Human body sensing method and system

Abstract

The invention provides a human body sensing method and system. The method comprises the following steps: S1, acquiring echo data; s2, processing the echo data to obtain a one-dimensional Fourier transform result and a one-dimensional Fourier transform mean value; s3, performing fast Fourier transform and windowing on the one-dimensional Fourier transform result to obtain a two-dimensional fast Fourier transform result, and detecting the two-dimensional fast Fourier transform result by adopting an SO-CFAR algorithm to generate a fast frame point cloud data set; s4, performing fast Fourier transform and windowing on the one-dimensional Fourier transform mean value to obtain a two-dimensional fast Fourier transform result, and detecting the two-dimensional fast Fourier transform result by adopting an SO-CFAR algorithm to generate a slow frame point cloud data set; s5, clustering and filtering the fast frame point cloud data set and the slow frame point cloud data set to obtain target information; s6, obtaining target keeping and physiological target position information; and S7, judging whether a person exists or not. The cooperation of the fast time frame and the slow time frame realizes complementation from two levels of motion and micro motion, and the accuracy of human body sensing and recognition is improved.

Description

Technical Field

[0001] The present application relates to the field of millimeter wave radar application technology, and more specifically, to a human body perception method and system. Background Art

[0002] In the field of smart home, human sensing technology has become one of the important directions for the intelligent upgrade of home appliances. The advancement of science and technology has led to the emergence of smart air conditioners, especially the ability to automatically turn on and off air conditioners with or without people, which can make people's lives more convenient. As the greenhouse effect becomes increasingly serious, this technology can effectively reduce carbon emissions. Automatically turning on and off air conditioners means that users do not need to worry about forgetting to turn off the air conditioner, which has a good market prospect.

[0003] Infrared sensors work by detecting infrared rays of about 10um emitted by the human body. Affected by interference from various heat sources, especially when the room temperature is close to the human body temperature, it is difficult to accurately distinguish the human body from the background temperature. The detection accuracy of the infrared sensor will drop significantly, and it cannot accurately distinguish the human body from other moving objects, resulting in a high false alarm rate. At the same time, the infrared emitted by the human body is easily blocked, resulting in missed alarms, and the working performance stability is poor. The visual recognition solution based on RGB camera for human perception through image processing is highly dependent on light. Visual recognition involves user privacy and security issues. Especially in a home environment, the use of such a camera often makes users feel uneasy and affects their acceptance. Microwave sensors can better protect user privacy, but the accuracy of human perception is low.

[0004] In summary, the perception accuracy of existing human body perception methods is low. Summary of the invention

[0005] The purpose of the present invention is to provide a human body perception method and system to address the deficiencies in the prior art, so as to solve the problem of low perception accuracy of the human body perception method in the prior art.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows: The present application provides a human body sensing method, the method comprising the steps of: S1, obtain echo data; S2, processing the echo data to obtain a one-dimensional Fourier transform result and a one-dimensional Fourier transform mean; S3, performing fast Fourier transform and windowing on the one-dimensional Fourier transform result to obtain a two-dimensional fast Fourier transform result, and using the SO-CFAR algorithm to detect it to generate a fast frame point cloud data set; S4, performing fast Fourier transform and windowing on the one-dimensional Fourier transform mean to obtain a two-dimensional fast Fourier transform result, and using the SO-CFAR algorithm to detect it to generate a slow-frame point cloud data set; S5, clustering and filtering the fast frame point cloud data set and the slow frame point cloud data set to obtain target information; S6, target retention and acquisition of physiological target position information; S7, determination of unmanned status.

[0007] Furthermore, in step S1, first, the ADC basic voltage is self-calibrated, the MCU controls the RF chip to transmit a linear frequency modulated continuous wave through the SPI interface protocol, and the ADC samples the echo signal.

[0008] Furthermore, step S2 includes performing non-coherent accumulation on multiple echo signals to obtain time domain signal data, subtracting the mean of the previous frame of sampling data from the time domain signal data, and then performing one-dimensional fast Fourier transform and windowing on the distance dimension to obtain a one-dimensional Fourier transform result, taking the average of the one-dimensional Fourier transform results to obtain a one-dimensional Fourier transform mean.

[0009] Furthermore, in step S3, the CFAR detector divides the window into a reference unit, a protection unit, and a detection unit. When the value of the detection unit is greater than the detection threshold S, the data is saved to form a fast frame point cloud data set.

