Computer vision-based inventory machine automatic identification counting method

Abstract

This invention discloses an automatic identification and counting method for inventory management machines based on computer vision, relating to the field of warehouse management technology. The method includes: forming identification features of materials; obtaining a target standard image; forming a top view of a slice of the goods pile and calculating the volume of the goods pile; analyzing and obtaining the number of identifiable target materials in the goods pile; dividing the volume of the goods pile into visible and invisible volumes; forming at least one volume allocation scheme and generating a verification coefficient for the volume allocation scheme; obtaining the newly added quantity of target materials; and obtaining the real-time quantity of target materials. By forming identification features of materials, obtaining the volume of the goods pile, obtaining the number of identifiable target materials, and forming a target volume allocation scheme, the influence of occlusion on feature recognition can be eliminated. The method estimates the number of each type of newly entered material for both partially and completely occluded materials.

Description

Technical Field

[0001] This invention relates to the field of warehouse management technology, specifically to a computer vision-based automatic identification and counting method for inventory management machines. Background Technology

[0002] Inventory counting machines refer to equipment or systems that utilize automation and information technology to replace or assist manual labor in counting, recording, and analyzing the inventory quantities of various materials in a warehouse. Since the existing inventory of various materials in the warehouse is known, only the number of newly arrived materials needs to be categorized and counted. However, newly arrived materials are usually stacked together, and this stacking can obscure identification features, affecting the counting process. Furthermore, some materials may be completely obscured, making them difficult to count using conventional methods. Summary of the Invention

[0003] To address the aforementioned technical problems, this paper provides a computer vision-based automatic identification and counting method for inventory management machines. This technical solution resolves the issues raised in the background section.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] Computer vision-based automatic identification and counting methods for inventory management include:

[0006] Obtain at least one type of material in the warehouse, obtain the current quantity of the material in the warehouse, and form identification features for the material;

[0007] Images of the goods entering the warehouse are captured to obtain at least one target image. The target image is then processed according to standards to obtain a target standard image.

[0008] Segment at least one cargo stack slice outline to form a target standard image; based on the cargo stack slice outline, form a top view of the cargo stack slice; and calculate the volume of the cargo stack.

[0009] Based on the identification features, at least one target material is found in the cargo pile, and the number of identifiable target materials in the cargo pile is analyzed.

[0010] Based on the number of identifiable items, the volume of the cargo stack is divided into visible volume and invisible volume;

[0011] Based on the invisible volume, at least one volume allocation scheme is formed, and a verification coefficient of the volume allocation scheme is generated. The volume allocation scheme with the smallest verification coefficient is taken as the target volume allocation scheme.

[0012] Based on the target volume allocation scheme, the new quantity of target materials is obtained. The existing quantity of target materials is added to the new quantity to obtain the real-time quantity of target materials.

[0013] Preferably, the identification features of the material formation include the following steps:

[0014] The surface of the material is evenly divided into at least one local block, and a front view image of the local block is obtained;

[0015] At least one sampling point is uniformly selected in the interval (0, 10), and at least one sampling point is uniformly selected in the interval (0, 360).

[0016] The front view image is scaled according to the values ​​of the sampling points to obtain at least one first image. The first image is rotated in the plane where the first image is located according to the values ​​of the sampling points to obtain at least one second image.

[0017] The second images generated from the frontal view images of the local blocks are summarized to form a verification set of the local blocks;

[0018] Take one of the at least one materials as the feature material and the remaining materials as the non-feature materials. If the intersection of the verification sets of two local blocks is empty, then the two local blocks are not associated.

[0019] If the local block of the characteristic material is not associated with any local block of the non-characteristic material, then the local block of the characteristic material is taken as the target local block, and the verification set of the target local block of the characteristic material is used to form the identification feature of the characteristic material.

[0020] When the characteristic material traverses at least one material, the identification characteristics of all materials are formed.

[0021] Preferably, the step of acquiring images of the pile of goods entering the warehouse to obtain at least one target image includes the following steps:

[0022] Choose any horizontal line as the baseline, set an arrow at one end of the baseline, take the horizontal plane where the baseline is located as the reference plane, and rotate the baseline within the reference plane according to the values ​​of the collection points to obtain at least one reference line.

