CN114038035A - Artificial intelligence recognition device based on big data - Google Patents
Artificial intelligence recognition device based on big data Download PDFInfo
- Publication number
- CN114038035A CN114038035A CN202111303955.0A CN202111303955A CN114038035A CN 114038035 A CN114038035 A CN 114038035A CN 202111303955 A CN202111303955 A CN 202111303955A CN 114038035 A CN114038035 A CN 114038035A
- Authority
- CN
- China
- Prior art keywords
- face
- data
- dimensional
- artificial intelligence
- image
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 33
- 230000000007 visual effect Effects 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 15
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000005516 engineering process Methods 0.000 claims abstract description 12
- 230000008859 change Effects 0.000 claims abstract description 6
- 238000011161 development Methods 0.000 claims abstract description 6
- 230000002452 interceptive effect Effects 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 43
- 241000282414 Homo sapiens Species 0.000 claims description 35
- 238000000034 method Methods 0.000 claims description 22
- 238000000605 extraction Methods 0.000 claims description 19
- 238000012937 correction Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000000513 principal component analysis Methods 0.000 claims description 11
- 238000004891 communication Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 7
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 230000001427 coherent effect Effects 0.000 claims description 5
- 210000004209 hair Anatomy 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000001360 synchronised effect Effects 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 230000003287 optical effect Effects 0.000 claims description 2
- 206010063385 Intellectualisation Diseases 0.000 abstract description 2
- 230000036544 posture Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000007306 turnover Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Abstract
The invention belongs to the technical field of face recognition, and discloses an artificial intelligence recognition device based on big data, which comprises: the three-dimensional laser scanning module acquires face information data through a dimensional micro-scanner; simultaneously, preprocessing the three-dimensional face point cloud; the three-dimensional image preprocessing module is used for analyzing the two aspects of extracting the characteristics of the visual two-dimensional multimedia target image and generating the change characteristics of the three-dimensional multimedia visual target image to extract the characteristics of the three-dimensional multimedia visual target image; the face model training module is used for calling an OpenCV internal function to obtain a trained model based on an OpenCV technology; the interactive recognition module is used for inputting face information data and recognizing the face information data; and the remote face recognition module is used for carrying out data processing and face recognition in a LabVIEW development environment and finally feeding back the recognized identity information to the slave machine. The invention can make the computer technology develop towards intellectualization and automation.
Description
Technical Field
The invention belongs to the technical field of face recognition; relates to an artificial intelligence recognition device based on big data.
Background
The intelligent human face recognition technology is characterized in that key parts of a human face are recognized and analyzed, expressions of a person are collected and stored in a database, and then the key parts of the human face and the expressions are recognized through comparison and analysis. Currently, the identification of strange faces by human beings is not completely accurate, and especially for some people with similar long faces, although the people can be distinguished, how to identify the faces is difficult to describe. That is more difficult for the vision system image capture device.
The problem of low recognition rate exists in the current face intelligent recognition technology.
Disclosure of Invention
In order to solve the technical problems, the invention provides an artificial intelligent identification device based on big data.
The invention is realized by the following steps: big data-based artificial intelligence recognition device, the big data-based artificial intelligence recognition device comprising:
the three-dimensional laser scanning module acquires face information data through the micro-dimensional scanner, and the face information data usually contains redundant information of hair and shoulders irrelevant to face recognition besides the face. In order to reduce the operation amount and improve the recognition rate, the three-dimensional face point cloud is preprocessed at the same time; the pretreatment comprises the following steps: face cutting and posture correction;
the three-dimensional image preprocessing module is used for extracting the three-dimensional multimedia visual target image characteristics by analyzing the two aspects of the extraction of the visual two-dimensional multimedia visual target image characteristics and the generation of the three-dimensional multimedia visual target image change characteristics; constructing a sparse representation algorithm-based model, limiting the minimum value of a target function by using gradient projection, optimizing the gradient direction to obtain a sparsely represented identification visual image, and realizing artificial intelligent identification of the three-dimensional multimedia visual image;
the face model training module is used for providing the algorithm in OpenCV2.4.9 based on the OpenCV technology, and only needs to classify faces in a database according to different people, then respectively read in the faces and call OpenCV internal functions to obtain a trained model;
the interactive recognition module is used for inputting face information data and recognizing the face information data;
the remote face recognition module is used for transmitting data to the host computer through a wireless channel by using the NI USRP equipment; the host receives the data of the slave from the connected NI USRP, performs data processing and face recognition in a LabVIEW development environment, and finally feeds back the recognized identity information to the slave.
