AGV markerless positioning and parking method and system based on storage location feature recognition

Abstract

The application provides an AGV non-mark positioning and parking method and system based on warehouse location feature recognition, which comprises the following steps: AGV receives the docking task issued by the dispatching system; AGV navigates to the preset recognition range of the target warehouse location according to the rough position, and starts the 3D vision acquisition module installed on the AGV; 3D point cloud data or depth image data is acquired through the 3D vision acquisition module; the acquired 3D point cloud data is processed to identify the inherent structural features of the target warehouse location; the center point coordinates of the target warehouse location are calculated through geometric fitting or calculation according to the identified inherent structural features; the deviation of the current pose of AGV relative to the center point of the warehouse location is calculated; and AGV is controlled to adjust the pose and accurately park at the target warehouse location according to the calculated deviation. The physical structure of the warehouse location is directly perceived through 3D image acquisition, and the center point of the warehouse location is calculated, so that high-precision parking without manual marking is realized.

Description

Technical Field

[0001] This invention relates to the field of automated guided vehicle (AGV) control technology, specifically to a method and system for unmarked positioning and docking of AGVs based on warehouse location feature recognition, which is particularly suitable for automated warehousing scenarios where there is no need for manual marking such as QR codes or magnetic strips on the ground or shelves. Background Technology

[0002] In automated warehousing and logistics systems, AGVs require extremely high final docking accuracy when they travel to target storage locations (such as racking locations, docking stations, charging stations, etc.) to dock with goods. The current mainstream solution is to affix QR codes to the ground or racking columns corresponding to each storage location. The AGV then uses an onboard camera to recognize the QR codes and obtain its precise position relative to the storage location, thereby achieving accurate docking.

[0003] However, this location method that relies on QR codes has the following inherent drawbacks: The initial deployment costs are high and the cycle is long: it requires pasting QR codes on the ground of thousands of storage locations one by one, and accurately measuring the world coordinates of each QR code to enter into the system, which is a huge project.

[0004] Post-maintenance is difficult: QR codes are easily crushed by forklift tires, covered by dust and oil, or worn or detached due to cargo handling, resulting in positioning failure and requiring frequent manual inspection and replacement.

[0005] Poor environmental adaptability: In harsh environments such as cold storage, humidity, and dust, the lifespan and recognition rate of QR codes will drop significantly.

[0006] Insufficient flexibility: When adjusting the warehouse layout, it is necessary to re-paste the QR code and update the map data, which is not flexible enough.

[0007] While there are visual positioning solutions in existing technologies, most still rely on manually set special markers (such as circular color blocks, coded points, etc.), failing to achieve true marker-free positioning.

[0008] Therefore, there is an urgent need for a method that does not rely on manual marking and can directly utilize the structural features of the storage location itself for high-precision positioning and docking. Summary of the Invention

[0009] To address the aforementioned problems in the existing technology, the present invention aims to provide a markerless positioning and docking method and system for AGVs based on warehouse location feature recognition. This method directly perceives the physical structure of the warehouse location using a 3D camera, calculates the center point of the warehouse location, and achieves high-precision docking without the need for manual marking.

[0010] The present invention achieves the above objectives through the following technical solutions: A markerless positioning and docking method for AGVs based on warehouse location feature recognition includes: Step S1: The AGV receives the docking task issued by the scheduling system. The task includes the identification information of the target storage location and its approximate location in the world coordinate system. Step S2: The AGV navigates to the preset recognition range of the target storage location based on the rough location, and the 3D vision acquisition module installed on the AGV is activated; Step S3: Acquire 3D point cloud data or depth image data containing the target storage location using the 3D vision acquisition module; Step S4: Process the collected 3D point cloud data to identify the inherent structural features of the target storage location; inherent structural features include the columns on both sides of the storage location, the bottom corners of the storage location, the back panel of the storage location, or the edges of the shelf panels; Step S5: Based on the identified inherent structural features, calculate the coordinates of the center point of the target storage location through geometric fitting or calculation, and further calculate the deviation of the AGV's current pose relative to the center point of the storage location. The current pose includes position and heading angle. Step S6: Based on the calculated deviation, control the AGV to adjust its posture and precisely stop at the target storage location.

[0011] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided. In step S1, the scheduling system receives a handling task issued by the upper management system. The handling task includes the identification information of the target material. The scheduling system queries the warehouse location management database according to the identification information of the target material to obtain the identification information of the target warehouse location where the material is currently located and its warehouse location coordinates in the warehouse map. The scheduling system maps the warehouse location coordinates to the world coordinate system to obtain the approximate location of the target warehouse location in the world coordinate system. The scheduling system sends the docking task containing the target warehouse location identification information and the approximate location to the AGV via wireless communication.

[0012] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided. In step S2, the AGV plans a navigation path from its current position to the area where the target warehouse is located by using a path planning algorithm based on its current pose and the approximate position of the target warehouse. The AGV travels along the navigation path. During the journey, the AGV's current position in the world coordinate system is obtained in real time through its own navigation system. Based on the current position and the approximate position of the target storage location, the remaining distance between the AGV and the target storage location is calculated in real time. When the AGV travels to a position less than or equal to the first preset distance threshold from the target storage location, the AGV decelerates to the first preset speed and enters the approach mode; when the AGV continues to travel to a position less than or equal to the second preset distance threshold from the target storage location, the AGV decelerates to the second preset speed. When the AGV travels to a position less than or equal to the third preset distance threshold from the target storage location, it is determined that the AGV has entered the preset recognition range of the target storage location. The AGV decelerates to a stop and starts the 3D vision acquisition module installed on the AGV to begin executing step S3.

