pooling on the feature map output by the convolution layer, compressing the features of each channel into a single value, thereby obtaining the global feature information of the channel, reducing the dimension of the data and the amount of computation; the Expand layer expands the features at different scales; this structural design enables the model to process features at different levels and scales, enhancing the feature expression capability, while reducing the complexity of the model and improving computational efficiency.

[0013] In addition, the fully connected layer integrates the features extracted by the previous layers and maps the high-dimensional feature vector to a low-dimensional output space to achieve the classification of living and non-living objects.

[0014] Furthermore, in the process of training the initial liveness detection model, it also includes: preliminarily evaluating the model performance of the initial liveness detection model based on the validation set, so as to adjust the model parameters according to the preliminary evaluation results.

[0015] In the above technical solution, preliminary evaluation through the validation set can promptly detect whether the model has a tendency to overfit. It can also try different hyperparameter combinations and observe the performance changes of the model on the validation set to find the optimal hyperparameter settings.

[0016] Furthermore, after the initial liveness detection model training is completed, it also includes: performing a final evaluation on the model performance based on the test set to obtain a target liveness detection model according to the final evaluation result.

[0017] In the above technical solution, test set evaluation can comprehensively verify the model's various performance indicators, such as accuracy, recall rate, F1 value, specificity, etc. These indicators reflect the performance of the model from different perspectives. By comprehensively evaluating these indicators on the test set, we can have a more comprehensive understanding of the strengths and weaknesses of the model, thereby making targeted improvements and optimizations to the model, thereby improving the model's reliability and accuracy.

[0018] Furthermore, based on the same concept, the present application also provides a liveness detection method, including: obtaining multiple frames of radar signals within a set time length, and inputting the radar signals into a target liveness detection model to output liveness detection results in the target vehicle; wherein, the target liveness detection model is obtained through the liveness detection model training method described above.

[0019] In the above technical solution, obtaining multiple frames of radar signals within a set time period can provide richer information than single-frame signals, making the detection results more stable and reliable, and improving the accuracy of liveness detection; the target liveness detection model has undergone a series of scientific training processes and can accurately distinguish between living and non-living targets in the vehicle, maintaining high detection accuracy even in complex environments.

[0020] Furthermore, based on the same concept, the present application also provides a liveness detection system, comprising: An acquisition module is used to acquire radar signals within a set time period and input the radar signals into a target liveness detection model; and a detection module is used to perform liveness detection in the target vehicle using the target liveness detection model and output the liveness detection result.

[0021] In the above technical solution, the intelligent liveness detection function can enhance the safety and technological sense of the vehicle and increase the competitiveness of the product; at the same time, users can also use the vehicle more confidently without worrying about potential dangers caused by negligence, thereby improving the overall user experience.

[0022] Furthermore, based on the same concept, the present application also provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned liveness detection method when running.

[0023] Furthermore, based on the same concept, the present application also provides a vehicle-mounted terminal, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the above-mentioned liveness detection method.

[0024] Compared with the prior art, the present invention has the following advantages: This application first obtains initial radar data within a preset area, and obtains liveness input data based on the initial radar data and the corresponding time series data; constructs an initial liveness detection model based on a one-dimensional convolutional network and a FIRE module; and trains the initial liveness detection model using the liveness input data to obtain a target liveness detection model. By incorporating time series data, this application can provide richer and more characteristic data, reduce misjudgments caused by environmental interference, and thus improve the model's ability to distinguish between live and non-live objects; and the use of a one-dimensional convolutional network and a FIRE module can not only improve prediction accuracy, but also reduce computational complexity and improve model prediction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1This is a flowchart of the liveness detection model training method described in this application.

[0026] Figure 2 This is a flow chart of the liveness detection method described in this application.

[0027] Figure 3 This is a framework diagram of the liveness detection system described in this application. DETAILED DESCRIPTION

[0028] The following is a further detailed description of a liveness detection model training method, detection method, system, medium and terminal of the present application in conjunction with specific embodiments and drawings.

