Human Presence Detection Method Based on Dual-Source Synergistic Fusion of Millimeter-Wave Radar and Infrared Pyroelectric Sensor

Abstract

This invention discloses a human presence perception method based on the dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensors. The method processes raw data collected by the millimeter-wave radar sensor receiver to obtain a high-quality range-velocity dimension data matrix; then performs SVD decomposition to reconstruct two new data matrices, corresponding to strong target signals and weak target signals respectively; the human presence perception output of the millimeter-wave radar sensor is obtained through trajectory features and breathing features; the human presence perception output of the infrared pyroelectric sensor is obtained by processing the signal from the infrared pyroelectric sensor; then, a heterogeneous data matrix is ​​constructed based on the outputs of the two sensors, and the heterogeneous data matrix is ​​processed by a fusion algorithm to obtain the fusion result. Finally, the fusion result is stored in a delayed memory matrix, and the timeliness is calculated according to the timeliness factor corresponding to different delay units, thus obtaining the final human presence perception result.

Description

Technical Field

[0001] This invention relates to the fields of fusion application technology of millimeter-wave radar and infrared pyroelectric sensors and the field of intelligent sensing technology, and in particular to a human presence sensing method based on the dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensors. Background Technology

[0002] With the rapid advancement of technology and the continuous evolution of society, the concept of smart living has taken root in people's hearts, especially in the field of smart homes, where people have an increasingly strong pursuit of intelligent and convenient lifestyles. Under this trend, human presence sensing technology, as a key component in building smart living environments, is becoming increasingly important. Traditional image-based human detection technology, while powerful, faces problems such as high cost, high algorithm complexity, privacy risks, and limited application in complex environments. In contrast, pyroelectric infrared (PIR) sensors, with their low cost, low energy consumption, simple installation process, and keen ability to capture human infrared signals, have gradually become the favorite in the field of indoor human presence detection. Especially when integrated with wireless distributed sensor networks and advanced machine learning algorithms, the application potential of PIR sensors is greatly expanded, making their deployment in various scenarios more widespread and efficient. However, the raw signal output of pyroelectric sensors is susceptible to changes in heat sources, target movement speed, and position, which limits their performance in high-precision human presence detection.

[0003] To overcome this limitation, a more stable and accurate method for detecting human presence is needed that can compensate for the shortcomings of a single sensor. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a human presence sensing method based on the dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor. This method utilizes human presence sensing technology and combines a dual-source collaborative sensing method of millimeter-wave radar sensor and pyroelectric infrared sensor.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The present invention provides a human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor, comprising the following steps:

[0007] S1. Obtain the raw data collected by the millimeter-wave radar sensor receiver and process it to obtain a data matrix in the range-velocity dimension;

[0008] S2. Perform SVD decomposition on the range-velocity spectrum data matrix, and reconstruct two new data matrices by comparing singular values ​​with thresholds and combining eigenvalues ​​and eigenvectors, which correspond to strong target signals and weak target signals, respectively.

[0009] S3. Process the strong target signal and the weak target signal respectively, and obtain the human presence perception output of the millimeter-wave radar sensor through trajectory features and breathing features;

[0010] S4. Acquire the signal from the infrared pyroelectric sensor and process it to obtain a voltage signal. Compare the voltage signal with a preset threshold to obtain the human presence sensing output result of the infrared pyroelectric sensor.

[0011] S5. Construct a heterogeneous data matrix based on the sensing output results obtained from the millimeter-wave radar sensor and the infrared pyroelectric sensor respectively, and process the heterogeneous data matrix through a fusion algorithm to obtain the dual-source fusion result.

[0012] S6. Store the fusion result into the delayed memory matrix, calculate the timeliness according to the timeliness factor corresponding to different delay units, and thus obtain the final human existence perception result.

[0013] Furthermore, the raw data in step S1 is processed according to the following steps:

[0014] S101. A one-dimensional fast Fourier transform is used to process the time-domain signal of each frame, converting the signal from the time domain to the frequency domain.

[0015] S102. Noise removal is performed using a dynamic filtering method;

[0016] S103. Perform incoherent superposition to accumulate the target's echo energy;

[0017] S104. The signal is extended from the one-dimensional frequency domain to the two-dimensional frequency domain using a two-dimensional fast Fourier transform, and a range-Doppler domain data matrix containing target distance and radial velocity information is constructed.

[0018] S105. Perform zero-velocity target elimination, that is, eliminate objects with no radial velocity.

[0019] Furthermore, the strong target signal in step S2 is processed according to the following steps:

[0020] S301. Perform 3DFFT processing on the separated strong target signal to construct a high-dimensional representation of the target in three dimensions: range, Doppler, and azimuth, and form point cloud data.

[0021] S302. After obtaining the point cloud data, perform clustering operations to group scattering points from the same target together to form independent point clusters;

[0022] S303. Use a Kalman filter to process each point cluster and update the slant range and azimuth of the observed target.

[0023] S304. The nearest neighbor rule is used to associate point clusters between consecutive frames, thereby establishing the target trajectory and realizing target tracking;

[0024] S305. Obtain a strong target displaying a continuous trajectory. Determine whether it is a human target based on the continuous trajectory. If it is, mark the detection result as "human presence" and output 1; otherwise, if it is not, mark it as an interference target and mark it as "human absence" and output 0.

[0025] Furthermore, the weak target signal in step S3 is processed according to the following steps:

[0026] S306. Extract feature vectors from the separated weak target signals to obtain phase data at the range gate where the target is located;

[0027] S307, extract phase unwinding, phase difference, smoothing filtering, and breathing value respectively;

[0028] S308. By analyzing the processed phase data, identify the periodic phase changes caused by respiration, and make judgments as follows:

[0029] If respiratory features are detected in a weak target signal, it indicates that the target has vital signs. Therefore, it is confirmed as a human target, and the final human presence detection result is marked as "human presence" and output 1.

[0030] Conversely, if no respiratory features are detected, the target is identified as an interference target and marked as "human body does not exist", and 0 is output.

[0031] Furthermore, the signal from the infrared pyroelectric sensor is processed in step S4, specifically as follows:

[0032] S401. Divide the space within the field of view of the infrared pyroelectric sensor into multiple independent monitoring sub-regions;

[0033] S402. When a human body or other warm target enters the sensor's field of view and crosses these monitoring sub-areas, the pyroelectric element inside the sensor acquires the temperature difference signal and converts the temperature difference signal into an electrical signal.

