CN117805764A - Millimeter wave radar anti-interference fall detection method based on three-dimensional time-frequency information matrix - Google Patents

Abstract

Compared with a method for inputting a plurality of two-dimensional short-time feature images to enter a neural network for identification, the method for detecting the interference of the millimeter wave radar based on the three-dimensional time-frequency information matrix, disclosed by the application, has the advantages that interference information in a current testing environment is firstly collected, and in the subsequent human body movement testing process, the influence of external interference on the extraction of human body target movement feature information is effectively inhibited through a Doppler matrix cancellation method with the interference environment. Meanwhile, in order to effectively fuse micro Doppler information in the human body movement process, the extracted human body Doppler information matrix is constructed into a three-dimensional time-frequency information matrix form, so that the complexity of multiple networks and multiple inputs is avoided, the identification effect of the millimeter wave radar on falling detection under different interference environments is effectively improved, and the falling detection accuracy is improved.

Description

Millimeter wave radar anti-interference fall detection method based on three-dimensional time-frequency information matrix

Technical field

The invention relates to the technical field of radar detection, specifically a millimeter wave radar anti-interference fall detection method based on a three-dimensional time-frequency information matrix.

Background technique

As the elderly population continues to grow, their health problems have attracted social attention. Falls are a particularly common cause of disability or even death among the elderly, so fall monitoring for the elderly is very necessary. The current common fall monitoring methods are: (1) Surveillance cameras: Install surveillance cameras in each room and use video sequence information to detect whether the elderly have fallen. (2) Wearable devices: The elderly need to wear sensor devices such as smart watches to use the changes in acceleration and gyroscope horizontal angle information when falling to determine whether they are currently in a fall state. (3) Millimeter wave radar: Install radar equipment in the room to identify the current fall behavior based on the significant changes in the short-term characteristics of the radar echo signal when the elderly fall.

In the fall detection method based on computer vision surveillance video, there are problems such as privacy protection for the elderly in private indoor environments, and unstable and deteriorating detection effects under different light environments; and in the fall detection method based on wearable sensor devices, due to the decline in memory and physical strength of the elderly, there are problems such as forgetting to wear the equipment and inconvenience in movement. Therefore, the present invention uses the echo signal of the millimeter-wave radar to achieve fall detection. Since the radar echo signal contains both target information and interference factors such as fans and air conditioners in the indoor environment, its motion feature information will also be collected by the radar and mixed with the human target signal, greatly affecting the effect of fall detection. If there is too much interference feature information, the fall recognition network will pay more attention to the interference information and ignore the time-frequency information containing the target, resulting in fall detection errors.

At present, most of the fall recognition methods based on millimeter wave radar adopt the method of inputting short-term feature maps into neural network recognition. After the extracted echo signals are processed, they are converted into feature maps such as time-frequency diagrams and range Doppler spectra, and then they are jointly input into one or several networks to synthesize the feature information extracted from the current human action. Discrimination. However, most current methods do not make full use of the micro-Doppler features during human movement. The method of using multiple feature maps is cumbersome and does not effectively suppress interference problems in the environment, which leads to low fall detection accuracy. question.

Contents of the invention

The purpose of the present invention is to propose a millimeter wave radar anti-interference fall detection method based on a three-dimensional time-frequency information matrix to address the problem that most current methods do not fully utilize the micro-Doppler characteristics of human body movement, the method of using multiple feature maps is cumbersome, and there is no effective suppression of interference problems in the environment, which leads to low accuracy of fall detection.

The technical solution adopted by the present invention to solve the above technical problems is:

The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix includes the following steps:

Step 1: In the same indoor environment, use the millimeter wave radar to obtain the transmission signal and echo signal when there is no human motion, and the transmission signal and echo signal when there is human motion;

Step 2: Mix the transmit signal and the echo signal when there is no human body movement to obtain the intermediate frequency signal, sample the intermediate frequency signal to obtain the discrete intermediate frequency signal, and perform window processing on the discrete intermediate frequency signal, and then perform Doppler The Fourier transform of Levey and distance dimensions is used to obtain the distance Doppler matrix of each frame of data, and the distance Doppler matrix RD _noise with the largest element value is selected;

Step 3: Process the transmit signal and echo signal when there is human body movement through the steps of step 2 to obtain the range Doppler matrix RD _k of each frame of data, and convert the range Doppler matrix RD _k of each frame of data into Difference with RD _noise respectively to obtain the range Doppler matrix for interference removal in each frame.

