A method and system for digital inventory of reusable items

Abstract

This invention relates to the field of data management technology and discloses a method and system for digital inventory of reusable items. The method includes: performing mean filtering on the original image of the inventory area to obtain standard image data of the inventory area; establishing a three-dimensional view benchmark library for reusable items; performing adaptive matching statistics between the standard image data and the three-dimensional view benchmark library to obtain the initial pose and initial quantity of the reusable items; performing state discrimination on the spatial relationship between items in the standard image data to obtain a dataset of items to be confirmed in the standard image data; performing multi-view feature analysis on the item areas in the dataset of items to be confirmed, and performing simulation verification on the initial pose to obtain verification data for the dataset of items to be confirmed; and performing collaborative evaluation on the initial quantity and verification data to obtain a digital inventory report of reusable items. This invention can improve the efficiency of digital inventory of reusable items.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a method and system for digital inventory of recyclable items. Background Technology

[0002] Existing digital inventory technologies for reusable items have significant shortcomings in image preprocessing. Noise suppression and feature enhancement of the original images of the inventory area are insufficient, failing to effectively eliminate image distortion caused by environmental interference. This results in inaccurate extracted item feature data, creating potential errors in subsequent matching and statistical processes. Furthermore, the construction of the 3D view benchmark library lacks a systematic integration of prior digital knowledge of reusable items with their common placement orientations. The multi-view coverage and standardization of the benchmark data are low, failing to provide comprehensive and accurate reference for item matching in different scenarios and limiting the adaptability of the matching process.

[0003] In the item identification and inventory verification stages, the limitations of existing technologies are even more pronounced. Their feature comparison methods are mostly limited to a single dimension or fixed pattern, making it difficult to handle different pose changes and complex placement scenarios for recurring items, resulting in significant errors in initial pose localization and quantity counting. Furthermore, there is a lack of effective state discrimination mechanisms for complex spatial relationships commonly found between items, such as occlusion, stacking, and boundary adhesion. Moreover, no targeted multi-view feature verification process has been established, making it impossible to accurately review and calibrate questionable items. Ultimately, this leads to insufficient completeness and reliability of inventory results, significantly reducing the efficiency and quality of digital inventory and failing to meet the actual needs for efficient and accurate inventory of recurring items. Summary of the Invention

[0004] This invention provides a method and system for digital inventory of recyclable items to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for digital inventory of reusable items, comprising: S1. Perform mean filtering on the original image of the inventory area to obtain standard image data of the inventory area; S2. Based on the conventional placement posture of the recyclable items and prior digital knowledge, establish a three-dimensional view reference library for the recyclable items; S3. Perform adaptive matching statistics between the standard image data and the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item; S4. Based on the initial pose, the spatial relationship between items in the standard image data is determined to obtain the dataset of items to be confirmed in the standard image data. S5. Perform multi-view feature analysis on the item regions in the dataset of items to be confirmed, and perform simulation verification on the initial pose based on the analysis results to obtain the verification data of the dataset of items to be confirmed. S6. Perform a collaborative evaluation of the initial quantity and the verification data to obtain a digital inventory report of the recyclable items.

[0006] In a preferred embodiment, the step of performing mean filtering on the original image of the inventory area to obtain standard image data of the inventory area includes: Obtain the original image of the inventory area. The original image is adjusted in grayscale to obtain the enhanced image data of the original image; The enhanced image data is subjected to edge filtering to obtain a smooth image of the inventory area; The smoothed image is subjected to mean filtering to obtain standard image data of the inventory area.

[0007] In a preferred embodiment, establishing a three-dimensional view reference library for the reusable items based on their conventional placement and prior digital knowledge includes: Parameter mining is performed on the three-dimensional geometric information and surface texture information of the circular item to obtain the prior digital knowledge data of the circular item; Based on the conventional placement posture of the cyclical items, the prior digital knowledge data is subjected to multi-view projection rendering to obtain multi-view reference data of the cyclical items. Key features are labeled on the multi-view benchmark data to obtain standardized feature description data of the cyclic items; The standardized feature description data, the multi-view benchmark data, and the prior digital knowledge data are associated and stored to obtain the three-dimensional view benchmark library of the cyclic item.

[0008] In a preferred embodiment, the step of adaptively matching and statistically analyzing the standard image data with the 3D view reference library to obtain the initial pose and initial quantity of the cyclic items includes: Multi-scale feature extraction is performed on the standard image data to obtain a global contour feature point set of the standard image data; Extract the reference feature set of the two-dimensional standard view from the three-dimensional view reference library; The global contour feature point set is compared with the reference feature set in a similarity traversal, and a mapping relationship between the standard image data and the two-dimensional standard view is established based on the comparison results. Based on the mapping relationship, geometric consistency screening is performed on the global contour feature point set to obtain the reliable mapping pairs of the cyclic items; The position and rotation angle of the cyclic item in the standard image data are determined based on the trusted mapping pair, which serves as the initial pose of the cyclic item. Based on the initial pose, instances of the same cyclic item in the standard image data are aggregated into regions, and the number of independent regions after aggregation is used as the initial number of cyclic items.

[0009] In a preferred embodiment, the similarity score in the similarity traversal comparison is calculated using the following formula: ; In the formula, The similarity score is given. The total number of matching feature point pairs in the mapping relationship. The summation index for matching feature points. The preset appearance similarity weighting coefficients, For the first The Euclidean distance between the matching feature points in the feature space, The preset distance scale parameters, The preset geometric similarity weight coefficients, For the first The change in the angle between the matched feature point and other matching point pairs in its neighborhood, representing the local geometric structure. It is a natural exponential function. For geometric consistency function, For the first The length of the longest continuous contour segment for matching feature points. This is the reward function for contour continuity.

[0010] In a preferred embodiment, the step of determining the spatial relationships between items in the standard image data based on the initial pose to obtain a dataset of items to be confirmed from the standard image data includes: Based on the initial pose, determine the candidate image region of the cyclic item in the standard image data; Spatial evaluation is performed on the relative positional relationship and boundary overlap state between the candidate image regions to obtain the three-dimensional state information of the circular items. The three-dimensional state information includes whether there is occlusion, stacking and boundary adhesion between the circular items. Based on the three-dimensional state information, candidate items in the standard image data that are in an occluded state, a highly stacked state, or a state with blurred boundaries are identified. The candidate items and candidate image regions in the occluded state, the highly stacked state, and the blurred boundary state are used to obtain the standard image data dataset of items to be confirmed.