[0013] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided. In step S3, after the AGV decelerates to a stop and determines that it has entered the preset recognition range of the target warehouse location, a collection trigger signal is sent to the 3D vision collection module. The 3D vision collection module responds to the collection trigger signal and performs one or more scans of the area where the target warehouse location is located according to preset collection parameters, generating original three-dimensional point cloud data or depth image data containing the target warehouse location. The collection parameters include collection frame rate, resolution, and exposure time. The acquired raw 3D point cloud data or depth image data is converted from the raw data format in the camera coordinate system to the required standardized data format. The converted 3D point cloud data or depth image data is output to the storage location feature recognition module to identify the inherent structural features of the target storage location.

[0014] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided. In step S4, the collected three-dimensional point cloud data is processed, including: using a voxel downsampling algorithm to reduce the point cloud density to reduce the amount of data, and using a statistical filtering algorithm to remove outlier noise points to obtain denoised three-dimensional point cloud data. The denoised 3D point cloud data is divided into several independent point cloud clusters using the Euclidean clustering algorithm, with each point cloud cluster corresponding to an independent physical target. Based on the geometric constraints of the storage location pillars, two point cloud clusters corresponding to the left and right pillars of the target storage location are selected from several independent point cloud clusters obtained by segmentation; the geometric constraints of the storage location pillars include the height range and width range of the pillars. The selected point cloud clusters for the left and right pillars are projected onto a horizontal plane. The least squares method is then used to fit the linear equation x·sinθ - y·cosθ + d = 0 to the projected point clouds, thus obtaining the central axis parameter (θ) of the left pillar. L , d L ) and the center axis parameters of the right column (θ) R , d R The center axis of the left column and the center axis of the right column together constitute the inherent structural feature identification result of the target storage location.

[0015] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided for a segmented m independent point cloud cluster. Calculate each point cloud cluster separately Geometric feature parameters: high: ,in, and Point cloud clusters Maximum and minimum height values ​​in the vertical direction; width: ,in,( , () is a cluster of point clouds The Middle The horizontal coordinates of the points, ( , () is a cluster of point clouds The horizontal centroid coordinates; volume: ,in, For point cloud clusters The number of points included The volume of a single voxel after voxel downsampling; Based on the preset column height range [ , ], Column width range , ] and column volume range [ , ],right m Initial screening is performed on individual point cloud clusters, and point cloud clusters that do not meet any of the following conditions are removed: The point cloud clusters that meet all the conditions are retained, and the set of point cloud clusters after the initial screening is denoted as:

[0016] in, k This represents the number of point cloud clusters retained after the initial screening.

[0017] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided, which performs pairwise pairing on the initially screened point cloud cluster set Ω, for any two point cloud clusters and Calculate its combined characteristic parameters: Horizontal spacing:

[0018] in,( , )and( , ( ) are point cloud clusters and The horizontal centroid coordinates; High consistency:

[0019] Width consistency:

[0020] Based on the preset standard width range of the storage location [ , Maximum permissible height deviation and maximum allowable width deviation Select a set Ψ of candidate pairs from all pairs that satisfy all of the following conditions: Perform optimal pairing selection on the candidate pairing set Ψ, selecting the pairing that minimizes the absolute value of the difference between the horizontal spacing and the standard width of the storage location. , As the optimal pairing:

[0021] in, This is the standard width of the storage location; Based on the horizontal centroid coordinates of the two point cloud clusters in the optimal pairing ( , )and( , The left-right positional relationship of the pillar points determines the cloud cluster on the left. And the right-side pillar dot cloud cluster : If the candidate pairing set Ψ is empty, the location feature recognition is determined to have failed, the current recognition process is terminated, and a recognition failure alarm signal is sent to the superior system.

[0022] According to the present invention, an AGV markerless positioning and docking method based on warehouse location feature recognition is provided. The method for calculating the coordinates of the warehouse location center point in step S5 includes the following steps: Obtain the fitted parameters of the center axis of the left column (θ) L , d L ) and the center axis parameters of the right column (θ) R ,d R Calculate the equations of the straight lines on the horizontal plane for the central axes of the left and right columns, respectively: The equation of the straight line of the central axis of the left column is: x·sinθ L - y·cosθ L + d L = 0; The equation of the straight line of the central axis of the right column is: x·sinθ R - y·cosθ R + d R = 0; Calculate the midpoint of the center axis of the left column and the center axis of the right column to form the center line of the storage location, specifically including: Calculate any point P on the central axis of the left column. L The coordinates of P, and the axis of the right column centerline. L At the same horizontal height Y H (i.e., the corresponding point P with the same y-coordinate) R The coordinates of P are calculated using the following formula. L and P R midpoint P M : P M = (P L + P R ) / 2; Let the y-coordinate value change from Y min To Y max Calculate the midpoint P at each horizontal height by continuously varying the preset step size ΔY. M (y), and all midpoints P M (y) The equation of the straight line connecting the two pillars forming the centerline of the storage location; the centerline of the storage location is the symmetrical centerline between the center axes of the left and right pillars, and its direction vector V mid = V L + V R V L and V R These are the unit vectors representing the directions of the center axes of the left and right columns, respectively.