Step 4: De-interference the range Doppler matrix of each frame Perform spectral peak search to obtain the distance between the human target and the radar, and compare this distance with the distance obtained in the previous frame to determine whether the difference between the current frame and the previous frame is greater than 30 range gates. If the difference is greater than 30 range gates, then Select the range Doppler matrix corresponding to the current frame, otherwise, select the range Doppler matrix corresponding to the previous frame;

Step 5: Based on the range Doppler matrix selected in step 4, obtain all the Doppler vectors within the range of the range gate number in the range Doppler matrix to construct a matrix, and then undergo the inverse Fourier transform and After short-time Fourier transform, it is used as the Doppler information matrix of the frame;

Step 6: According to steps 4 and 5, obtain the Doppler information matrix corresponding to all frames, and use this to construct a three-dimensional time-frequency information matrix. Then input the three-dimensional time-frequency information matrix into the neural network for action recognition and fall detection. detection.

Further, the transmission signal is expressed as:

Among them, A _T is the amplitude of the transmitted signal, f _c is the central slope of the signal, t is the transmission time, B is the bandwidth, T _m is the frequency modulation period, τ is the time delay, dτ is the differential of τ, and j is the imaginary unit.

Further, the echo signal is expressed as:

Among them, A _R is the amplitude of the echo signal, Δt is the time delay of the echo signal relative to the transmitted signal, and Δf _d is the Doppler frequency shift.

Further, the intermediate frequency signal is expressed as:

S _IF (t)＝S _T (t)S _R (t)≈A _T A _R exp{j2π[f _c Δt+(f _I -Δf _d )t]}

Among them, f _I represents the frequency of the intermediate frequency signal at time t,

Furthermore, the distance Doppler matrix RD _noise with the largest element value is expressed as:

RD _noise (x,y) = max(RD ₁ (x,y),RD ₂ (x,y),...,RD _n (x,y))

Wherein, RD _noise (x, y) is the element value of the xth row and yth column in the range Doppler matrix RD _noise , RD _k (x, y) is the element value of the xth row and yth column in the kth frame unmanned environment range Doppler matrix, n is the total number of frames, k = 1, 2, ..., n.

further, the Expressed as:

Further, the discrete intermediate frequency signal is expressed as:

S _I (n,m)＝A _I exp{j2π[f _I (n)-Δf _d (m)]/f _s }, (1≤n≤N), (1≤m≤M)

Among them, S _I (n, m) is the discrete expression of the m-th sampling point of the n-th cycle, A _I is the amplitude of the intermediate frequency signal, f _s is the sampling frequency, and N is the number of repeated cycles in each frame, that is, the transmitted signal The number of Chirps in, M is the number of sampling points in one frequency modulation cycle.

Further, the short-time Fourier transform is expressed as:

Where W(t) is the window function, is the motion echo signal of the target within this time range, and f is the signal frequency.

Further, the distance between the human target and the radar is expressed as:

Here, c represents the speed of light.

Further, the neural networks AlexNet and Resnet34.

The beneficial effects of the present invention are:

Compared with the method of inputting multiple two-dimensional short-time feature maps into neural network recognition, in this application, the interference information in the current test environment is first collected, and in the subsequent human motion test process, the distance Doppler matrix cancellation method with the interference environment is used to effectively suppress the influence of external interference on the extraction of human target motion feature information. At the same time, in order to effectively fuse the micro-Doppler information during human motion, this application constructs the extracted human Doppler information matrix into a three-dimensional time-frequency information matrix form, avoiding the complexity of multiple networks and multiple inputs, effectively improving the recognition effect of millimeter-wave radar for fall detection in different interference environments, and improving the accuracy of fall detection.

Description of drawings

Figure 1 is the flow chart of this application;

Figure 2 is the interference environment distance Doppler matrix diagram;

FIG3 is a diagram of the interference-removing range-Doppler matrix;

Figure 4 is a three-dimensional time-frequency information matrix diagram for interference removal.