[0010] Furthermore, the detection threshold S = kZ so , where k is the threshold factor, Z so =min(X, Y), where X and Y are the average values ​​of the reference cells on both sides of the detection cell, respectively. , Y= , where N represents the number of elements in the single-sided reference unit and i represents the i-th element.

[0011] Furthermore, in step S4, the one-dimensional Fourier transform mean is a data set of 40*1, and the one-dimensional Fourier transform means of the first 32 frames are filled in sequence to form a 40*32 data set, and fast Fourier transform and windowing are performed in the Doppler velocity dimension to obtain a two-dimensional fast Fourier transform result, and the two-dimensional fast Fourier transform result is detected to obtain a slow frame point cloud data set.

[0012] Furthermore, in step S5, the DBSCAN algorithm is used to cluster the fast frame point cloud data set and the slow frame point cloud data set respectively, and the clustering results are filtered using a mean filtering method.

[0013] Furthermore, in step S6, if the difference between the target information stored after mean filtering and the previous target information is less than 3 in distance units and less than 5 in speed units, the holding point of the previous target information is maintained; otherwise, a new holding point is established.

[0014] Furthermore, in step S7, when the moving target, the micro-moving target, and the physiological target all disappear, it is determined that there is no one within the detection range; otherwise, it is determined that there is a person within the detection range.

[0015] The present application also proposes a human perception system, which includes an accelerator and a processor. The accelerator is an integrated radar baseband processor algorithm accelerator including a data transfer module, a signal processing module, and a data processing module; the processor is a 32-bit CPU processor including a clustering module, a mean filtering module, a target holding module, a physiological signal detection module, and a target information output module; the system executes the above method.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The method of this application uses millimeter wave radar for detection, and is divided into fast time frame dimension and slow time frame dimension in signal processing. The fast time frame dimension focuses on macroscopic motion, captures significant motion information, and helps identify the existence and approximate position of the target; the slow time frame dimension focuses on tiny, physiological motion signals, and is used to extract more delicate micro-motion information, which helps to perceive people's subtle movements or physiological signals. The combination of fast time frame and slow time frame can achieve complementarity from the two levels of motion and micro-motion, which not only enriches the level of recognition information, but also enhances the system's perception of complex scenes and tiny physiological signals. Physiological signals such as heartbeat and breathing can enhance the stability of target recognition, help distinguish the human body from other moving objects, and avoid false alarms, thereby significantly improving the accuracy of human perception and recognition.

[0017] Improve signal-to-noise ratio and enhance anti-interference ability: The high sensitivity of the fast time frame can quickly filter out the main moving targets, while the slow time frame helps to detect tiny physiological signals such as heartbeat and breathing in the case of high noise; even in complex environments or noise interference, the system can still maintain a high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a human body sensing method provided by the present invention; Figure 2 A schematic diagram of the working state of the detector in step S3 of the human body sensing method provided by the present invention; Figure 3 A schematic diagram of the installation of a human body sensing system provided by the present invention; Figure 4 A schematic diagram of a detection area of ​​a human body perception system provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the implementation process of the present invention clearer, it will be described in detail below with reference to the accompanying drawings.

[0020] The present invention provides a human body sensing method, such as Figure 1 As shown, the method comprises the following steps: S1, obtaining echo data; The millimeter-wave radar chip is placed horizontally at the air outlet of the air conditioner and about 30 cm away from the air conditioner, and communicates with the air conditioner through a wiring harness; the radar board is fixed in a plastic casing, and the plastic casing is fixed to the wall and can be fixed with 3M adhesive. A microstrip patch antenna is used, and the transmitting antenna and receiving antenna are printed on the circuit board for transmitting and receiving echo signals. The millimeter-wave radar chip integrates an on-chip programmable FMCW frequency modulation controller, which can generate a variety of complex frequency-modulated continuous wave signal outputs; it integrates a radar baseband processor algorithm accelerator and a 32-bit CPU processor, and uses a 1-transmit 1-receive microstrip antenna array. The frequency sweep start frequency is 24.01 GHz, the frame period is 100ms, and the velocity resolution expression is △V =λ / 2N*T c , where λ is the wavelength, N is the number of FFT points, and T c is the sending time interval, and the velocity resolution is 0.315m / s. The distance accuracy is expressed as , where μ is the radar frequency modulation slope, c is the speed of light, and F s is the AD sampling rate, N fft is the number of points in FFT; the distance accuracy is about 22cm, that is, one distance unit of the radar is about 22cm. The expression of distance resolution is △r= C / 2B, where C is the speed of light and B is the spectrum bandwidth; the distance resolution is about 88cm, that is, the two targets can only be distinguished if the distance between the front and the back targets is greater than 88cm. The 24.0GHz to 24.25GHz frequency band is a universal frequency band open to the world and is a compliant frequency band. The frequency sweep starts at 24.01Ghz, leaving a certain space to prevent frequency leakage. Beyond 24.25Ghz, the distance and speed resolution are calculated based on the above parameters. If the frame period is too short, there will be no obvious difference from the previous frame, resulting in the inability to detect the target. If the frame period is too long, the data will not be updated in time.

