Radar identification and collision early warning method and system for two-wheeled vehicle

Abstract

The invention relates to the technical field of radar identification, and discloses a radar identification and collision early warning method and system for a two-wheeled vehicle, and the method comprises the steps: carrying out the synchronous collection of a radar echo signal and a vehicle body motion signal in a backward detection region of an electric two-wheeled vehicle, doppler frequency shift data, radar reflection sectional area data, radial distance data, reflection signal intensity data and motion state data are obtained; carrying out vehicle body motion compensation and time sequence variance calculation to obtain first static and dynamic target classification information; carrying out reflection characteristic calculation to obtain a candidate target data set; and performing consistency verification on the radial distance change rate and the Doppler velocity to obtain second static and dynamic target classification information, and outputting a target early warning signal in combination with the second static and dynamic target classification information. And the timeliness and reliability of collision early warning are improved.

Description

Technical Field

[0001] The present invention relates to the field of radar recognition technology, and in particular to a radar recognition and collision warning method and system for a two-wheeled vehicle. Background Art

[0002] Electric two-wheelers are prone to collisions with other vehicles on urban roads, with rear-end collisions, in particular, showing an increasing incidence. Traditional electric two-wheelers lack effective collision warning systems, forcing drivers to rely primarily on vision and hearing to detect approaching vehicles from behind. This passive approach often fails to detect potential collision threats at night, in inclement weather, or when attention is distracted.

[0003] Existing on-board radar collision warning technology is primarily used for four-wheeled vehicles. Its algorithms and system designs are optimized for the motion characteristics and usage scenarios of four-wheeled vehicles, but direct application to electric two-wheelers presents numerous technical obstacles. Electric two-wheelers frequently change their body posture during driving, tilt when turning, and employ complex maneuvers. These motion characteristics significantly interfere with radar signals, preventing traditional radar systems from accurately identifying and tracking targets behind them. Summary of the Invention

[0004] The present invention provides a radar identification and collision warning method and system for two-wheeled vehicles, which effectively reduces the misjudgment rate of static and dynamic targets, significantly enhances the ability to identify targets in different motion states, and improves the timeliness and reliability of collision warnings.

[0005] In a first aspect, the present invention provides a radar identification and collision warning method for a two-wheeled vehicle, the radar identification and collision warning method for a two-wheeled vehicle comprising: The radar echo signal and vehicle motion signal in the rear detection area of ​​the electric two-wheeled vehicle are synchronously collected to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data and motion status data; Performing vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information; performing reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information to obtain a candidate target data set; Based on the radial distance data, the consistency of the radial distance change rate and the Doppler velocity of the candidate target data set is verified to obtain second static and dynamic target classification information, and a target warning signal is output in combination with the second static and dynamic target classification information.

[0006] In combination with the first aspect, in a first implementation of the first aspect of the present invention, the radar echo signal and the vehicle body motion signal within the rear detection area of ​​the electric two-wheeled vehicle are synchronously collected to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data, and motion state data, including: Data collection is performed on the radar echo signal within the backward detection area received by the millimeter wave radar to obtain the radar echo signal; Based on the installation position relationship between the radar coordinate system and the vehicle coordinate system, the radar echo signal is subjected to frequency domain processing and coordinate system conversion calibration to obtain Doppler frequency shift data, radial distance data, radar reflection cross-sectional area data, and reflection signal intensity data; The three-axis acceleration signal and the three-axis angular velocity signal output by the IMU inertial measurement unit in the electric two-wheeled vehicle are synchronously collected to obtain a vehicle body motion signal, and the vehicle body motion signal is filtered to obtain motion state data.

[0007] In combination with the first aspect, in a second implementation of the first aspect of the present invention, performing vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information includes: Extracting the longitudinal velocity component, the lateral velocity component and the angular velocity component from the motion state data; Calculating the vehicle body Doppler frequency shift component based on the longitudinal velocity component and the target azimuth angle data to obtain a first Doppler frequency shift component corresponding to each target position; Setting segmented compensation coefficient data based on the longitudinal velocity component; performing a product operation on the first Doppler frequency shift component and the segmented compensation coefficient data to obtain a second Doppler frequency shift component, and performing a difference calculation on the Doppler frequency shift data and the second Doppler frequency shift component to obtain relative Doppler frequency shift data; Converting the relative Doppler frequency shift data into a radar operating wavelength to obtain Doppler velocity data; A time series variance calculation is performed based on the relative Doppler frequency shift data to obtain first static and dynamic target classification information.

[0008] In combination with the first aspect, in a third implementation of the first aspect of the present invention, calculating the vehicle body Doppler shift component based on the longitudinal velocity component and the target azimuth angle data to obtain a first Doppler shift component corresponding to each target position includes: Extracting target azimuth data from the radial distance data and the radar reflection cross-sectional area data; Performing Doppler frequency domain conversion on the longitudinal velocity component according to the radar operating frequency to obtain a reference Doppler frequency shift value corresponding to the longitudinal motion of the electric two-wheeled vehicle; Performing coordinate system correction on the target azimuth angle data based on the geometric relationship between the radar installation position and the longitudinal axis of the vehicle body to obtain detection azimuth angle data; A cosine function operation is performed on the reference Doppler frequency shift value and the detection azimuth angle data to obtain a first Doppler frequency shift component corresponding to each target position.

[0009] In combination with the first aspect, in a fourth implementation of the first aspect of the present invention, performing time series variance calculation based on the relative Doppler shift data to obtain first static and dynamic target classification information includes: Extracting M frames of continuous data from the relative Doppler frequency shift data to obtain M frames of relative Doppler frequency shift time series data corresponding to each target; Perform mean calculation and variance calculation on the M frames of relative Doppler frequency shift time series data to obtain the Doppler frequency shift mean and Doppler frequency shift stability coefficient of each target; Performing a threshold comparison on the Doppler frequency shift stability coefficient based on a preset static-dynamic discrimination threshold to obtain classification data of stationary targets having a stability coefficient less than the static-dynamic discrimination threshold and classification data of moving targets having a stability coefficient greater than or equal to the static-dynamic discrimination threshold; The M frames of relative Doppler frequency shift time series data are subjected to an inter-frame change rate analysis to obtain Doppler gradient change rate data, and the stationary target classification data is reconfirmed based on the judgment condition that the absolute values ​​of the gradient change rates of N consecutive frames are all less than a preset target value to obtain the first stationary and moving target classification information.

[0010] In combination with the first aspect, in a fifth implementation of the first aspect of the present invention, performing reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data based on the first static and dynamic target classification information to obtain a candidate target dataset includes: Normalizing the radar reflection cross-sectional area data based on the radial distance data and the road surface reflection coefficient to obtain normalized reflection cross-sectional area data; Performing frequency domain transformation and spectrum feature extraction on the reflected signal intensity data to obtain reflected signal spectrum feature data; The normalized reflection cross-sectional area data and the reflection signal spectrum feature data are feature matched with a preset guardrail target feature library, a traffic sign target feature library, and a vehicle target feature library to obtain false target identification mark data corresponding to each target; A fusion discrimination is performed based on the false target identification mark data and the first static and dynamic target classification information to obtain reflection feature discrimination data, and a candidate target data set is determined based on the reflection feature discrimination data.

[0011] In combination with the first aspect, in a sixth implementation of the first aspect of the present invention, performing consistency verification of the radial range change rate and the Doppler velocity on the candidate target data set based on the radial range data to obtain second static and dynamic target classification information, and outputting a target warning signal in combination with the second static and dynamic target classification information, includes: performing inter-frame differential analysis on radial distance data of each target in the candidate target data set to obtain a first radial distance change rate of each target, and performing angle compensation based on a lateral offset angle of the target relative to the longitudinal axis of the vehicle body to obtain a second radial distance change rate; Determining first target data that meets a consistency condition and second target data that does not meet the consistency condition according to the second radial distance change rate and Doppler velocity data of the corresponding target; Performing a time sequence consistency check on the first target data that meets the consistency condition to obtain third target data; Performing secondary verification on the second target data that does not meet the consistency condition to obtain fourth target data that passes the secondary verification, and merging the third target data with the fourth target data to obtain second static and dynamic target classification information; State tracking and collision warning are performed based on the second static and dynamic target classification information, and a target warning signal is output.