Detailed ways

It should be particularly noted that, in the absence of conflict, the various embodiments disclosed in this application can be combined with each other.

Specific Embodiment 1: This embodiment will be described in detail with reference to Figure 1. The millimeter wave radar anti-interference fall detection method based on a three-dimensional time-frequency information matrix described in this embodiment.

Step 1: Obtain the echo signals of interference factors in the unmanned moving environment within the set number of frames of the millimeter wave radar. The transmitted signal and the echo signal are mixed to obtain an intermediate frequency signal, and the intermediate frequency signal is sampled to obtain a discrete intermediate frequency signal. After windowing the discrete intermediate frequency signal, perform Fourier transform of Doppler dimension and range dimension respectively to obtain the range Doppler matrix of each frame of data. Then, within this period of time, the range Doppler of the environment containing interference information is The Le matrix RD _noise can be expressed as:

RD _noise (x, y) = max (RD ₁ (x, y), RD ₂ (x, y),..., RD _n (x, y))

Among them, RD _noise (x, y) is the element value of the x-th row and y-th column in the distance Doppler matrix RD _noise , and RD _k (x, y) is the k-th frame of the uninhabited environment distance Doppler matrix. The element value of row x and column y, n is the total number of frames, k=1,2,...,n.

Step 2: In the above environment, obtain the transmitted echo signal containing human movement within the set number of frames of the millimeter wave radar. Process the same as step 1 to obtain the range Doppler matrix of each frame of data. Subtract the range Doppler matrix in each frame from the range Doppler matrix RD _noise to obtain the interference-free range Doppler matrix, then the kth Range Doppler matrix for frame de-interference It can be expressed as:

Among them, RD _k is the range Doppler matrix of the kth frame containing the human target;

Step 3: Perform a spectral peak search on the interference-free range Doppler matrix of each frame to obtain the distance between the target and the radar, and compare this distance with the distance obtained in the previous frame. Since the speed of the human body when falling is about 5m/s, the radar range gate resolution used in this application is 7.14cm, and the acquisition time of one frame is 0.036s. Assuming that the target height is 1.8m, it can be obtained by calculation that the radar spectrum peak search The positioning target position difference is no more than 30 distance gates. If this number of distance gates is exceeded, it indicates that the peak search algorithm may mistakenly select the interference target. Therefore, in the human target distance estimation in this frame, the human target position in the previous frame is still selected.

Step 4: Since the human body movement amplitude is approximately 50cm, the radar range gate resolution used in this application is 7.14cm. Therefore, based on the estimated position of the target, the number of range gates is set to 7, and the range Doppler matrix of the frame is All Doppler vectors within the range of the range gate number are intercepted into a matrix, and the micro-Doppler information of the target's limbs except the trunk is obtained. After inverse Fourier transform and short-time Fourier transform, it is used as the Doppler information matrix of the frame;

Step 5: Superpose the Doppler information matrices of all frames to construct a three-dimensional time-frequency information matrix, and input it into the neural network for action recognition. The three-dimensional time-frequency information matrix here can use any network that can input a three-dimensional matrix, and the output is the action classification result. This application takes the neural networks AlexNet and Resnet34 as examples.

The steps for obtaining the distance Doppler matrix of each frame of data are:

The signal obtained by the frame dividing operation is sequentially subjected to window filtering, range-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform to obtain the range Doppler matrix of each frame of data.

The transmitted signal is expressed as:

Among them, A _T is the amplitude of the transmitted signal, f _c is the central slope of the signal, t is the transmission time, B is the bandwidth, T _m is the frequency modulation period, τ is the time delay, and dτ is the differential of τ.

The echo signal is expressed as:

Among them, A _R is the amplitude of the echo signal, Δt is the time delay of the echo signal relative to the transmitted signal, Δf _d is the Doppler frequency shift, and j is the imaginary unit.

The intermediate frequency signal is expressed as:

S _IF ( t ) = S _T ( t ) S _R ( t ) ≈ _AT A _R exp { j 2 π [ f _c Δ t + ( f _I - Δ _{f d} ) t ] }

in, Indicates the frequency of the intermediate frequency signal at time t.

