4D millimeter wave radar based static target deformation monitoring and power failure self-recovery method

Abstract

This invention discloses a static target deformation monitoring and power-off self-recovery method based on 4D millimeter-wave radar, relating to the fields of radar signal processing and intelligent monitoring technology. The method includes: automatically identifying and locking a preset static corner reflector using a movable dynamic auxiliary corner reflector, extracting its four-dimensional feature parameters: distance, horizontal angle, pitch angle, and relative velocity; sending these parameters as calibration reference points to the 4D millimeter-wave radar, and writing them into a non-volatile storage unit after high-computation matching within the local search space; continuously acquiring multiple frames of point cloud data, filtering neighboring strong reflection points, and obtaining a merged distance value through median filtering; calculating the cumulative change of the current high-precision distance value relative to the initial calibration value using a super-resolution ranging algorithm to obtain the sub-millimeter-level deformation (DEF); and automatically loading the calibration parameters from the non-volatile storage unit after a power outage and restart, re-matching and activating the target, and resuming real-time monitoring.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing and intelligent monitoring technology, and more specifically to a static target deformation monitoring and power-off self-recovery method based on 4D millimeter-wave radar. Background Technology

[0002] With the rapid development of infrastructure construction, the number of engineering projects such as mountain highways, railways, reservoirs, and mining areas is constantly increasing, making slope stability a critical aspect of engineering safety management. Under the combined effects of long-term natural erosion, rainfall, earthquakes, and human disturbance, slopes are highly susceptible to deformation and even landslides, posing a serious threat to people's lives and property. Therefore, higher demands are placed on real-time monitoring and early warning technologies for slope deformation.

[0003] Currently, slope deformation monitoring technologies mainly include GNSS (Global Navigation Satellite System) measurement, total station measurement, inclinometer monitoring, fiber optic grating monitoring, and InSAR (Interferometric Synthetic Aperture Radar) technology. However, these methods generally have significant limitations in practical applications: GNSS and total stations are greatly affected by weather and obstructions, making it difficult to achieve continuous monitoring around the clock; fiber optic sensors require the laying of a large number of cables, resulting in high installation and maintenance costs and susceptibility to environmental damage; while InSAR can cover a wide area, its low temporal resolution makes it unable to capture sudden, minute displacements, and its data processing is complex and has high latency.

[0004] Millimeter-wave radar, with its advantages of all-weather, all-day operation, high precision, and non-contact measurement, is gradually being introduced into the field of infrastructure health monitoring. However, existing millimeter-wave radar-based monitoring systems still face the following technical bottlenecks:

[0005] Static target identification is difficult: In complex slope environments, natural or artificial clutter such as rocks, guardrails, and vegetation can produce strong reflections, making it difficult for radar to accurately identify the preset static corner reflectors from massive point clouds.

[0006] Insufficient deformation monitoring accuracy: Due to the limitations of radar signal bandwidth and traditional FFT processing methods, the range resolution of conventional FMCW radar is usually only at the centimeter level (e.g., the theoretical resolution is 3.75 cm under a 4 GHz bandwidth), which cannot meet the early warning requirements for early slope slip (often at the sub-millimeter to millimeter level).

[0007] The system has poor robustness and lacks autonomous recovery capability: Existing solutions usually store calibration parameters on a host computer or in the cloud. Once the radar loses power and restarts, the reference information needs to be re-sent and manually reset. This makes it impossible to achieve seamless continuation of monitoring tasks in unattended scenarios, which seriously affects the continuity and reliability of long-term monitoring.

[0008] Therefore, there is an urgent need for a 4D millimeter-wave radar monitoring method that can autonomously identify static cooperative targets, achieve sub-millimeter-level high-precision deformation monitoring, and automatically restore the monitoring state after power failure, in order to solve the shortcomings of existing technologies in terms of automation, accuracy, and robustness. Summary of the Invention

[0009] In view of the above problems, this invention is proposed to provide a static target deformation monitoring and power-off self-recovery method based on 4D millimeter-wave radar to overcome or at least partially solve the above problems. This invention requires no manual intervention, has strong anti-environment interference capability, and has power-off self-recovery capability. It can realize all-weather, high-precision, continuous and reliable deformation monitoring and is suitable for safety early warning scenarios of key infrastructure such as slopes, bridges, and dams.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] In a first aspect, embodiments of the present invention provide a method for static target deformation monitoring and power-off self-recovery based on 4D millimeter-wave radar, comprising the following steps:

[0012] S1. Identify static corner reflectors: Use dynamic auxiliary targets to identify and lock onto static corner reflectors, determine their precise location, and extract four-dimensional feature parameters;

[0013] S2. Calibrate and store the static corner reflector: send the four-dimensional feature parameters as calibration reference point information to the 4D millimeter-wave radar; the 4D millimeter-wave radar performs high-performance matching analysis and consistency analysis in the local search space with the calibration reference point as the center; when the matching result meets the preset judgment conditions, the four-dimensional feature parameters are written into the non-volatile storage unit inside the radar.

[0014] S3. Merge multiple target data points generated by a single static corner reflector: Continuously collect multi-frame point cloud data returned by 4D millimeter-wave radar, filter out the set of strong reflection points in the neighborhood of the calibration reference point based on the calibration reference point defined by the four-dimensional feature parameters stored in the non-volatile storage unit, and filter the distance data at the same time to obtain the merged distance value of the static corner reflector at the current time.

