A pedestrian recognition and intrusion detection method and system based on millimeter wave radar
By using a neural network model based on millimeter-wave radar, the problem of poor pedestrian recognition performance of RGB cameras under large-area occlusion was solved, and effective identification of pedestrians behind occlusions and detection of intruders were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INSTITUTE OF TECHNOLOGY (SHENZHEN) (INSTITUTE OF SCIENCE AND TECHNOLOGY INNOVATION HARBIN INSTITUTE OF TECHNOLOGY SHENZHEN)
- Filing Date
- 2023-10-10
- Publication Date
- 2026-06-16
AI Technical Summary
Existing pedestrian recognition systems based on RGB cameras suffer from reduced image quality under large-area occlusion, resulting in poor pedestrian recognition performance and difficulty in accurately identifying pedestrians in complex scenes.
Millimeter-wave radar is used to acquire training signals and construct a neural network model. By acquiring a four-dimensional matrix, extracting point cloud information features, processing bidirectional LSTM, and supervising comparative learning, the identification of pedestrians and intruders is achieved.
It can effectively identify pedestrians even under large-area obstruction and detect intruders hiding behind obstructions, thus improving the accuracy and reliability of pedestrian identification.
Smart Images
Figure CN117310698B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pedestrian identification and intruder detection technology, and in particular to a pedestrian identification and intrusion detection method and system based on millimeter-wave radar. Background Technology
[0002] In the field of pedestrian recognition, traditional methods often use RGB cameras to perceive targets. This involves acquiring image or video data of pedestrians using an RGB camera, and then using feature detection algorithms to identify pedestrians based on their contours, faces, clothing, and other obvious features. However, RGB camera-based pedestrian recognition systems have high environmental requirements, such as needing sufficient lighting and avoiding rain, fog, or other adverse weather conditions. Furthermore, complex scenarios where objects unrelated to the target obstruct the camera lens are common in the real world. The presence of occlusions interferes with camera image quality, preventing subsequent detection algorithms from extracting effective features from the image or video, thus reducing the effectiveness of pedestrian recognition.
[0003] When there are large areas of obstruction, such as large potted plants or posters, the image quality of the camera will be severely degraded. Therefore, existing pedestrian recognition technology needs a system and method that can accurately identify pedestrians even under conditions of large-scale obstruction to improve the accuracy of pedestrian recognition. Summary of the Invention
[0004] This invention provides a pedestrian identification and intrusion detection method and system based on millimeter-wave radar, aiming to solve or partially solve the problems existing in the background art.
[0005] To solve the above-mentioned technical problems, the present invention is implemented as follows:
[0006] The first aspect of this invention provides a method for pedestrian identification and intrusion detection based on millimeter-wave radar, the method comprising:
[0007] Acquire training radar signals, wherein the training radar signals are millimeter-wave radar signals;
[0008] A neural network training model is constructed based on the trained radar signals;
[0009] Acquire test radar signals, import the test radar signals into the neural network training model, and perform iterative calculations;
[0010] Based on the calculation results, the pedestrian type and the intruder type were identified.
[0011] In conjunction with the first aspect, in some embodiments, constructing a neural network training model based on the training radar signals includes:
[0012] A four-dimensional matrix is obtained based on the training radar signal, wherein the four dimensions of the four-dimensional matrix are the number of transmission frames, the number of antennas, the number of chirps, and the number of sampling points per chirp, respectively.
[0013] The neural network model is constructed based on the four-dimensional matrix.
[0014] In conjunction with the first aspect, in some implementations, a four-dimensional matrix is obtained based on the training radar signal, wherein the four dimensions of the four-dimensional matrix are the number of transmission frames, the number of antennas, the number of chirps, and the number of sampling points per chirp, including:
[0015] Based on the radar signal, pedestrian morphology information and pedestrian motion information are acquired, wherein the pedestrian morphology information is point cloud information constituting the external outline of the pedestrian, and the pedestrian motion information is the motion information of multiple points corresponding to the point cloud information.
