Similar moving target identification method based on millimeter wave radar

Abstract

The invention discloses a similar moving target identification method and device based on a millimeter wave radar, and a computer readable storage medium. The method comprises the steps of obtaining millimeter wave radar original sensing data in a radar detection range; processing the original sensing data of the millimeter wave radar to obtain a target sensing data set for representing an independent moving target; extracting morphological characteristics reflecting the physical form of the independent moving target and motion characteristics reflecting the motion state of the independent moving target from the target sensing data set; and inputting the morphological features and the motion features into a pre-trained classification model to obtain a recognition result of the independent moving target. The method has the advantages of user privacy protection, high recognition precision, all-weather recognition and the like.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave radar technology, and in particular to a method, device, and computer-readable storage medium for identifying similar moving targets based on millimeter-wave radar. Background Technology

[0002] Currently, in scenarios requiring the differentiation of children from pets, the primary approach relies on camera-based or image recognition technologies. These solutions typically attempt to distinguish between children and pets by analyzing visual features in images or video streams, such as the target's shape, size, color, and behavioral patterns. Such methods can achieve a certain degree of differentiation in specific environments, such as well-lit public areas where privacy is not a primary concern.

[0003] However, existing camera- or image recognition-based solutions have significant drawbacks and limitations. First, in specific locations with high privacy requirements, such as bathrooms, bedrooms, or hotel rooms, deploying cameras for continuous monitoring is often unacceptable to the public and may even raise concerns about privacy violations. Second, the accuracy of cameras drops significantly at night or in low-light conditions due to decreased image quality, affecting their ability to operate around the clock. Furthermore, distinguishing between individuals of similar size, such as young children and large dogs, based solely on visual characteristics remains challenging.

[0004] Therefore, the industry urgently needs a technical solution that can achieve high-precision, all-weather differentiation between children and pets while protecting user privacy. Summary of the Invention

[0005] This application provides a method for identifying similar moving targets based on millimeter-wave radar, aiming to achieve high-precision, all-weather differentiation between children and pets while protecting user privacy.

[0006] To achieve the above objectives, embodiments of this application provide a method for identifying similar moving targets based on millimeter-wave radar, including:

[0007] Acquire raw sensing data from millimeter-wave radar within its detection range;

[0008] The raw sensing data from the millimeter-wave radar is processed to obtain a target sensing dataset for characterizing independent moving targets;

[0009] Extract morphological features reflecting the physical form of independent moving targets and motion features reflecting the motion state of independent moving targets from the target perception dataset;

[0010] The morphological features and motion features are input into a pre-trained classification model to obtain the recognition result of the independent moving target.

[0011] In one embodiment, the raw sensing data from the millimeter-wave radar is processed to obtain a target sensing dataset for characterizing independent moving targets, including:

[0012] The initial point cloud data of all objects within the radar's detection range are determined from the raw sensing data of the millimeter-wave radar.

[0013] Remove the point clouds with zero velocity information from the initial point cloud data to obtain the target point cloud data of all moving objects within the detection range;

[0014] Cluster analysis is performed on the target point cloud data to generate the target perception dataset.

[0015] In one embodiment, determining initial point cloud data for all objects within the radar detection range from the raw sensing data of the millimeter-wave radar includes:

[0016] The range-Doppler maps of all objects within the radar's detection range are obtained by performing range-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform on the raw sensing data of the millimeter-wave radar.

[0017] Two-dimensional constant false alarm rate (CFAR) detection is performed on the range-Doppler spectrum to identify all target units within the radar detection range and acquire the range data of the target units;

[0018] The azimuth and elevation angle data of the target unit are extracted from the raw sensing data of the millimeter-wave radar;

[0019] The initial point cloud data is generated based on the distance data, the azimuth data, and the pitch data.

[0020] In one embodiment, extracting morphological features reflecting the physical form of an independently moving target from the target perception dataset includes:

[0021] Based on the target perception dataset, extract the surface contour of the independent moving target;

[0022] Calculate the minimum bounding cube parameters of the surface profile;

[0023] The height and aspect ratio parameters of the minimum bounding cube are extracted as the morphological features.

[0024] In one embodiment, extracting motion features reflecting the motion state of an independently moving target from the target perception dataset includes:

[0025] Obtain the Doppler spectrum information of the independent moving target in the specified frequency band from the target perception dataset;

[0026] Based on the Doppler spectrum information, calculate the energy integral on the negative frequency side and the energy integral on the positive frequency side of the Doppler spectrum;

[0027] The harmonic asymmetry of the independently moving target is calculated based on the energy integrals on the negative frequency side and the positive frequency side, and is used as the motion characteristic.

[0028] In one embodiment, the pre-trained classification model is a support vector machine model;

[0029] Inputting the morphological features and the motion features into a pre-trained classification model includes: combining the morphological features and the motion features into a three-dimensional feature vector and inputting it into a pre-trained support vector machine model.

