Method for coordinating and scheduling multiple robots based on large model to realize vehicle detection

Abstract

The present application relates to the technical field of robot data monitoring, and particularly relates to a method for realizing vehicle detection and multi-robot coordination scheduling based on a large model, comprising: acquiring driving time window and driving speed data of a robot to be detected driving to a current section and background traffic logs; constructing a passing speed grid map of the current section according to the distribution of the current section and the driving time window, in combination with the background traffic logs; simulating driving time distribution to obtain a motion deduction deviation degree according to the driving speed data and the passing speed grid map; obtaining an identity verification result according to the angle deviation between the termination position and the initial position of the robot to be detected on the current section and the shooting position of a monitoring device, in combination with the motion deduction deviation degree, for coordinating and scheduling the robot to be detected. The present application provides a key identity benchmark for multi-robot coordination scheduling, and effectively avoids path conflicts, task discontinuities and other problems caused by identity loss.

Description

Technical Field

[0001] This invention relates to the field of robot data monitoring technology, specifically to a method for coordinated scheduling of multiple robots for vehicle detection based on a large model. Background Technology

[0002] In scenarios such as closed logistics parks and automated ports, the core prerequisite for the coordinated scheduling of robots such as AGVs and unmanned trucks is to achieve continuous and accurate vehicle identification. Roadside sensing units (such as cameras and millimeter-wave radar) are deployed at key nodes in the operating area. However, due to the large area and numerous obstructions, the field of view of these sensing units cannot completely cover the entire area, resulting in uncontrolled blind spots between observation points.

[0003] When a vehicle passes through a blind spot and enters the next observation point from a different angle, its visual appearance features drift significantly due to the drastic change in shooting perspective (such as from a frontal view to a side view). This problem directly causes traditional visual re-identification algorithms to fail. Furthermore, the blind spot contains uncertainties such as unconnected vehicles and temporary obstacles, making road conditions highly variable. The actual travel time of vehicles fluctuates non-linearly, and traditional prediction methods based on static distance or average speed cannot distinguish whether this fluctuation originates from road congestion or the vehicle's own performance limitations. The failure of visual recognition combined with inaccurate motion duration prediction leads to the system's inability to determine whether preceding and following observations across different viewpoints belong to the same vehicle, resulting in a loss of vehicle identity continuity. This causes a disconnect in scheduling contexts such as task priority and right-of-way locks, ultimately leading to scheduling chaos such as robot path conflicts and duplicate task allocation, severely impacting operational efficiency and stability. Summary of the Invention

[0004] To address the issue of poor accuracy in identity verification results for cross-view robotic vehicles using existing methods, this invention aims to provide a multi-robot coordinated scheduling method for vehicle detection based on a large model. The specific technical solution adopted is as follows:

[0005] Acquire the travel time window and travel speed data of the robot to be detected in the current road segment, and construct the background traffic log by recording the vehicle travel records in the current road segment within the travel time window;

[0006] Based on the current road segment distribution and the distribution of the robot's driving time window, combined with the spatiotemporal characteristics of vehicle driving records in the background traffic log, a grid map of the current road segment's traffic speed is constructed.

[0007] Based on the driving speed data of the robot under test and the traffic speed grid map, the driving time distribution of the robot under test in the current road segment is simulated to obtain the degree of deviation in the motion inference of the robot under test.

[0008] Based on the angular deviations between the robot's termination and initial positions on the current road segment and the camera's shooting position, and combined with the degree of motion deduction deviation, the robot's identity verification result on the current road segment is obtained, which is used for coordinated scheduling of the robot.

[0009] Preferably, the travel time window is the time period between the entry time and the exit time of the robot to be detected to the current road segment;

[0010] The driving speed data includes the robot's maximum driving speed, maximum acceleration, and maximum deceleration.

[0011] Vehicle driving records include the driving time, location, and speed of each authenticated device.

[0012] Preferably, the step of constructing a traffic speed grid map of the current road segment based on the current road segment distribution and the distribution of the robot's travel time window, combined with the spatiotemporal characteristics of vehicle travel records in the background traffic log, specifically includes:

[0013] Using the current road segment's mileage as the first dimension and the total travel time window's duration as the second dimension, the total mileage of the current road segment and the total travel time window are uniformly divided to construct a two-dimensional grid map.

[0014] Based on the vehicle speed and travel time recorded in the background traffic log, and combined with the speed limit information of the current road segment, the speed information of the unit grid in the two-dimensional grid map is updated to obtain the traffic speed grid map of the current road segment.

[0015] Preferably, the step of updating the speed information of the unit grids in the two-dimensional grid map based on the vehicle driving records in the background traffic log and the speed limit information of the current road segment to obtain the traffic speed grid map of the current road segment specifically includes:

[0016] The maximum speed limit for the current road segment is used as the speed information of the cell grid in the two-dimensional grid map. The cell grid corresponding to the vehicle's driving record is determined by the driving time and driving position of the vehicle in the current road segment. The speed information of the cell grid corresponding to the vehicle's driving record is updated by the minimum value between the driving speed of the vehicle's driving record and the maximum speed limit for the current road segment, thus obtaining the traffic speed grid map of the current road segment.

[0017] Preferably, the step of simulating the travel time distribution of the robot under test in the current road segment based on the travel speed data of the robot under test combined with the traffic speed grid map to obtain the degree of motion inference deviation of the robot under test specifically includes:

[0018] Initialize the iteration steps and travel distance of the robot under test in the current road segment, and obtain the preset time step and the simulated speed of the robot under test;

[0019] At each preset time step, the simulated speed of the robot under test is updated based on the driving speed data of the robot under test and the speed information of the unit grid in the current road segment's traffic speed grid diagram, and the driving distance of the robot under test at each preset time step is calculated.

[0020] The robot under test is iterated over the current road segment at a preset time step. The iteration stops when the travel distance is greater than or equal to the total mileage of the current road segment. The cumulative value of the preset time step under all iteration steps is obtained to obtain the simulated travel time of the robot under test.

[0021] Based on the proportion of cell grids in the traffic speed grid diagram where the speed information is updated, a gain coefficient is determined. The degree of motion inference deviation of the robot under test is obtained by weighting the proportion of the difference between the simulated driving time length of the robot under test and the driving time window using the gain coefficient.

[0022] Preferably, the step of updating the simulated speed of the robot under test based on the robot's driving speed data and the speed information of the unit grid in the current road segment's traffic speed grid map, and calculating the robot's driving mileage at each preset time step, specifically includes:

[0023] Obtain the initial speed of the robot to be tested as it travels to the current road segment;

[0024] For any iteration, based on the initial velocity and maximum acceleration of the robot under test, calculate the theoretical travel distance of the robot under test at a preset time step;

[0025] The theoretical travel distance under the preset time step is used to determine the cell grid range of the preset time step in the traffic speed grid diagram. The speed information within the cell grid range and the minimum value of the highest travel speed of the robot under test are obtained as the simulated speed under the preset time step.