Preferably, the three-dimensional laser scanning module includes:
the face cutting unit is used for determining the position of a nose tip quickly according to the fact that the nose tip has the maximum value on the face of the human face according to the geometric constraint of the human face, then drawing a sphere by taking the nose tip as a center, and the face area contained in the sphere is the effective area of the face of the human face;
the gesture correction unit is used for performing Principal Component Analysis (PCA) on the cut face point cloud, taking a feature vector corresponding to the maximum feature value as a Y axis of a new coordinate system, taking a feature vector corresponding to the minimum feature value as a Z axis, and establishing a right-hand coordinate system, wherein the coordinate system is called a posture coordinate system PCS; converting the cut human face point cloud into the PCS by taking the nose tip point as an original point of the PCS to finish the human face posture correction; all the human faces are converted into the front postures by establishing the PCS of the human face point cloud, and then the human face point cloud is converted into the same resolution ratio.
Preferably, the three-dimensional image preprocessing module includes:
the two-dimensional multimedia target image feature extraction unit is used for finding out a two-dimensional coordinate point by utilizing a visual target three-dimensional contour image and setting a multimedia geometric projection model capable of showing a perspective conversion model in a real environment;
and the three-dimensional multimedia visual target image unit is used for extracting three-dimensional multimedia visual dynamic image characteristics through optimization conversion of a nonlinear algorithm.
Preferably, the remote face recognition module adopts a coherent loss of a twin neural network: the device consists of 6 convolutional layers, 4 pooling layers and 1 full-connection layer. The 5 x 5 convolution kernel is split into two layers of 3 x 3 convolution kernels.
Preferably, the big data-based artificial intelligence recognition method for operating the big data-based artificial intelligence recognition apparatus includes:
the method comprises the steps that a micro-dimensional optical scanner obtains face information data, wherein the face information data comprise a face part and redundant information such as hair and shoulders which are irrelevant to face identification; simultaneously preprocessing the three-dimensional face point cloud; the pretreatment comprises the following steps: face cutting and posture correction;
extracting three-dimensional multimedia visual target image characteristics by analyzing the two aspects of visual two-dimensional multimedia target image characteristic extraction and three-dimensional multimedia visual target image change characteristic generation; constructing a sparse representation algorithm-based model, limiting the minimum value of a target function by utilizing gradient projection, optimizing the gradient direction to obtain a sparsely represented identification visual image, and realizing artificial intelligence identification of the three-dimensional multimedia visual image;
based on the OpenCV technology, the algorithm is provided in OpenCV2.4.9, and trained models can be obtained only by classifying human faces in a database according to different people, respectively reading the human faces and calling OpenCV internal functions;
inputting face information data for recognition;
the NI USRP equipment is used for sending data to the host through a wireless channel; the host receives the data of the slave from the connected NI USRP, performs data processing and face recognition in a LabVIEW development environment, and finally feeds back the recognized identity information to the slave.
Preferably, the PCA dimension reduction of the artificial intelligence recognition method based on big data comprises the following specific steps:
(1) removing a classification label of the data, and taking the d-dimensional data after removal as a sample;
(2) calculating a mean vector of d dimensions, the mean of each dimension vector of all data;
(3) calculating a dispersion matrix or covariance matrix of all data;
(4) calculating a feature value e1, e 2.., ed and a corresponding feature vector lambda1, lambda 2.., lambda d;
(5) sorting the eigenvectors in a descending order according to the magnitude of the eigenvalues, selecting the first k largest eigenvectors to form a matrix W with dimensions d x k, wherein each column represents one eigenvector;
(6) transforming the sample data into a new subspace by using the eigenvector matrix W of d × K:
y＝WT×x
where x is a vector in d x 1 dimensions representing one sample and y is a vector in K x 1 dimensions in the new subspace.
