A point cloud topological map generation method and system, and a storage medium

Abstract

The application discloses a kind of point cloud topology map generation method and system, storage medium, the method includes: analyzing warehouse scene image and extracting multiple shelf strips;Adjacent shelf strips that meet preset merging condition are merged into shelf group;Obstacle region and free region are determined based on shelf area and shelf group, and candidate passage is extracted;Side boundary is formed by expanding outside in the form of side boundary of shelf group, and longitudinal passage is obtained by screening according to the position relationship of candidate passage center line and side boundary;Transverse passage is generated outside the overall boundary of shelf group, and form closed loop passage network;Position marker point and connection relationship are generated along passage network, and storage location unit is identified in the region of shelf group and the associated action point is generated;Obstacle point set, position marker point, action point, connection relationship and storage location association are converted to map coordinate system, and point cloud topology map is generated.Thereby realize that warehouse scene image is automatically converted into the structured map usable by robot, and reduce false passage and passage fracture.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing and mobile robot map construction technology, and in particular to a point cloud topology map generation method and system, and storage medium based on image analysis and structured channel reconstruction. Background Technology

[0002] In smart warehousing scenarios, unmanned handling equipment such as mobile robots, forklift robots, and automated guided vehicles typically rely on structured maps for localization, path planning, task scheduling, and work point matching before performing tasks such as handling, shelving, picking, and replenishment. The maps required for actual deployment include not only obstacle boundaries or passable areas, but also path location points, action points associated with the operation, inter-point connections, and storage location information.

[0003] Existing methods for constructing warehouse maps mainly include: operators manually drawing obstacle boundaries, aisle paths, location points, and storage location information in map editing software, creating maps through on-site point collection, or manually converting graphic elements from CAD drawings and warehouse floor plans into map objects usable by robots. These methods are labor-intensive, time-consuming, and different operators have varying understandings of the relationships between aisle boundaries, storage location locations, and action points, resulting in poor consistency and reusability of the generated maps.

[0004] Furthermore, in actual warehouse deployment, it is common to encounter situations where only static images such as warehouse screenshots, warehouse planning maps, layout diagrams, or CAD-exported images are available. These images themselves do not contain path topology, action points, and storage location data that can be directly parsed by the robot system, making it difficult to directly generate deployable structured maps. Ordinary image segmentation methods can usually only identify color areas or blank areas in the image, lacking an understanding of the warehouse layout structure: on the one hand, there are often narrow gaps between shelves that do not actually meet the conditions for robot passage. These gaps appear as blank areas in the image and are easily misidentified as passable passages, resulting in false passages or broken passages in the generated map; on the other hand, storage locations are usually represented in the image as repetitive unit structures inside the shelves. Existing solutions cannot automatically recover storage location units from these and establish the correspondence between storage locations and robot operating positions, relying heavily on manual data entry.

[0005] Therefore, how to automatically extract shelf, aisle, and storage location information from warehouse scene images and organize it into a structured map that can be directly used by mobile robots is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] Therefore, the main objective of this invention is to provide a point cloud topology map generation method and system, and storage medium, so as to automatically extract shelf, aisle and storage location information from warehouse scene images and generate a point cloud topology map that can be directly used by mobile robots, thereby reducing the cost of manual mapping and improving map generation efficiency and result consistency.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for generating a point cloud topology map is provided, the steps of which include:

[0008] The acquired warehouse scene images are analyzed to identify shelf areas and extract multiple shelf strips from them;

[0009] Merge adjacent shelf strips that meet the preset merging conditions into a shelf group;

[0010] Based on the shelf area and the shelf group, an obstacle area is determined, the area outside the obstacle area is determined as a free area, and candidate channels are extracted based on the free area;

[0011] A side strip is formed by extending outward from the side boundary of the shelving group. The candidate channels are screened according to the positional relationship between the center line of the candidate channel and the side strip to obtain the longitudinal channel.

[0012] Based on the overall boundary of each of the aforementioned shelving groups, a transverse channel is generated on the outside of the overall boundary, and the transverse channel is connected with the longitudinal channel to form a closed-loop channel network.

[0013] Location markers and connections are generated along the channel network, storage units are identified within the shelving group area, and action points associated with the storage units are generated on the channel network.

[0014] The set of obstacle points extracted based on the obstacle area, the location markers, the action points, the connection relationships, and the association between the cargo space unit and the action points are uniformly converted to the map coordinate system to generate a point cloud topology map.

[0015] In a possible preferred embodiment, the preset merging conditions include: the spacing between adjacent shelf strips is less than a preset gap threshold, and the projection overlap rate of adjacent shelf strips in the length direction of the shelf strips is greater than a preset overlap threshold; the shelf group covers the gap between the adjacent shelf strips; the projection overlap rate is the ratio of the smaller of the overlap length of the adjacent shelf strips in the length direction and the length of the adjacent shelf strips in the length direction.