[0026] Based on the initial speed and simulated speed of the robot under test, the simulated acceleration of the robot under test from the initial speed to the simulated speed is calculated. Based on the initial speed and simulated speed of the robot under test, the travel distance of the robot under test at the preset time step and the updated simulated speed are calculated.

[0027] Preferably, the step of obtaining the identity verification result of the robot under test in the current road segment based on the angular deviation between the termination position and the initial position of the robot under test in the current road segment and the shooting position of the monitoring device, combined with the degree of motion deduction deviation, specifically includes:

[0028] Based on the deviation between the observation angles corresponding to the travel direction of the robot under test at the end position and the travel direction at the initial position in the current road segment, the visual weight of the robot under test in the current road segment is obtained.

[0029] The cosine similarity between the image feature vector of the image taken by the robot at the end position of the current road segment and the image feature vector of the image taken at the initial position is obtained and used as a visual similarity index of the robot in the current road segment.

[0030] Using the aforementioned visual weights, the visual similarity index and motion inference deviation of the robot to be tested are fused to obtain the verification confidence of the robot to be tested in the current road segment;

[0031] Based on the verification confidence level of the robot to be tested in the current road segment and the preset confidence threshold, obtain the identity verification result of the robot to be tested in the current road segment.

[0032] Preferably, obtaining the visual weight of the robot under test in the current road segment based on the deviation between the observation angles corresponding to the travel direction at the end position and the travel direction at the initial position of the robot under test in the current road segment specifically includes:

[0033] Based on the termination position and initial position of the robot to be detected in the current road segment, the lane tangential vector at the termination position and the lane tangential vector at the initial position of the robot to be detected in the current road segment are determined respectively, and the preset optical axis orientation vector of the detection system is obtained.

[0034] The angle between each lane tangential vector and the optical axis orientation vector is calculated as the observation angle corresponding to each lane tangential vector. The absolute value of the cosine of the absolute value of the difference between the observation angles corresponding to the termination position and the initial position of the robot under test in the current road segment is used as the visual weight of the robot under test in the current road segment.

[0035] Preferably, the step of using the visual weights to fuse the visual similarity index and motion inference deviation of the robot to be detected to obtain the verification confidence level of the robot to be detected in the current road segment specifically includes:

[0036] The verification confidence level of the robot under test in the current road segment is obtained by weighting and summing the negative correlation coefficient of the degree of deviation in motion inference of the robot under test and the visual similarity index.

[0037] Among them, the negative correlation coefficient of the degree of deviation in motion deduction corresponds to the negative correlation coefficient of visual weight, and the weight corresponding to the visual similarity index is the visual weight.

[0038] Preferably, obtaining the identity verification result of the robot to be detected in the current road segment based on the verification confidence level of the robot to be detected in the current road segment and a preset confidence threshold specifically includes:

[0039] When the verification confidence level of the robot to be tested in the current road segment is greater than or equal to the preset confidence threshold, the robot to be tested passes the identity verification.

[0040] When the verification confidence level of the robot to be tested in the current road segment is less than the preset confidence threshold, the robot to be tested fails the authentication.

[0041] The embodiments of the present invention have at least the following beneficial effects:

[0042] This invention first acquires the robot's driving time window and speed data, while integrating vehicle driving records for the current road segment during that time period to construct a background traffic log. This provides a complete and real-time data source for subsequent dynamic road condition modeling, ensuring a solid data foundation for environmental constraint analysis. Then, combining the spatial distribution of the current road segment, the robot's driving time window, and the spatiotemporal characteristics of the background traffic log, a traffic speed grid map is constructed. This transforms abstract traffic flow information into structured speed-spatiotemporal information, accurately representing the upper limit of actual traffic capacity at different times and on different road segments. This solves the problem that static speed limits cannot adapt to sudden scenarios such as temporary congestion, providing an accurate environmental benchmark for motion simulation. Furthermore, based on the robot's driving speed data and the traffic speed grid map, constrained kinematics simulation is performed to simulate the robot's theoretical driving time under current road conditions. Comparing this simulation with the actual driving time yields the degree of deviation in motion simulation, providing objective physical evidence for identity verification and avoiding the risk of misjudgment based solely on visual features. Finally, by combining the angular deviation of the robot's initial and final positions relative to the monitoring equipment (quantifying the credibility of visual features) and the degree of deviation in motion deduction (quantifying the matching degree of physical behavior), cross-view authentication is completed, providing a key identity benchmark for multi-robot coordinated scheduling, effectively avoiding problems such as path conflicts and task gaps caused by identity loss, and realizing a closed loop of scheduling logic. Attached Figure Description

[0043] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This invention provides a flowchart of the steps involved in a multi-robot coordinated scheduling method for vehicle detection based on a large model.

[0045] Figure 2 This is a flowchart of the steps for obtaining the degree of motion deduction deviation of the robot to be tested provided by the present invention;

[0046] Figure 3 This is a flowchart of the steps of the method for obtaining the identity verification result of the robot to be detected in the current road segment provided by the present invention. Detailed Implementation

[0047] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a multi-robot coordinated scheduling method for vehicle detection based on a large model proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] The following description, in conjunction with the accompanying drawings, details a specific scheme for a multi-robot coordinated scheduling method for vehicle detection based on a large model, provided by this invention.

[0050] Please see Figure 1 The diagram illustrates a flowchart of a multi-robot coordinated scheduling method for vehicle detection based on a large model, according to an embodiment of the present invention. The method includes the following steps:

[0051] Step S100: Obtain the driving time window and driving speed data of the robot to be detected traveling to the current road segment, and construct the background traffic log by recording the vehicle driving within the driving time window of the current road segment.

[0052] First, because key nodes within the operational area will deploy roadside sensing units (such as cameras and millimeter-wave radar), and because the installation height, elevation angle, and lighting conditions of each roadside sensing unit are independent and inconsistent, directly using pixel coordinates from the image plane cannot establish a unified spatial reference in multi-machine collaboration. Furthermore, collaborative scheduling relies on a globally consistent time base. Therefore, mapping the two-dimensional detection boxes to the topological coordinate system of the high-precision map using a pre-calibrated extrinsic parameter matrix, and labeling them based on a unified clock, is a prerequisite for achieving spatial location normalization.

[0053] Specifically, during the system initialization phase, a high-precision map of the entire field is pre-constructed, containing unique identifiers and topological length information for all drivable road segments. Simultaneously, the intrinsic and extrinsic parameter matrices of each roadside sensing unit are calibrated, establishing the transformation relationship between the camera coordinate system and the world coordinate system. All roadside sensing units maintain microsecond-level time synchronization with the central server via Precise Time Protocol (PTP).

