A general-purpose pallet 3D pose positioning method and system, and a storage medium

Abstract

The application provides a general pallet 3D pose positioning method and system and a storage medium, wherein the method steps comprise: collecting point cloud data and texture images in a scene, establishing a mapping relationship A between the point cloud and the texture image pixel coordinates according to the camera internal parameters; identifying the ROI region of the pallet in the texture image based on a learning network, so as to obtain the target point cloud in the ROI according to the mapping relationship A to perform grid division, and performing plane fitting on the point cloud in each grid and then performing tensor voting to select the pallet plane; extracting the point cloud near the pallet plane to project onto the pallet plane to obtain a projection image, so as to determine the pallet type through image recognition; taking the 3D point coordinates corresponding to the center pixel of the projection image as the recognition position; and calculating the included angle between the pallet plane and the Z-axis of the camera coordinate system to obtain the pallet pose. In this way, the space pose of the pallet can be obtained without relying on the pallet model parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to object space pose positioning technology, and in particular to a general pallet 3D pose positioning method and system and a storage medium. BACKGROUND

[0002] A pallet is a platform used for loading, transporting and storing goods, usually made of wood, plastic or metal. Pallets are widely used in the logistics industry, which can improve the efficiency and safety of goods transportation. Because the traditional manual operation of forklift taking and placing method is low in efficiency, and errors and accidents are prone to occur. Therefore, people began to look for an automatic technology that can quickly and accurately identify pallets, improve the efficiency and safety of goods transportation. At the same time, with the rapid development of the logistics industry and the explosive growth of logistics turnover demand, the industrial scene needs to introduce intelligent forklifts and other more automated and efficient goods loading and transportation methods to meet the safety and efficiency of logistics demand.

[0003] Therefore, as the eyes of intelligent forklifts to take and place goods, the current pallet recognition technology mainly relies on computer vision and image processing technology. For example, by using high-resolution cameras and professional image processing software, the image of the pallet can be obtained and analyzed and recognized. Among them, the pallet recognition technology can detect the size, shape, material and identification of the pallet, as well as the placement method and quantity of the goods. These information can be used to optimize the loading and transportation process of the goods, reduce human errors and improve work efficiency.

[0004] In practice, intelligent forklifts need to know the placement pose of pallets, and the current mainstream pallet recognition method is to use a 3D camera to obtain point cloud information of the scene. Then according to the scene point cloud structure information obtained by the 3D camera and the known pallet model, the pose of the pallet in the scanned scene is obtained by matching.

[0005] However, the disadvantage of this prior art is that the pose calculation method based on model matching requires the model file of the pallet in the warehouse, which restricts the use of multiple types of pallets in the factory. However, in actual scenarios, pallets need to be selected according to the size, weight and material of the transported materials, so it is often difficult to meet the different needs of the entire factory with one type of pallet. Therefore, how to adaptively recognize the spatial pose of different types of pallets based on vision has become the main research direction of the current intelligent logistics industry. SUMMARY

[0006] Therefore, the main purpose of the present application is to provide a general pallet 3D pose positioning method and system and a storage medium, so as to obtain the spatial pose of the pallet without relying on the pallet model parameters.

[0007] In order to achieve the above object, according to a first aspect of the present application, a general pallet 3D pose positioning method is provided, steps comprising:

[0008] Step S100 collects point cloud data and texture images in the scene, and establishes a mapping relationship A between 3D point coordinates in the point cloud data and pixel coordinates of the texture images according to camera intrinsic parameters;

[0009] Step S200 identifies the ROI region of the pallet in the texture image based on a learning network, so as to obtain the corresponding target point cloud in the ROI region according to the mapping relationship A;

[0010] Step S300 divides the target point cloud into grids, and performs plane fitting on the point cloud in each grid, and performs tensor voting on all fitted planes to select the pallet plane;

[0011] Step S400 extracts the point cloud near the pallet plane to project onto the pallet plane to obtain a projection image, so as to determine the pallet type through image recognition;

[0012] Step S500 takes the 3D point coordinates corresponding to the center pixel of the projection image as the recognition position , calculates the angle between the pallet plane and the Z-axis of the camera coordinate system , and obtains the pallet pose .