Preferably, the wireless channel of the remote face recognition based on the artificial intelligence recognition method of big data is realized by adopting a wireless communication system, and the information transmission is completed by adopting an asynchronous communication mode; the data is transmitted in blocks, and check bits are added when the data is transmitted; after the data to be sent is coded, a bit code is formed, then the data is divided into frames, and the data sequence in the frames is check, synchronous bits, the sequence of the data packet in the whole information sequence and data bits.
Preferably, the remote face recognition module of the artificial intelligence recognition method based on big data specifically includes:
(1) coherent loss using a twin neural network: the device consists of 6 convolutional layers, 4 pooling layers and 1 full-connection layer. Splitting a 5 × 5 convolution kernel into two layers of 3 × 3 convolution kernels, increasing the depth of a network, and taking P-RELU as an activation function without increasing the calculated amount:
(2) the whole network is trained by using CASIA-Webface:
firstly organizing training data, then performing feature learning on a network, learning to discrete Features by using the Microsoft max loss + center loss, and finally performing feature extraction or classification through a trained model;
in order to make the training result better, the types with too few samples are removed firstly, only 10010 people with the most images in 10575 are selected, and image cleaning is performed at the same time, so that the identity of the people in the test set is ensured not to be repeated. The data set is balanced by randomly selecting the same number of sheets in each class, and the data of the image is enhanced by horizontal turning; carrying out face key point detection and face normalization on the balanced data set, and then training a network after unifying the size 112 x 96;
(3) the feature comparison is carried out by adopting a method for calculating cosine similarity, the larger the cosine value is, the more similar the two human faces are, and the cosine formula is as follows:
if vectorization is performed on both sides b and c, the cosine values for calculating two vector angles can be written as:
and a feature extraction stage, namely extracting depth feature vectors from the test image through a trained network, and splicing and representing feature graphs of the original graph and a horizontal turnover graph of the original graph into representation of a modified graph.
Preferably, the feature extraction of the remote face recognition module of the artificial intelligence recognition method based on big data includes the steps of:
obtaining a plurality of local areas according to the segmented face images;
comparing one pixel point in each region with a pixel point in a neighborhood, if the gray value of the pixel point in the neighborhood is larger than that of the central pixel point, marking the position of the pixel point in the neighborhood as 1, and otherwise marking as 0, thereby obtaining the LBP value of the central pixel point;
calculating histograms of all neighborhoods, namely calculating the frequency of each number, and then normalizing the histograms;
and connecting the calculated histograms of each neighborhood into a feature vector, wherein the feature vector is a vector for describing LBP texture features of the whole image.
Preferably, the intelligent processing terminal carries the artificial intelligence recognition device based on big data; the intelligent processing terminal comprises: computer, panel, cell-phone.
In summary, the invention has the advantages and technical effects that: the invention converts all human faces to the front postures by establishing the PCS of the human face point cloud, and then converts the human face point cloud to the same resolution. PCA dimension reduction can map high-dimensional vectors to low-dimensional vectors while preserving principal component information of the vectors well. The method comprises the steps of preprocessing three-dimensional face point cloud; the pretreatment comprises the following steps: face cutting and posture correction; the calculation amount is reduced, the recognition rate is improved, and the subsequent processing of the face information data is facilitated.
The invention has the advantages of higher accuracy of visual image recognition, reduced recognition time and ensured stability of recognition rate. The wireless channel of the remote face recognition module provided by the invention is realized by adopting a wireless communication system, and the information transmission is finished by adopting an asynchronous communication mode. The data is transmitted in blocks, and check bits are added when the data is transmitted, so that the accuracy of data transmission is ensured. In order to make the training result better, the invention firstly removes the types with too few samples, only selects 10010 people with the most images in 10575, and simultaneously cleans the images, thereby ensuring that the number of the people in the test set is not repeated.