[0054] During operation, the roadside perception unit acquires video image data in real time. Using an object detection algorithm, it performs object detection on the video image data. When the detection result indicates that the detection box of a target robot vehicle touches a preset field of view boundary, the system triggers a stop-pass event and records the global UTC timestamp of the event. The preset field of view boundary is a virtual limit set within the camera's detection range to trigger the stop-pass event. It needs to be set according to the specific implementation scenario and can be set as a virtual line at the road segment entrance or exit, parallel to the lane width direction. The object detection algorithm is a known technology, such as existing CNN-based object detection neural networks used to track and identify target robot vehicles in video image data.

[0055] Furthermore, read the camera's intrinsic parameter matrix. and the extrinsic parameter matrix relative to the geodetic coordinate system Using the ray casting algorithm, the center pixel coordinates of the bottom edge of the target robot vehicle's bounding box are selected and projected backwards onto the road surface (i.e., the height). (on the plane), obtain the projection point of the detection box of the target robot vehicle, and calculate the three-dimensional coordinates of the projection point in the world coordinate system. Subsequently, the system locates the lane centerline of the road segment closest to the projection point in the high-precision map, obtains the road segment identifier to which the projection point belongs, and the cumulative mileage offset of the target robot vehicle relative to the starting point of the road segment.

[0056] In high-precision maps, road segments are typically described using the Frenet coordinate system. The method for obtaining the cumulative mileage offset of the target robot vehicle is as follows: the projection points of the target robot vehicle's detection box are orthogonally projected onto the nearest lane centerline in the world coordinate system to find the nearest point (perpendicular foot) on the centerline. The cumulative curve length from the starting position of the road segment to the perpendicular foot point along the centerline trajectory is calculated as the cumulative mileage offset of the target robot vehicle when a stop-pass event is triggered. Finally, the timestamp of the robot vehicle's stop-pass event, the road segment identifier, and the mileage offset constitute the spatiotemporal coordinate tuple of the robot vehicle.

[0057] The travel time window is the time interval between the entry and exit times of the robot under test on the current road segment. More specifically, the travel time window of the robot under test on the current road segment represents the range of travel time between two topologically adjacent observation points after the robot under test triggers a stop-passing event. Here, an observation point refers to the location where the roadside sensing unit is deployed. More specifically, for the robot under test, the timestamp corresponding to the stop-passing event recorded in the data collected from observation point A is recorded as the first time point (i.e., the entry time), and the timestamp corresponding to the stop-passing event recorded in the data collected from observation point B is recorded as the second time point (i.e., the exit time). The time range between the first and second time points constitutes the travel time window of the robot under test. Here, the current road segment refers to the road segment between observation point A and observation point B, and observation point B is topologically adjacent to observation point A. A stop-passing event refers to the robot vehicle's detection frame touching a preset field-of-view boundary as described above.

[0058] The process of obtaining the travel time window is essentially that the server, based on the topological relationship of the high-precision map, filters out the pairs of transit events that are topologically adjacent to the preceding observation point A and the following observation point B. This allows the server to calculate the shortest path and theoretical limit travel time between the preceding observation point A and the following observation point B for the spatiotemporal causal relationship generated by each transit event pair. Event pairs with time differences that do not conform to physical laws are excluded, further narrowing down the candidate range and providing a data foundation for the identity verification process of the robot to be detected.

[0059] The travel speed data includes the robot under test's maximum travel speed, maximum acceleration, and maximum deceleration. This data characterizes the inherent motion performance specifications of the robot vehicle. The maximum travel speed represents the robot's designed maximum speed, i.e., the upper limit of its physical speed determined by motor characteristics and gear ratios, measured in m / s; the maximum acceleration is the robot's maximum acceleration capability under standard operating conditions, measured in m / s². 2 Maximum deceleration: The maximum deceleration capacity of the vehicle's braking system (absolute value), in m / s². 2 The roadside sensing unit can run a lightweight classification network to identify the vehicle type or vehicle specification category of the robot under test. If the category output by the classification network is not in the device specification library (e.g., unregistered social vehicles), the system calls a preset default parameter set, which adopts the most conservative dynamic parameters of the entire vehicle model to ensure the safety coverage of subsequent simulations.

[0060] Vehicle driving records include the driving time, location, and speed of each authenticated device. The driving records of vehicles on the current road segment within the driving time window constitute the background traffic log. This log consists of real-time telemetry data reported by all authenticated operational equipment (such as AGVs with normal RTK-GPS signals and clearly defined task statuses). Specifically, vehicle driving records indicate the location and speed of a particular device at a given time. This log serves as a globally shared, read-only resource, providing data support for reconstructing the dynamic environment and resolving the cold start problem of simulation environment input.

[0061] Because of the physical limitations on vehicle mechanical performance, the movement of any entity in physical space is constrained by the shortest path and the maximum design speed across the entire field. To conserve computational resources, invalid event pairs caused by sensor false alarms or timestamp drift are first eliminated using static physical boundaries.

[0062] Specifically, based on the location information at the beginning and end of the travel time window, the shortest connected path is searched in the high-precision map topology network to obtain the total mileage of the shortest connected path. Simultaneously, the maximum design speed for all vehicle models in the equipment specification library is obtained. If the travel time window is shorter than the theoretical limit, it is determined that the two events do not have physical causal continuity, and subsequent processing is terminated directly. The theoretical limit is the ratio of this total mileage to the maximum design speed.

[0063] Step S200: Based on the distribution of the current road segment and the distribution of the driving time window of the robot to be detected, and combined with the spatiotemporal characteristics of vehicle driving records in the background traffic log, construct a grid map of the traffic speed of the current road segment.

[0064] The main purpose of this step is to quantify the dynamic traffic constraints of uncontrolled road sections, construct a traffic speed grid map that reflects real-time road conditions, and provide an accurate environmental benchmark for subsequent constrained kinematics deductions based on vehicle vehicle motion performance.

[0065] Traffic conditions on uncontrolled road sections are highly time-varying. Static speed limits or historical average speeds cannot match sudden congestion, obstacles, and other emergencies, directly leading to significant discrepancies between traditional travel time predictions and actual conditions. Therefore, this step requires a three-level logical progression to construct the grid map: first, based on the spatial distribution characteristics of the current road section, spatial grid units covering the entire path length are divided; second, the time dimension range of the grid is determined by combining the driving time window of the robot to be detected; finally, vehicle driving records falling within this spatiotemporal range are extracted from the background traffic log, and the upper limit of the travel speed of each grid unit is updated based on the "weakest link effect," that is, the initial value of the grid is corrected by the actual driving speed of other identified vehicles, simulating the spatiotemporal propagation characteristics of congestion. This process transforms abstract road condition information into a structured and refined speed spatiotemporal matrix, achieving accurate quantification of the dynamic environment of uncontrolled road sections, and providing a data foundation for subsequent analysis of environmental constraints and individual capabilities, and the generation of reliable physical and logical evidence.

