WO2021226893A1 - 物件辨识系统及相关装置 - Google Patents

物件辨识系统及相关装置 Download PDF

Info

Publication number
WO2021226893A1
WO2021226893A1 PCT/CN2020/090096 CN2020090096W WO2021226893A1 WO 2021226893 A1 WO2021226893 A1 WO 2021226893A1 CN 2020090096 W CN2020090096 W CN 2020090096W WO 2021226893 A1 WO2021226893 A1 WO 2021226893A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
radio frequency
key
frequency identification
fog
Prior art date
Application number
PCT/CN2020/090096
Other languages
English (en)
French (fr)
Inventor
张圻毓
Original Assignee
鸿富锦精密工业(武汉)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 鸿富锦精密工业(武汉)有限公司 filed Critical 鸿富锦精密工业(武汉)有限公司
Priority to US17/266,126 priority Critical patent/US11429800B2/en
Priority to PCT/CN2020/090096 priority patent/WO2021226893A1/zh
Priority to CN202080000831.2A priority patent/CN113939832A/zh
Publication of WO2021226893A1 publication Critical patent/WO2021226893A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K7/00—Methods or arrangements for sensing record carriers, e.g. for reading patterns
    • G06K7/10—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation
    • G06K7/10009—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves
    • G06K7/10297—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation sensing by radiation using wavelengths larger than 0.1 mm, e.g. radio-waves or microwaves arrangements for handling protocols designed for non-contact record carriers such as RFIDs NFCs, e.g. ISO/IEC 14443 and 18092
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/20—Scenes; Scene-specific elements in augmented reality scenes

