Non-standard warehouse AI intelligent checking system

Abstract

The method focuses on the non-standard warehouse checking problem in the warehouse logistics field. Aiming at the problems of low efficiency, poor accuracy and high labor cost of traditional manual checking and limitation of unmanned aerial vehicle checking in a non-standard warehouse, the invention provides a non-standard warehouse AI intelligent checking system. The system integrates robot / unmanned aerial vehicle automatic checking, AI algorithm and path planning, special scene processing and data acquisition and processing technologies. By customizing a robot chassis, a lifting device, an array camera, a multi-layer camera disc and other hardware and combining various AI recognition algorithms and path planning algorithms, efficient and accurate checking is achieved. The method has the advantages of improving the checking efficiency, improving the accuracy, reducing the labor cost, adapting to a complex environment, enhancing the reliability, facilitating algorithm iteration and deployment, being good in compatibility and high in expandability and the like, and an innovative and practical solution is provided for non-standard warehouse checking.

Description

Technical Field

[0001] The present invention relates to the technical field of warehousing and logistics, and in particular to an AI intelligent inventory counting system for non-standard warehouses. Background Art

[0002] In the field of warehouse management, many warehouses, such as large non-standard warehouses of 30,000 square meters, currently mainly use manual inventory. Figure 4 As shown, the staff needs to identify and count the number of goods on each pallet in the warehouse, and then compare it with the warehouse system data. This method has significant disadvantages:

[0003] Inefficiency: Manual inventory is slow. For example, a large rice warehouse needs to be shut down for two days to conduct a comprehensive inventory. During this period, operations such as goods entering and leaving the warehouse are suspended, which seriously reduces operational efficiency. Relevant research shows that the efficiency of manual inventory in large warehouses is far lower than that of automated inventory. Its speed is limited by the number of personnel and work intensity, and it is difficult to complete large-scale inventory tasks in a short time.

[0004] High labor costs: Manual inventory requires a lot of manpower. Take a warehouse with a 7.5-kilometer long route as an example. Assuming that workers walk to count goods using barcode scanning, and the warehouse is medium-sized and the goods are tightly stacked, and the inventory is counted every two days, each worker can complete about 800-1200 meters of inventory tasks in one working day (8 hours), and it takes about 20-30 people to complete the entire warehouse inventory. This means that companies need to pay high labor costs, including wages, benefits, etc.

[0005] Accuracy is difficult to guarantee: During the manual inventory process, staff are prone to make mistakes due to fatigue, negligence, etc., resulting in inaccurate inventory data. There may be cases of missed or wrong quantities of goods, which will affect the authenticity of inventory data and have an adverse impact on the company's procurement, sales and other links.

[0006] In addition, the drone inventory technology introduced by some warehouses also has limitations in non-standard warehouse environments. Although drone inventory has a fast movement speed and can theoretically quickly locate the inventory location, it has high maintenance costs, is easy to hit the shelves, does not have edge AI real-time analysis capabilities, has limited power and requires frequent charging, and has difficulty in positioning in the warehouse. Because the metal shelves in the warehouse reflect UWB signals, outdoor precise differential GPS technology cannot be used, which greatly reduces the positioning accuracy of drones in the warehouse and makes it difficult to meet the needs of accurate inventory. Related studies have pointed out that the application of drones in complex warehousing environments still faces many technical difficulties, and the reliability and stability of their inventory needs to be improved. Summary of the invention

[0007] The purpose of the present invention is to provide a non-standard warehouse AI intelligent inventory system with the following functions:

[0008] Improve inventory counting efficiency: Devote to breaking through the limitations of the traditional manual inventory counting mode. By introducing advanced automation technologies and AI intelligent algorithms, achieve rapid inventory counting of goods in non-standard warehouses. Change the current situation where inventory counting requires a 2-day shutdown every month, significantly shorten the inventory counting cycle, ensure the continuity of daily warehouse operations, and comprehensively improve the overall operation efficiency of the warehouse.

[0009] Enhance inventory counting accuracy: Aiming at the errors and omissions prone to occur in manual inventory counting, utilize the precise recognition ability of AI algorithms to conduct all-round and high-precision recognition and analysis of shelf, pallet, and goods information, ensuring the accuracy and real-time nature of warehouse inventory data, and providing reliable data support for the core operations of the enterprise such as procurement and sales.

[0010] Adapt to complex warehouse environments: Fully consider the characteristics of non-standard warehouses, such as large area (30,000 square meters), complex shelf structures (multi-level, multi-location), and the existence of non-standard shelves and irregularly stacked goods. Develop a highly adaptable inventory counting system. Through customized hardware devices and intelligent algorithms, achieve comprehensive coverage and precise inventory counting of complex warehouse environments.

[0011] Achieve unmanned inventory counting: Aiming to reduce manual dependence and lower labor costs, create an automated and intelligent inventory counting process. Eliminate subjective and unstable factors in the manual inventory counting process, realize unmanned operation of warehouse inventory counting, and at the same time improve the standardization and normalization levels of inventory counting work.

[0012] Solve the problems of inventory counting in special scenarios: Focus on special scenarios such as areas inaccessible to robots, multi-layer stacked goods with partial occlusion, and difficulty in observing goods on the middle shelf due to more than two shelves arranged side by side. Design targeted solutions. Comprehensively apply AI image recognition technology, intelligent algorithms, and human-machine collaboration models to ensure accurate inventory counting can be completed in various complex and special scenarios.

[0013] The present invention is realized through the following technical solutions:

[0014] An AI intelligent inventory counting system for non-standard warehouses, comprising:

[0015] A robot / drone automatic inventory counting system, which uses AI technology to cooperate with robots or drones to automatically identify and read structured data from the bin location number to the quantity of goods, and compare it with the warehouse system;

[0016] An AI algorithm and path planning module, including a shelf recognition algorithm, a shelf bin number recognition algorithm, a pallet recognition algorithm, a pallet code recognition algorithm, an algorithm for counting goods on the pallet, and a robot path planning algorithm; A special scenario processing solution module, which provides processing methods for special scenarios such as areas inaccessible to robots like corners, multi-layer stacked goods with partial occlusion, and more than two shelves arranged side by side.

[0017] The data acquisition and processing module collects image data through a customized camera on the robot, performs real-time analysis and processing using AI algorithms, identifies relevant information and compares it with the warehouse system. Meanwhile, it marks suspicious points and pushes photos for manual secondary confirmation.

