A high anti-interference millimeter wave radar parking space occupancy detection method

Abstract

The application discloses a kind of high anti-interference millimeter wave radar parking space occupation detection methods, it is related to radar parking space detection technical field, by radio frequency sensing unit emit frequency-modulated continuous wave and limit single-target detection range, collection echo signal and extract multidimensional physical characteristic vector;Establish environment situation awareness and double-scale dynamic threshold adaptive model, generate and environment base evolution matching detection constraint matrix;In feature matching and state discrimination feedback loop, using spatiotemporal feature continuity association mechanism and micro-motion feature compensation model carry out time domain cross verification, output regional occupation state mapping value;Simultaneously construct state-driven channel collaborative scheduling model, reverse regulation detection sampling period.The application effectively solves the technical problems of communication reliability, etc. of traditional detection scheme in static vehicle locking, non-standard parking identification and strong electromagnetic interference, realizes all-weather, high-precision, low-power parking space occupation detection, especially suitable for complex indoor and outdoor parking scene.

Description

Technical Field

[0001] This invention relates to the field of radar parking space detection technology, specifically a high-interference-resistant millimeter-wave radar method for detecting parking space occupancy. Background Technology

[0002] Current intelligent parking lot space detection solutions mainly include three mainstream technologies: inductive loop detection, video image detection, and ultrasonic detection. Each technology has obvious shortcomings and cannot simultaneously achieve a balance between detection accuracy, environmental adaptability, and deployment cost.

[0003] Inductive loop detectors determine parking space status by detecting electromagnetic changes generated when a vehicle drives over the loop. This technology is mature and highly stable, but it requires road construction and has high installation and maintenance costs. It can only detect vehicle driving actions and cannot distinguish between large vehicles and non-motorized vehicles, making it prone to misjudgments. Furthermore, construction will damage the road surface, making it unsuitable for underground parking garages and parking structures.

[0004] Video image detection: Parking spaces can be detected by relying on camera image recognition algorithms. Multiple parking spaces can be detected in a coordinated manner. However, it is extremely dependent on lighting conditions. The recognition accuracy drops significantly in nighttime, backlight, rain, snow, fog, and shadow environments. At the same time, there is a risk of privacy leakage of images. It also requires high computing power, consumes a lot of equipment power, and has high operation and maintenance costs.

[0005] Ultrasonic detection: It uses ultrasonic ranging to determine whether there is a car in the parking space. It is inexpensive and easy to install, but the detection angle is too wide and it is easily affected by temperature, humidity, airflow and dust. It has low accuracy in identifying low-profile vehicles and small electric vehicles, and it is prone to blind spots at close range, resulting in a high rate of false detection and missed detection.

[0006] However, the drawback of existing technologies is that:

[0007] 1. Weak anti-interference ability: Ultrasonic detection cannot filter pedestrians, debris, environmental noise, and crosstalk interference from adjacent ultrasonic waves, resulting in frequent false detections and missed detections;

[0008] 2. Limited identification dimensions: Ultrasonic detection relies solely on a single parameter, such as distance or echo intensity, making it impossible to accurately distinguish between vehicles, pedestrians, and obstacles;

[0009] 3. Static detection failure: Ultrasonic detection is designed for vehicles that have been parked for a long time. It is prone to misjudging the vacancy of the parking space due to signal attenuation or changes in ambient light.

[0010] 4. Poor adaptability: Ground loop detection requires installation on broken road surfaces, making it unsuitable for different scenarios such as indoor underground garages and parking garages, resulting in insufficient environmental adaptability;

[0011] 5. High construction and deployment costs: Video recognition and detection equipment has complex network deployment, high requirements for supporting operation and maintenance systems, and high material costs; at the same time, most detection equipment has poor compatibility, and the cost of adapting to new and old parking lots and the cost of large-scale deployment construction remain high, which is not conducive to large-scale popularization and implementation. Summary of the Invention

[0012] The purpose of this invention is to provide a highly anti-interference millimeter-wave radar parking space occupancy detection method. Through innovative mechanisms such as multi-dimensional physical feature fusion recognition, spatiotemporal feature continuous memory, dual-scale dynamic threshold adaptation, and end-to-end hardware and software collaborative anti-interference, this method eliminates the defects of traditional single-threshold detection from the algorithmic level, achieving all-weather, all-scenario, high-precision, and extremely low-power accurate detection of parking space vacancy / occupancy status, thereby solving the problems mentioned in the background art.

[0013] To achieve the above objectives, the present invention provides the following technical solution: a high anti-interference millimeter-wave radar method for detecting parking space occupancy, comprising the following steps:

[0014] S1: The radio frequency sensing unit transmits a frequency-modulated continuous wave signal and constructs a physical echo link containing a single target area through a spatial detection range limitation mechanism. It continuously collects raw echo data and performs pre-filtering and noise reduction mapping to output a continuous target spatial signal stream.

[0015] S2: Receive the target spatial signal stream, perform multi-dimensional joint analysis in the signal domain, extract and construct a multi-dimensional physical feature vector group representing the target, wherein the feature vector group is formed by the mutual coupling of spatial distance parameter, motion velocity parameter, reflection intensity parameter and frequency domain distribution parameter;

[0016] S3: Establish an environmental situation awareness and threshold adaptive interaction model, simultaneously extract the slope of environmental bottom-level noise fluctuation in the target spatial signal stream, and combine multi-scale dynamic threshold calculation to generate an adaptive detection constraint matrix that matches the evolution of the current physical environment base.

[0017] S4: Establish a feature matching and state discrimination feedback loop, using the adaptive detection constraint matrix as the dynamic decision boundary, to perform real-time mapping and clutter truncation on the multi-dimensional physical feature vector group; and trigger the spatiotemporal feature continuity correlation mechanism and micro-motion feature compensation model to perform temporal cross-verification at the feature level, thereby comprehensively calculating and outputting the regional space occupancy state mapping value.