[0016] In a possible preferred embodiment, the step of screening the candidate channels to obtain the longitudinal channels includes: retaining the candidate channels whose center lines are located within the side strips as the longitudinal channels, and eliminating the candidate channels whose center lines are located within the shelving group; the side strips include a left side strip and a right side strip formed by extending outwards by a preset width based on the left and right boundaries of the shelving group; the center line is the average value of the positions of the two boundary positions of the candidate channels.

[0017] In a possible preferred embodiment, the step of generating a transverse passage outside the overall boundary of each of the shelving groups includes: determining the overall upper boundary and overall lower boundary of each of the shelving groups, and generating a top transverse passage and a bottom transverse passage at positions offset by a preset distance from the overall upper boundary and the overall lower boundary, respectively.

[0018] In a possible preferred embodiment, the step of extracting candidate channels based on the free region includes: calculating the percentage of pixels belonging to the free region in each pixel column, and determining the continuous pixel column intervals in which the percentage of pixels is greater than a preset percentage threshold and the width meets the preset minimum channel width as candidate channels.

[0019] In a possible preferred embodiment, the step of identifying storage location units within the shelf group area includes: determining the central axis of the shelf group, dividing the shelf group into side areas on both sides of the central axis, extracting storage location candidate frames in each side area, and filtering the storage location candidate frames according to preset geometric conditions to obtain storage location units.

[0020] In a possible preferred embodiment, the step of generating action points associated with the storage unit on the aisle network includes: mapping the storage unit toward the aisle centerline closest to it, and determining the corresponding point mapped onto the aisle centerline as the action point; calculating the distance between the center point of the storage unit and each action point, and selecting the action point with the smallest distance from the action points located on the same shelf side as the storage unit as the associated action point of the storage unit.

[0021] In a possible preferred embodiment, the point cloud topology map generation method further includes the following steps: acquiring a reference image containing a planned route, extracting a reference channel centerline from the reference image; determining an automatically generated channel centerline corresponding to the reference channel centerline, and weighting and fusing the reference channel centerline and the corresponding channel centerline according to a preset weight to obtain a corrected channel centerline.

[0022] To achieve the above objectives, corresponding to the above method example, according to another aspect of the present invention, a point cloud topology map generation system is also provided, comprising:

[0023] The image analysis module is used to perform region analysis on the acquired warehouse scene images, identify the shelf areas, and extract multiple shelf strips from them;

[0024] The shelving group building module is used to merge adjacent shelving strips that meet preset merging conditions into a shelving group;

[0025] The aisle reconstruction module is used to determine the obstacle area based on the shelf area and the shelf group, define the area outside the obstacle area as the free area, and extract candidate aisles based on the free area; expand outward from the side boundary of the shelf group to form a side strip, filter the candidate aisles according to the positional relationship between the centerline of the candidate aisle and the side strip to obtain the longitudinal aisle; generate transverse aisles outside the overall boundary of each shelf group, and connect the transverse aisles with the longitudinal aisles to form a closed-loop aisle network;

[0026] The map generation module is used to generate location markers and connection relationships along the channel network, identify storage units within the shelf group area, and generate action points associated with the storage units on the channel network; and uniformly convert the obstacle point set extracted based on the obstacle area, the location markers, the action points, the connection relationships, and the association between the storage units and the action points to the map coordinate system to generate a point cloud topology map.

[0027] To achieve the above objectives, in accordance with the above method examples, according to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the point cloud topology map generation method as described in any of the above examples.

[0028] The point cloud topology map generation method, system, and storage medium provided by this invention utilize the structural patterns of shelf layout in a warehouse scenario to progressively constrain the map generation process, rather than simply determining passable spaces based on blank areas in the image. First, adjacent shelf strips meeting preset merging conditions are merged into shelf groups, thus incorporating gaps between shelves that are not actually passable into these groups, reducing interference from pseudo-free regions on subsequent channel extraction from the outset. Then, the side strips formed by extending outwards from the side boundaries of the shelf groups constrain the position of candidate channels, transforming channel recognition from searching for blank areas in the image to finding structured channels that conform to the shelf layout and robot movement patterns. Next, lateral channels are reconstructed outside the overall boundaries of the shelf groups, ensuring closed-loop connectivity of the channel network. Simultaneously, location identification is limited to the shelf group area, and action points associated with location units are generated on the channel network, automatically establishing a correspondence between warehouse operation information and navigation topology.

[0029] This enables the automatic generation of structured maps usable by robots from warehouse scene images, reducing the workload and deployment cycle of manual mapping. Moreover, the generation process is driven by unified rules, avoiding inconsistencies in results caused by differences in understanding among different operators, and improving map generation efficiency and consistency. Attached Figure Description

[0030] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0031] Figure 1 A flowchart illustrating a point cloud topology map generation method provided in an embodiment of the present invention;

[0032] Figure 2 This is a schematic diagram of the original warehouse scene provided in an embodiment of the present invention;

[0033] Figure 3 This is a schematic diagram of the original warehouse scene image after grayscale preprocessing provided in an embodiment of the present invention;

[0034] Figure 4 This is a schematic diagram of the original warehouse scene image after binary preprocessing provided in an embodiment of the present invention;

[0035] Figure 5 This is a schematic diagram of image region layering provided in an embodiment of the present invention;

[0036] Figure 6 A comparative schematic diagram showing the merging of adjacent shelf strips before and after, as provided in an embodiment of the present invention;

[0037] Figure 7 This is a comparative diagram showing the candidate channel selection process before and after the screening process, as provided in an embodiment of the present invention.