[0023] According to the present invention, an AGV markerless positioning and docking method based on storage location feature recognition is provided. The RANSAC random sampling consensus algorithm is used to perform plane fitting on the point cloud located in the rear region of the target storage location in the denoised 3D point cloud data, thereby identifying the plane equation of the rear baffle of the target storage location. A f ·x + B f ·y + C f ·z + D f = 0; Simultaneously, the RANSAC random sampling consensus algorithm is used to perform plane fitting on the point cloud located in the bottom region of the target storage location in the denoised 3D point cloud data, and to identify the bottom plane equation of the target storage location: A g ·x + B g ·y + C g ·z + Dg = 0; Determine the center point P of the storage location center The three-dimensional coordinates, where: The horizontal coordinates are determined by calculating the intersection point P of the centerline of the storage location and the plane of the rear baffle. intersect Take the vector V along the direction of the storage location centerline. mid Distance from intersection point P intersect Preset depth offset L offset The point at point P is used as the horizontal coordinate of the center point of the storage location. centerhorizontal ,Right now: P centerhorizontal = P intersect + L offset · V mid ; Among them, the preset depth offset L offset Pre-set the depth required for the AGV forks to insert into the storage location; The height coordinates are determined by taking the height value Z from the equation of the bottom plane. floor Superimposed preset lifting offset H lift The height coordinates of the center point of the storage location are obtained: Z center = Z floor + H lift ; Among them, Z floor = -D g / C g Preset lift offset H lift The lifting height required for the AGV forks to pick up goods is preset; Finally, the center point P of the storage location was obtained. center The three-dimensional coordinates are: P center = (x center , y center , z center ); Where, x center and y center For P centerhorizontal x and y coordinates, z center = Z center ; Obtain the current pose information of the AGV in the world coordinate system, including the coordinates of the AGV's center point P. agv = (x agv , y agv ,z agv ) and AGV heading angle θ agv Calculate the deviation of the AGV relative to the center point of the storage location: Lateral offset: ΔX = x agv- x center ; Vertical offset: ΔY = y agv - y center ; Angular deviation: Δθ = θ agv - θ target ; Where, θ target The target heading angle of the reservoir centerline normal in the world coordinate system is given by the direction vector V of the reservoir centerline. mid The calculation yields: θ target = atan2(V midy V midx ) + π / 2.

[0024] A markerless positioning and docking system for AGVs based on warehouse location feature recognition includes: The 3D vision acquisition module, installed on the AGV, is used to acquire three-dimensional point cloud data or depth image data containing the target storage location. The storage location feature recognition module is used to process the collected 3D point cloud data and identify the inherent structural features of the target storage location. The warehouse center calculation module is used to calculate the coordinates of the center point of the target warehouse location based on the identified inherent structural features through geometric fitting or calculation, and further calculate the deviation of the AGV's current pose relative to the center point of the warehouse location. The docking control module is used to control the AGV to adjust its posture and precisely dock at the target storage location based on the calculated deviation.

[0025] Therefore, compared with the prior art, the AGV markerless positioning and docking method and system based on warehouse location feature recognition proposed in this invention has the following beneficial effects: 1. This invention directly utilizes the inherent physical structure of the storage location (side columns, back panel, bottom corners, etc.) as a positioning reference, eliminating the need for any manual markings such as QR codes or magnetic strips on the ground or shelves, and also eliminating the need for auxiliary positioning equipment such as reflectors or UWB base stations around the storage location. This changes the existing technical approach of relying on manual marking for precise AGV positioning and docking. This invention only requires inputting the structural parameters of the storage location itself (column spacing, height range, width range, etc.) into the system to complete deployment.

[0026] 2. This invention, through its constructed geometric feature parameter system and two-level screening architecture, can automatically identify and locate warehouse pillars from point cloud data, significantly reducing the time and labor costs of system deployment. In the later maintenance phase, since there is no need to deal with issues such as QR code wear, soiling, or detachment, the frequency of manual inspection and replacement is significantly reduced, effectively controlling the overall operating cost of the warehousing system.

[0027] 3. This invention uses a 3D camera to acquire three-dimensional point cloud data of the target storage location. Location is achieved by identifying the geometric features of the storage location's physical structures, such as the pillars, back panel, and bottom plane. These structural features are made of metal profiles, possessing sufficient structural strength and stability, and are not easily altered or damaged by environmental factors. Furthermore, this invention employs the RANSAC algorithm to fit the back panel plane and bottom plane, exhibiting strong robustness to outlier noise in the point cloud data. It can stably fit the planar parameters even when a certain proportion of interfering points exist in the point cloud, further enhancing the reliability of this invention in harsh environments.

[0028] 4. In this invention, the storage location reference is the physical structure of the storage location itself, and the storage location coordinate information is only related to the location of the storage location in the world coordinate system. When the storage location layout is adjusted, it is only necessary to update the location coordinates of the target storage location in the storage location management database of the scheduling system, without any modification to the physical environment of the storage location. The storage location feature recognition and center point calculation method of this invention is entirely based on the real-time perception of the inherent structure of the storage location by a 3D camera, and does not rely on any pre-laid physical markers. Therefore, changes in the storage location will not affect the effectiveness of the recognition algorithm, enabling the warehousing system to respond to changes in business needs with extremely high flexibility and achieve dynamic optimization of warehousing resources.