Definitions

  • the present disclosure relates to an object identification system, and more specifically, to an object identification system for food delivery and logistics packages.
  • the delivery terminal still relies on manpower to complete the door-to-door signing.
  • the demand for delivery personnel and freight drivers is always constant, but because of the human factors in the middle of the logistics process, it is likely that the wrong goods will be delivered to consumers or food will suffer.
  • the existing shipping method is to pick up the goods in the warehouse and send them to the delivery staff to scan the shipment.
  • the delivery staff photographs the address on the list and arrives in front of the consumer's house, and then removes the goods from the minivan or locomotive and sends them to the consumer for signature collection.
  • the delivery staff go to the store to wait. After the consumer places the order, they pick up the food directly from the store, put it in the aluminum foil insulation bag on the motorcycle, and take out the food after riding to the destination and hand it over to the consumer for receipt.
  • the present disclosure provides an object identification system, which includes: an input device used to obtain an image data and a radio frequency identification (RFID) data of an object; a processing device connected to the input device to perform a Model training program, where the model training program includes: extracting an object feature based on the image data; generating a classification data for identifying the object based on the object feature and radio frequency identification, where the radio frequency identification is used to verify the correctness of the classification data, To generate a deep learning model; and an output device connected to the processing device for generating an object data corresponding to the object according to the classification data and radio frequency identification data generated by the deep learning model for the object.
  • RFID radio frequency identification
  • the present disclosure provides a fog terminal device for object identification, including: a processing unit for executing a program code; and a storage unit connected to the processing unit for storing program code; wherein the program code instructs the processing unit to execute
  • RFID Radio Frequency Identification
  • the present disclosure provides a mobile device including: a processing unit for executing a program code; and a storage unit connected to the processing unit for storing program code; wherein the program code instructs the processing unit to perform the following steps: obtain an object An image data and a radio frequency identification (RFID); send image data and radio frequency identification to a fog-end device for object identification; receive an object data and an augmented reality about the object from the fog-end device Data; and based on the object data and the augmented reality data corresponding to the object data, an augmented reality image is generated.
  • RFID radio frequency identification
  • FIG. 1 is a schematic diagram illustrating the architecture of an object recognition system according to an embodiment of the present disclosure.
  • FIG. 2 is a flowchart illustrating the operation of the object recognition system according to an embodiment of the present disclosure.
  • Fig. 3 is a flowchart illustrating front-end logistics shipment according to an embodiment of the present disclosure.
  • FIG. 4 is a flowchart illustrating the recognition of artificial intelligence objects according to an embodiment of the present disclosure.
  • FIG. 5 is a schematic diagram illustrating the combination of an object recognition system and augmented reality technology according to an embodiment of the present disclosure.
  • Fig. 6 is a schematic diagram illustrating a smart lock application according to an embodiment of the present disclosure.
  • Fig. 7 is a flow chart illustrating the operation of locking the smart lock according to an embodiment of the present disclosure.
  • Fig. 8 is a flowchart illustrating the unlocking operation of the smart lock according to an embodiment of the present disclosure.
  • Fig. 9 is a flowchart illustrating the unlocking operation of the smart lock according to an embodiment of the present disclosure.
  • Fig. 10 is a flowchart illustrating an artificial intelligence logistics system according to an embodiment of the present disclosure.
  • FIG. 11 is a schematic diagram illustrating an image displayed by the object recognition system combined with augmented reality according to an embodiment of the present disclosure.
  • FIG. 1 is a schematic diagram of the object recognition system architecture.
  • an object recognition system includes an input device, an output device, and a processing device.
  • the input device, output device, and processing device each include at least one processing unit and a storage unit.
  • the input device 10 may be a mobile device, such as a mobile phone or a tablet and other related interface devices.
  • the object 11 is a freight item or food, and does not have to be a square or a fixed shape. It only needs the characteristic size and color package to provide the processing device 12 for identification and use.
  • the user uses the input device 10, such as a lens on a mobile device, to take an image of the object 11 and send it to the processing device 12 for object identification.
  • the processing device 12 generates a deep learning model through classification and model training, and can more accurately determine what the object 11 is. It is worth noting that the processing device 12 of the present disclosure applies radio frequency identification (RFID) technology, such as near-field communication (NFC) identification data (or NFC tags), for model training, In order to realize the deep learning of artificial intelligence.
  • RFID radio frequency identification
  • the logistics personnel bring the input device 10 close to the object 11 to scan the radio frequency identification data (such as an NFC tag) of the object 11 and transmit the radio frequency identification data to the processing device 12 for the processing device 12 to perform data comparison.
  • the processing device 12 after the processing device 12 has identified the object 11, it will generate object data and send it to the output device to display the object identification result.
  • the output device and the input device may be the same or different devices, and may be devices with display functions such as mobile devices or screens.
  • the above-mentioned transmission may be any possible connection method such as wired, wireless network, bluetooth, etc.
  • FIG. 2 is a flowchart illustrating the operation of the object recognition system according to an embodiment of the present disclosure.
  • the process 200 shown in FIG. 2 can be executed by the processing unit in the processing device 12 shown in FIG. 1 to execute the program code in the storage unit, where the program code is used to instruct the processing unit to perform the following steps:
  • Step 202 Obtain image data and radio frequency identification data of the object.
  • Step 204 Perform a model training process, where the model training process includes: extracting an object feature based on the image data; generating a classification data for identifying the object based on the object feature and radio frequency identification data, where the radio frequency identification data is used Verify the correctness of the classified data to generate a deep learning model.
  • Step 206 Generate an object data corresponding to the object according to the classification data and the radio frequency identification data generated by the deep learning model for the object.
  • the processing device 12 conducts deep learning model training based on the image data of the object and the object radio frequency identification data, thereby generating high-precision object identification results, which can effectively reduce the situation of delivery personnel sending wrong products and improve The efficiency of delivery staff picking up goods.
  • the processing device 12 of the present disclosure uses marginal calculations to process data at the fog end, so it can be dispersed in the user terminal for object identification, that is, the user terminal does not need to send the original image data and radio frequency identification data back to the server, and the server does not need to be updated. Large processing efficiency to deal with the needs of user terminals, this method can also allow users to have no sense of delay, reduce the use threshold of network speed, and logistics companies can also reduce the cost of large-scale server computing.
  • the object identification system of the present disclosure realizes object identification through artificial intelligence data processing technology.
  • the object identification system of the present disclosure uses radio frequency identification data (such as NFC tags) as the control group of artificial intelligence deep learning, which can effectively improve the accuracy of object identification, and the deep learning model can be trained continuously for a long time.
  • radio frequency identification data such as NFC tags
  • the collected image data has high big data value, and provide more data for future retraining of artificial intelligence, so it can effectively obtain the expected results, and solve the innate object identification difficulties and logistics
  • the bottleneck of sorted picking improves the efficiency of fast shipping.
  • Figure 3 is a flow chart of front-end logistics shipment.
  • Each item AZ is a product on the logistics line or the number of a product that will be out of the warehouse.
  • the artificial intelligence processing unit 310 (as shown in Figure 1 The processing unit of the processing device 12) analyzes the data to realize the classification of the object AZ.
  • the artificial intelligence processing unit 310 can obtain the image of the object A-Z through the input device 300 and store it as a key feature image.
  • the artificial intelligence processing unit 310 includes multiple arithmetic units, and the detailed functions are as follows.
  • the recognition calculation unit 311 can confirm whether the image has characteristics or just environmental characteristic values through recognition calculations, and make a distinction judgment first. If there are no features of the object, the subsequent recognition calculations will not be actively performed.
  • the recognition calculation mainly uses artificial intelligence algorithms to judge the characteristic image, reducing the possibility of misjudgment and greatly improving the system
  • the accuracy is then transferred to the arithmetic unit 313 to calculate the probability of the coincidence of the feature value.
  • the feature meanings represented by each point of the object feature are different. Therefore, the probability can be effectively calculated through the arithmetic unit 313, and finally sent Go to the classification calculation unit 314 to generate object classification data.
  • the same object will be categorized. If there is a way to categorize, the data will be merged, and the images that cannot be identified will be separately archived in unknown objects for use in future AI deep learning.
  • the processed object classification data/results will be output to the radio frequency identification device 320.
  • the radio frequency identification data of the object itself can be captured by the radio frequency identification chip on the mobile device for the artificial intelligence processing unit 310 to confirm/compare the classification results. Then train a high-precision deep learning model.
  • the augmented reality processing system 330 generates augmented reality display data, so that it can finally display the object data (such as detailed object data, shipping-related information, object name and logistics sequence, etc.) through the augmented reality method. .
  • the object classification data and radio frequency identification data are transmitted back to the database 340 for storage for later use and analysis.
  • the object data and the augmented reality display data (such as the above detailed object data and shipping-related information) are transmitted to the output device 350 (such as the screen of a mobile device) to display the result.
  • this disclosure uses an object identification system to double-identify objects to reduce the uncertainty of human factors, thereby enhancing consumer confidence and protecting the rights and interests of both parties.
  • Artificial intelligence can be said to have evolved with each passing day.
  • Machine learning is a very important part of artificial intelligence. It can be divided into two parts: training and prediction.
  • the machine can be made to learn object judgments like humans.
  • the learning process corrects the correctness of the learning judgment, so that the accuracy of the judgment continues to rise and maintains a certain degree of correctness.
  • the part of the prediction is after the machine training, without assistance and correction. In this way, we can proactively predict what the object is, which is also a predictive matter in our understanding.
  • supervised learning means that all data has standard answers. It can be used for judging errors during machine learning, and it will be more accurate during prediction. Label the data and extract the features from the data. In future predictions, you only need to find this feature to identify the object or judge the result. This method is closer to manual classification. It is easier for machines but will increase the pre-processing burden for humans.
  • Unsupervised learning means that all data has no standard answers and cannot provide machine learning judgment errors. Therefore, the machine must Finding the answer by yourself will often make the predictions less accurate, but it can reduce the burden of collation after data collection, but the relative cost is that the prediction error is large, and it is suitable for classification of complex and difficult data sets.
  • Semi-supervised learning is a small amount of data. There are answers, you can provide machine learning reference, but most of them still have no answers. The machine must find the answer by itself, which is equivalent to combining the advantages of the above two. This method requires only a few manual classifications, and at the same time, it can improve the accuracy of some predictions. , Is one of the more commonly used methods at present.