[0012] In combination with the first aspect, in a seventh implementation manner of the first aspect of the present invention, determining the first target data that meets the consistency condition and the second target data that does not meet the consistency condition based on the second radial range change rate and the Doppler velocity data of the corresponding target includes: Calculating the absolute value of the difference between the second radial range change rate of each target in the candidate target data set and the corresponding Doppler velocity data to obtain physical quantity consistency deviation data of each target; The physical quantity consistency deviation data of each target is compared with a preset consistency verification threshold to obtain first target data that meets the consistency condition and second target data that does not meet the consistency condition.

[0013] In combination with the first aspect, in an eighth implementation of the first aspect of the present invention, performing state tracking and collision warning based on the second static and dynamic target classification information and outputting a target warning signal includes: Initializing the state vector of the moving target data in the second static and dynamic target classification information to obtain initial state vector data, and establishing a state transfer matrix and an observation matrix of an extended Kalman filter; Executing a prediction step and an update step of an extended Kalman filter based on the initial state vector data to obtain target trajectory tracking data, and performing linear extrapolation on the target trajectory tracking data to obtain target future position prediction data; Calculating a relative motion trajectory based on the target future position prediction data and the motion state data to obtain minimum approach distance data between the target and the electric two-wheeled vehicle and estimated time data for reaching the minimum distance; Based on the collision judgment conditions that the minimum approach distance data is less than a safety distance threshold and the estimated time data is less than a warning time threshold, a collision warning is performed and a target warning signal is output.

[0014] In a second aspect, the present invention provides a radar identification and collision warning system for a two-wheeled vehicle, the radar identification and collision warning system for a two-wheeled vehicle comprising: A synchronous acquisition module is used to synchronously acquire radar echo signals and vehicle motion signals within the rear detection area of ​​the electric two-wheeled vehicle to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data, and motion status data; a target classification module, configured to perform vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information; a reflection feature calculation module, configured to perform reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information, to obtain a candidate target data set; An output module is used to verify the consistency of the radial distance change rate and Doppler velocity of the candidate target data set based on the radial distance data, obtain second static and dynamic target classification information, and output a target warning signal in combination with the second static and dynamic target classification information.

[0015] The technical solution provided by the present invention achieves deep signal fusion by synchronously collecting radar echo signals and vehicle motion signals to obtain multi-dimensional information such as Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data, and motion state data. Based on the motion state data obtained by the IMU inertial measurement unit, the Doppler frequency shift data is accurately compensated for vehicle motion, effectively eliminating the interference of the electric two-wheeled vehicle's own motion on the radar signal, solving the problem of target detection errors caused by vehicle motion in traditional technologies, and improving the accuracy of target recognition. A dual verification mechanism of time series variance calculation and radial distance change rate and Doppler velocity consistency verification is used to establish a hierarchical recognition system for the first static and dynamic target classification information and the second static and dynamic target classification information, effectively reducing the misjudgment rate of static and dynamic targets and significantly enhancing the recognition ability of targets in different motion states. By extracting normalized reflection cross-sectional area data and spectral features, combined with pre-set guardrail, traffic sign, and vehicle target feature libraries for feature matching, effective recognition and suppression of false targets are achieved, solving the false alarm problem caused by fixed obstacles and non-vehicle targets in traditional technologies. The extended Kalman filter algorithm is used for multi-target state tracking. Through iterative calculation of state vector initialization, prediction step and update step, smooth tracking of target trajectory and accurate prediction of future position are achieved, which significantly improves the timeliness and reliability of collision warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 Schematic diagram of the steps of a radar identification and collision warning method for a two-wheeled vehicle according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of a radar identification and collision warning system for a two-wheeled vehicle in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] An embodiment of the present invention provides a radar identification and collision warning method and system for a two-wheeled vehicle. The terms "first," "second," "third," "fourth," and so on (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0019] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 An embodiment of a two-wheeled vehicle radar identification and collision warning method according to an embodiment of the present invention includes: Step S1, synchronously collecting radar echo signals and vehicle body motion signals within the rear detection area of ​​the electric two-wheeled vehicle to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data, and motion state data; It is understood that the execution subject of the present invention can be a two-wheeled vehicle radar identification and collision warning system, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0020] Specifically, the 77GHz millimeter-wave radar module at the rear of the electric two-wheeled vehicle boasts high resolution and long-range detection capabilities, enabling real-time detection of radar echo information from objects behind it, regardless of lighting conditions. When a target enters the rearward detection zone, the millimeter-wave radar receiving module samples multiple frames of echo signals in real time, generating a raw data stream containing Doppler shift characteristics, distance variation characteristics, and reflection amplitude variation characteristics. These signals are then input into the signal processing unit for frequency domain demodulation. Fast Fourier transforms are used to extract the target's velocity-related characteristics, namely the Doppler shift, and simultaneously analyze the radial distance from the target to the radar. Considering the spatial angle, tilt offset, and height difference between the radar's mounting position and the vehicle's coordinate system, calibration parameters are used to reposition the radar's coordinate system before target feature extraction. The polar coordinate data measured by the radar is mapped to a unified vehicle coordinate system using a coordinate system transformation matrix, achieving spatial geometric calibration of the radar detection data. This step embeds raw Doppler frequency shift data, multi-channel reflection cross-sectional area data, target distance information, and reflected signal strength into the vehicle's relative coordinate model, eliminating data errors and directional distortion caused by the misalignment of the radar and vehicle axes. Furthermore, vehicle motion data is acquired using an inertial measurement unit (IMU) mounted at the center of the vehicle or on the main control board. This unit synchronously outputs three-axis acceleration and three-axis angular velocity data every tens of milliseconds, representing the vehicle's linear acceleration state in the X, Y, and Z directions and its roll, pitch, and yaw angular velocity information, respectively. To improve the stability and physical consistency of the vehicle motion data, the IMU output is filtered using a first-order low-pass filter or Kalman filter to suppress the high-frequency noise generated by the acceleration sensor under vibrating road conditions. The angular velocity data is then bias-corrected for drift to produce the motion data.

[0021] Step S2: performing vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information; Specifically, the current three-axis motion state information is extracted from the inertial measurement unit (IMU) to obtain the longitudinal velocity component, lateral velocity component, and angular velocity component about the Z axis. The longitudinal velocity reflects the linear velocity behavior of the vehicle when moving forward or backward and is the core variable that determines the target Doppler shift compensation; the lateral velocity and angular velocity jointly determine the trajectory deviation behavior of the vehicle when turning or changing direction. Based on the azimuth angle data between the target and the vehicle body, that is, the angle of the target relative to the longitudinal axis of the vehicle, the vehicle's own longitudinal motion velocity is projected in the direction of the target. The cosine function is used to calculate the Doppler shift effect component of the vehicle in this direction on the radar echo, forming the first Doppler shift component. Because the vehicle's motion stability and measurement error exhibit a nonlinear relationship across different speed ranges, a segmented compensation mechanism is introduced based on the current longitudinal velocity. In low-speed ranges (e.g., 5-15 km / h), the compensation coefficient is increased to 1.2 to enhance the suppression of small velocity errors. In medium-speed ranges (15-25 km / h), the compensation coefficient is set to 1.0 to maintain equivalent compensation. In high-speed ranges (25-35 km / h), the compensation coefficient is set to 0.8 to prevent high-frequency fluctuations from interfering with the compensation results. The first Doppler shift component is multiplied by the segmented compensation coefficient corresponding to that speed range to obtain the compensated second Doppler shift component. This second Doppler shift component is then used as a reference for subtraction of the raw Doppler shift data collected by the radar to obtain the relative Doppler shift caused by the target's actual motion. The relative Doppler shift data is converted to the corresponding Doppler velocity data by converting it to the unit corresponding to the operating wavelength of the radar's center frequency. Statistical analysis is performed on the target's relative Doppler shift data within consecutive time frames. The standard variance model is used to calculate the Doppler variation within a fixed time window, which serves as a measure of the target's frequency shift stability along the time axis. If the variance is below a preset threshold (e.g., 0.85Hz), the target is considered stationary; if it exceeds the threshold, it is considered a moving target with stable frequency shift characteristics.