Further, in the sampling of the intermediate frequency signal, the discrete expression of each sampling point is:

_SI (n, m) = _AI exp{j2π[ _fI (n) - _Δfd (m)]/ _fs }, (1≤n≤N), (1≤m≤M)

Where S _I (n,m) is the discrete expression of the mth sampling point in the nth cycle, A _I is the intermediate frequency signal amplitude, _fs is the sampling frequency, N is the number of repeated cycles in each frame, that is, the number of chirps in the transmitted signal, and M is the number of sampling points in one frequency modulation cycle.

The short-time Fourier transform STFT is expressed as:

Among them, W(t) is the window function, x~(t) is the motion echo signal of the target within the time range, t is the time, and f is the signal frequency.

The millimeter wave radar transmitting signal is linear frequency modulation FMCW.

This application is a millimeter-wave radar anti-interference fall detection method based on a three-dimensional time-frequency information matrix. Compared with the method of inputting multiple short-term feature map network recognition, this application fully suppresses the echo signal after fast Fourier The interference information in the environment in the range Doppler matrix after leaf transformation is used to extract the Doppler information of human movement, and the time-frequency information in the entire process of human movement is constructed into a three-dimensional time-frequency information matrix, which is finally input into the neural network. The identification effectively improves the recognition effect of millimeter wave radar for fall detection in different interference environments.

The purpose of this application is to solve the problem that in different indoor environments, due to the existence of interference signals generated by moving targets such as fans and air conditioners, the extracted human target motion characteristics are mixed with interference information, resulting in unstable fall detection results.

Example:

This embodiment will be described in detail with reference to Figure 1. The millimeter wave radar anti-interference fall detection method based on a three-dimensional time-frequency information matrix described in this embodiment includes:

Stage 1: Determine the test environment distance Doppler matrix

The transmitted signal within one frequency modulation cycle of linear frequency modulation (FMCW) millimeter wave radar can be expressed as:

Among them, A _T is the amplitude of the transmitted signal, f _c is the central slope of the signal, t is the transmission time, B is the bandwidth, and T _m is the frequency modulation period.

After reflection from the target and the environment, the echo signal obtained by the millimeter wave radar receiving antenna is:

Among them, A _R is the amplitude of the echo signal, Δt is the time delay of the echo signal relative to the transmitted signal, and Δf _d is the Doppler frequency shift.

Step 1. Calculate the echo signal range Doppler matrix

The expression of the intermediate frequency signal obtained after mixing the transmitted signal and the echo signal is:

S _IF (t)＝S _T (t)S _R (t)≈A _T A _R exp{j2π[f _c Δt+(f _I -Δf _d )t]} (3)

in, represents the frequency of the intermediate frequency signal at time t. At this time, the distance of the target can be expressed as:

In the formula, c represents the speed of light, which is 3×10 ⁸ m/s.

When the intermediate frequency signal is sampled, the discrete expression of the mth sampling point in its nth cycle is:

S _I (n,m)＝A _I exp{j2π[f _I (n)-Δf _d (m)]/f _s }, (1≤n≤N), (1≤m≤M) (5)

Among them, _AR is the amplitude of the intermediate frequency signal, _fs is the sampling frequency, N is the number of repeated cycles in each frame, that is, the number of chirps in the transmitted signal, and M is the number of sampling points in one frequency modulation cycle.

At this time, if the M sampled discrete intermediate frequency signals with N linear frequency modulation periods in each frame are firstly subjected to window filtering to remove the high-frequency interference components, and then the range dimension FFT and Doppler dimension FFT are performed on them respectively, the range Doppler matrix of the frame data can be obtained.

Step 2: Estimate the test environment interference matrix

The maximum value of the range Doppler matrix obtained in each frame is calculated. The range Doppler matrix RD _noise containing interference information in the environment within the radar set frame number can be expressed as:

RD _noise (x, y) = max (RD ₁ (x, y), RD ₂ (x, y),..., RD _n (x, y)) (6)

Among them, RD _noise (x, y) is the element value of the x-th row and y-th column in the distance Doppler matrix RD _noise , and RD _k (x, y) is the k-th frame of the uninhabited environment distance Doppler matrix. The element value of row x and column y, n is the total number of frames, k=1,2,...,n.