[0029] See Figure 1 , this application provides a liveness detection model training method, including the following steps S100-S300.

[0030] In a preferred embodiment, radar data samples are collected by two UWB detection modules in the vehicle, such as mainly using tap20-tap32 corresponding to distance data of about 1.05m-2.85m, and the radar data samples are downsampled. Time series data is added to the samples after the downsampling operation, and then the data of multiple timestamps are converted into multi-channel data through dimensional reconstruction as live input data; the one-dimensional convolutional network introduces the FIRE module to complete the construction of the initial live detection model, and the initial live detection model is further trained with the live input data to obtain the final target live detection model. Among them, the convolution layer of the one-dimensional convolutional network uses a convolution kernel of size 3*1 to perform a sliding convolution operation on the input multi-channel data to generate multiple feature maps; the FIRE module first performs a Squeeze operation on the feature map output by the convolution layer, and then performs an Expand operation after the Squeeze operation to expand the global feature descriptor into two sets of feature maps. Finally, the generated multiple sets of feature maps are spliced ​​to obtain the feature representation optimized by the FIRE module, and the output feature representation is compared with the true label to update the model parameters and optimize the model.

[0031] It should be noted that the data selection range of the radar data samples can be set by those skilled in the art according to actual application requirements, and is not necessarily limited to the range described above.

[0032] Next, steps S100 to S300 are described in detail.

[0033] Step S100: Acquire initial radar data within a preset area, and obtain living body input data based on the initial radar data and corresponding time series data.

[0034] In some embodiments, close-range TAPs are affected by antenna characteristics and have large noise, and there is a certain correlation between adjacent TAPs. Radar data such as tap20-tap32 corresponding to a distance of about 1.05m-2.85m is obtained through two UWB detection modules in the vehicle as the initial radar data; wherein tap20 and tap32 represent time identifiers.

[0035] Among them, the UWB detection module can be installed in the center of the roof, on both sides of the roof, on the dashboard or seat backs, etc.; the two modules need to be at a certain distance from each other to ensure that the detection range can cover the interior space of the cabin and ensure the comprehensiveness of the detection data.

[0036] Furthermore, obtaining the living input data includes: performing a downsampling operation on the initial radar data to obtain a first radar data sample; and obtaining time series data corresponding to the first radar data sample, and embedding the time series data into the first radar data sample to obtain a second radar data sample.

[0037] In some embodiments, based on actual needs and data characteristics, downsampling is performed by selecting one data point every 10 data points, that is, reducing the original sampling frequency of 100 times per second to 10 times per second; the initial radar data is filtered according to the above rules; for example, the initial radar data is a sequence containing 60,000 data points (10 minutes × 60 seconds × 100 times / second), and after downsampling, a first radar data sample containing 6,000 data points (10 minutes × 60 seconds × 10 times / second) is obtained.

[0038] Furthermore, the acquisition time of each data point in the first radar data sample is recorded; for example, the acquisition time of each data point is represented by a timestamp (e.g., in seconds from the start of acquisition). For the first radar data sample containing 6,000 data points, we obtain a corresponding time series data sequence containing 6,000 timestamps.

[0039] Furthermore, the time series data is fused with the first radar data sample. For example, each data point of the original first radar data sample is a one-dimensional vector. Now the corresponding timestamp is used as the second dimension, and each data point is expanded into a two-dimensional vector, thereby obtaining the second radar data sample containing time series information.

[0040] In the above technical solution, the amount of data is reduced through downsampling operations, which greatly reduces the computing resources required for subsequent model training and data processing, and improves overall computing efficiency. This is especially true when processing large-scale radar data. Time series data reflects the changes in radar signals over time. Physiological activities such as movement and breathing of living organisms have unique patterns and regularities in the time dimension. Integrating time series information into radar data provides the model with richer feature information, enabling the model to better capture the dynamic characteristics of living organisms, enhance the feature expression ability of the data, and help improve the accuracy and robustness of the liveness detection model.

