Method for accurately positioning assets by combining machine room face recognition technology with UWB-RFID

Abstract

The invention relates to the technical field of face recognition application, and discloses a machine room face recognition technology and UWB-RFID combined asset accurate positioning method, which comprises the following steps: establishing a face recognition feature local dictionary based on a deep learning framework in a face image quality detection module; asset topology information is added into the feature local dictionary, before an administrator enters a machine room, all management asset information of the machine room and associated information between assets form an information chain, the information chain is put into the asset topology information, and the feature local dictionary is updated, so that face recognition data are unified and integrated. The face recognition feature local dictionary based on the deep learning framework is innovatively established in the face image quality detection module, asset topology information is added into the feature local dictionary, all management asset information and network association information of the machine room form an information chain before an administrator enters the machine room, and assistance is provided for the service quality of the machine room.

Description

Technical Field

[0001] The present invention relates to the field of face recognition application technology, and in particular to a method for accurately locating assets by combining computer room face recognition technology with UWB-RFID. Background Art

[0002] Deep learning is an important part of machine learning. It is a version of a deep network framework based on artificial neural networks. It forms more abstract deep attributes by combining underlying features and amplifies the distributed laws of nonlinear data. This deep automatic feature learning method has stronger data understanding and learning capabilities than shallow networks. Deep learning draws on the information processing process of the human brain to process information in layers, extract features, and classify information. The access control system, especially the computer room, has evolved from traditional handwritten registration to face recognition, swiping work cards, etc., which are all ways of access and asset supervision. Therefore, it is necessary to have a technical capability for automatically verifying the information of the person through face recognition. The staff can check whether there are any abnormalities in the information of the person entering the area in the background system, so as to carry out targeted personnel management.

[0003] The existing access and asset supervision methods often make it difficult to grasp the dynamic information of personnel due to incomplete input information. They can only be displayed after manual background updates, which is inconvenient. At the same time, the traditional RFID technology has significant drawbacks in the precise positioning of densely deployed sensors within one meter. The application of the next-generation UWB-RFID combined with the Star Flash technology has also transformed the world-recognized "smart dust" technology from theory to actual productivity. Based on this, we provide a next-generation UWB-RFID combined with Star Flash technology for the application of face recognition, and use it in the field of access management and asset supervision to help computer room services. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the existing technology, the present invention provides a method for accurately locating assets in a computer room by combining face recognition technology with UWB-RFID. It innovatively establishes a local dictionary of face recognition features based on a deep learning framework in the face image quality detection module, adds asset topology information to the local dictionary of features, and organizes all the management asset information and network-related information of the computer room into an information chain before the administrator enters the computer room, thereby providing support for the service quality of the computer room and solving the above-mentioned technical problems.

[0006] (II) Technical solution

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID includes the following steps:

[0009] S1. The computer room service management system detects whether the basic face recognition image of the personnel meets the specifications through the face image quality detection module to avoid low recognition rate due to the basic face image not meeting the specifications; a local dictionary of face recognition features based on the deep learning framework is established in the face image quality detection module;

[0010] S2. Add asset topology information to the feature local dictionary. Before the administrator enters the computer room, all the management asset information and the associated information between assets in the computer room are combined into an information chain. The information chain is placed in the asset topology information and the feature local dictionary is updated to unify and integrate the face recognition data.

[0011] S3, splitting each target of the information chain of the topological structure contained in the feature local dictionary, and using the cumulative probability distribution function (CDF) model to model the probability distribution diagram of the positioning error of the positioning information received by the UWB-RFID transmitter corresponding to the target;

[0012] S4, the UWB-RFID transmitter uses the next generation UWB-RFID tag and transmits data through the IEEE 802.15.3a protocol;

[0013] S5, the UHF collector deployment evaluation algorithm of the UWB-RFID transponder calibrates the probability of the positioning error between the predicted value calculated by the cumulative probability distribution function model and the true value, and updates the feature local dictionary;

[0014] S6. The staff will send the personnel information of this area to the facial access control integrated machine. When the person passes through by face scanning, the access control integrated machine will verify the information and upload the pass record containing personal information to the management platform. If the person passing through carries items such as UWB-RFID tags in the computer room, causing abnormal access control, he will not be allowed to pass, and an alarm will be generated in the background of the management system.