[0018] Further as an improvement of the technical solution of the present invention, the robot / drone automatic inventory system includes:

[0019] A robot chassis with automatic walking ability and a lifting device, which can be raised to a height of 4 meters on the shelf, supports 4G data transmission and indoor positioning, can remotely control path walking, provides a CAN interface, and the robot body can place a vision pre-processing server and an AI inference processing server to run relevant control and inventory logic programs;

[0020] A customized acquisition camera using an array camera and a combination of multiple focal lengths. Through multi-group camera and multi-focal length fusion stitching technology, it can identify label content and the global shelf vision at a distance of 2 meters, and can be connected to the vision pre-processing server;

[0021] The camera tray and the multi-layer camera tray. Multiple customized acquisition cameras can be placed at the same height and assembled into a multi-layer structure according to the shelf height and installed on the robot lifting arm;

[0022] A three-proof handheld PAD with a camera, running the Android system and a custom program, and linked with the inventory robot through a 4G data link;

[0023] A warehouse local server for running the inventory management program and data storage, which is linked with on-site devices through a 4G / 5G data link.

[0024] Further as an improvement of the technical solution of the present invention, in the AI algorithm and the path planning module: The shelf recognition algorithm is used to identify the position of the shelf in the current picture, judge whether the shelf is on the left or right, and whether it is aligned with the center of the camera, etc.;

[0025] The shelf library number recognition algorithm is used to identify the specific content of the library number of the shelf in the current picture;

[0026] The pallet recognition algorithm is used to identify the presence and position of the pallet;

[0027] The pallet code recognition algorithm is used to identify the specific content of the pallet code;

[0028] The goods counting algorithm on the pallet is used to identify the number of cuboid boxes on the shelf, judge the stacking state of the goods, and report the information to the background;

[0029] The robot path planning algorithm is used to realize the actual path planning of multiple robots in the warehouse and perform inventory inspection using the shortest path.

[0030] As a further improvement to the technical solution of the present invention, in the special scenario processing solution module:

[0031] The areas inaccessible to the corner robots are supplemented and inventoried manually.

[0032] For goods stacked in multiple layers with partial occlusion, the stacking method is judged according to the consistency of the sizes of the front boxes; when more than two shelves are arranged side by side, it is identified and processed by an AI algorithm or specially processed manually.

[0033] As a further improvement to the technical solution of the present invention, the system further includes:

[0034] The system architecture and hardware components, including the inventory robot body, handheld PAD, warehouse local server, and cloud business management system;

[0035] The inventory process and algorithm implementation part, including map modeling and path planning, image acquisition and preprocessing, AI algorithm recognition and data processing, data comparison and update, manual secondary confirmation, and data storage and report generation;

[0036] The system iteration and optimization part, which realizes the continuous improvement of the system through algorithm iteration, hardware upgrade, and user experience optimization.

[0037] As a further improvement to the technical solution of the present invention, the inventory robot body includes:

[0038] A customized mobile robot chassis with an automatic walking function, capable of rising to a height of 4 meters on the shelf, supporting 4G data transmission and indoor positioning functions, and a vision preprocessing server and an AI inference processing server are placed on the chassis;

[0039] A multi-layer camera disk installed on the robot's lifting arm, and each layer of the camera disk is equipped with multiple groups of customized acquisition cameras, adopting multi-focus fusion technology.

[0040] As a further improvement to the technical solution of the present invention, the handheld PAD is a three-proof Android tablet with a camera, communicates with the robot and the local server through a 4G data link, and displays the inventory status in real time.

[0041] As a further improvement to the technical solution of the present invention, the warehouse local server is used to run the inventory management program, store inventory data, and perform data interaction with the robot, PAD, and cloud management system, supporting dynamic expansion and cluster mode.

[0042] As a further improvement to the technical solution of the present invention, the cloud business management system is used to remotely monitor and manage inventory tasks, realize multi-site linkage work, and support one-key deployment of new algorithm models.

[0043] As a further improvement of the technical solution of the present invention, the algorithm iteratively trains and optimizes the AI algorithm regularly using the image data collected during the inventory process, supports user-defined algorithm modules, and realizes rapid deployment and application through a distributed cluster architecture.

[0044] It should be noted that:

[0045] Robot / UAV automatic inventory system:

[0046] Robot chassis and lifting device: Independently developed a customized chassis with high flexibility and stability, equipped with an advanced lifting device to build an intelligent inspection robot system. This robot chassis not only has efficient automatic walking ability and can shuttle freely in complex warehouse environments, but also can accurately rise to a height of 4 meters on the shelf to meet the inventory needs of shelves at different heights. At the same time, it supports 4G data transmission and indoor positioning functions, and can realize remote path control. It provides rich low-level CAN and other interfaces to facilitate the main control program to finely control the robot's walking. The internal space of the robot body is reasonably arranged, and it can accommodate at least two high-performance servers, one dedicated to visual pre-processing and the other focused on AI inference processing, ensuring that the robot can process various types of data in real time and efficiently during operation.

[0047] Customized acquisition camera: Adopt advanced array camera technology, and through a clever combination of multiple focal lengths, achieve accurate recognition of a global ultra-wide field of view and ultra-distant tags. Use multiple groups of cameras and multi-focal length fusion stitching technology, so that the camera can clearly capture the label content at a distance of 2 meters and comprehensively grasp the overall shelf situation. The camera is closely connected to the visual pre-processing server through network and other protocol interfaces to ensure that the collected image data can be transmitted to the server for processing in a timely and accurate manner.

[0048] Camera tray and multi-layer camera tray: Innovatively design the camera tray, and multiple customized acquisition cameras can be flexibly placed at the same height to ensure that when the robot walks in the aisle, it can comprehensively and completely collect data on the shelves on both sides. According to the different heights of the shelves, multiple camera trays are combined into a multi-layer structure and firmly installed on the robot's lifting arm to achieve a one-time efficient scan of multi-layer warehouse shelves, greatly improving the inventory efficiency.

[0049] Handheld PAD: Equipped with an Android handheld PAD with three-proof functions (waterproof, dustproof, and drop-proof), and an HD camera is built in. The PAD runs a specially customized inventory program and realizes real-time linkage with the inventory robot through a 4G data link. During the inventory process, the staff can use the PAD for manual secondary leakage compensation or confirmation of the inventory results, view the inventory situation in real time at any time, and promptly conduct leakage checking and manual inventory operations.

[0050] Warehouse Local Server: Deploy one or more high-performance servers as the warehouse local server to stably run the inventory management program, achieve efficient storage and secure management of massive data. Closely link with the inventory robots and inventory tablets on-site through 4G / 5G data links, comprehensively control all aspects of the inventory business, and achieve efficient control of the overall business.

