A millimeter wave radar-based eye capture and tracking method

Abstract

The application discloses a kind of based on millimeter wave radar's eyeball capture and tracking method, belong to radar detection technical field.The present application aims at solving the problem that existing optical eyeball tracking method is sensitive to external condition, robustness is insufficient under complex scene, including: utilize millimeter wave radar to emit frequency-modulated continuous wave signal to face, collect eye echo, generate distance dimension data and speed dimension data;From distance dimension data, determine candidate distance unit, extract its phase time sequence and analyze frequency characteristics, if frequency characteristics fall into the frequency range of eyeball physiological movement, then determine as eyeball target;Angle estimation is carried out to eyeball target to obtain three-dimensional coordinates, and filter is used to track and predict the position and speed of eyeball target.The present application does not depend on environment illumination, can penetrate glasses and slight obstruction, does not collect face image to protect privacy, realizes stable and reliable eyeball capture and tracking under complex dynamic scene.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave radar technology, and specifically to an eye-tracking method based on millimeter-wave radar. Background Technology

[0002] Eye-tracking technology is widely used in human-computer interaction, driving behavior monitoring, virtual reality, medical diagnosis, and psychological research. By acquiring the user's gaze direction or eye position in real time, it can enable functions such as gaze-controlled interactive interfaces, driver fatigue assessment, gaze behavior analysis of autistic children, and advertising effectiveness evaluation.

[0003] Typical optical eye-tracking methods use an infrared light source to illuminate the eye, creating a Purkinje spot on the corneal surface, while a camera captures images of the eye. The pupil center is extracted through image segmentation, and the corneal reflection point position is extracted using binarization or edge detection. The gaze point coordinates are then calculated based on the relative offset vectors of the two and a pre-calibrated model, and continuously output at a frame rate of 30 to 120 frames per second. However, this method is highly dependent on image quality, which is easily affected by changes in ambient lighting, the user wearing glasses or sunglasses, or the eye being obscured by eyelashes or a mask. Once pupil or reflection point features are lost or distorted, the accuracy of feature extraction, offset calculation, and gaze point mapping will gradually decrease or even fail completely. Therefore, existing optical eye-tracking methods are sensitive to external conditions, lack robustness in complex dynamic scenes, and struggle to maintain stable and reliable tracking performance. Summary of the Invention

[0004] The purpose of this invention is to provide an eye-tracking and capture method based on millimeter-wave radar to solve the technical problems of existing optical eye-tracking methods being sensitive to external conditions, lacking robustness in complex dynamic scenarios, and struggling to maintain stable and reliable tracking performance.

[0005] The technical solution of this invention is implemented as follows: In a first aspect, the present invention provides an eye-tracking and detection method based on millimeter-wave radar, comprising: Step S1: Use millimeter-wave radar to transmit frequency-modulated continuous wave signals to the face, collect eye echoes, and generate distance and velocity data. Step S2: Determine candidate distance units from the distance dimension data, extract the phase time series of the candidate distance units and analyze their frequency characteristics. If the frequency characteristics fall within the range of 0.5Hz to 5Hz of the physiological eye movement frequency, then it is determined to be an eye target. Step S3: Estimate the angle of the eye target to obtain the three-dimensional coordinates of pitch angle, azimuth angle and distance, and use a filter to track and predict the position and velocity of the eye target.

[0006] A further technical solution is that step S1 specifically includes: Step S11: Generate a linear frequency modulated continuous wave signal and radiate the signal toward the human face through a radar transmitting antenna; Step S12: Collect the echo signal reflected back from the eye area through the radar receiving antenna; Step S13: Mix the transmitted signal and the received signal to obtain the intermediate frequency signal; Step S14: Perform range-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform on the intermediate frequency signal to generate range-dimensional data and velocity-dimensional data.

[0007] A further technical solution is that step S14 specifically includes: Step S141: Sample the intermediate frequency signal within each frequency-modulated continuous wave cycle to obtain an intermediate frequency sampling sequence for one cycle; Step S142: Perform the distance dimension fast Fourier transform on the intermediate frequency sampling sequence to obtain complex data in each distance unit, which constitutes the distance dimension data; Step S143: Perform the Doppler Fast Fourier Transform on the complex data sequences of the same distance unit at different periods to obtain complex data for each distance and velocity unit; Step S144: Combine the results of the distance-dimensional fast Fourier transform with the results of the Doppler-dimensional fast Fourier transform to generate the distance-dimensional data and velocity-dimensional data.