[0030] In one embodiment, the support vector machine model processes the input three-dimensional feature vector to obtain the recognition result by including:

[0031] The three-dimensional feature vectors are standardized to obtain standardized feature data;

[0032] The standardized feature data is mapped to a high-dimensional space using a radial basis function (RBF) kernel to obtain a similarity metric. The formula for the RBF kernel is as follows:

[0033] K(x,x i )=exp(-γ||xx i || 2 ),

[0034] Where γ is the kernel parameter, which controls the width of the kernel function;

[0035] Based on the similarity metric, the data in the high-dimensional space is classified using an optimal hyperplane to obtain the recognition result. The optimal hyperplane follows the formula below:

[0036] w T x+b=0,

[0037] Where w represents the weight vector of the hyperplane, which determines the direction of the hyperplane, and b is the bias term, which determines the distance between the hyperplane and the origin.

[0038] Furthermore, the hyperplane satisfies the following constraints:

[0039] y i (w T x i +b)≥1,

[0040] Among them, y i ∈{1,-1} represents the corresponding category label.

[0041] In one embodiment, after determining the identification result of the independent moving target, the method further includes:

[0042] The recognition results are transmitted to an external device for display.

[0043] To achieve the above objectives, this application also proposes a similar moving target identification device based on millimeter-wave radar, including a memory, a processor, and a similar moving target identification program based on millimeter-wave radar stored in the memory and executable on the processor. When the processor executes the similar moving target identification program based on millimeter-wave radar, it implements the similar moving target identification method based on millimeter-wave radar as described in any of the above claims.

[0044] To achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a similar moving target recognition program based on millimeter-wave radar. When the similar moving target recognition program based on millimeter-wave radar is executed by a processor, it implements the similar moving target recognition method based on millimeter-wave radar as described in any of the preceding claims.

[0045] The method for identifying similar moving targets in this application has the following advantages:

[0046] 1. It combines strong privacy protection with high environmental adaptability, breaking through the limitations of application scenarios:

[0047] Unlike camera-based solutions, the millimeter-wave radar in this application perceives targets by analyzing electromagnetic wave echoes, without involving image acquisition at all, generating abstract three-dimensional point cloud data. This fundamentally solves user concerns and regulatory risks associated with deploying surveillance equipment in privacy-sensitive areas such as bedrooms, bathrooms, and hotel rooms. Furthermore, the physical properties of millimeter waves allow them to penetrate smoke and dust, and their performance is unaffected by environmental factors such as light intensity, darkness, and inclement weather, ensuring stable operation in all weather conditions and scenarios. Compared to visual solutions whose performance deteriorates drastically at night or in low light, this offers an overwhelming advantage in environmental adaptability.

[0048] 2. High discrimination accuracy, capable of precisely identifying targets of similar size:

[0049] Traditional identification methods rely solely on a single dimension such as height, making them prone to confusion when faced with targets of similar size (e.g., crawling infants versus medium-sized pets). This solution extracts the height and aspect ratio of the smallest bounding cube to construct morphological features that reflect the overall contour of the target's "height, size, and build." More importantly, this application introduces a motion feature dimension that is completely orthogonal to the physical morphology. By analyzing the Doppler spectrum and calculating harmonic asymmetry, it can capture the subtle differences in intrinsic movement patterns produced by different biological gaits (e.g., a child's bipedal walking versus a pet's quadrupedal movement), differences that are difficult to discern with the naked eye. Finally, by inputting these two dimensions (morphology + motion) into a pre-trained classification model such as a Support Vector Machine (SVM), even if one dimension of the target's features (e.g., morphology) is ambiguous or similar, the significant difference in the other dimension (motion) can provide the model with a decisive basis for accurate judgment. This multi-feature fusion design greatly enhances the discriminative power of the feature set, thereby fundamentally improving the accuracy and robustness of identification.

[0050] 3. Full-process automation and high practicality enable a closed loop from intelligent sensing to effective alarm:

[0051] On the one hand, this application achieves automatic and accurate locking of independent moving targets by filtering, denoising, and clustering the raw data, without any manual intervention. On the other hand, this application automatically transforms the internal, machine-readable classification results into timely alerts that are perceptible to users on external devices (such as mobile app push notifications, smart speaker broadcasts, etc.) by using standardized communication protocols such as Wi-Fi and Bluetooth for the transmission and display of recognition results.

[0052] This end-to-end automated design makes this application not just an algorithm model, but a practical product that can be seamlessly integrated into smart home and security systems. It effectively transforms abstract analysis and judgment into actionable information that can remind guardians and ensure children's safety, completing the key closed loop from "perception" to "action," and has significant social value and promising industrial application prospects. Attached Figure Description

[0053] 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 the structures shown in these drawings without creative effort.

[0054] Figure 1 This is a block diagram of a similar moving target identification device based on millimeter-wave radar according to an embodiment of the present invention;

[0055] Figure 2 This is a flowchart illustrating an embodiment of the similar moving target recognition method based on millimeter-wave radar according to the present invention.

[0056] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0057] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0058] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0059] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The quantifier "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of "first," "second," and "third," etc., does not indicate any order and can be interpreted as names.

[0060] like Figure 1 As shown, Figure 1 This is a schematic diagram of the hardware operating environment server 1 (also called a similar moving target identification device based on millimeter-wave radar) involved in the embodiments of the present invention.

[0061] The server in this embodiment of the invention includes devices with display functions such as "Internet of Things devices", smart air conditioners, smart lights, smart power supplies with network connectivity, AR / VR devices with network connectivity, smart speakers, autonomous vehicles, PCs, smartphones, tablets, e-book readers, and portable computers.

[0062] like Figure 1 As shown, the server 1 includes: a memory 11, a processor 12, and a network interface 13.