[0015] S4. Displacement deformation monitoring based on the merged distance value: Using the distance value corresponding to the four-dimensional feature parameters stored in the non-volatile storage unit as the initial calibration value, calculate the cumulative change of the high-precision distance value at the current moment relative to the initial calibration value to obtain the deformation DEF;

[0016] S5. Achieve autonomous recovery of monitoring data after power failure: After the 4D millimeter-wave radar is powered on again, it retrieves all stored four-dimensional feature parameters from the non-volatile storage unit, loads them one by one, and repeatedly executes the high-computing-power matching analysis of the local search space range in S2; if the matching is successful, the corresponding target point is activated and its real-time deformation monitoring is restored; after all target points have been matched and activated, the system automatically enters the normal monitoring mode.

[0017] Preferably, the specific implementation of S1 includes the following process:

[0018] S101: The 4D millimeter-wave radar and static corner reflector have been deployed in the field and are both in a stationary state.

[0019] S102: After the 4D millimeter-wave radar is started, it transmits FMCW frequency-modulated continuous wave signals, collects and generates point cloud data containing static environmental targets, and uses it as a static background reference.

[0020] S103: Introducing a movable dynamic auxiliary corner reflector as a dynamic auxiliary target, which moves laterally left and right and radially near and far within the monitoring range of the 4D millimeter-wave radar, and gradually approaches the static corner reflector.

[0021] S104: Based on continuous frame point cloud data, perform spatial correlation analysis between the motion trajectory of the dynamic auxiliary corner reflector and the strong reflection points in the static background reference:

[0022] When a static strong reflection point is identified in continuous frame point cloud data, its relative velocity is zero, the spatial distance between it and the dynamic auxiliary corner reflector decreases as the dynamic auxiliary corner reflector moves, and its echo signal strength increases accordingly, the static strong reflection point is determined to be the static corner reflector, and its position is determined.

[0023] S105: Extract the four-dimensional feature parameters of the static corner reflector, including distance, horizontal angle, pitch angle and relative speed.

[0024] Preferably, the specific implementation of S2 includes the following process:

[0025] S201: Use the four-dimensional feature parameters as calibration reference points;

[0026] S202: The host computer sends the calibration reference point information to the 4D millimeter-wave radar via the communication interface;

[0027] S203: After receiving the calibration reference point information, the 4D millimeter-wave radar performs high-computational matching analysis within a local search space centered on the calibration reference point. The local search space is limited to: distance deviation not exceeding ±0.5 meters, horizontal angle deviation not exceeding ±1°, pitch angle deviation not exceeding ±1°, and relative velocity being zero. Within this local search space, the target point with the strongest echo intensity is retrieved, the target point is identified as the static corner reflector, and its four-dimensional characteristic parameters are written into the non-volatile storage unit inside the radar to complete the calibration.

[0028] Preferably, the specific implementation of S3 includes the following process:

[0029] S301: Continuously receives multi-frame point cloud data returned by the 4D millimeter-wave radar to the calibrated static corner reflector. Each frame of point cloud contains multiple data points with distance, horizontal angle, pitch angle and velocity information.

[0030] S302: Based on the four-dimensional feature parameters of the calibration reference point pre-stored in S2, select the set of strong reflection points located in its neighborhood in each frame of point cloud;

[0031] S303: Set median filter parameters, including sliding window width and sliding step size;

[0032] S304: Based on the set of strong reflection points, extract the distance values ​​of each data point to form the distance observation sequence at the current time; perform median filtering on the distance observation sequence: sort the distance observation values ​​within a sliding window and take the median as the filtered output at that time.

[0033] S305: Use the filtered output as the combined distance value of the static corner reflector at the current moment.

[0034] Preferably, in step S4, the original intermediate frequency signal is obtained based on the timestamp corresponding to the merged distance value, and the high-precision distance value at the current moment is calculated based on the super-resolution ranging algorithm. The cumulative change of the high-precision distance value at the current moment relative to the initial calibration value is calculated to obtain the deformation variable DEF.

[0035] Preferably, the specific implementation of S4 includes the following process:

[0036] S401: Based on the timestamp corresponding to the merged distance value, obtain the original intermediate frequency signal collected at the same time, and perform N-point fast Fourier transform on the original intermediate frequency signal sampled in each Chirp period to obtain the distance domain spectrum.

[0037] The FFT spectrum Bin spacing is:

[0038]

[0039] Zero-filling technique is used to expand the number of FFT points, refining the spectral grid spacing without increasing the sampling rate or observation time; combined with the FMCW radar ranging model, the FFT spectral bin spacing is mapped to an equivalent range grid spacing, resulting in the FFT range resolution:

[0040]

[0041] Where c is the speed of light, and S is the Chirp slope, i.e., S = B / B is the signal bandwidth. Where N is the time period of the Chirp signal, and N is the number of FFT sampling points. For FFT distance resolution, For the spectral Bin spacing, The sampling frequency;

[0042] S402: Obtaining the range dimension amplitude spectrum based on FFT range resolution;

[0043] S403: In the range dimension amplitude spectrum, detect and locate the main peak frequency point corresponding to the static corner reflector, and determine its FFT Bin position;

[0044] S404: Using the FFT Bin where the main peak is located and its two adjacent Bins as sample points, the precise peak frequency value at the sub-Bin level is obtained by using the spectral peak precise positioning algorithm.

[0045] S405: Based on the accurate peak frequency value, combined with FMCW radar ranging, the frequency estimation result is mapped to the high-precision distance estimate of the static corner reflector at the current moment;

[0046] S406: Using the effective high-precision distance value obtained during the initial calibration phase of the system as a reference benchmark, the high-precision distance estimation result at the current moment is differentially calculated with it to obtain the deformation variable DEF at the corresponding moment.