[0011] In a preferred embodiment, the step of performing multi-view feature analysis on the item regions in the dataset of items to be confirmed, and performing simulation verification on the initial pose based on the analysis results to obtain verification data for the dataset of items to be confirmed, includes: Based on the item region in the dataset of items to be confirmed and the initial pose, a virtual observation view surrounding the item region is generated; Under the virtual observation perspective, the cyclic items in the 3D view reference library are subjected to 3D geometric information detection to obtain the simulated contour features and simulated texture features of the cyclic items; Based on the simulated contour features and the simulated texture features, feature adaptation and extraction are performed on the standard image data to obtain the actual image content of the standard image data; The simulated texture features and simulated contour features are compared and evaluated with the actual image content to obtain the matching degree evaluation result of the circular item; Based on the matching degree evaluation result, the initial pose is simulated and iteratively fine-tuned until the matching degree evaluation result meets the preset consistency condition. The final optimized pose that satisfies the consistency condition is recorded as the verified pose, and the verified pose and the matching degree evaluation result are integrated into the verification data of the dataset of items to be confirmed.

[0012] In a preferred embodiment, the step of performing three-dimensional geometric information detection on the circular object in the three-dimensional view reference library under the virtual observation perspective to obtain the simulated contour features and simulated texture features of the circular object includes: Based on the virtual observation perspective, the three-dimensional geometric information is observed and projected to obtain the depth map and surface visibility mask of the circular item; In the depth map and the surface visibility mask, the external geometric boundary of the circular object under the virtual observation view is extracted as the simulation contour data of the circular object; Key point detection is performed on the simulated contour data to obtain the simulated contour features of the cyclic item; Based on the surface texture information associated with the three-dimensional geometric information and the virtual observation viewpoint, a simulated texture image of the cyclic item under the virtual observation viewpoint is obtained through texture mapping and sampling. The simulated texture image is divided into regions and statistical features are extracted to obtain the simulated texture features of the cyclic item.

[0013] In a preferred embodiment, the step of co-evaluating the initial quantity and the verification data to obtain a digital inventory report of the recyclable items includes: The initial quantity is compared with the quantity of items verified in the verification data to obtain the quantity difference information of the cyclic items; Based on the quantity difference information, determine whether the initial quantity is consistent with the confirmed quantity in the verification data; If the results are consistent, the initial quantity is taken as the final inventory quantity of the recycle item, and the final inventory quantity and the verification data are integrated into the inventory result data of the recycle item. If an inconsistency is determined, the dataset of items to be confirmed is re-statistically analyzed and verified based on the verification data to obtain the corrected inventory count of the cyclical items. The corrected inventory count, the quantity difference information, and the verified item pose are then integrated into the inventory result data of the cyclical items. The inventory results data are formatted and packaged according to the preset report template to obtain a digital inventory report of the reusable items.

[0014] To address the above problems, the present invention also provides a digital inventory system for recyclable items, the system comprising: The image preprocessing module is used to perform mean filtering on the original image of the inventory area to obtain standard image data of the inventory area; The 3D view reference library construction module is used to establish a 3D view reference library for the cyclical items based on the conventional placement posture of the cyclical items and prior digital knowledge. The initial identification and positioning module is used to adaptively match and statistically analyze the standard image data with the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item. The item spatial relationship discrimination module is used to perform state discrimination on the spatial relationship between items in the standard image data based on the initial pose, and obtain the item dataset to be confirmed in the standard image data; The multi-view feature verification module is used to perform multi-view feature analysis on the item regions in the item dataset to be confirmed, and to perform simulation verification on the initial pose based on the analysis results, so as to obtain the verification data of the item dataset to be confirmed. The comprehensive report generation module is used to collaboratively evaluate the initial quantity and the verification data to obtain a digital inventory report of the recyclable items.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention employs a progressive preprocessing workflow involving grayscale adjustment, edge filtering, and mean filtering on the original image of the inventory area. This effectively removes image noise, enhances the recognizability of item features, and significantly improves the quality and stability of standard image data, providing highly reliable foundational data support for accurate processing in subsequent stages. Simultaneously, a 3D view benchmark library is constructed based on the 3D geometric information, surface texture information, and conventional placement posture of the cyclical items. This integrates multi-view benchmark data with standardized feature descriptions and stores them in association, giving the benchmark references more comprehensive coverage and accurate matching adaptability, providing a solid reference foundation for rapid item identification.

[0016] 2. This invention utilizes an adaptive matching mechanism combining multi-scale feature extraction and multi-dimensional similarity comparison to accurately pinpoint the initial pose of recurring items and complete initial quantity counting, thus improving the initial accuracy of item location and counting. Items in complex states are screened out through spatial relationship discrimination, and a verification process involving multi-view virtual observation, simulation feature comparison, and iterative pose fine-tuning enables precise review and calibration of questionable items, ensuring the completeness and accuracy of the inventory results. Finally, a standardized digital inventory report is generated through the collaborative evaluation of initial and verification data, constructing an efficient and closed-loop inventory process that significantly improves the overall efficiency and reliability of digital inventory of recurring items. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for digital inventory of recyclable items according to an embodiment of the present invention. Figure 2 This is a functional module diagram of a digital inventory system for recyclable items provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a method for digitally inventorying reusable items. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for digitally inventorying reusable items can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a method for digitally inventorying reusable items according to an embodiment of the present invention. In this embodiment, the method for digitally inventorying reusable items includes: S1. Perform mean filtering on the original image of the inventory area to obtain standard image data of the inventory area; In this embodiment of the invention, the step of performing mean filtering on the original image of the inventory area to obtain standard image data of the inventory area includes: Obtain the original image of the inventory area. The original image is adjusted in grayscale to obtain the enhanced image data of the original image; The enhanced image data is subjected to edge filtering to obtain a smooth image of the inventory area; The smoothed image is subjected to mean filtering to obtain standard image data of the inventory area.

[0021] The image acquisition device is used to capture images of a preset inventory area. During the capture process, the relative position of the image acquisition device and the inventory area is kept fixed. The focal length and exposure parameters of the image acquisition device are set to fixed values ​​that match the characteristics of the items in the inventory area. This ensures that the captured image completely covers the preset inventory area, and that the outline and texture information of the inventory items in the image are clearly distinguishable. Finally, the original image of the inventory area is obtained.

[0022] The red, green, and blue color channel values ​​of each pixel in the original image are weighted and calculated by multiplying the red channel value by 0.299, the green channel value by 0.587, and the blue channel value by 0.114. The three products are then added together to obtain the grayscale value of each pixel. This grayscale value replaces the corresponding three color channel values ​​of the original image, completing the grayscale conversion of the original image. Subsequently, the contrast of the converted grayscale image is adjusted to map the grayscale value of each pixel in the grayscale image to a grayscale range of 0 to 255, making the grayscale difference between objects and the background in the image clearer. Finally, the enhanced image data of the original image is obtained.

[0023] A filter template adapted to the pixel size of the enhanced image data is selected, and the filter template is slid pixel by pixel on the pixel matrix of the enhanced image data. During the sliding process, each pixel covered by the filter template is compared with the gray values ​​of adjacent pixels in the template based on its own gray value. The gray value characteristics of the object edge are preserved, while the pixel differences in areas with small gray value fluctuations in the template are weakened. After the sliding is completed, the gray values ​​of all pixels processed by the template are integrated to eliminate false edges caused by noise in the enhanced image data, and finally a smooth image of the inventory area is obtained.

