Robot unstacking method and related device
By acquiring the depth image to generate normal vector images and calculating the gradient image, the box contour lines are extracted, which solves the problem of weak light and similar texture in robot destanding, and achieves high-precision destanding operation.
Patent Information
- Application Number
- CN202510481239.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-07-04
AI Technical Summary
In the prior art, when robots destand, due to relying on 2D color RGB images, the light is dim or the texture of adjacent cartons is close, resulting in poor contour recognition effect, which reduces the success rate of destanding.
A 3D vision camera is used to acquire depth images, generate normal vector images, and calculate gradient images, thereby extracting the box target contour lines and controlling the robot to destack.
It improves the success rate of robot destacking, especially in dark light conditions, and has high contour recognition accuracy.
Smart Images
Figure CN120244965A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of robot control, and particularly to a method for robot palletizing and related devices. Background Art
[0002] In the prior art, the contour recognition method for a robot to depalletize boxes in a stack uses a 2D color RGB image as input. The visual image processing algorithm extracts the surface contour of the cardboard box by using feature point extraction, edge detection, etc. However, usually, the visual algorithm based on RGB image input has high requirements for the surface texture features of the cardboard box in the image. However, in the actual factory scenario, the light is dim, resulting in an unsatisfactory imaging of the object, which affects the contour recognition. In addition, if the textures of two adjacent cardboard boxes are similar, it will also greatly affect the contour recognition effect, thereby reducing the success rate of the robot's depalletizing. Summary of the Invention
[0003] This application provides a method for robot depalletizing and related devices, which can improve the success rate of the robot's depalletizing.
[0004] In the first aspect of this application, a method for robot depalletizing is provided. The method includes: obtaining a depth image of the stack to be depalletized; generating a normal vector image of the depth image according to the depth image; determining a gradient image of the normal vector image according to the normal vector image; extracting a target contour line of the surface of each box target in the stack to be depalletized from the gradient image; and controlling the robot to depalletize the stack to be depalletized according to the target contour line of each box.
[0005] In an embodiment, the step of generating a normal vector image of the depth image according to the depth image includes: for each first pixel point in the depth image, obtaining a first depth value of the first pixel point and second depth values of a plurality of second pixel points around the first pixel point, and determining a spatial coordinate of the first pixel point according to the first depth value of the first pixel point, and respectively determining spatial coordinates of the plurality of second pixel points according to the second depth values of the plurality of second pixel points, and then determining a normal vector of the first pixel point according to the spatial coordinate of the first pixel point and the spatial coordinates of the plurality of second pixel points; and generating the normal vector map according to the normal vector of each first pixel point.
[0006] In an embodiment, the step of determining a gradient image of the normal vector image according to the normal vector image includes: for each pixel point in the normal vector image, obtaining a second-order derivative of the pixel point to obtain a gradient value of the pixel point;
[0007] Generate the gradient image according to the gradient value of each pixel point in the normal vector image.
[0008] In one embodiment, the step of extracting the target contour lines of the target surfaces of each box in the to-be-unstacked stack from the gradient image includes: performing contour detection on the gradient image to generate the contour lines of each surface of each box in the gradient image; for each surface of each box, generating a minimum two-dimensional bounding box that frames the contour line of the surface; screening out some of the minimum two-dimensional bounding boxes according to the size of the box; and determining the contour lines in the screened minimum two-dimensional bounding boxes as the target contour lines.
[0009] In one embodiment, the step of screening out some of the minimum two-dimensional bounding boxes according to the size of the box includes: screening out some of the minimum two-dimensional bounding boxes according to the preset area and / or preset length-width ratio of the surface of the box.
[0010] In one embodiment, the step of controlling the robot to unstack the to-be-unstacked stack according to the target contour lines of each box includes: respectively establishing a coordinate system for each box in the gradient image according to the target contour line of the target surface of the box; determining the position of the box in the robot coordinate system according to the position of the coordinate system of the box in the coordinate system of the gradient image and the conversion relationship between the coordinate system of the gradient image and the robot coordinate system; and controlling the robot to unstack the to-be-unstacked stack according to the position of the box in the robot coordinate system.
[0011] In one embodiment, the step of obtaining the depth image of the to-be-unstacked stack includes: photographing the to-be-unstacked stack by a 3D vision camera to obtain the depth image of the to-be-unstacked stack.
