A multi-agv intersection dynamic priority passing coordination control system combined with visual detection

Abstract

The application discloses a kind of dynamic priority passing coordination control system of multi-AGV intersection combined with visual detection, it is related to AGV coordination control technical field, including, obtain visual image data, identify the real-time motion state of each AGV on real-time occupancy grid map;Based on AVG real-time motion state set, predict intersection occupancy time window, according to time window overlap, identify conflict AGV group, calculate dynamic passing priority in conflict AGV group, form dynamic priority queue;According to dynamic priority queue, extract high-priority AGV passing track, generate dynamic obstacle constraint set, execute local MPC avoidance planning to low-priority AGV, obtain low-priority AGV avoidance track;AGV track is rechecked twice, according to the rechecking result, track adjustment is carried out, and conflict-free track set is output;Based on conflict-free track set, generate passing control instruction and issue AGV execution, collect execution feedback and update intersection state.

Description

Technical Field

[0001] This invention relates to the field of AGV coordination control technology, and in particular to a dynamic priority passage coordination control system for multi-AGV intersections that combines visual detection. Background Technology

[0002] With the development of intelligent manufacturing, smart warehousing and flexible logistics systems, the collaborative scheduling technology of multiple Automated Guided Vehicles (AGVs) has gradually become an important research direction in the field of intelligent logistics. Existing AGV systems have gradually evolved from traditional fixed-path navigation to an autonomous collaborative operation mode that integrates LiDAR, visual perception and multi-sensor positioning. Especially in warehousing and sorting, unmanned workshops and automated delivery scenarios, a large number of AGVs need to frequently pass through shared spaces such as intersections, merging areas and narrow channels. Therefore, the problem of traffic coordination and control in the intersection area of ​​multiple AGVs has become a key factor affecting the system's operating efficiency and safety.

[0003] However, existing technologies still have the following shortcomings. First, most existing multi-AGV intersection coordination schemes rely on fixed priorities or static scheduling rules, making it difficult to dynamically adjust the passage order according to the real-time operating status of AGVs. When the traffic density at intersections increases, local congestion, frequent stops, and even system oscillations caused by repeated priority competition can easily occur. Second, although some schemes introduce visual detection or LiDAR perception, they are usually only used for target localization or obstacle detection. There is a lack of a dynamic coordination mechanism that deeply integrates visual detection results with the spatiotemporal occupancy status of intersections, resulting in insufficient accuracy in AGV motion state recognition. Especially in high-speed intersections or multi-vehicle concurrent scenarios, it is difficult to accurately predict conflict relationships and passage time windows. In addition, existing obstacle avoidance methods mostly adopt single-stage path planning or simple yielding strategies, lacking a collaborative optimization mechanism between high-priority trajectory protection and low-priority dynamic avoidance. When local obstacle avoidance fails, repeated replanning, trajectory jitter, and decreased passage efficiency are likely to occur. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a dynamic priority passage coordination control system for multi-AGV intersections that combines visual detection, solving the problems of insufficient dynamic priority adjustment, low real-time conflict prediction accuracy, and easy oscillation of multi-vehicle obstacle avoidance trajectories in existing multi-AGV intersection control.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a dynamic priority traffic coordination control system for multi-AGV intersections that combines visual detection, comprising, The status recognition module constructs a visual coordinate system for the intersection and generates a transformation matrix, acquires visual image data, identifies the real-time motion status of each AGV on the real-time occupied grid map, and outputs a set of AGV real-time motion statuses. The priority sorting module predicts the intersection occupancy time window based on the real-time motion state set of AVGs, identifies conflicting AGV groups based on the overlap of time windows, calculates the dynamic passage priority within the conflicting AGV groups, and forms a dynamic priority queue. The trajectory planning module extracts the high-priority AGV passage trajectory from the dynamic priority queue, generates a dynamic obstacle constraint set, performs local MPC avoidance planning on the low-priority AGV, and obtains the low-priority AGV avoidance trajectory. The secondary verification module performs a secondary verification of the AGV trajectory, adjusts the trajectory based on the verification results, and outputs a set of conflict-free trajectories. The execution feedback module generates traffic control commands based on a set of conflict-free trajectories and issues them to the AGV for execution, while collecting execution feedback to update the intersection status.