[0051] AI Algorithms and Path Planning:

[0052] Shelf Recognition Algorithm: Use deep learning algorithms to accurately identify the position of the shelves in the current image, accurately judge whether the shelves are on the left or right, and whether they are aligned with the center of the camera and other key information, providing a reliable reference for the robot to accurately align with the shelves, ensuring that the robot can quickly and accurately collect data from the shelves.

[0053] Shelf Warehouse Number Recognition Algorithm: Based on OCR (Optical Character Recognition) technology and deep learning models, quickly and accurately identify the warehouse number of the shelves in the current image, ensure that the inventory data is highly consistent with the warehouse location information in the warehouse system, and lay a solid foundation for subsequent data processing and analysis.

[0054] Pallet Recognition Algorithm: Use advanced image recognition algorithms to effectively identify the presence and position of pallets, provide accurate pallet information for the entire inventory process, and ensure the integrity and accuracy of the inventory data.

[0055] Pallet Code Recognition Algorithm: Through feature extraction and deep learning training of pallet codes, achieve high-precision recognition of the specific content of pallet codes, ensuring that the hierarchical structure of the inventory data is clear and complete.

[0056] Goods Counting Algorithm on Pallets: Comprehensively use AI technologies such as object detection and image segmentation to accurately identify the number of rectangular boxes on the shelves and the number of goods of each shape. At the same time, the algorithm can intelligently judge whether the goods are stacked completely, whether they are double-stacked, and whether the number needs to be reconfirmed, and report the stacking status of the goods on the pallet to the background in a timely and accurate manner.

[0057] Robot Path Planning Algorithm: Adopt advanced path planning algorithms, combined with the actual layout of the warehouse and the distribution of shelves, to achieve the optimal path planning of multiple robots in the warehouse. Use the shortest path inspection and inventory strategy to reasonably arrange the walking routes of the robots, avoid repeated paths and ineffective movements, and maximize the inventory efficiency.

[0058] Special Scenario Handling Solutions:

[0059] Areas inaccessible to robots such as corners: Establish a dedicated database of inaccessible areas in the system. When the robot detects inaccessible areas such as corners during the inventory process, it automatically marks and uploads the information of these areas to the system. According to the system prompts, the staff uses a handheld PAD to conduct manual supplementary inventory and enters the inventory results into the system in real time to ensure that the goods in all areas of the warehouse can be accurately inventoried.

[0060] Goods are stacked in multiple layers and partially blocked: Through in-depth analysis of the goods images, use AI algorithms to identify features such as the length-width ratio and area difference of the goods, and intelligently judge whether the goods are stacked in two layers. For the case where the sizes of the front boxes are the same, it is determined as single-layer stacking; for the case where the sizes of the front boxes are quite different, it is determined as cross-stacking and automatically defined as double-layer. For the goods determined to be double-layer stacked or with doubts, they are automatically marked as requiring manual confirmation to ensure the accuracy of the inventory results.

[0061] More than two shelves side by side: When encountering a special scenario of more than two shelves side by side, first use AI algorithms to conduct multi-dimensional analysis of the image and try to identify the goods situation on the middle shelf. If the algorithm cannot accurately identify, the area is automatically marked as a suspicious point, and the staff obtains the relevant image information through a handheld PAD for manual secondary confirmation and processing.

[0062] Data collection and processing: During the operation of the robot, continuously collect image data of the shelves and goods through the customized camera on it. These data are transmitted to the vision pre-processing server in real time. After pre-processing operations such as denoising, distortion correction, and ROI (region of interest) extraction, they are then transmitted to the AI inference processing server. The AI algorithm conducts real-time and in-depth analysis and processing of the pre-processed data, accurately identifies the information of the shelves, pallets, and goods, and comprehensively compares the identification results with the data in the warehouse system to achieve automatic inventory. At the same time, for the suspicious points that appear during the algorithm identification process, the system automatically marks and pushes the relevant photos to the terminal or computer for manual secondary confirmation to ensure the reliability of the inventory data.

[0063] In summary, the present invention has the following beneficial effects:

[0064] Significantly improve the inventory efficiency: Compared with the traditional manual inventory that takes 2 days, using the robot inventory system of the present invention, a single robot can efficiently complete an inventory within 7.3 hours. If multiple robots work in parallel, such as 5 robots working together, it only takes 1.46 hours to complete the inventory task of the entire warehouse. This greatly shortens the inventory time, effectively reduces the warehouse downtime, comprehensively improves the warehouse operation efficiency, enables the warehouse to respond to market demands faster, and enhances the competitiveness of the enterprise.

[0065] Improve inventory accuracy: The AI algorithm of the present invention has undergone extensive data training and optimization, enabling accurate identification of shelf, pallet, and cargo information, effectively avoiding errors and omissions caused by factors such as fatigue and negligence during manual inventory. By combining the AI algorithm with manual secondary confirmation, the high accuracy of inventory data is ensured, providing reliable data support for warehouse management and helping enterprises make more scientific and reasonable decisions.

[0066] Reduce labor costs: By implementing unmanned inventory and automated operations, the dependence on manual inventory is significantly reduced, and the labor input is decreased. During the entire inventory process, only a few personnel are required to be responsible for spot-checking and secondary confirmation of suspicious locations, greatly reducing the manual labor intensity, saving a large amount of labor costs for the enterprise, and improving the economic efficiency of the enterprise.

[0067] Adapt to complex warehouse environments: With innovative designs such as customized robot chassis, lifting devices, and multi-layer camera discs, as well as advanced AI algorithms, the inventory system of the present invention can perfectly adapt to large non-standard warehouses of 30,000 square meters and complex shelf structures and non-standard shelf conditions. Whether it is multi-layer shelves or complex storage locations, comprehensive and accurate inventory can be achieved, effectively solving the limitations of traditional inventory methods in complex environments and providing efficient and reliable inventory solutions for various non-standard warehouses.

[0068] Improve reliability: The AI system has a powerful function of marking suspicious points. For points with doubts in algorithm recognition, they are automatically marked and photos are pushed to the manual for secondary confirmation. At the same time, the system archives photos of each inventory point and can be retrieved and queried at any time. This dual guarantee mechanism naturally prevents the occurrence of cheating and inventory omission problems, greatly improving the reliability and credibility of the inventory work and providing a solid guarantee for the enterprise's warehousing management.

[0069] Fast algorithm iteration and efficient deployment: The present invention constructs a complete set of automatic model training and iteration systems, and through collecting video data during user use, continuously performs automatic iterative training on the AI algorithm. This enables the algorithm to continuously adapt to changes in the warehouse environment and updates in the way of goods stacking, continuously improving the algorithm accuracy. At the same time, the self-developed distributed cluster architecture realizes the rapid deployment and application of new algorithms, greatly improving the iteration efficiency of the product, being able to meet the diverse needs of users in a timely manner, and maintaining the technological leadership of the product.