[0016] Feature extraction is performed on the pedestrian morphology information and the pedestrian motion information.
[0017] In conjunction with the first aspect, in some implementations, feature extraction is performed on the pedestrian morphology information and the pedestrian motion information to satisfy the following:
[0018] ；
[0019] in, Represents a shared linear network. Represents network parameters, This represents the high-dimensional vector feature obtained after the point cloud of the i-th frame passes through a shared linear layer. This indicates that the i-th frame of the sample contains n points describing the pedestrian's pose.
[0020] In conjunction with the first aspect, in some embodiments, feature extraction of the pedestrian morphology information and the pedestrian motion information further includes:
[0021] For each of the current plurality of points, a weight is assigned, satisfying the following:
[0022] ;
[0023] in, This represents the attention network used for weight allocation. The parameters represent the attention network. This indicates the current weight.
[0024] In conjunction with the first aspect, in some embodiments, feature extraction of the pedestrian morphology information and the pedestrian motion information further includes:
[0025] Obtain the global feature vector of the current point. ,satisfy:
[0026]
[0027] as well as
[0028] ;
[0029] in, This represents the point cloud feature vector after semantic augmentation. express Network parameters of the coding layer.
[0030] In conjunction with the first aspect, in some embodiments, constructing a neural network training model based on the training radar signals further includes:
[0031] The current sample is sequentially input into a bidirectional LSTM in chronological order, and the sum of the hidden units of the two LSTM directions at the last moment is taken as the global feature vector description of the current input sample, satisfying the following:
[0032] ;
[0033] in, express The network parameter H represents the summation result of the bidirectional LSTM hidden unit vectors at the last moment.
[0034] In conjunction with the first aspect, in some embodiments, constructing a neural network training model based on the training radar signals further includes:
[0035] Correct the situation where the prediction fails, and satisfy the following conditions:
[0036] ;
[0037] in, Represents the cross-entropy loss function. This represents the predicted class label of the sample. This represents the actual category label.
[0038] A second aspect of this invention provides a pedestrian identification and intrusion detection system based on millimeter-wave radar, the system comprising:
[0039] A first acquisition module is used to acquire training radar signals, wherein the training radar signals are millimeter-wave radar signals.
[0040] The construction module is used to construct a neural network training model based on the training radar signals;
[0041] The second acquisition module is used to acquire the test radar signal, import the test radar signal into the neural network training model and perform iterative calculations.
[0042] The confirmation module is used to confirm the pedestrian type and the intruder type based on the calculation results.
[0043] A third aspect of this invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.
[0044] Memory, used to store computer programs;
[0045] When a processor executes a program stored in memory, it implements the method steps proposed in the first aspect of the embodiments of the present invention.
[0046] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention.
[0047] The embodiments of the present invention have the following advantages:
[0048] This invention provides a pedestrian identification and intrusion detection method based on millimeter-wave radar. First, a training radar signal, wherein the training radar signal is a millimeter-wave radar signal, is acquired. Then, a neural network training model is constructed based on the training radar signal. Next, a test radar signal is acquired, imported into the neural network training model, and iterative calculations are performed. Finally, based on the calculation results, the pedestrian type and intruder type are determined. The method proposed in this invention, using millimeter-wave radar, can effectively identify pedestrians behind obstructions and simultaneously detect intruders hiding behind obstructions. Attached Figure Description
[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0050] Figure 1 This is a flowchart illustrating a pedestrian identification and intrusion detection method based on millimeter-wave radar in an embodiment of the present invention. Detailed Implementation
[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] Chinese patent application CN114663839A discloses a method and system for pedestrian re-identification under occlusion based on deep neural networks. This method divides pedestrian images in the training set into multiple sequence blocks according to occlusion conditions, then inputs these sequence blocks into a recognition model for training, obtaining global and local information about the pedestrian for subsequent identity identification. However, this method is mainly effective for cases with limited occlusion. When there is large-area occlusion, such as large potted plants or posters, the camera's image quality will be severely degraded, and the above method will struggle to extract effective features. Therefore, existing pedestrian identification technology needs a system and method capable of accurately identifying pedestrians under large-scale occlusion conditions to improve the accuracy of pedestrian identification.