[0029] 5. This invention directly uses the actual physical structure of the storage location as a positioning reference. Based on real-time point cloud data perceived by a 3D camera, the center point of the storage location is calculated. The positioning benchmark is the actual physical location on-site, not the coordinates recorded in the database, achieving true WYSIWYG positioning. This invention constructs a virtual centerline using the symmetry of the left and right pillars. The accuracy of this virtual centerline depends on the accuracy of the pillar point cloud fitting, not the accuracy of manual measurement. By calculating the intersection of the centerline of the storage location and the back panel plane, and introducing a preset depth offset, this invention anchors the positioning benchmark to the actual physical structure of the storage location, fundamentally eliminating positioning deviations caused by measurement errors, marker displacement, or database coordinate errors. Simultaneously, this invention utilizes the RANSAC algorithm to simultaneously identify redundant designs in the back panel plane and the bottom plane. This ensures that even if the point cloud data of one plane is incomplete due to occlusion, the system can still complete the positioning calculation based on the information from the other plane, further improving the reliability of the positioning.

[0030] 6. This invention constructs a two-level positioning architecture of global coarse positioning and local fine positioning: At the global level, the AGV relies on the coarse position issued by the scheduling system and its own navigation system (such as laser SLAM) to travel to the vicinity of the target warehouse. At this stage, the requirements for navigation accuracy are low, and the AGV can maintain a high driving speed to improve work efficiency; At the local level, the AGV decelerates and stops after entering the preset recognition range, and high-precision pose calculation and fine-tuning of docking are performed by collecting point cloud data through a 3D camera.

[0031] 7. This invention decouples the positioning accuracy requirements of the final docking stage from the navigation efficiency requirements of the driving stage, enabling the AGV to use positioning methods with different accuracy levels and different technical paths in different operation stages. This ensures high accuracy at the final docking stage while also taking into account high efficiency for long-distance driving, and at the same time reduces the continuous occupation of 3D sensors and computing resources.

[0032] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0033] Figure 1 This is a flowchart of an embodiment of the AGV unmarked positioning and docking method based on warehouse location feature recognition according to the present invention.

[0034] Figure 2 This is a flowchart of step S1 in an embodiment of an AGV unmarked positioning and docking method based on warehouse location feature recognition according to the present invention.

[0035] Figure 3 This is a flowchart of step S2 in an embodiment of the AGV unmarked positioning and docking method based on warehouse location feature recognition of the present invention.

[0036] Figure 4 This is a flowchart of step S3 in an embodiment of the AGV unmarked positioning and docking method based on warehouse location feature recognition of the present invention.

[0037] Figure 5 This is a flowchart of step S4 in an embodiment of the AGV unmarked positioning and docking method based on warehouse location feature recognition of the present invention.

[0038] Figure 6 This is a flowchart of step S5 in an embodiment of the AGV unmarked positioning and docking method based on warehouse location feature recognition of the present invention.

[0039] Figure 7 This is a system schematic diagram of an embodiment of the AGV unmarked positioning and docking system based on warehouse location feature recognition according to the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0041] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0042] An Example of an AGV Markless Positioning and Docking Method Based on Warehouse Location Feature Recognition See Figure 1 This embodiment provides a markerless positioning and docking method for AGVs based on warehouse location feature recognition, including: Step S1: The AGV receives the docking task issued by the scheduling system. The task includes the identification information of the target storage location and its approximate location in the world coordinate system. Step S2: The AGV navigates to the preset recognition range of the target storage location based on the rough location, and the 3D vision acquisition module installed on the AGV is activated; Step S3: Acquire 3D point cloud data or depth image data containing the target storage location using the 3D vision acquisition module; Step S4: Process the collected 3D point cloud data to identify the inherent structural features of the target storage location; inherent structural features include the columns on both sides of the storage location, the bottom corners of the storage location, the back panel of the storage location, or the edges of the shelf panels; Step S5: Based on the identified inherent structural features, calculate the coordinates of the center point of the target storage location through geometric fitting or calculation, and further calculate the deviation of the AGV's current pose relative to the center point of the storage location. The current pose includes position and heading angle. Step S6: Based on the calculated deviation, control the AGV to adjust its posture and precisely stop at the target storage location.

[0043] In step S1, as Figure 2 As shown, the scheduling system receives handling tasks from the upper-level management system, which include the identification information of the target material. The scheduling system queries the storage location management database based on the identification information of the target material to obtain the identification information of the target storage location where the material is currently located and its storage location coordinates on the warehouse map. The scheduling system maps the storage location coordinates to the world coordinate system to obtain the approximate location of the target storage location in the world coordinate system. The scheduling system then sends the docking task containing the target storage location identification information and the approximate location to the AGV via wireless communication.

[0044] In step S2, as Figure 3 As shown, the AGV plans a navigation path from its current position to the area where the target storage location is located based on its current pose and the approximate location of the target storage location using a path planning algorithm. The AGV travels along the navigation path. During the journey, the AGV's current position in the world coordinate system is obtained in real time through its own navigation system. Based on the current position and the approximate position of the target storage location, the remaining distance between the AGV and the target storage location is calculated in real time. When the AGV travels to a position less than or equal to the first preset distance threshold from the target storage location, the AGV decelerates to the first preset speed and enters the approach mode; when the AGV continues to travel to a position less than or equal to the second preset distance threshold from the target storage location, the AGV decelerates to the second preset speed, which is less than the first preset speed. When the AGV travels to a position less than or equal to the third preset distance threshold from the target storage location, it is determined that the AGV has entered the preset recognition range of the target storage location. The AGV decelerates to a stop and starts the 3D vision acquisition module installed on the AGV to begin executing step S3. The third preset distance threshold is less than the second preset distance threshold, and the third preset distance threshold is preset according to the effective working distance of the 3D vision acquisition module.

