Carton recognition method suitable for unloading scene
By combining data processing of RGB and depth images with Canny edge detection and Hough transform, the problems of accuracy and robustness in carton recognition under complex unloading scenarios are solved, and efficient carton position and orientation calculation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONGCHU HENGKE INTERNET OF THINGS SYST CO LTD
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-17
AI Technical Summary
Existing technologies struggle to achieve high-precision, robust, and real-time cardboard box recognition in complex unloading scenarios. AI-based methods are susceptible to ambient lighting and occlusion, while traditional visual algorithms lack robustness when texture features are weakened.
The data is processed by combining RGB and depth images. The rectangular outline of the cardboard box is extracted by Canny edge detection and Hough transform. The image detection box and the binary outline image are fused together to determine the final position and orientation of the cardboard box.
It achieves high-precision and robust carton recognition in complex unloading environments, reduces redundant calculations, and ensures the continuity and smoothness of unloading operations.
Smart Images
Figure CN121883591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cardboard box recognition technology, specifically a cardboard box recognition method applicable to unloading scenarios. Background Technology
[0002] In automated logistics and intelligent warehousing systems, the automation and intelligence of truck unloading are key to improving overall supply chain efficiency. In this scenario, quickly and accurately identifying stacked or arranged cartons inside the truck and calculating the position and orientation of each carton in three-dimensional space in real time is the prerequisite and foundation for guiding subsequent gripping robots or robotic arms to perform precise unloading operations.
[0003] Currently, pose calculation solutions applied to such scenarios mainly rely on two types of technical paths: one is based on artificial intelligence image recognition and deep learning models, such as using object detection or instance segmentation networks to directly infer the two-dimensional boundary or mask of the cardboard box from the image, and then combining it with point cloud data provided by depth sensors for clustering and three-dimensional reconstruction to calculate the pose; the other focuses on traditional computer vision algorithms, such as locating the cardboard box from the image through edge detection, corner extraction, geometric shape matching, etc., and then fusing it with depth information to recover its three-dimensional structure.
[0004] However, in the complex unloading environment, both of the above-mentioned single technical paths have obvious limitations: Although AI-based methods have strong semantic understanding and generalization capabilities, their performance is greatly affected by training data, ambient lighting, occlusion, and printed patterns on the carton surface, making them prone to missed detections, false detections, or inaccurate boundary positioning, which directly affects the accuracy of subsequent point cloud matching and pose estimation; while methods that rely purely on traditional vision algorithms often suffer from insufficient robustness in recognition due to unstable feature extraction when goods are tightly stacked, the background is cluttered, or the texture features of the carton are weakened, and they are usually computationally inefficient and difficult to meet real-time requirements.
[0005] Therefore, existing technologies still lack a carton recognition method that can achieve high accuracy, high robustness, and high real-time performance in complex unloading scenarios. Summary of the Invention
[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this application provides a carton recognition method suitable for unloading scenarios.
[0007] (II) Technical Solution To address the above problems, this application provides the following technical solution: A method for recognizing cardboard boxes in unloading scenarios includes: Acquire the RGB image and corresponding depth image of the area to be unloaded; Data calculations are performed on the RGB and depth images to obtain a set of carton poses; The carton pose set is filtered based on the depth of the carton to obtain the nearest unloading face pose set and the image detection box set; The rectangular outline of the carton in the nearest unloading face pose set is extracted by Canny edge detection and Hough transform to obtain a binary outline image; The image detection box set is compared and fused with the contour binary image to obtain the final valid cardboard box, and the final position and orientation of the cardboard box are calculated.
[0008] Preferably, the acquisition of the RGB image and the corresponding depth image of the area to be unloaded specifically includes: acquiring the RGB image and the corresponding depth image of the area to be unloaded using an RGBD camera.