  • the present disclosure uses supervised machine learning, that is, uses radio frequency identification data, to determine the correctness/error of the object classification results identified by the artificial intelligence processing unit 310, and then continuously trains the deep learning model to improve the object recognition performance Accuracy.
  • the present disclosure displays the object data through augmented reality after the object is identified, so that the logistics personnel can effectively identify the item and detailed data of the object, such as the detailed data describing the object itself and the data related to the transportation of the object.
  • freight drivers or delivery personnel can use the lens of a mobile phone to obtain image data and send it to a processing device with marginal computing artificial intelligence.
  • the processing device does pre-processing for each image, eliminates the image background, removes noise, etc., and then extracts and analyzes the location-related information of the object from the image, and then uses the image coordinate value of the location to calculate and store the data content. Accumulate more data to further improve the accuracy of object judgment.
  • object classification data and object data can be combined and stored with augmented reality data when archiving. Therefore, the delivery personnel can quickly handle a large amount of transported goods or food, and realize the artificial intelligence logistics system from the delivery of the meal to the consumer, ensuring food safety and reducing the error rate of logistics. Therefore, through the object identification system proposed in this disclosure, after the front-end deep learning model and logistics data database are correctly established (such as classification results, image features, and augmented reality data, etc.), the delivery personnel can identify through mobile devices You don’t need to scan the NFC tags of the labeled objects one by one to confirm the objects.
  • FIG. 4 is a flowchart illustrating artificial intelligence object recognition according to an embodiment of the present disclosure.
  • the present disclosure can obtain image data through the input device (step N01), and then send it to the processing device to determine whether there is an object feature value (step N02), otherwise, return to step N01 to continue capturing image data. If an image feature is identified, the processing device stores the image feature (step N03) for future deep learning. After storing the image feature, the processing device determines whether the image feature is in the previously determined database (step N04). For example, there are different feature algorithms corresponding to different image features, which can improve the accuracy of object recognition.
  • step N05 When it is determined that the object has been identified in the database (step N05), a new artificial intelligence object identification is performed (step N07), and the object identification data is stored. Conversely, if the object is not in the identified database (step N06), the data of the unknown object is stored (step N08). Finally, reclassify and retrain the model through the object identification data and unknown object data (step N09), so as to improve the accuracy of object identification in the future.
  • the object identification system of the present disclosure can store and update the object identification data every time, so that the accuracy is continuously improved to reach the ultimate goal of artificial intelligence, and finally the object identification data and the augmented reality display data are stored in a comprehensive database In order to achieve the effect of data analysis on a single object or multiple objects at the same time.
  • FIG. 5 is a schematic diagram illustrating the combination of an object recognition system and augmented reality technology according to an embodiment of the present disclosure.
  • the object recognition data S001 after object recognition and the augmented reality display data S002 are integrated to generate data about the object recognition result and augmented reality.
  • the data S003 is stored in the database S004 to build a feature model (for example, there are different feature algorithms for different object image features), and a better deep learning model can be retrained.
  • the application of the database S004 can simplify a lot of judgment time and response time to provide more application space for deep learning in the future.
  • Fig. 6 is a schematic diagram illustrating a smart lock application according to an embodiment of the present disclosure.
  • the box body 63 is a container for transporting goods. The shape and material of this container are not restricted. It may be square, round, bag, box, or back bag. All of them are covered by this patent. Only a smart lock is required to protect the safety of the contents. sex.
  • the box 63 is provided with a smart lock 61 (in one embodiment, the smart lock 61 may be a two-dimensional barcode device). The two-dimensional bar code represents the smart number of the smart lock 61.
  • the unlocking information can be obtained by scanning the two-dimensional bar code with the mobile device 60, and the notification is transmitted through the signal 62 (the transmission may be any possible connection method such as wired, wireless network, Bluetooth, NFC, etc.)
  • the present invention does not limit the physical type of unlocking and unlocking, and the actual use condition depends on the box 63, which is within the scope of protection of this patent, and its physical state is not described in detail.
  • Fig. 7 is a flow chart illustrating the operation of locking the smart lock according to an embodiment of the present disclosure.
  • the mobile device scans the two-dimensional bar code of the smart lock (step C001), and sends the scanned two-dimensional code back to the logistics party's server to obtain the locked key (step C002), and then obtain The key is transmitted to the smart lock to make it lock (step C003).
  • the above-mentioned transmission may be any possible connection method such as wired, wireless network, Bluetooth, NFC, etc., which are all within the scope of protection of this patent, and its status is not described in detail.
  • Fig. 8 is a flowchart illustrating the unlocking operation of the smart lock according to an embodiment of the present disclosure.
  • the mobile device When goods or food are delivered to the client, and the client has a mobile device in hand, he can use his mobile device to pick up the goods.
  • the mobile device will display a code scanning screen and the client can scan the QR code on the smart lock Barcode (step C004).
  • the mobile device After the mobile device obtains the two-dimensional barcode, it will be combined with the random password of the order and uploaded to the network server, and the server will issue the key to the mobile device of the customer (step C005).
  • the mobile device transmits the secret key to the smart lock to unlock it (step C006).
  • the above transmission may be any possible connection method such as wired, wireless network, Bluetooth, NFC, etc. Therefore, this unlocking method must be used under the condition that the customer's mobile device has wireless transmission capabilities.
  • FIG. 9 is a flowchart illustrating the unlocking operation of the smart lock according to an embodiment of the disclosure.
  • the delivery staff will carry a mobile device to receive orders and track the progress of the shipment, and the company can also manage it conveniently. Therefore, this method is an inevitable and feasible unlocking method. Click through the mobile device in the delivery person's hand to select the customer arrival option, but this screen will pop up with a graphic sliding password or a numeric password.
  • This password is entered into the logistics system when the client places an order, and only the server This set of passwords is known to the client, so when the delivery person needs to obtain the unlocking key through the mobile device, he must send the set of passwords to the server at the same time (step C007) to retrieve the key from the server (step C008) , To ensure that the smart lock can be opened only after the customer is present or authorized by the customer, and finally the key is sent to the smart lock via the wireless mechanism through the mobile device of the delivery person to unlock (step C009), so as to achieve the entire process from the shipment to the customer All are safe and sound.
  • the boxes for goods transportation and food delivery are very diverse, but they all lack the smart lock mechanism.
  • the present disclosure only needs to redesign the transport box (not limited to various types of boxes or handbags), and then the smart lock can be added to it, so as to solve the doubts about food safety and improve the competitiveness of the logistics industry.
  • Fig. 10 is a flowchart illustrating an artificial intelligence logistics system according to an embodiment of the present disclosure.
  • the object recognition data is obtained in real time.
  • the NFC tag is compared with the recognition result to train a deep learning model and improve the accuracy of object recognition (step 102).
  • the object identification result combined with the augmented reality display method can make the object three-dimensional and real, and display its related shipping data in real time (step 104), which is completely different from the current traditional barcode scanning method on the goods individually. Therefore, the combination of the object identification result of the present disclosure and the augmented reality display method can greatly reduce the frequency and time of scanning goods for delivery personnel.
  • the two-dimensional bar code is used to realize the smart lock and unlock transportation method (steps 106-110), so that the entire logistics system can ensure the safety from the shipment to the consumer.
  • the artificial intelligence object recognition disclosed in this disclosure adopts a marginal computing method and can be directly performed on the fog side.
  • logistics personnel can use mobile devices to perform object recognition using artificial intelligence marginal operations to actively identify the appearance of goods. Because it’s not easy to distinguish between the various advertisements in the packaging when the goods are shipped, the identification through the appearance and size characteristics can be used for preliminary identification.
  • the integration of the NFC tag can further contactless scanning to verify the shipment data, and conversely, the high-precision object identification technology can also reduce it. The number of times the NFC tag is repeatedly sensed.
  • the integration of object identification data and augmented reality display data enables the content of the shipment to be displayed when the goods are taken, so that they can be cross-checked to avoid delivery errors when leaving the warehouse, put in the appropriate delivery box, and the food store or
  • the shipping center scans the two-dimensional bar code to lock the smart lock.
  • the delivery person does not have the authority to unlock it.
  • the smart lock can be opened through the consumer APP by scanning the QR code in front of the consumer, or the consumer can enter the order under it.
  • the code of time can only be used in the delivery person’s app to open the smart lock of the goods with the delivery person’s mobile phone, thus avoiding food safety concerns.
  • FIG. 11 is a result of the object recognition system of the present disclosure combined with augmented reality presented on a mobile device.
  • the present disclosure displays object data through augmented reality, allowing users to see the added images and information in a real state, and combines virtual images with real objects to synthesize images through (Video See-through) The display results.
  • the augmented reality of the present disclosure can display and interact with the user in real time, without special reaction time or post-production time, and can display the information of multiple items in real time.
  • most interactive technology experiences can only be used by a single product or for personal use before they can be replaced by another product to search. However, this situation can be solved by augmented reality, that is, if it is used on a truck or a warehouse.
  • the object data displayed in the augmented reality includes not only text descriptions, but also graphic display methods. Therefore, the freight driver or delivery person can quickly learn the content of the item from the pattern to increase the speed of picking up the goods.
  • the intelligentization of logistics systems is a very important development topic.
  • the characteristics of freight objects are initially classified through object identification. Since the characteristics of each object are unique, a deep learning model that combines image characteristics and radio frequency identification is adopted. , Can make the accuracy of object identification continue to rise, reducing the possibility of human misjudgment.
  • the delivery staff can use the inductive radio frequency identification to further confirm the verification of the object transfer information and record the shipment back to the warehouse management system, and cooperate with the augmented reality to provide real-time delivery details of the goods to achieve the ability to quickly divert shipments.
  • the delivery staff can scan the two-dimensional bar code to lock the logistics goods intelligently.
  • the intelligent lock mechanism can not be opened until the customer signs the receipt to ensure that the entire delivery process is safe and reliable until the customer's hands.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Toxicology (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Economics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Electromagnetism (AREA)
  • Computer Security & Cryptography (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Development Economics (AREA)
  • Biomedical Technology (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种物件辨识系统，包含有：一输入装置，用来取得一物件的一影像数据及一射频识别数据；一处理装置，连接输入装置，用来进行一模型训练程序，其中模型训练程序包含有：根据影像数据，撷取一物件特征；根据物件特征及射频识别数据，产生用来辨识该物件的一分类数据，其中射频识别数据用来验证分类数据的正确性，以产生一深度学习模型；以及一输出装置，连接处理装置，用来根据深度学习模型对物件所产生的分类数据及射频识别数据，产生对应物件的一物件数据。