[0034] S403. Perform threshold detection and logical judgment. Compare the generated voltage signal with the preset threshold. If the voltage signal exceeds the preset threshold, it indicates that there is a moving human target in the detection space. Mark the human presence detection result as "human presence" and output 1. Otherwise, if the voltage does not exceed the preset threshold, it is determined that there is no moving human target in the detection space and marked as "human absence", output 0.

[0035] Furthermore, the strong target signal and the weak target signal in step S2 are separated in the following manner:

[0036] The range-velocity spectrum data matrix is ​​decomposed using SVD. The mathematical form of SVD is:

[0037] ;

[0038] in: It is An orthogonal matrix whose column vectors are the left singular vectors of ;

[0039] It is A diagonal matrix, where the elements on the diagonal are the singular values ​​of , usually arranged in descending order;

[0040] It is An orthogonal matrix whose column vectors are the right singular vectors of ;

[0041] By comparing singular values ​​with a threshold, two data matrices corresponding to strong and weak targets are obtained, respectively.

[0042] Furthermore, in step S401, dividing the space within the field of view of the infrared pyroelectric sensor into multiple independent monitoring sub-regions is performed in the following manner:

[0043] By dividing the Fresnel lens into several different regions to monitor different areas within the field of view, each sub-region can be regarded as an independent detection unit, capable of independently sensing temperature changes;

[0044] When a human body or other warm target enters the sensor's field of view and crosses these monitoring sub-regions, a temperature difference is generated. The pyroelectric element inside the sensor can convert this temperature difference into an electrical signal.

[0045] Furthermore, the preset threshold in step S4 is calculated according to the following formula:

[0046] ;

[0047] in:

[0048] For the first The received voltage intensity of each unit;

[0049] The number of reference units;

[0050] This is a threshold factor, usually set according to the required false alarm probability.

[0051] Furthermore, the heterogeneous data matrix in step S5 is constructed in the following manner:

[0052] Acquire human presence perception results from millimeter-wave radar sensors Human presence perception results output by infrared pyroelectric sensor ;

[0053] The sampling timestamp is determined based on the sampling frequency of the two sensors, and the outputs of the millimeter-wave radar sensor and the infrared pyroelectric sensor are aligned using the timestamp.

[0054] The aligned results are merged into a single matrix to obtain a two-dimensional heterogeneous data matrix, where each row corresponds to a time point and each column corresponds to a sensor.

[0055] Furthermore, in step S5, obtaining the dual-source fusion result based on the heterogeneous data matrix is ​​specifically performed in the following manner:

[0056] The fusion results of the millimeter-wave radar sensor and the infrared pyroelectric sensor were obtained by using a weighted average method on the heterogeneous data matrix. ,Right now

[0057] ;

[0058] in, This represents the weighted average fusion result; This indicates the number of data points used to calculate the average. express The detection results of the infrared pyroelectric sensor at that time; express The detection results of the millimeter-wave radar sensor at that moment;

[0059] This represents the weighted average result obtained from millimeter-wave radar sensors; This represents the weighted average result obtained from the infrared pyroelectric sensor;

[0060] The fusion result is obtained based on the weighted average threshold method. ;

[0061] Then, the fusion result of the millimeter-wave radar sensor and the infrared pyroelectric sensor is obtained using the weight allocation method. ,Right now

[0062] ;

[0063] The trust ratios for millimeter-wave radar sensors and infrared pyroelectric sensors are as follows:

[0064] ;

[0065] The fusion result is obtained by setting a threshold based on the weight allocation method. ;

[0066] Finally, the fusion results were analyzed. and fusion results Perform a logical OR operation to obtain the dual-source fusion result. ,Right now:

[0067] ;

[0068] Among them, the logical "OR" operator This means that if any fusion result is 1, the final result is 1; if both are 0, the final result is 0.

[0069] The beneficial effects of this invention are as follows:

[0070] This invention provides a human presence sensing method based on the dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensors. This method utilizes the complementary characteristics of the two sensors: the millimeter-wave radar can operate independently of lighting conditions, while the infrared pyroelectric sensor responds rapidly to temperature changes and human movement. Through data fusion algorithms, this method not only improves detection accuracy and robustness, reducing false alarms and false negatives, but also adapts to various environmental conditions, enabling more complex behavior recognition, such as distinguishing between stationary and moving human bodies. Furthermore, by not capturing clear images, this method protects user privacy while optimizing efficiency through flexible energy management strategies, providing a more comprehensive, accurate, and privacy-friendly human presence sensing solution, demonstrating strong application potential in smart homes and other fields.

[0071] Meanwhile, by constructing a dual-source fusion system integrating millimeter-wave radar and infrared pyroelectric sensors, and designing a decision-level fusion algorithm, the shortcomings of a single sensor are compensated for, achieving more stable and accurate human presence perception. This effectively improves the accuracy and reliability of human presence detection, significantly reduces the false alarm rate, and provides strong technical support for the optimization and upgrading of smart home systems.

[0072] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0073] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0074] Figure 1 This is a system architecture diagram for this embodiment.

[0075] Figure 2 This is the algorithm flowchart in this embodiment.

[0076] Figure 3 This is a schematic diagram of the indoor testing scenario in this embodiment.

[0077] Figure 4 This is a schematic diagram of the monitoring area division in this embodiment.

[0078] Figure 5 This is a schematic diagram of the test category division in this embodiment.

[0079] Figure 6 This is an example diagram of the host computer's display interface in this embodiment.

[0080] Figure 7 This is an example diagram comparing the transmitted and received signals of the millimeter-wave radar sensor in this embodiment.

[0081] Figure 8 This is a thermal example diagram of the decomposed strong target signal RDM in this embodiment.

[0082] Figure 9 This is an example diagram of the weak target signal after decomposition in this embodiment.

[0083] Figure 10 This is an example diagram showing the trajectory results of a strong target obtained by the millimeter-wave radar sensor in this embodiment.

[0084] Figure 11 This is an example diagram showing the breathing results of weak targets obtained by the millimeter-wave radar sensor in this embodiment.

[0085] Figure 12 This is an example diagram of the visible field of view partitioning of the infrared pyroelectric sensor in this embodiment.

[0086] Figure 13 This is an example diagram of the output signal waveform of the infrared pyroelectric sensor in this embodiment.

[0087] Figure 14 This is an example diagram showing the output results of the weighted average method and the weight allocation method in this embodiment.

[0088] Figure 15 This is an example diagram showing the output result of the delayed memory matrix in this embodiment.