[0024] A 3×3 filtering window is selected and moved pixel by pixel across the pixel matrix of the smoothed image. During the movement, the average gray value of all pixels within the coverage area of ​​the filtering window is calculated. The gray values ​​of all pixels within the window are added together, and the sum is divided by the total number of pixels within the window. The calculated average value is used to replace the gray value of the pixel corresponding to the center position of the filtering window. After the window has moved to all pixel positions in the smoothed image and the replacement operation has been completed, the resulting image is the standard image data of the inventory area.

[0025] The beneficial effects include ensuring the integrity and clarity of the original image of the inventory area, improving the enhancement effect of the original image after grayscale adjustment, strengthening the distinction between items and background, eliminating noise and false edges in the enhanced image, achieving smooth image processing, further optimizing the uniformity of image pixel grayscale distribution, and finally obtaining standard image data of the inventory area with stable quality to meet the needs of subsequent related processing in this area.

[0026] S2. Based on the conventional placement posture of the recyclable items and prior digital knowledge, establish a three-dimensional view reference library for the recyclable items; In this embodiment of the invention, establishing a three-dimensional view reference library for the reusable items based on their conventional placement and prior digital knowledge includes: Parameter mining is performed on the three-dimensional geometric information and surface texture information of the circular item to obtain the prior digital knowledge data of the circular item; Based on the conventional placement posture of the cyclical items, the prior digital knowledge data is subjected to multi-view projection rendering to obtain multi-view reference data of the cyclical items. Key features are labeled on the multi-view benchmark data to obtain standardized feature description data of the cyclic items; The standardized feature description data, the multi-view benchmark data, and the prior digital knowledge data are associated and stored to obtain the three-dimensional view benchmark library of the cyclic item.

[0027] A full-dimensional 3D scan is performed on the recyclable item, and spatial coordinate information of the item's surface is collected point by point to obtain 3D geometric information. At the same time, information such as color distribution, texture direction, and unevenness of the item's surface is collected to obtain surface texture information. The collected 3D geometric information and surface texture information are classified and organized, invalid collected data are removed, and effective information that can completely represent the physical form of the recyclable item is retained to form a structured data set that can be directly used for subsequent processing. Finally, the prior digital knowledge data of the recyclable item is obtained.

[0028] The most frequent placement state of the cyclical item in the actual application scenario is determined as the normal placement posture. Based on this normal placement posture, the spatial position of the cyclical item is fixed. With the geometric center of the item as the origin of the viewpoint, the prior digital knowledge data is projected along the horizontal and vertical directions according to the evenly distributed viewpoint directions. During the projection process, the geometric and texture features in the prior digital knowledge data are completely preserved. The rendering optimization of the two-dimensional view formed by the projection under each viewpoint is performed to make the outline and texture details of the item in the viewpoint clearly presented. The two-dimensional views rendered under all views are summarized and organized to finally obtain the multi-view reference data of the cyclical item.

[0029] For each view in the multi-view benchmark data, key elements that can characterize the core form of the cyclical item are identified, including the item's outline boundary points, surface texture features, structural component splicing positions, and geometric abrupt change areas. The location information and morphological features of each identified key element are clearly described, and the description is bound to the corresponding view. At the same time, the descriptions of key elements in all views are standardized according to a unified feature description rule to eliminate description differences between different views, and finally, standardized feature description data of the cyclical item is obtained.

[0030] A unique identifier is assigned to each cyclical item. Standardized feature description data, multi-view benchmark data, and prior digital knowledge data are associated with this identifier to establish a one-to-one correspondence between the three types of data. A hierarchical storage architecture is used to store the associated data. The first layer stores the identifier and category information of the cyclical item; the second layer stores the prior digital knowledge data corresponding to the identifier; and the third layer stores the corresponding multi-view benchmark data and standardized feature description data. During the storage process, the integrity and retrieval of the data are ensured, so that any associated data can be quickly retrieved through the identifier. Finally, a three-dimensional view benchmark library of the cyclical item is obtained.

[0031] The beneficial effects include obtaining complete and effective three-dimensional geometric and surface texture information of recyclable items, generating multi-view benchmark data that clearly presents the characteristics of items based on conventional placement postures, forming standardized feature description data with unified specifications through key feature annotation, and achieving orderly integration and rapid retrieval of various types of data through associated storage. The final three-dimensional view benchmark library can provide stable and reliable data support for the subsequent identification and management of recyclable items.

[0032] S3. Perform adaptive matching statistics between the standard image data and the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item; In this embodiment of the invention, the step of adaptively matching and statistically analyzing the standard image data with the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item includes: Multi-scale feature extraction is performed on the standard image data to obtain a global contour feature point set of the standard image data; Extract the reference feature set of the two-dimensional standard view from the three-dimensional view reference library; The global contour feature point set is compared with the reference feature set in a similarity traversal, and a mapping relationship between the standard image data and the two-dimensional standard view is established based on the comparison results. Based on the mapping relationship, geometric consistency screening is performed on the global contour feature point set to obtain the reliable mapping pairs of the cyclic items; The position and rotation angle of the cyclic item in the standard image data are determined based on the trusted mapping pair, which serves as the initial pose of the cyclic item. Based on the initial pose, instances of the same cyclic item in the standard image data are aggregated into regions, and the number of independent regions after aggregation is used as the initial number of cyclic items.

[0033] The formula for calculating the similarity score in the similarity traversal comparison is as follows: ; In the formula, The similarity score is given. The total number of matching feature point pairs in the mapping relationship. The summation index for matching feature points. The preset appearance similarity weighting coefficients, For the first The Euclidean distance between the matching feature points in the feature space, The preset distance scale parameters, The preset geometric similarity weight coefficients, For the first The change in the angle between the matched feature point and other matching point pairs in its neighborhood, representing the local geometric structure. It is a natural exponential function. For geometric consistency function, For the first The length of the longest continuous contour segment for matching feature points. This is the reward function for contour continuity.

[0034] The similarity score is a quantified value obtained by summing and averaging the results of each pair of matching feature points in the global contour feature point set and each pair of matching feature points in the baseline feature set. The total number of matching feature point pairs is a specific value obtained by counting each matching feature point pair that has a one-to-one correspondence in the mapping relationship established between the global contour feature point set and the baseline feature set.

[0035] The summation index of matching feature points is an identifier used to sequentially traverse the pairs of matching feature points in the mapping relationship, starting from the first pair and corresponding to the last pair. The appearance similarity weight coefficient is based on historical feature matching data of the two-dimensional standard view of the cyclic item and the actual collected standard image data. The similarity comparison results under different weight coefficients are statistically analyzed to determine the degree of fit between the actual matching results. The value with a fit of 100% is selected as a fixed preset value, and the sum of this value and the preset value of the geometric similarity weight coefficient is 1.