[0015] Preferably, in S1, a method for establishing a local dictionary of face recognition features based on a deep learning framework uses a transfer learning method to optimize and improve parameters of an existing training model and applies it to the DLD_WJKCR algorithm model.

[0016] Preferably, the specific method for optimizing and improving the DLD_WJKCR algorithm model includes:

[0017] (1) Using a two-dimensional normal probability density function, randomly extract a series of image blocks near the center of the image and input them into the convolutional neural network. The probability density function is as follows:

[0018]

[0019] Where μ = [M / 2, N / 2] is the center of the face image, and p = [x, y] T is any point in the image, and n is an adjustable parameter;

[0020] (2) The training samples extract image blocks according to probability, and use convolutional neural networks to extract features to calculate the deep features of the image blocks. These deep features are then processed by correlation and screening to obtain the local facial feature dictionary atoms that are insensitive to intra-class changes required for recognition, and construct a deep local dictionary.

[0021] Preferably, in S3, the cumulative probability distribution function (CDF) is as follows:

[0022]

[0023] Among them, the CDF value of algorithm A under the specified accuracy threshold is Aor, N is the total number of positioning times, and n is the number of times the positioning error is lower than the selected threshold value; positioning with a positioning error higher than the selected threshold value is considered to be unsuccessful positioning, and the cumulative probability distribution coefficient can intuitively reflect the positioning performance of the algorithm.

[0024] Preferably, in S4, the next generation UWB-RFID tag meets the following five conditions:

[0025] (1) Even in indoor scenes or with obstacles, positioning can be performed with sub-meter accuracy;

[0026] (2) Small, with a maximum area of ​​no more than 10 square centimeters, and lightweight, without batteries;

[0027] (3) Made of recyclable materials;

[0028] (4) Conformable;

[0029] (5) Low cost to allow deployment of multiple UWB-RFID transponders in the environment.

[0030] Preferably, in S5, the UWB-RFID transponder includes a UWB / UHF antenna, a power management unit, a sensor, a control unit and a UWB backscatter modulator, and the power management unit includes a UHF collector, an energy storage device and a power converter, wherein the UWB / UHF antenna is respectively connected to the UHF collector and the UWB backscatter modulator, the UHF collector is connected to the power switch, the power switch is respectively connected to the energy storage device and the sensor, and the power switch is bidirectionally connected to the control unit, the sensor is connected to the control unit, and the control unit is connected to the UWB backscatter modulator.

[0031] Preferably, in S5, the evaluation algorithm is evaluated using the accuracy-RMSE metric, and the smaller the RMSE, the better; the formula of RMSE is as follows:

[0032]

[0033] Among them, r ij It is predicted that sample i has j classification categories, is the score given by the algorithm for predicting the asset coordinate location, Represents the total number of classification categories in the dataset.

[0034] RMSE is the most popular metric, which describes the gap between the results predicted by CDF and the actual asset coordinates. Optimizing the RMSE metric is actually predicting the health assessment score.

[0035] Preferably, the evaluation algorithm is evaluated using the accuracy-MAE metric;

[0036]

[0037] Among them, r ij It is predicted that sample i has j classification categories, is the score given by the algorithm for predicting the asset coordinate location, Represents the total number of classification categories in the dataset.

[0038] MAE is another commonly used metric. It directly calculates the absolute error between the predicted value and the true value and obtains the accuracy result through weighted average. The advantage of using weighted average is that it can make up for the shortcomings of RMSE and MAE that cause unclear prediction results due to oversimplified data classification. This improves the accuracy of the prediction results. The smaller the error, the higher the accuracy.

[0039] Preferably, in S6, the personnel information includes a face picture and ID card information.

[0040] Compared with the prior art, the present invention provides a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, which has the following beneficial effects:

[0041] The present invention innovatively establishes a local dictionary of face recognition features based on a deep learning framework in the face image quality detection module, adds asset topology information to the local dictionary of features, and organizes all the management asset information and network-related information of the computer room into an information chain before the administrator enters the computer room, thereby providing support for the service quality of the computer room.