[0013] In a possible preferred embodiment, wherein the step of performing tensor voting on all fitted planes to select the pallet plane in step S300 comprises:

[0014] Step S310 calculates the angles between the normal of each plane and the X-axis and Y-axis of the camera coordinate system and respectively:

[0015] ;

[0016] Step S320 equally divides the angles of the X-axis and Y-axis of the camera coordinate system according to a preset angle step a to construct an angle space coordinate system, and calculates according to the plane equation and the angles :

[0017]

[0018]

[0019] obtains the position of each plane in the angle space ;

[0020] Step S330 counts the grid with the most points in the angle space of all planes, and calculates the mean value of the plane equation in the grid to obtain the pallet plane.

[0021] In a possible preferred implementation, the step of obtaining the projection image in step S400 comprises:

[0022] Step S410 divides the pallet plane into a grid according to a preset size, wherein each grid corresponds to an image pixel point.

[0023] Step S420 projects the point cloud near the pallet plane onto the pallet plane, and when each grid is mapped to the point cloud, the pixel color of the grid is adjusted, and all grids are traversed to generate the projection image.

[0024] In a possible preferred implementation, the preset size of the grid is 0.5 mm*0.5 mm; and when each grid is mapped to the point cloud, the pixel gray value of the grid is adjusted to 255, and the remaining gray values are 0.

[0025] In a possible preferred implementation, the training step of the learning network in step S200 comprises:

[0026] Step S210 sets 3D cameras at the horizontal height of the pallet and a plurality of height positions upward, sets a downward inclination angle for the remaining 3D cameras except the horizontal height, to obtain pallet samples at different heights and camera installation position variables thereof;

[0027] Step S220 sets a distance interval of the camera, an angle interval and an angle step of the angle between the pallet plane and the camera plane at each 3D camera installation height, to collect pallet image samples, to make pallet deep learning samples, to train the learning network.

[0028] In a possible preferred implementation, the distance interval in step S220 is 0.2 m, the angle interval is , and the angle step is 10°.

[0029] To achieve the above-mentioned purpose, corresponding to the above-mentioned method, according to the second aspect of the present application, a general pallet 3D pose positioning system is provided, which comprises:

[0030] A storage unit for storing programs comprising steps of the general pallet 3D pose positioning method according to any one of the above, for the control unit, the image recognition unit, the texture recognition unit, and the processing unit to execute timely;

[0031] A control unit for controlling the camera to collect point cloud data and texture images in the scene;

[0032] A texture recognition unit for identifying the ROI region of the pallet in the texture image based on the learning network, to obtain the corresponding target point cloud in the ROI region according to the mapping relationship A between the 3D point coordinates in the point cloud data and the pixel coordinates in the texture image;

[0033] The processing unit divides the target point cloud into a mesh and performs plane fitting on the point cloud within each mesh. It then performs tensor voting on all fitted planes to select the stack plane. Afterward, it extracts the point cloud near the stack plane and projects it onto the stack plane to obtain a projection image. Once the image recognition unit determines the stack type, the 3D point coordinates corresponding to the center pixel of the projection image are used as the recognition location. ; and calculate the angle between the pallet plane and the Z-axis of the camera coordinate system. To obtain the attitude of the pallet .

[0034] In a possible preferred embodiment, the tensor voting process for selecting the stack plane from all fitted planes includes:

[0035] The processing unit calculates the angle between the normal of each plane and the camera's X-axis in the camera coordinate system. Angle with the Y-axis Then, the angle between the X and Y axes of the camera coordinate system is divided into equal intervals according to a preset angle step size 'a', constructing an angular space coordinate system. Based on the plane equation and the angle... Calculate the position of each plane in the angular space; then count the grids with the most points in the angular space for all planes, and average the plane equations in the grids to obtain the pallet plane.

[0036] In a possible preferred embodiment, obtaining the projection image includes:

[0037] The processing unit divides the stack plane into grids according to a preset size, where each grid corresponds to an image pixel. Then, it projects the point cloud near the stack plane onto the stack plane. When each grid is mapped to the point cloud, the pixel color of that grid is adjusted. The process is repeated for all grids to generate a projection map.

[0038] To achieve the above objectives, in accordance with the above methods, according to a third aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, wherein, when the computer program is executed by a processor, it implements the steps of any of the above-described general-purpose stack 3D pose localization methods.