[0036] No. The Euclidean distance between matching feature points in the feature space is obtained by extracting the first... For the 3D coordinates of the matching feature points in the feature space, the coordinate differences between the two feature points on the three coordinate axes are calculated. The squares of each coordinate difference are then summed, and the sum is squared to obtain the final value. The distance scale parameter is a fixed preset value obtained by calculating the arithmetic mean of the Euclidean distances of all matching feature point pairs in the historical matching process. This value is used to adjust the influence of the Euclidean distance on the similarity score.

[0037] The geometric consistency function is a monotonically decreasing function of local geometric deformation. Specifically, it is calculated by exponentiation of the negative absolute value of the local geometric deformation with the natural constant as the base. The smaller the local geometric deformation, the closer the function output value is to 1; the larger the local geometric deformation, the closer the function output value is to 0. The function's value is derived from the calculation of the change in the angle between the local geometric structure formed by each pair of matching feature points and other matching point pairs in their neighborhood, used to measure the local geometric similarity of matching point pairs. The geometric similarity weight is a preset weight coefficient that works in conjunction with the appearance similarity weight to adjust the contribution of local geometric consistency to the overall similarity score. The value of this parameter is based on historical geometric structure matching data of the two-dimensional standard view of the cyclic item and the actual collected standard image data. The matching results under different weights are statistically analyzed to determine the degree of fit with the actual results. A value with a fit of 100% is selected as a fixed preset value, and the sum of this value and the preset value of the appearance similarity weight is 1, remaining constant throughout the similarity traversal comparison process. (Including the...) The length of the longest continuous contour segment for matching feature points is determined by selecting the longest continuous contour segment from the global contour feature point set of the standard image data. Starting from the pixel corresponding to the matching feature point, traverse adjacent contour feature points along the two extension directions of the contour until a non-contour feature point is encountered. Count the number of consecutive contour feature points in the two directions respectively, and select the value of the side with the larger number as the fixed value to evaluate the continuity of the matching point in spatial distribution.

[0038] The contour continuity reward function is a function that includes the first contour. The function is a monotonically increasing function for the length of the longest continuous contour segment matching feature points. Specifically, it is calculated as the smaller of 1 and the product of the reward coefficient and the length of that longest continuous contour segment divided by the total contour length. The value of this function is derived from the sum of the lengths of the longest continuous contour segments containing the first segment. The calculation of the longest continuous contour segment length of the matching feature points is used to increase the contribution weight of all matching point pairs on the segment when there is a long continuous contour segment in the matching point set. This makes the algorithm give higher evaluation to reliable matches that can maintain contour continuity when facing partial occlusion or noise interference, thereby improving the robustness of the matching.

[0039] The similarity score is calculated by fusing quantitative results based on appearance similarity and those based on local geometric similarity, and then combining them with a reward mechanism for contour continuity. It performs a pairwise quantitative calculation and summation averaging of matching feature point pairs between the global contour feature point set and each two-dimensional standard view in the baseline feature set. This achieves precise quantification of the feature matching similarity between the standard image data and each two-dimensional standard view. The quantified similarity score serves as the core quantitative basis for determining whether the standard image data and the two-dimensional standard view are similar views. A higher score indicates a higher degree of feature matching. Based on this score, a mapping relationship between the standard image data and the two-dimensional standard view can be accurately established, providing a precise quantitative foundation for obtaining reliable mapping pairs of cyclic items through geometric consistency screening. This ensures the accuracy of subsequently determining the initial pose and initial quantity of cyclic items.

[0040] The standard image data is subjected to image pyramid downsampling at three preset fixed scales: 1x original scale, 0.5x original scale, and 0.25x original scale. Edge detection is performed on the image at each scale. Pixels with an absolute difference of gray value greater than or equal to 50 between adjacent pixels are identified as edge pixels by traversing the image pixel by pixel. The edge pixels obtained at each scale are then subjected to contour fitting. The pixels on the fitted continuous contour are sampled and selected according to the rule of every 5 pixels. The pixels obtained from downsampling at all scales are summarized and deduplicated. The set of pixels obtained after removing duplicate pixels is the global contour feature point set of the standard image data.

[0041] The system iterates through all two-dimensional standard views in the 3D view benchmark library. For each two-dimensional standard view, it performs the same multi-scale feature extraction operation as the standard image data, namely, it performs image pyramid downsampling processing at 1x original scale, 0.5x original scale, and 0.25x original scale. At each scale, it identifies adjacent pixels with an absolute difference in grayscale value greater than or equal to 50 by traversing pixel by pixel to obtain edge pixels. After fitting contours to the edge pixels, it samples and selects them at intervals of 5 pixels. Then, it summarizes and removes duplicates to obtain the contour feature point set of a single two-dimensional standard view. The contour feature point set of each two-dimensional standard view is then associated and bound with the corresponding cyclic item category identifier and view shooting angle information of the two-dimensional standard view. The contour feature point sets and corresponding association information of all two-dimensional standard views in the 3D view benchmark library that have completed association and binding are integrated. The integrated set is the benchmark feature set of two-dimensional standard views in the 3D view benchmark library.

[0042] Based on each pixel in the global contour feature point set, the pixel coordinate matching degree is verified sequentially with the pixels in the contour feature point set corresponding to each two-dimensional standard view in the baseline feature set. By checking the deviation values ​​of the horizontal and vertical coordinates of two pixels in the image coordinate system, when the absolute values ​​of the horizontal and vertical coordinate deviation values ​​are both less than or equal to 10 pixel units, the two pixels are determined to be matching pixels. Then, for each two-dimensional standard view, the number of pixels in its contour feature point set that match the global contour feature point set is counted, and the proportion of the number of matching pixels to the total number of pixels in the contour feature point set of the two-dimensional standard view is calculated as the feature similarity. When the feature similarity is greater than or equal to 70%, the two-dimensional standard view is determined to be a similar view to the standard image data. Then, the global contour feature point set of the standard image data and the contour feature point sets of all two-dimensional standard views determined to be similar views are associated one-to-one according to the matching pixels. This one-to-one association relationship is the mapping relationship between the standard image data and the two-dimensional standard view.

[0043] Based on the established mapping relationship, the coordinate information of each set of corresponding matching pixels is extracted. The geometric transformation relationship of the coordinate information of each set of matching pixels is verified. The verification process is as follows: taking the contour feature point set of the two-dimensional standard view as a reference, three sets of non-collinear matching pixels are arbitrarily selected. The relative position ratio and angle relationship of the three sets of pixels in the standard image data are checked to see if they are consistent with the correspondence in the two-dimensional standard view. When all three sets of non-collinear matching pixels selected meet the condition of consistent relative position ratio and angle relationship, the geometric transformation relationship of the set of matching pixels is determined to be valid. Then, the matching pixel pairs with valid geometric transformation relationship are filtered out from the mapping relationship. All the filtered matching pixel pairs are the reliable mapping pairs of the cyclic items. The coordinates of all matching pixels in the trusted mapping pair are summarized, and the average coordinates of the pixels corresponding to all trusted mapping pairs in the standard image data are calculated. This average coordinate is used as the center position coordinates of the cyclic item in the image coordinate system of the standard image data, thereby determining the position of the cyclic item in the standard image data. Then, taking the arrangement direction of the pixels of the corresponding trusted mapping pair in the two-dimensional standard view as the reference direction, the angle between the arrangement direction of the pixels of the corresponding trusted mapping pair in the standard image data and the reference direction is calculated. This angle value is used as the rotation angle of the cyclic item in the standard image data. The determined center position coordinates and rotation angle are used together as the initial pose of the cyclic item.