Description

物件辨识系统及相关装置 技术领域
本揭示是有关于一种物件辨识系统，更具体地，是关于一种用于外送餐饮及物流包裹的物件辨识系统。
背景技术
近几年来外送餐饮与货运的市场兴盛，外送餐饮与物流的包裹方式虽有差异，但过程中尽量的简化与快速到货的目标是一致的。现在产业变化快速，早年店面的经营模式渐渐转入电商模式，让商机可以不受时间与距离而影响，以往消费型态会在特定时段到特定店家的模式，逐渐转变为通过手机应用程序(application，APP)消费的到府交货型态，因应这种型态的转变，物流业者也受到更多竞争，早年的货运型态已经渐渐不符合时代趋势，但无论运送货品或食物的差异有多大，到了送货的终端仍是依靠人力去完成挨家挨户的签收，外送员与货运司机需求始终不断，但因为物流中间过程夹杂人为因素，很可能有送错货品给消费者或食物遭外送员偷吃与污染的疑虑，送错货品必须退回原发货中继站，造成物流成本的增加。此外，食品偷吃、偷开启和污染也很难厘清是在哪个环节出了问题，例如，是否在店家端就已经短少食物，还是外送过程中造成的损耗，因而造成食安上的疑虑。
现有的货运方式是在仓库进行捡货交给外送员扫描出货，外送员照列表上的地址到了消费者家门前，再从小货车或者机车上取下货品交与消费者签名收取，外送食物则是外送员前往店家等待，消费者下单后直接从店家手中领取食物，放入摩托车上铝箔保温袋中，骑到目的地之后取出食物交与消费者签收。
在现有的运送机制上，对于是店家还是外送过程出错的责任厘清很难判断。此外，在没有任何安全或检验机制的外送食物或包裹，如果碰到数张接单的迭单，外送员送货签收可能会造成安全上的漏洞。另外，没有检验确认的签 收机制，如果公司门口同时送好几个消费者，双方签收没看清楚姓名也可能会错拿其他人的包裹。因此，现有的物流方式实有改进的必要。
发明概述
本揭示提供一种物件辨识系统，包含有：一输入装置，用来取得一物件的一影像数据及一射频识别(Radio Frequency Identification，RFID)数据；一处理装置，连接输入装置，用来进行一模型训练程序，其中模型训练程序包含有：根据影像数据，撷取一物件特征；根据物件特征及射频识别，产生用来辨识物件的一分类数据，其中射频识别用来验证分类数据的正确性，以产生一深度学习模型；以及一输出装置，连接处理装置，用来根据深度学习模型对物件所产生的分类数据及射频识别数据，产生对应物件的一物件数据。
本揭示提供一雾端装置，用来进行物件辨识，包含有：一处理单元，用来执行一程序代码；以及一储存单元，连接处理单元，用来储存程序代码；其中程序代码指示处理单元执行以下步骤：取得一物件的一影像数据及一射频识别(Radio Frequency Identification，RFID)数据；进行一模型训练程序，其中模型训练程序包含有：根据影像数据，撷取一物件特征，以及根据物件特征及射频识别，产生用来辨识物件的一分类数据，其中射识别用来验证分类数据的正确性，以产生一深度学习模型；以及根据深度学习模型对物件所产生的分类数据及射频识别数据，产生对应物件的一物件数据。
本揭示提供一行动装置，包含有：一处理单元，用来执行一程序代码；以及一储存单元，连接处理单元，用来储存程序代码；其中程序代码指示处理单元执行以下步骤：取得一物件的一影像数据及一射频识别(Radio Frequency Identification，RFID)；传送影像数据及射频识别至一雾端装置，用来进行物件辨识；从雾端装置接收关于物件的一物件数据及一扩增实境数据；以及根据物件数据及对应物件数据的扩增实境数据，产生一扩增实境影像。
附图说明
当结合附图阅读时，从以下详细叙述中可最好地理解示例性公开的各面 向。各种特征未按比例绘制。为了清楚讨论，可任意增加或减少各种特征的维度。
图1是根据本揭示实施例说明物件辨识系统架构的示意图。
图2是根据本揭示实施例说明物件辨识系统运作的流程图。
图3是根据本揭示实施例说明前端物流出货的流程图。
图4是根据本揭示实施例说明人工智能物件辨识的流程图。
图5是根据本揭示实施例说明物件辨识系统与扩增实境技术结合的示意图。
图6是根据本揭示实施例说明智能锁应用的示意图。
图7是根据本揭示实施例说明智能锁上锁操作的流程图。
图8是根据本揭示实施例说明智能锁解锁操作的流程图。
图9是根据本揭示实施例说明智能锁解锁操作的流程图。
图10是根据本揭示实施例说明人工智能物流系统的流程图。
图11是根据本揭示实施例说明物件辨识系统结合扩增实境显示影像的示意图。
具体实施方式
以下叙述含有与本揭露中的示例性实施方式相关的特定信息。本揭露中的附图和其随附的详细叙述仅为示例性实施方式。然而，本揭露并且不局限于此些示例性实施方式。本领域技术人员将会想到本揭露的其他变化与实施方式。除非另有绘示，附图中相同或对应的组件可由相同或对应的附图标号表示。此外，本揭露中的附图与例示通常不是按比例绘制的，且非旨在对应于实际的相对维度。
出于一致性和易于理解的目的，在示例性附图中藉由标号以标示相同特征(虽在一些示例中并未如此标示)。然而，不同实施方式中的特征在其他方面可能不同，因此不应狭义地局限于附图所示的特征。
图1为物件辨识系统架构的示意图。简单来说，物件辨识系统包含输入 装置、输出装置及处理装置，其中输入装置、输出装置及处理装置分别包含至少一处理单元及储存单元。输入装置10可为行动装置，如手机或者平板等相关接口装置。物件11为货运物品或食物，不一定是方形或者固定型态，只需特征大小颜色包装可提供处理装置12识别使用。详细来说，用户使用输入装置10，如行动装置上的镜头，进行拍摄物件11的影像，并传送至处理装置12进行物件辨识。处理装置12通过分类与模型训练，产生深度学习模型，更能精确判断出物件11为何。值得注意的是，本揭示的处理装置12应用射频识别(Radio Frequency Identification，RFID)技术，如近距离无线通信(Near-field communication，NFC)识别数据(或称为NFC标签)，进行模型训练，以实现人工智能深度学习。举例来说，物流人员将输入装置10靠近物件11，以扫描物件11的射频识别数据(如NFC标签)，并将射频识别数据传送至处理装置12，以供处理装置12进行数据比对。此外，在处理装置12辨识完物件11后，会产生物件数据，并传送至输出装置来显示物件辨识结果。在一实施例中，输出装置与输入装置可为相同或不同装置，并可为行动装置或屏幕等具有显示功能的装置。上述传输可能为有线、无线网络、蓝芽等任何可能的连接方式。
图2为根据本揭示实施例说明物件辨识系统运作的流程图。图2中显示的流程200可由图1所示的处理装置12中的处理单元执行储存单元中的程序代码，其中程序代码用来指示处理单元执行以下步骤：
步骤202：取得物件的影像数据及射频识别数据。
步骤204：进行一模型训练程序，其中模型训练程序包含有：根据影像数据，撷取一物件特征；根据物件特征及射频识别数据，产生用来辨识物件的一分类数据，其中射频识别数据用来验证分类数据的正确性，以产生一深度学习模型。
步骤206：根据深度学习模型对物件所产生的分类数据及射频识别数据，产生对应物件的一物件数据。