[0022] Step S3: performing reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information to obtain a candidate target data set; Specifically, the first static and moving target classification information is used as the basis for data screening, and only entities preliminarily determined as moving targets are subjected to reflection characteristic analysis and processing. On this basis, the radar reflection cross-sectional area data of the target is subjected to normalization operation, the radial distance information of the target in the current frame and the preset road surface reflection coefficient are introduced, the original reflection cross-sectional area value is corrected in terms of distance attenuation and environmental influence factor, the reflection intensity is divided by the square of the distance and multiplied by the environmental correction factor, and the normalized reflection cross-sectional area data reflecting the intrinsic radar reflection capability of the target is obtained. The radar reflection signal strength data is subjected to frequency domain transformation operation, the frequency spectrum structure thereof is extracted through fast Fourier transform, and on this basis, key parameters such as the frequency spectrum center frequency offset, the frequency spectrum width and the frequency spectrum symmetry of the target reflection signal are calculated to form a frequency spectrum characteristic data set for target form analysis. The frequency spectrum symmetry is an important index for identifying linear structure targets (such as guardrails) and non-linear structure targets (such as vehicles), because linear reflectors have strong frequency spectrum symmetry, and real vehicle targets have irregularly distributed frequency spectrum due to the multi-faceted reflection structure. The normalized reflection cross-sectional area data and the frequency spectrum characteristic data are used as joint input, and a feature matching operation is performed with a built-in typical target characteristic database. The database has preset typical reflection characteristic distribution ranges of guardrail type, traffic sign type, parked vehicle type and moving vehicle type, and defines the normalized RCS range, the frequency spectrum symmetry threshold and the reflection form stability interval thereof, respectively. In the matching process, a weighted similarity calculation method is used to generate a set of matching scores of each target with each type of target in the database, and the corresponding false target recognition marker data is output according to the matching degree, i.e. marking the target as a guardrail, a sign, a parked vehicle or a moving vehicle. The false target recognition marker data is fused and analyzed with the first static and moving target classification information, and a conditional logic rule is used to comprehensively judge whether the target belongs to a real moving threat object. For example, when a target highly matches the guardrail type in the frequency spectrum characteristics, but its Doppler fluctuation exceeds the moving threshold, the system marks it as a contradictory state and triggers a high-confidence secondary recognition mechanism, and if a target has asymmetric frequency spectrum, a high RCS value and significant Doppler fluctuation, it is confirmed as a real candidate target. Through the fusion judgment process, false static structure objects or pseudo-moving reflectors are removed from the original moving target set, and a candidate target data set with real threat possibility is screened out.

[0023] Step S4, based on the radial distance data, the radial distance change rate and the Doppler velocity consistency of the candidate target data set are verified to obtain the second static and moving target classification information, and the target early warning signal is output in combination with the second static and moving target classification information.

[0024] Specifically, radial distance data for each target in consecutive time frames is extracted from the candidate target dataset. The first radial distance rate of change is calculated by dividing the frame-to-frame distance difference by the inter-frame time interval. This rate describes whether the target is approaching or receding along the vehicle's radar beam and is used to physically quantify its dynamic state. Considering the nonlinear trajectory motion and the lateral angle between the target and the vehicle during real-world driving, a compensation mechanism for the target's lateral offset angle is introduced. Based on the principle of geometric projection, the first radial distance rate of change is multiplied by a cosine factor for angle compensation to eliminate projection errors caused by non-radial motion, resulting in a second radial distance rate of change. The second radial distance rate of change is then compared with the Doppler velocity, and the absolute difference between the two is calculated to determine whether they meet the consistency condition. The consistency tolerance is set within 0.5 m / s. If the difference is less than this threshold, it is considered first target data, indicating that the target's range motion behavior is highly consistent with the frequency shift response, conforming to real-world motion patterns. Otherwise, it is classified as second target data, indicating a possible physical anomaly or data anomaly. For the first target data, a consistency verification mechanism in the time dimension is introduced, requiring the target to meet the above consistency conditions in at least three consecutive time windows, thereby eliminating short-term anomalies caused by transient noise, ranging errors or occasional interference. The final output is the third target data, which has a highly reliable motion trend. At the same time, for the second target data that does not meet the consistency conditions, a secondary verification mechanism is triggered. On the basis of expanding the time window to eight frames, the consistency threshold is relaxed (for example, to 0.8m / s), and its multiple frames of data are re-evaluated to determine whether it presents a stable physical change relationship over a longer observation period. If the re-verification conditions are met, it is confirmed as the fourth target data. Finally, the third target data and the fourth target data are merged to obtain the second static and dynamic target classification information, which represents the real moving target set confirmed by dual physical verification and timing verification. Based on this, an extended Kalman filter tracker is activated for each valid target in the second static and dynamic target classification information, constructing a six-dimensional state vector containing the target's position, velocity, and acceleration. This state vector is then combined with parameters such as the target's relative speed to the vehicle, current distance, and approach trend to perform a collision risk assessment. This system then generates corresponding target warning signals according to pre-set tiered warning rules, including Level 1 (rapid approach with a short distance), Level 2 (medium speed approach with a medium distance), and Level 3 (slow approach but close distance). Furthermore, the warning threshold is dynamically adjusted based on the vehicle's current operating speed to enhance the algorithm's adaptability under different operating conditions. A warning suppression mechanism is also implemented to prevent frequent alerts from borderline targets. The threat level, target relative position, velocity vector, and recommended action instructions are synchronously output to the two-wheeled vehicle control module.

[0025] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Data collection is performed on the radar echo signal within the backward detection area received by the millimeter wave radar to obtain the radar echo signal; Based on the installation position relationship between the radar coordinate system and the vehicle coordinate system, the radar echo signal is processed in the frequency domain and the coordinate system conversion calibration is performed to obtain Doppler frequency shift data, radial distance data, radar reflection cross-sectional area data and reflection signal intensity data; The three-axis acceleration signal and three-axis angular velocity signal output by the IMU inertial measurement unit in the electric two-wheeled vehicle are synchronously collected to obtain the vehicle body motion signal, and the vehicle body motion signal is filtered to obtain the motion state data.