[0041] Furthermore, the obtaining of live input data also includes: converting the second radar data sample into a multi-channel radar data sample through dimensional reconstruction to serve as the live input data; wherein, it also includes: dividing the live input data into a training set, a validation set, and a test set.

[0042] In some embodiments, the number of channels is first confirmed, such as converting the second radar data sample into a 3-channel radar data sample, where the first channel represents the amplitude information of the radar signal, the second channel represents the phase information of the radar signal, and the third channel represents the previously embedded timing information; then, the data features in the second radar data sample are analyzed, and the data related to the amplitude, phase, and timing are extracted respectively.

[0043] For example, for each data point in the second radar data sample, it is decomposed into an amplitude value, a phase value, and a timestamp according to its physical meaning and mathematical characteristics; then all the amplitude values ​​are combined into a data sequence for the first channel, the phase values ​​are combined into a data sequence for the second channel, and the timestamps are combined into a data sequence for the third channel.

[0044] The data sequences of these three channels are further arranged in order to form a three-dimensional data structure, namely the multi-channel radar data sample. Each data point in this three-dimensional structure has corresponding amplitude, phase and timing information, thus completing the dimensional reconstruction.

[0045] Furthermore, according to the set division rules, such as 70%, 15%, and 15%, the multi-channel radar data samples are divided into training set, validation set, and test set.

[0046] For example, multi-channel radar data samples are divided using random sampling. After the division is completed, the data distribution of the training set, validation set, and test set is checked to ensure that each data set contains live data in various situations, such as passenger data of different ages and postures, and pet data in different states, etc., to ensure that the model can perform well in various situations.

[0047] In the above technical solution, dimensional reconstruction converts the second radar data sample into a multi-channel radar data sample. Each channel can represent different types of feature information. The multi-channel setting enables the model to analyze and learn the data from multiple angles, greatly enriching the feature expression of the data, allowing the model to more comprehensively and accurately capture the characteristic patterns of living organisms, thereby improving the accuracy of living organism detection. The reasonable division of the training set, validation set, and test set helps the model learn the inherent laws of the data, rather than just memorizing specific samples in the training data.

[0048] Step S200: constructing an initial living body detection model based on a one-dimensional convolutional network and a FIRE module.

[0049] In some embodiments, the initial living body detection model includes at least a convolution layer, a squeeze layer, an expand layer, and a fully connected layer; and also includes an input layer, a pooling layer, and an output layer.

[0050] Among them, input layer: According to the dimension of the multi-channel radar data sample obtained by previous processing, the dimension of the input layer is determined, such as the number of channels is 3 and the time step is 100.

[0051] Convolutional Layer: Design multiple one-dimensional convolutional layers, each using kernels of different sizes to capture features at different scales. For example, the first convolutional layer uses a kernel of size 3 with a stride of 1. The number of kernels can be adjusted as needed; let's assume it's set to 16. This way, after the first convolutional layer, the data dimension becomes (16, 100). Add an activation function after each convolutional layer, and those skilled in the art can choose their own.

[0052] The FIRE module consists of a Squeeze layer and an Expand layer. The FIRE module can be reused as needed to further extract and enhance features. The Squeeze layer convolves the input feature map with a 1x1 convolution kernel, compressing the number of channels; for example, it compresses the input's 16 channels to 8. The Expand layer calculates the weight of each channel through global average pooling, a fully connected layer, and a sigmoid activation function. These weights are then multiplied by the compressed feature map to highlight important feature information.

[0053] A pooling layer is added after the FIRE module to reduce the dimension of the data and the amount of computation while retaining important feature information.

[0054] Fully connected layer: combines and classifies the extracted features; output layer: outputs the corresponding results according to the task requirements of liveness detection.

