Cooperative warning and anti-collision method fusing wide-area perception of unmanned aerial vehicle and vehicle-mounted radar

Abstract

This invention relates to the fields of intelligent transportation and autonomous driving technology, and discloses a collaborative early warning and collision avoidance method integrating wide-area perception from unmanned aerial vehicles (UAVs) and vehicle-mounted radar. The method first establishes a unified spatiotemporal reference and broadcasts a beacon; then, it acquires multipath radar signals through active transmission from the vehicle and passive reception from the UAV, separating single-base echoes, direct waves, and bistatic scattered waves; a collaborative processing module constructs a rotating ellipsoid based on the time difference of arrival, and determines the geometric intersection of the target by combining it with the single-base ranging sphere; further, it extracts the Doppler frequency shift to establish a coupled verification equation set, confirming the instantaneous position and velocity vector of the target in the geometric intersection; subsequently, it uses this position to guide the UAV to acquire images and perform semantic threat confirmation; finally, it generates a collaborative early warning package to guide collision avoidance decisions. This invention overcomes line-of-sight obstruction by utilizing a high-altitude perspective, improves detection accuracy through dual geometric constraints and Doppler verification, and reduces false alarm rates by combining visual semantic analysis, thus enhancing early warning capabilities in complex scenarios.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and autonomous driving technology, specifically to a collaborative early warning and collision avoidance method that integrates wide-area perception from unmanned aerial vehicles (UAVs) with vehicle-mounted radar. Background Technology

[0002] The development of autonomous driving technology and advanced driver assistance systems (ADAS) depends on the performance of onboard environmental perception systems. Currently, onboard perception systems integrate sensors such as millimeter-wave radar, lidar, and cameras to detect the vehicle's surroundings and track targets.

[0003] However, these vehicle-mounted sensors (especially radars) are installed at low heights, making their detection line of sight easily obstructed by large vehicles, buildings, or terrain undulations, severely limiting their ability to detect non-line-of-sight (NLOS) targets or obstructed targets. Furthermore, low-angle detection easily introduces strong ground clutter interference, which together limit the system's effective warning distance and detection range, making it difficult to meet the advanced collision avoidance requirements in complex urban or highway scenarios.

[0004] To extend the sensing range, existing technologies have attempted to utilize drones for wide-area sensing. However, in cooperative localization techniques, relying solely on a single geometric measurement constraint (e.g., using only Time Difference of Arrival (TDOA) or only single-base ranging) makes it difficult to accurately calculate the unique location of the target. For example, simple TDOA measurement can only confine the target to a rotating ellipsoid, resulting in localization ambiguity, which leads to low initial localization accuracy and difficulty in convergence.

[0005] Furthermore, existing radar sensing systems have shortcomings in confirming target status and threat levels. After acquiring the kinematic state of a target through geometric positioning and Doppler measurements, the system still struggles to distinguish between a detected target that is a genuine hazard (such as a pedestrian or obstructing vehicle) and a non-threatening obstacle (such as a roadside metal guardrail). This reliance solely on kinematic data leads to a high false alarm rate, affecting the reliability of collision avoidance decisions. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a collaborative early warning and collision avoidance method that integrates wide-area perception of UAVs and vehicle-mounted radar. This method solves the problems of existing vehicle-mounted radar being easily blocked by line of sight due to its low installation height, making it difficult to detect non-line-of-sight targets; the problems of positioning ambiguity and low accuracy caused by single geometric constraints in collaborative positioning; and the problems of traditional perception relying solely on kinematic states and being unable to distinguish real threats, resulting in a high false alarm rate.

[0007] To achieve the above objectives, the present invention provides a collaborative early warning and collision avoidance method integrating UAV wide-area perception and vehicle-mounted radar: In the technical solution provided by this invention, the vehicle-mounted execution unit and the UAV execution unit first establish a unified spatiotemporal reference through a precise time protocol and a high-precision positioning module. The vehicle-mounted execution unit, acting as an active radar signal source, broadcasts beacon information containing radar signal waveform parameters and its own instantaneous kinematic state (position, velocity) to the UAV execution unit while transmitting radar signals.

[0008] During the perception phase, the system utilizes multiple signal propagation paths in parallel. The vehicle-mounted actuator's single-base receiver receives the single-base echo signal reflected from the target. The UAV actuator's passive radar receiving module receives a mixed signal, which includes at least the direct wave signal emitted directly by the vehicle-mounted actuator and the bistatic scattered wave signal scattered by the target.

[0009] The collaborative processing module generates a local reference signal using the known waveform parameters in the beacon information, and separates the direct wave signal and the bistatic scattered wave signal from the mixed signal through cross-correlation processing.

[0010] During the geometric positioning phase, the system employs dual geometric constraints. The collaborative processing module first calculates a path distance difference based on the time difference of arrival (TDOA) of the direct wave signal and the bistatic scattered wave signal at the UAV. This distance difference defines a rotating ellipsoid (first geometric candidate set) with the instantaneous positions of the vehicle-mounted actuator and the UAV actuator as its foci. Simultaneously, the collaborative processing module calculates a single-base instantaneous distance based on the two-way propagation delay of the single-base echo signal received by the vehicle-mounted actuator. This distance defines a sphere (second geometric candidate set) with the instantaneous position of the vehicle-mounted actuator as its center. The intersection of these two geometric candidate sets (the final geometric intersection) defines the possible geometric positions of the target to be detected.