[0070] Good compatibility and user experience: Adopting the design concept of backend operation and lightweight frontend, the inventory system of the present invention has high compatibility, does not depend on the model and configuration of the user's tablet, and can achieve a consistent user experience on different devices. Whether on high-end devices or ordinary devices, staff can conveniently and quickly operate the handheld PAD for inventory-related work, reducing the user's usage threshold and improving user satisfaction.

[0071] Strong scalability: The present invention adopts a modular design, where each functional module is independent and works collaboratively, facilitating the addition of new algorithm types according to different user requirements. At the same time, through a distributed cluster architecture, the system can easily handle the needs of large-scale user growth and has good scalability. As the warehouse scale expands and the business develops, the system can be flexibly upgraded and expanded to continuously provide efficient inventory services for enterprises. Brief Description of the Drawings

[0072] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objectives, and advantages of the present invention will become more apparent:

[0073] Figure 1 It is the control flowchart of an AI intelligent inventory system for a non-standard warehouse according to an embodiment of the present invention;

[0074] Figure 2 It is the system hardware composition architecture diagram according to an embodiment of the present invention;

[0075] Figure 3 It is the schematic diagram of the hardware composition of the inventory robot according to an embodiment of the present invention;

[0076] Figure 4 It is the schematic diagram of the manual inventory scenario in the prior art;

[0077] Figure 5 It is the schematic diagram of the system architecture and hardware composition of an AI intelligent inventory system for a non-standard warehouse according to an embodiment of the present invention. Detailed Embodiments

[0078] To make the objectives, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application are described clearly and completely. Obviously, the embodiments described below are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0079] Refer to Figures 1 to 3 and Figure 5 , an AI intelligent inventory system for a non-standard warehouse, whose system architecture and hardware composition include:

[0080] Inventory robot body:

[0081] Chassis and lifting device: The customized mobile robot chassis adopts a high-performance drive system, with excellent obstacle avoidance and obstacle crossing capabilities, ensuring free movement in the complex environment of the warehouse and stable and reliable operation. Its lifting device uses high-precision screw drive technology, combined with advanced motor control algorithms, and can be smoothly raised to a height of 4 meters within seconds to accurately position to the target shelf position. The 4G data transmission module selects industrial-grade products to ensure the stability and real-time performance of data transmission, and can work normally even in the warehouse corner with weak signals. Indoor positioning adopts multi-modal positioning technology that combines laser navigation, visual navigation, and inertial navigation, with a positioning accuracy of up to ±5 mm, ensuring that the robot accurately travels to each inventory position. The two servers installed on the chassis both adopt high-performance industrial-grade servers, with powerful computing capabilities and data processing capabilities, and can meet long-term and high-intensity visual processing and AI inference tasks.

[0082] Multi-layer camera tray: The design of the multi-layer camera tray fully considers different shelf heights and cargo distributions. The customized acquisition cameras on each layer of the camera tray adopt the latest CMOS image sensors, with the characteristics of high resolution, low noise, and wide dynamic range. The multi-focal length fusion technology realizes seamless splicing through a unique algorithm, ensuring that both the tiny label details can be clearly captured at a distance of 2 meters and the entire shelf can be fully presented. The camera tray and the robot lifting arm adopt a quick-installation and quick-disassembly structure, which is convenient for quick installation and adjustment in different warehouse scenarios.

[0083] Handheld PAD: The three-proof handheld PAD adopts a sturdy and durable shell material, and has passed strict waterproof, dustproof, and anti-drop tests, and can adapt to the harsh working environment of the warehouse. Its built-in high-definition camera supports autofocus and optical image stabilization functions, ensuring clear and stable images. The running customized inventory program interface is simple and intuitive, and the staff can operate it proficiently through simple training. When the PAD communicates with the robot and the local server through a 4G data link, an encrypted transmission protocol is adopted to ensure the security and integrity of the data.

[0084] Warehouse local server: The warehouse local server adopts a high-performance enterprise-level server, with a large-capacity storage hard disk and high-speed CPU and memory. The server supports dynamic expansion and can add hard disks and memory at any time according to the growth of the warehouse scale and data volume. In the cluster mode, multiple servers are connected through a high-speed network to achieve load balancing and data redundancy backup, ensuring the stability and reliability of the system during large-scale data processing. The server is not only responsible for storing inventory data, but also runs an intelligent data analysis program, which can perform real-time analysis on the inventory data and provide decision-making support for warehouse management.

[0085] Cloud Business Management System: The cloud business management system is built on an advanced cloud computing platform and has powerful remote monitoring and management functions. Through this system, managers can view the inventory progress, inventory results, and equipment operation status of each warehouse in real time. The one-key deployment function of the new algorithm model adopts an automated deployment process, greatly shortening the algorithm update time and improving the system iteration efficiency. At the same time, the cloud system also supports multi-site collaborative work, and different warehouses can share data and algorithm models to achieve optimal resource allocation.

[0086] Inventory Process and Algorithm Implementation

[0087] Map Modeling and Path Planning: Before the inventory starts, the robot first conducts a 3D scan of the warehouse using a lidar to construct an accurate map of the warehouse. At the same time, using the image data collected by the camera, combined with deep learning algorithms, the positions and numbers of the shelves are identified and labeled, and are accurately matched with the location information in the warehouse system. The path planning algorithm combines the A* algorithm and the Dijkstra algorithm. According to the warehouse map and shelf distribution, it generates the optimal inventory path for the robot. During the path planning process, factors such as the robot's obstacle avoidance requirements, driving speed, and power consumption are fully considered to ensure the rationality and efficiency of the path.

[0088] Image Acquisition and Preprocessing: The robot travels along the planned path. Before reaching each shelf, it accurately positions itself through a high-precision position sensor and positioning system to ensure that the camera maintains the best shooting angle with the shelf. During the 10-second stop, the multi-layer camera disk quickly acquires image data on both sides of the shelf. The visual preprocessing server uses parallel computing technology to quickly denoise, correct distortion, and extract ROIs from the acquired images. The denoising algorithm uses a deep learning-based non-local means denoising algorithm, which can effectively remove noise in the image while retaining the detail information of the image; the distortion correction algorithm accurately corrects the image based on the pre-calibrated camera parameters to ensure the accuracy of the image; the ROI extraction algorithm automatically extracts the region of interest according to the position information of the shelf and the goods, reducing the data volume for subsequent processing.