[0045] In step S3, as Figure 4 As shown, after the AGV decelerates to a stop and determines that it has entered the preset recognition range of the target storage location, it sends a collection trigger signal to the 3D vision collection module; the 3D vision collection module responds to the collection trigger signal and performs one or more scans of the area where the target storage location is located according to the preset collection parameters, generating original three-dimensional point cloud data or depth image data containing the target storage location; the collection parameters include collection frame rate, resolution and exposure time; The acquired raw 3D point cloud data or depth image data is converted from the raw data format in the camera coordinate system to the required standardized data format. The converted 3D point cloud data or depth image data is output to the storage location feature recognition module to identify the inherent structural features of the target storage location.

[0046] In step S4, as Figure 5 As shown, the collected 3D point cloud data is processed, including: using a voxel downsampling algorithm to reduce the point cloud density to reduce the amount of data, and using a statistical filtering algorithm to remove outlier noise points to obtain denoised 3D point cloud data. The denoised 3D point cloud data is divided into several independent point cloud clusters using the Euclidean clustering algorithm, with each point cloud cluster corresponding to an independent physical target. Based on the geometric constraints of the storage location pillars, two point cloud clusters corresponding to the left and right pillars of the target storage location are selected from several independent point cloud clusters obtained by segmentation; the geometric constraints of the storage location pillars include the height range and width range of the pillars. The selected point cloud clusters for the left and right pillars are projected onto a horizontal plane. The least squares method is then used to fit the linear equation x·sinθ - y·cosθ + d = 0 to the projected point clouds, thus obtaining the central axis parameter (θ) of the left pillar. L , d L ) and the center axis parameters of the right column (θ) R , d R The center axis of the left column and the center axis of the right column together constitute the inherent structural feature identification result of the target storage location.

[0047] For the m independent point cloud clusters obtained from the segmentation Calculate each point cloud cluster separately Geometric feature parameters: high: ,in, and Point cloud clusters Maximum and minimum height values ​​in the vertical direction; width: ,in,( , () is a cluster of point clouds The Middle The horizontal coordinates of the points, ( , () is a cluster of point clouds The horizontal centroid coordinates; volume: ,in, For point cloud clusters The number of points included This represents the volume of a single voxel after voxel downsampling.

[0048] Based on the preset column height range [ , ], Column width range , ] and column volume range [ , ],right m Initial screening is performed on individual point cloud clusters, and point cloud clusters that do not meet any of the following conditions are removed: The point cloud clusters that meet all the conditions are retained, and the set of point cloud clusters after the initial screening is denoted as:

[0049] in, k This represents the number of point cloud clusters retained after the initial screening.

[0050] The point cloud clusters Ω after initial screening are paired up in pairs. For any two point cloud clusters... and ( p < q ), calculate its combined characteristic parameters: Horizontal spacing:

[0051] in( , )and( , ( ) are point cloud clusters and The horizontal centroid coordinates.

[0052] High consistency:

[0053] Width consistency:

[0054] Based on the preset standard width range of the storage location [ , Maximum permissible height deviation and maximum allowable width deviation Select a set Ψ of candidate pairs from all pairs that satisfy all of the following conditions: Perform optimal pairing selection on the candidate pairing set Ψ, selecting the pairing that minimizes the absolute value of the difference between the horizontal spacing and the standard width of the storage location. , As the optimal pairing:

[0055] in, This is the standard width for the storage location.

[0056] Based on the horizontal centroid coordinates of the two point cloud clusters in the optimal pairing ( , )and( , The left-right positional relationship of the pillar points determines the cloud cluster on the left. And the right-side pillar dot cloud cluster : If the candidate pairing set Ψ is empty, the location feature recognition is determined to have failed, the current recognition process is terminated, and a recognition failure alarm signal is sent to the superior system.

[0057] In this embodiment, as Figure 6 As shown, the method for calculating the coordinates of the storage location center point in step S5 includes the following steps: Obtain the fitted parameters of the center axis of the left column (θ) L , d L ) and the center axis parameters of the right column (θ) R ,d R Calculate the equations of the straight lines on the horizontal plane for the central axes of the left and right columns, respectively: The equation of the straight line of the central axis of the left column is: x·sinθ L - y·cosθ L + d L = 0; The equation of the straight line of the central axis of the right column is: x·sinθ R - y·cosθ R + d R = 0; Calculate the midpoint of the center axis of the left column and the center axis of the right column to form the center line of the storage location, specifically including: Calculate any point P on the central axis of the left column. L The coordinates of P, and the axis of the right column centerline. L At the same horizontal height Y H (i.e., the corresponding point P with the same y-coordinate) R The coordinates of P are calculated using the following formula. L and P R midpoint P M : P M = (P L + P R ) / 2; Let the y-coordinate value change from Y min To Y max Calculate the midpoint P at each horizontal height by continuously varying the preset step size ΔY. M (y), and all midpoints P M (y) The equation of the straight line connecting the two pillars forming the centerline of the storage location; the centerline of the storage location is the symmetrical centerline between the center axes of the left and right pillars, and its direction vector V mid = V L + V R V L and V R These are the unit vectors representing the directions of the center axes of the left and right columns, respectively.