[0008] A further technical solution is that step S2 specifically includes: Step S21: Perform constant false alarm rate detection on the distance dimension data and extract distance units whose signal amplitude exceeds the threshold as candidate distance units; Step S22: Collect echo complex values ​​on the candidate range unit for multiple consecutive frames, extract the phase component of each frame, and form a phase time series; Step S23: Use variational mode decomposition to decompose the phase time series into several intrinsic mode components, perform power spectrum estimation on each component, and screen out the dominant components whose spectral peaks are located in the 0.5Hz to 5Hz frequency band. Step S24: Calculate the energy percentage of the dominant component. When the energy percentage exceeds a preset threshold, determine that the target corresponding to the candidate distance unit is the eye target.

[0009] A further technical solution is that step S23 specifically includes: Step S231: Optimize the penalty factor and number of decomposition modes of the variational mode decomposition using the sparrow search algorithm to obtain the optimal decomposition parameters; Step S232: Perform variational mode decomposition on the phase time series according to the optimal decomposition parameters to obtain several intrinsic mode components; Step S233: Perform power spectrum estimation for each intrinsic mode component, calculate the peak amplitude of each component in the frequency band from 0.5Hz to 5Hz, and identify the component with the largest peak amplitude as a candidate component. Step S234: Perform singular value decomposition on the candidate component and calculate the singular spectral entropy. When the singular spectral entropy is lower than a preset threshold, determine the candidate component as the dominant component of eye movement.

[0010] A further technical solution is that step S24 specifically includes: Step S241: Calculate the sum of the squared values ​​of the signal strength of the dominant component on the time axis, and use it as the energy value of the dominant component; Step S242: Calculate the cumulative sum of the squared signal strength values ​​of the phase time series on the time axis, as the total energy value; Step S243: Divide the energy value of the dominant component by the total energy value to obtain the energy percentage of the dominant component; Step S244: Determine whether the energy percentage is greater than or equal to a preset threshold. If so, determine that the target corresponding to the candidate distance unit is the eye target.

[0011] A further technical solution is that step S3 specifically includes: Step S31: Use multiple receiving channels of millimeter-wave radar to acquire the echo signal of the eye target, and calculate the azimuth and elevation angles of the eye target relative to the radar by phase comparison; Step S32: Based on the distance value obtained in step S2, and combined with the azimuth and elevation angles, convert the distance, azimuth, and elevation angles into three-dimensional spatial coordinates in the radar coordinate system. Step S33: Use the three-dimensional spatial coordinates calculated at the current time as the observation value and input them into the pre-built filter to update the state estimate of the filter; Step S34: Extract the current position estimate and current velocity estimate of the eye target from the state vector of the filter, and output them as the tracking prediction result.

[0012] A further technical solution is that step S33 specifically includes: Step S331: Record the three-dimensional spatial coordinates of the current moment and multiple consecutive moments before, forming a position sequence; Step S332: Input the position sequence into a pre-trained classifier to identify whether the current eye movement belongs to saccade mode or smooth tracking mode; Step S333: If the scanning mode is identified, increase the process noise covariance value in the filter; if the smooth tracking mode is identified, decrease the process noise covariance value in the filter. Step S334: Using the current three-dimensional spatial coordinates as the observation value, and combining them with the adjusted process noise covariance value, perform filter state update to obtain the current state estimate of the eye target.

[0013] Secondly, the present invention provides an eye-tracking and detection system based on millimeter-wave radar, comprising: The signal acquisition and data generation module is used to transmit frequency-modulated continuous wave signals to the face using millimeter-wave radar, collect eye echoes, and generate distance and velocity data. An eye target capture module is used to determine candidate distance units from the distance dimension data, extract the phase time series of the candidate distance units and analyze their frequency characteristics. If the frequency characteristics fall within the range of physiological eye movement frequency, it is determined to be an eye target. The tracking and prediction module is used to estimate the angle of the eye target to obtain its three-dimensional coordinates, and to use a filter to track and predict the position and velocity of the eye target.