[0066] Therefore, the method for obtaining the traffic speed grid map of the current road segment can be achieved by steps S201 and S202.

[0067] Step S201: Using the mileage of the current road segment as the first dimension and the total time of the travel time window as the second dimension, the total mileage of the current road segment is uniformly divided, and the total time of the travel time window is uniformly divided to construct a two-dimensional grid map.

[0068] The main purpose of this step is to construct a structured grid carrier covering the entire spatiotemporal range of uncontrolled road segments, providing a unified spatiotemporal reference for the subsequent filling of dynamic road condition information. Traffic constraints on uncontrolled road segments involve both spatial location and temporal changes, requiring standardized grid partitioning to achieve full coverage of the shortest connected path. Therefore, this step uses the total mileage of the current road segment as the first spatial dimension and the total duration of the robot's travel time window as the second temporal dimension, uniformly dividing both dimensions to construct a two-dimensional grid map. This grid map accurately covers the total mileage of the shortest connected path corresponding to the initial and final positions within the travel time window, transforming the abstract spatiotemporal range into a computable and indexable set of structured units, laying the foundation for the subsequent injection of real-time speed constraints.

[0069] As a specific example, the total mileage of the current road segment is evenly divided with a preset length step, and the total time length of the travel time window is evenly divided with a preset time step. The length step can be 1 meter and the time step can be 1 second. The implementer can set these values ​​according to the specific implementation scenario.

[0070] Step S202: Based on the driving speed and driving time contained in the vehicle driving records in the background traffic log, and combined with the speed limit information of the current road segment, update the speed information of the unit grid in the two-dimensional grid map to obtain the traffic speed grid map of the current road segment.

[0071] The main purpose of this step is to inject real-time traffic flow information from the background into the grid framework, generating a traffic speed grid map that reflects the dynamic capacity of road segments, providing accurate environmental constraints for subsequent constrained kinematics simulations. The initial two-dimensional grid map only has a spatiotemporal framework and lacks speed constraints based on actual road conditions, making it unsuitable for the time-varying traffic conditions of uncontrolled road segments. Therefore, this step uses the spatiotemporal characteristics of identified vehicles in the background traffic logs, such as their speed and travel time, combined with the inherent design speed limit of the current road segment, to iteratively update the speed value of each cell in the two-dimensional grid map according to the "weakest link effect." That is, the initial grid values ​​are corrected using the actual traffic speeds of background vehicles, simulating the spatiotemporal propagation characteristics of traffic conditions such as congestion. The final generated traffic speed grid map accurately represents the upper limit of actual traffic speeds at different times and on different road segments, realizing the transformation of abstract traffic conditions into structured and quantifiable data.

[0072] Specifically, the maximum speed limit of the current road segment is used as the speed information of the unit grid in the two-dimensional grid map. The driving time and speed of the vehicle driving record are used to determine the unit grid corresponding to the vehicle driving record. The speed information of the unit grid corresponding to the vehicle driving record is updated by the minimum value between the driving speed of the vehicle driving record and the maximum speed limit of the current road segment, so as to obtain the traffic speed grid map of the current road segment.

[0073] It should be understood that the speed information initialized in the cell grid of the two-dimensional grid diagram is the speed limit information when the road segment was designed, which represents the maximum speed allowed to travel on that road segment.

[0074] Furthermore, the system searches the background traffic logs for all vehicle driving records that fall within the driving time window of the robot to be detected and whose location information is within the corresponding path range. For each vehicle driving record, the system calculates the index of the corresponding vehicle's mapping cell grid in the two-dimensional grid diagram, which is used to update the speed information in the two-dimensional grid diagram.

[0075] As a concrete example, for any given vehicle's driving record cell grid, the spatial index... Time Index ,in This represents the distance traveled by the vehicle at time t within the current road segment. This indicates the preset length step. This indicates the driving time recorded in the vehicle's driving log. Indicates the preset time step. This indicates the initial time of the travel time window. This indicates the length of time that the vehicle's driving time falls within the driving time window. This indicates a round-down operation. It rounds down the cell containing the vehicle's driving record. The speed information is updated to the minimum between the vehicle's recorded speed and the maximum speed limit for the current road segment. This update logic is based on the "bottleneck effect," meaning that if a background vehicle is traveling at a low speed at a certain point, it indicates the presence of congestion or an obstacle, and the speed limit for subsequent vehicles should be constrained accordingly. To simulate the persistence of congestion, this speed constraint is propagated along the time axis to future time steps until the preset congestion dissipation period ends.

[0076] It should be noted that if a certain cell grid has no background vehicle driving records within the driving time window of the robot to be tested (i.e., the background traffic log is empty), the speed information of that cell grid will remain the speed limit information of the current road segment to prevent the simulation from failing due to missing data.

[0077] Step S300: Based on the driving speed data of the robot under test and the traffic speed grid map, simulate the driving time distribution of the robot under test in the current road segment to obtain the degree of motion inference deviation of the robot under test.

[0078] The main purpose of this step is to decouple environmental constraints from the robot's individual motion performance, quantify the degree of deviation between the actual driving behavior and the theoretically deduced behavior of the robot under test, and provide key physical and logical evidence for subsequent cross-view vehicle identity determination.

[0079] The travel time on uncontrolled road segments is affected by both road congestion and vehicle performance. Traditional methods cannot distinguish between these two factors, leading to significant discrepancies between predicted and actual travel times, which makes reliable identity verification difficult. Therefore, this step first simulates the robot's travel process on the current road segment based on the robot's speed data and the dynamic road condition constraints represented by the traffic speed grid, obtaining the predicted travel time distribution. Then, the difference between the predicted and actual travel times is compared, and the deviation is weighted and corrected using the sample coverage of background traffic logs. Finally, the degree of motion prediction deviation that objectively reflects the rationality of the robot's driving behavior is calculated. This process combines abstract road condition constraints with concrete vehicle motion, achieving accurate quantitative analysis of robot driving behavior and providing a solid physical basis for solving the cross-view identity recognition problem.

[0080] In this regard, such as Figure 2 As shown, the method for obtaining the degree of motion deduction deviation of the robot under test can be implemented by steps S301 to S303.

[0081] Step S301: Initialize the iteration steps and travel distance of the robot to be tested in the current road segment, obtain the preset time step and the simulated speed of the robot to be tested; under each preset time step, update the simulated speed of the robot to be tested according to the travel speed data of the robot to be tested and the speed information of the cell grid in the traffic speed grid map of the current road segment, and calculate the travel distance of the robot to be tested under each preset time step.