[0042] The present invention updates the feature local dictionary by splitting the information chain of the topological structure contained in the feature local dictionary for each target, and adopts the probability distribution diagram of the cumulative probability distribution function (CDF) positioning error for the positioning information received by the UWB-RFID transmitter corresponding to the target. It solves the significant drawbacks of the traditional RFID coordinate positioning technology in the precise positioning of densely deployed RFID receivers within a certain distance. At the same time, the next generation of UWB-RFID combined with the application of star flash technology has also transformed the world-recognized "smart dust" technology from theory to actual productivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0044] Figure 1 This is a flow chart of a method for accurately locating assets in a computer room by combining face recognition technology with UWB-RFID in the present invention;

[0045] Figure 2 A schematic diagram of randomly extracting image blocks and extracting deep features in the present invention;

[0046] Figure 3 This is a structural diagram of the convolutional neural network migrated in the present invention;

[0047] Figure 4 This is a block diagram of the UWB-RFID transponder in the present invention. DETAILED DESCRIPTION

[0048] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0049] In the description of the present invention, unless otherwise specified, "plurality" means two or more than two; the orientations or positional relationships indicated by the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0050] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0051] like Figure 1 As shown, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID includes the following steps:

[0052] S1. The computer room service management system detects whether the basic face recognition image of the personnel meets the specifications through the face image quality detection module to avoid low recognition rate caused by the basic face image not meeting the specifications; a local dictionary of face recognition features based on the deep learning framework is established in the face image quality detection module.

[0053] Based on the method of establishing the local dictionary of face recognition features in the deep learning framework, the transfer learning method is used to optimize and improve the parameters of the existing training model and apply it to the DLD_WJKCR algorithm model; the optimization analysis of the deep local dictionary is based on the feature extraction process of the deep learning framework. The key to the algorithm lies in the training and migration of the model. Because the training model is too complex and time-consuming, the transfer learning method is used to optimize and improve the parameters of the existing training model and apply it to the DLD_WJKCR algorithm model.

[0054] The specific methods for optimizing and improving the DLD_WJKCR algorithm model include:

[0055] (1) Figure 2 As shown in the figure, in a continuous video image, many continuous images of a person's face at different angles, different postures and different expressions can be tracked on the time axis. When the image reaches the appropriate angle and proportion, these image sets are stored and identified and classified. In general, the face images that meet the requirements are in the middle and nearby positions of a picture, that is, near the middle position of the picture, the key information of the face is the most, and the value that can be mined is the greatest. Therefore, a two-dimensional normal probability density function is used to randomly extract a series of image blocks near the center of the picture and put them into the convolutional neural network. The probability density function is as follows:

[0056]

[0057] Where μ = [M / 2, N / 2] is the center of the face image, and p = [x, y] Tis any point in the image, and n is an adjustable parameter. Using this probability density function, a large number of image blocks near the center of the image are randomly and quickly extracted, which has a high probability of covering the key positions where the face can be identified. In this way, many pixels belonging to the background are filtered out, avoiding the interference and redundancy of irrelevant pixel blocks, making the facial features more prominent, and improving the feature extraction performance to a certain extent.

[0058] (2) Figure 3 As shown in the figure, the training samples extract image blocks by probability, use the convolutional neural network to extract features to calculate the deep features of the image blocks, and then perform correlation and screening on these deep features to obtain the face local feature dictionary atoms that are insensitive to intra-class changes required for recognition, and construct a deep local dictionary. This dictionary has stronger encoding capabilities than the dictionary obtained by ordinary sparse learning. The test samples are also subjected to similar extraction and calculation processing to adapt to the convolutional neural network structure of the training set change migration.

[0059] S2. Add asset topology information to the feature local dictionary. Before the administrator enters the computer room, organize all the management asset information and the relationship information between assets into an information chain. Put the information chain into the asset topology information and update the feature local dictionary to unify and integrate the face recognition data.

[0060] Benefits of unifying and integrating facial recognition data:

[0061] Unified data: The feature local dictionary can unify data from different sources and formats. Both structured and unstructured data can be effectively integrated into a knowledge base.