[0038] Figure 8 A schematic diagram of the structured channel network reconstruction results provided in an embodiment of the present invention;

[0039] Figure 9 This is a schematic diagram of the point cloud topology map generation system according to an embodiment of the present invention. Detailed Implementation

[0040] To enable those skilled in the art to better understand the technical solutions of the present invention, the specific technical solutions of the present invention will be clearly and completely described below in conjunction with embodiments, so as to help those skilled in the art further understand the present invention. Obviously, the embodiments described in this application are merely some embodiments of the present invention, and not all embodiments. It should be noted that, for those skilled in the art, the embodiments and features in the embodiments of this application can be combined with each other without departing from the concept of the present invention and without conflict. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the disclosure and protection scope of the present invention.

[0041] Furthermore, the terms "first," "second," "S100," "S200," etc., used in the specification, claims, and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such features can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those described herein. At the same time, the stages described in each step are not necessarily to be implemented in the same step; it should be understood that the implementation order of the contents of each step stage can be adjusted and interchanged without violating the inventive concept, so that embodiments of the invention described herein can be implemented in orders other than those described herein. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. Unless otherwise expressly specified and limited, the terms "set," "arrange," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two elements. Those skilled in the art can understand the specific meaning of the above terms in this case based on the specific circumstances and in conjunction with existing technology.

[0042] Before introducing specific embodiments, some terms involved in this invention will be explained. A point cloud topology map refers to a structured map file used for positioning, navigation, scheduling, and task execution by mobile robots, forklift robots, AGVs, or AMRs. It typically includes data such as obstacle point sets, location marker set, action point sets, connection relationship sets, curve relationship sets, and a list of storage locations. An action point refers to a map point associated with a specific task, such as a pickup point, drop point, docking point, barcode scanning point, waiting point, or other location points where the robot needs to perform an action. A shelf strip refers to an image area distributed in a long strip in a warehouse scene image, used to represent a single row of shelves or the side area of ​​a shelf. An obstacle mask refers to a binary image or marker matrix used to identify impassable areas. Projection overlap rate refers to the ratio between the projected overlap length of two target areas in a certain direction and the length of the target area, used to determine whether adjacent target areas have sufficient spatial correspondence.

[0043] Specifically, to address the problems of existing warehouse map construction methods that rely on manual drawing, are inefficient, have poor consistency, and make it difficult to directly convert static warehouse images into structured maps usable by robots, this invention provides a point cloud topology map generation method. For example... Figures 1 to 8 As shown, the method includes the following example steps:

[0044] S101 performs regional analysis on the acquired warehouse scene image, identifies the shelf area, and extracts multiple shelf strips from it.

[0045] In this example, the warehouse scene images obtained can be warehouse planning screenshots, warehouse floor plans, static images exported from a map editor, or CAD exported images. These images typically use different colors to represent shelves, boundaries, and other auxiliary information; this step utilizes this color coding pattern to perform region analysis on the images.

[0046] Specifically, the input image is first converted from its original color space to a target color space suitable for region segmentation, such as the HSV color space. For each pixel in the image, its color value is determined to fall within a threshold range to determine whether the pixel belongs to a shelf region, a boundary region, or another region. A shelf region can be represented as:

[0047] Formula 1

[0048] That is, when the pixel color value Falling into the shelf color threshold range season Otherwise ;

[0049] The boundary region can be represented as:

[0050] Formula 2

[0051] That is, when Falling into the boundary color threshold range season Otherwise .

[0052] Since the original image may contain annotations, local noise, color breaks, or discontinuous boundaries, morphological processing is required on the segmented shelf areas to eliminate isolated noise and fill local holes. The processed shelf areas serve as the basis for subsequent shelf strip extraction. Simultaneously, all shelf areas are merged to obtain the minimum bounding rectangle, which is then expanded outwards by 50-100 pixels to form a complete analysis region. Subsequent recognition and detection processes are performed within this analysis region, thereby reducing interference from irrelevant peripheral areas.

[0053] After obtaining the processed shelving area, perform connected component analysis or projection analysis on it to extract multiple shelving strips, each of which corresponds to a relatively continuous long strip of shelving in the image.

[0054] For example, suppose the input warehouse scene image size is 2400×1600 pixels, such as Figures 2 to 4 As shown, after performing grayscale and binarization preprocessing and color thresholding segmentation on the image in sequence, the following result is obtained: Figure 5 The shelving area and boundary area are shown, and then multiple shelving strips are extracted from the shelving area.

[0055] This step transforms the original warehouse scene image into clearly defined shelving areas, boundary areas, and shelving strips, providing an analytical foundation for subsequent map generation using shelving layout structure constraints.