[0007] As a preferred embodiment of the multi-AGV intersection dynamic priority traffic coordination control system combining visual detection described in this invention, the steps for constructing the intersection visual coordinate system and generating the transformation matrix are as follows: Establish a plane coordinate system for the intersection and match the camera image coordinates with the plane coordinates of the intersection to obtain the homography transformation matrix.

[0008] As a preferred embodiment of the multi-AGV intersection dynamic priority passage coordination control system combining visual detection described in this invention, the steps of identifying the real-time movement status of each AGV on the real-time occupied grid map and outputting the AGV real-time movement status set are as follows: The pre-trained AGV target recognition model is used to detect AGV targets in the current visual image data, and the visual detection box of each AGV in the image is obtained. The Shi-Tomasi corner detection algorithm is used to extract the set of stable feature points of AGV within the visual detection box. The multi-scale pyramid LK optical flow algorithm is used to construct a multi-scale image pyramid for the current frame and the next frame, and the set of stable feature points is projected onto the highest layer of the pyramid image to solve the optical flow displacement in the highest layer image. The optical flow displacement obtained from each layer is amplified and used as the initial displacement of the next layer. Then, local corrections are made to obtain the final displacement of the next layer. When iterating layer by layer to the original image layer, the final optical flow displacement of the stable feature points is obtained, and the predicted position of the stable feature points in the next frame is output. Based on the current frame position and the predicted position in the next frame of the stable feature point, a six-parameter affine model is fitted, and the affine model parameters are solved by the least squares method. Input the center point of the AGV visual detection box in the current frame into the six-parameter affine model to obtain the corrected center point of the next frame, and convert the corrected center point of the next frame into the intersection plane coordinate position through the homography transformation matrix. Calculate the visual velocity and visual heading angle based on the corrected changes in planar coordinates; The real-time motion status of each AGV is generated based on its AGV number, coordinate position, visual speed, and visual heading angle.

[0009] As a preferred embodiment of the multi-AGV intersection dynamic priority traffic coordination control system combining visual detection described in this invention, the steps for predicting the intersection occupancy time window based on the AVG real-time motion state set are as follows: Extract all potential conflict zones from the real-time occupancy raster map at intersections. The conflict zones include the central area of ​​the intersection, the left-turn merging area, the right-turn merging area, the straight-ahead merging area, and the loading / unloading point entrance area, forming a set of conflict zones. ; If the first The AGV route passes through the conflict area. Based on the AVG real-time motion state set, the first The AGV calculates its arrival time in the conflict zone. The estimated time, and based on the AGV's position in the conflict zone. The movement type within the area calculates the time taken to leave the conflict zone, forming the first... AGVs in the conflict area The time windows are defined, and the final output is a set of time windows for all AVGs.

[0010] As a preferred embodiment of the multi-AGV intersection dynamic priority passage coordination control system combining visual detection described in this invention, the steps of identifying conflicting AGV groups based on overlapping time windows, calculating the dynamic passage priority within each conflicting AGV group, and forming a dynamic priority queue are as follows. Based on the time window set of all AVGs, for any two AGVs in the same conflict zone Overlap detection is performed within a time window. If an overlap occurs, the distance between the two AGVs at that time is calculated, and it is determined whether the distance between the two AGVs at the overlap time is less than the sum of the dynamic safety radii of the two AGVs. If it is less, the two AGVs are determined to be in the conflict zone. There are potential conflicts within the memory; conflict zone A set of AGVs that have potential conflicts within the memory constitutes a conflicting AGV group; Calculate a dynamic priority score for all AGVs in the conflicting AGV group based on their arrival time, task urgency, remaining path length, current speed, and cargo status. The dynamic priority scores of all AGVs in the conflicting AGV group are sorted in descending order to form a dynamic priority queue.