[0062] Integrated data source: It supports mapping relationships across data entities, allowing information from different data sources to be interconnected to form a complete network of associated information.

[0063] The associated information network includes:

[0064] The first layer uses the correlation between database, middleware, and server basic monitoring indicators in the log service application database to mine the core application service topology and associate it with the network topology. The core application service topology records the relationship between the three types of indicators: database, middleware, and server basic monitoring indicators between different servers;

[0065] The second layer is to mine a data set based on the topological relationship of the core application services in the first layer from the logs of each service, called the topological data set. This data records the monitoring data and operating health status of the server and database, middleware, server CPU, memory, disk, and process in different time dimensions;

[0066] The third layer, after locating the network device through the first two layers, drill down to the board layer to obtain a more accurate positioning source. The network device consists of multiple boards, each of which includes independent CPU, memory, disk and other hardware.

[0067] S3, splitting each target of the information chain of the topological structure contained in the feature local dictionary, and using the cumulative probability distribution function (CDF) model to model the probability distribution diagram of the positioning error of the positioning information received by the UWB-RFID transmitter corresponding to the target;

[0068] The cumulative probability distribution function (CDF) is as follows:

[0069]

[0070] Among them, the CDF value of algorithm A under the specified accuracy threshold is Aor, N is the total number of positioning times, and n is the number of times the positioning error is lower than the selected threshold value; positioning with a positioning error higher than the selected threshold value is considered to be unsuccessful positioning, and the cumulative probability distribution coefficient can intuitively reflect the positioning performance of the algorithm.

[0071] S4. The UWB-RFID transmitter uses the next generation UWB-RFID tag and transmits data through the IEEE 802.15.3a protocol; the next generation UWB-RFID tag needs to meet the following five conditions:

[0072] (1) Even in indoor scenes or with obstacles, positioning can be performed with sub-meter accuracy;

[0073] (2) Small, with a maximum area of ​​no more than 10 square centimeters, and lightweight, without batteries;

[0074] (3) Made of recyclable materials;

[0075] (4) Conformable;

[0076] (5) Low cost to allow deployment of multiple UWB-RFID transponders in the environment.

[0077] S5, the UHF collector deployment assessment algorithm of the UWB-RFID transponder calibrates the probability of the positioning error between the predicted value and the true value calculated by the cumulative probability distribution function (CDF) model, and updates the feature local dictionary;

[0078] Among them, Figure 4As shown, the UWB-RFID transponder includes a UWB / UHF antenna, a power management unit, a sensor, a control unit and a UWB backscatter modulator, wherein the power management unit includes a UHF collector, an energy storage device and a power converter, wherein the UWB / UHF antenna is respectively connected to the UHF collector and the UWB backscatter modulator, the UHF collector is connected to the power switch, the power switch is respectively connected to the energy storage device and the sensor, and the power switch is bidirectionally connected to the control unit, the sensor is connected to the control unit, and the control unit is connected to the UWB backscatter modulator.

[0079] The evaluation algorithm uses the accuracy-RMSE metric for evaluation. The smaller the RMSE, the better. The formula for RMSE is as follows:

[0080]

[0081] Among them, r ij It is predicted that sample i has j classification categories, is the score given by the algorithm for predicting the asset coordinate location, Represents the total number of classification categories in the dataset.

[0082] RMSE is the most popular metric, which describes the gap between the results predicted by CDF and the actual asset coordinates. Optimizing the RMSE metric is actually predicting the health assessment score.

[0083] Evaluation algorithms can also be evaluated using the accuracy-MAE metric;

[0084]

[0085] Among them, r ij It is predicted that sample i has j classification categories, is the score given by the algorithm for predicting the asset coordinate location, Represents the total number of classification categories in the dataset.

[0086] MAE is another commonly used metric. It directly calculates the absolute error between the predicted value and the true value and obtains the accuracy result through weighted average. The advantage of using weighted average is that it can make up for the shortcomings of RMSE and MAE that cause unclear prediction results due to oversimplified data classification. This improves the accuracy of the prediction results. The smaller the error, the higher the accuracy.