[0021] After the millimeter wave radar is powered on, it first performs self-calibration of the ADC basic voltage. After the calibration is completed, the MCU controls the RF chip to transmit linear frequency modulation continuous waves through the SPI interface protocol, and when the time of transmitting the linear frequency modulation continuous wave and the electromagnetic wave reflected back after detecting the target object meet the preset conditions, the ADC is triggered to collect the echo data of the indoor target at this time to obtain the echo data. That is, the MCU controls the RF chip to send the linear frequency modulation continuous wave, and samples the echo signal through the ADC; after the linear frequency modulation continuous wave is transmitted, the radar receives the electromagnetic wave reflected back by the target object. The ADC collects the echo data. To store the echo data, the data collected by the ADC can be transferred to the specified memory space for storage through the data transfer module by using DMA (Direct Memory Access) transmission technology. S2. Process the echo data to obtain a one-dimensional Fourier transform result and a one-dimensional Fourier transform mean; The echo data obtained in step S1 is processed, and the intermediate frequency signal collected by ADC is used as the original data, that is, the signal formed by the transmitting signal and the receiving signal through the mixer. Multiple Chirp signals are accumulated non-coherently. The echo energy of a single radar pulse is limited, that is, the echo signals of multiple consecutive pulse repetition periods and uniform distance units in a wave position are superimposed to achieve energy accumulation. The non-coherent accumulation takes the amplitude of the signal into consideration and accumulates multi-cycle echoes. The accumulated target signal will be stronger, and the target can be detected more accurately. The time domain signal data after accumulation is obtained, and the mean value of the previous frame sampling data is subtracted from the signal. The current frame data minus the previous frame data can effectively remove noise. The collected data contains the echo reflection noise of objects in the room, and the static echo reflection objects. , its position and echo reflection power change is relatively small, so the stationary echo reflection object signal can be filtered out; when detecting the movement of the target, there is a certain difference between the previous frame and the current frame data. The previous frame data is averaged, and the current frame data is subtracted from the previous frame data, which can remove part of the noise of the current frame, highlight the target signal, facilitate data processing, and be more accurate; then perform one-dimensional fast Fourier transform and windowing on the distance dimension to obtain the one-dimensional Fourier transform result. Using the Fourier transform theory, the time domain signal is converted to the frequency domain signal for processing. The target is expressed in the frequency domain as a higher amplitude at its location, which can more conveniently and quickly obtain the target location. The window function is the Hanning window. Since the fast Fourier transform sometimes produces spectrum leakage, the windowing function can change the width and amplitude of the main lobe and side lobe of the spectrum to make the energy more concentrated, thereby reducing spectrum leakage. Take the average of the one-dimensional Fourier transform results to obtain the one-dimensional Fourier transform mean. The processed data is stored in the specified address space.