[0011] As a preferred embodiment of the multi-AGV intersection dynamic priority passage coordination control system combining visual detection described in this invention, the steps of extracting the passage trajectory of high-priority AGVs from the dynamic priority queue and generating a dynamic obstacle constraint set are as follows: Get the future trajectory of the highest priority AGV in the dynamic priority queue. ,Will It is expanded into a dynamic restricted area, forming a dynamic obstacle constraint set.

[0012] As a preferred embodiment of the multi-AGV intersection dynamic priority traffic coordination control system combining visual detection described in this invention, the steps for performing local MPC avoidance planning on low-priority AGVs to obtain their avoidance trajectories are as follows: Define the state for all AGVs in the dynamic priority queue except for the highest priority AGV. and control quantity ; Based on the dynamic obstacle constraint set, optimization objectives and control constraints are constructed, and local trajectory optimization is performed using improved MPC to obtain the low-priority AGV avoidance trajectory.

[0013] As a preferred embodiment of the multi-AGV intersection dynamic priority traffic coordination control system combining visual detection described in this invention, the steps of performing a secondary verification of the AGV trajectory, adjusting the trajectory based on the verification result, and outputting a set of conflict-free trajectories are as follows. Obtain the future trajectory of the high-priority AGV After avoiding the low-priority AGV trajectory, the trajectories of any two AGVs in the same conflict area are checked a second time at any prediction time. The second check includes point conflict check and edge conflict check. For any two AGVs within the same conflict zone, at any predicted time... If the edge conflict verification expression is satisfied, it means that an edge conflict still exists; If there are no conflicts during the second review, the trajectory of each AGV in the conflict area is output to obtain a set of conflict-free trajectories. If conflicts still exist after the second review, a set of candidate avoidance actions is generated for the low-priority AGVs among the conflicting AGVs, and a corresponding predicted trajectory is generated for each candidate action. The comprehensive cost is calculated based on the trajectory, and the candidate action that satisfies the control constraints and has the lowest comprehensive cost is selected as the final avoidance action. The corresponding avoidance trajectory is then output to obtain the adjusted set of conflict-free trajectories.

[0014] As a preferred embodiment of the multi-AGV intersection dynamic priority traffic coordination control system combining visual detection described in this invention, the steps of generating traffic control commands based on a set of conflict-free trajectories and issuing them to the AGVs for execution are as follows: After obtaining the set of conflict-free trajectories, an executable set of AGV control instructions is generated and issued for execution based on the behavior type of each AGV in the trajectory.

[0015] As a preferred embodiment of the multi-AGV intersection dynamic priority traffic coordination control system combining visual detection described in this invention, the steps for collecting, executing feedback, and updating the intersection status are as follows: After the AGV executes the control command, the visual inspection camera continues to collect images of the intersection. When the highest priority AGV completely leaves the conflict area, the AGV is removed from the dynamic priority queue, and the dynamic priority score calculation and trajectory planning are re-executed for the remaining AGVs until all AGVs in the conflict group have passed through the conflict area.

[0016] The beneficial effects of this invention are as follows: By fitting a six-parameter affine model and combining it with a homography transformation matrix to convert image coordinates into actual planar coordinates, robust and accurate perception of the microscopic motion state of each AGV is achieved under complex visual conditions, avoiding subsequent misjudgments or collisions caused by perception errors; by calculating priorities based on the urgency of the task and the remaining path length, task-driven dynamic priority allocation is realized, avoiding inefficient waiting or delays in important tasks; by transforming the trajectory of high-priority AGVs into dynamic obstacle constraints, local MPC avoidance planning is performed on low-priority AGVs to minimize the passage delay of low-priority AGVs, thereby improving the overall traffic efficiency of the intersection; through secondary verification and generation of candidate avoidance action sets and selection based on a multi-objective cost function, a safety closed loop of "planning-verification-correction" is formed. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0018] Figure 1This is a structural diagram of a multi-AGV intersection dynamic priority traffic coordination control system that combines visual detection, as described in Example 1.

[0019] Figure 2 This is a flowchart of a multi-AGV intersection dynamic priority traffic coordination control system that combines visual detection, as described in Example 1.