[0058] The RANSAC random sampling consensus algorithm is used to perform plane fitting on the point cloud located in the rear region of the target storage location in the denoised 3D point cloud data, and the plane equation of the rear baffle of the target storage location is identified: A f ·x + Bf ·y + C f ·z + D f = 0; Simultaneously, the RANSAC random sampling consensus algorithm is used to perform plane fitting on the point cloud located in the bottom region of the target storage location in the denoised 3D point cloud data, and to identify the bottom plane equation of the target storage location: A g ·x + B g ·y + C g ·z + D g = 0; Determine the center point P of the storage location center The three-dimensional coordinates, where: The horizontal coordinates are determined by calculating the intersection point P of the centerline of the storage location and the plane of the rear baffle. intersect Take the vector V along the direction of the storage location centerline. mid Distance from intersection point P intersect Preset depth offset L offset The point at point P is used as the horizontal coordinate of the center point of the storage location. centerhorizontal ,Right now: P centerhorizontal = P intersect + L offset · V mid ; Among them, the preset depth offset L offset Pre-set the depth required for the AGV forks to insert into the storage location; The height coordinates are determined by taking the height value Z from the equation of the bottom plane. floor Superimposed preset lifting offset H lift The height coordinates of the center point of the storage location are obtained: Z center = Z floor + H lift ; Among them, Z floor = -D g / C g (When C) g (≠ 0) Preset lifting offset H lift The lifting height required for the AGV forks to pick up goods is preset; Finally, the center point P of the storage location was obtained. center The three-dimensional coordinates are: P center = (x center , y center , z center ); Where, x center and y center For Pcenterhorizontal x and y coordinates, z center = Z center ; Obtain the current pose information of the AGV in the world coordinate system, including the coordinates of the AGV's center point P. agv = (x agv , y agv ,z agv ) and AGV heading angle θ agv Calculate the deviation of the AGV relative to the center point of the storage location: Lateral offset: ΔX = x agv - x center ; Vertical offset: ΔY = y agv - y center ; Angular deviation: Δθ = θ agv - θ target ; Where, θ target The target heading angle of the reservoir centerline normal in the world coordinate system is given by the direction vector V of the reservoir centerline. mid The calculation yields: θ target = atan2(V midy V midx ) + π / 2.

[0059] In practical applications, this embodiment is used with a stackable AGV equipped with a 3D vision camera (such as a binocular camera, TOF camera, or LiDAR) for pallet storage and retrieval operations in a heavy-duty racking area. The racking unit consists of two vertical columns on both sides and a rear baffle, without any QR code markings.

[0060] Coarse positioning approach steps The AGV receives the dispatch task: to go to warehouse location B-12 in the shelving area to pick up goods. The AGV travels to the aisle where warehouse location B-12 is located according to its own navigation system (such as laser SLAM, inertial navigation) and stops at the preset coarse positioning point about 1.5 meters in front of the shelving.

[0061] 3D vision acquisition steps The AGV decelerates to a stop and activates the 3D TOF camera mounted on the front of the vehicle at an upward angle. The camera scans the target storage area directly in front of it, acquiring dense three-dimensional point cloud data that includes information about the side pillars, rear panel, and ground.

[0062] Storage location feature identification steps The storage location feature recognition module preprocesses the point cloud data (filtering and downsampling), and sets regions of interest (ROIs) in the point cloud according to the preset storage location size range (column spacing approximately 2.3 meters, column height range 0-3 meters). Euclidean clustering is used to separate the point cloud clusters of the left and right columns from the ROI. Random Sample Consensus (RANSAC) algorithm is then used to fit spatial straight lines to the point cloud clusters on both sides, obtaining the left column axis L1 and the right column axis L2.

[0063] Storage location center calculation steps The storage location center calculation module calculates the center points of L1 and L2, and the line connecting these two points is the theoretical centerline of the storage location. Next, the point cloud plane of the rear baffle is identified, the intersection of this plane and the storage location centerline is taken, and the depth offset required for fork insertion is considered to finally determine the position of the three-dimensional coordinates (Xc, Yc, Zc) of the storage location center point P_center in the camera coordinate system.

[0064] Simultaneously, obtain the current center point coordinates P_agv and heading angle θ_agv of the AGV. Calculate the lateral deviation ΔX from P_agv to P_center, the longitudinal distance ΔY, and the angular deviation Δθ between the current heading and the normal to the centerline of the storage location.

[0065] Precise docking procedures The docking control module sends the calculated deviations (ΔX, ΔY, Δθ) to the AGV's motion controller. The controller plans a fine-tuning path, controlling the AGV to move slowly and adjust its direction until ΔX and Δθ approach zero and ΔY reaches the preset safe docking distance (such as the position where the forks are about to insert into the storage space). Finally, a stop command is issued to complete the precise docking.

[0066] If only one column is identified due to obstructed view or ambient light interference, the AGV can move slightly laterally according to a preset strategy to re-collect data until both sides are successfully identified. If multiple attempts fail, the AGV will stop and send an alarm message indicating that the warehouse location feature recognition has failed to be identified to the upper-level system.

[0067] An embodiment of an AGV unmarked positioning and docking system based on warehouse location feature recognition like Figure 7 As shown, this embodiment provides an AGV unmarked positioning and docking system based on warehouse location feature recognition, comprising: The 3D vision acquisition module, installed on the AGV, is used to acquire three-dimensional point cloud data or depth image data containing the target storage location. The storage location feature recognition module is used to process the collected 3D point cloud data and identify the inherent structural features of the target storage location. The warehouse center calculation module is used to calculate the coordinates of the center point of the target warehouse location based on the identified inherent structural features through geometric fitting or calculation, and further calculate the deviation of the AGV's current pose relative to the center point of the warehouse location. The docking control module is used to control the AGV to adjust its posture and precisely dock at the target storage location based on the calculated deviation.