[0018] S5: Establish a state-driven and channel-coordinated scheduling model, using the area space occupancy state mapping value as a feedback control parameter to adjust the detection sampling period of the radio frequency sensing unit in reverse; at the same time, transmit the area space occupancy state mapping value after link layer redundancy verification to the outside through an active anti-interference physical communication link with an embedded channel quality assessment mechanism.

[0019] Preferably, the interaction between the spatiotemporal feature continuity correlation mechanism and the micro-motion feature compensation model in the feature matching and state discrimination feedback loop in step S4 is specifically as follows:

[0020] When the multi-dimensional physical feature vector group represents the target in a continuous approaching state, the high-confidence dynamic multi-dimensional feature vector set within the time window is extracted and recorded into the spatiotemporal storage stack.

[0021] When the motion velocity parameter in the feature vector set is detected to decay to the zero limit range, a static spatiotemporal anchoring mapping is established to uniquely bind and associate the target's current dimensionality-reduced two-dimensional spatial feature parameter with the dynamic multidimensional feature vector set in the spatiotemporal storage stack.

[0022] During the periodic verification phase initiated by the system, the driving feature extraction unit captures the micro-Doppler frequency shift parameter of the target in a static state, and injects the frequency shift parameter as a compensation factor into the static spatiotemporal anchoring mapping for cross-validation, so as to maintain the effective locking of the regional spatial occupancy state mapping value.

[0023] Preferably, in step S4, during the real-time mapping and clutter truncation of the multi-dimensional physical feature vector group, an edge target soft-decision interaction logic is embedded:

[0024] When a high-frequency signal conforming to a preset reflection intensity parameter is continuously extracted within the effective detection critical boundary region of the physical echo link, the historical state time series and real-time reflection energy integral of the critical boundary region are extracted.

[0025] The historical state time series and the real-time reflection energy integral are input into the edge probability decision engine to dynamically calculate the actual occupancy probability metric of the signal within the critical boundary.

[0026] When the actual occupancy probability metric exceeds the system's preset warning probability envelope, the signal within the critical boundary is forcibly classified as an effective target intrusion, and the regional space occupancy status mapping value is updated.

[0027] Preferably, the edge target soft-decision interaction logic is coupled with a micro-motion feature orientation verification mechanism:

[0028] During the parallel cycle of the edge probability decision engine calculating the actual occupancy probability metric, the system generates a frequency domain high-sensitivity analytical window pointing to the specific spatial coordinates of the critical boundary based on the feature vector group;

[0029] In response to successfully matching a specific frequency domain distribution parameter characterizing physical micro-motion within the high-sensitivity resolution window in the frequency domain, the system generates a privilege override instruction, which is used to directly compensate the actual occupancy probability metric to the absolute confidence threshold.

[0030] Preferably, the specific interaction operation configuration of the environmental situation perception and threshold adaptive interaction model in step S3 is as follows:

[0031] In the background of the system, a short-term dynamic learning time window is used to respond to transient media disturbances, and a long-term historical evolution time window is used to characterize the absolute basis of physical space.

[0032] The first noise floor feature value output by the short-term dynamic learning time window and the second noise floor feature value output by the long-term historical evolution time window are extracted. The deviation gradient vector between the two in the time series is measured in real time through the difference comparison algorithm. The threshold parameter in the adaptive detection constraint matrix is ​​dynamically configured according to the deviation gradient vector.

[0033] Preferably, the environmental situation awareness and threshold adaptive interaction model is deeply coupled with an anti-environment drift circuit breaker interaction mechanism:

[0034] The system performs derivative calculations on the deviation gradient vector. When it is determined that the deviation gradient exhibits a continuous monotonically increasing trend and the cumulative change reaches the system's preset safety physical tolerance boundary limit, the system determines that a physical intrusion of a slowly changing environmental medium has occurred.

[0035] In response to the aforementioned physical intrusion determination, the system generates a blocking signal to cut off the data interaction link between the deviation gradient vector and the adaptive detection constraint matrix update module, thereby freezing the automatic increase of the detection determination threshold.

[0036] Preferably, the method also constructs a cross-level target state and environmental basis mutually exclusive decoupled feedback channel:

[0037] When the target completes a unique binding association through the static spatiotemporal anchoring mapping and the system continuously outputs the occupancy status mapping value, a spatial coordinate exemption command is sent to the environmental situation awareness and threshold adaptive interaction model through the mutual exclusion decoupling feedback channel.

[0038] In response to the exemption command, the environmental situation awareness and threshold adaptive interaction model, when performing iterative updates of the short-term dynamic learning time window, forcibly masks all signal data parameters in the physical coordinate domain where the target is located by setting a weight mask, thereby blocking the assimilation and transfer of target features to environmental noise features.

[0039] Preferably, the process of transmitting signals to the outside world through the active anti-interference physical communication link in step S5 includes interactive control of noise floor sensing and protocol reconstruction:

[0040] Before initiating the handshake for sending data frames on the physical link, the underlying communication control unit first performs a scan and evaluation of the current physical bus electromagnetic noise floor and channel spectrum purity.

[0041] When the quantization results of the scan evaluation indicate that there is a high-intensity harmonic injection or channel blocking risk in the link, the adaptive protocol reconstruction mechanism is triggered to switch the encapsulation format of the underlying data frame to a high-order cyclic redundancy check protocol with cascaded error correction capability, and simultaneously open multiple time diversity retransmission queues to combat sudden bit errors.

[0042] Preferably, the multi-node spatial awareness fusion network interaction logic is constructed based on the active anti-interference physical communication link extension:

[0043] When the actual occupancy probability metric value output by the edge probability decision engine falls to the system's preset fuzzy decision range, the main control processing node suspends the current decision and encapsulates a data request message containing the current spatial coordinate parameters and feature vector set.