[0011] In the physical verification and state confirmation stages, to resolve potential ambiguities in the geometric intersection and accurately calculate the target's motion state, this method introduces coupled Doppler verification. The collaborative processing module extracts the single-base Doppler frequency shift from the single-base echo signal and the bi-base Doppler frequency shift from the bi-base scattered wave signal in parallel. Since the theoretical models of both single-base and bi-base Doppler are associated with the instantaneous position and velocity vectors of the target, the collaborative processing module establishes a set of coupled verification equations. By solving this set of equations, not only can the instantaneous velocity vector of the target be obtained, but also, by evaluating the residuals between the theoretical model and the measured values, the unique instantaneous position vector satisfying the physical constraints can be identified in the final geometric intersection.

[0012] During the threat confirmation phase, to distinguish between real threats and non-threat targets, the system uses the confirmed instantaneous position vector of the target to be detected and the UAV's own position information to calculate the line-of-sight vector from the UAV to the target, and converts it into a guidance command in the aircraft's coordinate system. This guidance command is used to drive the UAV's onboard optical sensor (visual perception module) to align with the target and acquire optical images. The collaborative processing module then calls a pre-trained semantic recognition model to analyze the optical images, calculate the target's semantic label, and compare it with a pre-set threat target semantic database to determine the real threat level of the target to be detected.

[0013] Finally, the collaborative processing module integrates the confirmed instantaneous position vector, instantaneous velocity vector, semantic label, and system timestamp of the target to be detected, generates a collaborative warning package, and transmits it to the vehicle-mounted execution unit and the UAV execution unit. Each execution unit, based on the warning information and its own kinematic state, predicts the collision risk within the future time window and executes corresponding collision avoidance decisions and actions.

[0014] This invention provides a collaborative early warning and collision avoidance method that integrates wide-area perception from unmanned aerial vehicles (UAVs) and vehicle-mounted radar. It offers the following advantages: 1. This invention utilizes the wide-field perspective advantage of UAVs to effectively overcome the problems of line-of-sight obstruction (such as obstruction by large vehicles or buildings in front) and ground clutter interference caused by the low installation height of vehicle-mounted radar, by transmitting signals through the vehicle-mounted execution unit and passively receiving multi-path signals at a high altitude. This expands the effective range and coverage of the sensing system and enables the detection and early warning of non-line-of-sight or obstructed targets.

[0015] 2. In the geometric positioning stage, the present invention constructs dual geometric constraints, and intersects the rotating ellipsoid (first geometric candidate set) calculated based on the time difference of arrival (TDOA) with the sphere (second geometric candidate set) calculated based on the vehicle single-base echo time delay. By using these two geometric constraints derived from different physical measurements, the accuracy and convergence speed of the initial positioning of the target are improved, and the positioning ambiguity caused by a single geometric constraint is effectively suppressed.

[0016] 3. This invention establishes a coupled verification equation system of single-base Doppler and dual-base Doppler, which not only calculates the instantaneous velocity vector of the target but also identifies the unique instantaneous position that satisfies the physical constraints in the geometric intersection, thus improving the accuracy of state calculation. Subsequently, this high-precision position is used to guide the vision module to perform semantic analysis, enabling the determination of the target's true threat level and effectively distinguishing between real danger sources and non-threatening interference, thereby reducing the system's false alarm rate. Attached Figure Description

[0017] Figure 1 This is a flowchart of the collaborative early warning and collision avoidance method of the present invention; Figure 2 This is a schematic diagram of the collaborative sensing system structure of the present invention; Figure 3 This is a schematic diagram of the multipath radar signal propagation of the present invention; Figure 4 This is a schematic diagram illustrating the visual guidance principle of the present invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See attached document Figure 2 This invention provides a collaborative early warning and collision avoidance method that integrates wide-area perception of unmanned aerial vehicles (UAVs) and vehicle-mounted radar. This method can be executed by a collaborative perception system, which includes a vehicle-mounted execution unit and an UAV execution unit.

[0020] Among them, the vehicle-mounted execution unit (deployed in the vehicle) is one of the initiators and cooperating entities of this method, specifically including: a high-precision positioning module, a high-precision clock module, an active radar module, and a vehicle-to-air communication module.

[0021] High-precision positioning modules, such as a combination of a Differential Global Positioning System (RTK-GPS) and an Inertial Measurement Unit (IMU), are used to acquire the instantaneous position vector of the onboard actuator in the global coordinate system in real time. and instantaneous velocity vector .

[0022] High-precision clock modules, such as those supporting Precision Time Protocol (PTP) or Generalized Precision Time Protocol (gPTP), are used to synchronize with the clock modules of the UAV execution units to establish a unified, nanosecond-level system time base. .

[0023] The vehicle-to-air communication module (V2A) is used to broadcast a waveform beacon to the UAV execution unit. This waveform beacon contains radar signals that will be transmitted by the active radar module. The waveform parameters, and the instantaneous kinematic state obtained by the high-precision positioning module. .

[0024] Active radar modules, such as 77 GHz frequency modulated continuous wave (FMCW) radar, are configured as active illumination sources. In the system time base Under synchronization, the active radar module transmits known radar signals forward (including areas with obstructed line of sight). .

[0025] The active radar module is also configured as a monobase receiver. Used for parallel reception along the edge path( The single-base echo signal returned by the target to be detected .

[0026] The onboard execution unit (or subsequent collaborative processing module) receives the data. Doppler processing is performed to extract (measure) the single-base Doppler frequency shift of the target. ,Should The theoretical model is defined by the following formula: ; in, The center carrier frequency for the signals transmitted by the active radar module; The speed of light; Let be the instantaneous velocity vector of the target to be detected; This represents the instantaneous velocity vector of the onboard actuator. To point from the vehicle-mounted actuator to the target to be detected A unit vector, defined as . ,in Target to be detected The instantaneous position vector, Let be the instantaneous position vector of the on-board actuator. For vehicles and the target to be detected The instantaneous straight-line distance between them.