[0012] A second aspect of the present application provides a robot control device, and the system includes: an acquisition module for acquiring a depth image of a to-be-unstacked stack; a generation module connected to the acquisition module for generating a normal vector image of the depth image according to the depth image; a determination module connected to the generation module for determining a gradient image of the normal vector image according to the normal vector image; an extraction module connected to the determination module for extracting target contour lines of target surfaces of each box in the to-be-unstacked stack from the gradient image; and a control module connected to the extraction module for controlling a robot to unstack the to-be-unstacked stack according to the target contour lines of each box.
[0013] In a third aspect of the present application, a robot control device is provided. The robot control device includes a processor and a memory. The processor is configured to execute a computer program stored in the memory to implement the steps in the method described in any of the above embodiments.
[0014] In a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program that can be executed by a processor to implement the steps in the method described in any of the above embodiments.
[0015] Different from the prior art, the beneficial effects of the present application are as follows: The present application obtains a 3D depth image with depth values of the boxes in the stack to be disassembled, without obtaining a 2D color RGB image, and does not rely on conditions such as light scenes and deep learning data training. By generating a normal vector image of the depth image, the normal vector image can significantly increase the color mutation at positions with large depth value differences. As a result, the contour line features in the subsequent gradient image obtained based on the normal vector image are clearer and more obvious. Finally, the target contour line that meets the preset requirements is extracted by segmentation from the gradient image, and then, based on the target contour lines of each box, the robot is controlled to disassemble the stack to be disassembled. The method of the present application has a high accuracy in identifying the contours of the boxes and is not limited by dark light scenes, so the disassembly success rate of the robot can be improved compared with the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings, where:
[0017] Figure 1 is a flowchart showing an embodiment of the robot disassembly method of the present application;
[0018] Figure 2 is Figure 1 a color image of the stack to be disassembled in an embodiment;
[0019] Figure 3 is Figure 2 a depth image of a corresponding embodiment of the stack to be disassembled;
[0020] Figure 4 is based on Figure 3 a generated normal vector image of an embodiment;
[0021] Figure 5 is Figure 1Flow diagram of an embodiment of step S200;
[0022] Figure 6 is according to Figure 4 Gradient image of an embodiment determined;
[0023] Figure 7 is Figure 1 Flow diagram of an embodiment of step S300 in;
[0024] Figure 8 is from Figure 6 Schematic diagram of an embodiment for extracting the target contour line;
[0025] Figure 9 is Figure 1 Flow diagram of an embodiment of step S400 in;
[0026] Figure 10 is Figure 1 Flow diagram of an embodiment of step S500 in;
[0027] Figure 11 Schematic diagram of an embodiment of the robot control device of the present application;
[0028] Figure 12 Schematic diagram of another embodiment of the robot control device of the present application;
[0029] Figure 13 Schematic diagram of an embodiment of the computer-readable storage medium of the present application. Specific embodiments
[0030] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present application.
[0031] It should be noted that the terms "first" and "second" in this application are only for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In the description of this application, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically defined. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally further include steps or units not listed, or may optionally further include other steps or units inherent to these processes, methods, products, or devices.
[0032] Referring to Figure 1 , this application provides a method for a robot to unstack, and the method includes:
[0033] S100: Obtain the depth image of the stack to be unstacked.
[0034] First, it should be noted that the method of this application can be executed by a robot control device, such as a robot control cabinet.
[0035] Specifically, referring to Figure 2 and Figure 3 , Figure 2 shows a color RGB image of a stack to be unstacked, Figure 3 shows the corresponding depth image of the stack to be unstacked. It should be noted that in this step, only the depth image of the stack to be unstacked needs to be obtained, and it is not necessary to obtain the color RGB image of the stack to be unstacked. The corresponding color RGB image is shown here for the convenience of understanding the source of the objects in the depth image. The data dimension of the depth image is single-channel, and each pixel value in the depth image represents the vertical distance between each point in the stack to be unstacked and the camera, that is, the depth value. Therefore, the depth image has the 3D information of the stack to be unstacked.
[0036] Furthermore, since there are usually gaps between adjacent boxes in the stack to be unstacked, in the depth image, the depth value of the gap is quite different from that of the planes on both sides. Therefore, the information of different edges of different boxes can be identified by using the depth image.
[0037] In one embodiment, the stack to be unstacked is photographed by a 3D vision camera to obtain the depth image of the stack to be unstacked.