[0089] Figure 16 This is an example diagram of the output result of human presence perception in this embodiment. Detailed Implementation

[0090] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0091] Example 1

[0092] like Figure 1 As shown, Figure 1 This is a system architecture diagram for this embodiment. In practical applications, the orientation and installation position of the infrared pyroelectric sensor and millimeter-wave radar sensor in the system described in this embodiment can be set as follows. In this preferred embodiment, a binary pyroelectric sensor and a 60G millimeter-wave radar sensor will be used to transmit and receive signals.

[0093] The system includes a central processing unit connected to a millimeter-wave radar sensor, an infrared pyroelectric sensor, a human presence buzzer, and a human presence indicator light.

[0094] The central processing unit is equipped with a detection control module, an algorithm control module, and an output control module.

[0095] The detection and control module is used to collect data from the millimeter-wave radar sensor and the infrared pyroelectric sensor, and to perform preliminary processing on this data, such as filtering and amplification, to ensure the accuracy and reliability of the data. Simultaneously, this module is also responsible for coordinating the operating states of the two sensors to ensure they operate under optimal conditions, such as adjusting the sensor sensitivity or operating mode.

[0096] The algorithm control module is used to perform advanced data analysis and processing tasks. Based on data received from the detection control module, it uses specific algorithms to determine the presence of a human body. This may involve various techniques such as motion detection, temperature change analysis, and pattern recognition. Furthermore, this module can perform more in-depth analysis of the data, such as estimating the person's position and direction of movement.

[0097] The output control module is used to control the behavior of the human presence buzzer and human presence indicator light based on the analysis results of the algorithm control module. For example, when a person is detected, the buzzer can be activated to sound an alarm, and the indicator light can be illuminated to provide a visual cue.

[0098] The millimeter-wave radar sensor emits electromagnetic waves that propagate into the surrounding environment. When these waves encounter an object (especially a moving human body), they are reflected back. By analyzing the reflected waveform, the sensor can detect the object's presence, relative position, velocity, and other information. In this system, it is primarily used to detect human movement.

[0099] The infrared pyroelectric sensor is used to detect infrared radiation emitted by the human body. When a person enters the sensor's sensing range, the sensor can detect the change in temperature difference between the human body and the background temperature, thus enabling the detection of the human presence. This sensor is particularly suitable for detecting stationary or slowly moving targets.

[0100] The human body is equipped with a buzzer, which is used to sound an alarm when a human body is detected. As one of the system's alarm mechanisms, it helps users to know in a timely manner whether someone is present in the environment;

[0101] The human presence indicator light illuminates when a human body is detected, providing a direct visual signal. When used in conjunction with a buzzer, it can provide users with more comprehensive warning information.

[0102] like Figure 2 As shown, Figure 2 This is a flowchart of the algorithm processing in this embodiment, which includes a millimeter-wave radar sensor algorithm processing sub-process, an infrared pyroelectric sensor algorithm processing sub-process, and a decision-level algorithm processing process.

[0103] The human presence sensing method based on the dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor provided in this embodiment includes the following steps:

[0104] S1. Process the raw ADC (Analog-to-Digital Converter) data acquired by the radar to extract and analyze information about the presence of the human body, specifically:

[0105] S101 First, a one-dimensional fast Fourier transform (1D FFT) is used to process the time-domain signal of each frame, converting the signal from the time domain to the frequency domain.

[0106] S102 then implements pulse cancellation, a dynamic filtering method used to remove the effects of static targets (such as walls and furniture) and background noise (clutter) in the environment. This step is crucial for extracting weak dynamic target signals from the home environment;

[0107] S103 Secondly, incoherent superposition is performed to enhance the weak target signal. This process involves superimposing multiple consecutive radar echo signals to accumulate the target's echo energy, improve the signal-to-noise ratio, and make the weak target more obvious.

[0108] After completing the above steps, S104 further performs a two-dimensional fast Fourier transform (2D FFT) to extend the signal from the one-dimensional frequency domain to the two-dimensional frequency domain, namely the range-Doppler domain. This helps to construct a complete data matrix containing target range and radial velocity information, facilitating subsequent processing.

[0109] S105 Finally, zero-velocity target rejection, that is, rejecting objects that have no radial velocity. This operation is crucial for detecting moving targets (such as the human body) because the echo of a stationary target may mask or interfere with the true target signal;

[0110] S2. Perform singular value separation on the processed data matrix to identify strong target signals and weak target signals respectively.

[0111] S3. Process the strong target signal and the weak target signal respectively, and obtain the human presence perception output of the millimeter-wave radar sensor through trajectory features and breathing features;

[0112] Then, strong and weak targets are processed separately, specifically as follows:

[0113] Strong target processing procedure:

[0114] S301 first performs 3DFFT processing on the separated strong target signal to construct a high-dimensional representation of the target in three dimensions: range, Doppler, and azimuth, forming point cloud data;

[0115] After obtaining the point cloud, S302 performs clustering operations to group scattering points from the same target together, forming independent point clusters. This step helps distinguish different targets, especially when multiple targets are present in the radar's field of view simultaneously.

[0116] S303 then uses a Kalman filter to process each point cluster, updating the slant range and azimuth of the observed target. The Kalman filter is an effective recursive algorithm for estimating the system state over time, providing stable tracking performance even in noisy environments.

[0117] S304 then uses the nearest neighbor rule to associate point clusters between consecutive frames, thereby establishing the target trajectory and achieving target tracking.

[0118] S305 Finally, if a strong target shows a coherent trajectory, indicating that it has continuous movement in space, then it can be identified as a human target, and the final human presence detection result is marked as "human presence" (output 1); otherwise, if there is no trajectory, then it can be identified as an interfering target and marked as "human absence" (output 0).

[0119] Weak target handling process:

[0120] S306 performs feature vector extraction on the separated weak target signal to obtain the phase data value at the range gate where the target is located.

[0121] S307 then performs phase unwinding, phase difference, smoothing filtering, and breath value extraction processes respectively.

[0122] S308 can identify periodic phase changes caused by respiration by analyzing the processed phase data. If obvious respiratory features are detected in a weak target signal, it indicates that the target has vital signs and can be confirmed as a human target. The final human presence detection result is marked as "human presence" (output 1). Conversely, if no respiratory features are detected, it can be identified as an interfering target and marked as "human absence" (output 0).

