A Multi-Target Tracking and Respiration Detection Method and Device Based on Millimeter-Wave Radar

Abstract

The present invention discloses a multi-target tracking and breathing detection method and device based on millimeter-wave radar. The method includes the following steps: obtaining distance, speed, angle, and phase information from the sampled signal and completing the detection of target points; clustering the point cloud using the proposed improved fast DBSCAN algorithm; tracking the targets using a multi-target tracking algorithm and managing the trajectories; accurately obtaining the phase information corresponding to multiple targets using an algorithm for extracting target phase information based on secondary clustering and statistical information proposed by the present invention; and extracting breathing information on this basis. The present invention designs and integrates a set of multi-target tracking and breathing detection systems and integrates them into a set of devices, which has the advantages of high real-time performance, high integration, and high engineering value. Through this set of devices, the number and position of targets can be detected in real time at the terminal, and the targets can be accurately tracked, and the breathing information of each target can be detected during this process.

Description

Technical Field

[0001] The present invention relates to the fields of indoor personnel monitoring, medical treatment, post-disaster rescue, etc., and particularly relates to a multi-target tracking and breathing detection method and device based on a millimeter-wave radar. Background Art

[0002] Cameras have always been the main means for indoor personnel monitoring. However, the privacy protection issue of video monitoring brings great inconvenience to users, especially in private places such as homes, public bathrooms, toilets, and hospital wards, where there are great limitations. In recent years, the millimeter-wave radar technology has developed rapidly, making its cost and volume extremely close to those of cameras, while having unique advantages in performance, being unaffected by environmental factors such as weather, temperature, and light, not requiring imaging to solve privacy concerns, and having extremely high position resolution to accurately achieve positioning and easily detect the position where the target is located.

[0003] Breathing is one of the basic vital sign information parameters of the human body, and its change directly reflects whether the human body is healthy, playing a crucial role in the biomedical field. Currently, the commonly used breathing detection instruments mainly obtain human breathing information through sensors in contact with the human body, such as respiratory monitors, etc. Such contact measurement instruments are very inconvenient and cannot meet the requirements of some users (such as mental patients, infectious patients) under many conditions. Radar has advantages in breathing signal detection, such as not having to be in direct contact with the subject, being unaffected by environmental factors, and having strong penetration ability, which is of great significance for long-term physiological monitoring of the elderly in hospitals and at home.

[0004] Currently, radar systems applied in the field of life signal detection, such as continuous wave (CW) radar and ultra-wideband (UWB) radar, respectively have defects such as being unable to measure the propagation delay time and thus unable to obtain distance information, low spectrum utilization efficiency, high requirements for the sampling rate, resulting in low measurement accuracy, etc. The FMCW millimeter-wave radar has the advantages of both CW radar and UWB radar, can simultaneously achieve ranging and speed measurement of the target, thereby can distinguish multiple detection targets, and extract target micro-motion information (such as breathing information), and has strong anti-interference ability. However, the research on applying FMCW radar in the field of life signal detection started relatively late, and the signal processing process is relatively complex. Currently, most research has relatively large restrictions on the number and position of test personnel, and generally can only perform vital sign detection on a single person at a specified position (about 0.5 m directly facing the millimeter-wave radar), and there is basically no research on combining multi-target detection and breathing detection, and there is still great development potential in this field. Summary of the Invention

[0005] The object of the present invention is to provide a multi-target tracking and breathing detection method and device based on millimeter-wave radar in view of the deficiencies of the prior art, to propose a complete multi-target tracking and breathing detection method and finally integrate it into a set of devices, which can simultaneously detect the number and position of multiple targets in real time at the terminal and accurately track the targets, and detect the breathing information of each target during this process.