[0082] The main purpose of this step is to establish initial parameter benchmarks for the simulation of the robot's driving process, complete the iterative calculation of single-step motion states, and provide time-step state data support for subsequent full-process driving time extrapolation. Constrained kinematics extrapolation needs to consider both the robot's own performance and the dynamic road conditions of the road segment. Therefore, this step first initializes basic simulation parameters such as the number of iteration steps and driving distance, and obtains the preset time step and the robot's initial simulated speed. Then, within each time step, the simulated speed is dynamically updated by combining the robot's own driving speed data with the speed upper limit constraint of the corresponding cell in the traffic speed grid diagram. Simultaneously, the robot's driving distance within that time step is calculated, realizing the simulation start-up from the initial state to gradual dynamic updates, thus laying a solid data foundation for the full-process iterative extrapolation.

[0083] The first step is to obtain the initial speed of the robot to be tested as it travels to the current road segment.

[0084] It should be noted that during the simulation process, the iteration steps and travel distance of the robot under test in the current road segment are first initialized to 0, indicating that the initial iteration steps and initial travel distance of the simulation process are 0. Then, the simulation is performed iteratively with a preset time step as the time unit to obtain the simulated speed of the robot under test in each iteration. The iteration step is incremented by 1 at the beginning of each iteration. It should be understood that when the initial iteration step is 0, the simulation operation has not started. The simulated speed at this time is set as the actual travel speed of the robot under test at the initial moment of the travel time window corresponding to the current road segment.

[0085] It should be noted that in the first iteration, the initial speed of the robot under test is the actual speed at the beginning of the driving time window, and the initial speed of each subsequent iteration is the simulated speed after the end of the previous iteration.

[0086] The second step is to calculate the theoretical travel distance of the robot under test at a preset time step, based on the initial velocity and maximum acceleration of the robot under test in any iteration process.

[0087] The main purpose of this step is to estimate the mileage range covered by the robot under test within a preset time step and the corresponding cell grid interval. The theoretical travel distance represents the theoretical maximum travel distance of the robot under test within the current preset time step.

[0088] As a concrete example, the method for obtaining the theoretical travel distance of the robot under test at a preset time step can be expressed by the formula: ,in, This indicates the initial velocity of the robot under test. This represents the maximum acceleration of the robot under test. This indicates the preset time step, which is 1 second in this embodiment. This represents the theoretical travel distance of the robot under test at a preset time step.

[0089] The third step is to determine the cell grid range of the preset time step in the traffic speed grid map based on the theoretical travel distance under the preset time step, and obtain the speed information within the cell grid range and the minimum value of the highest travel speed of the robot under test as the simulated speed under the preset time step.

[0090] Specifically, for any given iteration, the theoretical travel distance at a preset time step can be used to determine the cell range it occupies in the traffic speed grid. ,in , Then, within the time range corresponding to the current iteration process, the traffic speed grid map is traversed. Within a certain range, the speed information and the minimum maximum speed of the robot under test are calculated. The simulated speed characterizes the simulation results of the robot under test in one iteration. This step reflects the physical logic that the entire journey is limited to the slowest section due to potential congestion.

[0091] The fourth step is to calculate the simulated acceleration of the robot from its initial speed to its simulated speed based on its initial speed and simulated speed. Based on the initial speed and simulated speed of the robot, the mileage traveled by the robot under test at the preset time step and the updated simulated speed are calculated.

[0092] Before each iteration begins, the initial velocity represents the speed of the simulated robot before it moves at a preset time step. The simulated velocity represents the target velocity of the simulated robot after it has moved for the preset time step. Therefore, based on the acceleration calculation formula, the acceleration required for the robot to reach the target velocity can be calculated. As a concrete example, the simulated acceleration required for the robot to reach the simulated velocity from the initial velocity can be expressed as: ,in This represents the simulated acceleration of the robot under test from its initial velocity to its simulated velocity during one iteration. This represents the simulated speed of the robot under test during one iteration. This represents the initial velocity of the robot to be detected during one iteration. This indicates the preset time step corresponding to one iteration process.

[0093] After the current iteration ends, the simulated velocity is updated according to the formula for calculating velocity from acceleration. After the current iteration ends, the updated simulated velocity is used as the initial velocity for the next iteration. As a concrete example, the formula for calculating the updated simulated velocity can be expressed as: , This represents the updated simulation velocity, which is the simulation velocity after the k-th iteration update, and can also be called the initial velocity of the (k+1)-th iteration. This represents the initial velocity of the k-th iteration. This represents the simulated acceleration during the k-th iteration. Indicates the preset time step.

[0094] Furthermore, the distance traveled by the robot under test within a preset time step can be calculated using the formula for calculating distance by acceleration, specifically: ,in This represents the distance traveled by the robot under test after the current iteration of the simulation process. This represents the initial velocity.

[0095] Step S302: Iterate the driving process of the robot to be tested on the current road segment with a preset time step. Stop when the driving distance is greater than or equal to the total mileage of the current road segment; obtain the cumulative value of the preset time step under all iteration steps to obtain the simulated driving time of the robot to be tested.

[0096] Specifically, the above iterative process continues until the cumulative mileage of the simulation process is greater than or equal to the total path length of the current road segment, at which point the iteration stops. The cumulative simulation time, which is also the simulated driving time of the robot under test, is the sum of the preset time steps across all iteration steps. Each iteration increments the iteration count by 1, and the cumulative mileage of the simulation process refers to the sum of the mileage traveled across all iteration steps.

[0097] The main purpose of this step is to complete the full kinematic simulation of the robot under test on the current road segment, and output the theoretical simulated travel time, providing a key theoretical time reference for subsequent deviation calculation. Since a single step cannot reflect the total travel time, this step uses a preset time step as the iteration unit, continuously accumulating the gradually increasing travel mileage output in step S301 to advance the robot's simulated travel process. When the accumulated travel mileage reaches or exceeds the total mileage of the current road segment, the iteration immediately terminates, and the time steps corresponding to all iteration steps are summed to finally obtain the simulated travel time of the robot under dynamic road condition constraints. This provides a core reference standard for comparing actual travel time and calculating the degree of deviation.

[0098] Step S303: Based on the proportion of cell grids in the traffic speed grid diagram where the speed information has been updated, determine the gain coefficient. Use the gain coefficient to weight the proportion of the difference between the simulated driving time length of the robot to be tested and the driving time window to obtain the degree of motion inference deviation of the robot to be tested.

[0099] The main purpose of this step is to calculate the statistically significant degree of motion deduction deviation by combining the reliability of road condition data, thus providing objective physical and logical evidence for cross-view robot identity determination. The raw difference between simulated and actual durations cannot reflect the coverage quality of road condition data. Therefore, this step first calculates the proportion of cell grids in the traffic speed grid map that have been updated by background traffic logs to determine the data reliability gain coefficient; then, it calculates the proportion of the difference between the simulated driving time length and the actual driving time window length, and uses the gain coefficient to weight and correct this proportion. The higher the data coverage, the larger the gain coefficient, and the stronger the penalty for deviation, ultimately obtaining the degree of motion deduction deviation that objectively reflects the degree of agreement between the robot's driving behavior and theoretical deduction.