[0022] S3, performing fast Fourier transform and windowing on the one-dimensional Fourier transform result to obtain a two-dimensional fast Fourier transform result, using the SO-CFAR algorithm to detect it and generate a fast frame point cloud data set; Perform fast Fourier transform and windowing on the one-dimensional Fourier transform result in the Doppler velocity dimension. The window function is a Chebyshev window to obtain a two-dimensional fast Fourier transform result, which is stored in the specified address space. Perform a two-dimensional Fourier transform on the one-dimensional Fourier transform result to obtain a two-dimensional distance-Doppler matrix, and use the SO-CFAR algorithm for constant false alarm detection. Figure 2 As shown, the CFAR detector divides the window into a reference unit, a protection unit P, and a detection unit D; wherein the reference unit is used to estimate the background clutter power value, and the protection unit is used to prevent the energy of the detection unit from leaking to the reference unit, that is, to prevent the units near it from affecting it, and the number of protection units is an integer. The protection units adjacent to the left and right of the detection unit are not included in the reference unit. On the left and right of the detection unit, N reference units are taken respectively, and the average values ​​of the N units are calculated to obtain X and Y, and the smaller average value of X and Y is selected as the background clutter power estimation value Z of the detection unit D. so =min(X,Y); where X= , Y= , detection threshold S = kZ so , where k is the threshold factor, N represents the number of elements in the single-sided reference unit, and i represents the i-th element. When the value of the detection unit is greater than S, the value of the unit is stored. The point cloud data set of the moving target is obtained, and a fast frame point cloud data set with distance, speed and energy information is generated. FFT is performed in the distance dimension and speed dimension respectively to generate the distance-Doppler two-dimensional matrix data. The SO-CFAR algorithm is used to detect each unit. If it exceeds the set threshold, the row, column and amplitude of this unit are output. For example, the output of x row and y column indicates that the unit is in the xth distance unit and the yth speed unit, and the amplitude is power. The fast time dimension refers to collecting data from the ADC every 100ms (each frame), subtracting the time domain mean, performing FFT in the distance dimension and Doppler speed dimension respectively, and then performing the SO-CFAR algorithm. In the slow time dimension, the one-dimensional FFT mean of each frame of data is used to accumulate and store 32 frames of data (equivalent to 3.2s of data collected in the slow time dimension), and then a Doppler velocity dimension FFT is performed. That is, after the system has run a loop for 32 frames, the first 32 frames of data are superimposed in sequence, and then a velocity dimension FFT is performed.

[0023] In the fast time dimension, the system samples once every 100ms, and through FFT operations in the distance dimension and Doppler velocity dimension, the distance and speed information of the moving target can be obtained in time. In this way, large-scale moving targets can be detected in real time, and the movement status of the human body in different positions can be quickly identified. The adaptive constant false alarm rate of the SO-CFAR algorithm further enhances the target detection capability of the fast time dimension. Especially in dynamic environments, CFAR can automatically adapt to background noise and environmental interference, detect target information generated by human motion, and thus generate fast-frame point clouds. The point cloud information is clustered, mean filtered, and target maintained, and the moving target can be accurately identified. The slow time dimension accumulates 32 frames (about 3.2 seconds of data) of one-dimensional FFT mean, and then performs speed dimension FFT. This can effectively capture the micro-motion characteristics of the human body, such as low-speed signals such as breathing and heartbeat. Since the frequency of physiological signals is usually low, the resolution and signal-to-noise ratio of micro-motion signals can be enhanced through 3.2 seconds of accumulation, further improving the stability and accuracy of heartbeat and breathing detection. This accumulation helps to reduce the impact of short-term noise and make low-frequency micro-motion signals more prominent.

[0024] S4, performing fast Fourier transform and windowing on the one-dimensional Fourier transform mean to obtain a two-dimensional fast Fourier transform result, using the SO-CFAR algorithm to detect it and generate a slow-frame point cloud data set; Based on the one-dimensional Fourier transform mean, which is a 40*1 data set, the one-dimensional Fourier transform mean of the first 32 frames is filled in order to form a 40*32 data set, and fast Fourier transform and windowing are performed in the Doppler velocity dimension; due to the increase in signal acquisition time, when the target is quiet, there will be some micro-movements in its body, making the amplitude at the target more prominent than the noise. The accumulated data is then passed through the SO-CFAR algorithm. When the amplitude of the detection unit is greater than a certain threshold, the value of the detection unit is output. Fast Fourier transform and windowing are performed in the Doppler velocity dimension to obtain a two-dimensional fast Fourier transform result, which is stored in the specified address space. The SO-CFAR algorithm with different parameters is used to detect the two-dimensional fast Fourier transform results to obtain a slow frame point cloud data set.