[0053] With the development of radio frequency (RF) technology, an increasing number of RF devices are being used for target perception, such as gesture recognition and motion recognition. Compared with other RF devices, millimeter-wave radar provides a more granular characterization of targets by emitting signals at higher frequencies and shorter wavelengths. Furthermore, RF signals have good penetration through non-metallic obstructions. Therefore, millimeter-wave radar can effectively identify pedestrians behind obstructions and detect intruders hiding behind them.
[0054] This application proposes a pedestrian identification and intrusion detection method based on millimeter-wave radar to improve the above-mentioned problems.
[0055] This application proposes a pedestrian identification and intrusion detection method based on millimeter-wave radar. Please refer to [link to relevant documentation]. Figure 1 This includes the following steps:
[0056] A method for pedestrian identification and intrusion detection based on millimeter-wave radar, the method comprising:
[0057] S101: Acquire training radar signal, wherein the training radar signal is a millimeter-wave radar signal.
[0058] Understandably, since radio frequency signals have good penetration through non-metallic obstructions, millimeter-wave radar is typically deployed behind such obstructions in a typical home environment to ensure that the obstruction blocks the line of sight between the radar and the target pedestrian. The millimeter-wave radar is then used to collect radar echo signals from multiple target pedestrians under different obstruction conditions.
[0059] S102: Construct a neural network training model based on the training radar signal.
[0060] Specifically, in this embodiment, step S102 may include the following steps:
[0061] S102-1: Obtain a four-dimensional matrix based on the training radar signal, wherein the four dimensions of the four-dimensional matrix are the number of transmission frames, the number of antennas, the number of chirps, and the number of sampling points per chirp, respectively;
[0062] S102-2: Construct the neural network model based on the four-dimensional matrix.
[0063] Specifically, the process of building a neural network model includes the following steps:
[0064] S102-2-1: Based on the radar signal, obtain pedestrian shape information and pedestrian motion information, wherein the pedestrian shape information is point cloud information constituting the external outline of the pedestrian, and the pedestrian motion information is the motion information of multiple points corresponding to the point cloud information.
[0065] S102-2-2: Extract features from the pedestrian morphology information and the pedestrian motion information.
[0066] Understandably, after inputting samples into a predefined neural network, the network can not only extract pedestrian features from a single point cloud but also integrate motion information from the point cloud sequence, ultimately forming high-dimensional feature vectors for different pedestrians and outputting pedestrian identity information through a linear classification layer. Specifically, given a point cloud sample containing multiple frames, let be the i-th frame of the sample containing points describing pedestrian poses, where represents the i-th point in the i-th frame. For a single frame of point cloud, a shared linear layer is first used as the basic feature extraction module. The shared linear layer extracts features from each point in the point cloud to obtain high-dimensional features.
[0067] Specifically, the process satisfies ；
[0068] in, Represents a shared linear network. Represents network parameters, This represents the high-dimensional vector feature obtained after the point cloud of the i-th frame passes through a shared linear layer. This indicates that the i-th frame of the sample contains n points describing the pedestrian's pose.
[0069] S102-2-3: Assign weights to each of the current plurality of points, satisfying the following conditions:
[0070] ;
[0071] in, This represents the attention network used for weight allocation. The parameters represent the attention network. This indicates the current weight.
[0072] Considering the limited number of antennas in existing radar equipment, resulting in sparse radar point clouds in a single frame, a Transformer spatial coding layer is introduced in the algorithm to enrich the semantic information of the encoded point cloud feature vector. This layer uses a self-attention mechanism to calculate the feature similarity between the current point cloud vector and each point cloud vector in the sample sequence, and performs a weighted summation of features for other point clouds based on the similarity, thereby adding spatial semantic information to each frame of the current input sequence. Therefore, the algorithm also includes the following steps:
[0073] SS102-2-3: Obtain the global feature vector of the current point. ,satisfy:
[0074]
[0075] as well as
[0076] ;
[0077] in, This represents the point cloud feature vector after semantic augmentation. express Network parameters of the coding layer.