[0089] AI Algorithm Recognition and Data Processing:

[0090] Shelf Recognition Algorithm: The convolutional neural network (CNN) is used to recognize and locate the shelves in the image. Through a large number of sample trainings, the CNN model can accurately judge the position, direction of the shelf, and whether it is aligned with the camera center. The recognition accuracy can reach more than 99%, providing an accurate reference basis for the robot to adjust its position.

[0091] Shelf Warehouse Number Recognition Algorithm: Based on OCR technology and deep learning models, first locate and segment the warehouse number area in the image, and then use the trained character recognition model to recognize the warehouse number. This algorithm adopts transfer learning technology and can achieve high-precision recognition even with a small number of samples. The recognition accuracy rate reaches over 98%, ensuring the consistency between the inventory data and the location information in the warehouse system.

[0092] Pallet and Pallet Code Recognition Algorithm: Adopt an algorithm based on object detection and feature matching to recognize pallets and pallet codes. By learning and training the features such as the shape and color of the pallet, the algorithm can accurately recognize the presence and location of the pallet. For the recognition of pallet codes, use image enhancement technology and character segmentation algorithms to improve the recognition accuracy. The recognition accuracy rate reaches over 97%, ensuring the integrity of the hierarchical structure of the inventory data.

[0093] Goods Counting Algorithm: Comprehensively use object detection, image segmentation, and deep learning algorithms to accurately recognize the number of goods on the pallet. For regularly stacked goods, adopt a deep learning-based object detection algorithm to quickly and accurately count the number of goods; for complex stacking situations (such as double-layer stacking or incomplete stacking), use image segmentation algorithms to segment and count the goods, and use a deep learning model to judge the stacking state of the goods. For situations where the algorithm cannot accurately judge, mark them as suspicious points and push the relevant pictures to the background or PAD for manual secondary confirmation.

[0094] Data Comparison and Update: Compare the recognized inventory data with the data in the warehouse system in real time, and use data comparison algorithms to quickly find out the difference information. For the difference data, the system automatically marks it and generates a detailed inventory report, including information such as the location, quantity, and type of the difference. At the same time, synchronize the updated inventory data to the warehouse system in a timely manner to ensure the real-time and accuracy of the inventory data.

[0095] Manual Secondary Confirmation: The inventory personnel receive the images of the suspicious points pushed by the system through the handheld PAD and use the operation interface on the PAD for manual confirmation or supplementary inventory. For corners or special areas that the robot cannot reach, the system makes obvious marks on the map, and the inventory personnel conduct manual inventory according to the prompts and input the results into the system in real time through the PAD. During the manual confirmation process, the system records and traces the operations to ensure the reliability and traceability of the data.

[0096] Data Storage and Report Generation: The inventory data is stored in a high-performance database on the local server. Distributed storage and backup technologies are adopted to ensure the security and reliability of the data. Meanwhile, the data is backed up to the cloud storage server through the cloud management system to achieve off-site backup and load tolerance of the data. The system automatically generates a detailed inventory report, and the report content includes information such as inventory time, inventory results, discrepancy details, and analysis of abnormal situations. The inventory report supports export in multiple formats, facilitating the viewing and analysis by warehouse management personnel.

[0097] Special Scenario Handling

[0098] Areas Inaccessible to Robots: During the process of building the warehouse map model, the system comprehensively scans the warehouse environment through lidar and vision sensors, automatically identifies the areas inaccessible to robots, and stores the information of these areas in a dedicated database for inaccessible areas. During the inventory process, when the robot detects an inaccessible area, the system automatically marks this area as an area to be manually inventoried and sends a reminder to the inventory personnel through a handheld PAD. According to the system prompt, the inventory personnel use the handheld PAD to conduct a manual inventory of this area and enter the inventory results into the system in real time through a 4G data link. The entered data is verified and reviewed by the system and then merged with the robot inventory data to ensure that the goods in all areas of the warehouse can be accurately inventoried.

[0099] Goods Stacked in Multiple Layers with Partial Occlusion: The algorithm conducts in-depth analysis of the goods images, uses deep learning models to extract information such as the length-width ratio, area difference, and texture features of the goods, and intelligently determines whether the goods are stacked in two layers. For the case where the sizes of the front boxes are the same, by calculating the spacing and arrangement rules between the boxes, it is determined as single-layer stacking; for the case where the sizes of the front boxes are quite different, combined with the spatial position relationship and occlusion situation of the goods, it is determined as cross-stacking and automatically defined as double-layer. For the goods determined to be stacked in two layers or with doubts, the system automatically marks them as requiring manual confirmation and pushes the relevant images and analysis results to the handheld PAD. The inventory personnel conduct manual confirmation of the goods stacking situation based on the images and prompt information to ensure the accuracy of the inventory results.

[0100] More than Two Shelves Side by Side: When encountering the special scenario of more than two shelves side by side, first, use the AI algorithm to conduct multi-dimensional analysis of the images. The algorithm fuses and analyzes the images from different angles and attempts to identify the goods situation on the middle shelf. If the algorithm cannot accurately identify, the system automatically marks this area as a suspicious point and pushes the relevant images and position information to the handheld PAD. The inventory personnel obtain the image information through the handheld PAD and conduct secondary manual confirmation and processing based on the actual on-site situation. During the manual confirmation process, the system provides auxiliary tools, such as magnifying and rotating the images, to help the inventory personnel better observe the goods situation.

[0101] Incomplete stacking of goods: The algorithm judges the integrity of the goods stacking through perspective distortion analysis and deep learning models. By learning and analyzing the shape, position, and arrangement rules of the goods, the algorithm can identify whether there are vacancies or incomplete situations in the goods stacking. If vacancies or incomplete stackings are found, the system automatically marks them as suspicious points and pushes relevant pictures and analysis results to the PAD or the background. The inventory personnel manually confirm and supplement the inventory of the goods stacking situation based on the pictures and prompt information to ensure the accuracy of the inventory data.

[0102] System iteration and optimization

[0103] Algorithm iteration: A large amount of image data collected during the inventory process is used to regularly train and optimize the AI algorithm. Incremental learning and transfer learning techniques are adopted to continuously update the algorithm model, improving the accuracy and generalization ability of the algorithm. Support for user-defined algorithm modules allows users to upload custom algorithm models according to the characteristics and requirements of their own warehouses. The system realizes rapid deployment and application through a distributed cluster architecture, and the new algorithm model can be put into use immediately after deployment, improving the adaptability and flexibility of the system.

[0104] Hardware upgrade: The inventory robot supports hardware expansion. Users can easily add more cameras or computing power devices to the robot according to changes in the warehouse environment and increasing inventory requirements. For example, in areas with high shelf heights or complex goods distributions, high-resolution cameras can be added to improve the quality of image acquisition; in cases with large data processing requirements, computing power devices can be added to enhance the processing ability of the robot. The hardware upgrade process is simple and convenient, without the need for large-scale transformation of the robot, reducing the user's usage cost.