[0006] The object of the present invention is achieved by the following technical solutions:

[0007] On the one hand, the present invention provides a multi-target tracking and breathing detection method based on millimeter-wave radar, including the following steps:

[0008] S1. Collect millimeter-wave radar signals to obtain intermediate frequency (IF) signals;

[0009] S2. Process the IF signals to obtain the distance, angle and phase information corresponding to each detection point;

[0010] S3. Use the CASO-CFAR algorithm to detect target points in the distance domain and the angle domain respectively;

[0011] S4. Use Doppler Fourier transform to obtain the velocity information of the detected target points:

[0012] S5. At the initial moment, use the DBSCAN algorithm to cluster the point cloud information to obtain the number of targets, the point cloud corresponding to each target and the location information;

[0013] S6. Track the targets and manage the trajectories through a multi-target tracking algorithm combining clustering, data association and extended Kalman filtering;

[0014] S7. Extract the phase information corresponding to each target based on the target phase information extraction algorithm of quadratic clustering and statistical information;

[0015] S8. Obtain the breathing information of the targets based on the peak search method and the discrete short-time variable base fast Fourier transform algorithm.

[0016] Further, the S2 specifically includes the following sub-steps:

[0017] S201. Perform distance Fourier transform on the IF signals to obtain the distance information of each detection point:

[0018] S202. Use the Capon beamforming algorithm to estimate the angle of arrival by searching for spectral peaks on the Capon spectrum to obtain the angle information of each detection point:

[0019] S203. Obtain the phase information of each detection point through arctangent transformation.

[0020] Further, in S4, N Chirps with a radar emission interval of T are transmitted. Each reflected Chirp pulse is processed by Range-FFT, and peaks with different phases will appear at the same position. The phase difference Δφ at the peak and the displacement Δd of the target movement satisfy Δφ = 4πΔd / λ, where λ is the wavelength of the radar signal. Since Δd = vT c , the speed can be obtained as: v = λΔφ / 4πT c ; Perform Dopplor-FFT on the result of the range FFT in the Chirp dimension to obtain a range-Doppler map. Extract the peak of the range-Doppler map to obtain the Doppler frequency of the target, and then the speed can be calculated. c ; Perform Dopplor-FFT on the result of the range FFT in the Chirp dimension to obtain a range-Doppler map. Extract the peak of the range-Doppler map to obtain the Doppler frequency of the target, and then the speed can be calculated.

[0021] Further, S5 specifically includes the following sub-steps:

[0022] S501: Set two parameters: the neighborhood radius eps and the minimum number of points min_samples required to form a class;

[0023] S502: Select an unmarked point a i in sequence from the detected target points as the clustering center, and initialize the number of points c i of this class to 1;

[0024] S503: Start traversing other unmarked points after step S502. If the distance from a certain point to the clustering center is less than or equal to eps, then assign it to the class corresponding to this clustering center, and increment the number of points c i of this class by 1;

[0025] S504: Recalculate the clustering center of this class C is the set of points in this class, that is, the new clustering center of this class is the centroid of all samples in this class;

[0026] S505: Repeat steps S503 and S504 until it is judged whether the number of points in this class satisfies being greater than or equal to min_samples after all points have been traversed. If so, mark all points in this class as having been clustered, take these point clouds as a new class, and take the final clustering center as the center of the target corresponding to this class;

[0027] S506: Repeat steps S502 to S505 until all detected target points have been traversed to obtain the number of targets, the point cloud corresponding to each target, and the target position information.

[0028] Further, in S6, the multi-target tracking algorithm that combines clustering, data association, and extended Kalman filtering is used to track the target, including:

[0029] S601. If this moment is the starting moment, use the multiple target center points obtained by clustering as the starting points of the trajectories; if this moment is not the starting moment, determine the target association thresholds for the predicted points corresponding to each trajectory obtained at the previous moment, select the points within each target association threshold at this moment as measurement points, and perform data association on the measurement points and the predicted points of this trajectory obtained at the previous moment using a data association algorithm.

[0030] S602. Perform extended Kalman filtering based on the result of data association to update the trajectory state at this moment, use the updated state as the tracking value of this trajectory, and predict the state of the target at the next moment according to the update result.

[0031] S603. Cluster the points that have not completed data association again to determine whether there are new targets. If there are new targets, update the trajectories according to the relevant information of the cluster centers.

[0032] Further, in S6, the trajectory management method is specifically as follows:

[0033] There are three states for the trajectory state corresponding to each target: being detected, being associated, and not associated.

