Parkinson's gait evaluation method and system based on millimeter wave radar signal

Abstract

The invention discloses a Parkinson's gait assessment method and system based on millimeter wave radar signals. The method comprises the following steps: acquiring a millimeter wave radio frequency signal in an environment, and performing data processing on the signal to obtain a distance spectrum of the change of the position of the body of the Parkinson patient and a Doppler spectrum reflecting the change of the physical degree of the Parkinson patient; extracting a peak value of the distance spectrum at each moment through a filtering method, calculating a cross correlation coefficient of a Doppler spectrum as an index of gait mode matching, and segmenting the Doppler spectrum according to each step; gait feature calculation is carried out according to the filtered distance spectrum peak value and the segmented Doppler spectrum; and normalizing the gait features obtained by calculation, and inputting the normalized gait features into the trained machine learning model for feature evaluation to obtain the UPDRS-III score of the patient. According to the method, non-invasive measurement and evaluation of the Parkinson's gait characteristics are achieved in a low-cost mode, operation is convenient, the evaluation effect is accurate, and quantitative data and diagnosis reference are provided for doctors.

Description

Technical Field

[0001] The present invention relates to the field of gait assessment, and in particular to a method and system for assessing Parkinson's gait based on millimeter-wave radar signals. Background Art

[0002] In the traditional Parkinson's gait assessment, doctors use phenotypic tests to observe and evaluate patients' behavioral characteristics, motor skills, posture and other related symptoms to evaluate Parkinson's gait. However, phenotypic tests rely on professional doctors' subjective and qualitative evaluation of Parkinson's symptoms, which has the disadvantages of low accuracy, large errors, and easy to be affected by doctors' psychological factors. Therefore, in recent years, this field has been seeking to introduce precise and quantitative data collection and analysis mechanisms to assist doctors in gait assessment. The number of application systems based on various sensors has grown rapidly, showing a booming trend. For example, the technology based on the three-dimensional motion capture system can track the position information of each joint of the body of Parkinson's patients. The changes in the position of human joints can quantify the entire walking process to achieve the extraction of gait characteristics. The technology based on inertial sensors can track the speed, acceleration, and angular velocity information of specific parts of the body of Parkinson's patients such as feet, calves, wrists, waists, etc. during walking, and calculate the gait characteristics such as step length and speed. The technology based on pressure sensors measures the changes in plantar pressure during the patient's walking process by installing special pressure sensing pads or requiring patients to wear pressure sensing insoles, and indirectly obtains gait characteristics by analyzing the changes in pressure.

[0003] Although existing application systems can already obtain relatively accurate gait characteristics, these systems are inconvenient to use: the use of 3D motion capture systems requires patients to wear tights with special markings, the use of inertial sensors requires patients to wear related equipment at the joints of the body, and the use of pressure-sensing pads or pressure-sensing insoles requires patients to change into special shoes. Parkinson's patients are usually elderly people. Wearing additional equipment is a burden and affects their normal walking, and even causes some patients to feel fear. On the other hand, due to the high price and complicated installation and use process, these devices are only suitable for one-time gait assessments in special examination rooms in hospitals, and cannot help doctors conduct long-term assessments of patients' gait.

[0004] Millimeter wave radar is a radar technology that uses millimeter wave frequency band radio frequency signals for detection and perception. It has the characteristics of high precision, high resolution and little environmental interference. In recent years, with the development of mobile computing and millimeter wave technology, millimeter wave radar has been widely used in the commercial field. However, the existing gait assessment technology based on millimeter wave radar signals is still stuck in measuring the gait characteristics of ordinary users. Not only does it not take into account the problem that Parkinson's patients have significantly different gait characteristics from normal people due to symptoms such as muscle rigidity, bradykinesia, and unstable posture; it also does not conduct further gait assessment and analysis based on the obtained gait characteristics. Summary of the invention

[0005] Aiming at the weaknesses of the existing millimeter-wave radar gait analysis, the present invention proposes a Parkinson's gait assessment method and system based on millimeter-wave radar signals, which solves the problem of how to obtain accurate abnormal gait characteristics of Parkinson's patients through millimeter-wave radar transmitting and receiving arrays, and uses machine learning methods to evaluate the UPDRS-III gait scores of Parkinson's patients.