[0047] Preferably, the specific implementation of S5 includes the following process:

[0048] After the S501:4D millimeter-wave radar is powered on again, it retrieves the four-dimensional feature parameters of all calibrated target points from the non-volatile storage unit and imports them into the calibration point data unit.

[0049] S502: Extract target point information one by one from the calibration point data unit, and perform high-computing-power matching analysis within the corresponding local search space; if the initial matching fails, record the exception and perform a second matching after completing the matching of the remaining target points;

[0050] S503: The matching process is repeated until all target points are recovered, after which the system automatically switches to normal monitoring mode.

[0051] Preferably, the non-volatile memory cell is an eMMC or a Flash chip.

[0052] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method for static target deformation monitoring and power-off self-recovery based on 4D millimeter-wave radar, which has the following effects:

[0053] 1. Achieve high-precision sub-millimeter deformation monitoring to meet early warning requirements.

[0054] By introducing super-resolution ranging algorithms (including zero-fill refined spectrum, three-point quadratic interpolation, etc.), the traditional FFT distance resolution limitation is broken, and sub-millimeter level (up to ±0.1 mm) distance measurement accuracy is achieved under conventional hardware conditions. This is significantly better than traditional GNSS (centimeter level), ordinary millimeter-wave radar (centimeter level) or InSAR (limited by time baseline) schemes. It can effectively capture the small initial displacements of infrastructure such as slopes and bridges, and provide reliable data support for early disaster warning.

[0055] 2. Automatically identifies and locks onto static corner reflectors without manual intervention.

[0056] An innovative dynamic auxiliary target guidance and identification mechanism is proposed: a movable dynamic auxiliary corner reflector is brought close to the static target to be calibrated. By analyzing the coupling relationship between spatial distance changes and echo intensity enhancement in continuous frame point clouds, and combining the "relative velocity is zero" criterion, the preset static corner reflector is automatically and robustly identified. This solves the problem of "target confusion" caused by clutter interference in complex field environments, avoids the drawbacks of traditional methods that rely on manual identification or prior coordinates, and greatly improves deployment efficiency and automation level.

[0057] 3. Possesses the ability to recover autonomously after a power outage, ensuring continuous long-term monitoring.

[0058] The calibrated four-dimensional characteristic parameters (range, horizontal angle, pitch angle, and relative velocity) are written into the radar's internal non-volatile storage unit (such as eMMC / Flash). When the system experiences a power outage and restart, it can automatically load historical parameters, perform high-computing-power matching analysis within the local search space, reactivate the target, and resume monitoring without requiring the host computer to reissue configuration or manual reset. This mechanism is particularly suitable for unattended, unstable power supply scenarios in field slope monitoring, ensuring long-term continuous and reliable data link.

[0059] 4. Strong resistance to environmental interference, enabling stable operation around the clock.

[0060] Based on the non-contact, active detection characteristics of 4D millimeter-wave radar, this method is not affected by weather conditions such as rain, fog, snow, dust, and sunlight, nor does it rely on satellite signals (such as GNSS). It can work continuously under all-day and all-weather conditions. At the same time, by merging multiple strong reflection points through median filtering, random noise and transient interference are effectively suppressed, and the stability of monitoring results is improved. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0062] Figure 1 This is a flowchart of a static target deformation monitoring and power-off self-recovery method based on 4D millimeter-wave radar provided in an embodiment of the present invention;

[0063] Figure 2 This is a flowchart of the power-off self-recovery method provided in the embodiments of the present invention;

[0064] Figure 3 This is a schematic diagram of the installation of a static corner reflector and a 4D millimeter-wave radar provided in an embodiment of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] This invention discloses a method for static target deformation monitoring and power-off self-recovery based on 4D millimeter-wave radar, such as... Figure 1 As shown, it includes:

[0067] S1. Identification of static corner reflectors: Static corner reflectors are identified and locked using dynamic auxiliary targets to determine their precise location and extract four-dimensional feature parameters; prior to identification, at least one static corner reflector is deployed as a steady-state target in the slope monitoring area, and a 4D millimeter-wave radar is installed on the opposite side so that the static corner reflector is within the monitoring field of view of the 4D millimeter-wave radar.

[0068] S2. Calibrate and store data for the static corner reflector: Send the four-dimensional feature parameters as calibration reference point information to the 4D millimeter-wave radar; The 4D millimeter-wave radar performs high-computing matching analysis and consistency analysis in the local search space with the calibration reference point as the center; If the matching is successful, the four-dimensional feature parameters are written into the non-volatile storage unit inside the radar.

[0069] S3. Merge multiple target data points generated by a single static corner reflector: Continuously collect multi-frame point cloud data returned by 4D millimeter-wave radar, filter out the set of strong reflection points in the neighborhood of the calibration reference point based on the calibration reference point defined by the four-dimensional feature parameters stored in the non-volatile storage unit, and filter the distance data at the same time to obtain the merged distance value of the static corner reflector at the current time.

[0070] S4. Displacement deformation monitoring based on merged distance values: Using the distance values ​​corresponding to the four-dimensional feature parameters stored in the non-volatile storage unit as the initial calibration values, calculate the cumulative change of the high-precision distance value at the current moment relative to the initial calibration value to obtain the deformation DEF;

[0071] S5. Achieve autonomous recovery of monitoring data after power failure: After the 4D millimeter-wave radar is powered on again, it retrieves all stored four-dimensional feature parameters from the non-volatile storage unit, loads them one by one, and repeatedly executes the high-computing-power matching analysis of the local search space range in S2; if the matching is successful, the corresponding target point is activated and its real-time deformation monitoring is restored; after all target points have been matched and activated, the system automatically enters the normal monitoring mode.