[0034] If the trajectory satisfies the condition of becoming a target trajectory after clustering, set its state to a detected trajectory; if it is continuously detected in multiple frames that there are points exceeding a specified threshold that can be associated with this detected target trajectory, update the state of this target trajectory to being associated; if there is not a single point associated with this detected target trajectory, directly set its state to not associated; if a trajectory with a state of being associated for multiple consecutive frames still fails to match measurement points, delete this trajectory; at the same time, record and store the trajectory information of the detected and associated trajectories, the matched point clouds, and their position and phase information.

[0035] Further, S7 specifically includes the following sub-steps:

[0036] S701. Perform secondary clustering according to the position information of the point clouds corresponding to each target, select the point clouds corresponding to the median interval of the distribution, and thus obtain the phase information of the corresponding target.

[0037] S702. Preprocess the phase values corresponding to each target respectively, including: performing phase unwrapping processing to obtain the actual phase value, and calculating the phase difference to eliminate phase offset.

[0038] Further, S8 specifically includes the following sub-steps:

[0039] S801. Use an IIR band-pass filter to obtain the respiration waveforms of each target.

[0040] S802. Obtain the breathing rate of each target by using the peak-seeking method and the discrete short-time variable-base fast Fourier transform operation;

[0041] The formula for obtaining the breathing rate by the peak-seeking method is:

[0042]

[0043] where Δt is the interval time, M is the number of peaks within the time of Δt, and f res can be regarded as the breathing rate within the time of Δt;

[0044] The formula for the discrete short-time variable-base fast Fourier transform is:

[0045]

[0046] where w is the window function, X(k) is the Fourier transform of x(n)w, x(n) is the time-domain sequence, and N is the number of points of the Fourier transform;

[0047] Use the mixed-base FFT algorithm, that is, if N = 4 Q *2, Q is a positive integer, the Fourier transform can be decomposed into log 4 N = Q butterfly FFTs with a base of 4 and one butterfly FFT with a base of 2 at the last stage. In the case where the number of points is an integer power of 4, use the butterfly FFT with a base of 4, and in the case where the number of points is an integer power of 2 but not an integer power of 4, use the mixed-base FFT, that is, use the butterfly FFT with a base of 2 at the last stage and use the butterfly FFT with a base of 4 for the previous stages;

[0048] After performing the discrete short-time variable-base fast Fourier transform on the phase information of each target, the position corresponding to the spectral peak is the breathing rate of the target.

[0049] On the other hand, the present invention provides a multi-target tracking and breathing detection device based on a millimeter-wave radar, and the device includes a radar front-end system, a DSP subsystem, and a main subsystem;

[0050] The radar front-end system includes a transmitting antenna, a receiving antenna, a synthesizer, a mixer, and an ADC module, and is used to collect millimeter-wave radar signals to obtain intermediate-frequency IF signals;

[0051] The DSP subsystem is used to implement steps S2 to S4 in the above method, and transmit the detected target point cloud and its related information to the main subsystem for subsequent processing;

[0052] The main subsystem is used to implement steps S5 to S8 in the above method.

[0053] Further, the DSP subsystem uses a high-performance C674x DSP, which has powerful data processing capabilities and a high operating speed. A large number of basic signal data can be processed in parallel and in real time in the DSP subsystem; the main subsystem uses an ARM Cortex-R4F processor that is clocked at a frequency of 200 MHz.

[0054] The beneficial effects of the present invention are as follows: A multi-target tracking and respiration detection method and device based on a millimeter-wave radar proposed by the present invention can track the number, position, and trajectory of indoor personnel using the millimeter-wave radar, is not affected by environmental factors, does not require imaging, solves privacy concerns, and the ultra-high position resolution can accurately achieve positioning and target tracking. In addition, the millimeter-wave radar is used to non-contactedly detect the respiration of targets, which is more convenient. Compared with existing non-contact detection technologies, it not only improves the accuracy but also can achieve respiration detection of multiple targets according to the positions of different detected targets. The present invention designs and integrates a multi-target tracking and respiration detection system based on a millimeter-wave radar. Through the design of the software architecture and algorithm optimization, the system is finally fully integrated on a set of devices. Multiple functions such as detecting the number, position, tracking, and respiration detection of multiple indoor personnel can be completed on the terminal, which can meet both the monitoring needs and the respiration detection requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is the overall flowchart of the method of the present invention;