[0026] Specifically, a millimeter-wave radar is mounted on the rear axis of the electric two-wheeled vehicle, or slightly to the left or right, to cover the entire rearward area of ​​the lane. It detects the motion and reflection characteristics of rearward-facing objects in space. In this configuration, the millimeter-wave radar operates at a frequency of 77 GHz and has a bandwidth exceeding 1 GHz, enabling centimeter-level range resolution and 10 centimeters per second velocity resolution for targets. When the vehicle is started or within the data acquisition cycle set by the control system, the radar actively transmits a continuous waveform or frequency-modulated continuous wave. This waveform is reflected by an obstacle behind it and received by the radar receiver. The echo signal is a frequency-aliased wave train with phase delays and amplitude variations. The system's internal DSP module applies a fast Fourier transform to the echo signal, mapping the mixed signal in the time domain to a frequency domain spectrum. By analyzing the frequency shift components in the spectrum, Doppler shift data is obtained to determine the target's relative motion. The system also determines the target's radial distance by measuring the phase difference or pulse-echo time difference of the mixed signal. During the demodulation process, the target echo amplitude and the reflection angle from the radar receiving array are recorded. This data is used to infer the reflection cross-sectional area and reflected signal strength. The radar reflection cross-sectional area is a key parameter for measuring a target's reflectivity, reflecting the amount of energy it returns to the radar wave. Because the radar uses its installation coordinates as a reference system for spatial ranging and direction estimation, the resulting data is in the radar's local polar coordinate system, containing the target's range and azimuth relative to the radar. However, the vehicle's control system and path planning system operate uniformly based on the vehicle's coordinate system. Therefore, the system performs a coordinate system calibration before each radar activation. By measuring the angle, horizontal offset, and height difference between the radar and the vehicle's longitudinal axis during installation, and constructing a hybrid rotation-translation transformation matrix, the radar coordinate system data is projected onto the vehicle coordinate system, ensuring consistency in distance and direction. After this coordinate transformation, each frame of radar perception data output by the system contains five parameters: relative range, relative angle, Doppler shift, reflection intensity, and reflection area for each target. These data are described in a unified spatial standard. At the same time, the vehicle's motion is sensed and dynamically compensated using an IMU (Inertial Measurement Unit) installed on the vehicle's main control board or near its center of gravity. This unit, consisting of a high-precision MEMS three-axis accelerometer and a three-axis gyroscope, outputs the vehicle's linear acceleration in the X, Y, and Z directions, as well as its angular velocity around these three axes. The acquisition frequency is kept consistent with the radar sampling frequency or precisely aligned through timestamps to ensure timing synchronization.Because two-wheeled vehicles face complex driving conditions such as bumpy roads, vibration disturbances, and sudden inertial changes, the raw IMU data contains significant high-frequency noise and zero-point drift. This data is filtered using a first-order low-pass filter to suppress high-frequency acceleration. A Kalman filter based on state estimation is then used to smooth and drift-correct the angular velocity data, ensuring stable, continuous, and physically plausible outputs of longitudinal, lateral, and angular velocities. During this process, the sensor zero point is dynamically calibrated, bias is removed through static detection in a motionless state, and a time alignment window is calculated for the inertial and radar data to ensure consistent sampling within a unit time period. The system then integrates these processing results to output a core input dataset for real-time perception and decision-making. This dataset includes Doppler shift information in the vehicle coordinate system, target radial range, target reflectance characteristics, and smoothed vehicle longitudinal, lateral, and angular velocities.

[0027] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Extracting longitudinal velocity component, lateral velocity component and angular velocity component from motion state data; The vehicle body Doppler frequency shift component is calculated based on the longitudinal velocity component and the target azimuth angle data to obtain the first Doppler frequency shift component corresponding to each target position; Setting segment compensation coefficient data based on the longitudinal velocity component; Performing a product operation on the first Doppler frequency shift component and the segmented compensation coefficient data to obtain a second Doppler frequency shift component, and performing a difference calculation on the Doppler frequency shift data and the second Doppler frequency shift component to obtain relative Doppler frequency shift data; The relative Doppler frequency shift data is converted to the radar working wavelength to obtain the Doppler velocity data; A time series variance calculation is performed based on the relative Doppler frequency shift data to obtain first static and dynamic target classification information.

[0028] Specifically, a compensation modeling process based on the vehicle's motion state is established. This process is based on the three-axis acceleration and three-axis angular velocity data output by the inertial measurement unit (IMU). After filtering and coordinate transformation, three core dynamic parameters are extracted: the longitudinal velocity component of the vehicle in the forward direction, the lateral velocity component corresponding to the vehicle's lateral drift or offset tendency, and the angular velocity component about the vertical axis. The longitudinal velocity, as the linear velocity on the vehicle's main motion axis, is a key variable, determining the velocity contribution of the vehicle in the directional projection between the vehicle and the target. The lateral velocity and angular velocity together constitute the nonlinear interference motion of the vehicle during cornering or track changes. The longitudinal velocity component and the azimuth angle of the target relative to the vehicle's longitudinal axis are jointly modeled. The cosine projection relationship between the motion direction and the target direction is used to calculate the Doppler component of the vehicle's own velocity in the target direction, namely the first Doppler frequency shift component. Considering the nonlinear differences in the radar system's sensitivity to vehicle dynamics at different speeds, a set of segmented compensation coefficients is introduced. These coefficients are dynamically set based on the vehicle's current longitudinal velocity range, typically classified into three ranges: low, medium, and high. For example, in the low-speed range of 5 to 15 km / h, the compensation coefficient is set to 1.2 to improve sensitivity to small speed changes. In the medium-speed range of 15 to 25 km / h, a coefficient of 1.0 is used to maintain system linearity. In the high-speed range of 25 to 35 km / h, the coefficient is reduced to 0.8 to mitigate the impact of high-speed changes on system stability. The compensation coefficient corresponding to this speed range is multiplied by the first Doppler shift component to obtain the corrected second Doppler shift component. The difference between the raw Doppler shift data measured by the radar and the second Doppler shift component is calculated. The compensation shift caused by the vehicle is subtracted from the target's total frequency shift to obtain the relative Doppler shift value. This value theoretically only reflects the target's true velocity relative to the vehicle and is the core basis for determining whether the target is actually moving or stationary. The relative Doppler shift data is converted to the radar operating wavelength. The operating wavelength of millimeter-wave radar is approximately 3.9 mm. Based on this wavelength constant, all relative Doppler shift values ​​are scaled to obtain Doppler velocity data in m / s. This converts the frequency characteristics into a visual physical velocity measure, improving the intuitiveness of the judgment and the adaptability of the algorithm. Using a time series variance analysis mechanism, the target's relative Doppler shift values ​​are evaluated for fluctuation within five frames of data, calculating the amplitude of the fluctuations to determine whether the target exhibits a stable motion trend over time. When the fluctuation range is lower than 0.85 Hz, it means that the target does not show obvious dynamic frequency shift behavior during the observation period and is preliminarily judged as a stationary target; if the fluctuation range exceeds this threshold, it indicates that the target presents continuous frequency shift changes and has a stable relative motion trend, and is judged as a moving target.

[0029] In a specific embodiment, the step of calculating the vehicle body Doppler shift component based on the longitudinal velocity component and the target azimuth angle data to obtain the first Doppler shift component corresponding to each target position may specifically include the following steps: Extract target azimuth data from radial distance data and radar cross-sectional area data; The longitudinal velocity component is converted into the Doppler frequency domain according to the radar operating frequency to obtain the reference Doppler frequency shift value corresponding to the longitudinal motion of the electric two-wheeled vehicle; Based on the geometric relationship between the radar installation position and the longitudinal axis of the vehicle body, the target azimuth angle data is corrected to obtain the detection azimuth angle data; The reference Doppler frequency shift value and the detection azimuth angle data are subjected to a cosine function operation to obtain the first Doppler frequency shift component corresponding to each target position.