According to the above steps, the range Doppler matrix RD _noise in the test environment containing interference can be calculated.

Stage 2: Range Doppler Matrix Cancellation

Step 1. Determine the interference removal range Doppler matrix

In the above environment, the transmitted echo signal containing human movement within the set number of frames of the millimeter wave radar is acquired. In the same stage of processing, the range Doppler matrix of each frame of data is obtained. The range Doppler matrix in each frame is subtracted from the range Doppler matrix RD _noise to obtain the interference-free range Doppler matrix, then the kth Range Doppler matrix for frame de-interference It can be expressed as:

Among them, RD _k is the range Doppler matrix of the kth frame containing the human target;

Step 2. Target lock

Perform a spectral peak search on the range Doppler matrix to remove interference in each frame to obtain the distance between the target and the radar, and compare this distance with the distance obtained in the previous frame. If the difference is greater than 30 range gates, it means The spectral peak search algorithm may mistakenly select interference targets. Therefore, in the human target distance estimation in this frame, the human target position in the previous frame is still selected.

Stage 3: Construct a three-dimensional time-frequency information matrix

Step 1. Micro-Doppler information fusion

Based on the estimated position of the target, set the number of range gates to 7, intercept all the Doppler vectors within the range of the range gate number in the range Doppler matrix of this frame into a matrix, and obtain the micro-Doppler of the target's limbs except the trunk. Doppler information (Doppler component mainly consists of trunk Doppler component and limb Doppler component), as the Doppler information matrix of the frame.

In order to process the time-frequency information of the continuously input time signal, the time domain of the entire signal is divided into time series of equal length through short-time Fourier transform, and fast Fourier operation is performed on each small time series. Because the signal is considered to be of finite length in a short time range, the short-time Fourier transform can effectively extract the changes in micro-Doppler information in each time period, and its expression is:

Where W(t) is the window function, It is the motion echo signal of the target within this time range.

According to the above steps, the target Doppler information in the single-frame millimeter wave radar echo signal can be calculated.

Step 2: Construct a three-dimensional time-frequency information matrix

The Doppler information matrices of all frames are superimposed to construct a three-dimensional time-frequency information matrix, which is input into the neural network for action recognition.

A deinterferenced three-dimensional time-frequency information matrix is obtained using primary environmental interference data and fall data to demonstrate the fall detection effect of this application:

Set the radar working parameters: radar starting frequency 60GHZ, FM bandwidth 2.1GHZ, each frame time 36ms, number of repetition cycles 255, a total of 120 frames of data.

specific process:

(1) Determine the interference environment distance Doppler matrix RD _noise , with a size of 256×255;

(2) Subtract the distance Doppler matrix of each frame in the fall data from the interference environment distance Doppler matrix RD _noise to obtain the interference-free distance Doppler matrix The size is 256×255;

(3) Range Doppler matrix for interference removal Conduct spectral peak search and apply target lock algorithm to prevent the search algorithm from mistakenly selecting interference targets, and determine the target position to be 57;

(4) Intercept the Doppler vectors of the seven range gates before and after the target position, with a size of 7×255, as the Doppler information matrix of the frame;

(5) The Doppler information matrix of each frame is superimposed after inverse Fourier transform and short-time Fourier transform to construct a three-dimensional time-frequency information matrix with a size of 7×255×120;

(6) Input the de-interferenced three-dimensional time-frequency information matrix into the neural network for recognition and obtain the recognition result;

Effect comparison: Compared with the previous input of multiple two-dimensional short-time feature maps, this application is suitable for indoor fall detection tasks under different interference environments; it can effectively suppress the influence of various indoor interferences on the extraction of human motion feature information, and at the same time effectively integrate the micro-Doppler information of human target motion.

It should be noted that the specific implementation is only an explanation and description of the technical solution of the present invention, and cannot be used to limit the scope of protection of the rights. Any partial changes made according to the claims and description of the present invention should still fall within the scope of protection of the present invention.