[0038] Specifically, the 3D vision camera includes a structured light camera, a binocular camera, a Tof (time of flight) camera, etc. The 3D vision camera performs image acquisition on the stack to be unstacked to obtain the depth image.
[0039] Of course, in some other embodiments, the depth image can also be directly obtained from other scanning devices, processed, and then sent to the control system of the robot of the present application.
[0040] S200: Generate a normal vector image of the depth image according to the depth image.
[0041] Specifically, referring to Figure 3 and Figure 4 , each pixel point in the normal vector image includes the values of the R, G, and B channels, which respectively correspond to the Nx, Ny, and Nz components of the normal vector of the plane where each pixel point in the depth image is located. Among them, the vertically upward normal vector (0, 0, 1) in the normal vector image corresponds to the horizontal plane in the depth image. The color displayed by the normal vector (0, 0, 1) in the normal vector image is blue. The horizontal normal vectors (1, 0, 0) and (0, 1, 0) in the normal vector image correspond to the vertical planes in the depth image. The color displayed by the normal vector (1, 0, 0) is red, and the color displayed by the normal vector (0, 1, 0) is green. Since the differences among the three colors of red, green, and blue are very large, at the positions where the depth values change greatly, the color contrast is relatively large. For example, it jumps from blue to red, or from blue to green. Thus, at the positions near the edge of the object, because the depth values are different, the colors of the corresponding normal vector map will change suddenly, forming an edge with a high-contrast color. Therefore, compared with the edge features of the depth map, the edge features of the normal vector map obtained after processing are enhanced, which is beneficial to subsequent contour extraction.
[0042] Referring to Figure 5 , in one embodiment, step S200 includes:
[0043] S210: For each first pixel point in the depth image, obtain the first depth value of the first pixel point and the second depth values of multiple second pixel points around the first pixel point, determine the spatial coordinates of the first pixel point according to the first depth value of the first pixel point, and respectively determine the spatial coordinates of the multiple second pixel points according to the second depth values of the multiple second pixel points. Then, determine the normal vector of the first pixel point according to the spatial coordinates of the first pixel point and the spatial coordinates of the multiple second pixel points.
[0044] Specifically, the depth value of each pixel point is stored in the depth image. The first pixel point can be any pixel point in the depth image, and the second pixel point is the pixel point around the first pixel point. For example, for the first pixel point at a non-edge position in the depth image, the number of the second pixel points in the surrounding circle corresponding to it is 8; for the first pixel point at a non-corner edge position in the depth image, the number of the second pixel points in the surrounding circle corresponding to it is 5; for the first pixel point at a corner position in the depth image, the number of the second pixel points in the surrounding circle corresponding to it is 3. If the selection of the second pixel points is two circles around the first pixel point, the number of the second pixel points can be increased again. The distance between the second pixel point and the first pixel point can be adjusted according to the actual situation, which is not limited in this application. After obtaining the first depth value of the first pixel point, the spatial coordinates of the first pixel point can be calculated according to the camera internal parameters and external parameters. After obtaining the second depth values of multiple second pixel points, the spatial coordinates of the second pixel points can be calculated according to the camera internal parameters and external parameters, and then the plane where the first pixel point is located can be determined according to the spatial coordinates of the first pixel point and the spatial coordinates of multiple second pixel points, and the normal vector of this plane can be further calculated by normalization.
[0045] S220: Generate a normal vector map according to the normal vector of each first pixel point.
[0046] Specifically, the three components of the normal vector of each pixel point are respectively stored as an image corresponding to the R, G, and B channels to obtain a normal vector map.
[0047] Refer to Figure 1 , after the above step S200, it includes:
[0048] S300: Determine the gradient image of the normal vector image according to the normal vector image.
[0049] Specifically, refer to Figure 4 and Figure 6 , by calculating the gradient of each pixel in the normal vector map, the gradient image of the normal vector image is obtained. The gradient image is preferably non-color, which can reduce the amount of data calculation. It can be seen that the lines formed by the white pixel points in the obtained gradient image are the edges of the box.
[0050] Refer to Figure 7 , in an embodiment, step S300 includes:
[0051] S310: For each pixel point in the normal vector image, obtain the second derivative of the pixel point to obtain the gradient value of the pixel point.