[0014] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the eye-capturing and tracking method based on millimeter-wave radar as described in any one of claims 1-8.

[0015] The beneficial effects of this invention are as follows: This invention replaces traditional optical cameras with millimeter-wave radar, utilizing radar echoes to generate range and velocity data. It identifies eye targets by extracting the phase time series of candidate range units and analyzing their frequency characteristics, and finally achieves continuous tracking through angle estimation and filtering. Compared to existing optical eye-tracking methods, this invention is completely independent of ambient light and can operate stably under extreme lighting conditions such as complete darkness, strong light, and backlight. Radar waves can penetrate eyeglass lenses, sunglasses, and lightly obstructing objects such as eyelashes and mask edges, effectively avoiding the loss or distortion of pupil or reflection point features in optical images. Furthermore, this invention does not collect any facial images, eliminating the risk of user privacy leakage at the source. Therefore, this invention fundamentally solves the technical problems of existing optical eye-tracking methods, such as sensitivity to external conditions, insufficient robustness in complex dynamic scenes, and difficulty in maintaining stable and reliable tracking performance, achieving all-weather, occlusion-resistant, and privacy-friendly high-robustness eye capture and tracking. Attached Figure Description

[0016] Figure 1The flowchart illustrates the steps of an eye-tracking method based on millimeter-wave radar, as provided in Embodiment 1 of the present invention. Figure 2 The flowchart of step S1 in the eye-tracking and tracking method based on millimeter-wave radar provided in Embodiment 1 of the present invention; Figure 3 The flowchart of step S2 in the eye-tracking and tracking method based on millimeter-wave radar provided in Embodiment 1 of the present invention; Figure 4 The flowchart of step S3 in the eye-tracking and tracking method based on millimeter-wave radar provided in Embodiment 1 of the present invention; Figure 5 This is a structural block diagram of an eye-tracking method system based on millimeter-wave radar provided in Embodiment 2 of the present invention; Figure 6 This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0017] To better understand the technical content of this invention, specific embodiments are provided below, and the invention will be further described in conjunction with the accompanying drawings. Example 1

[0018] This embodiment employs a 60 GHz millimeter-wave radar sensor. This sensor integrates four transmit channels and four receive channels, forming a multi-input multi-output virtual array with high angular resolution. The radar is mounted above the vehicle's dashboard, directly facing the driver, at a distance of 0.4 to 0.6 meters. The radar's frequency-modulated continuous wave (FM-CW) parameters are set as follows: starting frequency 60 GHz, bandwidth 4 GHz, ramp repetition period 100 microseconds, each frame containing 128 FM-CW cycles, and a frame rate of 100 Hz. Signal processing is handled by a digital signal processor (DSP) and an ARM core integrated within the radar chip. The DSP handles numerical calculations such as Fast Fourier Transform (FFT), while the ARM core handles target identification, tracking filtering, and coordinate mapping.

[0019] See Figures 1 to 4 , Figures 1 to 4 This is a flowchart illustrating the steps of an eye-tracking method based on millimeter-wave radar, as provided in Embodiment 1 of the present invention.

[0020] This invention provides an eye-tracking and detection method based on millimeter-wave radar, comprising: Step S1: Use millimeter-wave radar to transmit frequency-modulated continuous wave signals to the face, collect eye echoes, and generate distance and velocity data. This step includes step S11, where the radar internally generates a linear frequency modulated continuous wave signal, the frequency of which increases linearly with time, and radiates towards the driver's face through the transmitting antenna. Step S12, the receiving antenna collects the echo signal reflected from the facial area, focusing on the eye region. Step S13, the transmitted and received signals are mixed to obtain an intermediate frequency (IF) signal, the frequency of which is proportional to the target distance. Step S14, the IF signal is processed to generate range and velocity data.