[0078] To fully extract motion information between point cloud frames, the algorithm employs a bidirectional LSTM method to further mine the feature vectors of each frame in the input point cloud sequence. In some implementations, the following steps may also be included:
[0079] The current sample is sequentially input into a bidirectional LSTM in chronological order, and the sum of the hidden units of the two LSTM directions at the last moment is taken as the global feature vector description of the current input sample, satisfying the following:
[0080] ;
[0081] in, express The network parameter H represents the summation result of the bidirectional LSTM hidden unit vectors at the last moment.
[0082] In some implementations, to fully extract motion information between point cloud frames, the algorithm employs a bidirectional LSTM method to further mine the feature vectors of each frame in the input point cloud sequence. The current sample is sequentially input into the bidirectional LSTM in chronological order, and the sum of the hidden units in the two directions at the last moment is taken as the global feature vector description of the current input sample.
[0083]
[0084] in express The network parameters, H, represent the summation of the bidirectional LSTM hidden unit vectors at the last time step. Finally, a linear layer is used to classify the current global vector, and the model's final output is the number of predefined pedestrian categories.
[0085] In some implementations, a cross-entropy function is used as a penalty function to correct for prediction failures.
[0086] in Represents the cross-entropy loss function. This represents the predicted class label of the sample. This represents the actual category label.
[0087] Understandably, to effectively achieve intruder detection, this application introduces two sub-modules—supervised contrastive learning and distance measurement—building upon the target pedestrian identification section. Supervised contrastive learning narrows the distance between samples with the same category label within the feature space, while widening the distance between samples with different category labels. This means samples of the same category have a certain similarity in the feature space, while samples of different categories exhibit certain differences. When an intruder sample of unknown category exists, the sample is mapped to a high-dimensional feature space through a feature extractor. Since the intruder's category is not present in the current feature space, the features of the current sample differ from the feature centers of all known categories in the feature space. Distance learning further highlights these differences, determining whether the current sample belongs to an intruder sample. Specifically, a supervised momentum contrastive learning module is added to the pedestrian identification learning process. This module constructs a large-scale feature library for the samples in the current training set. When training on a sample in the training set, the obtained high-dimensional features are further compared with the features in the feature library to calculate similarity. Samples of the same category will receive higher similarity results, while samples of different categories will receive the opposite. The calculation process uses encoded vectors as label vectors for the contrastive learning module. Samples in the feature library with the same category label as the current sample are set to 1, while samples from different categories are set to -1. Feature similarity is calculated using an inner product approach, and the loss function is the mean squared error (MSE).
[0088]
[0089] in This represents the mean square error function. This represents the similarity vector obtained after calculating the similarity with all features in the feature library. This represents the label vector defined according to the sample category. This process is trained together with the pedestrian identification process; that is, the overall loss function is defined as:
[0090]
[0091] in This represents a predefined hyperparameter.
[0092] S103: Acquire the test radar signal, import the test radar signal into the neural network training model and perform iterative calculations.
[0093] S104: Based on the calculation results, confirm the pedestrian type and the intruder type.