[0123] like Figure 3 The diagram shows a fusion sensor installed laterally facing the monitoring area in this embodiment to ensure maximum coverage. The fusion sensor in this embodiment is installed in an open space, specifically at a fixed point below the ceiling. The sensor's position is chosen to maximize coverage of the monitored area, with the sensor facing the monitoring area, meaning its primary sensing zone points towards the interior space of the room. This arrangement ensures the sensor can effectively capture information from all angles, thereby reducing blind spots and improving overall monitoring effectiveness.

[0124] like Figure 4 The diagram shown is a top view of the monitoring area division in this embodiment.

[0125] Figure 4A rectangular grid-like plane exists, representing the specific area to be monitored. For ease of management and analysis, this large area is further divided into several smaller sub-regions, each a square cell. At the bottom center of the image, a small circle marks the actual installation location of the fusion sensor. The sensor is positioned at the edge of the entire monitoring area, but not directly against the boundary; it is slightly offset. This layout strategy helps to expand the sensor's effective coverage, enabling it to better monitor activity throughout the area. The sensor's coverage is fan-shaped, covering most of the grid cells. This means that as long as a target object is within these cells, it is likely to be detected by the sensor. However, some cells are outside the coverage area; these areas may become blind spots and cannot be reached by the current sensor configuration.

[0126] S4. Process the signal from the infrared pyroelectric sensor, specifically as follows:

[0127] S401 First, the space within the field of view (FOV) of the infrared pyroelectric sensor is finely divided to form multiple independent monitoring sub-regions;

[0128] S402 Then, when a human body or other warm target enters the sensor's field of view and crosses these monitoring sub-regions, a temperature difference is generated. The pyroelectric element inside the sensor converts this temperature difference into an electrical signal, typically a voltage change. This conversion is based on the thermoelectric effect, where temperature changes cause changes in the concentration of charge carriers in the material, leading to charge accumulation and voltage output.

[0129] S403 finally performs threshold detection and logical judgment. The generated voltage signal is compared with a preset threshold. If the voltage signal exceeds this threshold, it means that a significant temperature change has been detected, indicating that there is a moving human target in the detection space, and the final human presence detection result is marked as "human presence" (output 1); conversely, if the voltage does not exceed the threshold, it can be determined that there is no moving human target in the detection space, and it is marked as "human absence" (output 0).

[0130] S5. Construct a heterogeneous data matrix based on the output results of the millimeter-wave radar sensor and the infrared pyroelectric sensor; after processing the constructed heterogeneous data matrix using the weighted average method and the weight allocation method, obtain the dual-source fusion result.

[0131] S6. Based on the dual-source fusion results, the stability of the results is measured using the delayed memory matrix. The stabilized delayed memory matrix is ​​then substituted into the validity calculation formula to obtain the validity of the current result, thus obtaining the final judgment result.

[0132] The method provided in this embodiment utilizes the complementary characteristics of two sensors: millimeter-wave radar can operate independently of lighting conditions, while the infrared pyroelectric sensor responds rapidly to temperature changes and human movement. Through data fusion algorithms, this method not only improves detection accuracy and robustness, reducing false alarms and false negatives, but also adapts to various environmental conditions, enabling more complex behavior recognition, such as distinguishing between stationary and moving human bodies. Furthermore, by not capturing clear images, this method protects user privacy while optimizing efficiency through flexible energy management strategies, providing a more comprehensive, accurate, and privacy-friendly human presence sensing solution, demonstrating strong application potential in smart homes and related fields.

[0133] Example 2

[0134] Exemplary embodiments of the present invention will now be described in more detail with reference to the accompanying drawings.

[0135] like Figure 5 The diagram shown is a schematic of the test cases in this embodiment. Figure 5 The document showcases a series of test cases designed to evaluate sensor performance under various conditions. Each test case lists different heights and radii (Rm). These parameters can be used to define the sensor's detection range or the dimensions of the test environment. It also includes various interference sources and normal test items to simulate real-world scenarios and verify the sensor's reliability and accuracy. These scenarios cover both static and dynamic aspects, facilitating a comprehensive evaluation of the sensor's responsiveness to different types of actions and environmental changes. By organizing test cases in this way, researchers can systematically examine the sensor's performance in complex and ever-changing real-world situations, thereby improving design, product usability, and user experience.

[0136] like Figure 6 The image shown is the interface of the host computer in this embodiment. This embodiment can identify whether there is a human target in the monitoring area. For example, it can identify the presence of a human body in normal activities such as walking, standing, sitting, and squatting, as well as distinguish the absence of a human body in interfering activities such as fan shaking, curtain swaying, fan blowing on green plants, and chair sliding in, so as to achieve more stable and accurate human presence perception.

[0137] By processing measured data of the presence of a human body in a scene, according to the processing steps of this embodiment:

[0138] Step 1: Process the raw data collected by the radar.

[0139] First, the transmitted and received waveforms acquired by millimeter-wave radar are mixed and filtered to obtain the intermediate frequency (IF) signal. The time-domain waveforms of the transmitted and received signals are compared, as shown below. Figure 7As shown in a), the frequency changes with time, for example, as shown in a). Figure 7 As shown in b) of the diagram. An FFT is performed on the intermediate frequency signal to obtain a set of complex signals in the frequency domain. Subsequently, multiple sets of complex matrix signals with different distance and time units are acquired through slow-time data accumulation. Next, the echo signals at each sampling point in the frequency domain echo signal are subtracted from their corresponding signals to obtain the frequency domain echo signal after pulse cancellation filtering, thus removing static clutter. Then, the filtered echo signals undergo incoherent superposition processing to accumulate the target's echo energy and improve the signal-to-noise ratio. After two-dimensional FFT transformation, a range-velocity spectrum containing the target's distance and velocity is obtained. Finally, zero-velocity units containing objects without radial velocity in the spectrum are removed to prevent the echoes of stationary targets from masking or interfering with the true target signal.

[0140] Step 2: Separate strong and weak targets in the range-velocity spectrum.

[0141] Perform SVD decomposition on the range-velocity spectrum data matrix. The mathematical form of SVD can be expressed as:

[0142]

[0143] in: It is An orthogonal matrix whose column vectors are the left singular vectors of . It is A diagonal matrix, where the elements on the diagonal are the singular values ​​of , usually arranged in descending order. It is An orthogonal matrix is ​​, whose column vectors are the right singular vectors of .