根据流程200，处理装置12根据物件的影像数据及物件射频识别数据，来进行深度学习模型训练，进而产生高精确度的物件辨识结果，因此能有效降 低外送员送错商品的情况，并提升外送员捡货的效率。本揭示的处理装置12是应用边际运算在雾端进行数据处理，因此能分散在用户终端进行物件识别，即用户终端不需要将原始的影像数据及射频识别数据回传服务器，服务器也不需要更大的处理效能来处理用户终端的需求，这种方式也可以让使用者没有延迟感，降低网络速度的使用门坎，以及物流业者也可以减少大型服务器运算量的花费。
本揭示的物件辨识系统是通过人工智能数据处理技术，来实现物件辨识。详细来说，本揭示的物件辨识系统通过射频识别数据(如NFC标签)作为人工智能深度学习的对照组,可以有效的提升物件辨识的准确性，并且，长时间不断的训练深度学习模型，可以提升人工智能的精准度，搜集到的影像数据都有很高的大数据价值，提供未来重新训练人工智能更多数据量，因此是可以有效的获得预期的成果，并且解决先天物件辨识困难与物流分类捡货的瓶颈，提高快速出货的效率。
举例来说，图3为前端物流出货的流程图，每一件物件A-Z分别是在物流线上的货品或者在仓储即将出库的货物编号，通过人工智能处理单元310(如图1所示的处理装置12的处理单元)分析数据,以实现物件A-Z的分类。人工智能处理单元310可通过输入装置300取得物件A-Z的影像，并会转存为关键特征影像。人工智能处理单元310包含多个运算单元，详细功能如下。识别运算单元311，通过识别运算可以确认影像是否有特征还是仅仅只是环境特征值，先做出区隔的判断，要是没有物件的特征就不会主动进行接下来的辨识运算。当通过了识别运算确定有需要判断的物件在此影像中，将其数据传送至辨识运算单元312，辨识运算主要通过用人工智能算法对特征影像进行判断，减少误判的可能性，大幅提升系统准确性，之后转交给运算单元313，用来计算出特征值吻合度的机率，物件特征各点代表的特征意义各不相同，因此透过此运算单元313可以有效的计算出机率高低，最后传送到分类运算单元314以产生物件分类数据。同一个物件会进行归类，有办法分类的就进行数据整并，没办法辨识的影像其则分开归档至未知物件中，以便未来人工智能深度学习时新增分 类辨识时使用。大数据是未来的发展趋势，这些已经计算好的特征值与数据都是值得累积与保存的。处理好的物件分类数据/结果会输出至射频识别装置320，如通过行动装置上的射频识别芯片撷取物件本身的射频识别数据，以供人工智能处理单元310进行分类结果的确认/比对，进而训练出高精确度的深度学习模型。接着，通过扩增实境处理系统330产生扩增实境显示数据，使其最终可透过扩增实境方式来显示物件数据(如物件详细数据、货运相关信息、物件名称及物流顺序等)。物件分类数据与射频识别数据传回到数据库340中储存，以待之后的使用与分析。最后将物件数据与扩增实境显示数据(如上述物件详细数据与货运相关信息)传送至输出装置350(如行动装置的屏幕)来显示结果。
简单来说，本揭示通过物件辨识系统去双重确认物件，以降低人为因素的不确定性，藉以提升消费者信心与保障寄送双方权益。人工智能近几年随着类神经网络算法的快速推展，其进化速度可以说日新月异，机器学习是人工智能的很重要一环，可分为训练(Training)与预测(Predict)两大部分，训练可以使机器如同人类学习物体判断，学习过程纠正学习判断的正确性，使得判断准确率持续上升并保持一定的正确性，而预测的部分则是在通过机器训练之后，不须经过辅助与纠正的方式，就可以主动预测物体为何，这也是我们理解中的预测事务。机器学习种类又可分为监督式学习(supervised learning)、非监督式学习(un-supervied learning)与半监督式学习(semi-supervised learning)三种，监督式学习是所有数据都有标准答案，可以让机器学习时判断误差所用，预测时也会比较精准，将数据标注(label)，并且将数据中的特征(feature)取出来，将来预测只需找寻这个特征就可以辨识物体或者判断结果，这种方法比较接近人工分类，对机器较容易但是对人类的前置作业会增加较多预处理负担，非监督学习则是所有资料都没有标准答案，无法提供机器学习判断误差使用，因此机器必须自己找寻答案，往往预测也会较为不准确，但是可以减少数据搜集后的整理负担，但相对的代价就是预测误差较大，适合分类复杂且困难的数据集，半监督式学习则是少数的资料有答案，可以提供机器学习参考，但是大部分依然没有答案，机器必须自己找寻答案，等于是结合上述两种的优点，这 种方式只需少数的人工分类，同时又可以提升部分的预测精准度，是目前较常使用的一种方式。
举例来说，本揭示通过监督式的机器学习，即利用射频识别数据，判断人工智能处理单元310辨识出来的物件分类结果的正确性/误差，进而不断的训练深度学习模型，以提升物件辨识的准确度。
深度学习模型建立后，当物流人员使用本揭示的物件辨识系统来辨识物品时，不会受限于物品摆放的位置，甚至同时不只一件物品可进行辨识与交叉比对数据，因此能有效提升外送与货运的效率。详细来说，本揭示在物件辨识之后，通过扩增实境显示物件数据，能够使物流人员有效的辨识出物件的品项与详细数据，如描述物品本身的详细数据与物品转运相关数据等。
在扩增实境的操作上，货运司机或者外送员可利用手机镜头，取得影像数据，并传送给具有边际运算人工智能的处理装置。处理装置针对每张影像做前置处理，消除影像背景、清除噪声等等，再从影像中撷取并分析出物件的位置相关信息，再利用位置的影像坐标值来进行计算，储存数据内容，累积更多的数据，进一步提升物件判断的准确率。
举例来说，物件分类数据及物件数据，在在归档时，可以与扩增实境数据合并储存。因此，外送人员可以快速处理大量的运输货品或食物，实现从出货出餐到消费者手上的人工智能的物流系统，确保食品安全性与降低物流的错误率。因此，通过本揭示提出的物件辨识系统，在前端的深度学习模型及物流数据数据库正确建立后(如分类结果、影像特征及扩增实境数据等)，外送人员即可透过行动装置辨识出物件，而不需逐一扫标物件的NFC标签，来确认物件。
图4为根据本揭示实施例说明人工智能物件辨识的流程图。如图4所示，本揭示可通过输入装置取得影像数据(步骤N01)，接着传送到处理装置判断是否具有物件特征值(步骤N02)，否则退回到步骤N01继续撷取影像数据。如果有影像特征辨识出来，则处理装置储存影像特征(步骤N03)，以供未来深度学习之用。在储存完影像特征之后，处理装置判断影像特征是否在过往判断过的数据库中(步骤N04)。举例来说，针对不同的影像特征有不同的特征演算 法对应，可以提升物件辨识的正确性。当判断出物件已在辨识过的数据库时(步骤N05)，则进行新的人工智能物件辨识(步骤N07)，并储存物件辨识数据。相反的，如果物件不在已辨识过的数据库中(步骤N06)，则储存未知物件的数据(步骤N08)。最后，通过物件辨识数据及未知物件的数据进行重新分类与重新训练模型(步骤N09)，以便未来提高物件辨识的正确性。
再者，本揭示的物件辨识系统能将每一次的物件辨识数据储存并且更新，使得准确率不断提升达到人工智能的终极目标，最终将物件辨识数据与扩增实境显示数据储存于综合的数据库中，以实现对单一物件或者同时多物件进行数据分析的效果。