[0006] In order to achieve the above invention object, the technical solution of the present invention is as follows:

[0007] In a first aspect, a method for evaluating Parkinson's gait based on millimeter-wave radar signals comprises the following steps:

[0008] Acquire millimeter wave radio frequency signals in the environment, and process the signals to obtain a distance spectrum reflecting the changes in the position of the Parkinson's patient's body and a Doppler spectrum reflecting the changes in the Parkinson's patient's body speed;

[0009] The peak value of each moment is extracted from the distance spectrum by filtering method, the cross-correlation coefficient of Doppler spectrum is calculated as the index of gait pattern matching, and the Doppler spectrum is divided according to each step;

[0010] Gait features are calculated based on the filtered range spectrum peak and segmented Doppler spectrum;

[0011] The calculated gait features are normalized and input into the trained machine learning model for feature evaluation to obtain the patient's UPDRS-III score.

[0012] Preferably, the millimeter wave radio frequency signal in the environment is a frequency modulated continuous wave signal jointly emitted by multiple transmitting antennas of the transmitting antenna array, and the emitted signal is received by the millimeter wave receiving antenna array.

[0013] Preferably, the signal is processed to obtain a distance spectrum reflecting the change in the position of the Parkinson's patient's body and a Doppler spectrum reflecting the change in the Parkinson's patient's body speed, including:

[0014] The received signals are averaged to obtain the static components generated by the objects in the environment, and the static components are removed from the received signals to obtain a signal containing only the body movements of the Parkinson's patient;

[0015] For the signal with the static component removed, the phase difference between the signals received by multiple antennas in the receiving antenna array is used to supplement the signal received by a single antenna, thereby achieving multi-channel signal enhancement;

[0016] The enhanced signal is subjected to fast Fourier transform to obtain a distance spectrum reflecting the change in the position of the Parkinson's patient's body and a Doppler spectrum reflecting the change in the speed of the Parkinson's patient's body.

[0017] Preferably, the peak value at each moment is extracted from the distance spectrum by a filtering method, including:

[0018] Use average filtering on the distance spectrum to replace the distance spectrum at each moment with the average value of the distance spectrum energy within a time window of a specified length;

[0019] The position corresponding to the point with the largest distance spectrum energy at each moment is extracted to obtain the patient's position change curve. The curve is smoothed and outliers are eliminated using the Hempel filter and the average filter, and finally the filtered distance spectrum peak representing the patient's location information is obtained.

[0020] Preferably, the cross-correlation coefficient of the Doppler spectrum is calculated as an indicator of gait pattern matching, and the Doppler spectrum is segmented according to each step, including:

[0021] The Doppler spectrum is regarded as a two-dimensional image, and the cross-correlation coefficient of the image is calculated as the cross-correlation coefficient of the Doppler spectrum;

[0022] Determine a time range according to the reference gait cycle, and within the time range, find the Doppler spectrum of the next step with the largest cross-correlation coefficient with the Doppler spectrum of the current step to determine the gait cycle of the current step;

[0023] According to the determined gait cycle, the Doppler spectrum is segmented for each step.

[0024] Preferably, the gait feature calculation is performed according to the filtered range spectrum peak and the segmented Doppler spectrum, including:

[0025] The start and end time of each gait cycle are obtained according to the segmented Doppler spectrum, and the distance from the patient to the radar at the start and end time of each gait cycle is obtained according to the peak value of the filtered distance spectrum, so as to obtain the stride of each gait cycle;

[0026] The start time and end time of each gait cycle are subtracted to obtain the step time information, and the average step speed in each gait cycle is calculated based on the stride length and step time.

[0027] Preferably, the machine learning model adopts the XGBoost model.

[0028] In a second aspect, a Parkinson's gait assessment system based on millimeter wave radar signals is provided, comprising:

[0029] A transmitting antenna array, used for transmitting a frequency modulated continuous wave signal;

[0030] A receiving antenna array for receiving millimeter wave radio frequency signals in the environment;

[0031] A computing device is used to implement gait assessment of Parkinson's disease patients according to the Parkinson's disease gait assessment method based on millimeter-wave radar signals described in the first aspect.

[0032] In a third aspect, a computing device is provided, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the Parkinson's gait assessment method based on millimeter-wave radar signals as described in the first aspect are implemented.

[0033] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the method for assessing Parkinson's gait based on millimeter-wave radar signals as described above are implemented.