[0025] The SO-CFAR algorithm parameters are set to low, specifically, for the distance dimension search unit is 4, the protection unit is 10, and the threshold is 8, for the speed dimension search unit is 4, the protection unit is 10, and the threshold is 2, and the first slow-frame point cloud data set with distance, speed and energy information is generated. The SO-CFAR algorithm threshold parameter is set to high, specifically, for the distance dimension search unit is 6, the protection unit is 10, and the threshold is 8, for the speed dimension search unit is 6, the protection unit is 10, and the threshold is 2, and the second slow-frame point cloud data set with distance, speed and energy information is generated. Due to the different characteristics of the signal, the feature point clouds filtered out by different parameters represent different information. When the parameter is high, the filtered feature point cloud represents that the human body has micro-motion behavior (such as waving, kicking), and when the parameter is low, the filtered feature point cloud represents that the physiological human body only has breathing behavior. That is, in the slow time frame dimension, the one-dimensional Fourier transform mean is used as the original signal. After accumulating multiple frames of data, a two-dimensional Fourier transform is performed to obtain a distance-Doppler two-dimensional matrix. The SO-CFAR algorithm with different parameters is used to obtain a micro-motion target point cloud data set and a physiological information point cloud data set. The micro-motion target point cloud data set corresponds to the second slow frame point cloud data set, and the physiological information point cloud data set corresponds to the first slow frame point cloud data set. Fast and slow frames are distinguished by the acquisition signal duration. The slow frame accumulates the mean data of the first 32 frames. After 32 frames of accumulation, the target signal amplitude is characterized in the data set as being significantly higher than the surrounding units. Signals that are more complex to analyze in the time domain are easier to analyze in the frequency domain after Fourier transform.

[0026] S5, clustering and filtering the data set to obtain target information; First, the abnormal and biased Doppler velocity point clouds in the slow-frame point cloud data set and the fast-frame point cloud data set are deleted from the set to filter out the abnormal points; specifically, the point cloud information with Doppler velocity greater than 5 units and less than 27 units is deleted from the set, and the other information is retained. The velocity resolution is 0.315m / s; specifically, Doppler velocity units 0-31 are positive and negative, 1-15, for positive velocity, 16-31 for negative velocity, that is, 31 is -1, 30 is -2, 27 is -5, and the speed units with an absolute value of less than or equal to 5 are screened. It is relatively rare to see fast-moving targets in the home. This type of target may be generated by devices such as fans. Deleting this type of point cloud information can effectively filter out interference.

[0027] Then, the DBSCAN algorithm is used to cluster the slow frame point cloud data set and the fast frame point cloud data set respectively to obtain the clustering results. For the fast frame point cloud data set, a sample is selected as the core, and the point clouds with adjacent distance units within 5, speed units within 5, and more than 2 are identified as one class. For the micro-motion target point cloud data set, the point clouds with adjacent distance units within 5, speed units within 5, and more than 1 are identified as one class. The core unit data of each class is stored. By dividing the data into different classes, the complexity and noise of the data can be reduced, key features can be extracted, and subsequent data analysis tasks can be simplified. The DBSCAN algorithm is a density-based spatial clustering algorithm that divides areas with sufficient density into clusters and finds clusters of arbitrary shapes in a spatial database with noise. It defines a cluster as the maximum set of density-connected points. The concept of density-based clustering is used, that is, the number of objects (points or other spatial objects) contained in a certain area in the clustering space is required to be not less than a given threshold. The clustering speed is fast and can effectively process noise points and find spatial clusters of arbitrary shapes, and the clustering effect is good.

[0028] The clustering results are filtered by the mean filtering method. The mean filtering algorithm is based on the sliding window filtering method. The value of the center element of the window is replaced by the average value of the elements in the window. The size and shape of the filter window can be adjusted according to the specific task, which can retain the overall trend of the signal and has high algorithm efficiency. A core point is selected and its neighborhood is searched. The elements with less than 5 distance units and 10 speed units are considered as its neighborhood elements. The average value of no more than 5 elements is selected as the replacement value of the core point. The valid number of times within the range of each statistical point is calculated. If the conditions are met, the target information is stored, otherwise it is not stored. There must be more than three point clouds within the range of each statistical point, otherwise the statistical point is invalid. The target information includes distance, speed and power value, which are used as input values ​​for subsequent processing.