[0072] In this embodiment, the specific implementation of S1 includes the following process:

[0073] S101: The 4D millimeter-wave radar and static corner reflector have been deployed in the field and are both in a stationary state.

[0074] S102: After the 4D millimeter-wave radar is started, it transmits FMCW frequency-modulated continuous wave signals, collects and generates point cloud data containing static environmental targets, and uses it as a static background reference. Since all reflectors are stationary at this time, the point cloud data output by the radar contains multiple sets of static echo signals, making it difficult to distinguish the actual monitoring points.

[0075] S103: A movable dynamic auxiliary corner reflector or reflector is introduced as a dynamic auxiliary target within the radar monitoring range. It moves laterally (changing the azimuth angle) and radially (changing the distance from the radar) within the 4D millimeter-wave radar monitoring range, and gradually approaches the static corner reflector. This dynamic reflector will produce a significant Doppler frequency shift in the radar point cloud data, thus being stably identified as a dynamic target.

[0076] S104: Based on continuous frame point cloud data, perform spatial correlation analysis between the motion trajectory of the dynamic auxiliary corner reflector and the strong reflection points in the static background reference:

[0077] When a static strong reflection point is identified in continuous frame point cloud data, its relative velocity gradually decreases until it becomes zero, the spatial distance between it and the dynamic auxiliary corner reflector decreases as the dynamic auxiliary corner reflector moves, and its echo signal strength increases accordingly, the static strong reflection point is determined to be a static corner reflector, and its position is determined.

[0078] S105: After locking the static corner reflector, extract the four-dimensional feature parameters of the static corner reflector. The four-dimensional feature parameters include:

[0079] Range (R): Spatial distance from radar to corner reflector;

[0080] Relative velocity (v): zero in steady state;

[0081] Azimuth: The angle relative to the horizontal direction of the radar;

[0082] Elevation: The angle relative to the vertical direction of the radar.

[0083] At this point, the radar has successfully identified and locked onto the static corner reflector, which can then be used as a long-term monitoring reference point for continuous observation of deformation or displacement changes.

[0084] In this invention, whether using radar or corner reflectors as reference points, both are relatively stationary. Due to the strong electromagnetic wave reflection capability of corner reflectors against millimeter-wave radar, the radar can still detect the target point. However, since almost all target points are stationary, it is difficult to pinpoint the specific target point that represents the actual monitoring point. Furthermore, in practical use, although corner reflectors reflect electromagnetic waves more strongly than millimeter-wave radar, the complexity of the on-site environment is high. Trees, leaves, or even other metal objects near the radar may reflect electromagnetic waves with greater energy than those from more distant corner reflectors. Based on these various factors hindering the identification of the true target point, this invention proposes a method for locking steady-state targets using a dynamic target-assisted tracking approach.

[0085] The device is moved by hand using a movable, dynamic auxiliary corner reflector. By moving it left and right, as well as radially (relative to the radar), it gradually approaches the stationary corner reflector. By leveraging the more stable tracking and detection capabilities of the 4D millimeter-wave radar for dynamic targets, the movement of the dynamic target can help locate the stationary corner reflector. Furthermore, it can obtain relatively accurate distance, velocity (0 when stationary), horizontal angle, and pitch angle information of the stationary corner reflector relative to the radar position.

[0086] The following details the process by which the 4D millimeter-wave radar, after startup, transmits an FMCW (Frequency Modulated Continuous Wave) signal, acquires and generates point cloud data containing static environmental targets, and uses this data as a static background reference:

[0087] S1021 Chirp signal transmission phase:

[0088] The 4D millimeter-wave radar periodically transmits linear frequency modulated continuous wave signals (Chirp signals) from the transmitting antenna according to a preset time sequence.

[0089] S1022 Echo Reception and Intermediate Frequency Signal Generation:

[0090] When the transmitted signal encounters a target object, it will generate an electromagnetic reflection wave. After receiving the reflected signal, the receiving antenna mixes it with the current transmitted signal to obtain the original intermediate frequency (IF) signal.

[0091] S1023 Range FFT signal processing:

[0092] The 4D millimeter-wave radar performs an N-point Fast Fourier Transform (FFT) operation on the raw intermediate frequency signal acquired in each Chirp signal cycle to obtain the range domain spectrum.

[0093] S1024 Velocity Dimension Signal Processing:

[0094] In order to obtain information about the target's velocity, the system continuously transmits multiple chirp signals within one measurement cycle (called a frame).

[0095] Within the same range cell, the phase difference between different chirps accumulates over time due to the Doppler effect. Performing a second FFT on this phase sequence yields the Doppler frequency shift, allowing the calculation of the target's relative velocity.

[0096] S1025 Spatial Angle Estimation (Horizontal and Pitch Angles):

[0097] The 4D millimeter-wave radar employs a multiple-transmit, multiple-receive (MIMO) antenna array structure. Since the receiving antennas have a fixed spatial spacing, the echo signal from the same target will have a phase difference on different antenna elements. Based on this phase difference, the horizontal and vertical angles of the target can be estimated.

[0098] S1026 Target Point Cloud Generation and Coordinate Transformation:

[0099] In the three-dimensional domain of distance, velocity, and angle, constant false alarm rate (CFAR) or other threshold decisions are performed on the signal strength to screen out the true target reflection points.