[0052] Specifically, refer to Figure 4, calculate the second-order derivatives of the three components Nx, Ny, and Nz of the normal vector at each pixel point in the normal vector map respectively. The second-order derivatives are calculated through the Laplace operator, thereby calculating the gradient value of each pixel point. The advantage of the second-order derivative is that it detects the image edge through the change in the intensity of the pixel values of the image, and the image edge will be finer.
[0053] The calculation formula is as follows:
[0054]
[0055] Among them, dst refers to the gradient value corresponding to the pixel point in the gradient image, src refers to the normal vector value corresponding to the pixel point in the normal vector image, x and y are the position coordinates of the pixel point in the normal vector image in two directions, △ is the Laplace operator, is the partial derivative symbol.
[0056] S320: Generate a gradient image according to the gradient value of each pixel point in the normal vector image.
[0057] Specifically, refer to Figure 6 , take the gradient value of each pixel point, and then store the gradient value in the image in the form of grayscale to obtain the gradient image. It can be seen from the figure that determining the gradient image of the second-order derivative corresponding to the normal vector map can clearly display the contour line of the box body.
[0058] Of course, in some other embodiments, the first-order derivative of each pixel point in the normal vector map can also be calculated to obtain the gradient value of the pixel point, and a gradient image is generated according to the gradient value of each pixel point in the normal vector image. The first-order derivative can obtain the local gradient response value of the image gradient, and its final contour line is relatively thick, and the effect is worse than that of the previous embodiment.
[0059] Refer to Figure 1 , after the above step S300, it includes:
[0060] S400: Extract the target contour lines of the target surfaces of each box body in the stack to be disassembled from the gradient image.
[0061] Specifically, in this step, it is necessary to Figure 6 extract the target contour lines of the target surfaces of the box bodies in Figure 8 , in Figure 8 , because the camera shooting angle is top-down shooting, therefore, the target surface is the upper surface of the box body. In other embodiments, when the camera shooting angle changes, it can be any side surface of the box body, and the target surface refers to the surface of the box body facing the camera. The target contour line is the contour line of the target surface. The target contour lines of the target surfaces of each box body in the stack to be disassembled are usually extracted from the gradient image by using a contour detection algorithm.
[0062] In one embodiment, referring to Figure 9 , the above steps include:
[0063] S410: Perform contour detection on the gradient image to generate contour lines of each surface of each box in the gradient image.
[0064] Specifically, referring to Figure 6 , perform detection on Figure 6 through a contour detection algorithm to detect all the contour lines of each surface of all the boxes in the gradient image. At this time, all the contour lines in the gradient image can be detected, including the contour lines of multiple surfaces of the boxes.
[0065] S420: For each surface of each box, generate a minimum two-dimensional bounding box that encloses the contour line of the surface.
[0066] Specifically, since each box has multiple surfaces, and the adjacent surfaces of the box share an edge, and the surfaces of adjacent boxes may also intersect at the edge. Therefore, although all the contour lines are detected in the previous step, the contour lines cannot be identified as belonging to a certain box. In this step, by generating the minimum two-dimensional bounding box of the contour line, all the minimum surfaces of the box can be framed out. It can be understood that although the box has multiple surfaces, each surface is independent, and the adjacent surfaces share an edge. Therefore, each surface of each box can be framed out by the minimum two-dimensional bounding box, so as to extract the contour lines of each surface of all the boxes from all the contour lines in the gradient image.
[0067] S430: According to the size of the box, screen out some of the minimum two-dimensional bounding boxes.
[0068] Specifically, since the box has multiple surfaces, a surface in a certain orientation can be screened out according to a preset rule, and then the minimum two-dimensional bounding boxes of other surfaces are removed.
[0069] Preferably, in one embodiment, some of the minimum two-dimensional bounding boxes are screened out according to the preset area and / or preset length-width ratio of the surface of the box.
[0070] Specifically, the angles between the surfaces of the box and the camera are different. Among them, the surface facing the camera is generally perpendicular to the shooting direction of the camera. There is no change in the angle between the surface and the shooting direction of the camera due to the change of the shooting position of the camera, so that the surface is deformed. Therefore, the minimum two-dimensional bounding box can be screened out according to the preset area and / or preset length-width ratio of this surface. It can be understood that when the area differences of the target surfaces of different boxes are not significant, the preset area can be used for screening. When the length-width ratio differences of the target surfaces of different boxes are not significant, the preset length-width ratio can be used for screening.