The application of the artificial intelligent recognition technology is a great revolution in technological innovation, and the computer technology can be developed towards intellectualization and automation. The research on the artificial intelligent identification technology improves the technical content, breaks through the bottleneck existing in the application and further reaches the world leading level.
Drawings
FIG. 1 is a schematic structural diagram of an artificial intelligence identification device based on big data according to the present invention;
FIG. 2 is a schematic structural diagram of a three-dimensional laser scanning module according to the present invention;
FIG. 3 is a schematic structural diagram of a three-dimensional image preprocessing module according to the present invention;
FIG. 4 is a flow chart of wireless communication of the remote face recognition module of the present invention;
FIG. 5 is a schematic diagram of a remote face recognition module of the present invention;
FIG. 6 is a flow chart of feature extraction for the remote face recognition module of the present invention;
description of reference numerals:
1: a three-dimensional laser scanning module;
2: a three-dimensional image preprocessing module;
3: a face model training module;
4: an interaction identification module;
5: a remote face recognition module;
1-1: a face cutting unit;
1-2: an attitude correction unit;
2-1: a two-dimensional multimedia target image feature extraction unit;
2-2: and a three-dimensional multimedia visual target image unit.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The technical solution of the present invention is further described with reference to the following specific examples.
Examples
As shown in fig. 1, the artificial intelligence recognition apparatus based on big data provided by the present invention includes: the system comprises a three-dimensional laser scanning module 1, a three-dimensional image preprocessing module 2, a face model training module 3, an interactive recognition module 4 and a remote face recognition module 5.
The three-dimensional laser scanning module 1 acquires face information data through a micro-dimensional scanner, and the face information data usually contains redundant information such as hair and shoulders which are irrelevant to face recognition besides a face. In order to reduce the operation amount and improve the recognition rate, the three-dimensional face point cloud is preprocessed at the same time; the pretreatment comprises the following steps: face cutting and posture correction.
The three-dimensional image preprocessing module 2 is used for extracting the three-dimensional multimedia visual target image characteristics by analyzing the two aspects of the extraction of the visual two-dimensional multimedia visual target image characteristics and the generation of the three-dimensional multimedia visual target image change characteristics; and constructing a sparse representation algorithm-based model, limiting the minimum value of a target function by using gradient projection, optimizing the gradient direction to obtain a sparsely represented identification visual image, and realizing artificial intelligent identification of the three-dimensional multimedia visual image.
The face model training module 3 is used for providing the algorithm in opencv2.4.9 based on the OpenCV technology, and the trained model can be obtained only by classifying faces in a database according to different people, respectively reading in the faces and calling an OpenCV internal function.
And the interactive identification module 4 is used for inputting the face information data and identifying.
The remote face recognition module 5 is used for transmitting data to the host computer through a wireless channel by using NI USRP equipment; the host receives the data of the slave from the connected NI USRP, performs data processing and face recognition in a LabVIEW development environment, and finally feeds back the recognized identity information to the slave.
As shown in fig. 2, the three-dimensional laser scanning module 1 provided by the present invention includes:
the face cutting unit 1-1 is used for determining the position of a nose tip quickly according to the fact that the nose tip has the maximum value on the face of the human face according to the geometric constraint of the human face, then drawing a sphere by taking the nose tip as a center and taking the radius of the sphere as the center, and the face area contained in the sphere is the effective area of the face of the human face.
The pose correction unit 1-2 is configured to perform Principal Component Analysis (PCA) on the cut face point cloud, use a feature vector corresponding to the maximum feature value as a Y axis of a new coordinate system, use a feature vector corresponding to the minimum feature value as a Z axis, and establish a right-hand coordinate system, which is called a Pose Coordinate System (PCS). And converting the point cloud of the cut human face into the PCS by taking the nose tip point as the origin of the PCS to finish the correction of the human face posture. All the faces are converted to the frontal pose by establishing a PCS of the face point cloud, which is then converted to the same resolution.
PCA dimension reduction can map high-dimensional vectors to low-dimensional vectors while preserving the principal component information of the vectors well.