[0021] Specifically, in step S141, the intermediate frequency (IF) signal within each frequency-modulated continuous wave (FM) cycle is sampled at equal intervals, with 256 sampling points acquired per cycle, resulting in an IF sampling sequence for one cycle. In step S142, a range-dimensional Fast Fourier Transform (FST) is performed on the IF sampling sequence for each cycle, yielding complex data in 256 range cells. Each range cell corresponds to a radial distance range, and these complex data constitute the range-dimensional data. In step S143, for the same range cell, its complex data over 128 consecutive cycles is extracted to form a sequence. A Doppler-dimensional FFT is then performed on this sequence, yielding complex data in 128 Doppler cells, each corresponding to a radial velocity value. In step S144, the results of the range-dimensional FFT and the Doppler-dimensional FFT are combined to form a two-dimensional complex array of range and velocity data, i.e., range-dimensional data and velocity-dimensional data. Each element in this two-dimensional array reflects the echo intensity at a specific distance and velocity.

[0022] Step S2: Determine candidate distance units from the distance dimension data, extract the phase time series of the candidate distance units and analyze their frequency characteristics. If the frequency characteristics fall within the range of eye movement frequency, they are determined to be eye targets. This step includes step S21, which performs constant false alarm rate (CFAR) detection on the distance dimension data. Specifically, a detection window is slid across the distance dimension, the average power of the background noise within the window is calculated, and the signal amplitude of the detection unit is compared with a threshold calculated by multiplying the average power by a fixed coefficient. If the signal amplitude exceeds the threshold, the distance unit is marked as a candidate distance unit. This process can effectively suppress environmental clutter and highlight the distance range where moving targets exist. Because the eyeball exhibits micro-movements, its echo amplitude is slightly higher than that of static facial tissue; therefore, candidate distance units typically include the distance position of the eyeball.

[0023] Step S22: Acquire echo complex values ​​from candidate range cells across multiple consecutive frames. Since the radar frame rate is 100 Hz, one frame is acquired every 10 milliseconds. For each candidate range cell, the phase component of the echo complex value from each frame is extracted. The phase values ​​from 200 consecutive frames are arranged chronologically to form a phase time series. This series reflects the minute motion of the target on the time axis within that range cell.

[0024] Step S23: Variational mode decomposition (VMD) is used to decompose the phase time series into several intrinsic mode components. VMD is an adaptive signal decomposition method that can decompose non-stationary signals into multiple modes with sparse characteristics. First, step S231 uses the sparrow search algorithm to optimize two key parameters of VMD: the penalty factor and the number of decomposed modes. The sparrow search algorithm simulates the foraging behavior of sparrows and finds the optimal set of parameters through iterative search. Step S232: VMD is performed on the phase time series according to the optimized decomposition parameters to obtain several intrinsic mode components, each corresponding to a center frequency. Step S233: Power spectrum estimation is performed on each intrinsic mode component, and the peak amplitude of each component in the frequency band from 0.5Hz to 5Hz is calculated. The frequencies of human eye saccades, smooth tracking, and micro-jumping gaze all fall within this physiological range. The component with the largest peak amplitude is identified as a candidate component. Step S234: Singular value decomposition is performed on the candidate components, and the singular spectral entropy is calculated. Singular spectral entropy reflects the uncertainty and complexity of a signal. Eye movement signals exhibit strong regularity and low singular spectral entropy, while noise or interference signals are highly random and have high singular spectral entropy. When the singular spectral entropy is below a preset threshold, the candidate component is identified as the dominant component characterizing eye movement. Through a three-stage process of variational mode decomposition, sparrow search optimization, and singular spectral entropy screening, eye movement features can be reliably extracted from a noisy background.

[0025] Step S24: Calculate the energy percentage of the dominant component. Step S241: Calculate the sum of the squared signal strength values ​​of the dominant component over time, as the energy value of the dominant component. Step S242: Calculate the sum of the squared signal strength values ​​of the original phase time series over time, as the total energy value. Step S243: Divide the energy value of the dominant component by the total energy value to obtain the energy percentage. The larger this ratio, the greater the contribution of eye movement to the total echo signal. Step S244: Determine whether the energy percentage is greater than or equal to a preset threshold, which can be set to 0.6. If the condition is met, the target corresponding to the candidate distance unit is determined to be an eye target, completing eye capture. If the energy percentage is lower than the threshold, the candidate unit is determined to be static clutter or other interference and is discarded.