[0030] Specifically, for each target point associated with a valid echo signal, its spatial positioning angle, i.e., the target's azimuth in the current radar coordinate system, is inferred from the original radial distance data and radar cross-sectional area data. Because millimeter-wave radar uses an array antenna to estimate the angle of target reflection signals, each target's spatial position is represented not only by a distance value but also by a polar angle relative to the radar's front. This polar angle information serves as the initial target azimuth. The information implicit in the radar cross-sectional area data also indirectly provides the relative orientation of the target to the radar array. This angular variation causes directional attenuation of the received reflected power in the antenna pattern, thus changing the effective measured intensity of the reflection cross-sectional area. Therefore, the system uses the spatial distribution of the reflected energy combined with the radar beam direction to fit the maximum response angle of the reflection center, obtaining the original azimuth data relative to the radar device itself. Based on the initial azimuth, a geometric mapping relationship is established with the vehicle coordinate system. Because millimeter-wave radars are not always perfectly aligned with the vehicle's longitudinal axis, installation deviations may occur due to structural layout, interference avoidance, or device size limitations. These deviations can occur, such as leftward, rightward, upward, or downward tilts. Therefore, the system constructs a transformation matrix from the radar's polar coordinate system to the vehicle's Cartesian coordinate system based on the actual radar installation position. Using a combined rotation and translation transformation, the original target azimuth angle data is corrected for coordinate mapping. This results in the target's detection azimuth in the vehicle's coordinate system. This is the physical angle between the vehicle's own motion direction and the target, reflecting the degree to which the vehicle's forward velocity projects onto the target. Furthermore, based on the vehicle's current longitudinal velocity component and the millimeter-wave radar's operating frequency and wavelength, the system converts the linear velocity-level motion into a frequency domain representation. The vehicle's longitudinal velocity is then projected into the radar's frequency domain to determine the baseline Doppler shift corresponding to the current vehicle motion. This value represents the Doppler frequency shift that the radar would measure if the target were stationary and located directly in front of the vehicle. The calculation of this reference frequency shift value relies on a fixed proportional relationship between vehicle speed, radar carrier frequency, and wave velocity, and is updated in real time as speed changes. It serves as a frequency domain reference for the entire target group. A cosine function is applied to the reference Doppler frequency shift value and the detection azimuth angle data. The frequency shift value corresponding to the vehicle's longitudinal motion is projected along the direction of each target to obtain the first Doppler frequency shift component. This component represents the magnitude of the frequency shift interference exerted on the target by the vehicle's current speed at the angle with the target. When the azimuth angle is 0 degrees, meaning the target is directly behind the vehicle, the frequency shift component of the vehicle's speed in this direction is maximum. When the azimuth angle is 90 degrees, meaning the target is completely lateral to the vehicle, the projection of the vehicle's longitudinal velocity on the target's frequency shift approaches zero.

[0031] In a specific embodiment, the process of performing the time series variance calculation based on the relative Doppler frequency shift data to obtain the first static and dynamic target classification information may specifically include the following steps: Extract M frames of continuous data from the relative Doppler frequency shift data to obtain the M frames of relative Doppler frequency shift time series data corresponding to each target; The mean and variance of the relative Doppler shift time series data of M frames are calculated to obtain the mean Doppler shift and Doppler shift stability coefficient of each target; Based on the preset static and dynamic discrimination threshold, the Doppler frequency shift stability coefficient is compared with the threshold value to obtain the classification data of stationary targets with a stability coefficient less than the static and dynamic discrimination threshold and the classification data of moving targets with a stability coefficient greater than or equal to the static and dynamic discrimination threshold; The change rate between adjacent frames of the M-frame relative Doppler frequency shift time series data is analyzed to obtain the Doppler gradient change rate data. The stationary target classification data is reconfirmed based on the judgment condition that the absolute value of the gradient change rate of N consecutive frames is less than the preset target value to obtain the first static and dynamic target classification information.

[0032] Specifically, M consecutive frames of relative Doppler shift data are extracted. Using a sliding time window, the relative Doppler shift values ​​of each target in M ​​consecutive frames are extracted in time series. The length of the M frames is set based on the processing cycle, target duration, or system resource allocation, and is typically configured as 5 frames. Within this time window, each target forms a sequence of M relative Doppler shift values, which is the trajectory of the target's frequency shift changes in its current state. The M frames of relative Doppler shift time series data are averaged and averaged. Through mean calculation, the center value of the target's Doppler shift in the current time period is obtained, reflecting its approximate relative motion speed trend; variance calculation is performed to obtain the degree of fluctuation of the target's frequency shift changes over M consecutive frames. This variance value is defined as the Doppler shift stability coefficient, which effectively characterizes the temporal smoothness of the target's frequency shift and is an important indicator for distinguishing between static and moving states. For an ideal stationary target, its frequency shift fluctuates minimally across all time frames, with a variance approaching zero. However, for a moving target, the frequency shift sequence fluctuates due to slight changes in position, angle, and velocity, resulting in a significantly higher variance than for a stationary target. The Doppler shift stability coefficient is classified and screened based on a preset static-motion discrimination threshold. If a target's stability coefficient is less than the threshold, it is initially classified as a stationary target; if it is greater than or equal to the threshold, it is considered a moving target. This threshold classification strategy divides most targets into two groups: those with "small frequency shift fluctuation" and those with "large frequency shift fluctuation," forming the first-level static-motion screening structure and effectively eliminating occasional false frequency shifts caused by noise or background reflection disturbances. A secondary confirmation mechanism is introduced for data identified as stationary in the first round. This mechanism, based on a gradient analysis model, calculates the inter-frame differences of M frames of relative Doppler shift time series data to generate M-1 sets of Doppler shift change rate sequences, or Doppler gradient change rate data, representing the differential change in the target's frequency shift between two adjacent frames. To confirm whether a target is truly stationary, a threshold for the absolute value of the gradient change rate (e.g., 0.2 Hz / s) is set, along with a continuous frame judgment period of N frames (e.g., three). If the absolute value of the gradient change rate for a target over N consecutive frames is less than the threshold, the target is considered highly stable in time and is identified as a truly stationary target. Otherwise, the target is considered a false stationary target with slight movement or instability and is removed from the stationary target classification data. This two-layer screening mechanism ultimately yields the first classification of stationary and moving targets.

[0033] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Based on the radial distance data and the road surface reflection coefficient, the radar reflection cross-sectional area data is normalized to obtain normalized reflection cross-sectional area data; Performing frequency domain transformation and spectrum feature extraction on the reflected signal intensity data to obtain the reflected signal spectrum feature data; The normalized reflection cross-sectional area data and the reflection signal spectrum feature data are matched with the preset guardrail target feature library, traffic sign target feature library, and vehicle target feature library to obtain the false target identification mark data corresponding to each target; A fusion discrimination is performed based on the false target identification mark data and the first static and dynamic target classification information to obtain reflection feature discrimination data, and a candidate target data set is determined based on the reflection feature discrimination data.

[0034] Specifically, the radar-measured reflection cross-sectional area data is normalized. Because radar reflection cross-sectional area (RCS) is affected by distance attenuation and path energy loss, the same object, even with the same structure, can exhibit different RCS values ​​at different distances. Therefore, the system performs energy normalization based on the radial distance between the target and the radar and the current road reflection conditions. During this process, the system retrieves the current radial distance value for each target and, combined with the road reflection coefficient preset in the environmental configuration parameters, corrects the original RCS value to generate normalized reflection cross-sectional area data on a unified scale. This ensures that the reflection characteristics of similar targets at near and far locations are comparable, avoiding identification errors caused by distance errors. The reflected signal intensity data for each target is transformed into the frequency domain to obtain the structural characteristics of the reflected wave in the frequency domain. This process uses the fast Fourier transform algorithm to convert the time domain signal into a spectrum. Based on the spectrum, key indicators such as the energy distribution center, bandwidth range, and frequency response symmetry are analyzed to extract core parameters such as the target's spectral center offset, spectral bandwidth, and spectral symmetry. The spectral center offset reflects the location of the dominant frequency of the target's reflection, helping to distinguish homogeneous reflectors from multifaceted structures. Spectral bandwidth reflects the frequency spread of the reflected wave signal and is essential for identifying complex structures. Spectral symmetry, by comparing the energy distribution differences on either side of the spectrum, effectively distinguishes between linear, continuous structures (such as guardrails) and nonlinear structures (such as vehicles). Because guardrails exhibit high directional consistency and regularity of their reflecting surfaces from the radar's perspective, their spectral structure appears highly symmetrical. In contrast, real vehicles exhibit asymmetric spectral energy distribution due to body curvature, localized irregularities, and refractive effects from multiple surfaces. The normalized reflection cross-sectional area data and the reflected signal's spectral signature are compared with a pre-configured library of typical target signatures within the system. This library, developed through extensive experiments and field sampling, includes data samples from four target categories: guardrails, traffic signs, parked vehicles, and real dynamic vehicles. For each category, structural parameters such as the normalized RCS value range, spectral symmetry threshold, frequency offset range, and beamwidth distribution are defined. For example, the normalized reflection cross-sectional area of ​​guardrail targets ranges from 0.1 to 0.3 square meters, with a spectral symmetry greater than 0.8; the normalized RCS of traffic sign targets ranges from 0.05 to 0.15 square meters; while the normalized RCS of real moving vehicles ranges from 3.0 to 12.0 square meters, with a spectral symmetry less than 0.6. The system constructs a multidimensional similarity calculation function, compares and analyzes the feature vectors of each target with the aforementioned standard features, outputs a matching score, and assigns an identification label to each target based on the score, generating false target identification tag data.If a target simultaneously meets multiple characteristic conditions, such as a small RCS, high symmetry, and low-frequency offset, the system labels it as a "guardrail" target. If its spectral bandwidth is extremely narrow but its RCS is high, it is classified as a "signage" target. If the target has a high RCS, low spectral symmetry, and a discrete energy distribution, the system classifies it as a "real vehicle." The system fuses the false target identification labeling data with the first static and dynamic target classification information to form a unified discrimination data structure. This structure uses set logical rules to cross-check the labeling results with the dynamic state. If a target is determined to be stationary and its reflection characteristics strongly match those of a guardrail or sign, it is directly removed from the candidate target list. If a target is in motion but its reflection characteristics clearly match those of a static structure, it is marked as a suspicious target and enters the secondary verification process. Targets that are both mobile and possess real vehicle characteristics such as a high reflection area and spectral asymmetry are retained in the candidate target dataset. Based on the fusion discrimination results, the system constructs a final candidate target dataset. The targets in this dataset all have the following characteristics: they show a clear relative speed trend in Doppler dynamics, match vehicle-like characteristics in spatial energy characteristics, have asymmetric structures or strong reflection responses in reflection morphology, and are not covered by the static infrastructure feature library.