[0055] Step S300: training the initial liveness detection model using the liveness input data to obtain a target liveness detection model.

[0056] Furthermore, the training of the initial liveness detection model includes: initializing the model parameters of the initial liveness detection model; inputting the training set into the initial liveness detection model to perform feature extraction through the convolution layer, the squeeze layer and the expand layer in sequence, and performing classification operations through the fully connected layer to obtain a liveness prediction output result; and updating the model parameters based on the liveness prediction output result and a preset true label to iteratively train the initial liveness detection model until the model converges.

[0057] In some embodiments, to detect whether a living organism exists, the output of the fully connected layer is passed through a Sigmoid activation function to obtain a probability value between 0 and 1, indicating the probability of the existence of a living organism. This probability value is the liveness prediction output result.

[0058] Furthermore, the loss value is calculated based on the liveness prediction output and the preset true label. For binary classification problems, the binary cross-entropy loss function can be used. If the true label is 1 (indicating the presence of a live body) and the model predicts a probability of 0.3, then the corresponding loss value can be calculated based on the binary cross-entropy loss function. The backpropagation algorithm is used to calculate the gradient of the loss function for each parameter in the model. Based on the calculated gradient, the optimizer (such as the Adam optimizer) is used to update the model parameters.

[0059] In the above technical solution, the convolution layer can automatically extract local features from the in vivo input data, and perform convolution operations by sliding the convolution kernel on the data, which can capture feature patterns at different scales; the Squeeze layer performs global average pooling on the feature map output by the convolution layer, compressing the features of each channel into a single value, thereby obtaining the global feature information of the channel, reducing the dimension of the data and the amount of computation; the Expand layer expands the features at different scales; this structural design enables the model to process features at different levels and scales, enhancing the feature expression capability, while reducing the complexity of the model and improving computational efficiency.

[0060] In addition, the fully connected layer integrates the features extracted by the previous layers and maps the high-dimensional feature vector to a low-dimensional output space to achieve the classification of living and non-living objects.

[0061] Furthermore, in the process of training the initial liveness detection model, it also includes: preliminarily evaluating the model performance of the initial liveness detection model based on the validation set, so as to adjust the model parameters according to the preliminary evaluation results.

[0062] In some embodiments, a maximum number of iterations is set (e.g., 100 epochs), and each epoch represents a complete traversal of the entire training set. After each epoch, the model is evaluated using the validation set; the accuracy, loss value, and other indicators of the model on the validation set are calculated; if the loss value on the validation set no longer decreases, or the accuracy no longer increases, it means that the model may have converged.

[0063] In other embodiments, to prevent overfitting, an early stopping strategy may be employed. For example, if the loss value on the validation set does not decrease within 10 consecutive epochs, training is stopped and the model with the best performance on the validation set is selected as the preliminary target liveness detection model.

[0064] In the above technical solution, preliminary evaluation through the validation set can promptly detect whether the model has a tendency to overfit. It can also try different hyperparameter combinations and observe the performance changes of the model on the validation set to find the optimal hyperparameter settings.

[0065] Furthermore, after the initial liveness detection model training is completed, it also includes: performing a final evaluation on the model performance based on the test set to obtain a target liveness detection model according to the final evaluation result.

[0066] In some embodiments, various performance indicators of the model are verified based on the test set, such as accuracy, recall rate, F1 value, specificity, etc.; if the evaluation result of the model on the test set meets the preset performance requirements, then this preliminary target liveness detection model is determined as the final target liveness detection model; if the evaluation result does not meet the expected standard, the model needs to be optimized.

[0067] In the above technical solution, by comprehensively evaluating these indicators on the test set, we can have a more comprehensive understanding of the advantages and disadvantages of the model, so as to improve and optimize the model in a targeted manner and enhance the reliability and accuracy of the model.