[0039] The universal 3D pose localization method, system, and storage medium for pallets provided by this invention cleverly utilize the spatial localization capabilities of point clouds and the generalization capabilities of texture images simultaneously to calculate the spatial pose of the pallet without prior knowledge of the pallet model parameters. This eliminates the constraint of applicable pallet types, improving the versatility of pallet localization and recognition. Furthermore, this solution introduces a planar equation tensor voting technique, which significantly reduces the computational complexity of pallet plane extraction and improves the overall algorithm's operating efficiency. Attached Figure Description

[0040] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application, illustrate the preferred embodiments of the application and assist in the explanation of the application. In the drawings, the same reference numbers represent the same elements throughout the several views:

[0041] Figure 1 A general pallet 3D pose positioning method step schematic diagram of the application;

[0042] Figure 2 A general pallet 3D pose positioning method processing logic schematic diagram of the application;

[0043] Figure 3 A general pallet 3D pose positioning method, collecting training samples required by the learning network, example schematic diagram of the application;

[0044] Figure 4 A general pallet 3D pose positioning method, plane equation normal voting concept schematic diagram of the application;

[0045] Figure 5 A general pallet 3D pose positioning method, projection diagram schematic diagram of the application;

[0046] Figure 6 A general pallet 3D pose positioning system structure schematic diagram of the application. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the technical solutions of the present application, the specific technical solutions of the present application will be described clearly and completely below in conjunction with the embodiments, so as to help those skilled in the art to further understand the present application. Obviously, the embodiments described in the present application are only a part of the embodiments of the present application, not all the embodiments. It should be pointed out that, for those skilled in the art, the embodiments in the present application and the features in the embodiments can be combined with each other without departing from the concept of the present application and without being in conflict with each other. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the disclosure and protection scope of the present application.

[0048] Furthermore, the terms "first", "second", "S100", "S200" and the like in the description and in the claims of the present application and the drawings are used to distinguish similar objects, not necessarily describing a particular sequential or chronological order. It is to be understood that the data so used can be interchanged, where appropriate, so that the embodiments of the present application described herein can be carried out in other than the order described. Also, the terms "comprise", "comprising", "include", "including", "contain", "containing", "have", "having", "composed of", "comprised of", and the like are used synonymously to denote a certain presence within a thing and does not exclude additional, un-recited elements of like character. Unless otherwise defined, the terms "set", "arrange", "mount", "connect", "connected", "connectivity" are to be construed in their broadest form, for example, can be fixed connections, can be detachable connections, or integrally connected; can be mechanical connections, can be electrical connections; can be direct connections, or indirect connections through intermediate media, or internal communication between two elements. For those skilled in the art, the specific meanings of the above terms in the present application can be understood in combination with the prior art according to the specific circumstances.

[0049] Please refer to Figures 1 to 5 In order to obtain the spatial pose of the pallet without relying on the pallet model parameters, the present application provides a general pallet 3D pose positioning method, and the example steps thereof include:

[0050] Step S100 collects point cloud data and texture images in the scene by a 3D camera, and simultaneously establishes a mapping relationship A between 3D point coordinates in the point cloud data and pixel coordinates in the texture images according to the camera intrinsic parameters.

[0051] Specifically, the point cloud output by the 3D depth camera is The set of three-dimensional point coordinates of the object in the scene under the image coordinate system is The conversion relationship between the camera intrinsic parameters and the point cloud data is:

[0052]

[0053] Among them, is the mapping of the three-dimensional point coordinates on the texture image . According to the camera intrinsic parameter conversion relationship, the mapping of each three-dimensional point in the point cloud image to each pixel in the texture image can be obtained.

[0054] The storage mode of the point cloud data is stored in the form of ordered point cloud, and the width and height of the point cloud file are aligned with the width and height of the texture image, so that the mapping of each pixel point in the texture image to the three-dimensional point cloud can be obtained.

[0055] Step S200 identifies the ROI region of the pallet in the texture image based on the learning network to obtain the corresponding target point cloud in the ROI region according to the mapping relationship A.

[0056] In this example, the training step of the learning network preferably includes:

[0057] Step S210 sets 3D cameras at the horizontal height of the pallet and a plurality of height positions upward, sets a downward inclination angle for the remaining 3D cameras except the horizontal height, to obtain pallet samples at different heights and camera installation position variables.

[0058] For example, as shown in Figure 3 , three 3D camera installation height ranges of 0.3m, 0.6m and 1m can be set, wherein the 3D camera is installed horizontally at a height of 0.3m, the camera is installed at an inclination of 10 degrees downward at a height of 0.6m, and the camera is installed at an inclination of 30 degrees downward at a height of 1m. By setting such multi-angle and height camera installation position variables to collect pallet samples, the camera installation height and angle can be changed to affect the pallet recognition training, thereby improving the recognition rate of the subsequent learning network.