[0035] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Performing inter-frame differential analysis on the radial distance data of each target in the candidate target dataset to obtain a first radial distance change rate of each target, and performing angle compensation based on the lateral offset angle of the target relative to the longitudinal axis of the vehicle to obtain a second radial distance change rate; Determining first target data that meets the consistency condition and second target data that does not meet the consistency condition according to the second radial distance change rate and Doppler velocity data of the corresponding target; Performing a time sequence consistency check on the first target data that meets the consistency condition to obtain third target data; Performing secondary verification on the second target data that does not meet the consistency condition to obtain fourth target data that passes the secondary verification, and merging the third target data with the fourth target data to obtain second static and dynamic target classification information; Based on the second static and dynamic target classification information, status tracking and collision warning are performed, and a target warning signal is output.

[0036] Specifically, based on a dataset of candidate targets, radial distance information is extracted from multiple consecutive frames. A differential operation is performed on the distance values ​​of each target between two adjacent frames to obtain the target's first radial distance rate of change, which is the linear distance change trend of the target relative to the vehicle's radar coordinate system. Positive and negative values ​​represent the direction of the target moving away from or approaching the vehicle, respectively, while the magnitude reflects the target's velocity component. Considering that targets are offset from the vehicle's longitudinal axis, such as some targets slightly to the left or right of the rear of a two-wheeled vehicle, their motion behavior includes a certain proportion of lateral components. Directly using the radial change value to determine physical motion consistency can introduce errors. Therefore, the system applies an angle compensation to the first radial distance rate of change based on the target's lateral offset angle relative to the vehicle's longitudinal axis. This compensation method projectively adjusts the change rate by introducing a cosine angle factor, stripping the original change value of non-radial components and obtaining a second radial distance rate that more closely represents the target's true velocity along the rearward direction of the vehicle. The second radial range change rate is compared with the target's Doppler velocity data for consistency. A tolerance threshold (e.g., 0.5 meters per second) is set to determine whether the two match within a reasonable range. If the difference between the two velocity data for a particular target falls within this threshold, the system identifies it as the first target data that meets the physical consistency criteria. Conversely, if the difference exceeds the threshold, it is included in the second target data set that does not meet the consistency criteria. For targets identified as first target data, a time-based verification mechanism is introduced. Specifically, a time series consistency verification model is constructed to determine whether the target consistently meets the matching criteria for radial variation and Doppler velocity over multiple consecutive frames. This model is set up as a three-frame judgment window. Within this window, if the target passes the consistency comparison in each frame, it is considered to have a stable and continuous motion trend and is confirmed as the third target data. This verification strategy reduces false positives caused by occasional errors in a single frame, thereby improving the temporal robustness of the consistency judgment. For second target data that fail the initial consistency comparison, a secondary verification mechanism is triggered. This mechanism expands the observation time window, for example, from three frames to eight frames, and moderately relaxes the consistency tolerance threshold, for example, from 0.5 meters per second to 0.8 meters per second, and introduces angle trend filtering and time smoothing methods to recalculate the average consistency of the target over a longer time range. If the target shows a stable and consistent movement trend during the recalculation process, it will be included in the fourth target data set. The third target data and the fourth target data are merged to form the second static and dynamic target classification information. The classification result represents all high-confidence target sets confirmed by dual verification of distance change and Doppler velocity, time consistency detection and compensation correction. Based on the target classification information, the system activates the extended Kalman filter to construct the state vector, models the spatial position, velocity vector and acceleration trend of each target as continuous variables, and continuously tracks the target motion trajectory in subsequent data frames.The system outputs a warning level based on the distance, closing speed, and relative angle between the target and the vehicle, combined with collision risk level definition rules. For example, when the target distance is less than 15 meters and the closing speed is greater than 5 meters per second, it is defined as a level 1 threat; if the distance is between 15 and 25 meters and the closing speed is between 2 and 5 meters per second, it is defined as a level 2 threat; if the distance is less than 35 meters and the closing speed is less than 2 meters per second, it is defined as a level 3 threat. The system also automatically adjusts these distance thresholds based on the vehicle's current speed, reducing the warning distance at low speeds and increasing the warning sensitivity at high speeds, thereby improving the adaptability of the warning response. Combining dynamic tracking and threat modeling results, a complete warning signal containing the target number, current distance, relative speed, direction of movement, warning level, and recommended actions is ultimately output to the vehicle's main control unit or human-machine interface.

[0037] In a specific embodiment, the process of determining the first target data that meets the consistency condition and the second target data that does not meet the consistency condition based on the second radial distance change rate and the Doppler velocity data of the corresponding target may specifically include the following steps: The absolute value of the difference between the second radial range change rate of each target in the candidate target data set and the corresponding Doppler velocity data is calculated to obtain the physical quantity consistency deviation data of each target; The physical quantity consistency deviation data of each target is compared with a preset consistency verification threshold to obtain first target data that meets the consistency condition and second target data that does not meet the consistency condition.

[0038] Specifically, the absolute value of the difference between the second radial range rate of change and the corresponding Doppler velocity data for each target in the candidate target dataset is calculated. The former is based on the time derivative of spatial position change, while the latter is based on the physical mapping of electromagnetic frequency shift. Both describe the relative velocity of the target in the vehicle reference frame, and their values ​​should be highly consistent in physical logic. To evaluate this physical consistency at the algorithmic level, a difference operation is performed on these two velocity data for each target: the corresponding Doppler velocity value is subtracted from the second radial range rate of change. The absolute value of the difference result is then taken to obtain the target's physical quantity consistency deviation data, namely the velocity comparison error, which reflects the degree of deviation in the estimation of the same physical quantity between the two data sources. If the deviation value is small, it indicates that the target's motion trend is consistent from the perspective of both position derivative and frequency shift mapping, indicating a physically coherent and verifiable state. Conversely, if the deviation value is large, it indicates a contradiction between the two velocity sources, resulting from inconsistent velocity estimates caused by measurement noise, angle estimation errors, multipath reflections, data delays, or radar transient perturbations. To transform the continuous deviation data into structured discrimination results, a consistency verification threshold is introduced. The threshold is set based on experimental data and the system's error tolerance range. For example, if it is set to 0.5 meters per second, the system allows the target's distance change rate and its frequency shift mapping speed to differ by less than 0.5 meters per second and still be considered consistent. The physical quantity consistency deviation data of each target is compared with the threshold, and the system divides the targets into two categories based on the results: one category is targets that meet the consistency conditions, that is, those with deviation values ​​less than the threshold, which constitute the first target data set and are considered to have high reliability at the physical behavior level; the other category is targets that do not meet the consistency conditions, that is, targets with deviation values ​​greater than or equal to the threshold. These targets constitute the second target data set and are marked as suspicious or requiring further verification.