[0056] Figure 2 is a comparison diagram of the effects of the original DBSCAN algorithm and the clustering algorithm proposed by the present invention;

[0057] Figure 3 is a partial flowchart of the multi-target tracking algorithm;

[0058] Figure 4 is the flowchart of the trajectory management algorithm proposed by the present invention;

[0059] Figure 5 is a partial flowchart of the phase and respiration information extraction algorithm;

[0060] Figure 6 is the overall system structure diagram of the radar device;

[0061] Figure 7 is a diagram showing the tracking situation of multiple targets using the method and device proposed by the present invention;

[0062] Figure 8 and 9 is a diagram showing the respiration detection situation of multiple targets using the method and device proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0064] It should be clear that the described embodiments are only part of the embodiments of this application, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative work belong to the scope of protection of this application.

[0065] The terms used in the embodiments of this application are only for the purpose of describing specific embodiments, and are not intended to limit this application. The singular forms "a", "the", and "said" used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0066] The embodiments of the present invention provide a multi-target tracking and breathing detection method based on a millimeter-wave radar, as Figure 1 shown, and the specific implementation steps are as follows:

[0067] S1. Collect millimeter-wave radar signals;

[0068] The millimeter-wave radar transmits and receives FMCW frequency-modulated continuous waves. The characteristic of this electromagnetic wave is that the frequency of the signal linearly increases with time. The intermediate frequency IF signal generated after processing the transmitted waveform and the received waveform by the mixer is collected. Among them:

[0069]

[0070] X1 is the received signal, X2 is the transmitted signal, and Xout is the intermediate frequency IF signal; ω1, are the frequency and phase of the received signal respectively, and ω2, are the frequency and phase of the transmitted signal respectively.

[0071] S2. Perform corresponding processing on the collected intermediate frequency IF signal to obtain the distance, angle, and phase information corresponding to each detection point;

[0072] S201. Perform a distance Fourier transform on the collected intermediate frequency IF signal to obtain the distance information of each detection point:

[0073] d = fc / 2S

[0074] where f is the frequency of the intermediate frequency IF signal, c is the speed of light, and S is the change rate of the frequency of the FMCW frequency-modulated continuous wave emitted by the millimeter-wave radar;

[0075] S202. Use the Capon beamforming algorithm to estimate the angle of arrival by searching for spectral peaks on the Capon spectrum to obtain the angle information of each detection point:

[0076] When the target distance changes slightly, it will cause a large change in the phase at the peak of the Range-FFT. Therefore, the angle of arrival can be estimated by using the phase change caused by the distance difference between the target and the two antennas. The angle of arrival can be estimated by searching for the spectral peak on the Capon spectrum;

[0077] S203. Obtain the phase information of each detection point through the arctangent transformation;

[0078] S3. Detect target points using the CASO-CFAR algorithm in the range domain and the angle domain respectively;

[0079] S4. Obtain the velocity information of the detected target points using the Doppler Fourier transform:

[0080] The FMCW radar transmits N Chirps with an interval of T c Each reflected Chirp pulse is processed by Range-FFT, and peaks with different phases will appear at the same position. The phase difference Δφ at the peak is related to the displacement Δd of the target movement. Δφ = λ4πΔd, where λ is the wavelength of the radar signal. And since Δd = vT c , the velocity can be obtained as: v = λΔφ / 4πT c . Therefore, perform Dopplor-FFT on the result of the range FFT in the chirp dimension to obtain the range-Doppler map. Extract the peak of the range-Doppler map to obtain the Doppler frequency of the target, and then the velocity can be obtained.