[0100] After coordinate transformation, the polar coordinate information (R, Azimuth, Elevation) can be converted into the target point position in a spatial rectangular coordinate system:

[0101]

[0102] In the above formula, x, y, and z represent the relative spatial positions of the target point in a three-dimensional Cartesian coordinate system established with the radar as the origin;

[0103] R represents the straight-line spatial distance from the target point to the radar origin;

[0104] Azimuth represents the azimuth angle of the target point in the horizontal plane relative to the direction directly in front of the radar;

[0105] Elevation represents the pitch angle of the target point relative to the horizontal plane, with its positive direction being vertically upward.

[0106] This yields four-dimensional point cloud data containing information such as target distance, horizontal angle, pitch angle, and relative velocity, which can be used as a static background reference.

[0107] In this embodiment, the specific implementation of S2 includes the following process:

[0108] S201: Use the four-dimensional feature parameters as calibration reference points;

[0109] S202: The host computer sends the calibration reference point information to the 4D millimeter-wave radar through a communication interface (such as TCP / IP, CAN or CanFD);

[0110] S203: After receiving the calibration reference point information, the 4D millimeter-wave radar performs high-computation matching analysis within a local search space centered on the calibration reference point. The local search space is limited to the following range: target distance ±0.5 meters.

[0111] Horizontal angle range: target horizontal angle ±1°;

[0112] Pitch angle range: target pitch angle ±1°;

[0113] Speed ​​condition: Relative speed = 0.

[0114] Within this local search space, the target point corresponding to the strongest FMCW echo signal is retrieved, the target point is identified as a static corner reflector, and its four-dimensional characteristic parameters are written into the non-volatile storage unit inside the radar to complete the calibration.

[0115] In this embodiment, the specific implementation of S3 includes the following process:

[0116] S301: Continuously receives multi-frame point cloud data returned by a 4D millimeter-wave radar from a calibrated static corner reflector. Each frame contains multiple data points with range, horizontal angle, pitch angle, and velocity information. Example data is as follows (unit: meters): 85.028, 85.028, 85.031, 85.026, 85.027, 85.028, 85.029, 85.027, 85.028, 85.030, 85.026, 85.027, 85.028, 85.029

[0117] S302: Based on the four-dimensional feature parameters of the calibration reference point pre-stored in S2, select the set of strong reflection points located in its neighborhood in each frame of point cloud; specifically, extract the point cloud data related to the target point from the returned multi-frame point cloud data, and filter out weak reflection points unrelated to the target point according to the distance and angle matching relationship, retaining only the set of reflection points within the target reflection area.

[0118] S303: Set median filter parameters, including sliding window width and sliding step size; specifically, set the filter window size according to the monitored data frame rate and noise level, for example:

[0119] Window width: 3 data points (i.e., each time the current point and one point before and after it are retrieved).

[0120] Window sliding step: 1

[0121] The window size can be adaptively adjusted according to the noise characteristics of the data (e.g., using a 5-point or 7-point window to enhance smoothness).

[0122] S304: Based on the set of strong reflection points, extract the distance values ​​of each data point to form the distance observation sequence at the current time; perform median filtering on the distance observation sequence: sort the distance observation values ​​within a sliding window and take the median as the filtered output at that time.

[0123] The median is the value that is in the middle after sorting. If the window size is odd (e.g., 3x3 or 5x5), then the median is the middle value after sorting; if the window size is even, then the median is the average of the two middle values.

[0124] Slide the window to the next position and repeat the above process until all data points have been processed.

[0125] S305: Use the filtered output as the combined distance value of the static corner reflector at the current moment, take the mean or mode of the median filtered data, upload it to the monitoring platform and store the current monitoring data.

[0126] The filtered results are, for example: 85.028, 85.028, 85.027, 85.027, 85.028, 85.028, 85.028, 85.028, 85.027, 85.027, 85.028, 85.029, 85.029

[0127] Table 1 Comparison of Original Data and Data After Median Filtering

[0128]

[0129] By comparing the original data with the data after median filtering in the table above, the fluctuation range of the original data is ±0.005, that is, the original data fluctuates within 10 mm, while the fluctuation range of the filtered data is ±0.002, that is, the filtered data fluctuates within 4 mm. If a larger amount of data is used for median filtering, excellent monitoring data can be obtained.

[0130] By using median filtering for data processing, we can more accurately monitor the relative distance between monitoring data points. Furthermore, by using monitoring data of a single angle reflected by radar or small displacement deformation data of different data, we can more accurately determine whether a real displacement has occurred, thus better ensuring the accuracy of the monitoring data.

[0131] In this embodiment, the specific implementation of S4 includes the following process:

[0132] S401: Based on the timestamp corresponding to the merged distance value, obtain the original intermediate frequency signal collected at the same time, and perform N-point fast Fourier transform on the original intermediate frequency signal sampled in each Chirp period to obtain the distance domain spectrum.

[0133]

[0134] Zero-filling technique is used to expand the number of FFT points, refining the spectral grid spacing without increasing the sampling rate or observation time. Combined with a frequency-modulated linear wave (FMCW) radar ranging model, the FFT spectral bin spacing is mapped to an equivalent range grid spacing, yielding the FFT range resolution.

[0135]

[0136] Where c is the speed of light, S is the Chirp slope (S = B / Tc), B is the bandwidth, Tc is the time period of the Chirp signal, and N is the number of FFT sampling points. For FFT distance resolution, For the spectral Bin spacing, The sampling frequency is used; the basic ranging resolution is determined by the bandwidth B (B=4GHz). By using N-point FFT (or ultra-long FFT with zero padding), the spectral Bin spacing can be refined from B / Tc / N to fs / N, thus improving the equivalent ranging accuracy. Here, N=4096, B=4GHz, and Tc is the intermediate frequency signal time interval.