[0026] Step S3: Estimate the angle of the eye target to obtain its three-dimensional coordinates, and use a filter to track and predict the position and velocity of the eye target; This step includes step S31, which utilizes multiple receiving channels of the millimeter-wave radar to acquire the echo signal of the eye target. By comparing the phase difference between different receiving channels, a digital beamforming algorithm is used to calculate the azimuth and elevation angles of the eye target relative to the radar. Since the radar operates at a frequency of 60 GHz and a wavelength of approximately 5 millimeters, its angular resolution is ≥5 degrees, which is sufficient to distinguish between the left and right eyes.

[0027] Step S32: Based on the distance value obtained in step S2, and combined with the azimuth and elevation angles, convert the distance, azimuth, and elevation angles in the spherical coordinate system into three-dimensional spatial coordinates in the radar coordinate system. The conversion formula is: x equals distance multiplied by the cosine of the elevation angle, then multiplied by the cosine of the azimuth angle; y equals distance multiplied by the cosine of the elevation angle, then multiplied by the sine of the azimuth angle; z equals distance multiplied by the sine of the elevation angle. This yields the three-dimensional position of the eye center.

[0028] Step S33: The three-dimensional spatial coordinates calculated at the current moment are used as observation values ​​and input into the pre-constructed filter to update the filter's state estimate. This embodiment uses an extended Kalman filter, whose state vector includes six components: the three-dimensional position and three-dimensional velocity of the eyeball. To accommodate two different motion modes—eye saccades and smooth tracking—this step is further refined into the following sub-steps.

[0029] Step S331: Record the three-dimensional spatial coordinates of the current moment and the previous ten consecutive moments, forming a position sequence of length eleven. Step S332: Input this position sequence into a pre-trained classifier. The classifier uses a support vector machine model, and its training data comes from eye movement samples from different users, labeled as saccade mode or smooth tracking mode. The characteristics of saccade mode are rapid position changes, short duration, and large displacement; the characteristics of smooth tracking mode are uniform speed, smooth trajectory, and long duration. The classifier outputs which mode the current eye movement belongs to. Step S333: If identified as saccade mode, increase the value of the process noise covariance matrix in the filter, typically increasing it to five times its original value. This makes the filter more confident in the observations, thereby improving the response speed to rapid movements. If identified as smooth tracking mode, decrease the process noise covariance value, typically reducing it to one-fifth of its original value, making the filter more confident in the predictions, thereby effectively suppressing position output jitter. Step S334: Using the current three-dimensional spatial coordinates as the observation value, and combining them with the adjusted process noise covariance value, execute the standard extended Kalman filter state update equation to obtain the current state estimate of the eye target, including the three-dimensional position and three-dimensional velocity.

[0030] Step S34: Extract the current position estimate and current velocity estimate of the eye target from the filter's state vector, and output them as the tracking prediction result. This data can be sent to external devices, such as human-machine interface systems or driver monitoring systems, via a general-purpose input / output interface or a serial communication interface.

[0031] This embodiment can capture eye targets without relying on ambient light. The electromagnetic waves emitted by the frequency-modulated continuous wave radar are unaffected by ambient light intensity; the signal-to-noise ratio of the echo signal remains unchanged regardless of whether it is complete darkness or midday sunlight. Simultaneously, 60 GHz millimeter waves can penetrate ordinary glass lenses and light-colored sunglasses because the attenuation of electromagnetic waves in non-metallic media is far lower than the loss of light waves at the lens reflection interface. For physical obstructions such as eyelashes and mask edges, millimeter waves have stronger diffraction capabilities than visible light, and some eye echoes can still be received. At the signal processing level, variational mode decomposition and singular spectral entropy screening mechanisms can separate the dominant component of eye micro-movements from low signal-to-noise ratio echoes, avoiding tracking failures caused by the loss of traditional image features. The adaptive filter dynamically adjusts parameters according to saccade and smooth tracking modes, ensuring both fast response and stable output. Therefore, this embodiment effectively solves the technical problems of existing optical eye-tracking methods, such as sensitivity to external conditions, insufficient robustness in complex dynamic scenes, and difficulty in maintaining stable and reliable tracking performance. Example 2

[0032] Please see Figure 5 , Figure 5 This is a structural block diagram of an eye-tracking system based on millimeter-wave radar provided in Embodiment 3 of the present invention.