[0071] Of course, in some other embodiments, for example, the side surfaces of some boxes can also be used as target surfaces, and the minimum two-dimensional bounding box can be screened out by the included angle between the contour lines.
[0072] S440: Determine the contour lines in the screened minimum two-dimensional bounding box as the target contour lines.
[0073] Specifically, only a single surface is enclosed in the minimum two-dimensional bounding box, so the corresponding contour lines are also the contour lines of a single surface. Through the above screening of the minimum two-dimensional bounding box, the target contour lines of all target surfaces are obtained.
[0074] Please refer to Figure 1 , after the above step S400, it includes:
[0075] S500: Control the robot to unstack the stack to be disassembled according to the target contour lines of each box.
[0076] Specifically, after determining the target contour lines of each box, the positions of the target contour lines in the image are determined, that is, the positions of each box in the image. Then, the robot can be controlled to grab the boxes at the corresponding positions of each box, thus completing the unstacking.
[0077] In one embodiment, referring to Figure 10 , the above steps include:
[0078] S510: Establish a coordinate system for each box in the gradient image according to the target contour lines of the target surface of the box.
[0079] Specifically, each target contour line corresponds to a box and is the contour line of the target surface of the box. In the gradient image, since it is necessary to determine the position of each box in the gradient image, that is, to determine the position of the target contour line in the gradient image, a coordinate system is established for each target contour line, and thus a coordinate system is established for each box, and then the position of each box in the gradient image can be determined.
[0080] S520: Determine the position of the box in the robot coordinate system according to the position of the coordinate system of the box in the coordinate system of the gradient image and the conversion relationship between the coordinate system of the gradient image and the robot coordinate system.
[0081] Specifically, by determining the position of the box in the gradient image and the position of the object in the gradient image relative to the robot space, the robot can obtain the position of the box in the robot space. Among them, according to the conversion relationship between the coordinate system of the box and the coordinate system of the gradient image, the position of the box in the gradient image can be determined. According to the conversion between the coordinate system based on the gradient image and the robot coordinate system, the position of the box in the robot coordinate system can be determined, so that the robot can position the box.
[0082] S530: Control the robot to unstack the stack to be unstacked according to the position of the box in the robot coordinate system.
[0083] Specifically, after obtaining the position of the box in the robot coordinate system, the robot can be controlled to grasp the box at each corresponding position of the box, so as to complete the unstacking.
[0084] This application obtains the 3D depth image with depth values of the boxes in the stack to be unstacked, without obtaining the 2D color RGB image, and does not depend on conditions such as the light scene and deep learning data training. By generating the normal vector image of the depth image, the normal vector image can significantly increase the color mutation at positions with large depth value differences. Therefore, the gradient image determined subsequently based on the normal vector image has clearer and more obvious contour line features. Finally, the target contour line that meets the preset requirements is extracted from the gradient image through segmentation. Then, according to the target contour lines of each box, the robot is controlled to unstack the stack to be unstacked. The method of this application has high accuracy in contour recognition of boxes and is not limited by the dark light scene, so the unstacking success rate of the robot can be improved compared with the prior art.
[0085] Refer to Figure 11 , the second aspect of this application provides a robot control device 10, and the robot control device 10 includes:
[0086] The acquisition module 11 is used to acquire the depth image of the stack to be unstacked.
[0087] The generation module 12 is connected to the acquisition module 11 and is used to generate the normal vector image of the depth image according to the depth image.
[0088] The determination module 13 is connected to the generation module 12 and is used to determine the gradient image of the normal vector image according to the normal vector image.
[0089] The extraction module 14 is connected to the determination module 13 and is used to extract the target contour lines of the target surfaces of each box in the stack to be unstacked from the gradient image.
[0090] The control module 15 is connected to the extraction module 14 and is used to control the robot to unstack the stack to be unstacked according to the target contour lines of each box.
[0091] Among them, when the robot control device 10 of the present application works, it executes the method steps in any one of the above embodiments. For the detailed method steps, please refer to the relevant content above.
[0092] Among them, the robot control device 10 may be a robot control cabinet, a robot control box, or a robot control box, etc.
[0093] Refer to Figure 12 , a third aspect of the present application provides a robot control device 20, which includes a processor 21 and a memory 22. The processor 21 is configured to execute a computer program stored in the memory 22 to implement the steps in the method in any one of the above embodiments.