[0137] S402: Obtain the range dimension amplitude spectrum based on FFT range resolution, providing basic data for subsequent peak detection and precise positioning;

[0138] S403: In the range-dimensional amplitude spectrum, locate the main peak frequency point corresponding to the static corner reflector and determine its FFT Bin position;

[0139] S404: Using the FFT Bin where the main peak is located and its adjacent Bins as local observation samples, the precise peak frequency value at the sub-Bin level is obtained by using the spectral peak precise localization algorithm;

[0140] The precise peak localization algorithm includes, but is not limited to:

[0141] ① The quadratic interpolation method achieves continuous domain estimation of peak frequency by curve fitting of local peaks in the amplitude spectrum. Under high signal-to-noise ratio conditions, it can obtain an accuracy gain better than 1 / 100 to 1 / 1000 of the FFT grid spacing.

[0142] ②Compressed Sensing, based on the construction of an ultra-high resolution range grid, approximates the true peak position through an iterative reconstruction algorithm. It utilizes the sparse characteristics of the signal in the range domain to achieve high-precision positioning with fast convergence.

[0143] ③ The least squares fitting or low-rank Fourier transform (LRFT) method based on the signal model is used to perform parameterized modeling and fitting of the echo signal in a low signal-to-noise ratio environment, so as to reduce the impact of noise interference on the peak positioning accuracy.

[0144] S405: Based on precise peak frequency values, combined with the FMCW radar ranging formula. c represents the speed of light. This represents the beat frequency after the transmitted signal and the received echo signal are mixed. B represents the frequency modulation bandwidth. This represents the time span of a single chirp, and the high-precision distance value of the static corner reflector at the current moment is calculated, thus breaking through the FFT grid spacing limit;

[0145] S406: Using the effective high-precision distance value obtained during the initial calibration phase of the system as a reference benchmark, the high-precision distance estimation result at the current moment is differentially calculated with it to obtain the deformation DEF at the corresponding moment, thereby realizing deformation monitoring with millimeter-level and below accuracy.

[0146] This invention systematically demonstrates the mechanism for improving ranging accuracy from four aspects: bandwidth limit, large-scale FFT refinement, spectral peak super-resolution positioning, and statistical estimation theoretical limit.

[0147] According to the Cramér–Rao Bound theory, for an unbiased distance estimator, the lower bound of its ranging variance is closely related to the system bandwidth and signal-to-noise ratio. The larger the bandwidth and the higher the signal-to-noise ratio, the higher the theoretically achievable accuracy.

[0148] Under high signal-to-noise ratio conditions, by combining large-scale FFT processing (N=4096) with the interpolation gain G brought by spectral peak interpolation or sparse reconstruction, the distance estimation error can be further approximated to the theoretical lower bound in a statistical sense. Under typical parameter conditions, the ranging accuracy can be reduced to 0.3 mm or even lower, thus providing sufficient theoretical basis for millimeter-level and sub-millimeter-level deformation monitoring.

[0149] Table 2. Theoretical Accuracy Explanation of 4GHz Bandwidth for 77G Millimeter-Wave Radar

[0150]

[0151] This invention utilizes large-scale FFT (including zero-fill and extended observation), spectral peak super-resolution interpolation (parabolic, CS, etc.), and Cramér-Rao lower bound analysis to improve the ranging accuracy of a 77GHz / 4GHz bandwidth FMCW radar from the basic centimeter level (3.75cm) to the sub-millimeter level (≤0.3mm) under high SNR and reasonable parameter configuration, thus meeting the requirements for high-precision ranging.

[0152] Based on the above theoretical analysis, and considering the relationship between slope collapse and minute displacement deformation, the actual manifestation of the slope in the early stage of slope collapse is minute displacement deformation. Combined with the monitoring capabilities of 4D millimeter waves, which can achieve an accuracy of 0.3 millimeters, during long-term slope monitoring, the relative distance, relative horizontal angle, relative pitch angle, and relative speed of the corner reflectors can be observed. Multiple corner reflectors can be deployed on a single slope to form multi-point coverage of the slope surface. This not only reduces the dependence on traditional and unstable monitoring equipment such as GNSS, but also reduces costs.

[0153] Therefore, using 4D millimeter-wave radar to monitor for potential hazards on slopes is highly effective. Furthermore, the monitoring frequency of 4D millimeter-wave radar at target points can reach 5-10 Hz per second, which is crucial for real-time monitoring, enabling immediate reporting of potential hazards and timely warnings.

[0154] In this embodiment, as Figure 2 As shown, the specific implementation of S5 includes the following process:

[0155] After the S501 4D millimeter-wave radar is powered on again, it initializes the system parameters and internal communication module; it retrieves the four-dimensional feature parameters of all calibrated target points from the non-volatile storage unit and imports them into the calibration point data unit to prepare for subsequent matching analysis.

[0156] If no target point data is found in the storage unit, the system returns to calibration mode and waits for a new target point calibration operation.

[0157] S502: Extract target point information (including distance, relative velocity, horizontal angle, and pitch angle) one by one from the calibration point data unit, and perform high-performance matching analysis within the corresponding local search space (e.g., distance ±0.5m, horizontal angle ±1°, pitch angle ±1°).

[0158] Search for the point cloud echo data with the highest FMCW signal strength in this area and determine whether it matches the characteristics of the preset target point.

[0159] If a match is successful, the target point will be reactivated and marked as "monitoring valid status";

[0160] If a match fails, the error information is recorded and the next target point is matched. Error data needs to be matched again after the last target point is matched.