[0081] S5. At the initial moment, use the improved fast DBSCAN algorithm newly proposed in the present invention to cluster the point cloud information to obtain state information such as the number of targets, the point cloud corresponding to each target, and the location. This clustering algorithm combines the ideas and advantages of the density-based DBSCAN and the partition-based Kmeans clustering algorithms, has a low algorithm complexity and better clustering results in the research scenario of the present invention, can eliminate the target points generated by the misdetection of the local moving parts (head, hands, etc.) of the human body or the slight movement of the chair in the point cloud, and helps to improve the accuracy of the target position in the multi-target tracking algorithm and the phase acquisition accuracy in the respiration detection; specifically includes the following sub-steps:

[0082] S501. First, set two parameters: the neighborhood radius eps and the minimum number of points min_samples required to form a class;

[0083] S502. Select an unmarked point a i in sequence from the detected target points as the clustering center, and initialize the number of points c i = 1;

[0084] S503. Start traversing other unmarked points after step S502. If the distance from a certain point to the cluster center is less than or equal to eps, then assign it to the class corresponding to this cluster center, and increment the number of points c in this class. i by 1;

[0085] S504. Recalculate the cluster center for this class. Let C be the set of points in this class, that is, the new cluster center for this class is the centroid of all samples in this class.

[0086] S505. Repeat steps S503 and S504 until after all points are traversed, check whether the number of points in this class satisfies being greater than or equal to min_samples. If so, mark all points in this class as clustered, take these point clouds as a new class, and take the final cluster center as the center of the target corresponding to this class.

[0087] S506. Repeat steps S502 to S505 until all detected target points are traversed, to obtain status information such as the number of targets, the point clouds corresponding to each target, and the target positions.

[0088] Figure 2 It is the clustering result of the point clouds obtained by detecting two targets within a certain period of time. The left figure is the clustering effect obtained by using the original DBSCAN algorithm, and the right figure is the clustering result obtained by processing using the improved fast DBSCAN algorithm proposed by the present invention. It can be found that the clustering effect has a better improvement compared with the original algorithm. Some relatively discrete point clouds that may be due to local movements such as hands or misdetections such as chairs are better screened out, and the completed clustering is all the denser concentrated point clouds corresponding to the target main body.

[0089] S6. Use a multi-target tracking algorithm that combines clustering, data association, and extended Kalman filtering to track the targets, use the trajectory management scheme designed by the present invention to manage the trajectories and match information such as the point clouds corresponding to each target and the phase of the points. The main process of this part of the algorithm is as Figure 3 shown; specifically including the following sub-steps:

[0090] S601. If this moment is the starting moment, take the multiple target center points obtained by clustering as the starting points of the trajectories; if this moment is not the starting moment, then determine the target association thresholds for the predicted points corresponding to each trajectory obtained at the previous moment, select the points within the target association thresholds at this moment as the measurement points, and perform data association on the predicted points of this trajectory obtained at the previous moment using the data association algorithm. The data association algorithm can adopt the JPDA algorithm.

[0091] S602. Perform extended Kalman filtering based on the result of data association to update the trajectory state at this moment, use the updated state as the tracking value of this trajectory, and predict the state of the target at the next moment according to the update result;

[0092] S603. Cluster the points that have not completed data association again to determine whether a new target appears. If a new target appears, update the trajectory according to the relevant information of the cluster center;

[0093] S604. Manage the trajectories using the trajectory management scheme proposed by the present invention. Through this set of trajectory management schemes, the management and matching of trajectories can be better completed, and the situation where temporarily appearing noise points are misidentified as new target trajectories and the target reappears after being occluded for a short time can be better excluded. The trajectory management algorithm process proposed by the present invention is as Figure 4 shown. There are three states in total for the trajectory state corresponding to each target: being detected, being associated, and not associated. If the trajectory meets the conditions to become a target trajectory after clustering, its state is set to the detected trajectory; if it is continuously detected in multiple frames that there are points exceeding the specified threshold that can be associated with this detected target trajectory, the state of this target trajectory is updated to being associated; if there is not any point associated with the detected target trajectory, its state is directly set to the not associated state; if a certain trajectory with the state of being associated still fails to match the measurement points after multiple consecutive frames, delete this trajectory; at the same time, record and store the trajectory information of the detected and associated trajectories, the matched point cloud, and its position and phase and other information for use in subsequent multi-target tracking and phase and respiration information extraction processes.