[0056] Alternatively, in alternative implementations, image region layering can employ not only the HSV color space, but also the Lab color space, RGB component thresholding, grayscale thresholding, or region extraction based on a semantic segmentation model; and shelf strip extraction can be achieved not only through connected component analysis, but also through column projection analysis, edge detection, and contour fitting.

[0057] S102, merge adjacent shelf strips that meet the preset merging conditions into a shelf group.

[0058] Specifically, back-to-back shelf layouts are common in warehouse scenarios, where two shelf strips are close to each other in an image with only a narrow gap in between. This gap appears as a blank area on a pixel scale, but in actual warehouses, it often does not meet the conditions for robot passage. If this narrow gap is directly regarded as a free area, it is easy to be misidentified as a passage in subsequent passage extraction.

[0059] To address this, this embodiment introduces a back-to-back shelf merging mechanism. The principle behind this mechanism is based on the structural prior of back-to-back shelves in the warehouse layout: for two adjacent shelf strips, the lateral distance between them and their projected overlap rate along the length of the shelf strip are calculated. When the distance between adjacent shelf strips is less than a preset gap threshold and the projected overlap rate is greater than a preset overlap threshold, the two shelf strips are determined to be back-to-back shelves, and they are merged into a shelf group, so that the merged shelf group covers the gap between the adjacent shelf strips.

[0060] Among them, the projection overlap rate It can be represented as:

[0061] Formula 3

[0062] in Let be the overlap length between the i-th shelf strip and the j-th shelf strip along the length direction of the shelf strip. , These represent the lengths of the two shelf strips in that direction.

[0063] For example, such as Figure 6 As shown, suppose two adjacent shelf strips are extracted:

[0064] The left, right, top, and bottom boundaries of the first shelf strip are as follows: Left boundary right boundary upper boundary lower boundary ;

[0065] The left, right, top, and bottom boundaries of the second shelf strip are as follows: right boundary upper boundary lower boundary .

[0066] The horizontal spacing between the two is:

[0067]

[0068] Height of the first shelf strip:

[0069]

[0070] Second shelf strip height

[0071]

[0072] The longitudinal overlap length of the two

[0073]

[0074] Therefore, the vertical overlap rate between the two is:

[0075] .

[0076] If the gap threshold is set to 30 and the overlap rate threshold is set to 0.8, then:

[0077] and

[0078] Therefore, it can be determined that the two shelving strips are back-to-back shelving units, and they can be combined into one shelving group, such as... Figure 6 The results after merging are shown below.

[0079] Optionally, the preset gap threshold can be determined by converting the minimum physical width required for robot passage with the image resolution, that is, the pixel value corresponding to the gap width that does not actually meet the passage conditions is used as the basis for setting the gap threshold; the preset overlap threshold can be set according to the typical alignment degree of back-to-back shelves in the warehouse scenario.

[0080] This step incorporates the actually impassable gaps inside the back-to-back shelving units into the shelving group obstacle, reducing the interference of pseudo-free areas on subsequent aisle extraction from the source. Furthermore, the merged shelving group is used not only for obstacle area generation but also for subsequent aisle position constraints and the division of storage location identification areas, enabling the reuse of the same structural information across multiple stages.

[0081] S103, based on the shelf area and the shelf group, an obstacle area is determined, the area outside the obstacle area is determined as a free area, and candidate channels are extracted based on the free area.

[0082] Specifically, after constructing the shelf group, an obstacle region is generated based on the shelf area, boundary area, and shelf group filling area. The obstacle region represents locations where the robot cannot pass, including: the shelf body, wall boundaries, fence boundaries, and other impassable structures. In practical applications, a preset buffer distance can be added outwards to the obstacle region, and areas less than this buffer distance from obstacles can also be considered impassable, thus providing a safety margin for robot passage. Optionally, the preset buffer distance can be determined based on the robot's dimensions and the required safety clearance. Subsequently, within the analysis area, pixel areas that do not belong to the obstacle region are defined as free regions for subsequent channel extraction.

[0083] The goal of free area analysis is not simply to find all blank areas, but rather to find continuous, width-compliant passageways that conform to warehouse layout patterns. Therefore, this embodiment calculates the free area percentage for each pixel column within the analysis area: for the x-th pixel in the analysis area, its free area percentage qx is the ratio of the number of pixels belonging to the free area in that column to the height Ha of the analysis area.

[0084] Formula 4

[0085] in Indicates whether a pixel belongs to the free region.

[0086] When the proportion of free area in each column of a continuous column interval is greater than the preset proportion threshold, and the width of the continuous column interval meets the preset minimum channel width requirement, the interval is identified as a candidate channel region.

[0087] For example, in an analysis area of ​​x=820 in a certain column, the total pixel height Ha=1000, and the number of pixels in the free region is 900. Then, the proportion of the free region in this column is qx=900 / 1000=0.9. If the proportion threshold is set to 0.85, then this column meets the free region candidate channel condition. Optionally, the preset proportion threshold can be set according to the statistical situation of the proportion of shelf projection and annotation elements in the aisle area of ​​the warehouse image, and the preset minimum aisle width can be determined by converting the minimum physical width required for robot passage with the image resolution.