[0161] S503: After the first target point is successfully matched, the system automatically calls the data of the next target point and repeats S502;

[0162] This logic is repeated until all marked target points have been rematched and activated.

[0163] Once all target points have been restored, the system automatically switches to real-time monitoring mode to continue monitoring the displacement and deformation of the target points and uploading data.

[0164] To ensure that monitoring data is not lost after a system power outage or abnormal restart, this invention stores the target point data in an internal non-volatile storage unit (such as an eMMC or Flash chip) after calibration. The background is that if the radar cannot automatically restore the calibration point information and monitoring status after power-on, manual recalibration is still required, which is not only inefficient and wastes a lot of human effort, but may also cause data continuity interruptions.

[0165] To address this issue, a technical method for autonomously recovering target point data after radar power failure and subsequent power restoration has been invented. This method achieves autonomous recovery of radar monitoring functions through a systematic data storage, retrieval, loading, and matching process. During the radar calibration phase, the invention saves the characteristic parameters of all monitored target points to the radar's internal non-volatile storage unit. Upon radar power restoration, the invention automatically retrieves and loads target point information from the storage unit, and reconfirms the existence and location of the target points using computational focusing and local matching algorithms, thus achieving automatic recovery of the monitoring system without manual intervention.

[0166] Example 1:

[0167] The entire process of implementing a slope monitoring project, from "installation → calibration → monitoring → power failure → recovery," will be illustrated below:

[0168] The first step is the installation of static corner reflectors and 4D millimeter-wave radar, such as... Figure 3 As shown;

[0169] exist Figure 3 In this case, the 4D millimeter-wave radar is installed on a stable and risk-free location on the opposite side of the slope. The 4D millimeter-wave radar faces the slope directly. The circled locations are the installation points of the static corner reflectors, which are set at different fault locations on the slope and are all within the horizontal and vertical monitoring range of the 4D millimeter-wave radar.

[0170] The second step involves moving the dynamic auxiliary corner reflector while oscillating the fixed static corner reflector to determine the horizontal angle, elevation angle, and range information of the fixed static corner reflector relative to the 4D millimeter-wave radar.

[0171] Due to the large amount of content in this section, the following calibration process has been compiled for explanation.

[0172] When the movable dynamic auxiliary corner reflector moves around the fixed static corner reflector, the radar can identify relevant information and display it directly on the host computer, which includes distance information, horizontal angle information, and pitch angle information.

[0173] Once the host computer control software recognizes the relevant information, click "Confirm Calibration Position";

[0174] The host computer control software will automatically send relevant information to the radar, and the host computer will prompt that calibration is in progress and immediately display relevant real-time data;

[0175] After the radar's internal data processing and calibration are completed, a "Storage Confirmation" message will appear.

[0176] Clicking "Save Confirmation" will store the calibration information in the radar and display a message "Storage Complete";

[0177] The above completes the radar's tracking, identification, and calibration process for target points.

[0178] Step 3: Radar monitoring of the target point

[0179] Once the radar has calibrated the target point, it then begins continuous deformation and displacement monitoring. Through continuous monitoring of a single target point, combined with large-scale analysis and merging of the monitoring data, the distance between the radar and the angular reflection is reduced to the sub-millimeter level. Below, we present a data point from the current phase of the actual project, illustrating and analyzing the data:

[0180] By monitoring the distance, horizontal angle, pitch angle, and DEF of the slope in real time, and using the system's 4G capability, the data is uploaded to the platform. The platform's monitoring and early warning section analyzes the relevant data to determine if there are any potential risks based on different slope types, and then provides relevant management advice to the operation and maintenance department.

[0181] In addition, during real-time monitoring, the radar data, such as the information stored in real time for this location, will include: target ID, offset before and after power-on, cumulative change, cumulative value before reset, and backup of the cumulative value before reset.

[0182] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0183] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for static target deformation monitoring and power-off self-recovery based on 4D millimeter-wave radar, characterized in that, Includes the following steps: S1. Identify static corner reflectors: Use dynamic auxiliary targets to identify and lock onto static corner reflectors, determine their precise location, and extract four-dimensional feature parameters; S2. Calibrate and store the static corner reflector: The extracted four-dimensional feature parameters are sent to the 4D millimeter-wave radar as calibration reference point information; The 4D millimeter-wave radar performs high-computing matching and consistency analysis within a preset local search space, centered on the calibration reference point; When the matching result meets the preset judgment conditions, the corresponding four-dimensional feature parameters are written into the non-volatile storage unit inside the radar. S3. Merge multiple target data points generated by a single static corner reflector: Continuously collect multi-frame point cloud data returned by 4D millimeter-wave radar, filter out the set of strong reflection points in the neighborhood of the calibration reference point based on the calibration reference point defined by the four-dimensional feature parameters stored in the non-volatile storage unit, and filter the distance data at the same time to obtain the merged distance value of the static corner reflector at the current time. S4. Displacement deformation monitoring based on the merged distance value: Using the distance value corresponding to the four-dimensional feature parameters stored in the non-volatile storage unit as the initial calibration value, calculate the cumulative change of the high-precision distance value at the current moment relative to the initial calibration value to obtain the deformation DEF; S5. Achieve autonomous recovery of monitoring data after power failure: After the 4D millimeter-wave radar is powered on again, it retrieves all stored four-dimensional feature parameters from the non-volatile storage unit, loads them one by one, and repeatedly executes the high-computing-power matching analysis of the local search space range in S2; if the matching is successful, the corresponding target point is activated and its real-time deformation monitoring is restored; after all target points have been matched and activated, the system automatically enters the normal monitoring mode. The specific implementation of S1 includes the following process: S101: The 4D millimeter-wave radar and static corner reflector have been deployed in the field and are both in a stationary state. S102: After the 4D millimeter-wave radar is started, it transmits FMCW frequency-modulated continuous wave signals, collects and generates point cloud data containing static environmental targets, and uses it as a static background reference. S103: Introducing a movable dynamic auxiliary corner reflector as a dynamic auxiliary target, which moves laterally left and right and radially near and far within the monitoring range of the 4D millimeter-wave radar, and gradually approaches the static corner reflector. S104: Based on continuous frame point cloud data, perform spatial correlation analysis between the motion trajectory of the dynamic auxiliary corner reflector and the strong reflection points in the static background reference: When a static strong reflection point is identified in continuous frame point cloud data, its relative velocity is zero, the spatial distance between it and the dynamic auxiliary corner reflector decreases as the dynamic auxiliary corner reflector moves, and its echo signal strength increases accordingly, the static strong reflection point is determined to be the static corner reflector, and its position is determined. S105: Extract the four-dimensional feature parameters of the static corner reflector, including distance, horizontal angle, pitch angle and relative speed.