[0094] S7. The overall process of the phase and respiration information extraction algorithm part is as Figure 5 shown. Use a newly proposed target phase information extraction algorithm based on quadratic clustering and statistical information in the present invention to extract the phase information corresponding to each target; specifically, it includes the following sub-steps:

[0095] S701. Perform more strict clustering screening on the point cloud obtained by clustering or data association for each target, so that some point clouds with relatively low overall density are screened out, that is, the point cloud information corresponding to local movements such as hands and heads can be excluded as much as possible through quadratic clustering. Finally, the obtained point cloud is mostly distributed in the chest cavity. Perform quadratic clustering according to the position information of the point cloud corresponding to each target and select the point cloud corresponding to the median interval of the distribution to obtain the phase information corresponding to the target;

[0096] S702. Preprocess the phase values corresponding to each of the above targets respectively, mainly including: performing phase unwrapping processing to obtain the actual phase value, and calculating the phase difference to eliminate the phase shift;

[0097] S8. Use a respiration extraction scheme based on peak searching methods and algorithms such as the discrete short-time variable base fast Fourier transform proposed in the present invention to obtain the respiration information of the target; specifically, it includes the following sub-steps:

[0098] S801. Use an IIR band-pass filter (0.2 Hz - 0.5 Hz) to obtain the respiration waveforms of each target.

[0099] S802. Use operations such as peak searching methods and the discrete short-time variable base fast Fourier transform proposed in the present invention to obtain the respiration rates of each target. The main principle of using the peak searching method to obtain the respiration rate is:

[0100]

[0101] where Δt is the interval time, M is the number of peaks within the time interval Δt, and f res can be regarded as the respiration rate within the time interval Δt.

[0102] The discrete short-time variable base fast Fourier transform proposed in the present invention is mainly based on the discrete short-time Fourier transform:

[0103]

[0104] where w is the window function, X(k) is the Fourier transform of x(n)w, x(n) is the time-domain sequence, and N is the number of points of the Fourier transform.

[0105] To improve the operation efficiency and reduce the storage space, in the actual use of the present invention, a mixed base FFT algorithm is also designed, that is, if N = 4 Q * 2, Q is a positive integer, the Fourier transform can be decomposed into log 4 N = Q butterfly FFTs with a base of 4 and a final stage of a butterfly FFT with a base of 2. In the case where the number of points is an integer power of 4, the butterfly FFT with a base of 4 is used, and in the case where the number of points is an integer power of 2 but not an integer power of 4, the mixed base FFT is used, that is, the final stage uses a butterfly FFT with a base of 2, and the previous stages all use butterfly FFTs with a base of 4, so as to minimize the time complexity and improve the efficiency.

[0106] After performing the discrete short-time variable base fast Fourier transform on the phase information of each target, the position corresponding to the spectral peak is the respiration rate of the target.

[0107] Finally, a multi-target tracking and respiration detection system based on millimeter-wave radar is designed and integrated. Through the design of the software architecture and algorithm optimization, this method is finally integrated into a multi-target tracking and respiration detection device based on millimeter-wave radar. The overall system structure diagram of this device is as Figure 6 shown, and it mainly consists of three major modules: the radar front-end system, the DSP subsystem, and the main subsystem.

[0108] The radar front-end system mainly includes 2 transmitting antennas, 4 receiving antennas, a synthesizer, a mixer, and an ADC module, etc.

[0109] The DSP subsystem includes a high-performance C674x DSP, which has powerful data processing capabilities and high operating speed. In the DSP subsystem, a large number of basic signal data can be processed in parallel in real time, including performing operations such as FFT and Capon BF to obtain relevant range and angle information, using Doppler Fourier transform and arctangent transform to obtain the speed and phase information of each detection point, and on this basis, completing the detection of the target point cloud, that is, mainly corresponding to steps S2 to S4 in the aforementioned multi-target tracking and respiration detection method based on millimeter-wave radar. At this time, the data volume is greatly reduced, so the detected target point cloud and its related information are transmitted to the master subsystem through Mailbox for subsequent processing.

[0110] The master subsystem contains an ARM Cortex-R4F processor that clocks at a frequency of 200 MHz. This part mainly includes algorithms for multi-target tracking, special processing of phase information, and respiration information extraction, etc., which are relatively advanced parts. Through this set of devices, the number and position of targets can be detected in real time on the terminal and the targets can be accurately tracked, and during this process, the respiration information of each target is detected, that is, mainly corresponding to steps S5 to S8 in the aforementioned multi-target tracking and respiration detection method based on millimeter-wave radar.