[0063] The memory 11 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the server 1, such as the hard disk of the server 1. In other embodiments, the memory 11 can also be an external storage device of the server 1, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the server 1.

[0064] Furthermore, the memory 11 may include both internal storage units of the server 1 and external storage devices. The memory 11 can be used not only to store application software and various types of data installed on the server 1, such as the code of the similar moving target recognition program 10 based on millimeter-wave radar, but also to temporarily store data that has been output or will be output.

[0065] In some embodiments, processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 11 or process data, such as executing a similar moving target identification program 10 based on millimeter-wave radar.

[0066] The network interface 13 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface), which is typically used to establish communication connections between the server 1 and other electronic devices.

[0067] The network can be the Internet, a cloud network, a Wi-Fi network, a Personal Area Network (PAN), a Local Area Network (LAN), and / or a Metropolitan Area Network (MAN). Various devices in the network environment can be configured to connect to the communication network according to various wired and wireless communication protocols. Examples of such wired and wireless communication protocols may include, but are not limited to, at least one of the following: Transmission Control Protocol and Internet Protocol (TCP / IP), User Datagram Protocol (UDP), Hypertext Transfer Protocol (HTTP), File Transfer Protocol (FTP), ZigBee, EDGE, IEEE 802.11, Li-Fi, 802.16, IEEE 802.11s, IEEE 802.11g, multi-hop communication, wireless access points (APs), device-to-device communication, cellular communication protocols, and / or Bluetooth communication protocols, or combinations thereof.

[0068] Optionally, the server may also include a user interface, which may include a display, an input unit such as a keyboard, and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be referred to as a screen or display unit, used to display information processed in server 1 and to display a visual user interface.

[0069] Figure 1 Only server 1, which includes components 11-13 and a millimeter-wave radar-based similar moving target identification program 10, is shown. Those skilled in the art will understand that... Figure 1 The structure shown does not constitute a limitation on server 1 and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0070] In this embodiment, the processor 12 can be used to call the millimeter-wave radar-based similar moving target identification program stored in the memory 11 and perform the following operations:

[0071] Acquire raw sensing data from millimeter-wave radar within its detection range;

[0072] The raw sensing data from the millimeter-wave radar is processed to obtain a target sensing dataset for characterizing independent moving targets;

[0073] Extract morphological features reflecting the physical form of independent moving targets and motion features reflecting the motion state of independent moving targets from the target perception dataset;

[0074] The morphological features and motion features are input into a pre-trained classification model to obtain the recognition result of the independent moving target.

[0075] Based on the hardware architecture of the millimeter-wave radar-based similar moving target recognition device described above, an embodiment of the millimeter-wave radar-based similar moving target recognition method of the present invention is proposed.

[0076] Specifically, the "similar moving targets" mentioned in this application refer to two or more moving objects whose key physical and motion characteristics are similar under the perception of a specific sensor (such as the millimeter-wave radar in this application), making it difficult to effectively distinguish them using a single, conventional criterion.

[0077] In the home security or guardianship scenarios addressed in this application, "children" and "pets" are typical similar moving targets, and their similarity is mainly reflected in the following two aspects:

[0078] 1. Physical morphological similarity: A short child, especially an infant crawling, bending over, or sitting, may appear in radar point clouds with morphological features such as height, volume, and aspect ratio very similar to a medium or large pet (such as a Labrador Retriever or Golden Retriever). If the system relies solely on these morphological features for judgment, it could easily misidentify a crawling child as a pet or a standing large dog as a child.

[0079] 2. Similarity of macroscopic movements (speed and trajectory similarity): Children and pets may exhibit significant overlap in their indoor movement speed, activity range, and trajectory patterns. For example, both may run quickly or walk slowly in the living room. Therefore, distinguishing them solely based on the overall speed or trajectory of the target's movement lacks sufficient reliability.

[0080] Based on the aforementioned challenges—namely, the ease with which children and pets, as "similar moving targets," can be confused in a single dimension, leading to low reliability of traditional identification methods—the core objective of the similar moving target identification method proposed in this application is to achieve accurate, reliable, and all-weather differentiation of such targets with similar radar sensing characteristics through an innovative multi-dimensional feature fusion strategy, while protecting user privacy.

[0081] Reference Figure 2 , Figure 2 This is an embodiment of the similar moving target identification method based on millimeter-wave radar of the present invention. The similar moving target identification method based on millimeter-wave radar includes the following steps:

[0082] S10. Acquire raw sensing data of millimeter-wave radar within the radar detection range.

[0083] Specifically, this can be achieved by deploying a millimeter-wave radar sensor in the target monitoring area (such as a living room or bedroom). This radar sensor continuously emits electromagnetic waves into the detection space and receives the echo signals reflected from objects within that space; these raw echo signals constitute the raw sensing data. As a preferred approach, a frequency-modulated continuous wave (FMCW) radar can be used because it can simultaneously and accurately measure the target's range and velocity information. For example, a 60GHz FMCW radar equipped with a 3Tx4Rx time-division multiple access multiple input multiple output (TDM-MIMO) antenna array can acquire IQ (in-phase and quadrature) signals containing rich three-dimensional spatial information (range, azimuth, and elevation) and velocity information as raw sensing data, laying the foundation for subsequent high-precision target separation and feature extraction.