[0009] Preferably, the step of performing data calculation based on the RGB image and depth image to obtain the carton pose set specifically includes: Establish a vehicle coordinate system, where the X-axis points forward along the direction of the vehicle body, the Y-axis points to the left, and the Z-axis points upward; The Yolov8 Seg algorithm is used to segment the cartons in the RGB image to obtain a set of segmented pixels for each carton. A 3D point cloud is generated based on the depth image; The pixel set is segmented based on the instances of the cardboard boxes, and a subset of the 3D point cloud corresponding to the cardboard box is extracted from the 3D point cloud. Plane fitting, normal vector analysis, and principal component analysis are performed on the subset of the 3D point cloud, and the pose angle and center coordinates are calculated by combining the long side judgment, resulting in the cardboard box pose set S for all cardboard boxes. all .
[0010] Preferably, the step of filtering the carton pose set based on the depth of the carton to obtain the nearest unloading face pose set and the image detection box set specifically includes: For the carton pose set S all Classify along the X-axis, based on carton depth L d By sliding the step size along the positive X-axis, we find the minimum x-coordinate and obtain the classification set S. step ; Calculate the classification set S step The average x-coordinate of the cartons within each category interval is used to determine the minimum average x-coordinate as the distance L between the nearest unloading surface and the unloading robot. front ; From the classification set S step During the screening, if the carton orientation meets the requirements... Then, the set of poses of the nearest unloading surface S is formed. filterSimultaneously, the 2D detection image boxes corresponding to the poses are combined to form the image detection box set P. filter .
[0011] Preferably, the step of extracting the rectangular contour of the cardboard box from the nearest unloading surface pose set through Canny edge detection and Hough transform, and obtaining a binary contour image by valid contour determination, specifically includes: For the nearest unloading face pose set S filter Perform plane fitting to obtain the fitted plane SP Plan ; Traverse all the aforementioned 3D point clouds, retaining the points' paths to the fitting plane SP. Plan Points with a distance less than 0.02 are used to obtain the point cloud set SS. plan Simultaneously record SS plan Pix is a set of pixel coordinate values corresponding to each point. plan ; Retain Pix plan Take the corresponding image pixels, set the remaining pixels to 0, and generate a new image I. new ; For the new image I new Perform Canny edge detection; Performing Hough transform on the edge image to detect straight lines specifically includes: Use the probabilistic Hough transform to detect straight lines in the image; calculate the angle of each line, retain the lines with angles in the range of 0°±5° or 90°±5°, and group them into horizontal line groups and vertical line groups according to their angles. Within each group, perform distance clustering on the straight lines and retain the 2-4 longest straight lines as candidate carton edges; Select two straight lines from the horizontal line group and two straight lines from the vertical line group, and calculate the intersection of the four lines. If the four vertices satisfy the following conditions: the difference in length of opposite sides < 10%, the difference in length of diagonals < 5%, and the error between the interior angle and 90° < 8°, then the four vertices constitute the rectangular outline of the cardboard box. Filter the rectangular outlines to generate the final valid carton outline; Set the image pixel value of the final effective carton outline to 255, and set the remaining pixels to 0 to obtain a binary outline image I containing the rectangular outline of the carton. line .
[0012] Preferably, the step of filtering the rectangular outline to generate the final valid carton outline specifically includes: Calculate the bounding rectangle of the rectangular outline. If the aspect ratio of the bounding rectangle is between 0.5 and 2.0, and the pixel area inside the rectangular outline accounts for more than 80% of the area of the bounding rectangle, it is determined to be a candidate carton outline. If multiple overlapping contours are detected in the same carton, calculate the intersection over union (IOU) between the candidate carton contours. If the candidate carton contour with the largest area among those with IOU greater than 0.7 has an edge point ratio greater than 90%, it is determined as the final valid carton contour.