[0068] Based on the same concept, see Figure 2 The present application also provides a liveness detection method, comprising: obtaining a multi-frame radar signal within a set time length, and inputting the radar signal into a target liveness detection model to output a liveness detection result in a target vehicle; wherein the target liveness detection model is obtained by the liveness detection model training method described above.

[0069] In some embodiments, at least two UWB detection modules are installed at appropriate locations inside the vehicle, such as on the roof, under the seats, etc., to ensure full coverage of the vehicle interior.

[0070] Furthermore, according to actual needs, an appropriate duration is set, such as 10 seconds. This duration should be long enough to ensure that enough radar signal frames can be captured to accurately detect the presence of living things; at the same time, it should not be too long so as not to affect the real-time performance of the detection.

[0071] Furthermore, after the vehicle stops being used and the doors are closed, the UWB detection module begins to continuously collect radar signals according to the set duration, pre-processes the radar signals, and inputs them into the target liveness detection model to infer the input radar signals through the target liveness detection model and output the liveness detection results; a judgment is made based on the output results of the model. If the result output by the model indicates the presence of a living person (for example, the probability value is greater than 0.5), it is determined that there is a living person in the vehicle; if the result output by the model indicates the absence of a living person (for example, the probability value is less than or equal to 0.5), it is determined that there is no living person in the vehicle.

[0072] Furthermore, if a living being is detected in the vehicle, the system can immediately issue an alarm and notify the staff to conduct further inspection; if no living being is detected in the vehicle, the system can record the information and allow the vehicle to be dispatched and used next time.

[0073] In the above technical solution, obtaining multiple frames of radar signals within a set time period can provide richer information than single-frame signals, making the detection results more stable and reliable, and improving the accuracy of liveness detection; the target liveness detection model has undergone a series of scientific training processes and can accurately distinguish between living and non-living targets in the vehicle, maintaining high detection accuracy even in complex environments.

[0074] Based on the same concept, see Figure 3 , the present application also provides a liveness detection system, comprising: An acquisition module is used to acquire radar signals within a set time period and input the radar signals into a target liveness detection model; and a detection module is used to perform liveness detection in the target vehicle using the target liveness detection model and output the liveness detection result.

[0075] It should be noted that the liveness detection system is implemented using the liveness detection method described above, and its specific implementation is the same as that of the liveness detection method, which will not be described in detail here.

[0076] In the above technical solution, the intelligent liveness detection function can enhance the safety and technological sense of the vehicle and increase the competitiveness of the product; at the same time, users can also use the vehicle more confidently without worrying about potential dangers caused by negligence, thereby improving the overall user experience.

[0077] Based on the same concept, the present application also provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned liveness detection method when running.

[0078] In some embodiments, the storage medium stores several computer programs for causing an in-vehicle terminal to execute all or part of the steps of the methods described in various embodiments of the present application. The medium may include a USB flash drive, a mobile hard drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, among other media capable of storing program code.

[0079] Based on the same concept, the present application also provides a vehicle-mounted terminal, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the above-mentioned liveness detection method.

[0080] In some embodiments, the memory and processor are interconnected via a bus; the processor may be one or more CPUs. When the processor is a CPU, the CPU may be a single-core CPU or a multi-core CPU. The processor is used to control various functional modules of the vehicle terminal and process signals.

[0081] The memory includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), and CD-ROM (Compact Disc Read-Only Memory). The memory is used to store computer programs, operating systems, various applications, and data, such as a computer program for implementing the liveness detection method.

[0082] In summary, the present application provides a liveness detection model training method, detection method, system, medium and terminal; first, initial radar data within a preset area is obtained, and liveness input data is obtained based on the initial radar data and the corresponding time series data; an initial liveness detection model is constructed based on a one-dimensional convolutional network and a FIRE module; and the initial liveness detection model is trained with the liveness input data to obtain a target liveness detection model. The present application can provide richer and more characteristic data by incorporating time series data, reduce misjudgments caused by environmental interference, and thus improve the model's ability to distinguish between live and non-live objects; and the use of a one-dimensional convolutional network and a FIRE module can not only improve the prediction accuracy, but also reduce computational complexity and improve the model prediction efficiency.