[0027] The collaborative perception system of the present invention also includes a drone execution unit, which is deployed on the drone. The collaborative sensing end of this method specifically includes: a high-precision positioning module, a high-precision clock module, a passive radar receiving module, an empty vehicle communication module, and a visual sensing module.

[0028] High-precision positioning modules, such as a combination of a Differential Global Positioning System (RTK-GPS) and an Inertial Measurement Unit (IMU), are used to acquire the instantaneous position vector of the UAV execution unit in the same global coordinate system in real time. and instantaneous velocity vector .

[0029] A high-precision clock module is used to synchronize with the high-precision clock module of the on-board actuator, ensuring that both are on a unified system time base. .

[0030] The vehicle-to-vehicle (A2V) communication module is used to transmit the generated collaborative early warning information (such as a warning package containing the target status and category) back to the on-board execution unit after completing the perception calculation.

[0031] A visual perception module, such as an electro-optical / infrared (EO / IR) camera, is used in subsequent steps of this method to provide visual guidance and semantic verification of the physically verified target.

[0032] Passive radar receiver module ( The passive radar receiver module is configured to passively receive signals only and not transmit any signals. Its antenna array is tuned to the center carrier frequency of the onboard actuator's active radar module. .

[0033] The passive radar receiver module functions to receive (acquire) two signals in parallel: one is a direct wave signal from the onboard actuator. (along One path is the bistatic scattered wave signal scattered by the target to be detected; the other path is the bistatic scattered wave signal scattered by the target to be detected. (along path).

[0034] The UAV execution unit (or subsequent collaborative processing module) processes the received bistatic scattering wave signal. Processing is performed to extract (measure) the bistatic Doppler frequency shift of the target. ,Should The theoretical model is defined by the following formula: ; in, The center carrier frequency; The speed of light; Let be the instantaneous velocity vector of the target to be detected; This represents the instantaneous velocity vector of the onboard actuator. This represents the instantaneous velocity vector of the UAV execution unit; This is a unit vector pointing from the onboard execution unit to the target to be detected; Let be the unit vector pointing from the UAV execution unit to the target to be detected, and let it be defined as follows: ,in Target to be detected The instantaneous position vector, This represents the instantaneous position vector of the UAV execution unit. For drones and the target to be detected The instantaneous straight-line distance between them.

[0035] The collaborative sensing system of the present invention also includes a collaborative processing module, which is a functional module responsible for executing the core data processing and calculation steps in this method.

[0036] In terms of physical deployment, the collaborative processing module can be integrated onto the high-performance processor of the vehicle-mounted execution unit; alternatively, it can be integrated onto the onboard processor of the UAV execution unit; or, the functionality of the collaborative processing module can be distributed and collaboratively executed by the processors of the vehicle-mounted execution unit and the UAV execution unit. This collaborative processing module is configured as follows: Receive instantaneous kinematic state from the on-board actuator and Waveform parameters; Receive instantaneous kinematic state from UAV actuator ; Receive single-base echo signals from the active radar module of the on-board actuator. (or single-base Doppler measurements extracted from it) ); Receives a mixed signal (including direct wave signal) from the passive radar receiving module of the UAV actuator. and bistatic scattering wave signal ).

[0037] The collaborative processing module is further configured as follows: use The waveform parameters are used to perform cross-correlation processing on the received mixed signal in order to separate the waveform parameters. and .

[0038] calculate and Time difference between and according to Calculate the path distance difference .in At the speed of light, The time difference is the time of arrival.

[0039] Based on known and As the focus, and the calculated path distance difference Construct geometric constraints, which are defined by the following formula: ; in, Let be the instantaneous position vector of the target to be detected; This represents the instantaneous position vector of the onboard execution unit; This represents the instantaneous position vector of the UAV execution unit; Represents the magnitude of the vector (i.e., the Euclidean distance); For vehicles and the target to be detected The instantaneous straight-line distance between them; For drones and the target to be detected The instantaneous straight-line distance between them; For drones and vehicles The instantaneous straight-line distance between them.

[0040] This collaborative processing module generates a set of geometric candidates for the target by calculating the intersection of the hyperboloid of revolution defined by computational geometric constraints and the ground plane model (or digital elevation map). (It is represented by a hyperbolic arc).

[0041] The collaborative processing module is further configured as follows: from Extracting single-base Doppler measurements and from Extracting bipolar Doppler measurements ; Traversal Candidate position vectors on Substitute it into Theoretical models and In the theoretical model, a velocity of the unknown target is constructed. A system of linear equations.

[0042] The estimated target velocity can be obtained by solving this system of linear equations (e.g., using the least squares method). and residuals .

[0043] Based on the principle of minimum residual (i.e., finding the minimum residual) ),exist Lock the unique true target state that satisfies the physical constraints. .

[0044] Use confirmed Guide the visual perception module of the UAV execution unit and receive the semantic classification results returned by it.

[0045] Generate includes The system generates a collaborative early warning package with the semantic classification results and sends it to the main control system of the on-board execution unit.

[0046] See attached document Figure 1 This method provides a collaborative early warning and collision avoidance process, which is executed collaboratively by an onboard execution unit, an unmanned aerial vehicle execution unit, and a collaborative processing module.