[0013] Preferably, the comparison, fusion, and determination of the image detection frame set and the contour binary image are performed to obtain the final valid carton, and the final position and pose of the carton are calculated. Specifically, it includes: Project the image detection frame set P filter onto the contour binary image I line , and calculate the intersection over union (IOU) between each 2D detection frame in the image detection frame set P filter and the corresponding region contour in the contour binary image I line ; If IOU≥0.8, it is determined as the final valid carton box; If IOU≤0.3, perform geometric verification on the contour binary image I line . If the rectangularity is greater than or equal to 0.8 and the side lengths are within the carton size range, it is determined as the final valid carton box; If 0.3 < IOU < 0.8, calculate the number of the final valid carton contours within the 2D detection frame in the image detection frame set P filter . If there are more than 2 independent final valid carton contours within the 2D detection frame and the distance between the contours is greater than 1.5 times the carton thickness, it is determined as a multi-carton misdetection; split the independent sub-detection frames according to the spatial arrangement of the contours and replace the 2D detection frames in the image detection frame set P filter ; recalculate the IOU and determine whether to mark it as the final valid carton box according to the above steps; Perform clustering analysis on the point cloud inside the rectangle corresponding to the final valid carton box, and calculate the position and pose corresponding to the final valid carton box.
[0014] (III) Beneficial effects Compared with the prior art, the present application provides a carton recognition method applicable to the unloading scenario, having the following beneficial effects: 1. This method realizes the complementary advantages and cross-verification through the dual criteria of deep integration of AI instance segmentation and traditional geometric vision algorithms; the AI model provides global and semantic-level initial perception, effectively coping with texture changes and partial occlusions; the traditional algorithm performs fine geometric contour extraction and verification within the high-confidence region defined by the AI, which can effectively correct the missed detections and misdetections of the AI and accurately segment the stacked and adhered cartons; this fusion decision-making mechanism greatly enhances the stability and reliability of the system in complex unloading environments.
[0015] 2. This method innovatively introduces a dynamic filtering mechanism of "nearest unloading surface"; the system no longer processes the entire cargo compartment scene indiscriminately, but intelligently identifies and focuses on the cargo layer that most needs to be unloaded, so that subsequent computationally intensive traditional image processing and point cloud processing only need to be performed on the filtered, highly relevant local data, thereby greatly reducing redundant calculations; enabling the system to achieve high frame rate real-time processing even on limited embedded or industrial control computer computing platforms, effectively ensuring the continuity and smoothness of unloading operations.
[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic diagram of a carton recognition method applicable to unloading scenarios according to this application; Figure 2 This is a schematic diagram illustrating the principle of calculating the carton pose of a carton recognition method applicable to unloading scenarios according to this application. Detailed Implementation
[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] The terms "first" and "second" in the specification and claims of this application may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise stated, "multiple" means two or more. Furthermore, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0020] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.
[0021] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.
[0022] Please see Figures 1-2 This application provides a new technical solution: a method for recognizing cardboard boxes in unloading scenarios, comprising: Acquire the RGB image and corresponding depth image of the area to be unloaded; Data calculations are performed on the RGB and depth images to obtain a set of carton poses; The carton pose set is filtered based on the depth of the carton to obtain the nearest unloading face pose set and the image detection box set; The rectangular outline of the carton in the nearest unloading face pose set is extracted by Canny edge detection and Hough transform to obtain a binary outline image; The image detection box set is compared and fused with the contour binary image to obtain the final valid cardboard box, and the final position and orientation of the cardboard box are calculated.
[0023] In this invention, the acquisition of the RGB image and the corresponding depth image of the area to be unloaded specifically includes: acquiring the RGB image and the corresponding depth image of the area to be unloaded using an RGBD camera.