[0034] Beneficial effects: In view of the defects of the existing millimeter wave radar gait analysis technology, the present invention proposes to perform data enhancement on the frequency modulated continuous wave signal through the millimeter wave radar array, and then extract the specific Parkinson's gait features according to the micro-Doppler spectrum similarity, and use machine learning methods to evaluate the UPDRS-III gait score of Parkinson's patients. The present invention can realize Parkinson's gait feature measurement and UPDRS-III gait score evaluation on a general millimeter wave radar hardware platform through a simple software algorithm. The present invention can realize non-invasive measurement and evaluation of Parkinson's gait characteristics in a low-cost manner, providing doctors with quantitative data and diagnostic references, while the entire data collection process is also very convenient for Parkinson's patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a schematic diagram of the application scenario of Parkinson's gait assessment based on millimeter-wave radar signals.

[0036] Figure 2 It is an implementation flow chart of the Parkinson's gait assessment method based on millimeter-wave radar signals.

[0037] Figure 3 It is a flow chart of the method for generating data features.

[0038] Figure 4 is a flow chart of the method for assessing gait characteristics.

[0039] Figure 5 is an example of a Doppler spectrum segmented according to the gait cycle.

[0040] Figure 6 This is an example of the assessment of UPDRS-III stepping scores in a day for a Parkinson's patient. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0042] The existing gait assessment technology based on millimeter-wave radar signals is still limited to measuring the gait characteristics of ordinary users, without considering the abnormal gait of Parkinson's patients. Parkinson's patients' abnormal gaits such as dragging their feet and lack of hand-leg coordination are also fundamental problems that limit the general millimeter-wave radar signal gait assessment scheme. The present invention proposes a feature extraction algorithm for abnormal gait, and uses the extracted features to correctly assess the severity of Parkinson's gait. First, the multi-receiving antenna of the millimeter-wave radar is used to enhance the signal and improve the accuracy of feature extraction. A GPMR (Gait Pattern Match Rate) indicator for evaluating the similarity of micro-Doppler images is designed to achieve feature extraction of abnormal gait; then, a machine learning-based method is used to evaluate the UPDRS-III gait score of Parkinson's patients through gait features.

[0043] Figure 1 This is an application scenario of the present invention. The scenario includes a millimeter wave transmitting antenna array 102. The millimeter wave transmitting antenna array 102 is composed of three antennas with a spacing of 5 mm. The transmitted signal is a radio frequency signal of 77 GHz to 81 GHz, and it is transmitted 2000 times per second. The scenario includes a millimeter wave receiving antenna array 103. The millimeter wave receiving antenna array 103 is composed of three antennas with a spacing of 2.5 mm, and receives the signal transmitted by the millimeter wave transmitting antenna array 102. The millimeter wave receiving antenna array 103 transmits the received signal to the computing device 108, and the computing device 108 can obtain the distance spectrum and Doppler spectrum of the radio frequency signal through data processing, and further obtain the location and speed information of different objects. Figure 1As shown, there are two reflection paths from the millimeter wave transmitting antenna array 102 to the millimeter wave receiving antenna array 103: the environmental reflection path 104 is a path that reaches the millimeter wave receiving antenna array 103 after being reflected by the indoor object 101, and the target reflection path 105 is a path that reaches the millimeter wave receiving antenna array 103 after being reflected by the body 106 of the Parkinson's patient who is walking. The indoor object 101 is stationary, so the environmental reflection path 104 will not change, and the influence of the indoor object 101 on the received signal is unchanged. When the Parkinson's patient's body 106 is walking, the target reflection path 105 is constantly changing, which in turn causes the received signal to have a signal strength distribution and phase change with the movement of the Parkinson's patient's body 106. The present invention removes the influence of the indoor object 101 by removing the static component of the received signal, and enhances the influence of the Parkinson's patient's body 106 on the received signal through multiple antenna arrays, and extracts and evaluates the gait features through the distance spectrum and Doppler spectrum.

[0044] The present invention is characterized in that the entire data collection process is not disturbed by stationary objects in the environment, nor is it affected by light, noise, etc. The Parkinson's patient body 106 being tested does not need to wear any additional equipment. The algorithm for extracting the gait characteristics of Parkinson's patients is robust to Parkinson's patients with abnormal gait. The transmitting antenna array 102 and the receiving antenna array 103 are both general commercial devices, which only need to support the transmission of 77GHz to 81GHz radio frequency signals and support a sampling rate of 5000kps.