[0059] Step S220 sets the distance interval of the camera, the angle interval and the angle step of the angle between the pallet plane and the camera plane at each 3D camera installation height to collect pallet image samples to make pallet deep learning samples to train the learning network.

[0060] Preferably, at each camera installation height, the distance interval is 0.2m within 1-2m from the camera, the angle interval is , and the angle step is to collect pallet image samples and convert the pictures to grayscale images to make pallet deep learning samples. The learning network example can use yolo-v8 as the identification basis network of the pallet target, and the migration learning of the pallet identification based on the already trained model provided by yolo-v8 can achieve better recognition effect based on a small amount of pallet labeled samples as training data set.

[0061] Then, based on the trained learning network, the ROI region of the pallet in the texture image can be identified, and the ROI rectangular frame is composed of the starting point coordinates and the width and height .

[0062] Step S300 divides the target point cloud into grids, and performs plane fitting on the point cloud in each grid, and performs tensor voting on all fitted planes in the ROI region to select the pallet plane.

[0063] For example, the pallet ROI region identified by step S200 is divided into Based on the mapping relationship A between the image and the point cloud, the point cloud falling within each grid is extracted, and the point clouds within the same grid are marked as the same cluster point. .

[0064] Next, a plane fit is performed on the clustered point cloud within each grid. Assume the plane equation is... ,in Let be the parameters of the plane equation. Substituting these parameters into the plane equation for each point within the grid, the form is as follows:

[0065]

[0066] make , , .

[0067] achievable This is the plane equation.

[0068] Furthermore, step S300, which involves performing tensor voting on all fitted planes within the ROI region to select the stack plane, includes:

[0069] Step S310: Using the camera coordinate system as the reference coordinate system, calculate the angle between the normal of each plane and the camera X-axis. Angle with the Y-axis :

[0070] ;

[0071] Step S320 involves dividing the angle between the X and Y axes of the camera coordinate system at equal intervals according to a preset angle step size 'a' (e.g., using 30° intervals) to construct an angular space coordinate system. Then, the angle between each plane equation within the ROI region and the camera's X-axis is determined. Angle with the Y-axis Calculate its position in angular space. The formula is as follows:

[0072] Based on the plane equation and the included angle ,calculate:

[0073]

[0074]

[0075] Step S330: Calculate the grid with the most points in the angle space for all planes, and take the average of the plane equations in the grid to obtain the stack plane.

[0076] like Figure 4 As shown, based on the position of each plane equation in angular space The grid with the most points in the angle space of all planes in the ROI region is the grid with the highest votes, and the average of the plane equations in the grid can obtain the pallet plane equation.

[0077] Step S400 extracts the point cloud near the pallet plane to project onto the pallet plane to obtain a projection image, and determines the pallet type through image recognition.

[0078] wherein the distance threshold near the pallet plane can be determined The point cloud with a distance less than from the pallet plane in the ROI region is extracted as the pallet plane point. The calculation equation of the distance from the point to the plane is:

[0079] .

[0080] The step of obtaining the projection image in step S400 includes:

[0081] Step S410 divides the pallet plane into grids according to a preset size (for example, 0.5mm*0.5mm), wherein each grid corresponds to an image pixel point.

[0082] Step S420 projects the point cloud near the pallet plane onto the pallet plane, adjusts the pixel color of each grid when the grid is mapped to the point cloud (for example, when each grid is mapped to the point cloud, the pixel gray value of the grid is adjusted to 255, and the remaining gray values are 0, to obtain a binary image), and traverses all grids to generate a binary projection image as shown in Figure 5 .

[0083] Then, through the existing image recognition technology, the projection image is matched in similarity according to the supportable candidate pallet template image, if it is found that the similarity between a certain pallet template and the projection image is greater than a certain threshold, such as , it is considered that the pallet is the expected recognized pallet type.

[0084] Step S500 takes the 3D point coordinates corresponding to the center pixel of the projection image as the recognition position ; then calculates the angle between the pallet plane and the Z-axis of the camera coordinate system , wherein is the normal of the pallet plane, is the direction of the Z-axis of the camera. According to the forklift with only four dimensions, the final output of the 6DoF pose of the pallet is the pallet position and the attitude .