[0024] In this invention, the step of performing data processing on the RGB image and depth image to obtain a carton pose set specifically includes: Establish a vehicle coordinate system, where the X-axis points forward along the direction of the vehicle body, the Y-axis points to the left, and the Z-axis points upward; The Yolov8 Seg algorithm is used to segment the cartons in the RGB image to obtain a set of segmented pixels for each carton. A 3D point cloud is generated based on the depth image; The pixel set is segmented based on the instances of the cardboard boxes, and a subset of the 3D point cloud corresponding to the cardboard box is extracted from the 3D point cloud. Plane fitting, normal vector analysis, and principal component analysis are performed on the subset of the 3D point cloud, and the pose angle and center coordinates are calculated by combining the long side judgment, resulting in the cardboard box pose set S for all cardboard boxes. all .
[0025] In this invention, the step of filtering the carton pose set based on the depth of the carton to obtain the nearest unloading face pose set and the image detection box set specifically includes: For the carton pose set S all Classify along the X-axis, based on carton depth L d By sliding the step size along the positive X-axis, we find the minimum x-coordinate and obtain the classification set S. step ; Calculate the classification set S step The average x-coordinate of the cartons within each category interval is used to determine the minimum average x-coordinate as the distance L between the nearest unloading surface and the unloading robot. front ; From the classification set S step During the screening, if the carton orientation meets the requirements... Then, the set of poses of the nearest unloading surface S is formed. filter Simultaneously, the 2D detection image boxes corresponding to the poses are combined to form the image detection box set P. filter .
[0026] In this invention, the step of extracting the rectangular contour of the cardboard box from the nearest unloading surface pose set through Canny edge detection and Hough transform, and obtaining a binary contour image by valid contour determination, specifically includes: For the nearest unloading face pose set S filter Perform plane fitting to obtain the fitted plane SP Plan ; Traverse all the aforementioned 3D point clouds, retaining the points' paths to the fitting plane SP. Plan Points with a distance less than 0.02 are used to obtain the point cloud set SS. plan Simultaneously record SS plan Pix is a set of pixel coordinate values corresponding to each point. plan ; Retain Pix plan Take the corresponding image pixels, set the remaining pixels to 0, and generate a new image I. new ; For the new image I newCanny edge detection is performed, specifically including: The gradient magnitude and direction are calculated using the Sobel operator; edge points are preserved by improving nonmaximum suppression and combining neighborhood gradient consistency judgment. Connect broken edges by analyzing 8 connected components, and connect two edge points when the distance between them is less than 5 pixels and they are in the same direction. Morphological closing operations of 3×3 rectangular structural elements are used to fill the edge gaps, and opening operations are used to remove isolated noise points to obtain a continuous and complete edge map. Performing Hough transform on the edge image to detect straight lines specifically includes: Use the probabilistic Hough transform to detect straight lines in the image; calculate the angle of each line, retain the lines with angles in the range of 0°±5° or 90°±5°, and group them into horizontal line groups and vertical line groups according to their angles. Within each group, perform distance clustering on the straight lines and retain the 2-4 longest straight lines as candidate carton edges; Select two straight lines from the horizontal line group and two straight lines from the vertical line group, and calculate the intersection of the four lines. If the four vertices satisfy the following conditions: the difference in length of opposite sides < 10%, the difference in length of diagonals < 5%, and the error between the interior angle and 90° < 8°, then the four vertices constitute the rectangular outline of the cardboard box. Filter the rectangular outlines to generate the final valid carton outline; Set the image pixel value of the final effective carton outline to 255, and set the remaining pixels to 0 to obtain a binary outline image I containing the rectangular outline of the carton. line .
[0027] In this invention, the step of filtering the rectangular outline to generate the final valid carton outline specifically includes: Calculate the bounding rectangle of the rectangular outline. If the aspect ratio of the bounding rectangle is between 0.5 and 2.0, and the pixel area inside the rectangular outline accounts for more than 80% of the area of the bounding rectangle, it is determined to be a candidate carton outline. If multiple overlapping contours are detected for the same carton, the Intersection over Union (IOU) ratio between candidate carton contours is calculated. If the candidate carton contour with an IOU greater than 0.7 has the largest area and the proportion of edge points is greater than 90%, it is determined to be the final valid carton contour.