[0047] S1. Establishment of the spatiotemporal reference. The vehicle-mounted execution unit and the UAV execution unit use their respective high-precision clock modules to achieve nanosecond-level time synchronization and establish a unified system time reference. Simultaneously, each utilizes its own high-precision positioning module to acquire the instantaneous kinematic state in the same global coordinate system in real time, i.e. and The onboard actuator broadcasts information via the vehicle-to-air communication module. The waveform parameters and the waveform beacon of its own state.

[0048] S2, Execute multipath signal acquisition. In synchronization, the active radar module of the vehicle-mounted actuator transmits radar signals. The radar signal propagation path is received in parallel by two actuators: the single-base receiver of the vehicle-mounted actuator (…). )take over Single-base echo signal of the path Passive radar receiving module of UAV execution unit ( Simultaneously receive Direct wave signal of the path and Bistatic scattering signal of the path .

[0049] S3. Perform TDOA-based geometric positioning. The collaborative processing module utilizes known... Waveform parameters, for The received mixed signal is processed (e.g., cross-correlation) to separate the signals. and Through calculation and Time difference of arrival Calculate the path distance difference The collaborative processing module is based on Two focal positions and Geometric constraints are then constructed. These constraints define a hyperboloid of revolution, which intersects with the ground plane model to generate two-dimensional hyperbolic arcs, i.e., the geometric candidate set of the target. .

[0050] S4. Perform physical verification based on dual Doppler coupling. This step is used to select from the geometric candidate set. The system analyzes the real target and confirms its motion state. The collaborative processing module processes data in parallel from... Extracting single-base Doppler measurements and from Extracting bipolar Doppler measurements The physical constraint utilized by this method lies in a real instantaneous state. The TDOA geometric constraint must be satisfied simultaneously. lie in (above), single-base Doppler physical constraints (satisfying) Theoretical model) and bibasal Doppler physical constraints (satisfying) (Theoretical model). Co-processing module traversal. Candidate position vectors on Substitute it into Theoretical models and In the theoretical model, a velocity of the unknown target is constructed. A system of linear equations. Solve this system of equations and evaluate the residuals. (i.e., theoretical value and) (Differences between measurements) are used to identify the unique true target state in the geometric candidate set that simultaneously satisfies the dual Doppler physical constraints. .

[0051] S5. Perform visual guidance and semantic verification. The collaborative processing module utilizes this high-confidence candidate location vector. Actively guide the visual perception module of the drone's execution unit to... Targeted observation and semantic classification are performed within the region of interest (ROI) to identify the target type (e.g., pedestrian, vehicle, or non-obstacle).

[0052] S6. Perform collaborative early warning and decision-making. The collaborative processing module integrates high-confidence real target states. Based on the semantic classification results, a collaborative early warning package is generated. This early warning package is transmitted back to the on-board execution unit via the vehicle communication module, so that its main control system can make collision avoidance decisions.

[0053] In step S1, time synchronization is performed first.

[0054] Specifically, the high-precision clock module of the vehicle-mounted actuator and the high-precision clock module of the UAV actuator execute a precise time protocol (e.g., PTP, Precision Time Protocol, or gPTP, Generalized Precision Time Protocol).

[0055] During the execution of this precise time protocol, one clock module (e.g., a module in an onboard execution unit) acts as the master clock, and another clock module (e.g., a module in a drone execution unit) acts as the slave clock.

[0056] Two clock modules exchange a series of timing messages (such as Sync, Follow_Up, Delay_Req, and Delay_Resp messages, where Sync refers to a synchronization message, Follow_Up refers to a follow message, Delay_Req refers to a delay request message, and Delay_Resp refers to a delay response message) between them, enabling the slave clock to calculate its clock offset from the master clock and the message propagation delay.

[0057] The slave clock adjusts its local clock based on the calculated clock offset and propagation delay to align it with the master clock.

[0058] The result of this time synchronization process is that the vehicle-mounted execution unit and the UAV execution unit jointly establish and maintain a unified system time reference with nanosecond-level accuracy. .

[0059] this This provides a unified timestamp for all measurement data acquired in subsequent steps (including signal reception, position, and velocity measurements), which forms the basis for subsequent TDOA (Time Difference of Arrival) and Doppler resolution.

[0060] In step S1, the establishment of the spatial reference and beacon broadcasting are performed.

[0061] Specifically, the vehicle-mounted actuator and the drone actuator utilize their respective high-precision positioning modules (such as a combination of RTK-GPS and IMU) on a unified system time base. Under these conditions, the instantaneous kinematic state of the device is acquired and updated in real time in the same global coordinate system.

[0062] The onboard actuator acquires its instantaneous kinematic state. ,in, This is the instantaneous position vector of the on-board execution unit. This is the instantaneous velocity vector of the onboard actuator.

[0063] Simultaneously, the UAV execution unit acquires its instantaneous kinematic state. ,in This represents the instantaneous position vector of the UAV execution unit. This is the instantaneous velocity vector of the UAV execution unit.

[0064] Following this, or in parallel with status acquisition, the onboard execution unit performs beacon broadcasting.

[0065] The vehicle-mounted actuator uses its vehicle-to-air communication (V2A) module to broadcast a waveform beacon to the unmanned aerial vehicle (UAV) actuator. This waveform beacon contains the instantaneous kinematic state of the vehicle-mounted actuator. and the radar signals that its active radar module is about to transmit. The waveform parameters, for example, in the FMCW radar system, include the center carrier frequency. Sweep bandwidth Frequency sweep cycle and frequency modulation slope .

[0066] The UAV actuator receives this waveform beacon, thereby obtaining the instantaneous kinematic state of the vehicle-mounted actuator. And obtained the basis for generating the local reference signal required for signal separation (e.g., cross-correlation) in the subsequent step S3.