[0085] On the other hand, corresponding to the above method, the application also provides a general pallet 3D pose positioning system, which comprises:

[0086] The storage unit is used to store a program including the steps of the general stack 3D pose positioning method as described above, so that the control unit, image recognition unit, texture recognition unit, and processing unit can retrieve and execute it as needed.

[0087] The control unit is used to control the camera to acquire point cloud data and texture images in the scene;

[0088] The texture recognition unit is used to identify the ROI region of the stack in the texture image based on the learning network, so as to obtain the target point cloud corresponding to the ROI region according to the mapping relationship A between the 3D point coordinates in the point cloud data and the pixel coordinates of the texture image.

[0089] The processing unit divides the target point cloud into a mesh and performs plane fitting on the point cloud within each mesh. It then performs tensor voting on all fitted planes to select the stack plane. Afterward, it extracts the point cloud near the stack plane and projects it onto the stack plane to obtain a projection image. Once the image recognition unit determines the stack type, the 3D point coordinates corresponding to the center pixel of the projection image are used as the recognition location. ; and calculate the angle between the pallet plane and the Z-axis of the camera coordinate system. To obtain the attitude of the pallet .

[0090] The process of selecting the stack plane by tensor voting for all fitted planes includes: a processing unit that, in the camera coordinate system, calculates the angle between the normal of each plane and the camera's X-axis. Angle with the Y-axis Then, the angle between the X and Y axes of the camera coordinate system is divided into equal intervals according to a preset angle step size 'a', constructing an angular space coordinate system. Based on the plane equation and the angle... Calculate the position of each plane in the angular space; then count the grids with the most points in the angular space for all planes, and average the plane equations in the grids to obtain the pallet plane.

[0091] The process of obtaining the projection image includes: a processing unit that divides the stack plane into grids according to a preset size, where each grid corresponds to an image pixel; then projects the point cloud near the stack plane onto the stack plane; when each grid is mapped to the point cloud, the pixel color of that grid is adjusted; and the projection image is generated by traversing all grids.

[0092] On the other hand, corresponding to the above method, the present invention also provides a computer-readable storage medium storing a computer program thereon, wherein when the computer program is executed by a processor, it implements the steps of any of the above-described general-purpose stack 3D pose positioning methods.

[0093] In summary, by the universal pallet 3D pose positioning method and system and the storage medium provided by the application, the spatial positioning capability of the point cloud and the generalization capability of the texture image can be ingeniously utilized at the same time, so as to calculate the spatial pose of the pallet without preknowing the pallet model parameters. Therefore, the application is no longer constrained by the applicable pallet type, and the general applicability of the pallet positioning and identification is improved. In addition, in the scheme, the technical means of plane equation tensor voting is introduced, so that the calculation complexity of pallet plane extraction can be significantly reduced, and the running efficiency of the overall algorithm is improved.

[0094] The preferred embodiments help explain the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments described. Obviously, many modifications and variations can be made in light of the contents of the specification. The specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited by the claims and their entire scope and equivalents, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

[0095] Those skilled in the art can understand that, in addition to implementing the system, device and each module thereof provided by the application in the form of pure computer readable program code, the same program can also be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers, etc. by logically programming the method steps. Therefore, the system, device and each module thereof provided by the application can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures in the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing methods and structures in the hardware component.

[0096] In addition, all or part of the steps of the above-mentioned embodiment methods can be completed by programs instructing related hardware, and the programs are stored in a storage medium and include a plurality of instructions for causing a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the methods described in various embodiments of the application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk and various program code storage media.

[0097] In addition, various different embodiments of the embodiments of the application can also be combined arbitrarily, as long as they do not deviate from the idea of the embodiments of the application, and they should also be considered as the disclosed content of the embodiments of the application.