[0020] Figure 3 This is a flowchart of the priority sorting and trajectory planning module in Example 1. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1, referring to Figure 1 , Figure 2 and Figure 3 This is the first embodiment of the present invention, which provides a dynamic priority traffic coordination and control system for multi-AGV intersections that combines visual detection, including the following steps: S1, State Recognition Module: Constructs a visual coordinate system for the intersection and generates a transformation matrix, acquires visual image data, identifies the real-time motion state of each AGV on the real-time occupied grid map, and outputs a set of real-time AGV motion states. Specifically, an intersection plane coordinate system is established, and the camera image coordinates are matched with the intersection plane coordinates to obtain the homography transformation matrix; Furthermore, a pre-trained AGV target recognition model is used to perform AGV target detection on the current visual image data to obtain the visual detection box of each AGV in the image. Within the visual detection box, the Shi-Tomasi corner detection algorithm is used to extract the set of stable feature points of the AGV. The multi-scale pyramid LK optical flow algorithm is used to construct a multi-scale image pyramid for the current frame and the next frame, and the set of stable feature points is projected onto the highest layer of the pyramid image to solve the optical flow displacement in the highest layer image. The expression for a multi-scale image pyramid is: ; ; in, This represents the original image of the current frame. Indicates the current frame number Layer image, number The layer image is obtained by downsampling the image of the previous layer. This represents the highest-level low-resolution image. Indicates the number of pyramid levels. Indicates the original image of the next frame. Indicates the current frame. Indicates the next frame; The expression for solving the optical flow displacement in the highest layer image is: ; in, Indicates the first The first AGV The stable feature point at the th... Optical flow displacement in layered pyramid images Indicates the first The size of the neighborhood pixel window of a stable feature point in this invention is taken as a value. , Represents a neighborhood pixel window Pixel positions within, and These represent horizontal displacement and vertical displacement, respectively. The optical flow displacement obtained from each layer is amplified and used as the initial displacement for the next layer. Then, local corrections are performed to obtain the final displacement of the next layer, expressed as: ; ; ; in, Indicates the first The initial optical flow displacement of the layer, Indicates the first The final optical flow displacement of the layer, This represents the scale restitution coefficient between pyramid levels. Indicates the first The optical flow correction amount of the layer, and They represent the first The initial horizontal and initial vertical displacements of the layer and They represent the first Horizontal and vertical corrections for the layer Indicates the first The final optical flow displacement of the layer; When iterating layer by layer to the original image layer, the final optical flow displacement of the stable feature points is obtained, and the predicted position of the stable feature points in the next frame is output, as expressed by: ; ; in, This represents the final optical flow shift in the original image layer. and These represent the final displacement in the horizontal direction and the final displacement in the vertical direction, respectively. Indicates the predicted feature point position in the next frame. Indicates the position of the feature point in the current frame; Based on the current frame position and the predicted position in the next frame of the stable feature point, a six-parameter affine model is fitted. The parameters of the affine model are solved using the least squares method. The expression of the six-parameter affine model is: ; in, and These represent the horizontal and vertical positions of the feature points predicted by the affine model, respectively. , , , , and The parameters representing the affine model, and This indicates the horizontal and vertical positions of the feature points in the current frame; Input the center point of the AGV visual detection box in the current frame into the six-parameter affine model to obtain the corrected center point of the next frame, and convert the corrected center point of the next frame into the intersection plane coordinate position through the homography transformation matrix. Calculate the visual velocity and visual heading angle based on the corrected changes in planar coordinates; Based on the AGV number, coordinate position, visual speed, and visual heading angle, the real-time motion state of each AGV is generated, expressed as: ; ; in, Indicates the first Real-time movement status of the AGV. Indicates the first The serial number of the AGV. Indicates the first The horizontal coordinates of the AGV at the intersection. Indicates the first The visual speed of an AGV Indicates the first The visual heading angle of an AGV and Indicates the first The length and width of each AGV are obtained from the system database. Indicates the first The dynamic safety radius of an AGV. Indicates the system control response time. Indicates the first The maximum allowable deceleration of an AGV.