[0105] User experience optimization: The handheld PAD interface adopts a user-friendly design. According to the user's operation habits and feedback, the interface layout and operation process are continuously optimized. Support for multiple device sizes ensures a consistent user experience in different operating environments. The cloud management system provides real-time monitoring and data analysis functions. Through intuitive charts and data displays, it helps managers quickly grasp the inventory progress and results. At the same time, the system provides online help and training resources to facilitate users to obtain technical support and learn relevant knowledge during use.

[0106] Embodiment:

[0107] System architecture and hardware composition

[0108] Inventory robot body:

[0109] Chassis and Lifting Device: In a large e-commerce warehouse with an area of 50,000 square meters, the highest shelf height reaches 3.5 meters, the aisles are narrow and the goods are stacked densely. The customized mobile robot chassis, with its high-performance drive system, can flexibly shuttle through narrow aisles and successfully avoid various goods and obstacles. Once, a batch of goods was temporarily added in the warehouse, blocking the originally planned path. The robot relied on its excellent obstacle avoidance and obstacle crossing capabilities to quickly re-plan the path and successfully complete the inventory task. During a high-level shelf inventory, its lifting device smoothly rose to the 3-meter position in just 5 seconds, accurately positioning to the target shelf, providing guarantee for rapid data collection.

[0110] Multi-layer Camera Disk: In the warehouse of a mechanical parts manufacturing enterprise, the shelf heights are different, and the sizes and shapes of the parts vary. The customized acquisition cameras on the multi-layer camera disk, with the characteristics of high resolution, low noise and wide dynamic range, successfully identified the part labels of various sizes. For example, for the tiny labels of some small and precise parts, they can also be clearly captured. The seamless splicing achieved by its multi-focus fusion technology enables the staff to comprehensively understand the distribution of parts on the shelves. Moreover, when the warehouse shelf layout was adjusted, the staff quickly reinstalled and adjusted the camera disk through the quick installation and disassembly structure to adapt to the new warehouse scenario.

[0111] Handheld PAD: In a cold chain warehouse, the environmental temperature is low and the humidity is high. The three-proof handheld PAD operates stably in this harsh environment, and its rugged and durable shell effectively resists the erosion of low temperature and humidity. When the warehouse staff uses the PAD for manual secondary confirmation, the automatic focusing and optical image stabilization functions of the built-in high-definition camera can take clear and stable images even when the staff's hands tremble slightly due to cold. Through the encrypted transmission protocol, the PAD communicates with the robot and the local server safely and reliably, and there has never been a data leakage situation.

[0112] Warehouse Local Server: In the regional distribution center warehouse of a chain supermarket, with the continuous expansion of business, the inventory data volume has increased sharply. The warehouse local server adopts a high-performance enterprise-level server and easily copes with the growth of data volume. When the inventory data volume increased by 50% within a year, the server dynamically expanded the capacity and timely added hard disks and memory to ensure the stable operation of the system. During a large-scale promotion event, multiple servers achieved load balancing through the cluster mode and quickly processed a large amount of inventory data and sales data, providing strong support for the replenishment and distribution of the supermarket.

[0113] Cloud-based Business Management System: A multinational enterprise has multiple warehouses globally. Through the cloud-based business management system, managers can monitor the situation of each warehouse in real time. For example, managers located in Europe can view the inventory count progress and equipment operation status of Asian warehouses in real time. In an algorithm update, through the one-click deployment of new algorithm models, the automated deployment process completed the algorithm update for all warehouses in just 30 minutes, greatly improving the system's iteration efficiency. Data and algorithm models are also shared among different warehouses through this system, achieving optimized resource allocation and reducing operating costs.

[0114] Inventory Count Process and Algorithm Implementation

[0115] Map Modeling and Path Planning: In a newly built pharmaceutical warehouse, before the first inventory count, the robot used lidar to conduct a 3D scan of the warehouse and constructed an accurate map. When identifying the shelf positions and numbers, using deep learning algorithms, it completed the identification and labeling of thousands of shelves in the warehouse in just 2 hours and accurately matched them with the location information in the warehouse system. The path planning algorithm planned an optimal path for the robot based on the warehouse layout and shelf distribution. During the actual inventory count, the robot traveled along the planned path, successfully avoiding ongoing cargo handling operations in the warehouse and efficiently completing the inventory count task. The entire inventory count time was shortened by 70% compared to traditional manual inventory counting.

[0116] Image Acquisition and Preprocessing: In an electronics warehouse, before the robot reaches the shelf, it accurately locates through high-precision position sensors and positioning systems. During an inventory count of a certain shelf, the robot accurately stopped in front of the shelf for 10 seconds, and the multi-layer camera quickly acquired image data on both sides of the shelf. The visual preprocessing server uses parallel computing technology to quickly denoise, correct distortion, and extract ROIs from the images. For example, for an image with some areas overexposed due to light reflection, the denoising algorithm effectively removed the noise while retaining the product label information in the image; the distortion correction algorithm accurately corrected the distorted image caused by the camera angle problem, ensuring the accuracy of subsequent AI algorithm recognition.

[0117] AI Algorithm Recognition and Data Processing:

[0118] Shelf Recognition Algorithm: In a furniture warehouse, a convolutional neural network is used to identify and locate the shelves in the images. After training with a large number of samples, the accuracy of this algorithm in identifying the shelf position, orientation, and whether it is aligned with the camera center reached 99.5% during actual inventory counts. Once, when the robot was conducting an inventory count of a shelf in the corner of the warehouse, it accurately judged the tilt angle of the shelf and adjusted its own position in a timely manner to ensure the accuracy of data collection.

[0119] Warehouse Shelf Number Recognition Algorithm: In a clothing warehouse, a warehouse shelf number recognition algorithm based on OCR technology and deep learning models can accurately identify even when the warehouse number part is blurred. Through transfer learning technology, the recognition accuracy reaches 98.5% with a small number of samples. Once, due to the humid warehouse environment, the paper labels of some warehouse shelf numbers faded and blurred, but the algorithm still successfully identified the warehouse numbers, ensuring the consistency of the inventory data and the location information in the warehouse system.

[0120] Pallet and Pallet Code Recognition Algorithm: In a food warehouse, an algorithm based on object detection and feature matching is used to identify pallets and pallet codes. By learning and training the features such as the shape and color of the pallets, the algorithm can accurately identify the presence and location of the pallets. During an inventory, pallets of different specifications and colors were successfully identified. For the recognition of pallet codes, using image enhancement technology and character segmentation algorithms, the recognition accuracy reached 97.5%, ensuring the integrity of the hierarchical structure of the inventory data.