[0039] In a specific embodiment, the execution step performs state tracking and collision warning based on the second static and dynamic target classification information, and the process of outputting a target warning signal may specifically include the following steps: Initializing the state vector of the moving target data in the second static and dynamic target classification information to obtain initial state vector data, and establishing a state transfer matrix and an observation matrix of the extended Kalman filter; Executing the prediction step and the update step of the extended Kalman filter based on the initial state vector data to obtain target trajectory tracking data, and linearly extrapolating the target trajectory tracking data to obtain target future position prediction data; The relative motion trajectory is calculated based on the target's future position prediction data and motion state data to obtain the minimum approach distance data between the target and the electric two-wheeled vehicle and the estimated time data for reaching the minimum distance; Based on the collision judgment conditions that the minimum approach distance data is less than the safety distance threshold and the estimated time data is less than the warning time threshold, a collision warning is performed and a target warning signal is output.

[0040] Specifically, targets identified as exhibiting high-confidence motion behavior from the second static and dynamic target classification information are extracted, and a state vector initialization data structure is constructed for each target for dynamic tracking. The state vector includes the target's current plane coordinate position, velocity component, and necessary predicted acceleration components. For example, the target's lateral position, longitudinal position, lateral velocity, longitudinal velocity, and angular velocity in the vehicle coordinate system constitute five- or six-dimensional state variables. To improve the accuracy of modeling nonlinear motion processes, an extended Kalman filter (ERF) is used as the primary algorithm for target state estimation. The EKF defines two key matrices during the modeling process: a state transfer matrix, which predicts the target's future state based on its current state and time interval; and an observation matrix, which defines how current observed data is mapped into the state variable space. When constructing the state transfer matrix, the system sets the mapping relationship between position and velocity based on the vehicle radar frame rate and the target motion model, accounting for inertial effects and nonlinear behavior during motion. The observation matrix maps and integrates radar-observed target data such as radial range, Doppler velocity, and angular position to ensure that the observed data adequately corrects the state variables. After the state transition and observation models are set up, the system enters the extended Kalman filter (ERKF) phase. The filter uses the current state vector for prediction, extrapolating the state from the previous frame's target state using the state transition matrix to obtain a predicted state. The filter then enters the update phase, inputting observation data from the current radar sensor. By calculating the residual between the predicted and observed values ​​and incorporating the Kalman gain coefficients, the state is updated and corrected to obtain a more accurate representation of the target's actual state. This process continues throughout each radar data frame, ensuring stable tracking of the target's state with temporal continuity and dynamic consistency. While the Kalman filter estimation is complete, the current target state is linearly extrapolated to obtain a prediction of its motion trend over several future time periods. This extrapolation uses the target's position and velocity information from the current frame, combined with a preset time step and number of extrapolation steps, to calculate the target's estimated position at each future moment, thereby forming the target's future path data along the time axis. The target's future trajectory is integrated with the vehicle's own motion state data for analysis. A relative motion model between the target and vehicle is established in a unified vehicle coordinate system. Interpolation is performed on the spatial positions of each predicted frame to calculate the distance vector between the target and vehicle. This process then identifies the minimum approach distance data within the entire time prediction sequence—the closest spatial distance between the target and vehicle at a certain moment in the future. Furthermore, by analyzing the interval between the time point corresponding to this minimum approach distance and the current system time, the estimated time at which the minimum approach distance will be reached is determined. Together, these two data describe the spatial and temporal scales of future collision trends. Based on these minimum approach distance data and the estimated time data, the system invokes pre-set collision discrimination criteria and sets a set of threshold parameters for risk assessment.Typical thresholds are set at a safety distance threshold of 1.5 meters and a warning time threshold of 2.0 seconds. This means that if the system detects that a target is less than 1.5 meters away from the vehicle at a certain point in the future, and that the target will arrive within the next two seconds, the system deems a potential collision risk and immediately enters the warning output process. After confirming that the judgment conditions are met, the system generates a target warning signal and structures the output data according to the target level, including the target's unique number, relative position coordinates, current velocity vector, predicted collision time, warning level (e.g., level 1, 2, 3), and recommended actions (e.g., deceleration, lane change, braking, etc.). The signal is then sent to the electric two-wheeled vehicle's main control module, assisted driving system, or user interface via a communication interface.

[0041] The above describes the radar identification and collision warning method of a two-wheeled vehicle in an embodiment of the present invention. The following describes the radar identification and collision warning system of a two-wheeled vehicle in an embodiment of the present invention. Figure 2 In one embodiment of the present invention, a radar identification and collision warning system for a two-wheeled vehicle includes: A synchronous acquisition module is used to synchronously acquire radar echo signals and vehicle motion signals within the rear detection area of ​​the electric two-wheeled vehicle to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data, and motion status data; A target classification module is used to perform vehicle motion compensation and time series variance calculation on Doppler frequency shift data based on motion state data to obtain first static and dynamic target classification information; a reflection feature calculation module, configured to calculate the reflection features of the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information, to obtain a candidate target data set; The output module is used to verify the consistency of the radial range change rate and Doppler velocity of the candidate target data set based on the radial range data, obtain the second static and dynamic target classification information, and output the target warning signal in combination with the second static and dynamic target classification information.

[0042] Through the collaborative efforts of these components, the system simultaneously collects radar echo signals and vehicle motion signals to obtain multi-dimensional information, including Doppler shift data, radar cross-sectional area data, radial distance data, reflected signal strength data, and motion state data. This achieves deep signal-level fusion. Based on the motion state data captured by the IMU (Inertial Measurement Unit), the system accurately compensates for vehicle motion in the Doppler shift data, effectively eliminating the interference of the electric two-wheeled vehicle's own motion on the radar signal. This addresses the target detection error caused by vehicle motion in traditional technologies and improves target recognition accuracy. Using a dual verification mechanism involving time series variance calculation and consistency verification between the radial distance change rate and Doppler velocity, a hierarchical recognition system is established, combining primary static and dynamic target classification information with secondary static and dynamic target classification information. This effectively reduces the false positive rate for static and dynamic targets and significantly enhances the recognition capability for targets in different motion states. By extracting normalized cross-sectional area data and spectral features, combined with pre-defined target feature libraries for guardrails, traffic signs, and vehicles, the system effectively identifies and suppresses false targets, addressing the false positives caused by fixed obstacles and non-vehicle targets in traditional technologies. The extended Kalman filter algorithm is used for multi-target state tracking. Through iterative calculation of state vector initialization, prediction step and update step, smooth tracking of target trajectory and accurate prediction of future position are achieved, which significantly improves the timeliness and reliability of collision warning.

[0043] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0044] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0045] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A radar identification and collision warning method for two-wheeled vehicles, characterized in that: include: The radar echo signal and vehicle motion signal in the rear detection area of ​​the electric two-wheeled vehicle are synchronously collected to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data and motion status data; Performing vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information; performing reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information to obtain a candidate target data set; Based on the radial distance data, the consistency of the radial distance change rate and the Doppler velocity of the candidate target data set is verified to obtain second static and dynamic target classification information, and a target warning signal is output in combination with the second static and dynamic target classification information.