[0044] Using the center coordinates in the initial pose as the core, an initial detection region is defined with a preset pixel radius of 50 pixels. Within the initial detection region, pixels in the standard image data are divided into regions based on the rotation angle of the initial pose. Pixels that conform to the contour features of the cyclic item are grouped into the same region. Neighborhood detection is then performed on all defined initial detection regions in the standard image data. When the pixel distance between two detection regions is less than or equal to 30 pixels and the pixels in the regions all conform to the contour features of the same cyclic item, the two detection regions are merged. The neighborhood detection and merging operations are repeated on the merged region until no detection region that meets the merging conditions exists. At this point, the remaining unmerged independent regions in the standard image data are the aggregated regions corresponding to instances of the same cyclic item. The number of these independent aggregated regions is counted, and the specific value obtained is the initial number of cyclic items.

[0045] The beneficial effects are that by extracting and matching image features at multiple scales, the feature association between standard image data and two-dimensional standard view can be accurately obtained, a stable mapping relationship can be established, and reliable matching point pairs can be obtained through geometric consistency screening. The initial position and rotation state of the circular item can be accurately determined. Subsequently, through region aggregation and merging operations, the initial quantity of the circular item can be accurately counted, which improves the accuracy and robustness of the circular item detection and effectively ensures the reliability of pose and quantity recognition.

[0046] S4. Based on the initial pose, the spatial relationship between items in the standard image data is determined to obtain the dataset of items to be confirmed in the standard image data. In this embodiment of the invention, the step of determining the spatial relationships between items in the standard image data based on the initial pose to obtain the dataset of items to be confirmed in the standard image data includes: Based on the initial pose, determine the candidate image region of the cyclic item in the standard image data; Spatial evaluation is performed on the relative positional relationship and boundary overlap state between the candidate image regions to obtain the three-dimensional state information of the circular items. The three-dimensional state information includes whether there is occlusion, stacking and boundary adhesion between the circular items. Based on the three-dimensional state information, candidate items in the standard image data that are in an occluded state, a highly stacked state, or a state with blurred boundaries are identified. The candidate items and candidate image regions in the occluded state, the highly stacked state, and the blurred boundary state are used to obtain the standard image data dataset of items to be confirmed.

[0047] Using the center position coordinates in the initial pose as the origin, expand the rectangular area by 40 pixels in both the horizontal and vertical directions to form an initial rectangular area. Then, rotate the rectangular area according to the rotation angle in the initial pose so that the major axis of the rectangular area is consistent with the rotation angle of the cyclic item. The transformed rectangular area is the candidate image area of ​​the cyclic item in the standard image data.

[0048] For any two candidate image regions, first calculate the vertical distance between the centers of the two regions in the image coordinate system. When the vertical distance is greater than 60 pixels, it is determined that the two circular items have a stacking relationship. Then calculate the pixel overlap area of ​​the two regions. When the overlap area accounts for more than 20% of the total area of ​​either region, it is determined that the two circular items have an occlusion relationship. Finally, calculate the minimum distance between the edge pixels of the two regions. When the minimum distance is less than or equal to 5 pixels, it is determined that the two circular items have a boundary adhesion relationship. The information obtained by summarizing all the occlusion, stacking and boundary adhesion relationships is the three-dimensional state information of the circular items.

[0049] Based on the 3D state information, circular objects within candidate image regions exhibiting occlusion are identified as occluded candidate objects. Circular objects within candidate image regions exhibiting stacking relationships with a vertical distance greater than 60 pixels are identified as highly stacked candidate objects. Circular objects within candidate image regions exhibiting boundary adhesion relationships with an absolute difference in edge pixel grayscale values ​​less than 30 are identified as candidates with blurred boundaries. The occluded, highly stacked, and blurred-boundary candidate objects are then associated with the coordinate information and state labels of their respective candidate image regions. All associated and bound entries are then summarized, and the resulting set constitutes the dataset of objects to be confirmed in the standard image data.

[0050] The beneficial effects are that by defining a clear candidate image region, the corresponding region of the cyclical item in the standard image data can be accurately located. Combined with clear spatial relationship evaluation rules, the occlusion, stacking and boundary adhesion of the cyclical items can be accurately identified, thereby filtering out candidate items with abnormal states. Finally, a dataset of items to be confirmed containing candidate items and corresponding regions is generated, which effectively improves the accuracy of cyclical item state discrimination and anomaly identification, and provides a clear and reliable basis for subsequent item confirmation and processing.

[0051] S5. Perform multi-view feature analysis on the item regions in the dataset of items to be confirmed, and perform simulation verification on the initial pose based on the analysis results to obtain the verification data of the dataset of items to be confirmed. In this embodiment of the invention, the step of performing multi-view feature analysis on the item regions in the dataset of items to be confirmed, and performing simulation verification on the initial pose based on the analysis results to obtain verification data for the dataset of items to be confirmed, includes: Based on the item region in the dataset of items to be confirmed and the initial pose, a virtual observation view surrounding the item region is generated; Under the virtual observation perspective, the cyclic items in the 3D view reference library are subjected to 3D geometric information detection to obtain the simulated contour features and simulated texture features of the cyclic items; Based on the simulated contour features and the simulated texture features, feature adaptation and extraction are performed on the standard image data to obtain the actual image content of the standard image data; The simulated texture features and simulated contour features are compared and evaluated with the actual image content to obtain the matching degree evaluation result of the circular item; Based on the matching degree evaluation result, the initial pose is simulated and iteratively fine-tuned until the matching degree evaluation result meets the preset consistency condition. The final optimized pose that satisfies the consistency condition is recorded as the verified pose, and the verified pose and the matching degree evaluation result are integrated into the verification data of the dataset of items to be confirmed.

[0052] The process involves performing 3D geometric information detection on the circular objects in the 3D view reference library under the virtual observation perspective to obtain the simulated contour features and simulated texture features of the circular objects, including: Based on the virtual observation perspective, the three-dimensional geometric information is observed and projected to obtain the depth map and surface visibility mask of the circular item; In the depth map and the surface visibility mask, the external geometric boundary of the circular object under the virtual observation view is extracted as the simulation contour data of the circular object; Key point detection is performed on the simulated contour data to obtain the simulated contour features of the cyclic item; Based on the surface texture information associated with the three-dimensional geometric information and the virtual observation viewpoint, a simulated texture image of the cyclic item under the virtual observation viewpoint is obtained through texture mapping and sampling. The simulated texture image is divided into regions and statistical features are extracted to obtain the simulated texture features of the cyclic item.