[0111] Using the device integrated with the algorithm proposed in the present invention to perform tracking and respiration detection on multiple targets in the room, the tracking situation of two targets is as Figure 7 shown, and the two targets can be accurately tracked. The respiration waveforms and spectra of the two targets are obtained as Figure 8 and 9 shown. Good waveforms of each target can be obtained and there are very obvious spectral peaks in the spectrum, and the respiration rates of the two targets can be easily obtained as 21 bpm and 24 bpm respectively.

[0112] As described above, it is only the specific implementation manner of the present invention, but the implementation manner of the present invention is not limited thereto. Any other change, modification, or replacement method that does not deviate from the spirit and principle of the present invention is within the protection scope of the present invention, and specifically, it is subject to the protection scope of the claims.

Claims

1. A multi-target tracking and respiration detection method based on millimeter-wave radar, characterized in that, it includes the following steps: S1. Collect millimeter-wave radar signals to obtain intermediate frequency (IF) signals; S2. Process the IF signals to obtain the distance, angle, and phase information corresponding to each detection point; S3. Use the CASO-CFAR algorithm to detect target points in the distance domain and the angle domain respectively; S4. Use Doppler Fourier transform to obtain the velocity information of the detected target points: S5. At the initial moment, use the DBSCAN algorithm to cluster the point cloud information to obtain the number of targets, the point cloud corresponding to each target, and the location information: S6. Use a multi-target tracking algorithm combining clustering, data association, and extended Kalman filtering to track the targets and manage the trajectories; S7. Extract the phase information corresponding to each target based on the target phase information extraction algorithm of secondary clustering and statistical information, specifically including the following sub-steps: S701. Perform secondary clustering according to the position information of the point cloud corresponding to each target and select the point cloud corresponding to the median interval of the distribution to obtain the phase information corresponding to the target; S702. Preprocess the phase values corresponding to each target respectively, including: performing phase unwrapping processing to obtain the actual phase value, and calculating the phase difference to eliminate the phase offset; S8. Obtain the respiration information of the targets based on the peak search method and the discrete short-time variable base fast Fourier transform algorithm, specifically including the following sub-steps: S801. Use an IIR band-pass filter to obtain the respiration waveforms of each target; S802. Use the peak search method and the discrete short-time variable base fast Fourier transform operation to obtain the respiration rates of each target; The formula for obtaining the respiration rate by the peak search method is: where Δt is the interval time, M is the number of peaks within the time period of Δt, and f res is the breathing rate within the time period of Δt; The formula for the discrete short-time variable base fast Fourier transform is: where w is the window function, X(k) is the Fourier transform of x(n)w, x(n) is the time-domain sequence, and N is the number of points of the Fourier transform; Using a mixed - radix FFT algorithm, i.e., if N = 4 Q *2 and Q is a positive integer, the Fourier transform is decomposed into log 4 N = Q butterfly FFTs with a radix of 4 and one butterfly FFT with a radix of 2 at the last stage. When the number of points is an integer power of 4, the butterfly FFT with a radix of 4 is used. When the number of points is an integer power of 2 but not an integer power of 4, the mixed - radix FFT is used, i.e., the butterfly FFT with a radix of 2 is used at the last stage, and the butterfly FFTs with a radix of 4 are used in the previous stages; After performing the discrete short-time variable base fast Fourier transform on the phase information of each target, the position where the spectral peak is located corresponds to the respiration rate of the target.

2. The multi-target tracking and respiration detection method based on millimeter-wave radar according to claim 1, characterized in that, the specific steps of S2 include the following sub-steps: S201. Perform distance Fourier transform on the IF signals to obtain the distance information of each detection point: S202. Use the Capon beamforming algorithm to estimate the angle of arrival by searching for the spectral peak on the Capon spectrum to obtain the angle information of each detection point: S203. Obtain the phase information of each detection point through the arctangent transformation.