[0033] This invention provides an eye-tracking and detection system based on millimeter-wave radar, comprising: a signal acquisition and data generation module, used to transmit frequency-modulated continuous wave signals to the face using millimeter-wave radar, acquire eye echoes, and generate range and velocity data; an eye target acquisition module, used to determine candidate range units from the range data, extract the phase time series of the candidate range units and analyze their frequency characteristics, and if the frequency characteristics fall within the range of physiological eye movement frequencies, it is determined to be an eye target; and a tracking and prediction module, used to estimate the angle of the eye target to obtain three-dimensional coordinates, and use a filter to track and predict the position and velocity of the eye target.

[0034] It should be noted that, since the above is a system corresponding to a millimeter-wave radar-based eye capture and tracking method, its implementation principle is the same as that of a millimeter-wave radar-based eye capture and tracking method. For the sake of convenience and brevity, those skilled in the art can clearly understand that the specific working process of the system and modules described above can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here. Example 3

[0035] Please see Figure 6 , Figure 6 This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention.

[0036] The present invention provides an electronic device comprising: a memory 401 and a processor 402, wherein the memory 401 stores a computer program; when the computer program is executed by the processor 402, the processor 402 performs an eye-capturing and tracking method based on millimeter-wave radar as described in any of the above embodiments.

[0037] Memory 401 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 401 has storage space 403 for program code 413 for performing any of the method steps described above. For example, storage space 403 for program code may include individual program codes 413 for implementing the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When this code is run by a computing device, it causes the device to perform the various steps in the millimeter-wave radar-based eye capture and tracking method described above.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for eye capture and tracking based on millimeter-wave radar, characterized in that, include: Step S1: Use millimeter-wave radar to transmit frequency-modulated continuous wave signals to the face, collect eye echoes, and generate distance and velocity data. Step S2: Determine candidate distance units from the distance dimension data, extract the phase time series of the candidate distance units and analyze their frequency characteristics. If the frequency characteristics fall within the range of 0.5Hz to 5Hz of the physiological eye movement frequency, then it is determined to be an eye target. Step S3: Estimate the angle of the eye target to obtain the three-dimensional coordinates of pitch angle, azimuth angle and distance, and use a filter to track and predict the position and velocity of the eye target.

2. The eye-tracking and detection method based on millimeter-wave radar according to claim 1, characterized in that, Step S1 specifically includes: Step S11: Generate a linear frequency modulated continuous wave signal and radiate the signal toward the human face through a radar transmitting antenna; Step S12: Collect the echo signal reflected back from the eye area through the radar receiving antenna; Step S13: Mix the transmitted signal and the received signal to obtain the intermediate frequency signal; Step S14: Perform range-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform on the intermediate frequency signal to generate range-dimensional data and velocity-dimensional data.

3. The eye-tracking and detection method based on millimeter-wave radar according to claim 2, characterized in that, Step S14 specifically includes: Step S141: Sample the intermediate frequency signal within each frequency-modulated continuous wave cycle to obtain an intermediate frequency sampling sequence for one cycle; Step S142: Perform the distance dimension fast Fourier transform on the intermediate frequency sampling sequence to obtain complex data in each distance unit, which constitutes the distance dimension data; Step S143: Perform the Doppler Fast Fourier Transform on the complex data sequences of the same distance unit at different periods to obtain complex data for each distance and velocity unit; Step S144: Combine the results of the distance-dimensional fast Fourier transform with the results of the Doppler-dimensional fast Fourier transform to generate the distance-dimensional data and velocity-dimensional data.