[0029] S6, target retention and acquisition of physiological target position information; In the target holding stage, if the target information stored after mean filtering is within the close range of the previous output value, the previous target is held; if the target information stored after mean filtering is close to the previous frame data, it is held, otherwise a new holding point is established. Specifically, if the difference between the two target information is less than 3 in distance units and less than 5 in speed units, it is held, otherwise a new holding point is established. If the distance is far, a new holding point is established; even after filtering and other data processing, the target information measured by the radar changes within a certain range when the target is stationary, and this change is relatively small. Using this method, this micro-change jump can be eliminated. When the target starts to move, its change exceeds this threshold, and a new point is established to track the target. The stability and accuracy of target detection are increased.

[0030] The target information stored in the fast frame point cloud data set is calculated to update the "motion / micro-motion hold points" within the threshold range or to establish new "motion hold points" and mark them as "motion hold points". The target information stored in the slow frame point cloud data set is calculated to update the "motion hold points" within the range threshold or to update the "micro-motion hold points" within the range threshold and mark them as "micro-motion hold points" for the second slow frame point cloud. Due to the limitation of computing resources, the total number of hold points is limited to 10.

[0031] When only the "micro-motion hold point" exists, the first slow-frame point cloud data set is used as the original data, and the point cloud is filtered according to the point cloud characteristics of human physiological information to obtain a point cloud data set that meets the physiological characteristics; the elements in the filtered data set need to meet the following conditions: 3<Target[i].d1dx<26 Present. d1dx-4<Target[i].d1dx<Present. d1dx+4 Target[i].d2dx<4orTarget[i].d2dx>28 Target[i].pow>1500 Among them, Present. d1dx is the distance of the static holding point, Target[i].d1dx is the distance of the i-th element, Target[i].d2dx is the speed, and Target[i].pow is the power value. The physiological target position information is calculated by weighting the point cloud data set that meets the physiological characteristics; specifically, assuming that N points are selected from the target point cloud, the average distance of the N points is taken, and the expression is: TargetMean.d1dx=(Target[0].d1dx+ Target[1].d1dx +…+ Target[N-1].d1dx) / N , where Target[i].d1dx is the distance of the i-th element.

[0032] S7. Determination of whether there is an unmanned state.

[0033] When the motion holding point, micro-motion holding point and physiological target position information all disappear, it is determined that there is no one in the current detection range, and the air conditioner can be shut down and other related operations based on the result. Otherwise, it is determined that there is someone. In general, there are six personnel states. When there is a motion target, micro-motion target detection is performed, and physiological target detection is not performed; when there is no motion target, micro-motion target detection and physiological target detection are performed. When all three targets disappear, it is determined that there is no one in the current detection area, as shown in the following table. The method of the present invention can reduce costs and improve the range of human perception and detection stability on the basis of protecting user privacy.

[0034] The present application also provides a human perception system, which includes an integrated radar baseband processor algorithm accelerator and a 32-bit CPU processor. The integrated radar baseband processor algorithm accelerator includes a data transfer module, a signal processing module, and a data processing module. The human perception algorithm part is implemented based on the 32-bit CPU processor, including a clustering module, a mean filter module, a target retention module, a physiological signal detection module, and a target information output module. A 1-transmit 1-receive microstrip antenna array is used, the sweep start frequency is 24.01GHz, the sweep bandwidth is 220MHz, the distance resolution is 22cm, the speed resolution is 0.315m / s, and the frame period is 100ms. The millimeter-wave radar chip integrates an on-chip programmable FMCW frequency modulation controller, which can generate a variety of complex frequency-modulated continuous wave signal outputs. The millimeter-wave radar is placed in the horizontal direction of the air outlet of the air conditioner and about 30cm away from the air conditioner, and communicates with the air conditioner through a wiring harness; the radar board is fixed in a plastic shell, and the shell is fixed to the wall with 3M adhesive. Figure 3 and Figure 4 It is a schematic diagram of installation, the installation height is 2.7m, the angle between the radar antenna surface and the vertical plane is 50 degrees, and the antenna surface is tilted downward, so that the signal radiation range is large; the detection area is mapped to the ground plane, the azimuth angle range can reach ±80°, the micro-motion detection area is 4.5m, and the motion detection area is 7m, with good accuracy and stability. The beam distribution emitted by the microwave antenna is similar to an ellipse, the plane energy at its radio frequency normal is the most concentrated, the distribution range is wide, the detection distance is the farthest, and the effective distance can reach 7m; in the antenna design, by compressing the detection angle in the pitch direction and increasing the detection angle in the azimuth direction, the normal plane beam emitted by the radio frequency can be distributed in the azimuth of ±80. When the present invention is implemented, the radar is installed at a high place and the antenna surface is tilted downward, giving full play to the detection distance and angle performance of the antenna in the azimuth direction, the MCU can collect enough target signals, and the detection range in the azimuth direction can be increased from the traditional ±60° to ±80° through the cooperation of the algorithm for motion, micro-motion and physiological signal detection, which greatly improves the accuracy and stability of the detection.