[0067] See attached document Figure 3 In step S2, signal transmission and parallel reception are performed.

[0068] Specifically, on the established unified system time base Below, the active radar module of the vehicle-mounted actuator transmits radar signals. .

[0069] At the same time, the single-base receiver of the vehicle-mounted actuator ( Receive single-base echo signal The single-base echo signal Corresponding to radar signals Along Path (i.e., from the onboard execution unit to the target to be detected) The signal is then reflected back to the onboard actuator after being propagated.

[0070] Meanwhile, the passive radar receiving module of the UAV execution unit ( ) receives a mixed signal, the mixed signal containing at least two sources The main components.

[0071] The first component is the direct wave signal. The direct wave signal Corresponding to radar signals Along The signal received after propagation along the path (i.e., from the vehicle-mounted actuator directly to the UAV actuator).

[0072] The second component is the bistatic scattered wave signal. The bistatic scattering wave signal Corresponding to radar signals Along Path (i.e., from the onboard execution unit to the target to be detected) The signal received after being scattered and then propagated to the drone's execution unit.

[0073] In step S2, signal separation processing is performed.

[0074] Specifically, the collaborative processing module receives the passive radar receiver module of the UAV execution unit ( The mixed signal acquired is processed. The collaborative processing module utilizes the radar signal obtained from the waveform beacon in step S1. The waveform parameters are used to generate a local reference signal. .

[0075] Subsequently, the collaborative processing module executes the local reference signal. Cross-correlation processing between the received mixed signal and the signal. Because... Path (direct path) and The propagation distance of the scattering path varies, resulting in different direct wave signals. and bistatic scattering wave signal arrive There is a time difference between the moments.

[0076] Therefore, the output of this cross-correlation processing (e.g., in a distance-time profile or a delayed Doppler spectrum) appears as two or more separate peaks on the time delay axis.

[0077] Based on the cross-correlation result, the collaborative processing module identifies and separates the signals corresponding to the direct wave signals. The first peak and the corresponding bistatic scattered wave signal The second peak.

[0078] The output of this signal separation processing (i.e.) and (and their respective time delay information) are used for the Time Difference of Arrival (TDOA) calculation in the subsequent step S3.

[0079] In step S3, time difference measurement and distance difference calculation are performed first.

[0080] Specifically, the collaborative processing module extracts and measures the direct wave signal based on the signal separation results (e.g., two separated peaks in the cross-correlation output profile). and bistatic scattering wave signal In the passive radar receiving module of the UAV execution unit ( Time difference of arrival at ) The time difference of arrival The calculation is as follows: ; in, Direct wave signal (Right now Arrival time of the path; Two-base scattered wave signal (Right now The arrival time of the path. Subsequently, the collaborative processing module utilizes this arrival time difference. with the speed of light Solve for the bistatic scattering path ( ) and direct route ( Path distance difference between The calculation formula is: ; The path distance difference The following differences correspond to the geometric path lengths: ; in, For vehicles With the target to be detected The instantaneous distance between them; Target to be detected With drones The instantaneous distance between them; For vehicles With drones The instantaneous distance between them. This calculated path distance difference. It will be used as input for constructing the geometric constraints (hyperboloid of revolution).

[0081] In step S3, the generation of the geometric candidate set continues.

[0082] Specifically, the collaborative processing module utilizes the instantaneous position vector of the vehicle execution unit obtained in step S1. and the instantaneous position vector of the UAV execution unit and the calculated path distance difference Construct the geometric constraint equations. The geometric constraint equations are expressed as follows: The collaborative processing module restructures this relation as follows: ; in, For vehicles With the target to be detected The instantaneous distance between them; Target to be detected With drones The instantaneous distance between them; For vehicles With drones The instantaneous distance between them; This represents the path distance difference.

[0083] At that instant, the vehicle With drones instantaneous distance between (Based on known information) and (Calculated) and path distance difference (Measured) are all known constants.

[0084] Therefore, this equation (Constant) In three-dimensional space, an instantaneous position vector of the onboard execution unit is defined. and the instantaneous position vector of the UAV execution unit An ellipsoid of revolution with two foci.

[0085] The collaborative processing module defines this ellipsoid of revolution as the first geometric candidate set. The first geometric candidate set This represents the target to be detected based solely on bistatic TDOA measurements. All instantaneous position vectors The set of first geometric candidate sets. Mathematically described as: ; in, This is the first geometric candidate set; Target to be detected The instantaneous position vector; This represents the instantaneous position vector of the onboard execution unit; This represents the instantaneous position vector of the UAV execution unit. Represents three-dimensional Euclidean space; For the Euclidean norm (i.e., distance calculation), such that and ; This represents all conditions that are satisfied in three-dimensional space. The set that constitutes; For vehicles and the target to be detected The instantaneous straight-line distance between them; For drones and the target to be detected The instantaneous straight-line distance between them; For vehicles With drones The instantaneous distance between them; This represents the path distance difference.

[0086] In generating the first geometric candidate set Then, continue with the second geometric candidate set. The generation of .

[0087] Specifically, deployed in vehicles The vehicle-mounted actuator acquires the single-base echo signal. Process the signal to calculate the time it has traveled. Two-way propagation delay of the path .

[0088] The onboard actuator is based on this two-way propagation delay Solve for the vehicle With the target to be detected The instantaneous distance between them (i.e., the slope distance of a single base). : ; in, For measurement Two-way propagation delay; The speed of light; For the calculated vehicle With the target to be detected The instantaneous distance between them.