[0100] As a concrete example, the ratio of the number of cell grids in the traffic speed grid diagram whose speed information has been updated by background traffic logs to the total number of cell grids in the two-dimensional grid diagram is used as a gain coefficient, reflecting the sample coverage based on actual driving data. Only cell grids where background traffic flow vehicles have a driving record will have their speed information updated; cell grids without background data coverage retain their initial speed information. The closer the gain coefficient is to 1, the more sufficient the background data of the simulation path, and the higher the reliability of the inference results; conversely, a lower value indicates sparse data, and the reference value of the inference results is limited.

[0101] Furthermore, the method for obtaining the degree of motion deduction deviation of the robot under test can be expressed by the formula:

[0102]

[0103] in, This indicates the degree of deviation in the motion projection of the robot under test. This indicates the duration of the travel time window for the robot being tested. This indicates the length of the simulated driving time of the robot under test. Indicates the gain coefficient. To prevent division by zero of extremely small positive numbers (in this embodiment, the value is taken as...) ).

[0104] A larger gain coefficient corresponds to higher reference data coverage, greater system confidence in the simulation results, and therefore a heavier penalty for time deviations. Conversely, a smaller gain coefficient corresponds to lower reference data coverage, thus allowing for greater tolerance of deviations. The degree of motion simulation deviation will be transmitted as numerical modal evidence to subsequent steps.

[0105] It should be noted that the gain coefficient This reflects the confidence level of the background data. Gain coefficient. The larger the value, the more comprehensive the background data coverage, and the better the simulation results. The more reliable it is, the better. Therefore, in calculating the degree of deviation... Multiply by time This is to increase the penalty for bias at high confidence levels: that is, if a large time bias still occurs when the data is sufficient, it indicates that the possibility of identity mismatch is extremely high; conversely, if the data is sparse, the bias is appropriately tolerated.

[0106] Step S400: Based on the angular deviation between the termination position and the initial position of the robot to be tested in the current road segment and the shooting position of the monitoring device, and combined with the degree of motion deduction deviation, the identity verification result of the robot to be tested in the current road segment is obtained, which is used for the coordinated scheduling of the robot to be tested.

[0107] The main purpose of this step is to integrate the dual evidence of geometric perspective bias and physical motion deduction bias to complete the cross-view identity verification of the robot under test, output a high-confidence judgment result, and provide a key identity benchmark for multi-robot collaborative scheduling.

[0108] In cross-view scenarios, relying solely on visual appearance features is prone to misjudgment due to changes in shooting angle, while motion data alone cannot explain feature drift caused by differences in perspective. Isolated analysis of these two aspects is insufficient to support reliable identity determination. Therefore, this step first calculates the angle deviation based on the relative relationship between the robot's initial and final positions on the current road segment and the shooting position of the monitoring equipment. This deviation directly characterizes the credibility of the visual features (the larger the angle deviation, the more severe the visual feature drift, and the lower the credibility). Then, it combines this with the degree of motion deduction deviation output from step S300 (the smaller the deviation, the higher the consistency between the robot's actual driving behavior and the theoretical deduction, and the stronger the physical matching degree). Finally, through weighted fusion and comprehensive adjudication of multi-source heterogeneous evidence, the identity verification result is output, ensuring that the judgment result has both geometric rationality and physical logic. This result can be directly used to guide the coordinated scheduling of robots, effectively avoiding problems such as path conflicts, task gaps, or resource contention caused by identity loss, thus achieving a closed loop in the scheduling logic.

[0109] In this regard, such as Figure 3 As shown, the method for obtaining the identity verification result of the robot to be detected in the current road segment can be implemented by steps S401 to S404.

[0110] Step S401: Based on the deviation between the observation angles corresponding to the travel direction of the robot under test at the end position and the travel direction at the initial position in the current road segment, the visual weight of the robot under test in the current road segment is obtained.

[0111] The main purpose of this step is to quantify the credibility weights of visual features, providing dynamic geometric constraints for subsequent multi-source evidence fusion. In cross-view scenarios, the observation angle deviation between the robot's initial and final positions directly leads to visual appearance feature drift; the larger the angle deviation, the lower the reference value of the visual matching result. Therefore, this step calculates the observation angle deviations of the robot under test at its initial and final positions on the current road segment, converting the angle differences into quantifiable visual weights. The smaller the angle deviation, the higher the visual weight, indicating stronger credibility of the visual features; conversely, the larger the angle deviation, the lower the weight, weakening the proportion of visual evidence in the fusion, thereby solving the problem of misjudgment easily caused by relying solely on visual features.

[0112] Specifically, the first step is to determine the lane tangent vector at the end position and the lane tangent vector at the initial position of the robot under test in the current road segment based on the end position and the initial position of the robot under test in the current road segment, respectively, and obtain the preset optical axis orientation vector of the detection system.

[0113] More specifically, the termination position and initial position of the robot under test in the current road segment can be considered as the positions of observation point A and observation point B. Target points A and B, located at observation points A and B respectively, are obtained on the road centerline. If the road centerline of the current segment is a broken line, the direction vector of the line connecting the two adjacent coordinate points of the target point is calculated; this is the lane tangential vector. If the road centerline of the current segment is a curve, the tangential direction of the target point is calculated using the derivative of the curve, yielding the tangential vector. The tangential vector corresponding to the termination position of the robot under test in the current road segment is the lane tangential vector at the termination position, and the tangential vector corresponding to the initial position of the robot under test in the current road segment is the lane tangential vector at the initial position.

[0114] The optical axis orientation vector represents the shooting direction of the roadside sensing unit (RSU) camera, which is predetermined through device extrinsic calibration, that is, it represents which direction the camera is looking in.

[0115] The second step is to calculate the angle between the lane tangential vector and the optical axis orientation vector as the observation angle corresponding to each lane tangential vector, and to use the absolute value of the cosine of the difference between the observation angles corresponding to the termination position and the initial position of the robot under test in the current road segment as the visual weight of the robot under test in the current road segment.

[0116] The smaller the change in viewing angle Δφ (the absolute value of the difference between the observation angle at the end position and the observation angle at the initial position of the current road segment), the closer the cosine value is to 1, indicating that the observation views before and after are basically consistent, and the difference in vehicle appearance coding mainly stems from its own characteristics, resulting in high credibility of visual evidence. Conversely, the larger Δφ is, the closer the cosine value is to 0, indicating that drastic changes in viewing angle cause significant drift in visual appearance coding, resulting in low credibility of visual evidence. Therefore, in the subsequent comprehensive matching score calculation for identity determination, visual similarity is prioritized when the change in viewing angle is small; when the change in viewing angle is large, physical evidence derived from motion inference is relied upon first, effectively avoiding the problem of failure of a single visual modality.