[0053] The center coordinates of the item region in the dataset of items to be confirmed are extracted as the origin of the virtual observation. The rotation angle of the initial pose is used as the reference direction. The observation orientation is set at equal angular intervals along the horizontal circumference with the origin as the center in three-dimensional space. At the same time, a fixed observation height range is set in the vertical direction. Each combination of observation orientation and observation height forms an independent virtual observation view. The set of all such combinations is the virtual observation view around the item region.

[0054] For each virtual observation viewpoint, the 3D geometric model of the corresponding cyclic item is retrieved from the 3D view reference library. The 3D geometric model is then adapted to the observation parameters of the virtual observation viewpoint, ensuring that the 3D geometric model is completely within the observation range of the virtual observation viewpoint. The outer edge features of the adapted 3D geometric model are extracted, and continuous feature information of the outer edge of the model is extracted to form simulation contour features. At the same time, surface feature information such as texture arrangement and color distribution on the surface of the 3D geometric model is extracted to form simulation texture features. A set of simulation contour features and simulation texture features is generated for each virtual observation viewpoint.

[0055] Using the simulated contour features and simulated texture features from various virtual observation perspectives as a unified extraction template, feature matching and retrieval are performed within the object region corresponding to the object to be identified in the standard image data, according to the feature shape and arrangement rules of the extraction template. Image pixel information that matches the simulated contour feature shape and conforms to the arrangement rules of the simulated texture features within the object region is retrieved and extracted. All extracted pixel information is fully integrated according to its original spatial position in the standard image data. The integrated pixel information set is the actual image content of the standard image data.

[0056] The simulated contour features are compared point by point with the contour parts in the actual image content to confirm the overlap of the contour features. Then, the simulated texture features are compared region by region with the texture parts in the actual image content to confirm the matching of the texture features. Combining the overlap of the contour features and the matching of the texture features, the overall degree of feature matching is comprehensively judged according to the preset feature weights. The result of the overall degree of feature matching of the contour and texture is the matching degree evaluation result of the circular item.

[0057] When the matching degree evaluation result does not meet the preset consistency condition, based on the center position coordinates and rotation angle of the initial pose, the center position coordinates are first adjusted by a small range of position offset, and then the rotation angle is adjusted by a small range of angle deflection. After completing one adjustment, the virtual observation view is regenerated, and all operations of 3D geometric information detection, feature adaptation extraction and comparison evaluation are repeated to obtain a new matching degree evaluation result. If the new matching degree evaluation result still does not meet the preset consistency condition, the position and angle are adjusted in the same way again. The operation process of adjustment and evaluation is repeated until the matching degree evaluation result meets the preset consistency condition, at which point all adjustment operations are stopped.

[0058] When the matching degree evaluation result meets the preset consistency condition, record the position coordinates and rotation angle of the looped item after all fine-tuning is completed. The position coordinates and rotation angle are used together as the verified pose. The relevant information of the verified pose is associated with the matching degree evaluation result that meets the consistency condition. At the same time, a unique identifier for the item to be confirmed is added to each set of associated information. The association information of the verified poses and matching degree evaluation results of all items to be confirmed is summarized as a whole. The summarized information set is the verification data of the dataset of items to be confirmed.

[0059] Based on the observation azimuth, focal length, and projection plane parameters of the virtual observation perspective, all spatial coordinates of the three-dimensional geometric information are transformed to a two-dimensional projection plane according to the perspective projection rules. The vertical distance from each three-dimensional spatial point to the projection plane is calculated point by point. The vertical distance information is arranged according to the projection coordinates to form a two-dimensional image, which is the depth map of the circulating item. At the same time, it is determined point by point whether each surface point in the three-dimensional geometric information is occluded by other geometric structures. The corresponding coordinates of the unoccluded surface points on the projection plane are marked as visible, and the occluded ones are marked as invisible. The two-dimensional marked image formed according to the projection coordinates is the surface visibility mask of the circulating item.

[0060] Using a surface visibility mask as the filtering basis, pixel filtering is performed on the depth map, retaining only the depth map pixel information marked as visible areas in the mask. Edge recognition is then performed on the filtered depth map to identify continuous pixel lines in the image where the pixel grayscale values ​​show a step change. These continuous pixel lines are the external geometric boundaries of the cyclic object under the virtual observation view. All pixel coordinates, continuous arrangement order, and line thickness information of this external geometric boundary are fully integrated, and the integrated information is the simulation contour data of the cyclic object.

[0061] The continuous external geometric boundaries in the simulation contour data are traversed pixel by pixel and analyzed segment by segment to identify the inflection points, endpoints and pixels where the curvature changes abruptly on the boundary. These pixels are uniformly regarded as contour key points. The two-dimensional coordinate information, pixel distance with adjacent key points, connection relationship and local contour direction information of each contour key point are extracted. All the information of the contour key points is systematically summarized. The summarized contour key point information set is the simulation contour feature of the cyclic item.

[0062] The surface texture information of the cyclical item is retrieved from the pre-associated and bound 3D geometric information. The texture coordinates of the surface texture information are matched one-to-one with the spatial coordinates of the 3D geometric information, so that each 3D spatial point corresponds to a unique texture coordinate. Then, according to the perspective projection rules of the virtual observation perspective, the matched texture coordinates are transformed into a 2D projection plane along with the 3D spatial coordinates. In the 2D texture coordinate area after projection, texture sampling is performed point by point at a fixed pixel density to collect the texture color and texture detail information of each sampling point. The information of all sampling points is stitched together and integrated according to the coordinate arrangement rules of the 2D projection plane. The stitched 2D image is the simulated texture image of the cyclical item.

[0063] The simulated texture image is divided into multiple equal rectangular sub-regions according to its size. Statistical features are extracted for each rectangular sub-region. The extracted features include the color mean, color distribution range, and density of texture details within the sub-region. The statistical features of each sub-region are associated with its coordinate position in the simulated texture image. Then, the texture arrangement pattern, texture direction, and texture splicing features of the entire simulated texture image are extracted. The associated features of all sub-regions are fused and integrated with the texture features of the entire image. The fused set of all feature information is the simulated texture feature of the cyclic item.

[0064] The beneficial effects are as follows: by scientifically generating a virtual observation perspective around the object area, and combining perspective projection and feature extraction methods, the simulated contour features and simulated texture features of the cyclical object are accurately obtained. Relying on feature adaptation retrieval and point-by-point and region-by-region comparison and evaluation, the overall degree of feature matching can be accurately determined. Then, through pose iteration and fine-tuning, the matching degree is ensured to meet the consistency conditions. Finally, verification data that associates the verified pose and matching degree results is generated, which effectively improves the accuracy and reliability of the verification of the object dataset to be confirmed, and provides solid feature and data support for the subsequent object status confirmation and processing.