[0088] Alternatively, in alternative implementations, the centerline of the channel can be determined not only by free region proportion analysis, but also by skeleton extraction, distance transformation, centerline extraction, or Hough line detection.

[0089] S104, a side strip is formed by extending outward from the side boundary of the shelf group as a reference. The candidate channels are screened according to the positional relationship between the center line of the candidate channel and the side strip to obtain the longitudinal channel.

[0090] Specifically, candidate channels extracted solely based on the proportion of free area may not necessarily be channels that truly conform to warehouse operation patterns. Especially in back-to-back racking scenarios, gaps inside the racks, partial blank spaces between racks, and blank areas marked on drawings may all appear as "passable" candidate channels during free area analysis.

[0091] To address this issue, this step introduces shelf sideband constraints. The idea is to leverage the locational patterns of warehouse layouts: the longitudinal aisles of warehouse robots are typically located on the outside of shelf groups, rather than inside them. Specifically, for each shelf group, based on its left and right boundaries, a preset width is extended outwards to form left and right sidebands respectively. The centerline of each candidate aisle is calculated; the centerline can be taken as the average of the left and right boundary positions of the candidate aisle.

[0092] Formula 5

[0093] in , These are the coordinates of the left and right boundaries of the candidate channel, respectively. If the centerline of the candidate channel... If a candidate aisle is located within the left or right sideband of any shelving group, it is retained as a longitudinal aisle; if its centerline is located inside the shelving group, it is rejected.

[0094] For example, such as Figure 7 As shown, assuming a candidate longitudinal channel is obtained through free region analysis, with its left and right boundaries at 770 and 860 respectively, then the centerline... =(770+860) / 2=815. Assuming the right boundary of the merged shelving group is 760, and the side strip width is set to 80, then the right side strip range is [760, 840]. Because... =815 falls within this range. This candidate aisle is located within the right sideband of the shelving group and is retained as a valid longitudinal aisle. If the centerline of another candidate aisle is 700, it is discarded because the lateral range of the merged shelving group is 500~760, and this centerline is located inside the shelving group.

[0095] Optionally, the preset width of the side strip can be statistically set according to the relative positional relationship between the aisle and the shelving group in the warehousing scenario, for example, it can be set to be no less than half the pixel value corresponding to the physical width of the aisle.

[0096] As can be seen, by using sideband constraints, this step is not simply about finding blank paths in the image, but about finding structured channels that conform to the shelf layout and robot passage logic. This effectively avoids misidentifying areas such as gaps inside the shelf and blank areas in the drawings as channels. This is one of the key aspects that distinguishes this solution from ordinary image segmentation and mapping methods.

[0097] S105, based on the overall boundary of each of the shelving groups, a transverse channel is generated outside the overall boundary, and the transverse channel is connected with the longitudinal channel to form a closed-loop channel network.

[0098] In warehousing scenarios, vertical aisles are used by robots to move along the shelf direction, while horizontal aisles connect different vertical aisles to form a complete loop structure. However, the horizontal aisle lines in the original image are often incomplete, making it difficult to reconstruct the complete aisle network solely through image analysis.

[0099] To address this issue, this step automatically generates transverse aisles based on the overall boundary of the shelving group. The concept lies in utilizing the structural correspondence between the overall layout of the shelving group and the main aisles. Specifically, let the overall upper boundary of all shelving groups be... The overall lower boundary is ,but:

[0100] The centerline of the top horizontal channel can be represented as:

[0101] Formula 6

[0102] The centerline of the bottom horizontal channel can be represented as

[0103] Formula 7

[0104] in This is the preset offset distance of the horizontal aisle relative to the overall boundary of the shelving group. Optionally, if there is already a clear reference horizontal route or marked horizontal line in the original warehouse scene image, the generated top and bottom horizontal aisles can be corrected based on this reference position to ensure that the position of the horizontal aisles is consistent with the existing route information in the original image.

[0105] For example, suppose the overall upper boundary of all shelving groups is... =180, Overall Lower Boundary =1320, set the offset distance =100, then the center line of the top horizontal channel =180-100=80, center line of bottom horizontal channel =1320+100=1420, the reconstructed complete channel network is as follows: Figure 8 As shown.

[0106] This step allows for the reconstruction of the main horizontal channels at the top and bottom of the original image, even if the lines are incomplete. The overall boundary structure of the shelving group enables the vertical channels to connect, ensuring a complete closed-loop connectivity of the channel network and reducing channel breakage. Optionally, the preset offset distance can be set based on the actual physical width of the horizontal channel and the image resolution. For cases where a main horizontal channel exists in the middle of the warehouse scene, the proportion of free area can be statistically analyzed along the row direction, identifying continuous pixel row intervals that meet preset proportion and width conditions as the central horizontal channel.