[0121] Goods Counting Algorithm: In a daily necessities warehouse, for regularly stacked goods, a deep learning-based object detection algorithm is used to quickly and accurately count the number of goods. During an inventory of a toothpaste shelf, the number of toothpastes was accurately counted in just a few seconds. For complex stacking situations, such as double stacking or incomplete stacking, image segmentation algorithms and deep learning models are used to accurately judge the stacking state of the goods. During an inventory of a shampoo shelf, the double stacking situation of some goods was accurately identified, and for situations that the algorithm could not accurately judge, they were marked as suspicious points and relevant pictures were pushed to the background for manual secondary confirmation.

[0122] Data Comparison and Update: In an auto parts warehouse, the identified inventory data is compared with the data in the warehouse system in real time. After an inventory, the system quickly found the difference information using a data comparison algorithm and found that the actual number of a certain type of car tire was 5 less than the system record. The system automatically marked the difference data and generated a detailed inventory report, including information such as the location, quantity, and type of the difference. At the same time, the updated inventory data was timely synchronized to the warehouse system, ensuring the timeliness and accuracy of the inventory data and providing reliable data support for subsequent procurement and sales.

[0123] Manual secondary confirmation: In a chemical raw material warehouse, the inventory takers receive the images of suspicious points pushed by the system through a handheld PAD. During an inventory of barreled chemical raw materials, for the stacking situations of goods that the robot cannot accurately identify, the inventory takers use the operation interface on the PAD for manual confirmation. For the warehouse corners that the robot cannot reach, the system makes obvious marks on the map. The inventory takers conduct manual inventory according to the prompts and enter the results into the system in real time through the PAD. During the manual confirmation process, the system records and traces the operations, ensuring the reliability and traceability of the data and avoiding data errors caused by manual operation mistakes.

[0124] Data storage and report generation: In an overseas warehouse of an e-commerce company, the inventory data is stored in a high-performance database on the local server, adopting distributed storage and backup technologies. In a situation where a natural disaster occurred locally and caused a power outage, the data remained safe and reliable without any loss or damage. At the same time, the data is backed up to the cloud storage server through the cloud management system, realizing off-site backup of the data. The system automatically generates a detailed inventory report, and the report content includes information such as inventory time, inventory results, discrepancy details, and analysis of abnormal situations. The warehouse management personnel discovered the problems existing in the inventory management in a timely manner by exporting the inventory report and took corresponding measures for improvement.

[0125] Special scenario handling

[0126] Areas inaccessible to the robot: In a warehouse, due to some areas being occupied by large equipment, the robot cannot reach them. During the map modeling process, the system automatically identifies these inaccessible areas through lidar and vision sensors and stores their information in a dedicated inaccessible area database. During the inventory process, when the robot detects an inaccessible area, the system automatically marks this area as an area to be manually inventoried and sends a reminder to the inventory takers through the handheld PAD. The inventory takers conduct manual inventory of this area using the handheld PAD according to the prompts and enter the inventory results into the system in real time through the 4G data link. The data entered is verified and audited by the system and then merged with the robot inventory data, ensuring that the goods in all areas of the warehouse can be accurately inventoried.

[0127] Multi-layer stacking of goods with partial occlusion: In an electrical appliance warehouse, the algorithm performs in-depth analysis on the images of goods, uses a deep learning model to extract information such as the length-width ratio, area difference, and texture features of the goods, and intelligently determines whether the goods are stacked in two layers. For TV packaging boxes with the same size frames on the front, by calculating the spacing and arrangement rules between the boxes, it is determined to be single-layer stacking; for air conditioner packaging boxes with significantly different sizes on the front, combined with the spatial position relationship and occlusion situation of the goods, it is determined to be cross-stacked and automatically defined as double-layer. For goods determined to be double-layer stacked or with doubts, the system automatically marks them as requiring manual confirmation and pushes the relevant images and analysis results to the handheld PAD. The inventory takers make manual confirmation of the goods stacking situation based on the images and prompt information, ensuring the accuracy of the inventory results.

[0128] More than two shelves side by side: In a building materials warehouse, when encountering a special scenario of more than two shelves side by side, the AI algorithm is first used to perform multi-dimensional analysis on the images. During an identification of the tiles on the middle shelf, the algorithm successfully identified the information of some tiles by fusing and analyzing images from different angles. However, for some tiles that are severely occluded, the algorithm cannot accurately identify them, and the system automatically marks this area as a suspicious point and pushes the relevant images and position information to the handheld PAD. The inventory takers obtain the image information through the handheld PAD and perform manual secondary confirmation and processing based on the actual situation on site. During the manual confirmation process, the system provides auxiliary tools such as magnifying and rotating the images to help the inventory takers better observe the goods situation, and finally accurately completes the inventory task.

[0129] Incomplete stacking of goods: In a stationery warehouse, the algorithm judges the integrity of the goods stacking through perspective distortion analysis and a deep learning model. During an inventory of the notebook shelves, by learning and analyzing the shape, position, and arrangement rules of the goods, the algorithm identified that there were empty spaces on some shelves. The system automatically marks it as a suspicious point and pushes the relevant pictures and analysis results to the PAD. The inventory takers make manual confirmation of the goods stacking situation based on the pictures and prompt information and conduct supplementary inventory, ensuring the accuracy of the inventory data.

[0130] System iteration and optimization

[0131] Algorithm Iteration: In a fast-moving consumer goods warehouse, a large amount of image data collected during the inventory process is used to regularly train and optimize the AI algorithm. Incremental learning and transfer learning techniques are adopted to continuously update the algorithm model. For example, after a change in the stacking method of warehouse goods, the algorithm can quickly adapt to the new stacking method by learning the newly collected image data, improving the recognition accuracy. The system supports user-defined algorithm modules. Warehouse managers uploaded a user-defined algorithm model according to the characteristics of their warehouse, such as a large variety of goods and frequent updates. The system is quickly deployed and applied through a distributed cluster architecture. The new algorithm model is immediately put into use after deployment, improving the adaptability and flexibility of the system and effectively solving the problem of inaccurate recognition of new types of goods in previous inventories.