[0065] S6. Perform a collaborative evaluation of the initial quantity and the verification data to obtain a digital inventory report of the recyclable items.

[0066] In this embodiment of the invention, the step of co-evaluating the initial quantity and the verification data to obtain a digital inventory report of the recyclable items includes: The initial quantity is compared with the quantity of items verified in the verification data to obtain the quantity difference information of the cyclic items; Based on the quantity difference information, determine whether the initial quantity is consistent with the confirmed quantity in the verification data; If the results are consistent, the initial quantity is taken as the final inventory quantity of the recycle item, and the final inventory quantity and the verification data are integrated into the inventory result data of the recycle item. If an inconsistency is determined, the dataset of items to be confirmed is re-statistically analyzed and verified based on the verification data to obtain the corrected inventory count of the cyclical items. The corrected inventory count, the quantity difference information, and the verified item pose are then integrated into the inventory result data of the cyclical items. The inventory results data are formatted and packaged according to the preset report template to obtain a digital inventory report of the reusable items.

[0067] Extract the valid confirmation identifier for each item to be confirmed from the verification data after verifying pose matching. Count the number of items with valid confirmation identifiers as the number of verified and confirmed items in the verification data. Then, compare the initial quantity with the number of verified and confirmed items, record the difference between the two, and indicate the direction of the difference: either the initial quantity is more than the number of verified and confirmed items, or the initial quantity is less than the number of verified and confirmed items. Associate and integrate the difference and its direction to obtain the quantity difference information of the looping items. Extract the difference from the quantity difference information. When the difference is zero, it is determined that the initial quantity matches the confirmed quantity in the verification data; when the difference is not zero, it is determined that the initial quantity does not match the confirmed quantity in the verification data. The determination result is then associated and stored with the quantity difference information.

[0068] When the initial quantity matches the confirmed quantity in the verification data, the initial quantity is directly marked as the final inventory quantity of the cyclical items. The numerical information of the final inventory quantity is extracted and a quantity label is added. Then, the final inventory quantity with the label is associated with all the information in the verification data, including the verified pose and matching degree evaluation results of each item to be confirmed. All the associated information is then integrated into the system. The integrated information set is the inventory result data of the cyclical items.

[0069] When the initial quantity is inconsistent with the confirmed quantity in the verification data, the verified pose and matching degree evaluation result in the verification data are used as the basis for review. All items in the dataset to be confirmed are checked one by one. Items whose matching degree evaluation results do not meet the consistency condition are removed. Items whose matching degree evaluation results meet the consistency condition and have verified poses are retained. The retained items are counted one by one. The counted value is the corrected inventory quantity of the cyclical items. The numerical information of the corrected inventory quantity is extracted and a review label is added. Then, the corrected inventory quantity with the label, the complete quantity difference information, and the verified pose corresponding to the retained items are used as the verified item poses for overall association and integration. The integrated information set is the inventory result data of the cyclical items.

[0070] Retrieve a pre-set digital inventory report template, which contains fixed report columns, specifically: basic information of the inventory items, inventory quantity information, item position verification information, quantity difference information, and verification instructions. Fill in the various information in the inventory results data according to the pre-set column classification rules of the template. Standardize the format of each column after filling in the information to make the font, font size, and layout of the information uniform and standardized. Then add basic annotation information such as inventory time, inventory object, and data source to the standardized report. Digitally package the complete report after annotation. The packaged standardized digital file is the digital inventory report of the cyclical items.

[0071] The beneficial effects are that by comparing the deviations and judging the consistency between the initial quantity and the confirmed quantity in the verification data, accurate inventory results are generated according to different situations. When they are consistent, the initial quantity and verification data are directly integrated to ensure efficiency. When they are inconsistent, the verification data is used to review and correct the quantity to ensure accuracy. Then, the data is packaged into a digital inventory report through a standardized template. This not only improves the accuracy and completeness of the inventory data of cyclical items, but also ensures the standardization and readability of the report, providing a reliable inventory basis for the digital management of cyclical items.

[0072] like Figure 2 The diagram shown is a functional block diagram of a digital inventory system for recyclable items provided in an embodiment of the present invention.

[0073] The digital inventory system 100 for recyclable items described in this invention can be installed in an electronic device. Depending on the functions implemented, the digital inventory system 100 may include an image preprocessing module 101, a three-dimensional view reference library construction module 102, an initial identification and positioning module 103, an item spatial relationship discrimination module 104, a multi-view feature verification module 105, and a comprehensive report generation module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0074] In this embodiment, the functions of each module / unit are as follows: The image preprocessing module 101 is used to perform mean filtering on the original image of the inventory area to obtain standard image data of the inventory area. The three-dimensional view reference library construction module 102 is used to establish a three-dimensional view reference library for the cyclical items based on the conventional placement posture of the cyclical items and prior digital knowledge. The initial identification and positioning module 103 is used to adaptively match and statistically analyze the standard image data with the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item. The item spatial relationship discrimination module 104 is used to perform state discrimination on the spatial relationship between items in the standard image data based on the initial pose, and obtain the item dataset to be confirmed in the standard image data. The multi-view feature verification module 105 is used to perform multi-view feature analysis on the item region in the item dataset to be confirmed, and to perform simulation verification on the initial pose based on the analysis results, so as to obtain the verification data of the item dataset to be confirmed. The comprehensive report generation module 106 is used to collaboratively evaluate the initial quantity and the verification data to obtain a digital inventory report of the recyclable items.

[0075] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0076] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0077] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0078] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0079] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for digital inventory management of reusable items, characterized in that, The method includes: S1. Perform mean filtering on the original image of the inventory area to obtain standard image data of the inventory area; S2. Based on the conventional placement posture of the recyclable items and prior digital knowledge, establish a three-dimensional view reference library for the recyclable items; S3. Perform adaptive matching statistics between the standard image data and the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item; S4. Based on the initial pose, the spatial relationship between items in the standard image data is determined to obtain the dataset of items to be confirmed in the standard image data. S5. Perform multi-view feature analysis on the item regions in the dataset of items to be confirmed, and perform simulation verification on the initial pose based on the analysis results to obtain the verification data of the dataset of items to be confirmed. S6. Perform a collaborative evaluation of the initial quantity and the verification data to obtain a digital inventory report of the recyclable items.

2. The method for digital inventory management of reusable items as described in claim 1, characterized in that, The step of performing mean filtering on the original image of the inventory area to obtain standard image data of the inventory area includes: Obtain the original image of the inventory area. The original image is adjusted in grayscale to obtain the enhanced image data of the original image; The enhanced image data is subjected to edge filtering to obtain a smooth image of the inventory area; The smoothed image is subjected to mean filtering to obtain standard image data of the inventory area.