[0023] At least one acquisition direction is formed, and the acquisition direction is along the direction of the arrow pointing on the reference line;

[0024] Images of the cargo pile are acquired along the acquisition direction to obtain at least one target image.

[0025] Preferably, the standard processing of the target image to obtain a target standard image includes the following steps:

[0026] The outline of the cargo stack in the target image is identified, and the height of the cargo stack outline in the target image is used as the reference size of the target image.

[0027] The maximum value of the reference size is used as the baseline value. The baseline value is divided by the reference size of the target image to obtain the scaling ratio of the target image. The target image is scaled according to the scaling ratio of the target image to obtain the target standard image. The acquisition direction of the target image is matched to the target standard image formed by the target image.

[0028] Preferably, the segmentation to form at least one cargo stack slice outline of the target standard image includes the following steps:

[0029] The cargo stack outline in the target standard image is segmented using at least one horizontal parallel line to obtain at least one cargo stack slice outline, wherein the spacing between adjacent parallel lines is a preset distance, which is set based on experience.

[0030] At least one cargo stack slice outline of the target standard image is numbered from top to bottom. Cargo stack slice outlines with the same number correspond to the same cargo stack slice. The acquisition direction corresponding to the target standard image is matched to the cargo stack slice outline generated by the target standard image.

[0031] Preferably, the process of forming a top view of the cargo stack slice and calculating the volume of the cargo stack includes the following steps:

[0032] The maximum length of the outline of the cargo stack slice is used as a reference value to form a preset area, which is a square with a side length equal to the reference value.

[0033] At least one grid point is uniformly selected within a preset area to form at least one grid point pattern. The grid point pattern is formed by connecting grid points, and the vertices of the grid point pattern contain all possible combinations of grid points.

[0034] Project the grid pattern along the acquisition direction to obtain the projection length of the grid pattern in the acquisition direction;

[0035] The acquisition directions are arranged in ascending order of rotation angle at the time of formation to obtain an acquisition direction sequence. The projection lengths of the grid points in the acquisition directions are then arranged in the order of the acquisition direction sequence to obtain a projection length sequence.

[0036] According to the order of the acquisition direction sequence, the lengths of the cargo stack slice outlines corresponding to the cargo stack slices are arranged to obtain the outline length sequence;

[0037] The position with the smallest value in the projection length sequence is taken as the first position. The part of the projection length sequence before the first position is taken as the first front part. The part of the projection length sequence after the first position is taken as the first back part. The first position, the first back part, and the first front part are reassembled in order to obtain the projection correction sequence.

[0038] The position with the smallest value in the contour length sequence is taken as the second position. The part of the contour length sequence before the second position is taken as the second front part. The part of the contour length sequence after the second position is taken as the second back part. The second position, the second back part and the second front part are re-sembedded in order to obtain the contour correction sequence.

[0039] The absolute values ​​of the differences between the values ​​at the same position in the projection correction sequence and the contour correction sequence are accumulated to obtain the judgment value of the grid pattern. The grid pattern with the smallest judgment value is used as the top view of the cargo stack slice.

[0040] The top view of the cargo stack slice is identified in the top view of the cargo stack slice. At least one test point is uniformly selected on the top view of the cargo stack slice. The coordinate model of the test point is performed. The test point is fitted to obtain the contour function of the top view of the cargo stack slice. The area enclosed by the contour function is integrated to obtain the area of ​​the top view of the cargo stack slice.

[0041] The volume of the cargo stack slice is obtained by multiplying the area of ​​the top view contour of the cargo stack slice by a preset distance. The volumes of at least one cargo stack slice are summed to obtain the image volume of the cargo stack.

[0042] The actual height of the cargo stack is divided by the reference value to obtain the side length coefficient. The cube of the side length coefficient is used as the volume coefficient. The volume of the cargo stack image is multiplied by the volume coefficient to obtain the volume of the cargo stack.

[0043] Preferably, obtaining at least one target material present in the cargo pile and analyzing the identifiable number of target materials in the cargo pile includes the following steps:

[0044] Two target standard images are selected as reference images, and the acquisition directions of the two reference images are parallel.

[0045] If one of the images in the verification set of the target local block of the material appears in the reference image, then the material is taken as the target material.