图5为根据本揭示实施例说明物件辨识系统与扩增实境技术结合的示意图。如图5所示，为了建立物件辨识与扩增实境的数据库，将物件辨识后的物件辨识数据S001与扩增实境显示数据S002整合，以产生关于物件辨识结果与扩增实境的数据S003，最后将数据S003储存至数据库S004中，用来建立特征模型(如上述针对不同的物件影像特征有不同的特征算法)，并且能重新训练更好的深度学习模型。数据库S004的应用可以简化相当多判断的时间与反应时间提供未来更多深度学习的应用空间。
除此之外，为了保障食物与物品到消费者手上不受污染与安全性，毕竟人力运送都是临时雇员多，往往是货运与外送业最容易出现状况的一环，因此本揭示另提出结合二维条码(如QR code)的技术，以提升人工智能物流系统的安全性。
图6为根据本揭示实施例说明智能锁应用的示意图。箱体63为运送货品的容器，本容器不限制形状与材质，可能为方形、圆形、提袋、盒装、背袋，皆属本专利保障范围，只需提供智能锁保护内容物的安全性。箱体63上设置有智能锁61(在一实施例中，智能锁61可为二维条码的装置)。二维条码代表智能锁61的智能编号，使用行动装置60扫描二维条码可以得到解锁的信息，通过讯号62(传输可能为有线、无线网络、蓝芽、NFC等任何可能的连接方式)传输通知智能锁61开锁，本发明不限制开锁解锁的物理型态，实际使用状况要视 箱体63而改变，皆在本专利保障范围内，不详述其实体状态。
图7为根据本揭示实施例说明智能锁上锁操作的流程图。店家或者货运公司出货时，尤其是食物必须有安全保障，因此需通过店家或货运公司的行动装置进行上锁，而不能通过外送货物员进行上锁。如图7所示，行动装置扫描智能锁的二维条码(步骤C001)，并将扫描到的二维回传至物流方的服务器，以取得上锁的密钥(步骤C002)，然后将取得的密钥传输至智能锁使其进行上锁的动作(步骤C003)。上述传输可能为有线、无线网络、蓝芽、NFC等任何可能的连接方式，皆在本专利保障范围内，不详述其状态。
图8为根据本揭示实施例说明智能锁解锁操作的流程图。当货品或者食物外送至客户端，而客户有行动装置在手上时，可以使用自身行动装置领取货物的选项，此时，行动装置会出现扫码画面，客户可扫描智能锁上的二维条码(步骤C004)，行动装置在取得二维条码之后，会与自身订单的随机密码组合上传至网络服务器，由服务器发出密钥给客户的行动装置(步骤C005)。接着，行动装置传输密钥至智能锁使其解锁(步骤C006)。上述传输可能为有线、无线网络、蓝芽、NFC等任何可能的连接方式，因此此解锁方法必须在客户的行动装置具有无线传输能力的条件之下使用。
另一方面，如果客户端的行动装置没有无线传输能力，或者下楼没带行动装置无法再度回到高楼层取得，本揭示另提供智能锁解锁的备用方案，以避免到货无法开启的窘境。图9为本揭示实施例说明智能锁解锁操作的流程图。通常外送货物员会携带行动装置，才能够接单与追踪出货进度，公司也才能够方便管理，因此这个方式是必然可行的解锁方式。通过外送员手中的行动装置进行点选，选择客户到货选项，但是这个画面会跳出图形滑动密码或者数字密码，这个密码是客户端在下单时，就已经输入进入物流系统中，也只有服务器与客户端已知这组密码，因此当外送员需要通过行动装置取得解锁密钥时，必须与这组密码同时发送至服务器(步骤C007)，才能从服务器中取回密钥(步骤C008)，以确保客户在场或客户授权之后方能开启智能锁，最后通过外送员的行动装置将密钥通过无线机制发送至智能锁上解锁(步骤C009)，以达到整个流程 自出货到客户手中皆安全无虞。
目前货品运输与外送食物的箱子非常多元，但是都欠缺智能锁的机制。本揭示只需重新设计运输的箱体(不局限于箱子或者手提袋等各类型态)，就可以将智能锁增加至其中，解决食品安全的疑虑并提升物流业的竞争性。
图10是根据本揭示实施例说明人工智能物流系统的流程图。如图10所示，通过人工智能物件辨识(步骤100)，实时的得出物件辨识数据。进一步地，通过NFC标签与辨识结果比对，以训练出深度学习模型，并提升物件辨识的正确性(步骤102)。物件辨识结果搭配扩增实境显示方式，可以让物件立体和真实化，实时显示其相关运送数据(步骤104)，这与目前传统需个别扫描货品上的条码方式有着截然的不同。因此，本揭示物件辨识结果搭配扩增实境显示方式，能大大减少外送货物员的扫描货品次数及时间。最后，通过二维条码来实现智能锁上锁与解锁的运送方式(步骤106-110)，让整套物流系统可以确保从出货到消费者手中的安全性。
现有的人工智能物件识别技术大都需要回传至云端后台运算，本揭示的人工智能物件辨识是采取边际运算的方式，可以在雾端直接进行。物流人员在物流前端阶段，可通过行动装置进行人工智能边际运算的物件辨识，来主动辨识货品外观。因为货品运送时包装五花八门广告不容易区别，通过外表大小特征进行辨识可以进行初步的辨认，整合NFC标签可以进一步无接触扫描验证货品运送数据，反过来说通过高精确度的物件辨识技术也可以减少重复感应NFC标签的次数。此外，整合物件辨识数据与扩增实境显示数据，能使拍摄货品时显示货运内容，这样在出仓库时可以交叉确认避免配送错误发生，放入合适的外送盒并，且由食物店家或者出货中心扫描二维条码以上锁智能锁。外送员本身没有解开的权限，除非联系服务中心在特殊情况才能授权开启，否则要到消费者面前才可以通过消费者APP扫描二维条码开启智能锁，或者是消费者输入其下订货品时之密码于外送员APP中方可使用外送员的手机开启货品智能锁，因此能避免食安疑虑。
图11为本揭示物件辨识系统结合扩增实境在行动装置上呈现的结果。 本揭示通过扩增实境显示物件数据，可令使用者在现实的状态下可以看见所增加的影像及信息，将虚拟的影像与现实的物体结合的影像合成穿透式(Video See-through)的显示结果。本揭示的扩增实境可以实时与用户显示互动，不需特别的反应时间或者是后制时间，并且可实时地显示多物品的信息。目前多数的互动科技体验，通常都只能够让单一货品或者个人使用，才能够再换下一样货品搜寻，然而这样的情况可通过扩增实境来解决，也就是如果使用在货车上或者仓库要出库，可以同时扫描与辨识大量物品显示的成果，并非仅仅单一件物件的显示信息而已，因此能快速帮助货运司机或者外送员找到所要交货的物品，而且不会有人工捡货错误的问题产生。如图11所示，扩增实境的显示的物件数据除了文字描述，更包含有图型显示方式。因此，货运司机或者外送员能快速从图型得知物品内容，以提升捡货速度。
综上所述，物流系统的智能化是一个很重要的发展课题，通过物件辨识初步分类货运物件本身的特征，由于每个物件的特征是独特的，通过结合影像特征与射频识别的深度学习模型，可以让物件辨识的正确性持续上升，减少人为误判的可能性。在运送期间，外送员可通过感应射频识别进一步确认物件转运信息验证并且记录回仓管系统已出货，并配合扩增实境实时的提供货品运送详细信息，达到快速分流出货的能力。此外，外送员可扫描二维条码将物流商品智能锁上锁，直到客户签收时方能开启的智能锁的机制，以确保整个运送流程直到顾客手中皆是安全可靠的。
根据以上描述，在不脱离这些概念范围的情况下，可使用多种技术来实施本申请中叙述的概念。此外，虽然已经具体参考某些实施方式叙述了这些概念，但本领域具有通常知识者将认识到在不脱离这些概念范围的情况下可在形式和细节上进行改变。如此一来，所述的实施方式在各方面都将被视为是绘示性而非限制性的。并且，应理解本申请并且不限于上述的特定实施方式，且在不脱离本揭露范围的情况下，对此些实施方式进行诸多重新安排、修改和替换是可能的。