[0084] S20. The raw sensing data from the millimeter-wave radar is processed to obtain a target sensing dataset characterizing independent moving targets. Raw sensing data typically contains a large amount of environmental information and background noise, making direct target identification inefficient and prone to misjudgment. Therefore, by processing and refining this raw data, irrelevant information can be effectively filtered out, highlighting the characteristics of moving targets, thus forming a structured and analyzable dataset, i.e., a target sensing dataset.

[0085] In some embodiments, step S20 can be implemented by the following steps S21-S23:

[0086] In step S21, initial point cloud data for all objects within the radar's detection range are determined from the raw sensing data of the millimeter-wave radar. This step aims to transform the raw electromagnetic wave signal into a three-dimensional geometric representation of all objects in space.

[0087] In some embodiments, the determination of the initial point cloud data of all objects within the radar detection range in step S21 can be achieved through the following steps S211 to S214:

[0088] In step S211, the raw sensing data from the millimeter-wave radar undergoes range-dimensional Fast Fourier Transform (FFT) and Doppler-dimensional FFT to obtain a range-Doppler spectrum of all objects within the radar's detection range. Specifically, by analyzing the data in both range and Doppler (velocity) dimensions, the raw time-domain signal is transformed into a two-dimensional frequency-domain spectrum. In this spectrum, the position of each pixel corresponds to a specific range and velocity value, while its brightness represents the intensity of the signal reflected by the object at that position.

[0089] In step S212, two-dimensional constant false alarm rate (CFAR) detection is performed on the range-Doppler spectrum to identify all target units within the radar detection range and acquire the range data of the target units. Specifically, the range-Doppler spectrum contains a large amount of background noise. To accurately identify real objects, an adaptive detection algorithm such as two-dimensional constant false alarm rate (2D-CFAR) is required. This algorithm traverses the spectrum, dynamically sets a threshold by comparing the signal and noise levels in local areas to determine whether a signal peak is a valid target unit. Once a target unit is identified, its coordinates on the range axis of the spectrum are the range data of that target.

[0090] In step S213, the azimuth and elevation angle data of the target unit are extracted from the raw sensing data of the millimeter-wave radar. Specifically, for each valid target unit detected in step S212, the system needs to determine its orientation in space. This can be achieved by performing additional signal processing (such as performing a cross-antenna FFT) on the echo signals of the same target received by multiple receiving antennas of the radar. By analyzing the phase difference of the signal arriving at different antennas, the direction of arrival of the target, i.e., the azimuth (horizontal direction) and elevation (vertical direction) angle data, can be accurately calculated.

[0091] In step S214, the initial point cloud data is generated based on the distance data, the azimuth data, and the pitch data. Specifically, for each detected target unit, the system obtains a complete set of spherical coordinate system parameters: distance (from S212), azimuth (from S213), and pitch (from S213). Using standard coordinate system transformation formulas, this set of spherical coordinate data can be converted into a three-dimensional coordinate point (x, y, z) in the Cartesian coordinate system. Performing this transformation on all detected target units in the scene, the resulting set of all three-dimensional coordinate points constitutes the initial point cloud data capable of representing all objects within the detection range.

[0092] In step S22, the point cloud data with zero velocity information in the initial point cloud data is removed to obtain the target point cloud data of all moving objects within the detection range.

[0093] Specifically, the initial point cloud data contains a large number of static objects, which are interfering information for moving target recognition tasks. Therefore, by examining the velocity information of each point in the initial point cloud (which was obtained when generating the range-Doppler map) and removing point cloud data with zero or close to zero velocity, static backgrounds can be effectively filtered out, leaving only point clouds related to moving objects in the dataset, greatly reducing the computational load and complexity of subsequent processing.

[0094] In step S23, cluster analysis is performed on the target point cloud data to generate the target perception dataset.

[0095] Specifically, target point cloud data may still contain mixed point clouds of multiple moving targets (e.g., a child and a pet moving simultaneously). To distinguish them, these point clouds need to be clustered. For example, the DBSCAN (Density-Based Noise-Based Spatial Clustering) algorithm can be used, which can separate point cloud clusters belonging to different physical entities based on the density distribution of the point clouds in space. After clustering, each independent point cloud cluster represents an independent moving target, and the collection of all these point cloud clusters constitutes the final target-aware dataset.

[0096] Understandably, by employing a three-step processing flow—from generating the initial point cloud to filtering out static points, and then clustering dynamic points—this scheme can progressively and accurately separate each independent, moving target from the raw and complex radar signal. This refined preprocessing effectively eliminates interference from the static environment and the target itself, thus providing clear, explicit, and structured data objects for subsequent morphological and motion feature extraction.

[0097] S30. Extract morphological features reflecting the physical form of the independent moving target and motion features reflecting the motion state of the independent moving target from the target perception dataset.

[0098] Specifically, to accurately distinguish between similar targets such as children and pets, information from only one dimension is insufficient. One of the core concepts of this solution is to extract the essential differences between targets from two orthogonal dimensions: physical form and motion state.

[0099] In some embodiments, extracting morphological features that reflect the physical form of an independently moving target can be achieved through the following steps S31 to S33.

[0100] In step S31, the surface contour of the independent moving target is extracted based on the target perception dataset. Here, the target perception dataset is the point cloud cluster generated in step S20, representing a single independent moving target. Since the point cloud detected by millimeter-wave radar is itself a set of reflection points of electromagnetic waves from the object's surface, in practice, "extracting the surface contour" in this step usually refers to directly using all the three-dimensional coordinate points of the point cloud cluster as a discrete representation of the target's surface contour.