[0025] This step differs from directly calculating the AGV's motion state using the center difference of the detection frame. Instead, it extracts stable feature points within the AGV's detection frame, obtains the inter-frame displacement of these feature points through multi-scale pyramid LK optical flow, and then uses a six-parameter affine model to fit the overall motion of the optical flow points. Finally, it calculates the visual velocity and visual heading angle from the affine-corrected AGV center position, thereby improving the accuracy of motion state estimation for the AGV under conditions of rapid turning, partial occlusion, and detection frame jitter.

[0026] S2, Priority sorting module, predicts the intersection occupancy time window based on the real-time motion state set of AVG, identifies conflicting AGV groups according to the overlap of time windows, calculates the dynamic passage priority within the conflicting AGV groups, and forms a dynamic priority queue. Specifically, all potential conflict zones are extracted from the real-time occupancy grid map at intersections. The conflict zones include the central area of ​​the intersection, the left-turn merging area, the right-turn merging area, the straight-ahead merging area, and the loading / unloading point entrance area, forming a set of conflict zones. ; If the first The AGV route passes through the conflict area. Based on the AVG real-time motion state set, the first The AGV calculates its arrival time in the conflict zone. The estimated time, and based on the AGV's position in the conflict zone. The movement type within the area calculates the time taken to leave the conflict zone, forming the first... AGVs in the conflict area The time windows are defined, and the final output is a set of time windows for all AVGs. For the The AGV calculates its arrival time in the conflict zone. The estimated time is expressed as: ; in, Indicates the first A vehicle AGV arrived at the conflict zone. The estimated time, Indicates the current moment. Indicates the first The AGV is currently located in the conflict area. Path distance, Indicates the current speed. This represents the minimum speed threshold to prevent division by zero. If the first AGV in the conflict zone The movement type within the area is straight ahead, and the expression for calculating the time to leave the conflict zone is: ; in, Indicates the first AGV leaves the conflict area The estimated time, Indicates conflict area The length along the direction of AGV travel; If the first AGV in the conflict zone The movement type within the area is turning, and the expression for calculating the time to leave the conflict zone is: ; in, Indicates the turning angle. Indicates the turning radius. The turning speed of the AGV is determined by historical data; Furthermore, based on the time window set of all AVGs, for any two AGVs in the same conflict area... Overlap detection is performed within a time window. If an overlap occurs, the distance between the two AGVs at that time is calculated, and it is determined whether the distance between the two AGVs at the overlap time is less than the sum of the dynamic safety radii of the two AGVs. If it is less, the two AGVs are determined to be in the conflict zone. There are potential conflicts within the memory; conflict zone A set of AGVs that have potential conflicts within the memory constitutes a conflicting AGV group; A dynamic priority score is calculated for all AGVs in the conflicting AGV group based on their arrival time, task urgency, remaining path length, current speed, and cargo status. The expression is: ; in, Indicates the first Dynamic priority scoring for each AGV. Indicates the first The AGV is expected to reach the conflict zone. The remaining time, Represents a very small positive number. Indicates the first The task urgency of each AGV is predefined based on the current task type of the AGV. Indicates the first The loading status of an AGV is set to 1 when loaded and 0 when unloaded. Indicates the first The remaining path length of the AGV from the mission endpoint. Indicates the first The congestion contribution of the AGV, if the first If a waiting AGV will block subsequent AGVs, the urgency level of the task of all subsequently blocked AGVs is summed. , , , and The weighting coefficients are determined through cross-validation. The dynamic priority scores of all AGVs in the conflicting AGV group are sorted in descending order to form a dynamic priority queue.