[0046] The position of the image in the verification set of the target local block of the target material in the reference image is taken as the equivalent position of the target local block;

[0047] The number of equivalent locations of the target local block is taken as the number of occurrences of the target local block, and the maximum number of occurrences of the target local block is taken as the number of identifiable targets in the cargo pile.

[0048] Preferably, dividing the volume of the cargo stack into visible and invisible volumes based on the identifiable number includes the following steps:

[0049] Multiply the number of identifiable target materials in the cargo pile by the volume of the target materials to obtain the first volume of the target materials. Sum the first volumes to obtain the visible volume. Subtract the volume of the cargo pile from the visible volume to obtain the invisible volume.

[0050] Preferably, the process of forming at least one volume allocation scheme and generating verification coefficients for the volume allocation scheme includes the following steps:

[0051] The invisible volume is uniformly divided into at least one local volume, and the local volume is randomly allocated to the target material. At least one allocation method forms at least one volume allocation scheme.

[0052] The partial volumes allocated to the target material in the volume allocation scheme are accumulated to obtain the second volume of the target material. The second volume of the target material is divided by the volume of the target material to obtain the verification value of the target material.

[0053] The integer closest to the verification value of the target material is taken as the feature value of the target material;

[0054] The absolute values ​​of the differences between the characteristic values ​​and the verification values ​​of the target material are accumulated to obtain the verification coefficient of the volume distribution scheme.

[0055] Preferably, obtaining the additional quantity of the target material based on the target volume allocation scheme includes the following steps:

[0056] The number of identifiable target materials in the cargo pile is superimposed with the characteristic value of the target materials in the target volume allocation scheme to obtain the new quantity of target materials.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] By forming identification features of materials, obtaining the volume of the cargo pile, determining the identifiable number of target materials, and formulating a target volume allocation scheme, the influence of occlusion on feature recognition can be eliminated. Then, materials that are not completely occluded are counted to form an identifiable number. At the same time, the cargo pile is reconstructed in three dimensions through multi-directional image acquisition, and the size difference between the image and the actual size is eliminated to obtain the volume of the cargo pile. Through volume calculation and verification, the situation of completely occluded materials is estimated, thereby obtaining the number of each type of newly entered material, thus completing the inventory update. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the automatic identification and counting method for material inventory machines based on computer vision according to the present invention.

[0060] Figure 2 This is a schematic diagram of the process for forming the identification features of materials according to the present invention;

[0061] Figure 3 This is a schematic diagram illustrating the process of acquiring images of a pile of goods entering a warehouse to obtain at least one target image according to the present invention.

[0062] Figure 4 This is a schematic diagram of the process of performing standard processing on a target image to obtain a target standard image according to the present invention;

[0063] Figure 5 This is a schematic diagram of the process of segmenting at least one cargo stack slice outline to form a target standard image according to the present invention;

[0064] Figure 6 This is a top view of the process of forming a slice of the cargo stack according to the present invention, and a schematic diagram of the process of calculating the volume of the cargo stack.

[0065] Figure 7 This is a schematic diagram illustrating the process of obtaining at least one target material present in a cargo pile and analyzing the number of identifiable target materials in the cargo pile according to the present invention.

[0066] Figure 8 This is a schematic diagram illustrating the process of forming at least one volume allocation scheme and generating verification coefficients for the volume allocation scheme according to the present invention. Detailed Implementation

[0067] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0068] Reference Figure 1 As shown, the automatic identification and counting method for material inventory machines based on computer vision includes:

[0069] Obtain at least one type of material in the warehouse, obtain the current quantity of the material in the warehouse, and form identification features for the material;

[0070] Images of the goods entering the warehouse are captured to obtain at least one target image. The target image is then processed according to standards to obtain a target standard image.

[0071] Segment at least one cargo stack slice outline to form a target standard image; based on the cargo stack slice outline, form a top view of the cargo stack slice; and calculate the volume of the cargo stack.

[0072] Based on the identification features, at least one target material is found in the cargo pile, and the number of identifiable target materials in the cargo pile is analyzed.

[0073] Based on the number of identifiable items, the volume of the cargo stack is divided into visible volume and invisible volume;

[0074] Based on the invisible volume, at least one volume allocation scheme is formed, and a verification coefficient of the volume allocation scheme is generated. The volume allocation scheme with the smallest verification coefficient is taken as the target volume allocation scheme.