Claims (17)

  1. 一种物件辨识系统，包含有：
    一输入装置，用来取得一物件的一影像数据及一射频识别(Radio Frequency Identification，RFID)数据；
    一处理装置，连接所述输入装置，用来进行一模型训练程序，其中所述模型训练程序包含有：根据所述影像数据，撷取一物件特征；根据所述物件特征及所述射频识别数据，产生用来辨识所述物件的一分类数据，其中所述射频识别数据用来验证所述分类数据的正确性，以产生一深度学习模型；以及
    一输出装置，连接所述处理装置，用来根据所述深度学习模型对所述物件所产生的所述分类数据及所述射频识别数据，产生对应所述物件的一物件数据。
  2. 如权利要求1所述的物件辨识系统，其中所述输出装置更用来通过一扩增实境方式显示所述物件数据。
  3. 如权利要求2所述的物件辨识系统，更包含有一储存装置，连接所述输出装置，用来储存一扩增实境数据、所述物件数据、所述影像数据、所述射频识别数据、所述分类数据及所述物件特征的至少其中之一。
  4. 如权利要求3所述的物件辨识系统，其中所述处理装置包含有一云端服务器及一雾端服务器的至少其中之一，以及所述储存装置包含有一云端数据库及一雾端数据库的至少其中之一。
  5. 如权利要求3所述的物件辨识系统，其中所述处理装置更用来判断所述储存装置是否包含所述物件的所述物件数据、当判断所述储存装置不包含所述物件的所述物件数据时，进行所述模型训练程序，以及当判断所述储存装置包含所述物件的所述物件数据时，从所述储存装置取得所述物件数据。
  6. 如权利要求1所述的物件辨识系统，其中所述物件数据包含有一物件品称、运送数据及物流顺序的至少其中之一。
  7. 如权利要求1所述的物件辨识系统，其中所述输入装置更用来取得所述物件上的一二维条码、所述处理装置更用来根据所述二维条码，产生用来将所述 物件上的一智能锁上锁或解锁的一密钥，以及所述输出装置更用来接收所述密钥，用来加锁或解锁所述智能锁。
  8. 如权利要求1所述的物件辨识系统，其中所述输入装置更用来取得所述物件上的一二维条码及产生一随机密码、所述处理装置更用来根据所述随机密码及所述二维条码，产生用来将所述物件上的一智能锁上锁或解锁的一密钥，以及所述输出装置更用来接收所述密钥，用来加锁或解锁所述智能锁。
  9. 如权利要求1所述的物件辨识系统，其中所述射频识别数据包含有一近距离无线通信(Near-field communication，NFC)识别数据。
  10. 一雾端装置，用来进行物件辨识，包含有：
    一处理单元，用来执行一程序代码；以及
    一储存单元，连接所述处理单元，用来储存所述程序代码；
    其中所述程序代码指示所述处理单元执行以下步骤：
    取得一物件的一影像数据及一射频识别(Radio Frequency Identification，RFID)数据；
    进行一模型训练程序，其中所述模型训练程序包含有：根据所述影像数据，撷取一物件特征，以及根据所述物件特征及所述射频识别数据，产生用来辨识所述物件的一分类数据，其中所述射识别数据用来验证所述分类数据的正确性，以产生一深度学习模型；以及
    根据所述深度学习模型对所述物件所产生的所述分类数据及所述射频识别数据，产生对应所述物件的一物件数据。
  11. 如权利要求10所述的雾端装置，其中所述储存单元更用来储存对应所述物件数据的一扩增实境数据，以及所述程序代码更指示所述处理单元执行以下步骤：
    根据所述物件数据及对应所述物件数据的扩增实境数据，产生一扩增实境影像。
  12. 如权利要求10所述的雾端装置，其中所述储存单元更用来储存所述物件 数据，以及所述程序代码更指示所述处理单元执行以下步骤：
    判断所述储存单元是否包含所述物件的所述物件数据；
    当判断所述储存单元不包含所述物件的所述物件数据时，进行所述模型训练程序；以及
    当判断所述储存单元包含所述物件的所述物件数据时，从所述储存单元取得所述物件数据。
  13. 如权利要求10所述的雾端装置，其中所述程序代码更指示所述处理单元执行以下步骤：
    从一行动装置接收关于所述物件的一二维条码；
    根据所述二维条码，产生一密钥，其中所述密钥用来上锁或解锁所述物件上的一智能锁；以及
    传送所述密钥至所述行动装置。
  14. 如权利要求10所述的雾端装置，其中所述程序代码更指示所述处理单元执行以下步骤：
    从一行动装置接收关于所述物件的一二维条码及一随机密码；
    根据所述二维条码及所述随机密码，产生一密钥，其中所述密钥用来上锁或解锁所述物件上的一智能锁；以及
    传送所述密钥至所述行动装置。
  15. 一行动装置，包含有：
    一处理单元，用来执行一程序代码；以及
    一储存单元，连接所述处理单元，用来储存所述程序代码；
    其中所述程序代码指示所述处理单元执行以下步骤：
    取得一物件的一影像数据及一射频识别(Radio Frequency Identification，RFID)数据；
    传送所述影像数据及所述射频识别数据至一雾端装置，用来进行物件辨识；
    从所述雾端装置接收关于所述物件的一物件数据及一扩增实境数据；以及
    根据所述物件数据及对应所述物件数据的所述扩增实境数据，产生一扩增实境影像。
  16. 如权利要求15所述的行动装置，其中所述程序代码更指示所述处理单元执行以下步骤：
    取得关于所述物件的一二维条码；
    传送所述二维条码至所述雾端装置；
    从所述雾端装置接收用来上锁或解锁所述物件上的一智能锁的一密钥；以及
    传送所述密钥至所述智能锁，以上锁或解锁所述物件。
  17. 如权利要求15所述的行动装置，其中所述程序代码更指示所述处理单元执行以下步骤：
    取得关于所述物件的一二维条码及一随机密码；
    传送所述二维条码及所述随机密码至所述雾端装置；
    从所述雾端装置接收用来上锁或解锁所述物件上的一智能锁的一密钥；以及
    传送所述密钥至所述智能锁，以上锁或解锁所述物件。
PCT/CN2020/090096 2020-05-13 2020-05-13 物件辨识系统及相关装置 WO2021226893A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US17/266,126 US11429800B2 (en) 2020-05-13 2020-05-13 Object recognition system and related device
PCT/CN2020/090096 WO2021226893A1 (zh) 2020-05-13 2020-05-13 物件辨识系统及相关装置
CN202080000831.2A CN113939832A (zh) 2020-05-13 2020-05-13 物件辨识系统及相关装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/090096 WO2021226893A1 (zh) 2020-05-13 2020-05-13 物件辨识系统及相关装置