[0035] The present invention proposes a human body perception method and device for intelligent air conditioning based on millimeter wave radar. Under the specified installation angle and height, human body perception can be performed at an azimuth angle of ±80°, the motion detection range can reach 7m, the static detection range can reach 4.5m, and it has high stability and accuracy.

[0036] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A human body sensing method, characterized in that: The method comprises the steps of: S1, obtain echo data; S2, processing the echo data to obtain a one-dimensional Fourier transform result and a one-dimensional Fourier transform mean; S3, performing fast Fourier transform and windowing on the one-dimensional Fourier transform result to obtain a two-dimensional fast Fourier transform result, and using the SO-CFAR algorithm to detect it to generate a fast frame point cloud data set; S4, performing fast Fourier transform and windowing on the one-dimensional Fourier transform mean to obtain a two-dimensional fast Fourier transform result, and using the SO-CFAR algorithm to detect it to generate a slow-frame point cloud data set; S5, clustering and filtering the fast frame point cloud data set and the slow frame point cloud data set to obtain target information; S6, target retention and acquisition of physiological target position information; S7, determination of unmanned status.

2. The human body sensing method according to claim 1, characterized in that: In the step S1, first, the ADC basic voltage is self-calibrated, the MCU controls the radio frequency chip to transmit a linear frequency modulated continuous wave through the SPI interface protocol, and the ADC samples the echo signal.

3. The human body sensing method according to claim 2, characterized in that: The step S2 includes performing non-coherent accumulation on multiple echo signals to obtain time domain signal data, subtracting the mean of the sampling data of the previous frame from the time domain signal data, performing one-dimensional fast Fourier transform and windowing on the distance dimension to obtain the one-dimensional Fourier transform result, and taking the mean of the one-dimensional Fourier transform results to obtain the one-dimensional Fourier transform mean.

4. The human body sensing method according to claim 3, characterized in that: In step S3, the CFAR detector divides the window into a reference unit, a protection unit, and a detection unit. When the value of the detection unit is greater than the detection threshold S, the data is saved to form a fast frame point cloud data set.

5. The human body sensing method according to claim 4, characterized in that: The detection threshold S=kZ so , where k is the threshold factor, Z so =min(X, Y), X and Y are the average values ​​of the reference cells on both sides of the detection cell, respectively, X= , Y= , where N represents the number of elements in the reference unit on one side, and i represents the i-th element.

6. The human body sensing method according to claim 5, characterized in that: In step S4, the one-dimensional Fourier transform mean is a data set of 40*1, the one-dimensional Fourier transform means of the first 32 frames are filled in sequence to form a 40*32 data set, and fast Fourier transform and windowing are performed in the Doppler velocity dimension to obtain a two-dimensional fast Fourier transform result, and the two-dimensional fast Fourier transform result is detected to obtain a slow frame point cloud data set.

7. The human body sensing method according to claim 6, characterized in that: In the step S5, the DBSCAN algorithm is used to cluster the fast frame point cloud data set and the slow frame point cloud data set respectively, and the clustering results are filtered by a mean filtering method.

8. The human body sensing method according to claim 7, characterized in that: In step S6, if the difference between the target information stored after mean filtering and the previous target information is less than 3 in distance units and less than 5 in speed units, the holding point of the previous target information is maintained; otherwise, a new holding point is established.

9. The human body sensing method according to claim 8, characterized in that: In step S7, when the moving target, the micro-moving target, and the physiological target all disappear, it is determined that there is no one within the detection range; otherwise, it is determined that there is a person within the detection range.

10. A human body sensing system, characterized in that: The system includes an accelerator and a processor. The accelerator is an integrated radar baseband processor algorithm accelerator, including a data transfer module, a signal processing module, and a data processing module; the processor is a 32-bit CPU processor including a clustering module, a mean filtering module, a target holding module, a physiological signal detection module, and a target information output module; the system executes any method described in claims 1-9.

Images