[0075] Based on the target volume allocation scheme, the new quantity of target materials is obtained. The existing quantity of target materials is added to the new quantity to obtain the real-time quantity of target materials.

[0076] Newly received goods are stacked together, and some materials may be partially exposed. Because the stacking obscures certain features, different features will be counted differently. Furthermore, this solution only counts partially obscured or unobscured materials. It is difficult to count materials that are completely obscured. This situation is prone to occur when goods are stacked. Therefore, this solution includes a series of steps to address this issue.

[0077] Reference Figure 2 As shown, the steps to form the identification features of materials include:

[0078] The surface of the material is evenly divided into at least one local block, and a front view image of the local block is obtained;

[0079] At least one sampling point is uniformly selected in the interval (0, 10), and at least one sampling point is uniformly selected in the interval (0, 360).

[0080] The front view image is scaled according to the values ​​of the sampling points to obtain at least one first image. The first image is rotated in the plane where the first image is located according to the values ​​of the sampling points to obtain at least one second image.

[0081] The second images generated from the frontal view images of the local blocks are summarized to form a verification set of the local blocks;

[0082] Take one of the at least one materials as the feature material and the remaining materials as the non-feature materials. If the intersection of the verification sets of two local blocks is empty, then the two local blocks are not associated.

[0083] If the local block of the characteristic material is not associated with any local block of the non-characteristic material, then the local block of the characteristic material is taken as the target local block, and the verification set of the target local block of the characteristic material is used to form the identification feature of the characteristic material.

[0084] When the characteristic material traverses at least one material, the identification characteristics of all materials are formed.

[0085] After scaling and rotation comparison, if one local block is still different from another local block, it can be said that the two local blocks are different; otherwise, they are the same. For this purpose, a verification set of local blocks is formed, which includes the possible scaling ratios and all rotation cases of the local blocks. Therefore, when determining whether two local blocks are the same, the comparison of their verification sets is sufficient. As long as the intersection of their verification sets is not empty, it means that the two are the same. When forming the identification features of materials, it is necessary to ensure that the features used to identify materials are unique, that is, they only appear in the material and do not appear in other materials.

[0086] Reference Figure 3 As shown, acquiring images of a pile of goods entering a warehouse to obtain at least one target image includes the following steps:

[0087] Choose any horizontal line as the baseline, set an arrow at one end of the baseline, take the horizontal plane where the baseline is located as the reference plane, and rotate the baseline within the reference plane according to the values ​​of the collection points to obtain at least one reference line.

[0088] At least one acquisition direction is formed, and the acquisition direction is along the direction of the arrow pointing on the reference line;

[0089] Images of the cargo pile are acquired along the acquisition direction to obtain at least one target image.

[0090] The acquisition direction is obtained by rotating the baseline. Images of the cargo pile are acquired along the acquisition direction, which is equivalent to taking pictures around the cargo pile. Thus, the acquired target images can be used to create a 3D model of the cargo pile.

[0091] Reference Figure 4 As shown, the standard processing of the target image to obtain the target standard image includes the following steps:

[0092] The outline of the cargo stack in the target image is identified, and the height of the cargo stack outline in the target image is used as the reference size of the target image.

[0093] The maximum value of the reference size is used as the baseline value. The baseline value is divided by the reference size of the target image to obtain the scaling ratio of the target image. The target image is scaled according to the scaling ratio of the target image to obtain the target standard image. The acquisition direction of the target image is matched to the target standard image formed by the target image.

[0094] Because the distance to the cargo pile varies during shooting, the image size of the cargo pile outline in the target image is different. It is necessary to standardize the target images to make the image size of the cargo pile outline in the target images the same. Since the actual height of the cargo pile is constant when shooting around the cargo pile, each target image is scaled up so that the height of the cargo pile outline in the target standard image is equal to the reference value after scaling. In this way, the image size in the target standard image is consistent.

[0095] Reference Figure 5 As shown, segmenting at least one cargo stack slice outline to form a target standard image includes the following steps:

[0096] The cargo stack outline in the target standard image is segmented using at least one horizontal parallel line to obtain at least one cargo stack slice outline, wherein the spacing between adjacent parallel lines is a preset distance, which is set based on experience.