Publications (1)

Publication Number Publication Date
WO2021226893A1 true WO2021226893A1 (zh) 2021-11-18

Family

ID=78526171

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/090096 WO2021226893A1 (zh) 2020-05-13 2020-05-13 物件辨识系统及相关装置

Country Status (3)

Country Link
US (1) US11429800B2 (zh)
CN (1) CN113939832A (zh)
WO (1) WO2021226893A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019127618A1 (zh) * 2017-12-25 2019-07-04 图灵通诺（北京）科技有限公司 结算方法、装置和系统
CN109919276A (zh) * 2019-01-13 2019-06-21 湖南省农业信息与工程研究所 一种基于产品表面纹理图像特征的防伪方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107341538A (zh) * 2017-06-30 2017-11-10 上海上大海润信息系统有限公司 一种基于视觉的统计数量方法
CN108320404A (zh) * 2017-09-27 2018-07-24 缤果可为（北京）科技有限公司 基于神经网络的商品识别方法、装置、自助收银台
CN109299748A (zh) * 2018-10-26 2019-02-01 苏州浪潮智能软件有限公司 一种基于本地及云服务的物体检测识别方法
CN109615292A (zh) * 2018-11-30 2019-04-12 周强 外卖配送系统及配送方法
CN109635705A (zh) * 2018-12-05 2019-04-16 上海交通大学 一种基于二维码和深度学习的商品识别方法及装置
US20190332854A1 (en) * 2018-04-25 2019-10-31 Shutterfly, Inc. Hybrid deep learning method for recognizing facial expressions

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11625761B2 (en) * 2007-09-06 2023-04-11 Mohammad A. Mazed System and method for machine learning and augmented reality based user application

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107341538A (zh) * 2017-06-30 2017-11-10 上海上大海润信息系统有限公司 一种基于视觉的统计数量方法
CN108320404A (zh) * 2017-09-27 2018-07-24 缤果可为（北京）科技有限公司 基于神经网络的商品识别方法、装置、自助收银台
US20190332854A1 (en) * 2018-04-25 2019-10-31 Shutterfly, Inc. Hybrid deep learning method for recognizing facial expressions
CN109299748A (zh) * 2018-10-26 2019-02-01 苏州浪潮智能软件有限公司 一种基于本地及云服务的物体检测识别方法
CN109615292A (zh) * 2018-11-30 2019-04-12 周强 外卖配送系统及配送方法
CN109635705A (zh) * 2018-12-05 2019-04-16 上海交通大学 一种基于二维码和深度学习的商品识别方法及装置

Also Published As

Publication number Publication date
CN113939832A (zh) 2022-01-14
US20220114352A1 (en) 2022-04-14
US11429800B2 (en) 2022-08-30

Similar Documents

Publication Publication Date Title
CN104885110B (zh) 用于物品的递送的系统和方法
CN109829677B (zh) 一种物流运输过程监控方法及系统
US20120051643A1 (en) Method and system for capturing and inventoring railcar identification numbers
WO2021226893A1 (zh) 物件辨识系统及相关装置
WO2020160405A1 (en) Tamper-resistant item transport systems and methods
M’hand et al. A scalable real-time tracking and monitoring architecture for logistics and transport in RoRo terminals
CN110555356A (zh) 自助结帐系统、方法与装置
CN109081028A (zh) 一种基于机器人的物品输送方法和系统
US20210271704A1 (en) System and Method for Identifying Objects in a Composite Object
CN105184538A (zh) 一种货物到达时间预估的物流配送方法及系统
CN103996102A (zh) 一种集装箱箱体查验方法
CN106779565A (zh) 基于智能监控的物流仓储信息管理系统
CN113495979A (zh) 唯一的对象面id
CN107358279B (zh) 一种基于rfid的快递收发系统及收发方法
US12093880B2 (en) Edge computing device and system for vehicle, container, railcar, trailer, and driver verification
US20230004933A1 (en) Method and system for autonomous authentication
CN107704907A (zh) 一种基于识别码的快递运输方法、系统及快递包装
KR102429288B1 (ko) 택배물의 삼차원 이미지 분석방법 및 이를 이용한 택배물의 도난방지방법
Shahin et al. Robotics multi-modal recognition system via computer-based vision
KR20220021970A (ko) 무인택배 관리시스템
CN108376326A (zh) 物联网监管平台
WO2023028507A1 (en) System for asset tracking
CN111260291B (zh) 一种基于区块链技术的无人物流配送系统及配送方法
KR20050062727A (ko) Ｒｆｉｄ를 이용한 수하물 관리시스템
WO2024044174A1 (en) System and method for loading a container

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20935293

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20935293

Country of ref document: EP

Kind code of ref document: A1