[0045] Figure 2 201 is a flowchart of the method of the present invention. First, gait data is collected, millimeter wave transmission is performed in step 201, and then data collection is performed in step 202. The collected millimeter wave radio frequency signal will be used for data feature generation, and the collected signal is processed in step 203 to obtain the distance spectrum and Doppler spectrum, and the processed data is extracted in step 204 to obtain the peak information of the distance spectrum and the speed information of the Doppler spectrum. Finally, gait feature extraction and evaluation are performed based on the data features, and gait features such as stride and pace are extracted in step 205, and then gait feature evaluation is performed in step 206 to obtain the UPDRS-III gait score. It is worth noting that the above process is only a possible implementation method of the present invention, and the Parkinson's gait assessment system can also be implemented in various other ways.

[0046] In step 201, the millimeter wave radio frequency signal transmission is realized by the transmitting antenna array 102, and the transmitted signal is a frequency modulated continuous wave (FMCW) signal jointly emitted by multiple transmitting antennas. Specifically, a frequency modulated continuous wave with a frequency of 77 GHz to 81 GHz is transmitted, with a repetition period of 0.5 milliseconds and 256 sampling points per period. Of course, the frequency modulated continuous wave frequency and bandwidth used can be other values, which need to be set according to the desired accuracy of the final calculation.

[0047] After data collection in step 202 , the collected raw data needs to be processed (corresponding to step 203 ) to obtain data features (corresponding to step 204 ), including the range spectrum peak and the Doppler spectrum segmented according to gait. Figure 3 This is the implementation process of data feature generation in the present invention. Figure 3 As shown, first, in step 301, regarding the influence of the indoor object 101 on the received signal, since the indoor object 101 remains stationary, its influence on the received signal remains unchanged; while the body 106 of the Parkinson's patient undergoes a process of stationary-moving-stationary during the process of collecting gait data, and the sum of the generated acceleration dynamic component and deceleration dynamic component is 0. Therefore, the received signal is averaged to obtain the static component generated by the indoor object 101, and the static component is removed from the received signal to obtain a signal containing only the movement of the Parkinson's patient's body 106. While removing the influence of objects in the environment, the device is made robust to different deployment environments. Then, in step 302, the signal with the static component removed is subjected to multi-channel signal enhancement, and the phase difference between the signals received by multiple antennas in the receiving antenna array 103 is used to supplement the signal received by a single antenna, thereby achieving signal enhancement without changing the hardware and detection conditions. Next, in step 303, the enhanced signal is subjected to a Fast Fourier Transform (FFT) operation to obtain a range profile reflecting the position change of the Parkinson's patient's body 106 and a Doppler profile reflecting the speed change of the Parkinson's patient's body 106. After removing the static component, the Parkinson's patient's body 106 is the main reflection source. In step 304, the peak value at each moment is extracted from the range spectrum to obtain the position information of the target; in step 305, the Doppler spectrum is matched by GPMR (Gait Pattern Match Rate), the two-dimensional correlation coefficient of the Doppler spectrum is calculated, and the Doppler spectrum is separated according to each step. According to the peak information of the range spectrum and the segmented Doppler spectrum, the data features are obtained.

[0048] In step 302, the present invention uses signals from different receiving antennas to perform multi-channel signal enhancement, and can use the phase difference between the signals received by multiple antennas to supplement the signal received by a single antenna. First, there is a 2.5 mm interval between the receiving antennas, which causes the transmission time of the signal received by each antenna to be different; however, 2.5 mm is half a wavelength, which makes the signal received by each antenna basically the same. Specifically, the amplitudes of the signals received by different antennas are almost the same, but the phases are quite different. Considering the RF signal as a far field, the phase difference of the signal received by each antenna can reflect the delay of the signal reaching each antenna. By eliminating this delay and adding the signals received by multiple antennas, multi-channel signal enhancement can be achieved.

[0049] The present invention uses the distance spectrum to extract the peak value at each moment in step 304, and the position change information of the Parkinson's patient can be obtained. Because the position of each point on the distance spectrum corresponds to the radial distance from the position of the object in reality to the radar, the position of the peak of the distance spectrum corresponds to the position of the patient's body in reality. First, the distance spectrum is averaged. For the distance spectrum at each moment, the energy of each position is the average value of the distance spectrum energy in a window with a length of 5. This can prevent the patient from swinging his arms when he is very close to the radar, causing the maximum energy to appear in a non-trunk position. Then, the position corresponding to the point with the maximum energy of the distance spectrum at each moment is extracted to obtain the patient's position change curve. The Hampel filter and the average filter are used to smooth the curve and exclude outliers, and finally the filtered distance spectrum peak representing the patient's position information is obtained. Using the filtered distance spectrum peak can prevent the problem of misidentifying the position of the patient's hand as the patient's body when the patient is very close to the radar.

