Shape data analysis operation system based on deep learning technology

Abstract

The invention relates to the technical field of operation analysis, and discloses a shape data analysis operation system based on a deep learning technology. The body data analysis operation system comprises a body data acquisition and processing system, a deep learning and image data processing system, an image data quantification and precise search system based on deep learning, and a wearing and building recommendation system based on machine learning and data analysis; through deep learning and image processing technologies, a clothes-valley science and technology type body data operation platform is built, and the problem of inconvenience in clothes purchase in the prior art is solved; intelligent fitting is realized through technologies of optical scanning, deep learning, image recognition processing and the like, a one-stop platform for clothes selection, fitting, clothes purchasing, matching and recommendation is provided for a user, enough user body data is acquired in a later project and is used as a comprehensive operator of the body data, and data mining and decision support are provided for industries of public security, aviation, insurance, medical treatment and the like.

Description

Technical Field

[0001] The present invention relates to the field of operation analysis technology, and specifically to a body data analysis operation system based on deep learning technology. Background Art

[0002] Today, with the widespread application of intelligent technology, the existing clothing shopping methods still have the following problems: When shopping in physical stores: the store's clothing styles are limited; waiting in line and changing clothes are cumbersome and time-consuming; people don't know what clothes look best on them, nor do they know how to match the clothes they already have; when shopping online: the size is not suitable, and returns and exchanges are troublesome; the quality is uneven.

[0003] In view of this, we proposed a body data analysis and operation system based on deep learning technology. Summary of the invention

[0004] The purpose of the present invention is to provide a body data analysis and operation system based on deep learning technology to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a body data analysis and operation system based on deep learning technology, the body data analysis and operation system comprising: Body data collection and processing system: users first collect their own body data in our smart fitting rooms in shopping malls, communities, schools, etc. The deep learning and image data processing system realizes algorithm-automated image clipping and image retouching through facet matching of the human body white light real scene model and the clothing white light real scene model. Based on deep learning, the image data quantification and precise search system, and the classification retrieval algorithm based on deep convolutional neural network, we established a B_Data Clothing image database of different clothing styles with tens of millions of images, trained the deep convolutional neural network model, and verified it on the database test set. We extracted the convolution features of all clothing data and constructed a hash index using the iterative quantization method; The outfit recommendation system based on machine learning and data analysis includes: a general fuzzy recommendation system based on machine learning algorithms and a personalized and precise recommendation system based on outfit experts.

[0006] Optionally, the body data collection and processing system includes an AI fitting room system. The user opens the access control system of the AI ​​fitting room by scanning the QR code on the display screen outside the AI ​​fitting room. After entering the room, the voice prompt system reminds the user to perform relevant operations. The collection device starts collecting human body data when standing in the scanning area. The user can complete the whole process of trying on, changing, selecting and purchasing clothes on the mobile phone.

[0007] Optionally, the acquisition device is surrounded by 24 cameras, which create a body data for the user every 15° and automatically stitch a 360° full-scale body data set for the user. After 2-3 seconds, the scanning is completed and the system will send the collected body model and data to the client on the user's mobile phone.

[0008] Optionally, through the deep learning and image data processing system, when the user is scanned in the AI ​​fitting room, the camera in the fitting room will capture images of the user in a fixed position, fixed light source, fixed posture, and fixed angle every 15°. A total of 24 images are collected, and the 24 images will be transmitted to the cloud. Through deep learning, the N-Best human body recognition model is used to train the model with a deep confidence neural network, which can detect various body parts and key points in any posture.

[0009] Optionally, through the deep learning-based image data quantification and precise search system, a user uploads a test image, first extracts its CNN features, and maps it and the CNN features of the image in the database to the Hamming space, calculates the Hamming distance, and finally obtains the query image after retrieval in the database.

[0010] Optionally, the background of the general fuzzy recommendation system based on machine learning algorithm labels each piece of clothing differently according to its characteristics.

[0011] Optionally, the characteristics include: style, suitable for occasion, suitable for skin color, suitable for body shape, and suitable for weather.

[0012] Optionally, the personalized and precise recommendation system based on dressing experts will recruit a large number of third-party cooperation teams such as designers, clothing stylists, and clothing consultants according to the user volume in the later stage, and provide customized clothing matching design and plan recommendations for each user on the platform in a professional capacity.