3. The multi-target tracking and respiration detection method based on millimeter-wave radar according to claim 1, characterized in that, In S4, the radar transmission interval is T c Each reflected Chirp pulse is processed by Range-FFT, and peaks with different phases appear at the same position. The phase difference Δφ at the peak and the displacement Δd of the target satisfy Δφ=λ4πΔd, λ is the wavelength of the radar signal, and since Δd=vT c , the velocity is: v = λΔφ / 4πT c Perform Doppler-FFT on the result of range FFT in the Chirp dimension to obtain a range-Doppler graph. Extract the peak of the range-Doppler graph to obtain the Doppler frequency of the target, and then calculate the speed.

4. The multi-target tracking and respiration detection method based on millimeter-wave radar according to claim 1, characterized in that, the specific steps of S5 include the following sub-steps: S501. Set two parameters: the neighborhood radius eps and the minimum number of points min_samples required to form a class; S502. Select an unmarked point a from the detected target points in sequence i , as the clustering center, and initialize the number of points c in this class i = 1; S503. Start traversing other unmarked points after step S502. If the distance from a certain point to the cluster center is less than or equal to eps, then assign it to the class corresponding to this cluster center, and increment the number of points c in this class i by 1; S504. Recalculate its clustering center for this category C is the set of points in this category, that is, the new clustering center of this category is the centroid of all samples in this category; S505. Repeat steps S503 and S504 until all points are traversed. Then, determine whether the number of points in this category satisfies the condition of being greater than or equal to min_samples. If so, mark all the points in this category as clustered, regard these point clouds as a new category, and use the last clustering center as the center of the target corresponding to this category. S506. Repeat steps S502 to S505 until all detected target points are traversed, to obtain the number of targets, the point clouds corresponding to each target, and the target position information.

5. A multi-target tracking and breathing detection method based on a millimeter-wave radar according to claim 1, characterized in that in S6, a multi-target tracking algorithm combining clustering, data association, and extended Kalman filtering is used to track the targets, including: S601. If this moment is the starting moment, use the multiple target center points obtained by clustering as the starting points of the trajectories; if this moment is not the starting moment, determine the target association thresholds for the predicted points corresponding to each trajectory obtained at the previous moment, select the points within each target association threshold at this moment as the measurement points, and perform data association on the measurement points and the predicted points of this trajectory obtained at the previous moment using the data association algorithm. S602. Perform extended Kalman filtering according to the results of data association to update the trajectory state at this moment, use the updated state as the tracking value of this trajectory, and predict the state of the target at the next moment according to the update result. S603. Cluster the points that have not completed data association again to determine whether there are new targets. If there are new targets, update the trajectories according to the relevant information of the clustering centers.

6. A multi-target tracking and breathing detection method based on a millimeter-wave radar according to claim 5, characterized in that in S6, the trajectory management method is specifically as follows: There are three states for the trajectory state corresponding to each target: being detected, being associated, and not associated; If the trajectory satisfies the condition of becoming a target trajectory after clustering, set its state to the detected trajectory; if more than a specified threshold of points are detected and can be associated with this detected target trajectory for multiple consecutive frames, update the state of this target trajectory to being associated; if none of the points in the detected target trajectory are associated with this trajectory, directly set its state to the not associated state; if a certain trajectory with the state of being associated for multiple consecutive frames still does not match the measurement points, delete this trajectory; at the same time, record and store the trajectory information, the matched point clouds, and their position and phase information of the detected and associated trajectories.

7. A multi-target tracking and breathing detection device based on a millimeter-wave radar, characterized in that the device includes a radar front-end system, a DSP subsystem, and a main subsystem; the radar front-end system includes a transmitting antenna, a receiving antenna, a synthesizer, a mixer, and an ADC module, and is used to collect millimeter-wave radar signals to obtain intermediate frequency (IF) signals; the DSP subsystem is used to implement steps S2 to S4 in the method according to any one of claims 1-6, and transmit the detected target point clouds and their relevant information to the main subsystem for subsequent processing. The master subsystem is used to implement steps S5 to S8 in the method according to any one of claims 1-6.

8. The apparatus according to claim 7, wherein, the DSP subsystem uses a high-performance C674x DSP, which has powerful data processing capabilities and high operating speeds, and can parallelly and real-timely complete the processing of a large amount of basic signal data; the master subsystem uses an ARM Cortex-R4F processor that is timed at a frequency of 200 MHz.

Images