[0050] The present invention uses GPMR matching to segment the Doppler spectrum in step 307, and can obtain the start and end time of each step of the patient and the speed change information of the patient during this period. First, the walking process is a repetitive action. Even if Parkinson's patients have abnormal gait problems, the walking speed of the patient is still periodic. Then, by treating the Doppler spectrum as a two-dimensional image, and using the change of the cross-correlation coefficient of the image to reflect the periodicity of the patient's gait. This can prevent the problem of not being able to accurately extract the patient's leg speed when extracting the speed; at the same time, the cross-correlation coefficient is not sensitive to the starting speed of the Doppler spectrum of each step. Using the cross-correlation coefficient as an indicator of GPMR matching can ensure that the gait cycle with a slower speed when the patient just starts walking and the faster gait cycle when walking at a uniform speed can be equally applicable. In the specific matching process, a reference gait cycle duration of 0.5 seconds is first determined, and then a search is performed within the range of 0.35 seconds to 0.9 seconds to find the Doppler spectrum of the next step with the largest cross-correlation coefficient with the Doppler spectrum of the current step, and the gait cycle of the current step can be determined.

[0051] Figure 4 This is the implementation process of gait feature evaluation in the present invention. Figure 4 As shown, first, in step 401, the gait feature calculation is performed by combining the filtered range spectrum peak and the segmented Doppler spectrum, thereby obtaining the gait features such as the stride and speed of the detection target. Then, in step 402, the obtained gait features are normalized, and finally in step 403, the normalized features are input into the trained machine learning model for feature evaluation to obtain the patient's UPDRS-III score.

[0052] The present invention performs gait feature calculation in step 401. First, the start time and end time of each gait cycle can be obtained based on the segmented Doppler spectrum, and the step time information can be obtained by subtracting them. Then, the curve is smoothed using a Hample filter and an average filter and outliers are eliminated to obtain the filtered distance spectrum peak, as described above. The actual distance from the patient to the radar can be obtained by the formula: index of the distance spectrum peak × speed of light ÷ (2 bandwidth), and the stride of each gait cycle can be obtained by combining the start time and end time of each gait cycle of the patient. Finally, the average walking speed in each gait cycle can be obtained by (stride ÷ step time).

[0053] The present invention performs feature normalization in step 402. Feature normalization can ensure that different gait features such as stride and pace, although in different units, are distributed in the range of 0 to 1, so that the machine learning model can correctly evaluate each feature. In step 403, the normalized gait features are used for feature evaluation. A pre-trained XGBoost model is used to evaluate the UPDRS-III gait scores of Parkinson's patients. Compared with deep learning models, XGBoost has less demand for computing power, faster derivation speed, and does not require a large-scale gait feature data set for training, which is suitable for Parkinson's gait evaluation scenarios. At the same time, the XGBoost model can use regression results as output. The output result of the feature evaluation can be a number such as 2.3 or 1.6, which has higher accuracy than the 1 point, 2 points or 3 points obtained by doctors observing patients with the naked eye and referring to the UPDRS-III scale.

[0054] Figure 5 It is an example of the Doppler spectrum segmented according to the gait cycle in the present invention. The horizontal axis is time, the vertical axis is speed, and the color depth of each pixel represents the magnitude of the patient's body reflected energy at the corresponding position. It can be seen from the figure that although the abnormal gait of Parkinson's patients will cause their gait cycle to be unstable, and there is a phenomenon that the speed is faster in some cycles and slower in other cycles, the GPMR matching algorithm based on the mutual correlation coefficient can ensure the correct segmentation of abnormal Parkinson's gait.

[0055] Figure 6 This is an example of evaluating the UPDRS-III gait score of Parkinson's patients in one day. By measuring the gait characteristics for a long time and multiple times, and performing characteristic evaluation, the fluctuation of the UPDRS-III gait score of Parkinson's patients in one day is obtained. It can be seen that when Parkinson's patients take drugs, the results of the characteristic evaluation will immediately decrease; and as time goes by, the efficacy of the drug gradually disappears, and the results of the characteristic evaluation will gradually increase. This shows that the characteristic evaluation can correctly reflect the UPDRS-III gait condition of Parkinson's patients.