[0117] Thus, this visual weight is used to dynamically determine the reliability of the vehicle's visual features. The value is determined by the range of change in the viewing angle between the starting and ending observation points. The smaller the change in viewing angle, the higher the weight, and vice versa.

[0118] Step S402: Obtain the cosine similarity between the image feature vector of the image taken by the robot at the end position of the current road segment and the image feature vector of the image taken at the initial position, as a visual similarity index of the robot to be detected in the current road segment.

[0119] The main purpose of this step is to extract visual matching quantification indicators to provide intuitive visual feature basis for identity verification. Visual similarity is a fundamental dimension for determining whether robots are consistent, but it needs to be quantified through standardized calculations. Therefore, this step obtains the feature vectors of images taken by the robot to be detected at its initial and final positions on the current road segment. By calculating the cosine similarity between the two feature vectors, a visual similarity index is obtained. The closer the index value is to 1, the higher the matching degree of the appearance features of the preceding and following images; conversely, the lower the matching degree, the lower the matching degree, providing clear visual dimension quantification data for subsequent multi-source evidence fusion.

[0120] This visual similarity index avoids the subjectivity of human eyes or simple feature comparison by calculating the cosine similarity of high-dimensional image feature vectors, and accurately measures the degree of appearance matching of the robot at two observation points (the end position and the initial position of the current road segment).

[0121] Furthermore, in the identity determination process, this indicator serves as a direct basis for visual judgment. When the indicator is close to 1, it indicates that the appearance of the vehicles observed before and after are highly consistent, supporting the determination that they are the same robot; when the indicator is close to 0, it indicates that the appearance differences are extremely large, and visual evidence does not support the same determination.

[0122] It should be noted that the image feature vector is a high-dimensional floating-point vector generated by the roadside perception unit through a feature extraction network (such as a ResNet network, or a pre-trained large visual model, such as the Image Encoder of the CLIP model or DINOv2) encoding the captured image. More specifically, the high-dimensional floating-point vector generated by encoding the image captured by the robot under test at the end position of the current road segment through the feature extraction network is the image feature vector corresponding to the end position. The high-dimensional floating-point vector generated by encoding the image captured by the robot under test at the initial position of the current road segment through the feature extraction network is the image feature vector corresponding to the initial position.

[0123] Step S403: Using the visual weights, the visual similarity index and motion deduction deviation of the robot to be tested are fused to obtain the verification confidence of the robot to be tested in the current road segment.

[0124] The main purpose of this step is to integrate visual and physical evidence to generate a comprehensive verification confidence score, addressing the problem that a single evidence dimension cannot handle complex scenarios. Visual similarity indicators are easily affected by perspective, and the degree of motion deduction deviation only reflects the rationality of physical behavior; neither can support a high confidence score in isolation. Therefore, this step utilizes the visual weights output from step S401 to dynamically allocate the proportion of visual similarity indicators and motion deduction deviation in the fusion calculation. When the visual weight is high, the focus is on referencing the visual matching degree; when the weight is low, the focus is on relying on the physical deduction results. Finally, through weighted fusion, a verification confidence score that balances geometric rationality and physical logic is obtained, providing a core quantitative basis for identity determination.

[0125] Specifically, the verification confidence of the robot under test in the current road segment is obtained by weighted summation of the negative correlation coefficient of the degree of motion inference deviation of the robot under test and the visual similarity index; wherein, the weight corresponding to the negative correlation coefficient of the degree of motion inference deviation is the negative correlation coefficient of the visual weight, and the weight corresponding to the visual similarity index is the visual weight.

[0126] More specifically, the method for obtaining the verification confidence level of the robot to be detected in the current road segment can be expressed by the formula:

[0127]

[0128] in, This indicates the verification confidence level of the robot to be tested in the current road segment. Indicates visual weight, The negative correlation coefficient representing visual weight. Indicators of visual similarity The negative correlation coefficient, representing the degree of deviation in the motion projection of the robot under test, is... , This indicates the degree of deviation in the motion projection of the robot under test.

[0129] Weights corresponding to visual similarity index The larger the value, the smaller the change in the observation perspective (e.g., always viewing the car from the front). In this case, the visual features are most reliable, and the weight corresponding to the degree of motion inference deviation is... The smaller the value, the more weight is given to visual similarity; that is, when the same vehicle is viewed from the same angle, the matching degree of appearance features has the greatest reference value. Conversely, the weight of the visual similarity index is lower. The smaller the value, the greater the change in perspective (e.g., from looking at the front of the car to looking at the side). In this case, the visual features become unreliable due to distortion, and the weight corresponding to the degree of motion inference deviation is adjusted accordingly. The larger the value, the more likely the motion deduction results will be accepted, indicating that the visual feature drift is severe at this time, and the matching degree of the physical trajectory is more reliable.

[0130] Step S404: Based on the verification confidence level of the robot to be detected in the current road segment and the preset confidence threshold, obtain the identity verification result of the robot to be detected in the current road segment.

[0131] The main purpose of this step is to output a high-confidence authentication result, providing a direct decision-making basis for multi-robot coordinated scheduling. The verification confidence level must pass a preset threshold before it can be converted into an executable identity conclusion. Therefore, this step compares the verification confidence level obtained in step S403 with the system's preset confidence threshold. If the confidence level is higher than the threshold, it is determined that the observed robots are the same; if it is lower than the threshold, it is determined that they are different robots. This completes the closed loop of cross-view robot authentication, effectively avoiding scheduling conflicts caused by identity loss.

[0132] Specifically, before threshold determination, the validation confidence level is normalized. The normalization method can be minimization normalization. It should be understood that the maximum and minimum values ​​of this normalization process are validation confidence levels statistically derived from a large amount of historical data. The implementer needs to calculate this according to the specific implementation scenario, which will not be elaborated here. Furthermore, in this embodiment, the confidence threshold is set to 0.85.

[0133] More specifically, when the verification confidence level of the robot to be tested in the current road segment is greater than or equal to the preset confidence threshold, the robot to be tested passes the authentication; when the verification confidence level of the robot to be tested in the current road segment is less than the preset confidence threshold, the robot to be tested fails the authentication.

[0134] After verifying the identity of the robot to be tested, if the verification is successful, the robot can inherit the original task execution context and right-of-way without needing to be re-authenticated, and can directly execute the original scheduling instructions. If the verification fails, the system will forcibly release the originally associated mutex lock resources, treat it as a newly registered entity, and require it to reapply for right-of-way and plan a scheduling path.