[0027] S3, the trajectory planning module extracts the high-priority AGV passage trajectory based on the dynamic priority queue, generates a dynamic obstacle constraint set, performs local MPC avoidance planning on the low-priority AGV, and obtains the low-priority AGV avoidance trajectory. Specifically, obtain the future trajectory of the highest priority AGV in the dynamic priority queue. ,Will This is expanded into a dynamic restricted area, forming a dynamic set of obstacle constraints. Furthermore, for the AGVs in the dynamic priority queue other than the highest priority AGV, define the state. and control quantity ; Based on the dynamic obstacle constraint set, an optimization objective and control constraints are constructed, and an improved MPC is used for local trajectory optimization to obtain the low-priority AGV avoidance trajectory. The improved MPC optimization objective expression is as follows: ; ; in, Indicates the first The MPC cost function of an AGV. Indicates the length of the prediction time domain. Indicates a time index. Indicates the first AGV in the future The predicted state at each time step Indicates the first AGV in the future The target state corresponding to the original reference path at each time step. Indicates the first AGV in the future Control quantity per time step Indicates the first AGV in the future Control quantity per time step This indicates the dynamic obstacle avoidance penalty between low-priority AGVs and high-priority AGVs. Indicates the high-priority AGV number. and They represent the first AGV and the first The dynamic safety radius of an AGV. and They represent the first AGV and the first AGV in the future Predicted location at each time step; The control constraint expression is: ; ; in, Indicates the first The next predicted state of the AGV. This represents the AGC motion state transition function. Indicates the first The current linear speed of the AGV. This indicates the maximum permissible linear speed of the AGV. Indicates the first The current angular velocity of the AGV. This indicates the maximum permissible angular velocity of the AGV. This represents the rate of change during the continuous control cycle. This indicates the maximum allowable acceleration of the AGV. Indicates the control cycle.

[0028] S4, Secondary Verification Module: Performs secondary verification on the AGV trajectory, adjusts the trajectory based on the verification results, and outputs a set of conflict-free trajectories. Specifically, the future trajectory of the high-priority AGV is obtained. After avoiding the low-priority AGV trajectory, the trajectories of any two AGVs in the same conflict area are checked a second time at any prediction time. The second check includes point conflict check and edge conflict check. The point conflict verification expression is: ; in, and They represent the first AGV and the first AGV at the predicted time Location coordinates, and Indicates the first AGV and the first AGV at the predicted time The dynamic safety radius; For any two AGVs within the same conflict zone, at any predicted time... If the point conflict verification expression is satisfied, it means that a point conflict still exists; The edge conflict verification expression is: ; in, and They represent the first AGV path segment and the first Interpolation coefficients on the AGV path segment. Indicates the range of values ​​for the interpolation coefficients. and They represent the first AGV and the first The position increment of an AGV between adjacent time points; For any two AGVs within the same conflict zone, at any predicted time... If the edge conflict verification expression is satisfied, it means that an edge conflict still exists; If there are no conflicts during the second review, the trajectory of each AGV in the conflict area is output to obtain a set of conflict-free trajectories. If there are still conflicts after the second review, a set of candidate avoidance actions is generated for the low-priority AGVs in the conflicting AGVs, and a corresponding predicted trajectory is generated for each candidate action. The comprehensive cost is calculated based on the trajectory, and the candidate action that satisfies the control constraints and has the lowest comprehensive cost is selected as the final avoidance action. The corresponding avoidance trajectory is then output to obtain the adjusted set of conflict-free trajectories. The expression for the candidate avoidance action set is: ; in, Indicates the first A set of candidate obstacle avoidance maneuvers for an AGV. This indicates that you should slow down and proceed. This indicates that the system has entered a waiting state. Indicates lateral detour. This indicates a return to the safe waiting area; The expression for calculating the comprehensive cost based on the trajectory is as follows: ; in, Indicates the first The AGV is performing the first The combined cost of each candidate action Indicates the cost of delay, by the first The AGV is performing the first The difference between the estimated time to leave the conflict zone after each candidate action and the original time to leave the conflict zone is obtained. This represents the cost of deviating from the path, and its acquisition method is consistent with the first calculation term in the improved MPC optimization objective. and The weighting coefficients representing delay costs and deviation costs are determined through cross-validation.

[0029] S5, Execution Feedback Module: Generates traffic control instructions based on the set of conflict-free trajectories and issues them to the AGV for execution, and collects execution feedback to update the intersection status; Specifically, after obtaining the set of conflict-free trajectories, an executable set of AGV control instructions is generated and issued for execution based on the behavior type of each AGV in the trajectory; Furthermore, after the AGV executes the control command, the visual inspection camera continues to collect images of the intersection. When the highest priority AGV completely leaves the conflict area, the AGV is removed from the dynamic priority queue, and the dynamic priority score calculation and trajectory planning are re-executed for the remaining AGVs until all AGVs in the conflict group have passed through the conflict area.