[0044] The data request message is broadcast to adjacent subordinate probe nodes in the physical space via the physical communication link; the feature observation sets returned by each subordinate probe node for the overlapping area in the space are collected, and the joint weighted fusion operation of the multi-source feature data matrix is ​​performed to break the blind spot of the single node spatial perspective and output the calibrated regional spatial occupancy status mapping value.

[0045] Preferably, the process of reverse adjustment of the detection sampling period in step S5 is specifically executed by an event-driven feedback mechanism:

[0046] A state flip timer is set up to count the steady-state holding time of the space occupancy state mapping value in real time. If the time exceeds the system's preset idle tolerance period, the system kernel issues a power management command to drive the radio frequency sensing unit and signal processing bus to reduce the frequency to sleep detection mode.

[0047] In the sleep detection mode, if the rising edge feature that matches the sudden change event is extracted from the front-end envelope detector or the basic spatial distance parameter, the hardware interrupt controller immediately responds and generates a millisecond-level wake-up interrupt, driving the full module link of the system to fully reset and smoothly restore to the high frame rate physical feature full analysis sampling state.

[0048] Preferably, the specific process of extracting and constructing a multi-dimensional physical feature vector group representing the target includes: performing a multi-dimensional fast Fourier transform (FFT) on the effective echo signal to extract the intermediate frequency difference frequency signal matrix; and extracting the spatial distance parameter based on the spectral peak coordinate mapping of the intermediate frequency difference frequency signal matrix.

[0049] The motion velocity parameter is extracted by calculating the rate of change of phase difference between adjacent frequency-modulated continuous wave (FMCW) sweep cycles;

[0050] The reflection intensity parameter is extracted by integrating the energy density around the target's corresponding spectral peak.

[0051] And the frequency domain distribution parameters are extracted by statistically analyzing the spectral width dispersion and Doppler frequency shift distribution characteristics of the spectral peaks;

[0052] The system normalizes the above four heterogeneous physical parameters and concatenates them according to preset weight coefficients to construct the multi-dimensional physical feature vector group that represents the absolute physical properties of the target.

[0053] Preferably, the specific process of generating an adaptive detection constraint matrix that matches the evolution of the current physical environment substrate by combining multi-scale dynamic thresholding includes:

[0054] An initial high-dimensional detection threshold matrix containing the distance boundary lower limit, energy reflection basis, and velocity tolerance envelope is pre-set in the system kernel;

[0055] Based on the slope of the underlying environmental noise fluctuation, the transient environmental disturbance component and the long-term absolute background component are extracted.

[0056] The transient environmental disturbance component and the long-term absolute background component are input into an adaptive filter to deduce the environmental compensation weight matrix corresponding to each dimension of distance, intensity and velocity.

[0057] The initial high-dimensional detection threshold matrix is ​​updated by matrix multiplication or nonlinear mapping using the environmental compensation weight matrix, so that the decision envelope of the adaptive detection constraint matrix can adaptively expand or contract dynamically in accordance with the evolution of the physical environment base (such as snowfall or leaf accumulation).

[0058] Preferably, the specific logic for the triggering spatiotemporal feature continuity correlation mechanism and the micro-motion feature compensation model to perform temporal cross-verification at the feature level, and to calculate and output the regional spatial occupancy state mapping value accordingly, includes:

[0059] Initial probability calculation: Substitute the current multi-dimensional physical feature vector group into the adaptive detection constraint matrix for boundary verification to generate basic target existence probability values;

[0060] Spatiotemporal continuity weighting: If the target successfully triggers the spatiotemporal feature continuity association mechanism (i.e., completes the binding of static features with historical dynamic high-confidence features), the system applies a temporal continuity positive gain weight to the basic target existence probability value;

[0061] Micro-motion confidence compensation: In the time-domain cross-verification stage, if the micro-motion feature compensation model successfully matches the frequency shift component that characterizes the vehicle's physical micro-motion, the system applies an absolute confidence multiplier to the current probability value;

[0062] Anti-shake mapping output: The final comprehensive probability value after weighting and compensation is input into a Schmitt trigger or time smoothing hysteresis model. When the comprehensive probability value exceeds the high-bit trigger threshold and is maintained within the set anti-shake time window, the system flips and outputs a determined area occupancy state mapping value; when it falls to the low-bit release threshold, it outputs the area idle state mapping value.

[0063] In summary, the beneficial effects of this invention are:

[0064] This invention breaks through the bottleneck of misjudgment caused by dimensionality reduction of physical features under pure static conditions in traditional equipment. It innovatively adopts a multi-dimensional recognition architecture that combines dynamic and static features with "spatiotemporal feature memory + micro-motion compensation". It uses high-confidence features dynamically stored in the target database to anchor static targets, and is supplemented by cross-validation of micro-motion features such as engine residual vibration. The long-term accuracy of locking static vehicles approaches 100%.

[0065] Breaking away from the traditional hard boundary logic of physical beams that is "black and white", it introduces edge soft decision and non-standard parking probability model, and is supplemented by micro-motion frequency domain sensitivity amplification resonance technology. Even in extreme non-standard parking scenarios such as when only a small part of the vehicle has entered (such as 1 / 5 of the vehicle body), it can still accurately lock the occupancy status and effectively eliminate edge blind spots.

[0066] Abandoning fixed detection thresholds, the system employs a dual-scale adaptive dynamic algorithm with a "fuse-out anti-drift" mechanism. This effectively identifies and filters interference from slowly changing environmental media such as falling snow and accumulated leaves, avoiding the "boiling frog" effect of erroneous learning and ensuring the radar exhibits extremely high environmental robustness under both strong indoor noise and extreme outdoor weather conditions.

[0067] The radio frequency end employs narrow beamwidth and layered filtering to mitigate spatial crosstalk and human-induced interference; the data output end incorporates channel quality awareness and adaptive protocol reconfiguration mechanisms. Even when faced with strong electromagnetic harmonic injection from modern high-density parking garages and high-power charging piles, it can still ensure the absolute reliability of the underlying communication link and prevent state jumps.