[0135] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for coordinated scheduling of multiple robots for vehicle detection based on a large model, characterized in that, The method includes: Acquire the travel time window and travel speed data of the robot to be detected in the current road segment, and construct the background traffic log by recording the vehicle travel records in the current road segment within the travel time window; Based on the current road segment distribution and the distribution of the robot's travel time window, combined with the spatiotemporal characteristics of vehicle travel records in the background traffic log, a traffic speed grid map of the current road segment is constructed. Specifically, this includes: uniformly dividing the total mileage of the current road segment into two dimensions, using the mileage length of the current road segment as the first dimension and the total travel time window as the second dimension; and uniformly dividing the total travel time window into two dimensions to construct a two-dimensional grid map. The travel time window is the time interval between the robot's entry and exit from the current road segment. Based on the travel speed and travel time contained in the vehicle travel records in the background traffic log, and combined with the speed limit information of the current road segment, the speed information of the unit grids in the two-dimensional grid map is updated to obtain the traffic speed grid map of the current road segment. Based on the driving speed data of the robot under test and the traffic speed grid map, the driving time distribution of the robot under test in the current road segment is simulated to obtain the degree of deviation in the motion inference of the robot under test. Based on the angular deviations between the robot's termination and initial positions on the current road segment and the camera's shooting position, and combined with the degree of motion deduction deviation, the robot's identity verification result on the current road segment is obtained, which is used for coordinated scheduling of the robot.

2. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 1, characterized in that, The driving speed data includes the robot's maximum driving speed, maximum acceleration, and maximum deceleration. Vehicle driving records include the driving time, location, and speed of each authenticated device.

3. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 1, characterized in that, The process involves updating the speed information of each cell in the two-dimensional grid map based on the vehicle speed and travel time recorded in the background traffic log, combined with the speed limit information for the current road segment, to obtain the current road segment's traffic speed grid map. Specifically, this includes: The maximum speed limit for the current road segment is used as the speed information of the cell grid in the two-dimensional grid map. The cell grid corresponding to the vehicle's driving record is determined by the driving time and driving position of the vehicle in the current road segment. The speed information of the cell grid corresponding to the vehicle's driving record is updated by the minimum value between the driving speed of the vehicle's driving record and the maximum speed limit for the current road segment, thus obtaining the traffic speed grid map of the current road segment.

4. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 3, characterized in that, The step involves simulating the travel time distribution of the robot under test in the current road segment based on the robot's travel speed data and the traffic speed grid map, thereby obtaining the degree of deviation in the robot's motion projection. Specifically, this includes: Initialize the iteration steps and travel distance of the robot under test in the current road segment, and obtain the preset time step and the simulated speed of the robot under test; At each preset time step, the simulated speed of the robot under test is updated based on the driving speed data of the robot under test and the speed information of the unit grid in the current road segment's traffic speed grid diagram, and the driving distance of the robot under test at each preset time step is calculated. The robot under test is iterated over the current road segment at a preset time step. The iteration stops when the travel distance is greater than or equal to the total mileage of the current road segment. The cumulative value of the preset time step under all iteration steps is obtained to obtain the simulated travel time of the robot under test. Based on the proportion of cell grids in the traffic speed grid diagram where the speed information is updated, a gain coefficient is determined. The degree of motion inference deviation of the robot under test is obtained by weighting the proportion of the difference between the simulated driving time length of the robot under test and the driving time window using the gain coefficient.

5. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 4, characterized in that, The step of updating the simulated speed of the robot under test based on its driving speed data and the speed information of the unit grid in the current road segment's traffic speed grid map, and calculating the robot's travel distance at each preset time step, specifically includes: Obtain the initial speed of the robot to be tested as it travels to the current road segment; For any iteration, based on the initial velocity and maximum acceleration of the robot under test, calculate the theoretical travel distance of the robot under test at a preset time step; The theoretical travel distance under the preset time step is used to determine the cell grid range of the preset time step in the traffic speed grid diagram. The speed information within the cell grid range and the minimum value of the highest travel speed of the robot under test are obtained as the simulated speed under the preset time step. Based on the initial speed and simulated speed of the robot under test, the simulated acceleration of the robot under test from the initial speed to the simulated speed is calculated. Based on the initial speed and simulated speed of the robot under test, the travel distance of the robot under test at the preset time step and the updated simulated speed are calculated.

6. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 1, characterized in that, The method involves determining the identity verification result of the robot under test in the current road segment based on the angular deviations between its initial and final positions on the current road segment and the camera position of the monitoring device, combined with the degree of motion deduction deviation. Specifically, this includes: Based on the deviation between the observation angles corresponding to the travel direction of the robot under test at the end position and the travel direction at the initial position in the current road segment, the visual weight of the robot under test in the current road segment is obtained. The cosine similarity between the image feature vector of the image taken by the robot at the end position of the current road segment and the image feature vector of the image taken at the initial position is obtained and used as a visual similarity index of the robot in the current road segment. Using the aforementioned visual weights, the visual similarity index and motion inference deviation of the robot to be tested are fused to obtain the verification confidence of the robot to be tested in the current road segment; Based on the verification confidence level of the robot to be tested in the current road segment and the preset confidence threshold, obtain the identity verification result of the robot to be tested in the current road segment.

7. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 6, characterized in that, The step of obtaining the visual weight of the robot under test in the current road segment based on the deviation between the observation angles corresponding to the travel direction at the end position and the travel direction at the initial position of the robot under test in the current road segment specifically includes: Based on the termination position and initial position of the robot to be detected in the current road segment, the lane tangential vector at the termination position and the lane tangential vector at the initial position of the robot to be detected in the current road segment are determined respectively, and the preset optical axis orientation vector of the detection system is obtained. The angle between each lane tangential vector and the optical axis orientation vector is calculated as the observation angle corresponding to each lane tangential vector. The absolute value of the cosine of the absolute value of the difference between the observation angles corresponding to the termination position and the initial position of the robot under test in the current road segment is used as the visual weight of the robot under test in the current road segment.

8. The method for coordinated scheduling of multiple robots for vehicle detection based on a large model according to claim 6, characterized in that, The process of using the visual weights to fuse the visual similarity index and motion inference deviation of the robot to be detected to obtain the verification confidence level of the robot in the current road segment specifically includes: The verification confidence level of the robot under test in the current road segment is obtained by weighting and summing the negative correlation coefficient of the degree of deviation in motion inference of the robot under test and the visual similarity index. Among them, the negative correlation coefficient of the degree of deviation in motion deduction corresponds to the negative correlation coefficient of visual weight, and the weight corresponding to the visual similarity index is the visual weight.

9. A method for coordinated scheduling of multiple robots for vehicle detection based on a large model, as described in claim 6, is characterized in that... The step of obtaining the identity verification result of the robot to be detected in the current road segment based on the verification confidence level of the robot to be detected in the current road segment and a preset confidence threshold specifically includes: When the verification confidence level of the robot to be tested in the current road segment is greater than or equal to the preset confidence threshold, the robot to be tested passes the identity verification. When the verification confidence level of the robot to be tested in the current road segment is less than the preset confidence threshold, the robot to be tested fails the authentication.

Images