[0107] S106, generate location markers and connection relationships along the channel network, identify storage units within the shelf group area, and generate action points associated with the storage units on the channel network.

[0108] After the channel network is determined, it is necessary to further generate location markers and action points in the robot map. Location markers are used for path navigation and topology representation, while action points are used to indicate the positions where the robot performs actions such as picking up, placing, scanning, or docking.

[0109] For example, sampling is performed along the centerline of each channel at a preset physical interval to generate several location markers. If the map resolution is r and the preset physical sampling interval is D, then the corresponding pixel sampling interval is:

[0110] Formula 8

[0111] Simultaneously, the intersections between channels are determined based on the geometric intersections of the channel centerlines, and additional position markers are generated at these intersections for robot turning. Subsequently, adjacent position markers on the same channel are connected sequentially to form path connections; to meet the robot's bidirectional passage requirements, each connection can generate two curves, one in the forward direction and one in the reverse direction.

[0112] For example, if the map resolution r = 0.02 meters / pixel and the preset physical sampling interval D = 1.0 meter, then the pixel sampling interval P = 1.0 / 0.02 = 50 pixels. If the vertical range of a certain vertical channel is y = 100 to y = 1400, then a location marker point is generated every 50 pixels, in the order of (815, 100), (815, 150), (815, 200)... and so on.

[0113] In addition to obstacles and passageways, warehouse maps also need to identify storage location units. Storage location units are typically represented in images as repeating small squares, cross boxes, cell structures, or regularly arranged small rectangles within shelving units. This embodiment preferably performs storage location identification within the shelving group area, rather than performing a global search across the entire image. This reduces irrelevant interference and improves the effectiveness of candidate boxes. Specifically, edge detection and contour extraction are performed on the image within the shelving group area to obtain multiple storage location candidate boxes. These boxes are then filtered based on their width, height, area, and aspect ratio, retaining only those that match the geometric characteristics of the storage location as storage location units. Optionally, the filtering range for width, height, area, and aspect ratio can be set according to the actual physical dimensions of the storage location unit and the image resolution.

[0114] Furthermore, for back-to-back shelving units, a dual-sided identification method is preferred: first, determine the central axis of the shelving unit, then divide the unit into side areas on either side of the central axis, and extract and filter candidate storage locations within each side area. This avoids missed detections caused by treating the entire back-to-back shelving unit as a single area, thus improving the stability of dual-sided storage location identification.

[0115] The motion point is determined based on the identified location of the storage unit: the storage unit is mapped to the center line of the nearest aisle, and the corresponding point mapped to the center line of the aisle is set as the motion point, so that the motion point is located near the position where the robot can perform storage operations.

[0116] After obtaining the storage location unit and the action point, establish the mapping relationship between them: for the i-th storage location frame, its center point Average values ​​of the left and right boundaries and the top and bottom boundaries of the accessible storage location frame:

[0117] Formula 9

[0118] Calculate the distance between the center point and each action point aj=(xj,yj):

[0119] Formula 10

[0120] in,( , The coordinates of the center point of the storage location are shown in the diagram. The action point with the smallest distance is selected as the associated action point of this storage location unit. Preferably, a same-side constraint can also be added, that is, among the action points located on the same shelf side as the storage location unit, the one with the smallest distance is selected, so that the correspondence between storage locations and action points is more in line with the actual operation logic.

[0121] For example, suppose the left, right, top, and bottom boundaries of a identified storage location are 530, 570, 400, and 440 respectively. Then the pixel coordinates of the storage location's center point are (550, 420). Assuming there are action points 1 (815, 400) and 2 (815, 450) nearby, the distance between the storage location's center point and action point 1 is approximately 265.75, and the distance between the center point and action point 2 is approximately 266.69. Therefore, action point 1 is taken as the associated action point for this storage location.

[0122] This step automatically establishes a correspondence between the location business information and the navigation topology, eliminating the need for manual input of locations and configuration of work points. The generated map has both path navigation and warehousing operation capabilities.

[0123] Alternatively, in alternative implementations, in addition to edge detection and candidate box filtering, location identification can also be achieved using template matching, corner detection, or deep learning object detection; the mapping between locations and action points can be achieved not only using nearest neighbor distance mapping, but also combining shelf side constraints, path topology distance, orientation constraints, or task rules.

[0124] S107, the set of obstacle points extracted based on the obstacle area, the location markers, the action points, the connection relationships, and the association between the cargo space unit and the action points are uniformly converted to the map coordinate system to generate a point cloud topology map.

[0125] Specifically, after mapping obstacle areas, passageways, location markers, action points, and cargo locations, it is necessary to uniformly transform the pixel coordinates in the image to the map coordinate system. For example, let the pixel coordinates be (u, v), the map resolution be r, and the map origin be (u, v). , Then the corresponding world coordinates on the map are:

[0126] Formula 11

[0127] Formula 12

[0128] The map resolution can be determined by using reference objects with known physical dimensions in the image (such as standard shelf dimensions or image scale), and the map origin can be set according to the correspondence between the warehouse site coordinate system and the image.