[0068] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0069] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A markerless positioning and docking method for AGVs based on warehouse location feature recognition, characterized in that, include: Step S1: The AGV receives the docking task issued by the scheduling system. The task includes the identification information of the target storage location and its approximate location in the world coordinate system. Step S2: The AGV navigates to the preset recognition range of the target storage location based on the rough location, and the 3D vision acquisition module installed on the AGV is activated; Step S3: Acquire 3D point cloud data or depth image data containing the target storage location using the 3D vision acquisition module; Step S4: Process the collected 3D point cloud data to identify the inherent structural features of the target storage location; inherent structural features include the columns on both sides of the storage location, the bottom corners of the storage location, the back panel of the storage location, or the edges of the shelf panels; Step S5: Based on the identified inherent structural features, calculate the coordinates of the center point of the target storage location through geometric fitting or calculation, and further calculate the deviation of the AGV's current pose relative to the center point of the storage location. The current pose includes position and heading angle. Step S6: Based on the calculated deviation, control the AGV to adjust its posture and precisely stop at the target storage location.

2. The method according to claim 1, characterized in that: In step S1, the scheduling system receives a handling task from the upper-level management system, which includes the identification information of the target material. The scheduling system queries the storage location management database based on the identification information of the target material to obtain the identification information of the target storage location where the material is currently located and its storage location coordinates on the warehouse map. The scheduling system maps the storage location coordinates to the world coordinate system to obtain the approximate location of the target storage location in the world coordinate system. The scheduling system sends the docking task containing the target storage location identification information and the approximate location to the AGV via wireless communication.

3. The method according to claim 1, characterized in that: In step S2, the AGV plans a navigation path from its current position to the area where the target storage location is located using a path planning algorithm, based on its current pose and the approximate location of the target storage location. The AGV travels along the navigation path. During the journey, the AGV's current position in the world coordinate system is obtained in real time through its own navigation system. Based on the current position and the approximate position of the target storage location, the remaining distance between the AGV and the target storage location is calculated in real time. When the AGV travels to a position less than or equal to the first preset distance threshold from the target storage location, the AGV decelerates to the first preset speed and enters the approach mode; when the AGV continues to travel to a position less than or equal to the second preset distance threshold from the target storage location, the AGV decelerates to the second preset speed. When the AGV travels to a position less than or equal to the third preset distance threshold from the target storage location, it is determined that the AGV has entered the preset recognition range of the target storage location. The AGV decelerates to a stop and starts the 3D vision acquisition module installed on the AGV to begin executing step S3.

4. The method according to claim 1, characterized in that: In step S3, after the AGV decelerates to a stop and determines that it has entered the preset recognition range of the target storage location, a collection trigger signal is sent to the 3D vision acquisition module. The 3D vision acquisition module responds to the acquisition trigger signal and performs one or more scans of the target storage location area according to preset acquisition parameters, generating raw 3D point cloud data or depth image data containing the target storage location; the acquisition parameters include acquisition frame rate, resolution and exposure time. The acquired raw 3D point cloud data or depth image data is converted from the raw data format in the camera coordinate system to the required standardized data format. The converted 3D point cloud data or depth image data is output to the storage location feature recognition module to identify the inherent structural features of the target storage location.

5. The method according to claim 1, characterized in that: In step S4, the acquired 3D point cloud data is processed, including: using a voxel downsampling algorithm to reduce the point cloud density to reduce the amount of data, and using a statistical filtering algorithm to remove outlier noise points to obtain denoised 3D point cloud data. The denoised 3D point cloud data is divided into several independent point cloud clusters using the Euclidean clustering algorithm, with each point cloud cluster corresponding to an independent physical target. Based on the geometric constraints of the storage location pillars, two point cloud clusters corresponding to the left and right pillars of the target storage location are selected from several independent point cloud clusters obtained by segmentation; the geometric constraints of the storage location pillars include the height range and width range of the pillars. The selected point cloud clusters for the left and right pillars are projected onto a horizontal plane. The least squares method is then used to fit the linear equation x·sinθ - y·cosθ + d = 0 to the projected point clouds, thus obtaining the central axis parameter (θ) of the left pillar. L , d L ) and the center axis parameters of the right column (θ) R , d R The center axis of the left column and the center axis of the right column together constitute the inherent structural feature identification result of the target storage location.

6. The method according to claim 5, characterized in that: For the m independent point cloud clusters obtained from the segmentation Calculate each point cloud cluster separately Geometric feature parameters: high: ,in, and Point cloud clusters Maximum and minimum height values ​​in the vertical direction; width: ,in,( , () is a cluster of point clouds The Middle The horizontal coordinates of the points, ( , () is a cluster of point clouds The horizontal centroid coordinates; volume: ,in, For point cloud clusters The number of points included The volume of a single voxel after voxel downsampling; Based on the preset column height range [ , ], Column width range , ] and column volume range [ , ],right m Initial screening is performed on individual point cloud clusters, and point cloud clusters that do not meet any of the following conditions are removed: The point cloud clusters that meet all the conditions are retained, and the set of point cloud clusters after the initial screening is denoted as: in, k This represents the number of point cloud clusters retained after the initial screening.