[0028] In this invention, the step of comparing and fusing the image detection box set with the contour binary image to obtain the final valid cardboard box, and calculating the final position and orientation of the cardboard box, specifically includes: The image detection box set P filter Projected onto the binary contour image I line Calculate the image detection box set P filterThe intersection over union of each 2D detection box in line with the corresponding regional contour in the binary contour image I; If IOU≥0.8, it is determined as the final valid carton box; If IOU≤0.3, geometric verification is performed on the binary contour image I line If the rectangularity is greater than or equal to 0.8 and the side lengths are within the carton size range, it is determined as the final valid carton box; If 0.3 < IOU < 0.8, calculate the number of the final valid carton contours within the 2D detection boxes in the image detection box set P filter If there are more than 2 independent final valid carton contours within the 2D detection box and the distance between the contours is greater than 1.5 times the carton thickness, it is determined as a multi - carton mis - detection; split out independent sub - detection boxes according to the spatial arrangement of the contours and replace the 2D detection boxes in the image detection box set P filter Recalculate the IOU and determine whether to mark it as the final valid carton box according to the above steps; Perform clustering analysis on the point cloud inside the rectangular box corresponding to the final valid carton box, and solve the position and pose corresponding to the final valid carton box.
[0029] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "schematic embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.
[0030] Although the embodiments of this application have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of this application. The scope of this application is defined by the appended claims and their equivalents.
Claims
1. A method for recognizing cardboard boxes in unloading scenarios, characterized in that, include: Acquire the RGB image and corresponding depth image of the area to be unloaded; Data calculations are performed on the RGB and depth images to obtain a set of carton poses; The carton pose set is filtered based on the depth of the carton to obtain the nearest unloading face pose set and the image detection box set; The rectangular outline of the carton in the nearest unloading face pose set is extracted by Canny edge detection and Hough transform to obtain a binary outline image; The image detection box set is compared and fused with the contour binary image to obtain the final valid cardboard box, and the final position and orientation of the cardboard box are calculated.
2. The carton identification method for unloading scene according to claim 1, characterized in that, The acquisition of the RGB image and corresponding depth image of the area to be unloaded specifically includes: acquiring the RGB image and corresponding depth image of the area to be unloaded using an RGBD camera.
3. The method for identifying cardboard boxes in unloading scenarios according to claim 1, characterized in that, The step of performing data calculation based on the RGB image and depth image to obtain the carton pose set specifically includes: Establish a vehicle coordinate system, where the X-axis points forward along the direction of the vehicle body, the Y-axis points to the left, and the Z-axis points upward; The Yolov8 Seg algorithm is used to segment the cartons in the RGB image to obtain a set of segmented pixels for each carton. A 3D point cloud is generated based on the depth image; According to the example of the carton, a three-dimensional point cloud subset of a corresponding carton is extracted from the three-dimensional point cloud; a plane fitting, a normal vector analysis, and a principal component analysis are performed on the three-dimensional point cloud subset, a long side judgment is combined to complete a pose angle and a center coordinate calculation, and a carton pose set S of all cartons is obtained all .
4. The carton identification method for unloading scene according to claim 3, characterized in that, The process of filtering the carton pose set based on the carton's depth to obtain the nearest unloading face pose set and image detection box set specifically includes: To the set of carton poses S all Sort along the X-axis direction, with the carton depth L d Slide to the positive direction of the X-axis with a step size, find the minimum x-coordinate, and get the sorted set S step ; Calculate the classification set S step The average value of the x coordinate of the cartons in each classification interval, and the smallest average value of the x coordinate is determined as the distance L between the nearest unloading surface and the unloading robot front ; From the classification set S step During the screening, if the carton orientation meets the requirements... Then, the set of poses of the nearest unloading surface S is formed. filter Simultaneously, the 2D detection image boxes corresponding to the poses are combined to form the image detection box set P. filter .