[0101] In step S32, the minimum bounding cube parameters of the surface profile are calculated. After obtaining the surface profile of the target (i.e., the point cloud cluster), in order to summarize its overall size and shape in a simple and standardized way, the system uses an algorithm from computational geometry to find a cuboid (also called the minimum bounding cube) that can completely enclose the point cloud. The output of this algorithm is the parameters of this cuboid, such as its center point position in three-dimensional space, the orientation of the three axes, and most importantly—the length, width, and height dimensions along these three axes.

[0102] In step S33, the height and aspect ratio parameters of the smallest circumscribed cube are extracted as the morphological features. Specifically, the dimensions of the cube in the vertical direction (usually the Z-axis) are extracted as the target's "height" parameter. Simultaneously, the ratio of its length to width in the horizontal plane (usually the XY plane) is calculated as the target's "aspect ratio" parameter. These two parameters [height, aspect ratio] together constitute the final features characterizing the physical form of the independently moving target.

[0103] Optionally, the height H and aspect ratio R of the minimum circumscribed cube can be set. HW The specific calculations are as follows:

[0104] H = Δz;

[0105] R HW =H / max(Δx,Δy);

[0106] Where Δz refers to the difference in position of the target point cloud below the radar, i.e., the target height; Δx refers to the difference in horizontal position of the target point cloud above the radar, i.e., the target length; and Δy refers to the difference in vertical position of the target point cloud above the radar, i.e., the target width. The specific calculation formulas for these three are as follows:

[0107] Δz=z max -z min ;

[0108] Δx=|x max -x min |;

[0109] Δy=|y max -y min |;

[0110] Among them, z max z min This refers to the maximum and minimum values ​​of the target point cloud at the radar altitude boundary; x max x min This refers to the maximum and minimum values ​​of the target point cloud at the left and right boundaries of the radar; y max y minThis refers to the maximum and minimum values ​​of the target point cloud at the front and rear boundaries of the radar.

[0111] Understandably, by employing a method that geometrically normalizes discrete, irregular point cloud contours using the smallest bounding cube and extracts key dimensions and scale parameters, this approach efficiently abstracts the complex three-dimensional morphological information of a target into a set of low-dimensional but physically meaningful numerical features. This set of features can stably quantify the overall contours of the target, making it particularly effective in distinguishing similar targets with different body structures, such as children (who are typically closer to a "cube") and pets (who are typically more "slender"), and it is insensitive to minute changes in the target's pose, ensuring the robustness of the features.

[0112] In some embodiments, the extraction of motion features reflecting the motion state of an independently moving target in step S30 can be achieved through steps S34 to S36. This process aims to extract deep information from the dynamic changes of the target that can reveal its essential category, serving as a powerful supplement to morphological features.

[0113] In step S34, the Doppler spectrum information of the independent moving target within a specified frequency band is obtained from the target perception dataset. For an independent moving target (i.e., a point cloud cluster), the system analyzes its velocity changes within a short time window. This can be achieved by applying a short-time Fourier transform (STFT) to the target's raw echo signal. The output is a Doppler spectrum (or spectrogram), which shows the distribution of the target's micro-Doppler signal energy over time and frequency. Here, the "specified frequency band" typically refers to the frequency range associated with typical biological movements such as walking and running, for example, 0-25Hz, preferably 6-10Hz, to focus on key gait information.

[0114] In step S35, based on the Doppler spectrum information, the energy integral on the negative frequency side and the energy integral on the positive frequency side of the Doppler spectrum are calculated. The Doppler spectrum is centered at zero frequency; the positive frequency portion typically corresponds to the approaching motion of the target relative to the radar (or the approaching motion component of its body parts), while the negative frequency portion corresponds to its moving away motion. This step calculates the energy integral E on the negative frequency side by integrating the energy values ​​on both sides of zero frequency within a specified frequency band of the spectrum. left (i.e., the energy on the left) and the energy integral E on the positive frequency side right (i.e., energy on the right side).

[0115] Optionally, the energy integral E on the negative frequency side left Energy integral E on the positive frequency side right The calculation formula is as follows:

[0116]

[0117] Where f represents the frequency of the target walking, begin and end are the starting intervals of the energy integral frequency to be calculated, and S(f) is the spectral information obtained after the short-time Fourier transform of the signal to be detected.

[0118] In step S36, the harmonic asymmetry of the independently moving target is calculated based on the energy integrals on the negative frequency side and the positive frequency side, serving as the motion characteristic. To obtain a standardized index that can stably measure motion symmetry, the system calculates the harmonic asymmetry A based on the two energy values ​​obtained in the previous step. The calculation formula can be:

[0119]

[0120] The calculated result A is a value between 0 and 1. Its magnitude intuitively reflects the symmetry of the target's motion energy in the Doppler spectrum, and this value is used as the final motion characteristic.