7. The method according to claim 6, characterized in that: The point cloud clusters Ω after initial screening are paired up in pairs. For any two point cloud clusters... and Calculate its combined characteristic parameters: Horizontal spacing: in,( , )and( , ( ) are point cloud clusters and The horizontal centroid coordinates; High consistency: Width consistency: Based on the preset standard width range of the storage location [ , Maximum permissible height deviation and maximum allowable width deviation Select a set Ψ of candidate pairs from all pairs that satisfy all of the following conditions: Perform optimal pairing selection on the candidate pairing set Ψ, selecting the pairing that minimizes the absolute value of the difference between the horizontal spacing and the standard width of the storage location. , As the optimal pairing: in, This is the standard width of the storage location; Based on the horizontal centroid coordinates of the two point cloud clusters in the optimal pairing ( , )and( , The left-right positional relationship of the pillar points determines the cloud cluster on the left. And the right-side pillar dot cloud cluster : If the candidate pairing set Ψ is empty, the location feature recognition is determined to have failed, the current recognition process is terminated, and a recognition failure alarm signal is sent to the superior system.

8. The method according to any one of claims 1 to 7, characterized in that, The method for calculating the coordinates of the storage location center point in step S5 includes the following steps: Obtain the fitted parameters of the center axis of the left column (θ) L , d L ) and the center axis parameters of the right column (θ) R , d R Calculate the equations of the straight lines on the horizontal plane for the central axes of the left and right columns, respectively: The equation of the straight line of the central axis of the left column is: x·sinθ L - y·cosθ L + d L = 0; The equation of the straight line of the central axis of the right column is: x·sinθ R - y·cosθ R + d R = 0; Calculate the midpoint of the center axis of the left column and the center axis of the right column to form the center line of the storage location, specifically including: Calculate any point P on the central axis of the left column. L The coordinates of P, and the axis of the right column centerline. L At the same horizontal height Y H (i.e., the corresponding point P with the same y-coordinate) R The coordinates of P are calculated using the following formula. L and P R midpoint P M : P M = (P L + P R ) / 2; Let the y-coordinate value change from Y min To Y max Calculate the midpoint P at each horizontal height by continuously varying the preset step size ΔY. M (y), and all midpoints P M (y) The equation of the straight line connecting the two pillars forming the centerline of the storage location; the centerline of the storage location is the symmetrical centerline between the center axes of the left and right pillars, and its direction vector V mid = V L + V R V L and V R These are the unit vectors representing the directions of the center axes of the left and right columns, respectively.

9. The method according to claim 8, characterized in that: The RANSAC random sampling consensus algorithm is used to perform plane fitting on the point cloud located in the rear region of the target storage location in the denoised 3D point cloud data, and the plane equation of the rear baffle of the target storage location is identified: A f ·x + B f ·y + C f ·z + D f = 0; Simultaneously, the RANSAC random sampling consensus algorithm is used to perform plane fitting on the point cloud located in the bottom region of the target storage location in the denoised 3D point cloud data, and to identify the bottom plane equation of the target storage location: A g ·x + B g ·y + C g ·z + D g = 0; Determine the center point P of the storage location center The three-dimensional coordinates, where: The horizontal coordinates are determined by calculating the intersection point P of the centerline of the storage location and the plane of the rear baffle. intersect Take the vector V along the direction of the storage location centerline. mid Distance from intersection point P intersect Preset depth offset L offset The point at point P is used as the horizontal coordinate of the center point of the storage location. centerhorizontal ,Right now: P centerhorizontal = P intersect + L offset · V mid ; Among them, the preset depth offset L offset Pre-set the depth required for the AGV forks to insert into the storage location; The height coordinates are determined by taking the height value Z from the equation of the bottom plane. floor Superimposed preset lifting offset H lift The height coordinates of the center point of the storage location are obtained: Z center = Z floor + H lift ; Among them, Z floor = -D g / C g Preset lift offset H lift The lifting height required for the AGV forks to pick up goods is preset; Finally, the center point P of the storage location was obtained. center The three-dimensional coordinates are: P center = (x center , y center , z center ); Where, x center and y center For P centerhorizontal x and y coordinates, z center = Z center ; Obtain the current pose information of the AGV in the world coordinate system, including the coordinates of the AGV's center point P. agv = (x agv , y agv ,z agv ) and AGV heading angle θ agv Calculate the deviation of the AGV relative to the center point of the storage location: Lateral offset: ΔX = x agv - x center ; Vertical offset: ΔY = y agv - y center ; Angle deviation:Δθ = θ agv - i target ; Where, θ target The target heading angle of the reservoir centerline normal in the world coordinate system is given by the direction vector V of the reservoir centerline. mid The calculation yields: θ target = atan2(V midy V midx ) + π / 2.

10. A markerless positioning and docking system for AGVs based on warehouse location feature recognition, characterized in that, include: The 3D vision acquisition module, installed on the AGV, is used to acquire three-dimensional point cloud data or depth image data containing the target storage location. The storage location feature recognition module is used to process the collected 3D point cloud data and identify the inherent structural features of the target storage location. The warehouse center calculation module is used to calculate the coordinates of the center point of the target warehouse location based on the identified inherent structural features through geometric fitting or calculation, and further calculate the deviation of the AGV's current pose relative to the center point of the warehouse location. The docking control module is used to control the AGV to adjust its posture and precisely dock at the target storage location based on the calculated deviation.

Images