[0083] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely illustrative and are not intended to limit the scope of the present application. Various changes and modifications may be made therein by those skilled in the art without departing from the scope and spirit of the present application. All such changes and modifications are intended to be included within the scope of the present application as required by the appended claims.

[0084] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units described is merely a logical function division. In actual implementation, other division methods may be used, such as combining or integrating multiple units or components into another device, or ignoring or not performing some features.

[0086] The various component embodiments of the present application can be implemented in hardware, or in a software module running on one or more processors, or in a combination thereof. Those skilled in the art will appreciate that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some modules according to the embodiments of the present application. The application can also be implemented as a part or all of a device program (e.g., a computer program and a computer program product) for performing the method described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0087] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0088] Although the present application is described in conjunction with the above specific embodiments, it is obvious that those skilled in the art can make many substitutions, modifications and variations based on the above content. Therefore, all such substitutions, improvements and variations are included in the spirit and scope of the appended claims.

Claims

1. A method for training a liveness detection model, characterized in that: The following steps are involved: Acquire initial radar data within a preset area, and obtain living body input data based on the initial radar data and corresponding time series data; Build an initial liveness detection model based on a one-dimensional convolutional network and the FIRE module; Furthermore, the initial living body detection model is trained using the living body input data to obtain a target living body detection model.

2. The method for training a liveness detection model according to claim 1, wherein: The obtaining of living body input data comprises: performing a downsampling operation on the initial radar data to obtain a first radar data sample; Acquire time series data corresponding to the first radar data sample, and embed the time series data into the first radar data sample to obtain a second radar data sample.

3. The method for training a liveness detection model according to claim 2, wherein: The obtaining of living body input data further includes: Converting the second radar data samples into multi-channel radar data samples by dimensional reconstruction to serve as living body input data; The method further includes dividing the living body input data into a training set, a verification set and a test set.

4. The method for training a liveness detection model according to claim 3, wherein: The initial liveness detection model includes at least a convolutional layer, a squeeze layer, an expand layer, and a fully connected layer; and the training of the initial liveness detection model includes: Initializing model parameters of the initial living body detection model; Inputting the training set into the initial liveness detection model to perform feature extraction through the convolution layer, squeeze layer, and expand layer in sequence, and performing classification operation through the fully connected layer to obtain a liveness prediction output result; The model parameters are updated based on the liveness prediction output result and the preset true label to iteratively train the initial liveness detection model until the model converges.

5. The method for training a liveness detection model according to claim 4, wherein: The process of training the initial living body detection model also includes: The model performance of the initial living body detection model is preliminarily evaluated based on the validation set, so as to adjust the model parameters according to the preliminary evaluation results.

6. The method for training a liveness detection model according to claim 5, wherein: After the initial liveness detection model is trained, the following steps are also included: A final evaluation is performed on the model performance based on the test set to obtain a target living body detection model according to the final evaluation result.

7. A method for detecting a living body, characterized in that: include: Acquire multiple frames of radar signals within a set time period and input the radar signals into a target liveness detection model to output a liveness detection result in the target vehicle; The target liveness detection model is obtained by the liveness detection model training method according to any one of claims 1 to 6.

8. A system using the liveness detection method according to claim 7, characterized in that: include: An acquisition module is used to acquire radar signals within a set time period and input the radar signals into a target liveness detection model; The detection module is used to perform liveness detection in the target vehicle through the target liveness detection model and output the liveness detection result.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the living body detection method according to claim 7 when running.

10. A vehicle-mounted terminal, characterized in that: The vehicle-mounted terminal includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the liveness detection method as described in claim 7.

Images