The PCA dimension reduction comprises the following specific steps:
(1) removing a classification label (label) of the data, and taking the d-dimensional data after removal as a sample;
(2) calculating a d-dimensional mean vector (i.e., the mean of each-dimensional vector of all data);
(3) calculating a dispersion matrix (or covariance matrix) for all data;
(4) calculating a feature value (e1, e 2.., ed) and a corresponding feature vector (lambda1, lambda 2.., lambda d);
(5) sorting the eigenvectors in descending order according to the magnitude of the eigenvalue, selecting the first k largest eigenvectors to form a d x k dimensional matrix W (wherein each column represents an eigenvector)
(6) Transforming the sample data into a new subspace by using the eigenvector matrix W of d × K:
y＝WT×x
where x is a vector in d x 1 dimensions representing one sample and y is a vector in K x 1 dimensions in the new subspace.
The method comprises the steps of preprocessing three-dimensional face point cloud; the pretreatment comprises the following steps: face cutting and posture correction; the calculation amount is reduced, the recognition rate is improved, and the subsequent processing of the face information data is facilitated.
As shown in fig. 3, the three-dimensional image preprocessing module 2 provided by the present invention includes:
the two-dimensional multimedia target image feature extraction unit 2-1 is used for finding a two-dimensional coordinate point by utilizing a visual target three-dimensional contour image and setting a multimedia geometric projection model capable of showing a perspective conversion model in a real environment;
and the three-dimensional multimedia visual target image unit 2-2 is used for extracting the characteristics of the three-dimensional multimedia visual dynamic image through the optimization conversion of a nonlinear algorithm.
The invention has the advantages of higher accuracy of visual image recognition, reduced recognition time and ensured stability of recognition rate.
As shown in fig. 4, the wireless channel of the remote face recognition module 5 provided by the present invention is implemented by using a wireless communication system, and the information transmission is completed by using an asynchronous communication mode. The data is transmitted in blocks, and check bits are added when the data is transmitted, so that the accuracy of data transmission is ensured. After the data to be sent is coded, bit codes are formed, then the data are packaged into frames, and the data in the frames sequentially comprise check bits, synchronous bits, the sequence of the data packets in the whole information sequence and data bits.
As shown in fig. 5, the remote face recognition module 5 provided by the present invention specifically includes:
(1) coherent loss using a twin neural network: the device consists of 6 convolutional layers, 4 pooling layers and 1 full-connection layer. Splitting a 5 × 5 convolution kernel into two layers of 3 × 3 convolution kernels, increasing the depth of a network, and taking P-RELU as an activation function without increasing the calculated amount:
(2) the whole network is trained by using CASIA-Webface:
firstly organizing training data, then carrying out feature learning on the network, learning to discrete Features by using the Microsoft max loss + center loss, and finally carrying out feature extraction or classification through a trained model.
In order to make the training result better, the types with too few samples are removed firstly, only 10010 people with the most images in 10575 are selected, and image cleaning is performed at the same time, so that the identity of the people in the test set is ensured not to be repeated. The data set is balanced by randomly selecting the same number of sheets in each class, and the image is subjected to horizontally-reversed data enhancement. And (4) carrying out face key point detection and face normalization on the balanced data set, and training the network after the size is unified by 112 x 96.
(3) The feature comparison is carried out by adopting a method for calculating cosine similarity, the larger the cosine value is, the more similar the two human faces are, and the cosine formula is as follows:
if vectorization is performed on both sides b and c, the cosine values for calculating two vector angles can be written as:
and a feature extraction stage, namely extracting depth feature vectors from the test image through a trained network, and splicing and representing feature graphs of the original graph and a horizontal turnover graph of the original graph into representation of a modified graph.
As shown in fig. 6, the feature extraction of the remote face recognition module 5 provided by the present invention mainly includes the following steps:
obtaining a plurality of local areas according to the segmented face images;
comparing one pixel point in each region with a pixel point in a neighborhood, if the gray value of the pixel point in the neighborhood is larger than that of the central pixel point, marking the position of the pixel point in the neighborhood as 1, and otherwise marking as 0, thereby obtaining the LBP value of the central pixel point;
calculating histograms of all neighborhoods, namely calculating the frequency of each number, and then normalizing the histograms;
and connecting the calculated histograms of each neighborhood into a feature vector, wherein the feature vector is a vector for describing LBP texture features of the whole image.