[0097] At least one cargo stack slice outline of the target standard image is numbered from top to bottom. Cargo stack slice outlines with the same number correspond to the same cargo stack slice. The acquisition direction corresponding to the target standard image is matched to the cargo stack slice outline generated by the target standard image.

[0098] It is easy to know that cargo stack slices with the same number are cargo stack slices at the same height. Therefore, the only difference between the cargo stack slices is the acquisition direction.

[0099] Reference Figure 6 As shown, forming a top view of a slice of the cargo stack and calculating the volume of the cargo stack includes the following steps:

[0100] The maximum length of the outline of the cargo stack slice is used as a reference value to form a preset area, which is a square with a side length equal to the reference value.

[0101] At least one grid point is uniformly selected within a preset area to form at least one grid point pattern. The grid point pattern is formed by connecting grid points, and the vertices of the grid point pattern contain all possible combinations of grid points.

[0102] Project the grid pattern along the acquisition direction to obtain the projection length of the grid pattern in the acquisition direction;

[0103] The acquisition directions are arranged in ascending order of rotation angle at the time of formation to obtain an acquisition direction sequence. The projection lengths of the grid points in the acquisition directions are then arranged in the order of the acquisition direction sequence to obtain a projection length sequence.

[0104] According to the order of the acquisition direction sequence, the lengths of the cargo stack slice outlines corresponding to the cargo stack slices are arranged to obtain the outline length sequence;

[0105] The position with the smallest value in the projection length sequence is taken as the first position. The part of the projection length sequence before the first position is taken as the first front part. The part of the projection length sequence after the first position is taken as the first back part. The first position, the first back part, and the first front part are reassembled in order to obtain the projection correction sequence.

[0106] The position with the smallest value in the contour length sequence is taken as the second position. The part of the contour length sequence before the second position is taken as the second front part. The part of the contour length sequence after the second position is taken as the second back part. The second position, the second back part and the second front part are re-sembedded in order to obtain the contour correction sequence.

[0107] The absolute values ​​of the differences between the values ​​at the same position in the projection correction sequence and the contour correction sequence are accumulated to obtain the judgment value of the grid pattern. The grid pattern with the smallest judgment value is used as the top view of the cargo stack slice.

[0108] The top view of the cargo stack slice is identified in the top view of the cargo stack slice. At least one test point is uniformly selected on the top view of the cargo stack slice. The coordinate model of the test point is performed. The test point is fitted to obtain the contour function of the top view of the cargo stack slice. The area enclosed by the contour function is integrated to obtain the area of ​​the top view of the cargo stack slice.

[0109] The volume of the cargo stack slice is obtained by multiplying the area of ​​the top view contour of the cargo stack slice by a preset distance. The volumes of at least one cargo stack slice are summed to obtain the image volume of the cargo stack.

[0110] The actual height of the cargo stack is divided by the reference value to obtain the side length coefficient. The cube of the side length coefficient is used as the volume coefficient. The volume of the cargo stack image is multiplied by the volume coefficient to obtain the volume of the cargo stack.

[0111] The image size of the top view of the cargo stack slice is the same as the image size of the target standard image. Therefore, if the top view of the cargo stack slice is projected in the acquisition direction, the length of the projection is the same as the length of the cargo stack slice outline in the target standard image acquired in the same acquisition direction. Therefore, this principle can be used to select the grid point pattern to obtain the top view of the cargo stack slice.

[0112] Each cargo stack slice outline corresponds to a collection direction. Therefore, the lengths of the cargo stack slice outlines corresponding to the cargo stack slices can be arranged according to their corresponding collection directions and in the order of the collection direction sequence to obtain the outline length sequence.

[0113] When comparing them here, it is assumed that the grid pattern is the top view of the cargo stack slice. However, since there may be a difference in the rotation angle between the grid pattern and the cargo stack slice, the sequence of its projected lengths in the acquisition direction cannot be compared with the contour length sequence. It is necessary to rotate it until there is no difference with the cargo stack slice before comparison can be performed.

[0114] Therefore, after forming the projection length sequence and the contour length sequence, the two sequences are processed so that the position with the smallest length in the sequence is placed at the beginning of the sequence, and the remaining parts are changed in order. This is equivalent to adjusting the projection length sequence and the contour length sequence to be in the same state, that is, there is no rotation difference between the two. If the grid pattern is the top view of the cargo stack slice, the difference of the adjusted sequence is the smallest, that is, the judgment value is the smallest. Therefore, the required grid pattern can be selected as the top view of the cargo stack slice.