5. The carton identification method for use in a de-stacking scenario according to claim 4, wherein, The process of extracting the rectangular contour of the cardboard box from the nearest unloading face pose set using Canny edge detection and Hough transform, and then performing effective contour determination to obtain a binary contour image, specifically includes: to the set of poses S of the most recent unloading face position filter performing a plane fitting to obtain a fitted plane SP Plan ; Traverse all the aforementioned 3D point clouds, retaining the points' paths to the fitting plane SP. Plan Points with a distance less than 0.02 are used to obtain the point cloud set SS. plan Simultaneously record SS plan Pix is a set of pixel coordinate values corresponding to each point. plan ; Retain Pix plan Take the corresponding image pixels, set the remaining pixels to 0, and generate a new image I. new ; performing Canny edge detection on the new image I new performing Canny edge detection; Performing Hough transform on the edge image to detect straight lines specifically includes: Use the probabilistic Hough transform to detect straight lines in the image; calculate the angle of each line, retain the lines with angles in the range of 0°±5° or 90°±5°, and group them into horizontal line groups and vertical line groups according to their angles. Within each group, perform distance clustering on the straight lines and retain the 2-4 longest straight lines as candidate carton edges; Select two straight lines from the horizontal line group and two straight lines from the vertical line group, and calculate the intersection of the four lines. If the four vertices satisfy the following conditions: the difference in length of opposite sides < 10%, the difference in length of diagonals < 5%, and the error between the interior angle and 90° < 8°, then the four vertices constitute the rectangular outline of the cardboard box. Filter the rectangular outlines to generate the final valid carton outline; The image pixels where the final effective carton contour is located are set to 255, and the rest of the pixels are set to 0, to obtain a contour binary image I containing the carton rectangular contour line .
6. The carton identification method for use in a de-stacking scenario according to claim 5, wherein, The process of filtering the rectangular outline to generate the final valid carton outline specifically includes: Calculate the bounding rectangle of the rectangular outline. If the aspect ratio of the bounding rectangle is between 0.5 and 2.0, and the pixel area inside the rectangular outline accounts for more than 80% of the area of the bounding rectangle, it is determined to be a candidate carton outline. If multiple overlapping contours are detected for the same carton, the Intersection over Union (IOU) ratio between candidate carton contours is calculated. If the candidate carton contour with an IOU greater than 0.7 has the largest area and the proportion of edge points is greater than 90%, it is determined to be the final valid carton contour.
7. The carton identification method for use in a de-stacking scenario according to claim 5, wherein, The step of comparing and fusing the image detection box set with the contour binary image to obtain the final valid cardboard box, and calculating the final position and orientation of the cardboard box, specifically includes: detecting a set of image bounding boxes P filter projecting to the contour binary image I line , calculating the intersection over union of each 2D bounding box in the set of image bounding boxes P filter and the contour of the corresponding region in the contour binary image I line ; If IOU ≥ 0.8, it is determined to be a valid cardboard box. If IOU ≤ 0.3, then for the binary image I of the contour line Perform geometric verification; if the rectangularity is greater than or equal to 0.8 and the side length is within the carton size range, it is determined to be a final valid carton. If 0.3 < IOU < 0.8, calculate the number of the final effective carton contours within the 2D detection boxes in the set P of image detection boxes filter If there are more than two independent final effective carton contours within the 2D detection box, and the distance between the contours is greater than 1.5 times the carton thickness, it is determined as a multi-box misdetection; split out independent sub-detection boxes according to the spatial arrangement of the contours, and replace the 2D detection boxes in the set P of image detection boxes filter in the 2D detection boxes; calculate the IOU again, and determine whether to label it as the final effective carton box according to the above steps Cluster analysis is performed on the point cloud inside the rectangle corresponding to the final valid cardboard box to calculate the position and pose of the final valid cardboard box.