[0087] S6. The staff will send the personnel information of this area to the facial access control integrated machine. The personnel information includes facial pictures and ID card information. When the person passes through by face scanning, the access control integrated machine will verify the information and upload the pass record containing personal information to the management platform. If the person passing through carries items such as UWB-RFID tags in the computer room, causing abnormal access control, he / she will not be allowed to pass, and an alarm will be generated in the background of the management system.

[0088] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID. Features: S1. The computer room service management system detects whether the basic face recognition image of the personnel meets the specifications through the face image quality detection module; establishes a local dictionary of face recognition features based on the deep learning framework in the face image quality detection module; S2. Add asset topology information to the feature local dictionary. Before the administrator enters the computer room, all the management asset information and the associated information between assets in the computer room are combined into an information chain. The information chain is placed in the asset topology information and the feature local dictionary is updated to unify and integrate the face recognition data. S3, splitting each target of the information chain of the topological structure contained in the feature local dictionary, and using the cumulative probability distribution function model to model the probability distribution diagram of the positioning error of the positioning information received by the UWB-RFID transmitter corresponding to the target; S4, UWB-RFID transmitter uses the next generation UWB-RFID tag and transmits data through IEEE 802.15.3a protocol; S5, the UHF collector deployment assessment algorithm of the UWB-RFID transponder calibrates the probability of the positioning error between the predicted value and the true value calculated by the cumulative probability distribution function (CDF) model, and updates the feature local dictionary; S6. The staff will send the personnel information of this area to the facial access control integrated machine. When the person passes through by face scanning, the access control integrated machine will verify the information and upload the pass record containing personal information to the management platform. If the person passing through carries items such as UWB-RFID tags in the computer room, causing abnormal access control, he will not be allowed to pass, and an alarm will be generated in the background of the management system.

2. According to claim 1, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: In S1, a method for establishing a local dictionary of face recognition features based on a deep learning framework uses transfer learning to optimize and improve the parameters of an existing training model and applies it to the DLD_WJKCR algorithm model.

3. According to claim 2, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: The specific methods for optimizing and improving the DLD_WJKCR algorithm model include: (1) Using a two-dimensional normal probability density function, randomly extract a series of image blocks near the center of the image and input them into the convolutional neural network. The probability density function is as follows: Among them, μ = [M / 2, N / 2] is the center of the face image, p [x, y] T is any point in the image, and n is an adjustable parameter; (2) The training samples extract image blocks according to probability, and use convolutional neural networks to extract features to calculate the deep features of the image blocks. These deep features are then processed by correlation and screening to obtain the local facial feature dictionary atoms that are insensitive to intra-class changes required for recognition, and construct a deep local dictionary.

4. According to claim 1, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: In S3, the cumulative probability distribution function (CDF) is as follows: Among them, the CDF value of algorithm A under the specified accuracy threshold is Aor, N is the total number of positioning times, and n is the number of times the positioning error is lower than the selected threshold value.

5. According to claim 1, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: In S5, the UWB-RFID transponder includes a UWB / UHF antenna, a power management unit, a sensor, a control unit and a UWB backscatter modulator, wherein the power management unit includes a UHF collector, an energy storage device and a power converter, wherein the UWB / UHF antenna is respectively connected to the UHF collector and the UWB backscatter modulator, the UHF collector is connected to the power switch, the power switch is respectively connected to the energy storage device and the sensor, and the power switch is bidirectionally connected to the control unit, the sensor is connected to the control unit, and the control unit is connected to the UWB backscatter modulator.

6. According to claim 1, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: In S5, the evaluation algorithm uses the accuracy-RMSE metric for evaluation. The smaller the RMSE, the better. The formula for RMSE is as follows: Among them, r ij It is predicted that sample i has j classification categories, is the score given by the algorithm for predicting the asset coordinate location, Represents the total number of classification categories in the dataset.

7. According to claim 1, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: In S5, the algorithm is evaluated using the accuracy-MAE metric; Among them, r ij It is predicted that sample i has j classification categories, is the score given by the algorithm for predicting the asset coordinate location, Represents the total number of classification categories in the dataset.

8. According to claim 1, a method for accurately locating assets by combining face recognition technology in a computer room with UWB-RFID, Features: In S6, the personnel information includes face pictures and ID card information.

Images