[0144] By comparing singular values ​​with a threshold, it's possible to determine which signals correspond to strong targets and which to weak targets. Based on this classification, we can then proceed from... and Select the corresponding column vector from the list and combine it with... To reconstruct two new data matrices, corresponding to the strong target and the weak target respectively.

[0145] Therefore, after processing in step 1, the dimension is obtained as follows: The range-velocity spectrum data matrix is ​​a two-dimensional array. Each row and column consists of a series of floating-point numbers containing target-related information. SVD decomposition of the range-velocity spectrum data matrix yields a matrix with dimensions of [missing information]. diagonal matrix Its main characteristics are as follows: Non-zero elements on the diagonal: These elements are called singular values, and they reflect the main energy distribution of the original data. Larger singular values ​​correspond to the most important components in the data, while smaller singular values ​​correspond to minor components or noise.

[0146] Sparsity: The observation of many singular values ​​close to zero indicates that the original data may have some redundancy or low-rank characteristics. This property allows for effective dimensionality reduction by retaining only a few of the largest singular values ​​without losing too much information.

[0147] Energy concentration: Most of the energy is concentrated in the large singular values ​​in the first few rows / columns, while the smaller singular values ​​at the end decay rapidly to almost zero. This phenomenon is consistent with the characteristics of many natural signals and image data, where important information is often concentrated in a few key features.

[0148] Arranging the singular values ​​in the diagonal matrix in descending order yields the largest singular value, which is then taken. As a threshold for segmentation, two data matrices corresponding to strong and weak targets can be obtained after threshold comparison.

[0149] like Figure 8 The image shown is a thermal example diagram of the decomposed strong target signal RDM in this embodiment. Figure 8 The chart presents data on strong target signals after SVD decomposition. In this example, the horizontal axis represents velocity information, and the vertical axis represents range information. The color coding of the chart reflects the energy intensity, with darker colors indicating greater intensity. It can be seen that distinct bright areas appear under certain time and frequency combinations, indicating higher signal intensity in these areas, which may be the locations of targets of interest. Additionally, lighter-colored stripes are distributed elsewhere, which may represent background noise or other irrelevant signal components. By comparing the differences between these areas, it is easier to identify which parts contain truly useful information.

[0150] like Figure 9 The image shown is an example of the weak target signal after decomposition in this embodiment. Figure 9 The chart presents data on weak target signals after SVD decomposition. It includes three curves, each representing signal strength changes over different time periods. Some curves show sharp peaks, reflecting the instantaneous gain effect caused by the weak target's breathing and heartbeat; others show gentler fluctuations, more likely reflecting background noise levels. By comparing the similarities and differences between these curves, we can attempt to identify features that match the expected pattern and develop corresponding filtering algorithms to extract the signal components of interest.

[0151] In the strong target processing flow, the separated strong target signals are first processed by 3DFFT to construct a high-dimensional representation of the target in three dimensions: range, Doppler, and azimuth, thus forming point cloud data.

[0152] After obtaining the point cloud, clustering is performed to group scattering points from the same target together, forming independent point clusters. This step helps distinguish different targets, especially when multiple targets are present in the radar's field of view simultaneously.

[0153] The point cloud dataset that has been obtained is , where each point It contains three coordinate components: ;in, Indicates the first The distance between points; Indicates the first The azimuth angle of each point; Indicates the first The pitch angle of each point;

[0154] The goal of K-means is to divide these points into... Cluster This minimizes the squared error within the cluster. Cluster center The update rules are as follows:

[0155] ;in, express The number of points in the middle, Indicates the first One cluster;

[0156] Then, a Kalman filter is used to process each point cluster, updating the observed target's slant range and azimuth to estimate the system state over time, providing stable tracking performance even in noisy environments; the obtained target state vector is... The observation vector is The state transition matrix is ​​set as follows: The observation matrix is The process noise covariance matrix is The observation noise covariance matrix is The status update rules are as follows:

[0157] ;

[0158] in, This represents the state estimate at time t, taking into account all measurements up to time t. This is the final estimate, i.e., the corrected state estimate; This represents the state prediction at time t, considering only all previous measurements (i.e., measurements up to time t-1). This is a prediction of future states based on the state estimation from the previous step and the system model. This represents the Kalman gain, which determines the degree to which a new measurement affects the state estimate. It is a matrix whose calculation depends on the prediction error covariance matrix and the observation matrix H; This represents the value actually measured at time t.

[0159] Then, the nearest neighbor algorithm is used to associate point clusters between consecutive frames, thereby establishing the target trajectory and achieving target tracking. There is one point cluster in the current frame. We need to find the cluster of points in the previous frame. As its matching object. The matching criterion is usually based on minimizing the distance, that is, finding the nearest cluster of points:

[0160] ;

[0161] in, and These are the coordinate components of the point clusters in the current frame and the previous frame, respectively. This represents the set of point clusters in the previous frame, that is, all target locations detected in the previous frame; Indicates from A specific cluster of points selected from the previous frame, that is, the location of a specific target in the previous frame; This represents an index used to traverse each coordinate component in a point cluster;

[0162] Finally, if a strong target shows a coherent trajectory, indicating that it is moving continuously in space, it can be identified as a human target, and the human presence detection result output by the millimeter-wave radar sensor is marked as "human presence" (output 1); conversely, if there is no trajectory, it can be identified as an interfering target and marked as "human absence" (output 0).

[0163] like Figure 10 The image shown is an example of the trajectory results of a strong target obtained by the millimeter-wave radar sensor in this embodiment.

[0164] In the weak target processing flow, the feature vector of the separated weak target signal is first extracted to obtain the phase value at the distance gate where the target is located.

[0165] Due to the periodicity of the phase value (usually...) The phase extracted directly from the signal may be discontinuous. Therefore, phase unwrapping is required to recover the true phase changes. The resulting original phase sequence... According to the recursive formula:

[0166] ;

[0167] The phase obtained after phase unwinding is It is worth noting that if Then you need to add or subtract. To ensure phase continuity. This represents the phase difference between adjacent time points.

[0168] Next, the untangled phase needs to be determined. Phase difference is performed. Phase difference is used to remove the influence of static background, and is usually achieved by using the phase difference between adjacent sampling points. Phase difference can be expressed as: .

[0169] To further reduce the impact of noise, a moving average filter is typically applied to the phase-differentiated data. The corresponding formula is:

[0170] ;

[0171] in, It is the window length; This represents the phase difference after filtering, i.e., the phase difference value after the moving average filtering process. Indicates a point in time Phase difference at; Indicates a point in time, from arrive There are N time points in total.