[0089] The collaborative processing module uses this calculated instantaneous distance and the instantaneous position vector of the vehicle-mounted execution unit obtained in step S1. Construct a second geometric constraint.

[0090] This second geometric constraint, in three-dimensional space, defines the instantaneous position vector of the onboard actuator. With the center of the ball, A sphere with radius r. The collaborative processing module defines this sphere as the second set of geometric candidates. .

[0091] The second geometric candidate set Mathematically described as: ; in, This is the second geometric candidate set; Target to be detected The instantaneous position vector; This represents the instantaneous position vector of the onboard execution unit; Represents three-dimensional Euclidean space; For Euclidean norm (i.e., distance calculation); The instantaneous distance of the single basis is calculated; This represents all conditions that are satisfied in three-dimensional space. The set that constitutes the collection.

[0092] Finally, the collaborative processing module computes the first geometric candidate set. (Rotating ellipsoid) and the second geometric candidate set The geometric intersection of (spherical surfaces).

[0093] This geometric intersection is defined as the final geometric intersection. This represents the simultaneous satisfaction of the TDOA constraint ( ) and single-base ranging constraints ( All candidate location points A set of.

[0094] The final geometric intersection Mathematically described as: ; in, The final geometric intersection; This is the first geometric candidate set; This is the second geometric candidate set; The symbol represents the intersection operation of sets.

[0095] The final geometric intersection This will be used as input for physical verification in step S4.

[0096] In step S4, single-base and dual-base Doppler measurements are performed first.

[0097] Specifically, deployed in vehicles The vehicle-mounted actuator acquires the single-base echo signal. Perform time-frequency processing (e.g., range-Doppler RD processing) to measure the target. Single-base Doppler frequency shift .Should The theoretical model is defined by the following formula: ; At the same time, the collaborative processing module processes the separated bistatic scattering wave signal. Perform time-frequency processing to measure the target. Bibase Doppler frequency shift .Should The theoretical model is defined by the following formula: ; In S4, the establishment of the coupled verification equation system continues.

[0098] Specifically, the collaborative processing module extracts from the set of geometric candidates calculated in step S3 (e.g., from the final geometric intersection). (In the middle), select a candidate position vector to be verified. .

[0099] The collaborative processing module utilizes this candidate location vector and the instantaneous position vector of the vehicle-mounted execution unit obtained in step S1. and the instantaneous position vector of the UAV execution unit Solve for the instantaneous direction unit vector corresponding to the candidate position vector. and The unit vector is calculated as follows: ; ; in, The candidate position vectors to be verified are selected from step S3; This is the instantaneous position vector of the vehicle's actuator; This represents the instantaneous position vector of the UAV execution unit; This is the Euclidean norm operation, used to calculate the magnitude (i.e., distance) of a vector. Based on candidate position vectors Calculated target With vehicles The geometric distance between them; Based on candidate position vectors Calculated target With drones The geometric distance between them.

[0100] Subsequently, the collaborative processing module will measure the single-base Doppler frequency shift. and bipolar Doppler frequency shift As known observations, and using theoretical models, we establish information about the target to be detected. instantaneous velocity vector The coupled equation set is established as follows: ; ; In step S4, the calculation and confirmation of the target state continue.

[0101] Specifically, the collaborative processing module (for a given input from...) Selected candidate position vector Solve the coupled verification equations established by ).

[0102] The solution aims to calculate the target to be detected in the coupled verification equations. instantaneous velocity vector .

[0103] The collaborative processing module examines the solution process. If the coupled verification equation set (based on...) and It can calculate a unique, physically consistent instantaneous velocity vector. ,Should If both Doppler theoretical model equations can be satisfied simultaneously, the collaborative processing module confirms: the candidate position vectors used to construct this coupled verification equation set. It is a physically verified, real instantaneous target position.

[0104] At this point, the instantaneous velocity vector is solved. Confirmed as the target to be detected The instantaneous speed, Confirmed as the target to be detected The instantaneous position.

[0105] Conversely, if the coupled verification equation set (based on and , , , )for and No solution, or the solution is not unique (e.g., leading to a theoretical model that is inconsistent with the actual solution). , If there is a contradiction, the collaborative processing module determines the candidate position vector. For a spurious solution that fails physical verification, remove it from the final geometric intersection of step S3. Remove from the list.

[0106] The collaborative processing module performs a final geometric intersection. Repeat the verification, solving, and confirmation steps for other candidate positions.

[0107] Ultimately, the collaborative processing module outputs the only result that passes both geometric constraints and dual Doppler coupling physics verification. It was identified as the target to be detected. The instantaneous state (i.e., instantaneous position and instantaneous velocity).

[0108] See attached document Figure 4 In step S5, visual-guided solution is first performed.

[0109] Specifically, the collaborative processing module utilizes the target to be detected that was finally confirmed in step S4. instantaneous position vector and the instantaneous position vector of the UAV execution unit obtained (or updated in real time) in step S1. and the rotation matrix representing its instantaneous attitude .

[0110] The collaborative processing module calculates the instantaneous position vector from the UAV execution unit. Pointing to the target instantaneous position vector The gaze vector. In a global coordinate system (e.g., WGS-84 or NED), it is calculated as: ; in, In the global coordinate system, from the UAV Pointing to the target The line-of-sight vector; The instantaneous position vector confirmed in step S4; This is the instantaneous position vector of the UAV execution unit.

[0111] Subsequently, the collaborative processing module utilizes the rotation matrix of the instantaneous attitude of the UAV execution unit. , the line of sight vector Switch to drones In the body coordinate system, the body line-of-sight vector is obtained. The calculation formula is as follows: ; in, For drones The line-of-sight vector in the body coordinate system; Let be the rotation matrix from the global coordinate system to the body coordinate system, representing the UAV. The instantaneous posture.