[0056] The present invention also provides a computing device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the Parkinson's gait assessment method based on millimeter-wave radar signals as described in the first aspect are implemented.

[0057] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the Parkinson's gait assessment method based on millimeter-wave radar signals as described above are implemented.

[0058] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for assessing Parkinson's gait based on millimeter wave radar signals, characterized in that: The following steps are involved: Acquire millimeter wave radio frequency signals in the environment, and process the signals to obtain a distance spectrum reflecting the changes in the position of the Parkinson's patient's body and a Doppler spectrum reflecting the changes in the Parkinson's patient's body speed; The peak value of each moment is extracted from the distance spectrum by filtering method, the cross-correlation coefficient of Doppler spectrum is calculated as the index of gait pattern matching, and the Doppler spectrum is divided according to each step; Gait features are calculated based on the filtered range spectrum peak and segmented Doppler spectrum; The calculated gait features are normalized and input into the trained machine learning model for feature evaluation to obtain the patient's UPDRS-III score.

2. The method according to claim 1, characterized in that The millimeter wave radio frequency signal in the environment is a frequency modulated continuous wave signal jointly emitted by multiple transmitting antennas of the transmitting antenna array, and the transmitted signal is received by the millimeter wave receiving antenna array.

3. The method according to claim 2, characterized in that The signal is processed to obtain the distance spectrum of the change of the position of the Parkinson's patient's body and the Doppler spectrum reflecting the change of the Parkinson's patient's body, including: The received signals are averaged to obtain the static components generated by the objects in the environment, and the static components are removed from the received signals to obtain a signal containing only the body movements of the Parkinson's patient; For the signal with the static component removed, the phase difference between the signals received by multiple antennas in the receiving antenna array is used to supplement the signal received by a single antenna, thereby achieving multi-channel signal enhancement; The enhanced signal is subjected to fast Fourier transform to obtain a distance spectrum reflecting the change in the position of the Parkinson's patient's body and a Doppler spectrum reflecting the change in the speed of the Parkinson's patient's body.

4. The method according to claim 1, characterized in that: The peak value of each moment is extracted from the distance spectrum by filtering method, including: Use average filtering on the distance spectrum to replace the distance spectrum at each moment with the average value of the distance spectrum energy within a time window of a specified length; The position corresponding to the point with the largest distance spectrum energy at each moment is extracted to obtain the patient's position change curve. The curve is smoothed and outliers are eliminated using the Hempel filter and the average filter, and finally the filtered distance spectrum peak representing the patient's location information is obtained.

5. The method according to claim 1, characterized in that The cross-correlation coefficient of the Doppler spectrum is calculated as an indicator of gait pattern matching, and the Doppler spectrum is segmented according to each step, including: The Doppler spectrum is regarded as a two-dimensional image, and the cross-correlation coefficient of the image is calculated as the cross-correlation coefficient of the Doppler spectrum; Determine a time range according to the reference gait cycle, and within the time range, find the Doppler spectrum of the next step with the largest cross-correlation coefficient with the Doppler spectrum of the current step to determine the gait cycle of the current step; According to the determined gait cycle, the Doppler spectrum is segmented for each step.

6. The method according to claim 1, characterized in that The gait features are calculated based on the filtered range spectrum peak and segmented Doppler spectrum, including: The start and end time of each gait cycle are obtained according to the segmented Doppler spectrum, and the distance from the patient to the radar at the start and end time of each gait cycle is obtained according to the peak value of the filtered distance spectrum, so as to obtain the stride of each gait cycle; The start time and end time of each gait cycle are subtracted to obtain the step time information, and the average step speed in each gait cycle is calculated based on the stride length and step time.

7. The method according to claim 1, characterized in that The machine learning model uses the XGBoost model.

8. A Parkinson's gait assessment system based on millimeter wave radar signals, characterized in that: include: A transmitting antenna array, used for transmitting a frequency modulated continuous wave signal; A receiving antenna array for receiving millimeter wave radio frequency signals in the environment; A computing device, used for implementing gait assessment of Parkinson's disease patients according to the Parkinson's disease gait assessment method based on millimeter-wave radar signals according to any one of claims 1 to 7.

9. A computing device, characterized in that include: one or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the Parkinson's gait assessment method based on millimeter-wave radar signals as described in any one of claims 1-7 are implemented.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for assessing Parkinson's gait based on millimeter-wave radar signals as described in any one of claims 1 to 7 are implemented.

Images