[0172] By analyzing the processed phase data, periodic phase changes caused by respiration can be identified. If obvious breathing features are detected in a weak target signal, it indicates that the target has vital signs and can be confirmed as a human target. The final human presence detection result output by the millimeter-wave radar sensor is marked as "human presence" (output 1). Conversely, if no breathing features are detected, it can be identified as an interfering target and marked as "human absence" (output 0).

[0173] like Figure 11 As shown, Figure 11 This is an example diagram showing the breathing results of weak targets obtained by the millimeter-wave radar sensor in this embodiment. Figure 11 The horizontal axis represents time, and the vertical axis represents amplitude. The curve shows a periodic fluctuation pattern, which is typical of the phase changes caused by respiratory movements. Several key characteristics are as follows:

[0174] Periodic fluctuations: The curve exhibits regular up-and-down fluctuations, reflecting the expansion and contraction of the chest or abdomen during respiration. Each rise and fall represents a complete respiratory cycle.

[0175] Amplitude variation: Although the overall trend is periodic, significant variations in amplitude can occur within certain time periods. These variations may be due to different breathing depths or rates.

[0176] Time interval: As can be seen from the figure, the time interval between the two peaks is roughly the same, indicating that the respiratory rate is relatively stable.

[0177] By analyzing such phase data, periodic phase changes caused by respiration can be identified. If obvious breathing features are detected in a weak target signal, it indicates that the target has vital signs and can be confirmed as a human target. The final human presence detection result output by the millimeter-wave radar sensor is marked as "human presence" (output 1). Conversely, if no breathing features are detected, it can be identified as an interfering target and marked as "human absence" (output 0).

[0178] Step 3: Process the signal from the infrared pyroelectric sensor.

[0179] First, the field of view (FOV) of an infrared pyroelectric sensor is the range within which it can effectively detect. To improve detection accuracy and sensitivity, the space of different sensing surfaces within the field of view is usually finely divided into multiple independent monitoring sub-regions.

[0180] like Figure 12 As shown, Figure 12 This is an example diagram of the visible field of view partitioning for an infrared pyroelectric sensor. The region division method is as follows:

[0181] a. For the radial sensing area: the field of view is divided into several concentric annular regions based on the distance to the distance sensor. For example, Figure 12 The E1, E2, and E3 zones on the left are divided according to the different distances of the distance sensor.

[0182] b. For the tangential sensing area: the field of view is divided into several sector-shaped regions according to the angular direction. For example, Figure 12 The A1, A2, and A3 areas on the right are divided according to angle and direction.

[0183] c. Combined partitioning: Radial and tangential partitioning can also be combined to create a more refined gridded area. This method can improve spatial resolution while maintaining a certain coverage area.

[0184] The advantages of dividing the area into regions are as follows:

[0185] a. Improved positioning accuracy: By subdividing a large field of view into multiple smaller regions, the location of the heat source can be located more accurately. This is crucial for applications that require precise knowledge of the heat source's location.

[0186] b. Enhanced responsiveness: Each sub-region has its own sensing element, meaning even small temperature changes can be detected quickly. This is extremely useful for applications requiring high sensitivity.

[0187] c. Reduce false alarm rate: By independently monitoring each sub-area, the system can more effectively distinguish between real targets and background noise or other interference sources, thereby reducing the possibility of false alarms.

[0188] d. Adapt to different scenario requirements: System performance can be optimized by adjusting the method and size of region partitioning according to the needs of different application scenarios.

[0189] This division can be achieved using a Fresnel lens, which divides the light into several regions. A Fresnel lens is a special optical element that can focus light from different directions onto a single point. In infrared pyroelectric sensors, the Fresnel lens concentrates the received infrared radiation energy onto the sensor's sensitive area, thereby improving the sensor's sensitivity and detection range.

[0190] By designing the shape and structure of a Fresnel lens, the field of view can be divided. Specifically, the Fresnel lens can be divided into several regions, each corresponding to a sensing surface of the sensor. In this way, when an object enters a certain region, the corresponding sensing surface will receive more infrared radiation energy, enabling the sensor to monitor different areas within the field of view separately. Each sub-region can be regarded as an independent detection unit, capable of independently sensing temperature changes.

[0191] Then, when a human body or other warm target enters the sensor's field of view and crosses these monitoring sub-regions, a temperature difference is generated. The pyroelectric elements inside the sensor can convert this temperature difference into an electrical signal, typically a voltage change. This conversion is based on the thermoelectric effect, where temperature changes cause changes in the concentration of charge carriers in the material, leading to charge accumulation and voltage output.

[0192] Finally, threshold detection and logical judgment are performed. The generated voltage signal is compared with a preset threshold.

[0193] The preset threshold calculation formula is: ;

[0194] in:

[0195] For the first The received voltage intensity of each unit;

[0196] The number of reference units (excluding the center unit);

[0197] This is a threshold factor, usually set according to the required false alarm probability.

[0198] like Figure 13 The figure shown is an example of the output voltage signal and threshold waveform of an infrared pyroelectric sensor;

[0199] If the voltage signal exceeds this threshold, it means that a significant temperature change has been detected, indicating that there is a moving human target in the detection space, and the final human presence detection result is marked as "human presence" (output 1); conversely, if the voltage does not exceed the threshold, it can be determined that there is no moving human target in the detection space, and it is marked as "human absence" (output 0).

[0200] Step 4: Construct a heterogeneous data matrix based on the output results.

[0201] In multi-sensor fusion technology, combining data from different types of sensors can help improve the accuracy and robustness of the system. Steps 2 and 3 respectively yield the human presence perception results output by the millimeter-wave radar sensor. Human presence perception results output by infrared pyroelectric sensor The result only distinguishes between 0 and 1, where 0 represents no one exists and 1 represents someone exists. That is:

[0202]

[0203]

[0204] Because the two sensors have different sampling frequencies, their timestamps need to be aligned to compare and fuse the data from both sensors at the same time point. If the timestamps of the two sensors do not match perfectly, a linear interpolation method can be used to generate data for the missing time points. After aligning the outputs of the millimeter-wave radar sensor and the infrared pyroelectric sensor through timestamps, merging them into the same matrix yields a two-dimensional heterogeneous data matrix, where each row corresponds to a time point and each column corresponds to a sensor. That is:

[0205]

[0206] Here, M represents the final generated two-dimensional heterogeneous data matrix. Each row corresponds to a time point, and each column corresponds to a sensor. Indicates a point in time At that time, the output value of the millimeter-wave radar sensor; This indicates a point in time. At that time, the output value of the infrared pyroelectric sensor.