2. The method as described in claim 1, characterized in that, The specific implementation of S2 includes the following process: S201: Use the four-dimensional feature parameters as calibration reference points; S202: The host computer sends the calibration reference point information to the 4D millimeter-wave radar via the communication interface; S203: After receiving the calibration reference point information, the 4D millimeter-wave radar performs high-performance matching and consistency analysis within a local search space centered on the calibration reference point. The local search space is defined as follows: distance deviation not exceeding ±0.5 meters, horizontal angle deviation not exceeding ±1°, pitch angle deviation not exceeding ±1°, and relative velocity being zero. Within this local search space, the target point with the strongest echo intensity is retrieved, the target point is identified as the static corner reflector, and its four-dimensional characteristic parameters are written into the non-volatile storage unit inside the radar to complete the calibration.

3. The method as described in claim 1, characterized in that, The specific implementation of S3 includes the following process: S301: Continuously receives multi-frame point cloud data returned by the 4D millimeter-wave radar to the calibrated static corner reflector. Each frame of point cloud contains multiple data points with distance, horizontal angle, pitch angle and velocity information. S302: Based on the four-dimensional feature parameters of the calibration reference point pre-stored in S2, select the set of strong reflection points located in its neighborhood in each frame of point cloud; S303: Set median filter parameters, including sliding window width and sliding step size; S304: Based on the set of strong reflection points, extract the distance values ​​of each data point to form the distance observation sequence at the current time; perform median filtering on the distance observation sequence: sort the distance observation values ​​within a sliding window and take the median as the filtered output at that time. S305: Use the filtered output as the combined distance value of the static corner reflector at the current moment.

4. The method as described in claim 1, characterized in that, In step S4, the original intermediate frequency signal is obtained based on the timestamp corresponding to the merged distance value, and the high-precision distance value at the current moment is calculated based on the super-resolution ranging algorithm. The cumulative change of the high-precision distance value at the current moment relative to the initial calibration value is calculated to obtain the deformation variable DEF.

5. The method as described in claim 4, characterized in that, The specific implementation of S4 includes the following process: S401: Based on the timestamp corresponding to the merged distance value, obtain the original intermediate frequency signal collected at the same time, and perform N-point fast Fourier transform on the original intermediate frequency signal sampled in each Chirp period to obtain the distance domain spectrum. The FFT spectrum Bin spacing is: Zero-filling technique is used to expand the number of FFT points, refining the spectral grid spacing without increasing the sampling rate or observation time; combined with the FMCW radar ranging model, the FFT spectral bin spacing is mapped to an equivalent range grid spacing, resulting in the FFT range resolution: Where c is the speed of light, and S is the Chirp slope, i.e., S = B / B is the signal bandwidth. Where N is the time period of the Chirp signal, and N is the number of FFT sampling points. For FFT distance resolution, For the spectral Bin spacing, The sampling frequency; S402: Obtaining the range dimension amplitude spectrum based on FFT range resolution; S403: In the range dimension amplitude spectrum, detect and locate the main peak frequency point corresponding to the static corner reflector, and determine its FFT Bin position; S404: Using the FFT Bin where the main peak is located and its two adjacent Bins as sample points, the precise peak frequency value at the sub-Bin level is obtained by using the spectral peak precise positioning algorithm. S405: Based on the accurate peak frequency value, combined with FMCW radar ranging, the frequency estimation result is mapped to the high-precision distance estimate of the static corner reflector at the current moment; S406: Using the effective high-precision distance value obtained during the initial calibration phase of the system as a reference benchmark, the high-precision distance estimation result at the current moment is differentially calculated with it to obtain the deformation variable DEF at the corresponding moment.

6. The method as described in claim 1, characterized in that, The specific implementation of S5 includes the following process: After the S501:4D millimeter-wave radar is powered on again, it retrieves the four-dimensional feature parameters of all calibrated target points from the non-volatile storage unit and imports them into the calibration point data unit. S502: Extract target point information one by one from the calibration point data unit, and perform high-computing-power matching analysis within the corresponding local search space; if the initial matching fails, record the exception and perform a second matching after completing the matching of the remaining target points; S503: The matching process is repeated until all target points are recovered, after which the system automatically switches to normal monitoring mode.

7. The method as described in claim 1, characterized in that, The non-volatile memory cells are eMMC or Flash chips.

Images