[0132] Hardware Upgrade: In a high-end electronics warehouse, due to the high shelf height and complex goods placement, higher quality requirements for image acquisition are needed. Users added high-resolution cameras to the inventory robots according to the changes in the warehouse environment and the increasing inventory requirements. During an inventory of new mobile phone packaging boxes, the newly added high-resolution cameras clearly captured the tiny markings on the packaging boxes, improving the accuracy of the inventory. During promotional activities with high data processing requirements, computing power equipment was added to enhance the processing ability of the robots, ensuring that the inventory tasks could be efficiently completed in a short time. The hardware upgrade process was simple and convenient, and the installation and debugging were completed in only half a day, reducing the user's usage cost.

[0133] User Experience Optimization: In a comprehensive warehouse, the handheld PAD interface continuously optimizes the interface layout and operation process according to the user's operating habits and feedback. Previously, when staff operated the PAD, they often made incorrect operations resulting in data entry errors. After optimization, the interface is more concise and clear, and the operation process is more straightforward, effectively reducing the occurrence of incorrect operations. The cloud management system provides real-time monitoring and data analysis functions. Through intuitive charts and data displays, it helps managers quickly grasp the inventory progress and results. During a large-scale inventory, managers used the cloud management system to real-time understand the inventory progress in each area and timely adjusted the allocation of personnel and equipment to ensure that the inventory tasks were completed on time. At the same time, the system provides online help and training resources. Newly recruited staff quickly mastered the use method of the system through online learning, improving work efficiency.

[0134] The above has introduced in detail the technical solutions provided by the embodiments of the present invention. Specific examples are used herein to elaborate on the principles and implementation manners of the embodiments of the present invention. The description of the above embodiments is only applicable to helping understand the principles of the embodiments of the present invention; at the same time, for those of ordinary skill in the art, according to the embodiments of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A non-standard warehouse AI intelligent inventory system, characterized by: include: Robot / drone automatic inventory counting system, which uses AI technology and robots or drones to automatically identify and read structured data from storage location numbers to cargo quantities, and compares them with the warehouse system; AI algorithm and path planning module, including shelf recognition algorithm, shelf library number recognition algorithm, pallet recognition algorithm, pallet code recognition algorithm, pallet cargo counting algorithm and robot path planning algorithm; special scenario processing solution module, providing processing methods for special scenarios such as corners that are inaccessible to robots, multi-layer stacking of cargo and partial obstruction, and more than two shelves side by side; The data collection and processing module collects image data through the customized camera on the robot, uses AI algorithms to perform real-time analysis and processing, identifies relevant information and compares it with the warehouse system, marks suspicious points and pushes photos for manual secondary confirmation.

2. According to claim 1, a non-standard warehouse AI intelligent inventory system is characterized by: The robot / drone automatic inventory system includes: The robot chassis has the ability to walk automatically and has a lifting device. It can be raised to a height of 4 meters on the shelf. It supports 4G data transmission and indoor positioning. It can remotely control the path walking and provides a CAN interface. The robot body can be equipped with a visual pre-processing server and an AI reasoning processing server to run related control and inventory logic programs. Adopting array cameras and customized acquisition cameras with multiple focal lengths, it can recognize label content and global shelf vision at a distance of 2 meters through multiple camera groups and multi-focal length fusion stitching technology, and can be connected to the visual pre-processing server; Camera tray and multi-layer camera tray, multiple customized acquisition cameras can be placed at the same height, combined into a multi-layer structure according to the shelf height and installed on the robot lifting arm; A three-proof handheld PAD with a camera, running the Android system and custom programs, linked to the inventory robot via a 4G data link; The warehouse local server used to run inventory management programs and data storage is linked to on-site equipment via 4G / 5G data links.

3. According to claim 1, the AI ​​intelligent inventory counting system for non-standard warehouses is characterized in that , in the AI ​​algorithm and path planning module: Shelf recognition algorithm, used to identify the location of the shelf in the current picture, determine whether the shelf is on the left or right, whether it is aligned with the center of the camera, etc.; Shelf stock number recognition algorithm, used to identify the specific content of the stock number of the shelf in the current picture; Pallet recognition algorithm to identify the presence and location of pallets; Pallet code recognition algorithm, used to identify the specific content of the pallet code; The cargo counting algorithm on the pallet is used to identify the number of rectangular boxes on the shelf, determine the cargo stacking status, and report the information to the backend; The robot path planning algorithm is used to implement the actual path planning of multiple robots in the warehouse, using the shortest path for inspection and inventory.

4. According to claim 1, the AI ​​intelligent inventory counting system for non-standard warehouses is characterized in that , in the special scenario processing solution module: Corner areas that cannot be reached by robots are supplemented by manual inventory; When goods are stacked in multiple layers and partially blocked, the stacking method is determined based on the consistency of the box size frame on the front. When more than two shelves are placed side by side, they are identified and processed through AI algorithms or specially handled manually.

5. According to claim 1, the AI ​​intelligent inventory counting system for non-standard warehouses is characterized in that: The system further comprises: The system architecture and hardware components include the inventory robot, handheld PAD, warehouse local server and cloud business management system; Inventory process and algorithm implementation, including map modeling and path planning, image acquisition and preprocessing, AI algorithm recognition and data processing, data comparison and update, manual secondary confirmation, data storage and report generation; In the system iteration and optimization part, continuous improvement of the system is achieved through algorithm iteration, hardware upgrades and user experience optimization.

6. According to claim 5, a non-standard warehouse AI intelligent inventory system is characterized in that: The inventory counting robot body comprises: The customized mobile robot chassis has the function of automatic walking, can be raised to 4 meters high on the shelf, supports 4G data transmission and indoor positioning functions, and has visual pre-processing servers and AI reasoning processing servers placed on the chassis; A multi-layer camera tray is installed on the robot's lifting arm. Each layer of the camera tray is equipped with multiple sets of customized acquisition cameras and uses multi-focal length fusion technology.

7. According to claim 5, the AI ​​intelligent inventory counting system for non-standard warehouses is characterized by: The handheld PAD is an Android tablet with three-proof functions and a camera. It communicates with the robot and the local server through a 4G data link to display the inventory status in real time.

8. According to claim 5, the AI ​​intelligent inventory counting system for non-standard warehouses is characterized by: The warehouse local server is used to run the inventory management program, store inventory data, and interact with robots, PADs and cloud management systems, supporting dynamic expansion and cluster mode.

9. The non-standard warehouse AI intelligent inventory counting system according to claim 5 is characterized by: The cloud-based business management system is used to remotely monitor and manage inventory tasks, realize multi-site linkage work, and support one-click deployment of new algorithm models.

10. The non-standard warehouse AI intelligent inventory counting system according to claim 5 is characterized by: The algorithm iteratively utilizes the image data collected during the inventory process to regularly train and optimize the AI ​​algorithm, supports user-defined algorithm modules, and achieves rapid deployment and application through a distributed cluster architecture.

Images