[0121] Specifically, when a child walks on two legs, the swinging of their torso and limbs exhibits high symmetry relative to their overall forward speed. This results in a near-zero phase difference in their Doppler spectrum caused by arm swings and leg strides. This is reflected in the Doppler spectrum as a relatively balanced energy distribution of the micro-Doppler signal on both the positive and negative frequencies, leading to a low calculated harmonic asymmetry (A) value. Conversely, a pet moves on four legs, exhibiting a more complex gait. The phase relationships of its limbs and the undulations of its body generate significant harmonic components. This is reflected in the Doppler spectrum as a marked asymmetry in the distribution of its micro-Doppler signal, resulting in a higher A value. This significant difference in A values ​​provides a strong basis for classification models.

[0122] Understandably, by employing a method that analyzes the target's micro-Doppler spectrum and quantifies its harmonic asymmetry, this scheme can capture the inherent differences in movement patterns determined by different biomechanical structures (such as bipedal and quadrupedal). This movement characteristic is independent of the target's absolute velocity or size, but rather reflects its "how it moves," thus providing another dimension of information that is completely orthogonal to physical morphological features. This significantly enhances the accuracy and robustness of identification in ambiguous situations such as morphological similarity (e.g., a crawling baby and a pet).

[0123] S40. Input the morphological features and motion features into a pre-trained classification model to obtain the recognition result of the independent moving target. After extracting features that can characterize the multi-dimensional information of the target through step S30, an intelligent classification model is needed to comprehensively analyze these features and make a final, automated category judgment.

[0124] In some embodiments, the pre-trained classification model is a support vector machine (SVM) model. Therefore, inputting the morphological features and the motion features into the pre-trained classification model specifically involves combining the morphological features and the motion features into a three-dimensional feature vector and inputting it into the pre-trained support vector machine model.

[0125] In a specific implementation, the processing of the input 3D feature vector by the support vector machine model may include the following steps:

[0126] S41. Standardize the three-dimensional feature vectors to obtain standardized feature data. The standardization formula is as follows. Since the physical meanings and numerical ranges of the three features—height, aspect ratio, and harmonic asymmetry—are different, directly using the original values ​​may result in features with larger values ​​having excessive weight in the model. Therefore, standardization (e.g., subtracting the mean and dividing by the standard deviation) scales all features to a similar numerical range, eliminating this dimensional difference and helping to improve the model's training efficiency and final performance.

[0127] Alternatively, the formula for standardization is:

[0128]

[0129] Where x′ represents the standardized feature data; x represents the original feature data to be standardized; μ represents the mean of the input feature data; and σ represents the standard deviation of the input feature data. The specific formulas for calculating μ and σ are as follows:

[0130]

[0131]

[0132] Where N is the length of the input data, x i Let i be the feature data of the i-th input.

[0133] S42. The standardized feature data is mapped to a high-dimensional space using a radial basis function (RBF) to obtain a similarity measure. Considering that the boundary between children and pets in the three-dimensional feature space may not be a simple straight line or plane, but a complex nonlinear curve, the RBF kernel function implicitly maps the data from a low-dimensional space to a higher-dimensional feature space. In this high-dimensional space, data points that were originally linearly inseparable become more likely to be clearly separated by a hyperplane, which greatly enhances the model's ability to handle complex classification tasks.

[0134] Alternatively, the formula for the radial basis kernel function is:

[0135] K(x,x i )=exp(-γ||xx i || 2 ),

[0136] Where γ is the kernel parameter, which controls the width of the kernel function;

[0137] S43. Based on the similarity metric, the data in the high-dimensional space is classified using the optimal hyperplane to obtain the recognition result. The core idea of ​​Support Vector Machines (SVMs) is to find a decision boundary, i.e., the optimal hyperplane, that can separate samples of different categories with the maximum margin, while ensuring the correct classification of all training samples. In actual classification, the model determines which side of this optimal hyperplane a new feature vector falls on after being mapped by the kernel function, thereby determining its category.

[0138] Alternatively, in this embodiment, the optimal hyperplane is a 2-dimensional linear decision boundary, which follows the formula:

[0139] w T x+b=0,

[0140] Where w represents the weight vector of the hyperplane, which determines the direction of the hyperplane, and b is the bias term, which determines the distance between the hyperplane and the origin.

[0141] Furthermore, since the core idea is to maximize the margin while ensuring that all classification results in the test set are correct, optimizing the optimal hyperplane becomes obtaining the maximum margin under constraints, where the constraints are:

[0142] y i (w T x i +b)≥1,

[0143] Among them, y i ∈{1,-1}, representing the corresponding category label (1 for children, -1 for pets), a single sample x iThe formula for calculating the distance to the hyperplane is:

[0144]

[0145] It is worth noting that the classification model design in this application is not limited to this. In other embodiments, it can also be set as an artificial neural network model, such as a multilayer perceptron (MLP); an ensemble learning model based on decision trees, such as a random forest model; a gradient boosting decision tree (GBDT) model, such as XGBoost or LightGBM; or a model such as logistic regression, k-nearest neighbors (k-NN), or naive Bayes. The selection of these models can be flexibly adjusted according to different requirements for computing resources, model interpretability, and performance requirements in specific scenarios.

[0146] In some embodiments, after determining the identification result of the independent moving target, the method further includes:

[0147] S50. Transmit the recognition result to an external device for display. This step aims to present the system's internal analysis and judgment results to the user in an intuitive and timely manner, or to link with other intelligent systems, thereby realizing the application value of this technical solution.