[0129] For example, if the map origin is (-10, 20) and the map resolution is r = 0.02 m / pixel, then for the center point of the cargo location (550, 420), its world map coordinates are x = -10 + 550 × 0.02 = 1 and y = 20 - 420 × 0.02 = 11.6, which is (1, 11.6).

[0130] Through the coordinate transformation described above, the obstacle point set, location marker points, action points, and cargo location outline points are uniformly transformed into the point cloud topology map coordinate system. For obstacle areas, the obstacle point set can be extracted according to a preset sampling step size; then, the various structured results are semantically organized and topologically generated: the obstacle point set is organized into an obstacle point list, the location marker points are organized into a location point list, the action points are organized into an action point list, the mapping relationship between cargo locations and action points is organized into a cargo location list, and a connection relationship list and a curve relationship list are generated.

[0131] For example, if the world coordinates of the center point of a cargo location are (1, 11.6), and its associated action point number is 1, then the corresponding cargo location record can be represented as BIN=(1,1,11.6,1); if two adjacent location markers are LM1=(1,11.6) and LM2=(1,10.6), then the connection relationship E=(LM1,LM2) can be generated. After unified verification, completion, and structured encapsulation, the final output point cloud topology map file includes at least the map header information, obstacle point list, location point list, action point list, connection relationship list, curve relationship list, and cargo location list.

[0132] Through this step, the visual information in the warehouse scene images is automatically converted into a structured map result that can be directly parsed and used by the robot system, realizing end-to-end generation from static images to deployable point cloud topology maps.

[0133] Alternatively, in alternative implementations, in addition to outputting point cloud topology map files, other structured map file formats suitable for robot navigation and warehouse scheduling systems can also be output.

[0134] On the other hand, in some application scenarios, in addition to the main warehouse scene image, reference images with route annotations may also be provided. The route information in these images reflects the user's existing channel planning intentions and can be used to assist in correcting the automatically generated channel positions.

[0135] Specifically, for example, extracting the reference channel centerline from the reference route image, and determining the automatically generated channel centerline corresponding to the reference channel centerline; let the reference channel centerline be... The corresponding automatically generated channel centerline is The corrected channel centerline can then be obtained through a weighted average:

[0136] Formula 13

[0137] in This refers to the weighting coefficient. Optionally, the weighting coefficient can be determined based on the correspondence between the reference channel centerline and the automatically generated channel centerline, according to the principle of closest distance between centerlines and consistency of direction. The settings can be adjusted based on the emphasis placed on automatic recognition results and human planning intentions. This approach preserves the structured results of automatic recognition while incorporating location information from reference routes, ensuring that the final generated map conforms to both the actual shelving layout in the warehouse image and the human planning intentions.

[0138] On the other hand, corresponding to the above method examples, such as Figure 9As shown, the present invention also provides a point cloud topology map generation system, an example of which includes: an image parsing module, a shelf group construction module, a passage reconstruction module, and a map generation module. The image parsing module is used to perform step S101, the shelf group construction module is used to perform step S102, the passage reconstruction module is used to perform steps S103 to S105, and the map generation module is used to perform steps S106 to S107. The specific implementation methods of each module and the resulting technical effects can be found in the corresponding descriptions in the foregoing method embodiments, and will not be repeated here.

[0139] In summary, the point cloud topology map generation method and system provided by this invention utilizes the structural patterns of shelf layout in a warehouse scenario to impose progressive constraints on the map generation process, rather than simply determining passable spaces based on blank areas in the image: by merging adjacent shelf strips that meet preset merging conditions into shelf groups, gaps between shelves that do not actually allow passage are classified as obstacles, reducing interference from pseudo-free areas from the source; by constraining the position of candidate channels with side strips formed by extending outward from the side boundaries of shelf groups, channel recognition changes from searching for blank areas to searching for structured channels that conform to shelf layout and robot passage logic; by reconstructing lateral channels on the outside based on the overall boundary of the shelf group, closed-loop connectivity of the channel network is ensured; and by limiting the identification of storage locations to the area of ​​the shelf group and generating associated action points on the channel network, a correspondence is automatically established between warehouse operation information and navigation topology.

[0140] Based on the above design, this invention effectively reduces the situation where impassable areas such as shelf gaps are misidentified as passageways, as well as the problem of passageway breaks. The location information can be incorporated into the map structure without manual input. The final output is a point cloud topology map containing obstacle points, location markers, action points, connection relationships, curve relationships, and a list of locations. This map can be directly used by mobile robots for positioning, navigation, scheduling, and picking and placing operations. It realizes the automatic generation of a structured map usable by robots from warehouse scene images, reduces the cost of manual mapping, shortens the deployment cycle, and the generation process is driven by unified rules, which improves the efficiency of map generation and the consistency of results.