[0013] Compared with the prior art, the present invention provides a body data analysis and operation system based on deep learning technology, which has the following beneficial effects: This body data analysis and operation system based on deep learning technology builds the Yigu Technology body data operation platform through deep learning and image processing technology, which solves the inconvenience of buying clothes in existing technologies; through optical scanning, deep learning, image recognition processing and other technologies, it realizes intelligent fitting, and provides users with a one-stop platform for selecting, fitting, buying and matching clothes. In the later stage of the project, enough user body data will be collected, and it will serve as a comprehensive operator of body data to conduct data mining and decision-making support for public security, aviation, insurance, medical and other industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the process structure of the present invention; Figure 2 This is a schematic diagram of the human body parameter measurement process of the present invention; Figure 3 This is a schematic diagram of the process of using Open CV to perform human body segmentation on human body pictures collected in the AI ​​fitting room of the present invention; Figure 4 This is a schematic diagram of the human body segmentation results using Open CV in the present invention; Figure 5 This is a schematic diagram of the result of the feature point recognition of the obtained human body contour using the model reference method in the present invention; Figure 6 This is a schematic diagram of the search results of men's clothing of the present invention; Figure 7 This is a schematic diagram of the SSDH network framework of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] like Figure 1-Figure 7 As shown, the present invention provides a technical solution: a body data analysis and operation system based on deep learning technology, the body data analysis and operation system comprising: a body data acquisition and processing system, a deep learning and image data processing system, an image data quantification and precise search system based on deep learning, and a dressing recommendation system based on machine learning and data analysis.

[0017] Body data collection and processing system, users first collect their own body data in our smart fitting rooms in shopping malls, communities, schools, etc.; the body data collection and processing system includes an AI fitting room system. Users open the access control system of the AI ​​fitting room by scanning the QR code on the display screen outside the AI ​​fitting room. After entering the room, the voice prompt system reminds the user to perform relevant operations, and the collection device starts collecting human body data when standing in the scanning area.

[0018] The collection device consists of 24 cameras surrounding a circle, which creates a body data for the user every 15° and automatically stitches a 360° full-scale body data set for the user. After 2-3 seconds, the system will send the collected body model and data to the client on the user's mobile phone. The user can complete the whole process of trying on clothes, changing clothes, selecting clothes, and buying clothes on the mobile phone; the human body parameter measurement method based on photos is adopted. The specific process is as follows Figure 2As shown in the figure, the human body images collected in the AI ​​fitting room are segmented using Open CV, as shown in the figure. Figure 3 The processing effect is shown as Figure 4 As shown; The background noise is removed by using the corrosion cvEwde() method and the expansion cvDilate() method. Before the corrosion and expansion, all the contours are painted white to prevent the small black holes near the edge of the human body from being removed as background noise and affecting the smoothness of the edge. The contour with the largest area in the image, that is, the human body / clothing part, is retained, and its rectangular spatial position is calibrated and recorded; the model reference method is used to identify the feature points of the obtained human body contour. The key feature points are as follows: Figure 5 shown.

[0019] The deep learning and image data processing system realizes the algorithm's automatic cutout and retouching processing by matching the facets of the human body white light real-scene model and the clothing white light real-scene model. When the user is scanned in the AI ​​fitting room, the camera in the fitting room will capture images of the user at a fixed position, fixed light source, fixed posture, and fixed angle every 15°. A total of 24 images are collected, and the 24 images will be transmitted to the cloud. Through deep learning, the N-Best human body recognition model is used to train the model with a deep confidence neural network, which can detect various body parts and key points in any posture. For situations with high real-time requirements, the SVM model is used to determine several common actions people take when trying on clothes. In the initial post-rule re-judgment, after detecting abnormal models, the key point positions are manually corrected to ensure the accuracy of the extracted data; By implementing the model of the established clothing library, the key points of the human body model are in one-to-one correspondence with the key points of the clothing model. The clothing model has already been established, which we call the "two-way modeling, one-to-one matching" implementation framework, and then through the key point matching technology in the image processing algorithm. While the clothing is fitted to the human body, virtual stitching is performed, and then based on collision detection technology, the real appearance of the virtual clothing worn on the human body model is realized, such as wrinkles, draping and other effects. In this way, a relatively realistic fitting effect is simulated; Solution to the over-stretching problem: There are multiple fixed force points between the human body and the clothes. In some areas where the force is concentrated, obvious over-stretching often occurs. Near the hanging point, the cloth has obvious bending deformation due to gravity. The hanging point bears all the weight of the cloth, and the force is very concentrated. During the graphics processing process, the local collision between the clothes and the human body will adjust the clothes particles with a larger supporting force to avoid penetration. We sort the clothes particles according to the force size and adjust the over-stretching of the clothes particles with larger force to a limited extent; Image data quantization and precise search system based on deep learning, classification retrieval algorithm based on deep convolutional neural network. Establish a B_Data Clothing image database of different clothing styles with tens of millions of images, train a deep convolutional neural network model, and verify it on the database test set. Extract the convolution features of all clothing data, and use the iterative quantization method to build a hash index.