[0115] However, the image size of the top view of the cargo stack slice is consistent with the image size of the target standard image. Therefore, the ratio of the image volume of the cargo stack to the actual volume of the cargo stack is equal to the cube of the ratio of its side lengths. Based on this, the actual volume of the cargo stack can be calculated.

[0116] Reference Figure 7 As shown, obtaining at least one target material present in the cargo pile and analyzing the identifiable number of target materials in the cargo pile includes the following steps:

[0117] Two target standard images are selected as reference images, and the acquisition directions of the two reference images are parallel.

[0118] If one of the images in the verification set of the target local block of the material appears in the reference image, then the material is taken as the target material.

[0119] The position of the image in the verification set of the target local block of the target material in the reference image is taken as the equivalent position of the target local block;

[0120] The number of equivalent locations of the target local block is taken as the number of occurrences of the target local block, and the maximum number of occurrences of the target local block is taken as the number of identifiable targets in the cargo pile.

[0121] For materials that are not completely obscured or are not obscured, the target partial block of the material may or may not appear because it may be obscured. As a result, the count of different target partial blocks of the material will be different. Therefore, by taking the maximum number of occurrences of the target partial block as the number of identifiable target materials in the cargo pile, we can count materials that are not completely obscured or are not obscured.

[0122] Based on the identifiable number of items, the volume of the cargo stack is divided into visible volume and invisible volume, including the following steps:

[0123] Multiply the number of identifiable target materials in the cargo pile by the volume of the target materials to obtain the first volume of the target materials. Sum the first volumes to obtain the visible volume. Subtract the volume of the cargo pile from the visible volume to obtain the invisible volume.

[0124] When counting, it is also necessary to count materials that are completely obscured, as this cannot be done through feature recognition.

[0125] Reference Figure 8 As shown, forming at least one volume allocation scheme and generating verification coefficients for the volume allocation scheme includes the following steps:

[0126] The invisible volume is uniformly divided into at least one local volume, and the local volume is randomly allocated to the target material. At least one allocation method forms at least one volume allocation scheme.

[0127] The partial volumes allocated to the target material in the volume allocation scheme are accumulated to obtain the second volume of the target material. The second volume of the target material is divided by the volume of the target material to obtain the verification value of the target material.

[0128] The integer closest to the verification value of the target material is taken as the feature value of the target material;

[0129] The absolute values ​​of the differences between the characteristic values ​​and the verification values ​​of the target material are accumulated to obtain the verification coefficient of the volume distribution scheme.

[0130] The invisible volume is allocated to resources. Therefore, if the invisible volume is perfectly accurate, the volume allocated to resources must be an integer multiple of the resource's own volume. However, since the acquisition of the invisible volume is not perfectly accurate, the volume allocated to resources must be closest to an integer multiple of the resource's own volume. Therefore, this is used as a basis to determine the scheme for allocating the invisible volume to resources, and then, based on the allocation of volume, the number of resources that are completely obscured is determined.

[0131] Based on the target volume allocation scheme, the additional quantity of target materials is obtained through the following steps:

[0132] The number of identifiable target materials in the cargo pile is superimposed with the characteristic value of the target materials in the target volume allocation scheme to obtain the new quantity of target materials.

[0133] Furthermore, this solution also proposes a storage medium on which a computer-readable program is stored. When the computer-readable program is invoked, the aforementioned computer vision-based automatic identification and counting method for material inventory machines is executed.

[0134] It is understandable that the storage medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid-state drive (SSD).

[0135] In summary, the advantages of this invention are as follows: by forming identification features of materials, obtaining the volume of the goods pile, obtaining the identifiable number of target materials, and forming a target volume allocation scheme, the influence of occlusion on feature recognition can be eliminated. Furthermore, materials that are not completely occluded are counted to form an identifiable number. At the same time, by acquiring images from multiple angles, the goods pile is reconstructed in three dimensions, and the size difference between the image and the actual size is eliminated, thereby obtaining the volume of the goods pile. Through volume calculation and verification, the situation of completely occluded materials is estimated, thereby obtaining the number of each type of newly entered material, thus completing the inventory update.