[0030] In summary, this invention achieves robust and accurate perception of the microscopic motion state of each AGV under complex visual conditions by fitting a six-parameter affine model and combining it with a homography transformation matrix to convert image coordinates into actual planar coordinates, thus avoiding subsequent misjudgments or collisions caused by perception errors. By comprehensively calculating priorities based on task urgency and remaining path length, it realizes task-driven dynamic priority allocation, avoiding inefficient waiting or delays in important tasks. By transforming the trajectory of high-priority AGVs into dynamic obstacle constraints, it performs local MPC avoidance planning for low-priority AGVs, minimizing the passage delay of low-priority AGVs and thereby improving the overall traffic efficiency of intersections. Through secondary verification and generation of candidate avoidance action sets, and selection based on a multi-objective cost function, a safety closed loop of "planning-verification-correction" is formed.

[0031] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic priority traffic coordination control system for multi-AGV intersections combining visual detection, characterized in that: include, The status recognition module constructs a visual coordinate system for the intersection and generates a transformation matrix, acquires visual image data, identifies the real-time motion status of each AGV on the real-time occupied grid map, and outputs a set of AGV real-time motion statuses. The priority sorting module predicts the intersection occupancy time window based on the real-time motion state set of AVGs, identifies conflicting AGV groups based on the overlap of time windows, calculates the dynamic passage priority within the conflicting AGV groups, and forms a dynamic priority queue. The trajectory planning module extracts the high-priority AGV passage trajectory from the dynamic priority queue, generates a dynamic obstacle constraint set, performs local MPC avoidance planning on the low-priority AGV, and obtains the low-priority AGV avoidance trajectory. The secondary verification module performs a secondary verification of the AGV trajectory, adjusts the trajectory based on the verification results, and outputs a set of conflict-free trajectories. The execution feedback module generates traffic control commands based on a set of conflict-free trajectories and issues them to the AGV for execution, while collecting execution feedback to update the intersection status.

2. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 1, characterized in that: The steps for constructing the visual coordinate system of the intersection and generating the transformation matrix are as follows: Establish a plane coordinate system for the intersection and match the camera image coordinates with the plane coordinates of the intersection to obtain the homography transformation matrix.

3. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 2, characterized in that: The steps for identifying the real-time movement status of each AGV on the real-time occupied grid map and outputting the AGV real-time movement status set are as follows. The pre-trained AGV target recognition model is used to detect AGV targets in the current visual image data, and the visual detection box of each AGV in the image is obtained. The Shi-Tomasi corner detection algorithm is used to extract the set of stable feature points of AGV within the visual detection box. The multi-scale pyramid LK optical flow algorithm is used to construct a multi-scale image pyramid for the current frame and the next frame, and the set of stable feature points is projected onto the highest layer of the pyramid image to solve the optical flow displacement in the highest layer image. The optical flow displacement obtained from each layer is amplified and used as the initial displacement of the next layer. Then, local corrections are made to obtain the final displacement of the next layer. When iterating layer by layer to the original image layer, the final optical flow displacement of the stable feature points is obtained, and the predicted position of the stable feature points in the next frame is output. Based on the current frame position and the predicted position in the next frame of the stable feature point, a six-parameter affine model is fitted, and the affine model parameters are solved by the least squares method. Input the center point of the AGV visual detection box in the current frame into the six-parameter affine model to obtain the corrected center point of the next frame, and convert the corrected center point of the next frame into the intersection plane coordinate position through the homography transformation matrix. Calculate the visual velocity and visual heading angle based on the corrected changes in planar coordinates; The real-time motion status of each AGV is generated based on its AGV number, coordinate position, visual speed, and visual heading angle.

4. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 3, characterized in that: The steps for predicting the intersection occupancy time window based on the AVG real-time motion state set are as follows: Extract all potential conflict zones from the real-time occupancy raster map at intersections. The conflict zones include the central area of ​​the intersection, the left-turn merging area, the right-turn merging area, the straight-ahead merging area, and the loading / unloading point entrance area, forming a set of conflict zones. ; If the first The AGV route passes through the conflict area. Based on the AVG real-time motion state set, the first The AGV calculates its arrival time in the conflict zone. The estimated time, and based on the AGV's position in the conflict zone. The movement type within the area calculates the time taken to leave the conflict zone, forming the first... AGVs in the conflict area The time windows are defined, and the final output is a set of time windows for all AVGs.

5. The multi-AGV intersection dynamic priority passage coordination control system combining visual detection as described in claim 4, characterized in that: The steps for identifying conflicting AGV groups based on overlapping time windows, calculating dynamic passage priorities within each conflicting AGV group, and forming a dynamic priority queue are as follows. Based on the time window set of all AVGs, for any two AGVs in the same conflict zone Overlap detection is performed within a time window. If an overlap occurs, the distance between the two AGVs at that time is calculated, and it is determined whether the distance between the two AGVs at the overlap time is less than the sum of the dynamic safety radii of the two AGVs. If it is less, the two AGVs are determined to be in the conflict zone. There are potential conflicts within the memory; conflict zone A set of AGVs that have potential conflicts within the memory constitutes a conflicting AGV group; Calculate a dynamic priority score for all AGVs in the conflicting AGV group based on their arrival time, task urgency, remaining path length, current speed, and cargo status. The dynamic priority scores of all AGVs in the conflicting AGV group are sorted in descending order to form a dynamic priority queue.

6. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 5, characterized in that: The steps for extracting high-priority AGV trajectories from a dynamic priority queue and generating a dynamic obstacle constraint set are as follows. Get the future trajectory of the highest priority AGV in the dynamic priority queue. ,Will It is expanded into a dynamic restricted area, forming a dynamic obstacle constraint set.

7. The multi-AGV intersection dynamic priority passage coordination control system combining visual detection as described in claim 6, characterized in that: The steps for performing local MPC avoidance planning on low-priority AGVs to obtain their avoidance trajectories are as follows. Define the state for all AGVs in the dynamic priority queue except for the highest priority AGV. and control quantity ; Based on the dynamic obstacle constraint set, optimization objectives and control constraints are constructed, and local trajectory optimization is performed using improved MPC to obtain the low-priority AGV avoidance trajectory.

8. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 7, characterized in that: The steps for performing a second review of the AGV trajectory, adjusting the trajectory based on the review results, and outputting a set of conflict-free trajectories are as follows. Obtain the future trajectory of the high-priority AGV After avoiding the low-priority AGV trajectory, the trajectories of any two AGVs in the same conflict area are checked a second time at any prediction time. The second check includes point conflict check and edge conflict check. For any two AGVs within the same conflict zone, at any predicted time... If the edge conflict verification expression is satisfied, it means that an edge conflict still exists; If there are no conflicts during the second review, the trajectory of each AGV in the conflict area is output to obtain a set of conflict-free trajectories. If conflicts still exist after the second review, a set of candidate avoidance actions is generated for the low-priority AGVs among the conflicting AGVs, and a corresponding predicted trajectory is generated for each candidate action. The comprehensive cost is calculated based on the trajectory, and the candidate action that satisfies the control constraints and has the lowest comprehensive cost is selected as the final avoidance action. The corresponding avoidance trajectory is then output to obtain the adjusted set of conflict-free trajectories.

9. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 8, characterized in that: The steps for generating passage control commands based on the set of conflict-free trajectories and issuing them to the AGV for execution are as follows: After obtaining the set of conflict-free trajectories, an executable set of AGV control instructions is generated and issued for execution based on the behavior type of each AGV in the trajectory.

10. The multi-AGV intersection dynamic priority traffic coordination control system combining visual detection as described in claim 9, characterized in that: The steps for collecting, executing, and updating the intersection status are as follows: After the AGV executes the control command, the visual inspection camera continues to collect images of the intersection. When the highest priority AGV completely leaves the conflict area, the AGV is removed from the dynamic priority queue, and the dynamic priority score calculation and trajectory planning are re-executed for the remaining AGVs until all AGVs in the conflict group have passed through the conflict area.

Images