[0068] Based on a state-driven event feedback engine, it automatically enters low-frequency sleep mode during prolonged periods of idle time and achieves full-load wake-up within milliseconds when a sudden change signal is detected. This significantly reduces the overall energy consumption and hardware aging rate of high-density cluster deployments without compromising detection accuracy. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the 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 some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a schematic diagram of the overall process framework of the high anti-interference millimeter-wave radar parking space occupancy detection method of the present invention;

[0071] Figure 2 This is a schematic diagram of the system module connection in the high anti-interference millimeter-wave radar parking space occupancy detection method of the present invention;

[0072] Figure 3 This is a schematic diagram of point cloud information of pedestrians in a parking space in a high anti-interference millimeter-wave radar parking space occupancy detection method of the present invention;

[0073] Figure 4 This is a schematic diagram of point cloud information of pedestrians and vehicles in a parking space in a high anti-interference millimeter-wave radar parking space occupancy detection method of the present invention. The red triangle represents a car and the blue dots represent pedestrians.

[0074] Figure 5 This is a schematic diagram of point cloud information of an SUV vehicle in a high anti-interference millimeter-wave radar parking space occupancy detection method of the present invention. Detailed Implementation

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0076] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate several embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0077] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0078] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0079] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; they can refer to the internal communication of at least two elements or the interaction relationship of at least two elements, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0080] Please see Figure 1 - Figure 5 One embodiment of the present invention is a high anti-interference millimeter-wave radar parking space occupancy detection system, the core of which is the coordination of a radio frequency sensing unit, a main control processing module, and an active anti-interference physical communication link based on power line carrier (PLC).

[0081] Specifically, the radio frequency sensing unit in actual engineering uses a 24GHz FMCW (Frequency Modulated Continuous Wave) radar single chip. The spatial detection range limitation mechanism is achieved by integrating a narrow-beam microstrip patch antenna array with a horizontal angle of 15° and a vertical angle of 30° into the radar radio frequency front end. This narrow-beam antenna, through physical reconstruction, limits the energy of the main lobe of the beam to cover only the geometric boundary range of a single parking space, thereby physically cutting off the overlap of radio frequency signals between adjacent parking spaces and forming an independent physical echo link.

[0082] When the radar chip transmits a frequency sweep signal, the signal is transmitted via a physical echo link to the ground or vehicle chassis in the parking area and then returns. The main control processing module drives a high-precision ADC (analog-to-digital converter) to continuously acquire the intermediate frequency difference signal and performs pre-processing digital bandpass filtering and mean noise reduction mapping, thereby outputting a continuous target airspace signal stream on the system bus that has been freed from high-frequency random noise interference.

[0083] After receiving the continuous target spatial signal stream, the main control processing module inputs it into the signal domain multidimensional joint analysis unit. The specific calculation process is as follows:

[0084] Spatial distance parameter extraction: First, a one-dimensional fast Fourier transform (Range-FFT) is performed on the time-domain difference frequency signal within a single frequency sweep cycle. The system retrieves the spectral peak coordinates from the one-dimensional spectrum and then uses the formula... Where c is the speed of light. Where is the peak frequency and K is the frequency sweep slope, the precise absolute physical distance between the target and the radar equipment is calculated and the spatial distance parameter is formed.

[0085] Motion velocity parameter extraction: Next, a two-dimensional fast Fourier transform (Doppler-FFT) is performed on the multidimensional matrix corresponding to multiple consecutive frequency-modulated sweep cycles (Chirps), or the phase difference rate of change at the peak of the spectral line in the same distance bin is calculated between adjacent Chirp sweep cycles. The phase difference is then utilized. Based on the relationship with the Doppler frequency shift, the radial velocity of the target relative to the radar is calculated, forming the velocity parameter.

[0086] Reflection intensity parameter extraction: The system locks the distance unit corresponding to the peak coordinate of the spectral line, and performs complex modulus square integral calculation on the energy density in the peak and its adjacent preset number of frequency point units, such as two on the left and right, thereby calculating the radar cross section equivalent energy value of the target and forming the reflection intensity parameter.

[0087] Frequency domain distribution parameter extraction: In order to distinguish targets with different physical properties, the system further performs statistical analysis on the spectral morphology of Range-FFT. By calculating the spectral width dispersion at a 3dB decrease in spectral peak, the spectral steepness, and the clustering dispersion of the Doppler spectrum, the frequency domain distribution parameters of the target are extracted.

[0088] For example, when a real car drives into a parking space, due to the wide metal surface and varied shape of the car chassis, the constructed feature vector group shows that: the distance parameter is stable in the range of the installation height (e.g., 2.5 meters) minus the car height; the speed parameter shows a continuous deceleration discrete trajectory; the reflection intensity parameter is extremely high, and the RCS integral value is usually greater than 10 dBsm; the frequency domain distribution parameter shows multi-peak clustering and a wide spectral width, due to the presence of multiple strong reflection points on the car chassis.

[0089] When a pedestrian passes by a parking space, the characteristic vector group shows the following: the distance parameter changes rapidly; the speed parameter fluctuates irregularly with the human walking frequency; the reflection intensity parameter is extremely weak, and the RCS integral value is usually less than -5dBsm; the frequency domain distribution parameter shows a single-peaked dispersion, extremely narrow spectral width, and is very easy to dissipate over time.

[0090] The main control module performs maximum and minimum value normalization on the four types of heterogeneous physical parameters extracted above: spatial distance, motion speed, reflection intensity, and frequency domain distribution. Then, it concatenates and splices them according to the system's preset weight coefficients, such as distance 0.3, speed 0.2, intensity 0.3, and frequency domain 0.2, and finally constructs a multi-dimensional physical feature vector group that represents the absolute physical properties of the current target online.