[0136] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A computer vision-based automatic identification and counting method for material inventory machines, characterized in that, include: Obtain at least one type of material in the warehouse, obtain the current quantity of the material in the warehouse, and form identification features for the material; Images of the goods entering the warehouse are captured to obtain at least one target image. The target image is then processed according to standards to obtain a target standard image. Segment at least one cargo stack slice outline to form a target standard image; based on the cargo stack slice outline, form a top view of the cargo stack slice; and calculate the volume of the cargo stack. Based on the identification features, at least one target material is found in the cargo pile, and the number of identifiable target materials in the cargo pile is analyzed. Based on the number of identifiable items, the volume of the cargo stack is divided into visible volume and invisible volume; Based on the invisible volume, at least one volume allocation scheme is formed, and a verification coefficient of the volume allocation scheme is generated. The volume allocation scheme with the smallest verification coefficient is taken as the target volume allocation scheme. Based on the target volume allocation scheme, the new quantity of target materials is obtained. The existing quantity of target materials is added to the new quantity to obtain the real-time quantity of target materials. The process of forming a top view of the cargo stack slice and calculating the volume of the cargo stack includes the following steps: The maximum length of the outline of the cargo stack slice is used as a reference value to form a preset area, which is a square with a side length equal to the reference value. At least one grid point is uniformly selected within a preset area to form at least one grid point pattern. The grid point pattern is formed by connecting grid points, and the vertices of the grid point pattern contain all possible combinations of grid points. Project the grid pattern along the acquisition direction to obtain the projection length of the grid pattern in the acquisition direction; The acquisition directions are arranged in ascending order of rotation angle at the time of formation to obtain an acquisition direction sequence. The projection lengths of the grid points in the acquisition directions are then arranged in the order of the acquisition direction sequence to obtain a projection length sequence. According to the order of the acquisition direction sequence, the lengths of the cargo stack slice outlines corresponding to the cargo stack slices are arranged to obtain the outline length sequence; The position with the smallest value in the projection length sequence is taken as the first position. The part of the projection length sequence before the first position is taken as the first front part. The part of the projection length sequence after the first position is taken as the first back part. The first position, the first back part, and the first front part are reassembled in order to obtain the projection correction sequence. The position with the smallest value in the contour length sequence is taken as the second position. The part of the contour length sequence before the second position is taken as the second front part. The part of the contour length sequence after the second position is taken as the second back part. The second position, the second back part and the second front part are re-sembedded in order to obtain the contour correction sequence. The absolute values ​​of the differences between the values ​​at the same position in the projection correction sequence and the contour correction sequence are accumulated to obtain the judgment value of the grid pattern. The grid pattern with the smallest judgment value is used as the top view of the cargo stack slice. The top view of the cargo stack slice is identified in the top view of the cargo stack slice. At least one test point is uniformly selected on the top view of the cargo stack slice. The coordinate model of the test point is performed. The test point is fitted to obtain the contour function of the top view of the cargo stack slice. The area enclosed by the contour function is integrated to obtain the area of ​​the top view of the cargo stack slice. The volume of the cargo stack slice is obtained by multiplying the area of ​​the top view contour of the cargo stack slice by a preset distance. The volumes of at least one cargo stack slice are summed to obtain the image volume of the cargo stack. The actual height of the cargo stack is divided by the reference value to obtain the side length coefficient. The cube of the side length coefficient is used as the volume coefficient. The volume of the cargo stack image is multiplied by the volume coefficient to obtain the volume of the cargo stack. The process of forming at least one volume allocation scheme and generating verification coefficients for the volume allocation scheme includes the following steps: The invisible volume is uniformly divided into at least one local volume, and the local volume is randomly allocated to the target material. At least one allocation method forms at least one volume allocation scheme. The partial volumes allocated to the target material in the volume allocation scheme are accumulated to obtain the second volume of the target material. The second volume of the target material is divided by the volume of the target material to obtain the verification value of the target material. The integer closest to the verification value of the target material is taken as the feature value of the target material; The absolute values ​​of the differences between the characteristic values ​​and the verification values ​​of the target material are accumulated to obtain the verification coefficient of the volume distribution scheme.