2. The two-wheeled vehicle radar identification and collision warning method according to claim 1, characterized in that: The radar echo signal and the vehicle body motion signal in the rear detection area of ​​the electric two-wheeled vehicle are synchronously collected to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data and motion state data, including: Data collection is performed on the radar echo signal within the backward detection area received by the millimeter wave radar to obtain the radar echo signal; Based on the installation position relationship between the radar coordinate system and the vehicle coordinate system, the radar echo signal is subjected to frequency domain processing and coordinate system conversion calibration to obtain Doppler frequency shift data, radial distance data, radar reflection cross-sectional area data, and reflection signal intensity data; The three-axis acceleration signal and the three-axis angular velocity signal output by the IMU inertial measurement unit in the electric two-wheeled vehicle are synchronously collected to obtain a vehicle body motion signal, and the vehicle body motion signal is filtered to obtain motion state data.

3. The two-wheeled vehicle radar identification and collision warning method according to claim 1, characterized in that: The performing vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information includes: Extracting the longitudinal velocity component, the lateral velocity component and the angular velocity component from the motion state data; Calculating the vehicle body Doppler frequency shift component based on the longitudinal velocity component and the target azimuth angle data to obtain a first Doppler frequency shift component corresponding to each target position; Setting segmented compensation coefficient data based on the longitudinal velocity component; performing a product operation on the first Doppler frequency shift component and the segmented compensation coefficient data to obtain a second Doppler frequency shift component, and performing a difference calculation on the Doppler frequency shift data and the second Doppler frequency shift component to obtain relative Doppler frequency shift data; Converting the relative Doppler frequency shift data into a radar operating wavelength to obtain Doppler velocity data; A time series variance calculation is performed based on the relative Doppler frequency shift data to obtain first static and dynamic target classification information.

4. The two-wheeled vehicle radar identification and collision warning method according to claim 3, characterized in that: The calculating of the vehicle body Doppler frequency shift component according to the longitudinal velocity component and the target azimuth angle data to obtain the first Doppler frequency shift component corresponding to each target position includes: Extracting target azimuth data from the radial distance data and the radar reflection cross-sectional area data; Performing Doppler frequency domain conversion on the longitudinal velocity component according to the radar operating frequency to obtain a reference Doppler frequency shift value corresponding to the longitudinal motion of the electric two-wheeled vehicle; Performing coordinate system correction on the target azimuth angle data based on the geometric relationship between the radar installation position and the longitudinal axis of the vehicle body to obtain detection azimuth angle data; A cosine function operation is performed on the reference Doppler frequency shift value and the detection azimuth angle data to obtain a first Doppler frequency shift component corresponding to each target position.

5. The two-wheeled vehicle radar identification and collision warning method according to claim 3, characterized in that: The performing time series variance calculation based on the relative Doppler frequency shift data to obtain first static and dynamic target classification information includes: Extracting M frames of continuous data from the relative Doppler frequency shift data to obtain M frames of relative Doppler frequency shift time series data corresponding to each target; Perform mean calculation and variance calculation on the M frames of relative Doppler frequency shift time series data to obtain the Doppler frequency shift mean and Doppler frequency shift stability coefficient of each target; Performing a threshold comparison on the Doppler frequency shift stability coefficient based on a preset static-dynamic discrimination threshold to obtain classification data of stationary targets having a stability coefficient less than the static-dynamic discrimination threshold and classification data of moving targets having a stability coefficient greater than or equal to the static-dynamic discrimination threshold; The M frames of relative Doppler frequency shift time series data are subjected to an inter-frame change rate analysis to obtain Doppler gradient change rate data, and the stationary target classification data is reconfirmed based on the judgment condition that the absolute values ​​of the gradient change rates of N consecutive frames are all less than a preset target value to obtain the first stationary and moving target classification information.

6. The two-wheeled vehicle radar identification and collision warning method according to claim 1, characterized in that: The step of performing reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information to obtain a candidate target data set includes: Normalizing the radar reflection cross-sectional area data based on the radial distance data and the road surface reflection coefficient to obtain normalized reflection cross-sectional area data; Performing frequency domain transformation and spectrum feature extraction on the reflected signal intensity data to obtain reflected signal spectrum feature data; The normalized reflection cross-sectional area data and the reflection signal spectrum feature data are feature matched with a preset guardrail target feature library, a traffic sign target feature library, and a vehicle target feature library to obtain false target identification mark data corresponding to each target; A fusion discrimination is performed based on the false target identification mark data and the first static and dynamic target classification information to obtain reflection feature discrimination data, and a candidate target data set is determined based on the reflection feature discrimination data.

7. The two-wheeled vehicle radar identification and collision warning method according to claim 1, characterized in that: The step of performing consistency verification of radial distance change rate and Doppler velocity on the candidate target data set based on the radial distance data to obtain second static and dynamic target classification information, and outputting a target warning signal in combination with the second static and dynamic target classification information, includes: performing inter-frame differential analysis on radial distance data of each target in the candidate target data set to obtain a first radial distance change rate of each target, and performing angle compensation based on a lateral offset angle of the target relative to the longitudinal axis of the vehicle body to obtain a second radial distance change rate; Determining first target data that meets a consistency condition and second target data that does not meet the consistency condition according to the second radial distance change rate and Doppler velocity data of the corresponding target; Performing a time sequence consistency check on the first target data that meets the consistency condition to obtain third target data; Performing secondary verification on the second target data that does not meet the consistency condition to obtain fourth target data that passes the secondary verification, and merging the third target data with the fourth target data to obtain second static and dynamic target classification information; State tracking and collision warning are performed based on the second static and dynamic target classification information, and a target warning signal is output.

8. The two-wheeled vehicle radar identification and collision warning method according to claim 7, characterized in that: The determining, based on the second radial distance change rate and the Doppler velocity data of the corresponding target, first target data that meets the consistency condition and second target data that does not meet the consistency condition includes: Calculating the absolute value of the difference between the second radial range change rate of each target in the candidate target data set and the corresponding Doppler velocity data to obtain physical quantity consistency deviation data of each target; The physical quantity consistency deviation data of each target is compared with a preset consistency verification threshold to obtain first target data that meets the consistency condition and second target data that does not meet the consistency condition.

9. The two-wheeled vehicle radar identification and collision warning method according to claim 7, characterized in that: The performing state tracking and collision warning based on the second static and dynamic target classification information and outputting a target warning signal includes: Initializing the state vector of the moving target data in the second static and dynamic target classification information to obtain initial state vector data, and establishing a state transfer matrix and an observation matrix of an extended Kalman filter; Executing a prediction step and an update step of an extended Kalman filter based on the initial state vector data to obtain target trajectory tracking data, and performing linear extrapolation on the target trajectory tracking data to obtain target future position prediction data; Calculating a relative motion trajectory based on the target future position prediction data and the motion state data to obtain minimum approach distance data between the target and the electric two-wheeled vehicle and estimated time data for reaching the minimum distance; Based on the collision judgment conditions that the minimum approach distance data is less than a safety distance threshold and the estimated time data is less than a warning time threshold, a collision warning is performed and a target warning signal is output.

10. A radar identification and collision warning system for two-wheeled vehicles, characterized in that: A method for performing radar identification and collision warning of a two-wheeled vehicle according to any one of claims 1 to 9, comprising: A synchronous acquisition module is used to synchronously acquire radar echo signals and vehicle motion signals within the rear detection area of ​​the electric two-wheeled vehicle to obtain Doppler frequency shift data, radar reflection cross-sectional area data, radial distance data, reflection signal intensity data, and motion status data; a target classification module, configured to perform vehicle motion compensation and time series variance calculation on the Doppler frequency shift data based on the motion state data to obtain first static and dynamic target classification information; a reflection feature calculation module, configured to perform reflection feature calculation on the radar reflection cross-sectional area data and the reflected signal intensity data according to the first static and dynamic target classification information, to obtain a candidate target data set; An output module is used to verify the consistency of the radial distance change rate and Doppler velocity of the candidate target data set based on the radial distance data, obtain second static and dynamic target classification information, and output a target warning signal in combination with the second static and dynamic target classification information.

Images