[0020] When a user uploads a test image, its CNN features are first extracted and mapped to the Hamming space with the CNN features of the images in the database. The Hamming distance is calculated and finally the query image is retrieved from the database.

[0021] When the system receives a test image, it calculates the Hamming distance between the input test sample and each sample in the same class using a loop-up table after obtaining the hash code of the test image, sorts and returns the K nearest neighbor samples. Since the Hamming distance can be calculated directly by XOR, such binary operations are very fast, so the entire process takes less than 0.5 seconds on average; the design of the image processing module is as follows Figure 6 As shown in the figure, it mainly includes using a deep convolutional neural network model to generate hash codes, calculating the Hamming distance sorting, re-sorting the sorted results by color, putting pictures with high similarity in the front, and finally outputting the server address of similar pictures; The retrieval model of the web-based clothing retrieval system adopts the supervised semantics-preserving deep hashing (SSDH) model framework. The network framework of this model is as follows: Figure 7 As shown in the figure, it is similar to the deep convolutional neural network classification model based on clothing attributes, except that a new hidden layer is added after the seventh layer. The purpose of the new hidden layer is to binary encode the features. The value output by the function is either 1 or 0, thereby generating a binary hash code for the number of bits of the image. The B_Data Clothing database is still used to train the model, and images are retrieved from its database. After the hash code of the test image is completed, the Hamming distance between it and the hash code of the image in the database is calculated, and the top 100 similar image results are retrieved. Then, using the color feature information, the designed system re-sorts the retrieval results by color features, and finally generates the latest sorting results; The server image processing module will finally output the addresses of the most similar images in the server in order. The network end will read the search results, and the web page and APP will display the final results.

[0022] A dressing recommendation system based on machine learning and data analysis, wherein the dressing recommendation system based on machine learning and data analysis comprises: a general fuzzy recommendation system based on a machine learning algorithm and a personalized and precise recommendation system based on dressing experts.

[0023] Based on the general fuzzy recommendation system of machine learning algorithm, the system background labels each piece of clothing according to its characteristics, including style, suitable occasion, suitable skin color, suitable body shape, and suitable weather. The main labeling methods are: a. User UGC is generated by users. Each piece of clothing on the app has a label writing function. Users and the original manufacturers of the clothing can write labels according to their preferences and actual experience, and the background will automatically record and enter the labels into the database.

[0024] b. Machine learning based on basic learning rules in the background. The most basic label correspondence rules are set for different sizes, colors, and styles. The machine learning algorithm automatically identifies the characteristics of each piece of clothing and attaches corresponding labels.

[0025] c. Extract keywords from user comments and descriptions of the corresponding pages of the same items online. For each piece of clothing, users can freely post their own comments below. The system will extract tag keywords related to outfit recommendations from these comments and automatically generate tags for this piece of clothing; at the same time, for some clothing styles from e-commerce channels, the system will automatically extract keywords related to outfit recommendations from the details page of the modified clothing on the e-commerce platform and generate tags for the modified clothing.

[0026] d. Manual tagging: The company will arrange a product operation team to continuously tag the clothes in the database according to the trained tagging rules.

[0027] After tagging, whenever a user starts searching or raises their clothing needs, including body shape, skin color, face shape, weather on the day, and occasions, the system will quickly search for tags that match these needs and show the user the clothes with the highest matching degree and the highest comprehensive score. This service is mainly based on machine learning and is free for every user of the platform.

[0028] Based on the personalized and precise recommendation system of fashion experts, the platform will recruit a large number of third-party cooperation teams such as designers, clothing stylists, clothing consultants, etc. according to the user volume in the later stage. In the platform, they can provide customized clothing matching design and plan recommendations for each user in a professional capacity.

[0029] Each designer or fashion expert recommends outfits to users for a fee. Users can choose the appropriate designer or fashion expert according to their needs.

[0030] Ultimately, through the fuzzy recommendations of intelligent algorithms and the precise design of dressing experts, a one-stop platform for trying on, selecting and buying clothes is created for every user, becoming a dressing assistant that understands users better than they do.

[0031] At the same time, based on our core technology, we have also made the following three derivative designs to enhance user experience: 1. Users can ask their relatives and friends to help choose clothes through internal authorization of the APP.

[0032] Imagine a scenario where I want to buy clothes for my parents, friends or girlfriend without Yigu Technology, but whether the size of the clothes is suitable or whether the clothes look good can only be imagined. However, in the scenario of Yigu Technology, we can authorize the body model and other accounts to push each other to help others dress or choose clothes.

[0033] 2. After users have selected several pieces of clothing they like, they can share them on WeChat Moments or Weibo and let their friends vote for the best-looking one.