[0094] Understandably, the trained recognition network will be used for intruder detection during testing. First, all samples from each category in the training set are input into the trained recognition network, retaining the activation vectors of correctly classified samples. Then, the mean of the activation vectors for each category is calculated according to the category label, and the result is used as the centroid vector (MAV) for each category. For the k-th category (… Using the activation vectors of samples belonging to this class, calculate the distance between the vector and the centroid vector, denoted as . , where M represents the total number of samples in the current category k. This represents the distance between the activation vector and the centroid vector of the m-th sample in the k-th class of the current training set. A Weibull distribution is used to fit the maxima of the distances for each class. The fitted result is the cumulative distribution function of the Weibull distribution for each class. After feature encoding, the activation vector of the current input sample P is defined as:
[0095]
[0096] Calculate the centroid of each category based on the activation vector. Distance between The above distance values are input into the cumulative function of the Weibull distribution for each category, which returns the probability that the predicted sample does not belong to any category. ,but 1- This represents the probability that the current sample belongs to the k-th class. This value is used as the weight adjustment for the corresponding dimension of the activation vector, thus the final output indicating the probability that the sample belongs to the k-th class is... The probability that the current sample belongs to the intruder, i.e., the unknown class sample, is:
[0097]
[0098] Finally, we obtain the output containing all class probabilities, including the unknown class, and then transform this output into the final probability distribution using the Softmax function. .
[0099]
[0100] When the maximum classification probability of the above results is the largest in the unknown class or the maximum classification probability is less than a certain threshold, it is identified as the unknown category, i.e., the intruder category.
[0101] This invention provides a pedestrian identification and intrusion detection method based on millimeter-wave radar. First, a training radar signal, wherein the training radar signal is a millimeter-wave radar signal, is acquired. Then, a neural network training model is constructed based on the training radar signal. Next, a test radar signal is acquired, imported into the neural network training model, and iterative calculations are performed. Finally, based on the calculation results, the pedestrian type and intruder type are determined. The method proposed in this invention, using millimeter-wave radar, can effectively identify pedestrians behind obstructions and simultaneously detect intruders hiding behind obstructions.
[0102] Based on the same inventive concept, a second aspect of the present invention proposes a pedestrian identification and intrusion detection system based on millimeter-wave radar, the system comprising:
[0103] A first acquisition module is used to acquire training radar signals, wherein the training radar signals are millimeter-wave radar signals.
[0104] The construction module is used to construct a neural network training model based on the training radar signals;
[0105] The second acquisition module is used to acquire the test radar signal, import the test radar signal into the neural network training model and perform iterative calculations.
[0106] The confirmation module is used to confirm the pedestrian type and the intruder type based on the calculation results.
[0107] This invention provides a pedestrian identification and intrusion detection system based on millimeter-wave radar. First, a training radar signal, wherein the training radar signal is a millimeter-wave radar signal, is acquired. Then, a neural network training model is constructed based on the training radar signal. Next, a test radar signal is acquired, imported into the neural network training model, and iterative calculations are performed. Finally, based on the calculation results, the pedestrian type and intruder type are determined. The method proposed in this invention, using millimeter-wave radar, can effectively identify pedestrians behind obstructions and simultaneously detect intruders hiding behind obstructions.
[0108] Based on the same inventive concept, embodiments of this application also propose an electronic device, which includes:
[0109] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the pedestrian identification and intrusion detection method based on millimeter-wave radar according to the embodiments of this application.
[0110] Furthermore, to achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the pedestrian identification and intrusion detection method based on millimeter-wave radar according to embodiments of this application.