Claims

1. A general-purpose pallet 3D pose positioning method, comprising the following steps: Step S100: collecting point cloud data and texture images in a scene, and establishing a mapping relationship A between 3D point coordinates in the point cloud data and pixel coordinates of the texture images according to camera intrinsic parameters; Step S200: identifying a ROI region of a pallet in the texture images based on a learning network, so as to obtain corresponding target point cloud in the ROI region according to the mapping relationship A, and the step comprises: Step S210: setting 3D cameras at a horizontal height of the pallet and a plurality of height positions upward, and setting a downward inclination angle of the 3D cameras other than the horizontal height, so as to obtain pallet samples at different heights and camera installation position variables; Step S220: setting a distance interval of the camera, an angle interval and an angle step of a pallet plane and a camera plane at each 3D camera installation height, so as to collect pallet image samples, to make pallet deep learning samples, and to train the learning network; Step S300: dividing the target point cloud into grids, fitting a plane for the point cloud in each grid, and performing tensor voting on all fitted planes to select a pallet plane, and the step comprises: Step S310 calculates the angle between each plane normal and the camera X axis and the camera Y axis in the camera coordinate system, respectively :​ ； Step S320 divides the angle between the X-axis and the Y-axis of the camera coordinate system according to a preset angle step a to construct an angle space coordinate system, and calculates the angle between the normal vector of the plane and the Y-axis of the camera coordinate system according to the plane equation and the angle space coordinate system. , calculate: ； ； acquiring the position of each plane in angle space ; Step S330: counting all planes with the most points in the grid in the angle space, and obtaining a pallet plane by averaging the plane equations in the grid; Step S400: extracting point cloud near the pallet plane, projecting the point cloud onto the pallet plane to obtain a projection image, and determining a pallet type through image recognition; Step S500 takes the 3D point coordinates corresponding to the center point pixel of the projection map as the recognition position ; calculate the angle between the stack plane and the Z axis of the camera coordinate system to obtain the stack pose . 2.The general-purpose pallet 3D pose positioning method according to claim 1, wherein the step of obtaining the projection image in step S400 comprises: Step S410: dividing the pallet plane into grids according to a preset size, wherein each grid corresponds to an image pixel point; Step S420: projecting the point cloud near the pallet plane onto the pallet plane, adjusting the pixel color of each grid when the grid is mapped to the point cloud, and generating the projection image by traversing all grids. 3.The general-purpose pallet 3D pose positioning method according to claim 2, wherein the preset size of the grid is 0.5mm*0.5mm; when each grid is mapped to the point cloud, the pixel gray value of the grid is adjusted to 255, and the remaining gray values are 0.

4. The universal pallet 3D pose positioning method according to claim 1, wherein the distance interval in step S220 is 0.2 m, the included angle interval is , and the angle step is 10°. 5.A general-purpose pallet 3D pose positioning system, comprising: a storage unit configured to store a program comprising steps of the general-purpose pallet 3D pose positioning method according to any one of claims 1 to 4, so as to be executed by a control unit, an image recognition unit, a texture recognition unit, and a processing unit in time; the control unit configured to control a camera to collect point cloud data and texture images in a scene; the texture recognition unit configured to identify a ROI region of a pallet in the texture images based on a learning network, so as to obtain corresponding target point cloud in the ROI region according to a mapping relationship A between 3D point coordinates in the point cloud data and pixel coordinates of the texture images; the processing unit configured to divide the target point cloud into grids, fit a plane for the point cloud in each grid, and perform tensor voting on all fitted planes to select a pallet plane; Then the point cloud near the pallet plane is extracted to project onto the pallet plane to obtain a projection image, and after determining the pallet type by the image recognition unit, the 3D point coordinates corresponding to the center pixel of the projection image are taken as the recognition position ; and the angle between the pallet plane and the Z axis of the camera coordinate system is calculated to obtain the pallet posture . 6.The general-purpose pallet 3D pose positioning system according to claim 5, wherein the tensor voting on all fitted planes to select the pallet plane comprises: A processing unit calculates an included angle between each plane normal and the X-axis and the Y-axis of the camera coordinate system respectively The X-axis and the Y-axis of the camera coordinate system are then divided at equal intervals according to a preset angle step a to construct an angle space coordinate system, and the position of each plane in the angle space is calculated according to the plane equation and the included angle The grid with the most points in the angle space is then counted, and the mean value of the plane equation in the grid is calculated to obtain the stack plane.​ 7. The universal pallet 3D pose positioning system according to claim 5, wherein the obtaining the projection map comprises: a processing unit, which grids the pallet plane according to a preset size, wherein each grid corresponds to an image pixel point; then projects the point cloud near the pallet plane onto the pallet plane, adjusts the pixel color of the grid when each grid is mapped to the point cloud, and traverses all grids to generate the projection map.

8. A computer readable storage medium having stored thereon a computer program, wherein, The computer program is executed by the processor to implement the steps of the universal pallet 3D pose positioning method in any one of claims 1 to 4.

Images