4. The eye-tracking and detection method based on millimeter-wave radar according to claim 1, characterized in that, Step S2 specifically includes: Step S21: Perform constant false alarm rate detection on the distance dimension data and extract distance units whose signal amplitude exceeds the threshold as candidate distance units; Step S22: Collect echo complex values ​​on the candidate range unit for multiple consecutive frames, extract the phase component of each frame, and form a phase time series; Step S23: Use variational mode decomposition to decompose the phase time series into several intrinsic mode components, perform power spectrum estimation on each component, and screen out the dominant components whose spectral peaks are located in the 0.5Hz to 5Hz frequency band. Step S24: Calculate the energy percentage of the dominant component. When the energy percentage exceeds a preset threshold, determine that the target corresponding to the candidate distance unit is the eye target.

5. The eye-tracking and detection method based on millimeter-wave radar according to claim 4, characterized in that, Step S23 specifically includes: Step S231: Optimize the penalty factor and number of decomposition modes of the variational mode decomposition using the sparrow search algorithm to obtain the optimal decomposition parameters; Step S232: Perform variational mode decomposition on the phase time series according to the optimal decomposition parameters to obtain several intrinsic mode components; Step S233: Perform power spectrum estimation for each intrinsic mode component, calculate the peak amplitude of each component in the frequency band from 0.5Hz to 5Hz, and identify the component with the largest peak amplitude as a candidate component. Step S234: Perform singular value decomposition on the candidate component and calculate the singular spectral entropy. When the singular spectral entropy is lower than a preset threshold, determine the candidate component as the dominant component of eye movement.

6. The eye-tracking and detection method based on millimeter-wave radar according to claim 4, characterized in that, Step S24 specifically includes: Step S241: Calculate the sum of the squared values ​​of the signal strength of the dominant component on the time axis, and use it as the energy value of the dominant component; Step S242: Calculate the cumulative sum of the squared signal strength values ​​of the phase time series on the time axis, as the total energy value; Step S243: Divide the energy value of the dominant component by the total energy value to obtain the energy percentage of the dominant component; Step S244: Determine whether the energy percentage is greater than or equal to a preset threshold. If so, determine that the target corresponding to the candidate distance unit is the eye target.

7. The eye-tracking and detection method based on millimeter-wave radar according to claim 1, characterized in that, Step S3 specifically includes: Step S31: Use multiple receiving channels of millimeter-wave radar to acquire the echo signal of the eye target, and calculate the azimuth and elevation angles of the eye target relative to the radar by phase comparison; Step S32: Based on the distance value obtained in step S2, and combined with the azimuth and elevation angles, convert the distance, azimuth, and elevation angles into three-dimensional spatial coordinates in the radar coordinate system. Step S33: Use the three-dimensional spatial coordinates calculated at the current time as the observation value and input them into the pre-built filter to update the state estimate of the filter; Step S34: Extract the current position estimate and current velocity estimate of the eye target from the state vector of the filter, and output them as the tracking prediction result.

8. The eye-tracking and detection method based on millimeter-wave radar according to claim 7, characterized in that, Step S33 specifically includes: Step S331: Record the three-dimensional spatial coordinates of the current moment and multiple consecutive moments before, forming a position sequence; Step S332: Input the position sequence into a pre-trained classifier to identify whether the current eye movement belongs to saccade mode or smooth tracking mode; Step S333: If the scanning mode is identified, increase the process noise covariance value in the filter; if the smooth tracking mode is identified, decrease the process noise covariance value in the filter. Step S334: Using the current three-dimensional spatial coordinates as the observation value, and combining them with the adjusted process noise covariance value, perform filter state update to obtain the current state estimate of the eye target.

9. An eye-tracking and detection system based on millimeter-wave radar, characterized in that, include: The signal acquisition and data generation module is used to transmit frequency-modulated continuous wave signals to the face using millimeter-wave radar, collect eye echoes, and generate distance and velocity data. An eye target capture module is used to determine candidate distance units from the distance dimension data, extract the phase time series of the candidate distance units and analyze their frequency characteristics. If the frequency characteristics fall within the range of physiological eye movement frequency, it is determined to be an eye target. The tracking and prediction module is used to estimate the angle of the eye target to obtain its three-dimensional coordinates, and to use a filter to track and predict the position and velocity of the eye target.

10. An electronic device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the eye-capture and tracking method based on millimeter-wave radar as described in any one of claims 1-8.

Images