[0111] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0112] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0113] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0115] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0116] The present invention provides a detailed description of a pedestrian identification and intrusion detection method and system based on millimeter-wave radar. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for pedestrian identification and intrusion detection based on millimeter-wave radar, characterized in that, The method includes: Acquire training radar signals, wherein the training radar signals are millimeter-wave radar signals; construct a neural network training model based on the training radar signals; acquire test radar signals; import the test radar signals into the neural network training model and perform iterative calculations; and determine the pedestrian type and intruder type based on the calculation results. The neural network training model constructed based on the training radar signal further includes: acquiring pedestrian morphology information and pedestrian motion information based on the training radar signal, wherein the pedestrian morphology information is point cloud information constituting the external contour of the pedestrian, and the pedestrian motion information is motion information of multiple points corresponding to the point cloud information, and performing feature extraction on the pedestrian morphology information and the pedestrian motion information. Feature extraction is performed on the pedestrian morphology information and the pedestrian motion information to satisfy: ； in, Represents a shared linear network. Represents network parameters, This represents the high-dimensional vector feature obtained after the point cloud of the i-th frame passes through a shared linear layer. This represents the point cloud of the i-th frame in the sample, which contains N points describing the pedestrian's pose; Feature extraction of the pedestrian morphology information and the pedestrian motion information further includes: assigning weights to each of the current plurality of points, and satisfying the following: ； in, This represents the attention network used for weight allocation. The parameters represent the attention network. Indicates the current weight; The feature extraction process for the pedestrian morphology information and the pedestrian motion information further includes: obtaining the global feature vector of the current point. ,satisfy: as well as ； in, This represents the point cloud feature vector after semantic augmentation. express The network parameters of the encoding layer, based on the training radar signals to construct a neural network training model, also include: The current sample is sequentially input into a bidirectional LSTM in chronological order, and the sum of the hidden units of the two LSTM directions at the last moment is taken as the global feature vector description of the current input sample, satisfying the following: ； in, express The network parameters, H, represent the summation result of the bidirectional LSTM hidden unit vectors at the last moment; During testing, a trained recognition network will be used to detect intruders. First, all samples from each category in the training set will be input into the trained recognition network, retaining the activation vectors of correctly classified samples. Then, the mean of the activation vectors for each category will be calculated, and the result will be used as the centroid vector (MAV) for each category. For the k-th category, the distance between the activation vector of each sample belonging to that category and the centroid vector will be calculated, denoted as [equation missing]. ,in M represents the total number of samples in the current category k. This represents the distance between the activation vector and the centroid vector of the m-th sample in the k-th class of the current training set. A Weibull distribution is used to fit the maximum value among the distance results for each class. The fitted result is the cumulative distribution function of the Weibull distribution for each class. After feature encoding, the activation vector of the current input sample P is defined as: Calculate the centroid of each category based on the activation vector. Distance between The above distance values are input into the cumulative function of the Weibull distribution for each category, which returns the probability that the predicted sample does not belong to any category. ,but 1- This represents the probability that the current sample belongs to the k-th class. This value is used as the weight adjustment for the corresponding dimension of the activation vector, thus determining the final probability that the sample belongs to the k-th class. The probability that the current sample belongs to the intruder, i.e., the unknown class sample, is: Finally, we obtain the output containing all class probabilities, including the unknown class, and then transform this output into the final probability distribution using the Softmax function. , When the maximum classification probability of the above results is the largest in the unknown class or the maximum classification probability is less than a certain threshold, it is identified as the unknown category, i.e., the intruder category.
2. The pedestrian identification and intrusion detection method based on millimeter-wave radar according to claim 1, characterized in that, Based on the training radar signals, a neural network training model is constructed, including: A four-dimensional matrix is obtained based on the training radar signal, wherein the four dimensions of the four-dimensional matrix are the number of transmission frames, the number of antennas, the number of chirps, and the number of sampling points per chirp, respectively. The neural network training model is constructed based on the four-dimensional matrix.
3. The pedestrian identification and intrusion detection method based on millimeter-wave radar according to claim 2, characterized in that, The neural network training model constructed based on the training radar signals also includes: Correct the situation where the prediction fails, and satisfy the following conditions: ； in, Represents the cross-entropy loss function. This represents the predicted class label of the sample. This represents the actual category label.
4. A detection system applied to the pedestrian identification and intrusion detection method as described in any one of claims 1-2, characterized in that, The system includes: A first acquisition module is used to acquire training radar signals, wherein the training radar signals are millimeter-wave radar signals. The construction module is used to construct a neural network training model based on the training radar signals; The second acquisition module is used to acquire the test radar signal, import the test radar signal into the neural network training model and perform iterative calculations. The confirmation module is used to confirm the pedestrian type and the intruder type based on the calculation results.
5. An electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein, The processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the method as described in any one of claims 1-2.