3. The method for digital inventory of reusable items as described in claim 1, characterized in that, The step of establishing a three-dimensional view reference library for the reusable items based on their conventional placement and prior digital knowledge includes: Parameter mining is performed on the three-dimensional geometric information and surface texture information of the circular item to obtain the prior digital knowledge data of the circular item; Based on the conventional placement posture of the cyclical items, the prior digital knowledge data is subjected to multi-view projection rendering to obtain multi-view reference data of the cyclical items. Key features are labeled on the multi-view benchmark data to obtain standardized feature description data of the cyclic items; The standardized feature description data, the multi-view benchmark data, and the prior digital knowledge data are associated and stored to obtain the three-dimensional view benchmark library of the cyclic item.

4. The method for digital inventory of reusable items as described in claim 1, characterized in that, The step of adaptively matching and statistically analyzing the standard image data with the 3D view reference library to obtain the initial pose and initial quantity of the cyclic items includes: Multi-scale feature extraction is performed on the standard image data to obtain a global contour feature point set of the standard image data; Extract the reference feature set of the two-dimensional standard view from the three-dimensional view reference library; The global contour feature point set is compared with the reference feature set in a similarity traversal, and a mapping relationship between the standard image data and the two-dimensional standard view is established based on the comparison results. Based on the mapping relationship, geometric consistency screening is performed on the global contour feature point set to obtain the reliable mapping pairs of the cyclic items; The position and rotation angle of the cyclic item in the standard image data are determined based on the trusted mapping pair, which serves as the initial pose of the cyclic item. Based on the initial pose, instances of the same cyclic item in the standard image data are aggregated into regions, and the number of independent regions after aggregation is used as the initial number of cyclic items.

5. The method for digital inventory of reusable items as described in claim 4, characterized in that, The formula for calculating the similarity score in the similarity traversal comparison is as follows: ； In the formula, The similarity score is given. The total number of matching feature point pairs in the mapping relationship. The summation index for matching feature points. The preset appearance similarity weighting coefficients, For the first The Euclidean distance between the matching feature points in the feature space, The preset distance scale parameters, The preset geometric similarity weight coefficients, For the first The change in the angle between the matched feature point and other matching point pairs in its neighborhood, representing the local geometric structure. It is a natural exponential function. For geometric consistency functions, For the first The length of the longest continuous contour segment for matching feature points. This is the reward function for contour continuity.

6. The method for digital inventory of reusable items as described in claim 1, characterized in that, Based on the initial pose, the spatial relationships between items in the standard image data are determined to obtain a dataset of items to be confirmed from the standard image data, including: Based on the initial pose, determine the candidate image region of the cyclic item in the standard image data; Spatial evaluation is performed on the relative positional relationship and boundary overlap state between the candidate image regions to obtain the three-dimensional state information of the circular items. The three-dimensional state information includes whether there is occlusion, stacking and boundary adhesion between the circular items. Based on the three-dimensional state information, candidate items in the standard image data that are in an occluded state, a highly stacked state, or a state with blurred boundaries are identified. The candidate items and candidate image regions in the occluded state, the highly stacked state, and the blurred boundary state are used to obtain the standard image data dataset of items to be confirmed.

7. The method for digital inventory of reusable items as described in claim 1, characterized in that, The process involves performing multi-view feature analysis on the item regions in the dataset of items to be confirmed, and based on the analysis results, performing simulation verification on the initial pose to obtain verification data for the dataset of items to be confirmed, including: Based on the item region in the dataset of items to be confirmed and the initial pose, a virtual observation view surrounding the item region is generated; Under the virtual observation perspective, the cyclic items in the 3D view reference library are subjected to 3D geometric information detection to obtain the simulated contour features and simulated texture features of the cyclic items; Based on the simulated contour features and the simulated texture features, feature adaptation and extraction are performed on the standard image data to obtain the actual image content of the standard image data; The simulated texture features and simulated contour features are compared and evaluated with the actual image content to obtain the matching degree evaluation result of the cyclic item; Based on the matching degree evaluation result, the initial pose is simulated and iteratively fine-tuned until the matching degree evaluation result meets the preset consistency condition. The final optimized pose that satisfies the consistency condition is recorded as the verified pose, and the verified pose and the matching degree evaluation result are integrated into the verification data of the dataset of items to be confirmed.

8. The method for digital inventory of reusable items as described in claim 7, characterized in that, The process involves performing 3D geometric information detection on the circular objects in the 3D view reference library under the virtual observation perspective to obtain the simulated contour features and simulated texture features of the circular objects, including: Based on the virtual observation perspective, the three-dimensional geometric information is observed and projected to obtain the depth map and surface visibility mask of the circular item; In the depth map and the surface visibility mask, the external geometric boundary of the circular object under the virtual observation view is extracted as the simulation contour data of the circular object; Key point detection is performed on the simulated contour data to obtain the simulated contour features of the cyclic item; Based on the surface texture information associated with the three-dimensional geometric information and the virtual observation viewpoint, a simulated texture image of the cyclic item under the virtual observation viewpoint is obtained through texture mapping and sampling. The simulated texture image is divided into regions and statistical features are extracted to obtain the simulated texture features of the cyclic item.

9. The method for digital inventory of reusable items as described in claim 1, characterized in that, The step of collaboratively evaluating the initial quantity and the verification data to obtain a digital inventory report of the recyclable items includes: The initial quantity is compared with the quantity of items verified in the verification data to obtain the quantity difference information of the cyclic items; Based on the quantity difference information, determine whether the initial quantity is consistent with the confirmed quantity in the verification data; If the results are consistent, the initial quantity is taken as the final inventory quantity of the recycle item, and the final inventory quantity and the verification data are integrated into the inventory result data of the recycle item. If an inconsistency is determined, the dataset of items to be confirmed is re-statistically analyzed and verified based on the verification data to obtain the corrected inventory count of the cyclical items. The corrected inventory count, the quantity difference information, and the verified item pose are then integrated into the inventory result data of the cyclical items. The inventory results data are formatted and packaged according to the preset report template to obtain a digital inventory report of the reusable items.

10. A digital inventory system for recyclable items, characterized in that, The system for implementing the digital inventory method for reusable items as described in claim 1 includes: The image preprocessing module is used to perform mean filtering on the original image of the inventory area to obtain standard image data of the inventory area; The 3D view reference library construction module is used to establish a 3D view reference library for the cyclical items based on the conventional placement posture of the cyclical items and prior digital knowledge. The initial identification and positioning module is used to adaptively match and statistically analyze the standard image data with the three-dimensional view reference library to obtain the initial pose and initial quantity of the cyclic item; The item spatial relationship discrimination module is used to perform state discrimination on the spatial relationship between items in the standard image data based on the initial pose, and obtain the item dataset to be confirmed in the standard image data; The multi-view feature verification module is used to perform multi-view feature analysis on the item regions in the item dataset to be confirmed, and to perform simulation verification on the initial pose based on the analysis results, so as to obtain the verification data of the item dataset to be confirmed. The comprehensive report generation module is used to collaboratively evaluate the initial quantity and the verification data to obtain a digital inventory report of the recyclable items.

Images