[0141] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only by the claims and their full scope and equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

[0142] Those skilled in the art will understand that, besides implementing the system, apparatus, unit, and its modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and its modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0143] Furthermore, all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0144] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. A method for generating a point cloud topology map, characterized in that the steps include: include: The acquired warehouse scene images are analyzed to identify shelf areas and extract multiple shelf strips from them; Merge adjacent shelf strips that meet the preset merging conditions into a shelf group; Based on the shelf area and the shelf group, an obstacle area is determined, the area outside the obstacle area is determined as a free area, and candidate channels are extracted based on the free area; A side strip is formed by extending outward from the side boundary of the shelving group. The candidate channels are screened according to the positional relationship between the center line of the candidate channel and the side strip to obtain the longitudinal channel. Based on the overall boundary of each of the aforementioned shelving groups, a transverse channel is generated on the outside of the overall boundary, and the transverse channel is connected with the longitudinal channel to form a closed-loop channel network. Location markers and connections are generated along the channel network, storage units are identified within the shelving group area, and action points associated with the storage units are generated on the channel network. The set of obstacle points extracted based on the obstacle area, the location markers, the action points, the connection relationships, and the association between the cargo space unit and the action points are uniformly converted to the map coordinate system to generate a point cloud topology map.

2. The point cloud topology map generation method according to claim 1, characterized in that, The preset merging conditions include: the spacing between adjacent shelf strips is less than a preset gap threshold, and the projection overlap rate of adjacent shelf strips in the length direction of the shelf strips is greater than a preset overlap threshold; the shelf group covers the gap between the adjacent shelf strips; the projection overlap rate is the ratio of the smaller of the overlap length of the adjacent shelf strips in the length direction and the length of the adjacent shelf strips in the length direction.

3. The point cloud topology map generation method according to claim 1, characterized in that, The step of filtering the candidate channels to obtain the vertical channels includes: Candidate channels with their center lines located within the side strips are retained as the longitudinal channels, while candidate channels with their center lines located within the shelving group are eliminated. The side strips include a left side strip and a right side strip formed by extending outwards by a preset width based on the left and right boundaries of the shelf group; the center line is the average value of the positions of the two boundary positions of the candidate channel.

4. The point cloud topology map generation method according to claim 1, characterized in that, The step of generating a transverse aisle outside the overall boundary based on the overall boundary of each of the shelving groups includes: Determine the overall upper boundary and overall lower boundary of each of the aforementioned shelving groups, and generate a top transverse channel and a bottom transverse channel at positions offset by a preset distance from the outer side of the overall upper boundary and the overall lower boundary, respectively.

5. The point cloud topology map generation method according to claim 1, characterized in that, The step of extracting candidate channels based on the free region includes: The percentage of pixels belonging to the free region in each pixel column is counted. The consecutive pixel column intervals in which the percentage of each pixel is greater than a preset percentage threshold and the width meets the preset minimum channel width are determined as candidate channels.

6. The point cloud topology map generation method according to claim 1, characterized in that, The step of identifying the storage location unit within the shelving group area includes: Determine the central axis of the shelving group, divide the shelving group into side areas on both sides of the central axis, extract the candidate storage location frames in each side area, and filter the candidate storage location frames according to preset geometric conditions to obtain storage location units.

7. The point cloud topology map generation method according to claim 1, characterized in that, The steps for generating action points associated with the storage unit on the channel network include: Map the storage unit to the nearest aisle centerline, and determine the corresponding point mapped to the aisle centerline as the action point; Calculate the distance between the center point of the storage unit and each action point. Among the action points located on the same shelf side as the storage unit, select the action point with the smallest distance as the associated action point of the storage unit.

8. The point cloud topology map generation method according to claim 1, characterized in that, The steps also include: Obtain a reference image containing the planned route, and extract the centerline of the reference channel from the reference image; Determine the automatically generated channel centerline corresponding to the reference channel centerline, and then weight and fuse the reference channel centerline and the corresponding channel centerline according to a preset weight to obtain the corrected channel centerline.

9. A point cloud topology map generation system, characterized in that, include: The image analysis module is used to perform region analysis on the acquired warehouse scene images, identify the shelf areas, and extract multiple shelf strips from them; The shelving group building module is used to merge adjacent shelving strips that meet preset merging conditions into a shelving group; The aisle reconstruction module is used to determine the obstacle area based on the shelf area and the shelf group, define the area outside the obstacle area as the free area, and extract candidate aisles based on the free area; expand outward from the side boundary of the shelf group to form a side strip, filter the candidate aisles according to the positional relationship between the centerline of the candidate aisle and the side strip to obtain the longitudinal aisle; generate transverse aisles outside the overall boundary of each shelf group, and connect the transverse aisles with the longitudinal aisles to form a closed-loop aisle network; The map generation module is used to generate location markers and connection relationships along the channel network, identify storage units within the shelf group area, and generate action points associated with the storage units on the channel network; and uniformly convert the obstacle point set extracted based on the obstacle area, the location markers, the action points, the connection relationships, and the association between the storage units and the action points to the map coordinate system to generate a point cloud topology map.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the point cloud topology map generation method as described in any one of claims 1 to 8.

Images