[0091] It is worth mentioning that, in this embodiment, in order to adapt to the ever-changing and complex parking lot environment, such as outdoor wind and snow and indoor clutter, the system establishes an environmental situation awareness and threshold adaptive interaction model. In the system kernel, an initial high-dimensional detection threshold matrix is ​​pre-set, which includes the lower limit of the basic distance boundary, the energy reflection noise floor, and the Doppler velocity tolerance envelope.

[0092] During operation, the system background runs two sets of sliding time windows with different time scales in parallel:

[0093] Short-term dynamic learning time window: The time span is configured in minutes to capture and respond to transient environmental physical disturbances, such as sudden strong winds, plastic bags temporarily floating on the ground, and fallen leaves.

[0094] Long-term historical evolution time window: The time span is configured in hours or days to depict the gradual change of the absolute foundation of physical space.

[0095] The system extracts the slope of the bottom noise fluctuation in the target airspace signal stream under vehicle-free conditions, and the differential comparison algorithm calculates the deviation gradient vector between the first noise floor feature value output by the short-term window and the second noise floor feature value output by the long-term window in real time.

[0096] In outdoor parking scenarios, when continuous snowfall or gradual accumulation of fallen leaves occurs, the physical surface height and reflectivity of the ground change extremely slowly. Because this change is gradual, the short-term dynamic learning window continuously absorbs this change, causing the deviation gradient vector to exhibit a continuous unidirectional increasing trend. When the cumulative change in this deviation gradient reaches the system's preset safety physical tolerance boundary limit, such as when the simulated snow thickness exceeds 10 centimeters, the anti-environmental drift circuit breaker interaction mechanism is triggered. The main control module immediately generates a blocking signal, cutting off the data flow of the deviation gradient vector to the update module, forcibly freezing the current detection threshold update, and outputting an "abnormal environmental occlusion" warning signal to the cloud.

[0097] If the gradient deviates from the safe boundary, the gradient vector is input to an adaptive filter, which derives the environmental compensation weight matrix for each physical dimension. This matrix multiplication is then used to update the initial high-dimensional detection threshold matrix, thereby dynamically generating an adaptive detection constraint matrix that matches the current environmental evolution. This allows the decision envelope to dynamically expand or contract in response to environmental changes, where expansion raises the threshold to suppress noise interference, and contraction lowers the threshold to improve sensitivity.

[0098] It should be noted that, in this embodiment, the feature matching and state discrimination feedback loop established by the main control module uses the adaptive detection constraint matrix as the dynamic decision boundary. First, the real-time multi-dimensional physical feature vector group is substituted into this matrix for boundary truncation, filtering out pure clutter signals that do not conform to vehicle characteristics, and generating a basic target existence probability value.

[0099] To address the missed detections caused by static vehicle signal attenuation, the system initiates a cross-verification interaction between a spatiotemporal feature continuity correlation mechanism and a target micro-motion feature compensation model. The specific interaction data flow is as follows:

[0100] Dynamic trajectory locking and stack storage: When the distance and velocity parameters in the multi-dimensional physical feature vector group represent a target moving towards the center of the parking space with a continuously decelerating trajectory, and it is determined to be a vehicle entering the parking space event, the system captures the high-confidence dynamic multi-dimensional feature vector set within the time window in real time, including its huge RCS feature and continuous Doppler spectrum at the time of entering the parking space, and pushes it into the internal spatiotemporal storage stack.

[0101] Static spatiotemporal anchoring binding: When the motion velocity parameter in the feature vector group is detected to have completely decayed and stabilized in the zero limit range (determining that the vehicle has come to a complete stop), the system activates static spatiotemporal anchoring mapping, which converts the two-dimensional feature parameters that have undergone dimensionality reduction under the current pure static state into zero. At this time, the radar can only measure the distance and static reflection intensity, while the speed and dynamic spectrum are returned to zero. It is uniquely bound and associated with the high-confidence dynamic features stored in the stack before they are added to the database. Through this continuous memory of historical features, the system applies a strong temporal continuity positive gain weight to the basic target existence probability value.

[0102] Micro-motion confidence compensation: When the vehicle is stationary, the system triggers a micro-motion feature compensation model. The radar periodically emits specific high-density timing pulses. The feature extraction unit directionally captures and extracts the micro-Doppler frequency shift parameter (Micro-DopplerCluster) generated by residual engine vibration after the car is turned off and deformation and vibration of the passenger compartment caused by passengers leaving the vehicle when the vehicle is stationary. This micro-motion frequency shift parameter is injected into the static spatiotemporal anchoring mapping as a confidence compensation factor. Once a match is successful, the system applies an absolute confidence multiplier to the current probability value, thereby locking the decision probability at the highest bit.

[0103] Since the target is identified as a high-confidence vehicle through spatiotemporal anchoring, and the system continuously outputs the area occupancy status, to prevent the long-term stationary vehicle from being mistakenly filtered out as a new fixed building background by the aforementioned "short-term dynamic learning time window," the system sends a spatial coordinate exemption command to the threshold adaptive model through a mutual exclusion decoupling feedback channel. In response to this command, the threshold learning model, during its sliding window iterative update, forcibly masks and removes all echo data within the three-dimensional physical coordinate domain of the vehicle, including the corresponding distance bin and angle bin, by setting a weight mask. This prevents the assimilation and transfer of vehicle features to background noise features, ensuring the absolute stability of ultra-long-term static parking recognition.

[0104] Finally, the system inputs the weighted and compensated composite probability value into the Schmitt trigger. When the composite probability value rises above the set high-bit trigger threshold and remains consistent for a set anti-jitter time window, such as 3 seconds, for several consecutive frames, the trigger state flips, and the system outputs the occupied state mapping value (logic "1") to the bus. Similarly, when the composite probability value falls below the low-bit release threshold, the idle state mapping value of the output area flips (logic "0").