[0034] It turns the process of selecting and purchasing clothes into an opportunity to interact and communicate with friends, and at the same time increases the effective way for users to socialize through clothes and find circles with similar tastes and preferences as themselves.

[0035] 3. Cooperate with third-party logistics, users place orders online, and logistics delivers to their homes Yigu Technology adopts the principle of nearest delivery. When a user places an order, the backend will find the nearest physical store with inventory of the clothes the user needs based on the user's delivery address. If a corresponding store in the same city is found, the backend will automatically send a notification to the store's salesperson to pack the corresponding clothes. At the same time, the system automatically places an order on a third-party logistics platform, and the delivery staff will pick up the goods and deliver them to the customer's home. If there is no such store, it will directly connect to the manufacturer's inventory and the original manufacturer will ship the goods as soon as possible.

[0036] As an application of this embodiment: This body data analysis and operation system based on deep learning technology builds the Yigu Technology body data operation platform through deep learning and image processing technology, which solves the inconvenience of buying clothes in existing technologies; through optical scanning, deep learning, image recognition processing and other technologies, it realizes intelligent fitting, and provides users with a one-stop platform for selecting, fitting, buying and matching clothes. In the later stage of the project, enough user body data will be collected, and it will serve as a comprehensive operator of body data to conduct data mining and decision-making support for public security, aviation, insurance, medical and other industries.

[0037] The above generally describes the present invention in detail, but it is obvious to a person skilled in the art that some modifications or improvements can be made to the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.

Claims

1. A body data analysis and operation system based on deep learning technology, characterized by: The body data analysis operation system comprises: Body data collection and processing system: users first collect their own body data in our smart fitting rooms in shopping malls, communities, schools, etc. The deep learning and image data processing system realizes algorithm-automated image clipping and image retouching through facet matching of the human body white light real scene model and the clothing white light real scene model. Based on deep learning, the image data quantification and precise search system, and the classification retrieval algorithm based on deep convolutional neural network, we established a B_Data Clothing image database of different clothing styles with tens of millions of images, trained the deep convolutional neural network model, and verified it on the database test set. We extracted the convolution features of all clothing data and constructed a hash index using the iterative quantization method; The outfit recommendation system based on machine learning and data analysis includes: a general fuzzy recommendation system based on machine learning algorithms and a personalized and precise recommendation system based on outfit experts.

2. The body data analysis and operation system based on deep learning technology according to claim 1, characterized in that: The body data collection and processing system includes an AI fitting room system. The user opens the access control system of the AI ​​fitting room by scanning the QR code on the display screen outside the AI ​​fitting room. After entering the room, the voice prompt system reminds the user to perform relevant operations. When standing in the scanning area, the collection device starts to collect human body data. The user can realize the whole process of trying on, changing, selecting and purchasing clothes on the mobile phone.

3. The body data analysis and operation system based on deep learning technology according to claim 2, characterized in that: The acquisition device is surrounded by 24 cameras in a circle, which creates a body data for the user every 15° and automatically stitches a 360° full-scale body data set for the user. After 2-3 seconds, the scanning is completed and the system will send the collected body model and data to the client on the user's mobile phone.

4. The body data analysis and operation system based on deep learning technology according to claim 1, characterized in that: Through the deep learning and image data processing system, when the user is scanned in the AI ​​fitting room, the camera in the fitting room will capture images of the user in a fixed position, fixed light source, fixed posture, and fixed angle every 15°. A total of 24 images are collected, and the 24 images will be transmitted to the cloud. Through deep learning, the N-Best human body recognition model is used to train the model with a deep confidence neural network, which can detect various body parts and key points in any posture.

5. The body data analysis and operation system based on deep learning technology according to claim 1, characterized in that: Through the deep learning-based image data quantification and precise search system, a user uploads a test image, first extracts its CNN features, and maps it and the CNN features of the images in the database to the Hamming space, calculates the Hamming distance, and finally obtains the query image after retrieval in the database.

6. The body data analysis and operation system based on deep learning technology according to claim 1, characterized in that: The background of the general fuzzy recommendation system based on machine learning algorithm labels each piece of clothing differently according to its characteristics.

7. The body data analysis and operation system based on deep learning technology according to claim 6, characterized in that: Said characteristics include: style, suitable for the occasion, suitable for skin color, suitable for body type, suitable for weather.

8. The body data analysis and operation system based on deep learning technology according to claim 7, characterized in that: The personalized and precise recommendation system based on dressing experts will recruit a large number of third-party cooperation teams such as designers, clothing stylists, and clothing consultants according to the user volume in the later stage, and provide customized clothing matching design and plan recommendations for each user in a professional capacity on the platform.

Images