[0112] The collaborative processing module is based on the machine's line-of-sight vector. Solve the control of the drone The onboard optical sensors (e.g., photoelectric pods or cameras) are aimed at the target to be detected. Required guidance commands, such as target azimuth. and target pitch angle .

[0113] Assumption ,in Forward-facing, The body is facing to the right. If the coordinates are in the bottom orientation (e.g., NED body coordinate system), then the guidance command is calculated as follows: ; ; in, The target azimuth angle of the optical sensor (relative to the body coordinate system); The target pitch angle of the optical sensor (relative to the body coordinate system); For the line-of-sight vector Components in the body coordinate system; It is a two-parameter arctangent function.

[0114] The collaborative processing module will solve the boot instructions (e.g., and Send to drone The execution unit drives the optical sensor, ensuring its line of sight is precisely aligned with the instantaneous position vector confirmed in step S4. The physical space region where it is located prepares for semantic confirmation.

[0115] In step S5, semantic verification continues.

[0116] Specifically, deployed on drones Optical sensors, in guiding instructions (e.g., and Driven by the optical sensor, its line of sight is aligned with the physical space region where the instantaneous position vector of the target position confirmed in step S4 is located. The optical sensor acquires one or more frames of optical image data of this region. .

[0117] Collaborative processing module (or drone) The onboard processing unit acquires the optical image data. And call the pre-trained semantic recognition model. (For example, deep learning object detection or semantic segmentation networks) analyze the image data.

[0118] This semantic recognition model For image data The image is processed to calculate the corresponding instantaneous position vector captured in the image. semantic tags : ; in, Optical image data acquired by an optical sensor; For semantic recognition model; For the semantic tags calculated (e.g., birds, drones, buildings, tree branches, etc.).

[0119] The collaborative processing module will solve the semantic tag. With the pre-defined threat target semantic library Perform a comparison. If the... match The collaborative processing module determines the instantaneous state calculated in step S4 based on the entries in the table (e.g., tagged as drones or obstacles). A real threat confirmed through semantics. If this... Mismatch The co-processing module determines the instantaneous state calculated in step S4 based on the entries in the data (e.g., labeled as birds or clutter). The radar is flagged as a false alarm and marked as invalid. The collaborative processing module then finalizes the instantaneous state of the real threat, which has undergone semantic verification in step S5. The output is then sent to the subsequent collision avoidance decision module.

[0120] In step S6, the early warning packet is generated and transmitted first.

[0121] Specifically, the collaborative processing module acquires the target to be detected that represents a real threat and has finally passed semantic confirmation in step S5. instantaneous state .

[0122] The collaborative processing module simultaneously obtains the semantic tags of the target calculated in step S5. The collaborative processing module also obtains the current system timestamp. This is to ensure the timeliness of the early warning information.

[0123] The collaborative processing module is based on semantic tags. and system timestamp Generate standardized collaborative early warning packages. This collaborative early warning package It contains the following data fields: ; in, For collaborative early warning packages; The instantaneous state (position vector and velocity vector) of the threat target finally confirmed in step S5. Semantic labels (e.g., drones) for the threat targets finally confirmed in step S5; This is the system timestamp when the warning package was generated.

[0124] Generate this collaborative early warning package Subsequently, the collaborative processing module transmits the collaborative early warning package via the collaborative data link established in step S1. Broadcast or directed transmission to vehicles Vehicle-mounted actuators and drones The execution unit.

[0125] This transmission provides a unified, multi-validated (geometric, physical, and semantic) threat situation data for subsequent collaborative collision avoidance decisions.

[0126] In step S6, the decision execution continues.

[0127] Specifically, deployed in vehicles The vehicle-mounted execution unit, and the drone deployed All execution units receive collaborative early warning packets. .

[0128] The on-board execution unit obtains this instantaneous state And combined with the vehicle information obtained in step S1 Its own instantaneous kinematic state .

[0129] The onboard actuator is based on a kinematic model (e.g., a uniform linear motion model) to control the vehicle. and the target to be detected The future trajectory is calculated to obtain the respective future moments. Predicted position vector and Perform the calculation. The trajectory calculation formula is: ; ; in, For vehicles In the future The predicted position vector; Target to be detected In the future The predicted position vector; This represents the instantaneous position vector of the onboard execution unit; This represents the instantaneous velocity vector of the onboard actuator. Target to be detected The instantaneous position vector; Target to be detected The instantaneous velocity vector; This is the system timestamp when the warning package was generated.

[0130] The onboard execution unit calculates the instantaneous distance between two predicted position vectors at a future time. : ; in, This is an Euclidean norm operation used to calculate the magnitude (i.e., distance) of a vector.

[0131] The on-board actuator will measure the instantaneous distance With the preset vehicle safe distance threshold Compare them.

[0132] If within the preset future time window Inside (e.g.) ),exist If the vehicle's actuator determines that there is a collision risk, it will execute preset vehicle collision avoidance actions (e.g., issuing a warning to the driver, activating automatic emergency braking, or performing steering avoidance).

[0133] Meanwhile, deployed on drones The execution units execute decisions in the same way.

[0134] The execution unit of the drone solves the drone Predicted position vector : ; in, For drones In the future The predicted position vector; This represents the instantaneous position vector of the UAV execution unit; This is the instantaneous velocity vector of the UAV execution unit.

[0135] drones The execution unit calculates its relationship with the target to be detected. In the future instantaneous distance : ; in, This is an Euclidean norm operation used to calculate the magnitude (i.e., distance) of a vector.