[0148] In some embodiments, this transmission and display process can encompass various technical paths and presentation formats. The transmission of identification results can be achieved through various wired or wireless communication protocols. For example, after obtaining the classification result (such as "child" or "pet"), the processing unit within the system can encapsulate it into a data packet and send it to the home's wireless local area network via a built-in Wi-Fi module, or conduct point-to-point communication with nearby devices via a Bluetooth module. In other scenarios requiring high stability, wired connections such as Ethernet can also be used, or other communication protocols suitable for smart home scenarios such as Zigbee and Z-Wave can be employed.

[0149] External devices can take many forms, including personal smart terminals such as smartphones or tablets with accompanying applications (Apps) installed; central control devices in the home environment such as smart speakers, smart home control panels, or wall-mounted displays; and in professional security scenarios, computers in security monitoring centers.

[0150] The display method is not limited to visual presentation. It can be:

[0151] Visual displays: Prompts are given on smartphone or computer screens using text (e.g., directly displaying "Child detected in balcony area"), specific icons (e.g., a child or a paw print icon), or by highlighting the target location on an interior floor plan.

[0152] Auditory display: Play preset voice alerts (e.g., "Attention, a child is approaching the door") via smart speakers or mobile phone speakers, or play different types of alert sounds depending on the recognition results, such as playing a soft alert sound for pets, while playing an emergency alarm sound for children entering dangerous areas.

[0153] Other forms: For example, sending vibration alerts to the user's smartwatch or mobile phone as an auxiliary or silent warning method.

[0154] For example, suppose the identification system in this solution is deployed in a family living room, and the balcony area is pre-designated as a "danger zone" for children. When the system identifies a moving target as a "child" through the aforementioned steps S10 to S40, and the child enters the balcony area, step S50 is triggered. The system sends out an alarm data containing information such as "child," "balcony," and "danger" via Wi-Fi. At this time, the guardian resting in the bedroom receives this data on their smartphone's accompanying app, and a full-screen, high-priority push notification immediately pops up with the message "Level 1 Alert: Child has entered the balcony danger zone!" Simultaneously, the phone emits a continuous, high-decibel alarm sound and vibrates until the guardian manually confirms.

[0155] It is understandable that by integrating standardized communication protocols and enabling interaction with various smart devices used by users daily, this application can transform the results of internally abstract recognition algorithms into concrete, perceptible, and attention-grabbing actionable information for users. This step ensures that the system's perception and analysis capabilities can be effectively translated into proactive intervention and security in the real world, completing the key closed loop from "intelligent perception" to "effective alarm," greatly enhancing the sense of security and convenience for users (especially guardians of children).

[0156] Based on the above embodiments, it can be seen that the method for identifying similar moving targets in this application has the following beneficial effects:

[0157] 1. It combines strong privacy protection with high environmental adaptability, breaking through the limitations of application scenarios:

[0158] Unlike camera-based solutions, the millimeter-wave radar in this application perceives targets by analyzing electromagnetic wave echoes, without involving image acquisition at all, generating abstract three-dimensional point cloud data. This fundamentally solves user concerns and regulatory risks associated with deploying surveillance equipment in privacy-sensitive areas such as bedrooms, bathrooms, and hotel rooms. Furthermore, the physical properties of millimeter waves allow them to penetrate smoke and dust, and their performance is unaffected by environmental factors such as light intensity, darkness, and inclement weather, ensuring stable operation in all weather conditions and scenarios. Compared to visual solutions whose performance deteriorates drastically at night or in low light, this offers an overwhelming advantage in environmental adaptability.

[0159] 2. High discrimination accuracy, capable of precisely identifying targets of similar size:

[0160] Traditional identification methods rely solely on a single dimension such as height, making them prone to confusion when faced with targets of similar size (e.g., crawling infants versus medium-sized pets). This solution extracts the height and aspect ratio of the smallest bounding cube to construct morphological features that reflect the overall contour of the target's "height, size, and build." More importantly, this application introduces a motion feature dimension that is completely orthogonal to the physical morphology. By analyzing the Doppler spectrum and calculating harmonic asymmetry, it can capture the subtle differences in intrinsic movement patterns produced by different biological gaits (e.g., a child's bipedal walking versus a pet's quadrupedal movement), differences that are difficult to discern with the naked eye. Finally, by inputting these two dimensions (morphology + motion) into a pre-trained classification model such as a Support Vector Machine (SVM), even if one dimension of the target's features (e.g., morphology) is ambiguous or similar, the significant difference in the other dimension (motion) can provide the model with a decisive basis for accurate judgment. This multi-feature fusion design greatly enhances the discriminative power of the feature set, thereby fundamentally improving the accuracy and robustness of identification.

[0161] 3. Full-process automation and high practicality enable a closed loop from intelligent sensing to effective alarm:

[0162] On the one hand, this application achieves automatic and accurate locking of independent moving targets by filtering, denoising, and clustering the raw data, without any manual intervention. On the other hand, this application automatically transforms the internal, machine-readable classification results into timely alerts that are perceptible to users on external devices (such as mobile app push notifications, smart speaker broadcasts, etc.) by using standardized communication protocols such as Wi-Fi and Bluetooth for the transmission and display of recognition results.

[0163] This end-to-end automated design makes this application not just an algorithm model, but a practical product that can be seamlessly integrated into smart home and security systems. It effectively transforms abstract analysis and judgment into actionable information that can remind guardians and ensure children's safety, completing the key closed loop from "perception" to "action," and has significant social value and promising industrial application prospects.