[0105] In actual parking lot operations, non-standard parking scenarios are frequently encountered, such as vehicles crossing lines, eccentric parking, or vehicles only partially entering the parking space (e.g., occupying only 1 / 5 of the parking space area). To address these issues, the system incorporates edge target soft-decision interaction logic and a cross-device edge computing network based on an active anti-interference communication link.

[0106] Specifically, when a signal conforming to a preset reflection intensity parameter is continuously emitted from the critical boundary region of the defined horizontal 15° physical beam, it indicates the presence of high-density metallic reflection, and the edge probability decision engine is activated. The engine retrieves the historical state time series of the boundary coordinates and integrates the real-time reflection energy to dynamically calculate the actual occupancy probability metric of the boundary signal.

[0107] It should be noted that during the parallel calculation of this edge metric, the system generates a highly sensitive frequency domain analytical window based on the feature vector set, specifically pointing to the specific spatial coordinates of the critical boundary. If a specific micro-Doppler frequency domain distribution parameter characterizing physical micro-motion (such as edge airflow caused by a car door closing or edge metal deformation fluctuation) is successfully matched within this window, the system directly generates a privilege overriding instruction, bypassing the conventional probability layering calculation process, and forcibly compensates the actual occupancy probability metric to the absolute certainty threshold.

[0108] It should also be noted that if the actual occupancy probability metric of the edge target falls within the preset critical fuzzy decision range, such as 45% to 60%, the single-node radar cannot determine whether the cross-line vehicle is parked in its own parking space or the next parking space. The master control node suspends the current decision and encapsulates a data request message containing the current parking space spatial coordinate parameters and a multi-dimensional feature vector set.

[0109] Because this system employs an active anti-interference physical communication link with embedded noise floor awareness at the physical transmission layer, the underlying communication control unit scans and evaluates the strong harmonic electromagnetic noise injected into the current physical bus during the high-frequency switching operation of high-power charging piles in the garage before packet transmission. When an electromagnetic congestion risk is detected, it adaptively switches to a high-order cyclic redundancy check protocol with cascaded error correction capabilities and activates multiple time diversity retransmission queues, thereby establishing a 100% reliable low-error-rate PLC wired communication link even in a strong interference environment.

[0110] The master control node broadcasts data request messages to adjacent subordinate radar detection nodes in the physical space via this highly reliable, active anti-jamming physical communication link. Upon receiving the request, the adjacent detection nodes exchange feature observation data for the overlapping area at the underlying level, such as whether the neighboring radar detected another 4 / 5 of the vehicle's features at its edge. The master control node collects this collaborative data and performs joint weighted fusion of the multi-source feature data matrix. Once the neighboring radar confirms the presence of a large, high-confidence vehicle body at its edge and determines that the vehicle is actually occupying the current parking space across the lane, the system breaks through the blind spot of a single node, accurately flips the display, and outputs the calibrated area occupancy status mapping value.

[0111] In another embodiment, the system achieves extremely low operating power consumption through a state-driven and channel-coordinated scheduling model. A state flip-flop timer is internally set up to continuously monitor the steady-state duration of the region's spatial occupancy state mapping value.

[0112] When a parking space remains vacant for an extended period, and the steady-state duration of the state mapping value exceeds the system's preset idle tolerance period, the system kernel considers the current state to be in a safe phase without event triggering and immediately issues a power management command. The system forcibly reduces the frequency modulation and sweep cycle of the RF sensing unit from a high frame rate to a low-frequency detection sleep mode, such as 0.5Hz, sampling once every 2 seconds. Simultaneously, it shuts down the high-speed signal processing bus and most feature understanding algorithm units, resulting in a power consumption reduction of over 90% for the entire system.

[0113] In this dormant detection mode, the front-end analog envelope detector or basic spatial distance parameter extraction unit maintains continuous monitoring with extremely low power consumption. Once a car drives into the parking space, causing a sudden change in the basic spatial distance parameter, or when the envelope detector detects a signal with the characteristics of a sudden rising edge extracted from the RF energy flow, the hardware interrupt controller is triggered, generating a global wake-up interrupt within milliseconds. This interrupt drives a full-load reset of all disabled modules and the full feature analysis link in the system, smoothly restoring to a high frame rate sampling state with a millisecond-level response speed, and entering the full feature analysis process. This ensures that the dynamic trajectory feature data stored in the database is not lost, achieving a lossless fusion of energy efficiency and detection accuracy.

[0114] In addition, in order to more intuitively demonstrate the collaborative process of the core mechanisms of this invention, such as multi-dimensional feature fusion, dual-scale adaptive threshold, edge soft decision and full-link anti-interference, this invention also provides examples of complex working conditions in underground parking lots for new energy vehicles.

[0115] Scenario: In an underground parking garage, parking space A1-273 has a high-power DC fast charging station deployed in the adjacent parking space A1-274 on the left. The radar equipment is installed directly above the parking space and connected to the gateway via a 2-core power line carrier (PLC).

[0116] Detailed operation steps:

[0117] Phase 1: Idle sleep and background noise inspection, 02:00 AM, no vehicle in operation;

[0118] When a parking space remains vacant for an extended period, the main control module triggers an event-driven energy efficiency management mechanism, reducing the sampling frequency of the 24GFMCW radar to a low-frequency sleep mode.

[0119] During this period, the environmental situational awareness model continued to operate. Due to the drop in temperature at night and the shutdown of the garage ventilation system, the ambient noise level was extremely low. The system recorded this extremely pure physical basis into the "long-term absolute history window".

[0120] Phase Two: Dynamic Entry and Feature Spatiotemporal Memory, 08:30 AM, Vehicles Enter;

[0121] An all-electric sedan began to drive into parking space A1-273. A sudden change occurred in the radar's basic spatial distance parameter, triggering a hardware interrupt. The system smoothly woke up to high-frequency full-load sampling mode within milliseconds.