[0136] drones The execution unit will measure the instantaneous distance With the pre-set drone safe distance threshold Compare them.

[0137] If in that future time window Inside, there exists Then drone The execution unit determines that there is a collision risk and executes the preset drone collision avoidance actions (e.g., automatic climb, hovering, or maneuvering avoidance).

Claims

1. A method for cooperative pre-warning anti-collision of fusing wide-area perception of UAV and vehicle-mounted radar, characterized in that, Includes the following steps: S1. A unified spatiotemporal reference is established by the vehicle-mounted execution unit and the UAV execution unit. The vehicle-mounted execution unit broadcasts beacon information containing radar signal parameters and its own kinematic state. S2. Multipath radar signals are acquired through the active radar module of the vehicle-mounted execution unit and the passive radar receiving module of the UAV execution unit, and the mixed signals are separated. S3. Based on the separated signal components and the kinematic state information of the vehicle-mounted execution unit and the UAV execution unit, the collaborative processing module calculates the path distance difference and constructs a geometric candidate set to determine the final geometric intersection of the target to be detected. S4. The collaborative processing module extracts the Doppler frequency shift measurement value from the multipath radar signal, and establishes a coupled verification equation set based on the final geometric intersection and the Doppler frequency shift measurement value to confirm the instantaneous position vector and instantaneous velocity vector of the target to be detected; S5. The UAV execution unit guides the visual perception module to acquire optical images, and confirms the real threat of the target to be detected through semantic analysis; S6. The collaborative processing module integrates the confirmed instantaneous position vector, instantaneous velocity vector and semantic tag of the target to be detected, generates and transmits a collaborative warning package, and the vehicle-mounted execution unit and the UAV execution unit each perform collision avoidance decision and collision avoidance action.

2. The method of claim 1, wherein the method further comprises: In step S1, the high-precision clock module of the vehicle-mounted execution unit and the high-precision clock module of the UAV execution unit execute a precise time protocol. By exchanging multiple timing messages, the clock offset between the slave clock and the master clock and the message propagation delay are reduced, and the local clock is adjusted to achieve time synchronization.

3. The method of claim 1, wherein the method further comprises: In step S1, the waveform parameters of the radar signal include the center carrier frequency, sweep bandwidth, sweep period, and frequency modulation slope. The instantaneous kinematic state of the vehicle-mounted actuator includes an instantaneous position vector and an instantaneous velocity vector.

4. The method of claim 1, wherein the method further comprises: In step S2, the single-base receiver of the vehicle-mounted execution unit receives the single-base echo signal; The passive radar receiving module of the UAV execution unit receives the mixed signal, which includes a direct wave signal and a bistatic scattered wave signal; The collaborative processing module generates a local reference signal using the waveform parameters of the radar signal, and identifies and separates the direct wave signal and the bistatic scattered wave signal by performing cross-correlation processing between the local reference signal and the mixed signal.

5. The method of claim 4, wherein the method further comprises: In step S3, the collaborative processing module calculates the path distance difference based on the time difference of arrival of the direct wave signal and the bistatic scattered wave signal. The first geometric candidate set is a rotating ellipsoid, the foci of which are the instantaneous position vectors of the vehicle-mounted execution unit and the UAV execution unit; The second geometric candidate set is a sphere, the center of which is the instantaneous position vector of the vehicle-mounted execution unit, and the radius of which is the instantaneous distance of the single-base echo signal calculated from the two-way propagation delay.

6. The method of claim 4, wherein the method further comprises: In step S4, the collaborative processing module extracts the single-base Doppler frequency shift from the single-base echo signal and the bibase Doppler frequency shift from the bibase scattered wave signal in parallel. The coupled verification equation set includes both theoretical models of single-base Doppler frequency shift and theoretical models of double-base Doppler frequency shift. The collaborative processing module obtains the instantaneous velocity vector of the target to be detected by solving the coupled verification equations, and confirms the instantaneous position vector and the instantaneous velocity vector that uniquely satisfy the physical constraints in the final geometric intersection by evaluating the residual between the theoretical model and the measured value.

7. The method of claim 1, wherein the method further comprises: In step S5, the collaborative processing module uses the confirmed instantaneous position vector of the target to be detected, the instantaneous position vector of the UAV execution unit, and the rotation matrix representing the instantaneous attitude of the UAV execution unit to calculate the line-of-sight vector from the UAV execution unit to the target to be detected, and transforms the line-of-sight vector into the body coordinate system of the UAV execution unit to obtain the body line-of-sight vector. The collaborative processing module calculates the guidance commands required by the optical sensor carried by the UAV execution unit based on the body line-of-sight vector.

8. The method of claim 1, wherein the method further comprises: In step S5, the collaborative processing module calls a pre-trained semantic recognition model to analyze the optical image to calculate the semantic label, and compares the semantic label with a preset threat target semantic database to determine the true threat level of the target to be detected.

9. The method of claim 1, wherein the method further comprises: In step S6, the collaborative early warning package includes the confirmed instantaneous position vector and instantaneous velocity vector of the target to be detected, the semantic tag, and the system timestamp when the collaborative early warning package was generated.

10. The method of claim 1, wherein the method further comprises: In step S6, the vehicle-mounted execution unit and the UAV execution unit each predict their respective trajectories with the target to be detected within a future time window based on the instantaneous position vector and instantaneous velocity vector received from the cooperative early warning package, as well as their respective instantaneous kinematic states. They then determine whether there is a collision risk by comparing the predicted distance with their respective safe distance thresholds and execute corresponding collision avoidance actions.

Images