[0207] Step 5: Obtain the dual-source fusion result based on the heterogeneous data matrix.

[0208] For the heterogeneous data matrix obtained in step 4, the weighted average method is first used to obtain the fusion result of the millimeter-wave radar sensor and the infrared pyroelectric sensor. ,Right now

[0209] ;

[0210] in, This represents the weighted average fusion result; This indicates the number of data points used to calculate the average. express The detection results of the infrared pyroelectric sensor at that time; express The detection results of the millimeter-wave radar sensor at that moment.

[0211] This represents the weighted average result obtained from millimeter-wave radar sensors; This represents the weighted average result obtained from the infrared pyroelectric sensor.

[0212] Weighted average threshold set Because at this time Therefore, the fusion result at this time ;

[0213] Then, the fusion result of the millimeter-wave radar sensor and the infrared pyroelectric sensor is obtained using the weight allocation method. ,Right now

[0214] ;

[0215] The trust ratios for millimeter-wave radar sensors and infrared pyroelectric sensors are as follows:

[0216] ;

[0217] in, This indicates that the fusion result of the millimeter-wave radar sensor and the infrared pyroelectric sensor was obtained using the weighted average method; This indicates the fusion result of the millimeter-wave radar sensor and the infrared pyroelectric sensor obtained using the weighting method; This indicates the trust weight ratio for millimeter-wave radar sensors; This indicates the trust weight ratio of the infrared pyroelectric sensor;

[0218] The threshold for the weighted allocation method is set to Because at this time Therefore, the fusion result at this time ;

[0219] like Figure 14 As shown, Figure 14 The weights and processing results are obtained from the weighted average method and the weight allocation method. Figure 14 The image illustrates two different weight calculation methods: the weighted average method and the weight allocation method. The two images on the left show the results of the weighted average method: the upper image shows the trend of the weight values ​​of each element in the original input data over time; the lower image shows the final output result based on these weight values. The two images on the right show the results of the weight allocation method: the upper image similarly shows the trend of the weight values ​​of each element in the original input data over time; the lower image also shows the final output result based on these weight values. In summary, this image presents the importance (represented numerically) of each sample point in the entire dataset under each method, and also provides a comparison of the overall performance after processing by the corresponding algorithm.

[0220] Finally, the fusion results were analyzed. and fusion results Perform a logical OR operation to obtain the dual-source fusion result. .Right now:

[0221] ;

[0222] Among them, the logical "OR" operator This means that if any fusion result is 1, the final result is 1; if both are 0, the final result is 0.

[0223] Step 6: Output the timeliness determination results

[0224] The fusion result Stored in chronological order into dimension [dimension name]. The result in the delayed memory matrix is ​​as follows. Figure 15 As shown, Figure 15 This is an example diagram showing the output result of the delayed memory matrix in this embodiment. Figure 15The image shows an example of the output of a time-delayed memory matrix used to store and manage data generated over time. In this example, the rows of the matrix represent different timestamps, and the columns may represent different data types or sources. The intensity of the color in each cell indicates the strength or importance of the corresponding data type at that moment. Specifically, the image shows continuous band patterns, meaning that certain types of data have consistently appeared and remained at a high level over a period of time. On the other hand, there are also blank areas indicating that no relevant data was recorded. This visual presentation makes it easy to identify which time periods saw significant events or anomalies.

[0225] Furthermore, the formula for calculating the time factor corresponding to each delay unit is as follows:

[0226] ;

[0227] in, Indicates the first The timeliness factor of each delay unit is used to measure the importance of the data at that point in time; Indicates the first The time step of each delay unit ranges from 1 to 100; This represents the index variable in the summation, ranging from 1 to 100;

[0228] The specific values ​​are shown in the table below:

[0229] Table 1 Comparison of Delay Units and Time Efficiency Factors

[0230] Delay unit 1 2 ... 50 ... 99 100 Effective factors 0.000198 0.000396 ... 0.0099 ... 0.0196 0.0198

[0231] The timeliness score is obtained by multiplying the fusion result in the delayed memory matrix by the corresponding timeliness factor, i.e.:

[0232] ;

[0233] This indicates the final timeliness, which is a comprehensive evaluation of the fusion result after taking time factors into account; Indicates a point in time The result of fusion at that time;

[0234] Based on experience, the timeliness threshold is set as follows: .

[0235] After calculating the validity period, these elements are judged according to the following judgment rules.

[0236] Rule 1: If timeliness Below the threshold If the number of people is less than 1, it is assumed that no human body exists in the monitoring area.

[0237] Rule 2: If timeliness Above the threshold If so, it is assumed that a human body exists in the current monitoring area.

[0238] According to the rules of judgment, because Therefore, the output result is that a human body exists.

[0239] The output results for the human body in the current monitoring area are as follows: Figure 16 As shown, Figure 16 This is an example diagram of the output result of human presence perception in this embodiment.

[0240] The output results show that the human presence detection method based on the dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor provided in this embodiment can quickly and effectively detect the presence of human bodies in the monitoring area, verifying the correctness and effectiveness of this embodiment.

[0241] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor, characterized in that: Includes the following steps: S1. Obtain the raw data collected by the millimeter-wave radar sensor receiver and process it to obtain a data matrix in the range-velocity dimension; S2. Perform SVD decomposition on the range-velocity spectrum data matrix, and reconstruct two new data matrices by comparing singular values ​​with thresholds and combining eigenvalues ​​and eigenvectors, which correspond to strong target signals and weak target signals, respectively. S3. Process the strong target signal and the weak target signal respectively, and obtain the human presence perception output of the millimeter-wave radar sensor through trajectory features and breathing features; S4. Acquire the signal from the infrared pyroelectric sensor and process it to obtain a voltage signal. Compare the voltage signal with a preset threshold to obtain the human presence sensing output result of the infrared pyroelectric sensor. S5. Construct a heterogeneous data matrix based on the sensing output results obtained from the millimeter-wave radar sensor and the infrared pyroelectric sensor respectively, and process the heterogeneous data matrix through a fusion algorithm to obtain the dual-source fusion result. S6. Store the fusion results into the delayed memory matrix in chronological order, calculate the timeliness according to the timeliness factor corresponding to different delay units, and thus obtain the final human existence perception result. The heterogeneous data matrix in step S5 is constructed in the following manner: Acquire human presence perception results from millimeter-wave radar sensors Human presence perception results output by infrared pyroelectric sensor ; The sampling timestamp is determined based on the sampling frequency of the two sensors, and the outputs of the millimeter-wave radar sensor and the infrared pyroelectric sensor are aligned using the timestamp. The aligned results are merged into a single matrix to obtain a two-dimensional heterogeneous data matrix, where each row corresponds to a time point and each column corresponds to a sensor. The formula for calculating the time factor for each delay unit is: ； in, Indicates the first The timeliness factor of each delay unit is used to measure the importance of the data at that point in time; Indicates the first The time step of each delay unit ranges from 1 to 100; This represents the index variable in the summation, ranging from 1 to 100; The timeliness score is obtained by multiplying the fusion result in the delayed memory matrix by the corresponding timeliness factor, i.e.: ； This indicates the final timeliness, which is a comprehensive evaluation of the fusion result after taking time factors into account; Indicates a point in time The result of fusion at that time; The timeliness is compared with the threshold to obtain the final result of human presence perception.

2. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: The raw data in step S1 is processed according to the following steps: S101. A one-dimensional fast Fourier transform is used to process the time-domain signal of each frame, converting the signal from the time domain to the frequency domain. S102. Noise removal is performed using a dynamic filtering method; S103. Perform incoherent superposition to accumulate the target's echo energy; S104. The signal is extended from the one-dimensional frequency domain to the two-dimensional frequency domain using a two-dimensional fast Fourier transform, and a range-Doppler domain data matrix containing target distance and radial velocity information is constructed. S105. Perform zero-velocity target elimination, that is, eliminate objects with no radial velocity.

3. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: The strong target signal in step S2 is processed according to the following steps: S301. Perform 3DFFT processing on the separated strong target signal to construct a high-dimensional representation of the target in three dimensions: range, Doppler, and azimuth, and form point cloud data. S302. After obtaining the point cloud data, perform clustering operations to group scattering points from the same target together to form independent point clusters; S303. Use a Kalman filter to process each point cluster and update the slant range and azimuth of the observed target. S304. The nearest neighbor rule is used to associate point clusters between consecutive frames, thereby establishing the target trajectory and realizing target tracking; S305. Obtain a strong target displaying a continuous trajectory. Determine whether it is a human target based on the continuous trajectory. If it is, mark the detection result as "human presence" and output 1. Conversely, if not, it will be marked as an interference target, marked as "human body does not exist", and output 0.

4. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: The weak target signal in step S3 is processed according to the following steps: S306. Extract feature vectors from the separated weak target signals to obtain phase data at the range gate where the target is located; S307, extract phase unwinding, phase difference, smoothing filtering, and breathing value respectively; S308. By analyzing the processed phase data, identify the periodic phase changes caused by respiration, and make judgments as follows: If respiratory features are detected in a weak target signal, it indicates that the target has vital signs. Therefore, it is confirmed as a human target, and the final human presence detection result is marked as "human presence" and output 1. Conversely, if no respiratory features are detected, the target is identified as an interference target and marked as "human body does not exist", and 0 is output.

5. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: In step S4, the signal from the infrared pyroelectric sensor is processed, specifically as follows: S401. Divide the space within the field of view of the infrared pyroelectric sensor into multiple independent monitoring sub-regions; S402. When a human body or other warm target enters the sensor's field of view and crosses these monitoring sub-areas, the pyroelectric element inside the sensor acquires the temperature difference signal and converts the temperature difference signal into an electrical signal. S403. Perform threshold detection and logic judgment. Compare the generated voltage signal with the preset threshold. If the voltage signal exceeds the preset threshold, it indicates that there is a moving human target in the detection space. Mark the human presence detection result as "human presence" and output 1. Conversely, if the voltage does not exceed the preset threshold, it is determined that there is no moving human target in the detection space and is marked as "human body does not exist", and 0 is output.

6. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: The strong target signal and the weak target signal in step S2 are separated in the following manner: The range-velocity spectrum data matrix is ​​decomposed using SVD. The mathematical form of SVD is: ； in: It is An orthogonal matrix whose column vectors are the left singular vectors of ; It is A diagonal matrix, where the elements on the diagonal are the singular values ​​of , usually arranged in descending order; It is An orthogonal matrix whose column vectors are the right singular vectors of ; By comparing singular values ​​with a threshold, two data matrices corresponding to strong and weak targets are obtained, respectively.

7. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 5, characterized in that: The step S401, which divides the space within the field of view of the infrared pyroelectric sensor into multiple independent monitoring sub-regions, is performed in the following manner: By dividing the Fresnel lens into several different regions to monitor different areas within the field of view, each sub-region is regarded as an independent detection unit that can independently sense temperature changes; When a human body or other warm target enters the sensor's field of view and crosses these monitoring sub-regions, a temperature difference is generated. The pyroelectric element inside the sensor can convert this temperature difference into an electrical signal.

8. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: The preset threshold in step S4 is calculated according to the following formula: ； in: For the first The received voltage intensity of each unit; The number of reference units; This is a threshold factor, usually set according to the required false alarm probability.

9. The human presence sensing method based on dual-source collaborative fusion of millimeter-wave radar and infrared pyroelectric sensor as described in claim 1, characterized in that: In step S5, obtaining the dual-source fusion result based on the heterogeneous data matrix is ​​performed in the following manner: The fusion results of the millimeter-wave radar sensor and the infrared pyroelectric sensor were obtained by using a weighted average method on the heterogeneous data matrix. ,Right now ； in, This represents the weighted average fusion result; This indicates the number of data points used to calculate the average. express The detection results of the infrared pyroelectric sensor at that time; express The detection results of the millimeter-wave radar sensor at that moment; This represents the weighted average result obtained from millimeter-wave radar sensors; This represents the weighted average result obtained from the infrared pyroelectric sensor; The fusion result is obtained based on the weighted average threshold method. ; Then, the fusion result of the millimeter-wave radar sensor and the infrared pyroelectric sensor is obtained using the weight allocation method. ,Right now ； in, and These represent the trust ratios of the millimeter-wave radar sensor and the infrared pyroelectric sensor, respectively. ； The fusion result is obtained based on the weight allocation method threshold. ; Finally, the fusion results were analyzed. and fusion results Perform a logical OR operation to obtain the dual-source fusion result. ,Right now: ； Among them, the logical "OR" operator This means that if any fusion result is 1, the final result is 1; if both are 0, the final result is 0.

Images