[0122] During the dynamic process of a vehicle reversing into a parking space, the radar extracts a huge radar cross section (RCS), a continuously decelerating radial velocity, and wide-band spectral characteristics.

[0123] The main control module timestamps this high-confidence "dynamic multidimensional feature vector set" and pushes it into the internal spatiotemporal storage stack.

[0124] Phase 3: Static anchoring and micro-motion confirmation, 08:31 AM, the vehicle came to a complete stop and the engine was turned off;

[0125] Once the vehicle comes to a complete stop, the radial velocity parameter decays to zero. Addressing the weakness of traditional ultrasonic radar, which is prone to missed detections due to vehicle stationaryness and weakened features, this system triggers a "spatiotemporal feature continuity correlation mechanism."

[0126] The system uniquely spatiotemporally anchors and binds the current static two-dimensional features to the dynamic high-confidence features in the stack.

[0127] At the same time, the driver opens the door, gets out of the car, and closes the door. The radar frequency domain high-sensitivity resolution window captures the micro-Doppler frequency shift caused by the micro-vibration of the car body metal at the moment of closing the door. The system directly increases the occupancy probability to the absolute certainty threshold, outputs the "parking space occupancy" status through the PLC link, and classifies the three-dimensional coordinates of the vehicle into the "environmental feature immune zone".

[0128] Phase 4: Anti-drift and link preservation under strong interference environment; at 12:00 noon, the charging piles were running at full capacity.

[0129] A large SUV drove into the adjacent parking space on the left and started a high-power fast charger, causing electromagnetic harmonics and hot air to rise sharply in the garage.

[0130] At the underlying physical link layer: the communication control unit detects strong harmonic interference on the 2-core physical bus and immediately switches the PLC protocol to the high-level cascade error correction and multiple diversity retransmission mode to thoroughly filter out underlying errors and ensure that the occupancy status data of parking space A1-273 does not experience any communication jumps.

[0131] At the radio frequency analysis layer: due to the large size of the SUV and its irregular cross-line parking, it only occupies 1 / 5 of the parking space size, and the edge of its body intrudes into the critical detection area of ​​the narrow beam at a horizontal angle of 15° for parking space A1-273.

[0132] Short-term dynamic learning time windows revealed a sharp increase in edge noise. However, since the sedan in parking space A1-273 had been classified as an environmental feature immune zone, the system triggered a mutual exclusion decoupling feedback channel, shielding the background filtering of the core area and ensuring that the original sedan was not missed. At the same time, the edge probability decision engine calculated that the actual occupancy probability of the SUV crossing the line was only 15%, which did not reach the warning value. Furthermore, through joint weighted calculation interaction between the PLC and the left radar, it was confirmed that the main body of the SUV was in the left parking space. Thus, the crosstalk of the neighboring vehicle crossing the line was successfully filtered out by utilizing the dual anti-interference architecture of software and hardware.

[0133] Phase 5: State transition and low-power recovery, 18:00, the vehicle departs;

[0134] As the car started and drove away, the radar once again detected the reverse speed parameters and continuous spatial distance changes.

[0135] Once the vehicle has completely left the detection area, the feature vector group matching fails. The overall probability value rapidly drops below the release threshold.

[0136] The system uses a smooth hysteresis model to output a "parking space vacant" status after confirming the idle state and maintaining the anti-jitter time window. Subsequently, the state flip timer restarts, driving the system to smoothly reduce its frequency and switch back to low-frequency sleep mode, achieving adaptive voltage drop for overall power consumption.

[0137] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any variations or substitutions conceived without inventive effort should be included within the scope of protection of the invention. Therefore, the scope of protection of the invention should be determined by the scope defined in the claims.

Claims

1. A method for detecting parking space occupancy using millimeter-wave radar with high anti-interference capabilities, characterized in that: Includes the following steps: S1: The radio frequency sensing unit transmits a frequency-modulated continuous wave signal and constructs a physical echo link containing a single target area through a spatial detection range limitation mechanism. It continuously collects raw echo data and performs pre-filtering and noise reduction mapping to output a continuous target spatial signal stream. S2: Receive the target spatial signal stream, perform multidimensional joint analysis in the signal domain, and extract and construct a multidimensional physical feature vector group representing the target; S3: Establish an environmental situation awareness and threshold adaptive interaction model, simultaneously extract the slope of environmental bottom-level noise fluctuation in the target spatial signal stream, and combine multi-scale dynamic threshold calculation to generate an adaptive detection constraint matrix that matches the evolution of the current physical environment base. S4: Establish a feature matching and state discrimination feedback loop, use the adaptive detection constraint matrix as the dynamic decision boundary, perform real-time mapping and clutter truncation on the multi-dimensional physical feature vector group, and trigger the spatiotemporal feature continuity correlation mechanism and micro-motion feature compensation model to perform temporal cross-verification at the feature level, and comprehensively calculate and output the regional space occupancy state mapping value accordingly. S5: Establish a state-driven and channel-coordinated scheduling model, use the area space occupancy state mapping value as a feedback control parameter to adjust the detection sampling period of the radio frequency sensing unit; at the same time, transmit the area space occupancy state mapping value after link layer redundancy verification to the outside through an active anti-interference physical communication link with an embedded channel quality assessment mechanism.

2. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 1, characterized in that: The spatiotemporal feature continuity correlation mechanism and the micro-motion feature compensation model in step S4 specifically include the following steps: When the multi-dimensional physical feature vector group represents the target in a continuous approaching state, the high-confidence dynamic multi-dimensional feature vector set within the time window is extracted and recorded into the spatiotemporal storage stack. When the motion velocity parameter in the feature vector set is detected to decay to the zero limit range, a static spatiotemporal anchoring mapping is established to uniquely bind and associate the target's current dimensionality-reduced two-dimensional spatial feature parameter with the dynamic multidimensional feature vector set in the spatiotemporal storage stack. During the periodic verification phase, the driving feature extraction unit captures the micro-Doppler frequency shift parameter of the target in a static state and injects the frequency shift parameter as a compensation factor into the static spatiotemporal anchoring mapping for cross-validation to maintain the effective locking of the regional spatial occupancy state mapping value.

3. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 2, characterized in that: The process of real-time mapping and clutter truncation of the multi-dimensional physical feature vector group in step S4 includes the following steps: When a high-frequency signal conforming to a preset reflection intensity parameter is continuously extracted within the effective detection critical boundary region of the physical echo link, the historical state time series and real-time reflection energy integral of the critical boundary region are extracted. The historical state time series and the real-time reflection energy integral are input into the edge probability decision engine to dynamically calculate the actual occupancy probability metric of the signal within the critical boundary. When the actual occupancy probability metric exceeds the system's preset warning probability envelope, the signal within the critical boundary is classified as an effective target intrusion, and the area space occupancy status mapping value is updated.

4. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 3, characterized in that: The edge probability decision engine includes a micro-motion feature orientation verification mechanism, which includes: During the parallel cycle of the edge probability decision engine calculating the actual occupancy probability metric, the system generates a frequency domain high-sensitivity analytical window pointing to the specific spatial coordinates of the critical boundary based on the feature vector group; In response to successfully matching a specific frequency domain distribution parameter characterizing physical micro-motion within the high-sensitivity resolution window in the frequency domain, the system generates a privilege override instruction, which is used to directly compensate the actual occupancy probability metric to the absolute confidence threshold.

5. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 1, characterized in that: It also includes the construction of mutually exclusive channels between target states and the environment base across different levels: When the target completes a unique binding association through the static spatiotemporal anchoring mapping and the system continuously outputs the occupancy status mapping value, a spatial coordinate exemption command is sent to the environmental situation awareness and threshold adaptive interaction model through the mutual exclusion decoupling feedback channel. In response to the exemption command, the environmental situation awareness and threshold adaptive interaction model forcibly masks all signal data parameters in the physical coordinate domain of the target by setting a weight mask during the iterative update of the short-term dynamic learning time window.

6. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 1, characterized in that: The specific process of generating an adaptive detection constraint matrix that matches the evolution of the current physical environment substrate by combining multi-scale dynamic threshold calculation includes: An initial high-dimensional detection threshold matrix is ​​pre-set, which includes the lower limit of the distance boundary, the energy reflection substrate, and the velocity tolerance envelope; Based on the slope of the underlying environmental noise fluctuation, the transient environmental disturbance component and the long-term absolute background component are extracted. The transient environmental disturbance component and the long-term absolute background component are input into an adaptive filter to deduce the environmental compensation weight matrix corresponding to each dimension of distance, intensity and velocity. The initial high-dimensional detection threshold matrix is ​​updated by matrix multiplication or nonlinear mapping using the environmental compensation weight matrix, so that the decision envelope of the adaptive detection constraint matrix can adaptively expand or contract dynamically in accordance with the evolution of the physical environment substrate.

7. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 1, characterized in that: The method for comprehensively calculating and outputting the regional space occupancy status mapping value in S4 includes the following steps: Substitute the current multi-dimensional physical feature vector group into the adaptive detection constraint matrix to perform boundary verification and generate a basic target existence probability value. If the target successfully triggers the spatiotemporal feature continuity association mechanism, a temporal continuity positive gain weight is applied to the basic target existence probability value; During the time-domain cross-verification stage, if the micro-motion feature compensation model successfully matches the frequency shift component characterizing the vehicle's physical micro-motion, the system applies an absolute confidence multiplier to the current probability value. The final comprehensive probability value after weighting and compensation is input into a Schmitt trigger or a time-smoothing hysteresis model. When the comprehensive probability value exceeds the high-bit trigger threshold and is maintained within the set anti-shake time window, it is flipped and the determined area occupancy state mapping value is output. When it falls to the low-bit release threshold, the area idle state mapping value is output.

8. The high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 3, characterized in that: The active anti-interference physical communication link includes interactive control of noise floor sensing and protocol reconstruction: Before initiating the handshake for sending data frames on the physical link, the underlying communication control unit first performs a scan and evaluation of the current physical bus electromagnetic noise floor and channel spectrum purity. When the quantization results of the scan evaluation indicate that there is a high-intensity harmonic injection or channel blocking risk in the link, the adaptive protocol reconstruction mechanism is triggered to switch the encapsulation format of the underlying data frame to a high-order cyclic redundancy check protocol with cascaded error correction capability, and simultaneously open multiple time diversity retransmission queues to combat sudden bit errors.

9. A high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 8, characterized in that: Based on the aforementioned active anti-interference physical communication link extension, a multi-node spatial awareness fusion network interaction logic is constructed: When the actual occupancy probability metric value output by the edge probability decision engine falls to the system's preset fuzzy decision range, the main control processing node suspends the current decision and encapsulates a data request message containing the current spatial coordinate parameters and feature vector set. The data request message is broadcast to the physically adjacent subordinate probe nodes via the physical communication link; Collect feature observation sets for the spatially overlapping region returned by each subordinate detection node, perform joint weighted fusion operation of multi-source feature data matrices to break the blind spot of single-node spatial perspective and output calibrated regional spatial occupancy status mapping value.

10. A high anti-interference millimeter-wave radar parking space occupancy detection method according to claim 9, characterized in that: The specific method for generating an adaptive detection constraint matrix that matches the evolution of the current physical environment basis includes the following steps: The system operates a short-term dynamic learning time window to respond to transient media disturbances, and a long-term historical evolution time window to characterize the absolute basis of physical space. The first noise floor feature value output by the short-term dynamic learning time window and the second noise floor feature value output by the long-term historical evolution time window are extracted. The deviation gradient vector between the two in the time series is measured in real time through the difference comparison algorithm. The threshold parameter in the adaptive detection constraint matrix is ​​dynamically configured according to the deviation gradient vector.

Images