[0094] In this embodiment, the processor 21 may also be referred to as a CPU (Central Processing Unit). The processor 21 may be an integrated circuit chip with signal processing capabilities. The processor 21 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor 21 may also be any conventional processor, etc.
[0095] Among them, the robot control device 20 may specifically be a robot control cabinet, a robot control box, or a robot control box, etc.
[0096] Refer to Figure 13 , a fourth aspect of the present application provides a computer-readable storage medium 30, which stores a computer program 31. The computer program 31 can be executed by a processor to implement the steps in the method in any one of the above embodiments.
[0097] Among them, the computer-readable storage medium 30 may specifically be a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc, etc., which can store the computer program 31, or it may also be a server storing the computer program 31. The server can send the stored computer program 31 to other devices for running, or it can also run the stored computer program 31 by itself.
[0098] The above are only embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.
Claims
1. A method for robot palletizing, characterized in that, The method includes: Obtaining a depth image of the stack to be disassembled; Generating a normal vector image of the depth image according to the depth image; Determining a gradient image of the normal vector image according to the normal vector image; Extracting target contour lines of the target surfaces of each box body in the stack to be disassembled from the gradient image; Controlling a robot to disassemble the stack to be disassembled according to the target contour lines of each box body.
2. The method according to claim 1, wherein The step of generating a normal vector image of the depth image according to the depth image includes: For each first pixel point in the depth image, obtaining a first depth value of the first pixel point and second depth values of a plurality of second pixel points around the first pixel point, and determining a spatial coordinate of the first pixel point according to the first depth value of the first pixel point, and respectively determining spatial coordinates of the plurality of second pixel points according to the second depth values of the plurality of second pixel points, and then determining a normal vector of the first pixel point according to the spatial coordinate of the first pixel point and the spatial coordinates of the plurality of second pixel points; Generating the normal vector map according to the normal vector of each first pixel point.
3. The method according to claim 1, wherein The step of determining a gradient image of the normal vector image according to the normal vector image includes: For each pixel point in the normal vector image, obtaining a second-order derivative of the pixel point to obtain a gradient value of the pixel point; Generating the gradient image according to the gradient value of each pixel point in the normal vector image.
4. The method according to claim 1, wherein The step of extracting target contour lines of the target surfaces of each box body in the stack to be disassembled from the gradient image includes: Performing contour detection on the gradient image to generate contour lines of each surface of each box body in the gradient image; For each surface of each box body, generating a minimum two-dimensional bounding box enclosing the contour line of the surface; Filtering out some of the minimum two-dimensional bounding boxes according to the size of the box body; Determining the contour lines in the filtered minimum two-dimensional bounding boxes as the target contour lines.
5. The method according to claim 4, wherein The step of filtering out some of the minimum two-dimensional bounding boxes according to the size of the box body includes: Filtering out some of the minimum two-dimensional bounding boxes according to a preset area and / or a preset length-width ratio of the surface of the box body.
6. The method according to claim 1, wherein The step of controlling a robot to disassemble the stack to be disassembled according to the target contour lines of each box body includes: Respectively establishing a coordinate system of each box body in the gradient image according to the target contour lines of the target surfaces of the box body; Determining the position of the box body in the robot coordinate system according to the position of the coordinate system of the box body in the coordinate system of the gradient image and the conversion relationship between the coordinate system of the gradient image and the robot coordinate system. Control the robot to unstack the stack to be unstacked according to the position of the box body in the robot coordinate system.
7. The method for robotic palletizing according to claim 1, wherein The step of obtaining the depth image of the stack to be unstacked includes: Shoot the stack to be unstacked through a 3D vision camera to obtain the depth image of the stack to be unstacked.
8. A robot control device, characterized in that, The system includes: An acquisition module for acquiring the depth image of the stack to be unstacked; A generation module, connected to the acquisition module, for generating a normal vector image of the depth image according to the depth image; A determination module, connected to the generation module, for determining the gradient image of the normal vector image according to the normal vector image; An extraction module, connected to the determination module, for extracting the target contour lines of the target surfaces of each box body in the stack to be unstacked from the gradient image; A control module, connected to the extraction module, for controlling the robot to unstack the stack to be unstacked according to the target contour lines of each box body.
9. A robot control device, characterized in that, The robot control device includes a processor and a memory, and the processor is configured to execute a computer program stored in the memory to implement the steps in the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program can be executed by a processor to implement the steps in the method according to any one of claims 1-7.