[0164] Furthermore, this invention also proposes a computer-readable storage medium, which can be any one or any combination of several of the following: hard disk, multimedia card, SD card, flash memory card, SMC, read-only memory (ROM), erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, etc. The computer-readable storage medium includes a millimeter-wave radar-based similar moving target recognition program 10. The specific implementation of the computer-readable storage medium of this invention is largely the same as the aforementioned millimeter-wave radar-based similar moving target recognition method and the specific implementation of server 1, and will not be repeated here.

[0165] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0166] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), 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 apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, 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.

[0167] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function 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.

[0168] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable 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.

[0169] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0170] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying similar moving targets based on millimeter-wave radar, characterized in that, include: Acquire raw sensing data from millimeter-wave radar within its detection range; The raw sensing data from the millimeter-wave radar is processed to obtain a target sensing dataset for characterizing independent moving targets; Extract morphological features reflecting the physical form of independent moving targets and motion features reflecting the motion state of independent moving targets from the target perception dataset; The morphological features and motion features are input into a pre-trained classification model to obtain the recognition result of the independent moving target.

2. The similar moving target identification method based on millimeter-wave radar as described in claim 1, characterized in that, The raw sensing data from the millimeter-wave radar is processed to obtain a target sensing dataset for characterizing independent moving targets, including: The initial point cloud data of all objects within the radar's detection range are determined from the raw sensing data of the millimeter-wave radar. Remove the point clouds with zero velocity information from the initial point cloud data to obtain the target point cloud data of all moving objects within the detection range; Cluster analysis is performed on the target point cloud data to generate the target perception dataset.

3. The similar moving target identification method based on millimeter-wave radar as described in claim 2, characterized in that, The initial point cloud data of all objects within the radar detection range is determined from the raw sensing data of the millimeter-wave radar, including: The range-Doppler maps of all objects within the radar's detection range are obtained by performing range-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform on the raw sensing data of the millimeter-wave radar. Two-dimensional constant false alarm rate (CFAR) detection is performed on the range-Doppler spectrum to identify all target units within the radar detection range and acquire the range data of the target units; The azimuth and elevation angle data of the target unit are extracted from the raw sensing data of the millimeter-wave radar; The initial point cloud data is generated based on the distance data, the azimuth data, and the pitch data.

4. The similar moving target identification method based on millimeter-wave radar as described in claim 3, characterized in that, The morphological features reflecting the physical form of independently moving targets are extracted from the target perception dataset, including: Based on the target perception dataset, extract the surface contour of the independent moving target; Calculate the minimum bounding cube parameters of the surface profile; The height and aspect ratio parameters of the minimum bounding cube are extracted as the morphological features.

5. The similar moving target identification method based on millimeter-wave radar as described in claim 4, characterized in that, Motion features reflecting the motion state of independently moving targets are extracted from the target perception dataset, including: Obtain the Doppler spectrum information of the independent moving target in the specified frequency band from the target perception dataset; Based on the Doppler spectrum information, calculate the energy integral on the negative frequency side and the energy integral on the positive frequency side of the Doppler spectrum; The harmonic asymmetry of the independently moving target is calculated based on the energy integrals on the negative frequency side and the positive frequency side, and is used as the motion characteristic.

6. The method for identifying similar moving targets based on millimeter-wave radar as described in any one of claims 1 to 5, characterized in that, The pre-trained classification model is a support vector machine model; Inputting the morphological features and the motion features into a pre-trained classification model includes: combining the morphological features and the motion features into a three-dimensional feature vector and inputting it into a pre-trained support vector machine model.

7. The similar moving target identification method based on millimeter-wave radar as described in claim 6, characterized in that, The support vector machine model processes the input three-dimensional feature vector to obtain the recognition result, including: The three-dimensional feature vectors are standardized to obtain standardized feature data; The standardized feature data is mapped to a high-dimensional space using a radial basis function (RBF) kernel to obtain a similarity metric. The formula for the RBF kernel is as follows: K(x,x i )=exp(-γ||x-x i || 2 ), Where γ is the kernel parameter, which controls the width of the kernel function; Based on the similarity metric, the data in the high-dimensional space is classified using an optimal hyperplane to obtain the recognition result. The optimal hyperplane follows the formula below: w T x+b=0, Where w represents the weight vector of the hyperplane, which determines the direction of the hyperplane, and b is the bias term, which determines the distance between the hyperplane and the origin. Furthermore, the hyperplane satisfies the following constraints: y i (w T x i +b)≥1, Among them, y i ∈{1,-1} represents the corresponding category label.

8. The similar moving target identification method based on millimeter-wave radar as described in claim 1, characterized in that, After determining the identification result of the independent moving target, the method further includes: The recognition results are transmitted to an external device for display.

9. A similar moving target identification device based on millimeter-wave radar, characterized in that, The method includes a memory, a processor, and a millimeter-wave radar-based similar moving target identification program stored in the memory and executable on the processor. When the processor executes the millimeter-wave radar-based similar moving target identification program, it implements the millimeter-wave radar-based similar moving target identification method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a similar moving target recognition program based on millimeter-wave radar, which, when executed by a processor, implements the similar moving target recognition method based on millimeter-wave radar as described in any one of claims 1-8.

Images