2. The automatic identification and counting method for material inventory machines based on computer vision according to claim 1, characterized in that, The identification features of the material formation include the following steps: The surface of the material is evenly divided into at least one local block, and a front view image of the local block is obtained; At least one sampling point is uniformly selected in the interval (0, 10), and at least one sampling point is uniformly selected in the interval (0, 360). The front view image is scaled according to the values ​​of the sampling points to obtain at least one first image. The first image is rotated in the plane where the first image is located according to the values ​​of the sampling points to obtain at least one second image. The second images generated from the frontal view images of the local blocks are summarized to form a verification set of the local blocks; Take one of the at least one materials as the feature material and the remaining materials as the non-feature materials. If the intersection of the verification sets of two local blocks is empty, then the two local blocks are not associated. If the local block of the characteristic material is not associated with any local block of the non-characteristic material, then the local block of the characteristic material is taken as the target local block, and the verification set of the target local block of the characteristic material is used to form the identification feature of the characteristic material. When the characteristic material traverses at least one material, the identification characteristics of all materials are formed.

3. The automatic identification and counting method for material inventory machines based on computer vision according to claim 2, characterized in that, The process of acquiring images of the piles of goods entering the warehouse to obtain at least one target image includes the following steps: Choose any horizontal line as the baseline, set an arrow at one end of the baseline, take the horizontal plane where the baseline is located as the reference plane, and rotate the baseline within the reference plane according to the values ​​of the collection points to obtain at least one reference line. At least one acquisition direction is formed, and the acquisition direction is along the direction of the arrow pointing on the reference line; Images of the cargo pile are acquired along the acquisition direction to obtain at least one target image.

4. The automatic identification and counting method for material inventory machines based on computer vision according to claim 3, characterized in that, The standard processing of the target image to obtain the target standard image includes the following steps: The outline of the cargo stack in the target image is identified, and the height of the cargo stack outline in the target image is used as the reference size of the target image. The maximum value of the reference size is used as the baseline value. The baseline value is divided by the reference size of the target image to obtain the scaling ratio of the target image. The target image is scaled according to the scaling ratio of the target image to obtain the target standard image. The acquisition direction of the target image is matched to the target standard image formed by the target image.

5. The automatic identification and counting method for material inventory machines based on computer vision according to claim 4, characterized in that, The segmentation process to form at least one cargo stack slice outline of the target standard image includes the following steps: The cargo stack outline in the target standard image is segmented using at least one horizontal parallel line to obtain at least one cargo stack slice outline, wherein the spacing between adjacent parallel lines is a preset distance, which is set based on experience. At least one cargo stack slice outline of the target standard image is numbered from top to bottom. Cargo stack slice outlines with the same number correspond to the same cargo stack slice. The acquisition direction corresponding to the target standard image is matched to the cargo stack slice outline generated by the target standard image.

6. The automatic identification and counting method for material inventory machines based on computer vision according to claim 5, characterized in that, The process of obtaining at least one target material present in the cargo pile and analyzing the number of identifiable target materials in the cargo pile includes the following steps: Two target standard images are selected as reference images, and the acquisition directions of the two reference images are parallel. If one of the images in the verification set of the target local block of the material appears in the reference image, then the material is taken as the target material. The position of the image in the verification set of the target local block of the target material in the reference image is taken as the equivalent position of the target local block; The number of equivalent locations of the target local block is taken as the number of occurrences of the target local block, and the maximum number of occurrences of the target local block is taken as the number of identifiable targets in the cargo pile.

7. The automatic identification and counting method for material inventory machines based on computer vision according to claim 6, characterized in that, The process of dividing the volume of a stack of goods into visible and invisible volumes based on the number of identifiable items includes the following steps: Multiply the number of identifiable target materials in the cargo pile by the volume of the target materials to obtain the first volume of the target materials. Sum the first volumes to obtain the visible volume. Subtract the volume of the cargo pile from the visible volume to obtain the invisible volume.

8. The automatic identification and counting method for material inventory machines based on computer vision according to claim 7, characterized in that, The method for obtaining the additional quantity of target materials based on the target volume allocation scheme includes the following steps: The number of identifiable target materials in the cargo pile is superimposed with the characteristic value of the target materials in the target volume allocation scheme to obtain the new quantity of target materials.

Images