Describing and matching the human face: the LBP texture feature vector obtained by feature extraction can represent a face image, and an SVM classifier in Open CV can be used for classification, so that the purpose of face recognition is achieved.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (10)
1. An artificial intelligence recognition device based on big data, characterized in that, the artificial intelligence recognition device based on big data includes:
the three-dimensional laser scanning module acquires face information data through the micro-dimensional scanner, wherein the face information data usually comprises redundant information of hair and shoulders which are irrelevant to face recognition besides a face; in order to reduce the operation amount and improve the recognition rate, the three-dimensional face point cloud is preprocessed at the same time; the pretreatment comprises the following steps: face cutting and posture correction;
the three-dimensional image preprocessing module is used for analyzing the two aspects of extracting the characteristics of the visual two-dimensional multimedia target image and generating the change characteristics of the three-dimensional multimedia visual target image to extract the characteristics of the three-dimensional multimedia visual target image; constructing a sparse representation algorithm-based model, limiting the minimum value of a target function by utilizing gradient projection, optimizing the gradient direction to obtain a sparsely represented identification visual image, and realizing artificial intelligent identification of the three-dimensional multimedia visual image;
the face model training module is used for providing the algorithm in OpenCV2.4.9 based on the OpenCV technology, only the faces in the database are classified according to different people and then read in respectively, and an OpenCV internal function is called to obtain a trained model;
the interactive recognition module is used for inputting face information data and recognizing the face information data;
the remote face recognition module is used for transmitting data to the host computer through a wireless channel by using the NI USRP equipment; the host receives the data of the slave from the connected NI USRP, performs data processing and face recognition in a LabVIEW development environment, and finally feeds back the recognized identity information to the slave.
2. The artificial intelligence identification device based on big data as claimed in claim 1, wherein the three-dimensional laser scanning module comprises:
the human face cutting unit is used for determining the position of a nose tip quickly according to the fact that the nose tip has the maximum value on a human face under the geometric constraint of the human face, then drawing a sphere by taking the nose tip as a center and taking a human face area contained in the sphere as an effective area of the human face;
the gesture correction unit is used for analyzing PCA (principal component analysis) by taking the cut face point cloud as a principal component, taking the feature vector corresponding to the maximum feature value as a Y axis of a new coordinate system, taking the feature vector corresponding to the minimum feature value as a Z axis, and establishing a right-hand coordinate system which is called a posture coordinate system PCS (personal computing system); taking the nose tip point as the origin of the PCS, converting the point cloud of the cut human face into the PCS, and finishing the correction of the human face posture; all the faces are converted to the frontal pose by establishing a PCS of the face point cloud, which is then converted to the same resolution.
3. The artificial intelligence recognition apparatus based on big data according to claim 1, wherein the three-dimensional image preprocessing module comprises:
the two-dimensional multimedia target image feature extraction unit is used for finding a two-dimensional coordinate point by utilizing a visual target three-dimensional contour image and setting a multimedia geometric projection model capable of showing a perspective conversion model in a real environment;
and the three-dimensional multimedia visual target image unit is used for extracting the characteristics of the three-dimensional multimedia visual dynamic image through the optimization conversion of a nonlinear algorithm.
4. The artificial intelligence based recognition device of claim 1, wherein the remote face recognition module employs a coherent loss of a twin neural network: the device consists of 6 convolution layers, 4 pooling layers and 1 full-connection layer; the 5 x 5 convolution kernel is split into two layers of 3 x 3 convolution kernels.