Claims (10)

1. Millimeter wave radar anti-interference fall detection method based on three-dimensional time-frequency information matrix, which is characterized by including the following steps: Step 1: In the same indoor environment, use the millimeter wave radar to obtain the transmission signal and echo signal when there is no human motion, and the transmission signal and echo signal when there is human motion; Step 2: Mix the transmit signal and the echo signal when there is no human body movement to obtain the intermediate frequency signal, sample the intermediate frequency signal to obtain the discrete intermediate frequency signal, and perform window processing on the discrete intermediate frequency signal, and then perform Doppler The Fourier transform of Levey and distance dimensions is used to obtain the distance Doppler matrix of each frame of data, and the distance Doppler matrix RD _noise with the largest element value is selected; Step 3: Process the transmit signal and echo signal when there is human body movement through the steps of step 2 to obtain the range Doppler matrix RD _k of each frame of data, and convert the range Doppler matrix RD _k of each frame of data into Difference with RD _noise respectively to obtain the range Doppler matrix for interference removal in each frame. Step 4: De-interference the range Doppler matrix of each frame Perform spectral peak search to obtain the distance between the human target and the radar, and compare this distance with the distance obtained in the previous frame to determine whether the difference between the current frame and the previous frame is greater than 30 range gates. If the difference is greater than 30 range gates, then Select the range Doppler matrix corresponding to the current frame, otherwise, select the range Doppler matrix corresponding to the previous frame; Step 5: Based on the range Doppler matrix selected in step 4, obtain all the Doppler vectors within the range of the range gate number in the range Doppler matrix to construct a matrix, and then undergo the inverse Fourier transform and After short-time Fourier transform, it is used as the Doppler information matrix of the frame; Step 6: According to steps 4 and 5, obtain the Doppler information matrix corresponding to all frames, and use this to construct a three-dimensional time-frequency information matrix. Then input the three-dimensional time-frequency information matrix into the neural network for action recognition and fall detection. detection. 2. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 1, characterized in that the emission signal is expressed as: Among them, A _T is the amplitude of the transmitted signal, f _c is the central slope of the signal, t is the transmission time, B is the bandwidth, T _m is the frequency modulation period, τ is the time delay, dτ is the differential of τ, and j is the imaginary unit. 3. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 2, characterized in that the echo signal is expressed as: Among them, A _R is the amplitude of the echo signal, Δt is the time delay of the echo signal relative to the transmitted signal, and Δf _d is the Doppler frequency shift. 4. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 3, characterized in that the intermediate frequency signal is expressed as: S _IF ( t ) = S _T ( t ) S _R ( t ) ≈ _AT A _R exp { j 2 π [ f _c Δ t + ( f _I - Δ _{f d} ) t ] } Among them, f _I represents the frequency of the intermediate frequency signal at time t, 5. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 4, characterized in that the distance Doppler matrix RD _noise with the largest selected element value is expressed as: RD _noise (x,y) = max(RD ₁ (x,y),RD ₂ (x,y),...,RD _n (x,y)) Among them, RD _noise (x, y) is the element value of the x-th row and y-th column in the distance Doppler matrix RD _noise , and RDk (x, y) is the x-th element of the k-th frame uninhabited environment distance Doppler matrix. The element value of the yth column of the row, n is the total number of frames, k=1,2,...,n. 6. The millimeter wave radar anti-interference fall detection method based on three-dimensional time-frequency information matrix according to claim 5 is characterized in that Expressed as: 7. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 6, characterized in that the discrete intermediate frequency signal is expressed as: S _I (n,m)＝A _I exp{j2π[f _I (n)-Δf _d (m)]/f _s }, (1≤n≤N), (1≤m≤M) Among them, S _I (n, m) is the discrete expression of the m-th sampling point of the n-th cycle, A _I is the amplitude of the intermediate frequency signal, f _s is the sampling frequency, and N is the number of repeated cycles in each frame, that is, the transmitted signal The number of Chirps in, M is the number of sampling points in one frequency modulation cycle. 8. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 7, characterized in that the short-time Fourier transform is expressed as: Among them, W(t) is the window function, is the motion echo signal of the target within this time range, and f is the signal frequency. 9. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 8, characterized in that the distance between the human target and the radar is expressed as: Among them, c represents the speed of light. 10. The millimeter wave radar anti-interference fall detection method based on the three-dimensional time-frequency information matrix according to claim 1, characterized by the neural networks AlexNet and Resnet34.