5. A big data based artificial intelligence recognition method for operating the big data based artificial intelligence recognition device according to any one of claims 1 to 4, wherein the big data based artificial intelligence recognition method comprises the following steps:
the method comprises the steps that a micro-dimensional optical scanner obtains face information data, wherein the face information data comprise redundant information such as hairs and shoulders which are irrelevant to face recognition besides face parts; simultaneously, preprocessing the three-dimensional face point cloud; the pretreatment comprises the following steps: face cutting and posture correction;
extracting three-dimensional multimedia visual target image features by analyzing the two aspects of visual two-dimensional multimedia target image feature extraction and three-dimensional multimedia visual target image change feature generation; constructing a sparse representation algorithm-based model, limiting the minimum value of a target function by utilizing gradient projection, optimizing the gradient direction to obtain a sparsely represented identification visual image, and realizing artificial intelligent identification of the three-dimensional multimedia visual image;
based on the OpenCV technology, the algorithm is provided in OpenCV2.4.9, and trained models can be obtained only by classifying human faces in a database according to different people, respectively reading the human faces and calling OpenCV internal functions;
inputting face information data for recognition;
the NI USRP equipment is used for sending data to the host through a wireless channel; the host receives the data of the slave from the connected NI USRP, performs data processing and face recognition in a LabVIEW development environment, and finally feeds back the recognized identity information to the slave.
6. The artificial intelligence recognition method based on big data according to claim 5, wherein the PCA dimension reduction of the artificial intelligence recognition method based on big data comprises the following steps:
(1) removing a classification label of the data, and taking the d-dimensional data after removal as a sample;
(2) calculating a mean vector of d dimensions, the mean of each dimension vector of all data;
(3) calculating a dispersion matrix or covariance matrix of all data;
(4) calculating a feature value e1, e 2.., ed and a corresponding feature vector lambda1, lambda 2.., lambda d;
(5) sorting the eigenvectors in a descending order according to the magnitude of the eigenvalues, selecting the first k largest eigenvectors to form a matrix W with dimensions d x k, wherein each column represents one eigenvector;
(6) transforming the sample data into a new subspace by using the eigenvector matrix W of d × K:
y＝WT×x
where x is a vector in d x 1 dimensions representing one sample and y is a vector in K x 1 dimensions in the new subspace.
7. The artificial intelligence recognition method based on big data according to claim 5, wherein the wireless channel of the remote face recognition of the artificial intelligence recognition method based on big data is realized by a wireless communication system, and the information transmission is completed by asynchronous communication; the data is transmitted in blocks, and check bits are added when the data is transmitted; after the data to be sent is coded, bit codes are formed, then the data are packaged into frames, and the data sequence in the frames is check, synchronous bits, the sequence of the data packet in the whole information sequence and data bits.
8. The artificial intelligence recognition method based on big data according to claim 5, wherein the remote face recognition module of the artificial intelligence recognition method based on big data specifically comprises:
(1) coherent loss using a twin neural network: the device consists of 6 convolution layers, 4 pooling layers and 1 full-connection layer; splitting a 5 × 5 convolution kernel into two layers of 3 × 3 convolution kernels, increasing the depth of a network, and taking P-RELU as an activation function without increasing the calculated amount:
(2) the whole network is trained by using CASIA-Webface:
firstly organizing training data, then performing feature learning on a network, learning to discrete Features by using the Microsoft max loss + center loss, and finally performing feature extraction or classification through a trained model;
in order to make the training result better, firstly removing the types with too few samples, only selecting 10010 people with the most images in 10575, and simultaneously cleaning the images to ensure that the identities of the people in the test set are not repeated; the data set is balanced by randomly selecting the same number of sheets in each class, and the data of the image is enhanced by horizontal turning; carrying out face key point detection and face normalization on the balanced data set, and then training a network after unifying the size 112 x 96;
(3) the feature comparison is carried out by adopting a method for calculating cosine similarity, the larger the cosine value is, the more similar the two human faces are, and the cosine formula is as follows:
if vectorization is performed on both sides b and c, the cosine values for calculating two vector angles can be written as:
and in the characteristic extraction stage, extracting depth characteristic vectors from the test image through a trained network, and splicing and representing the original image and the characteristic graphs of the horizontal turning graph of the original image into the representation of the changed graph.
9. The artificial intelligence recognition method based on big data according to claim 5, wherein the feature extraction of the remote face recognition module of the artificial intelligence recognition method based on big data comprises the steps of:
obtaining a plurality of local areas according to the segmented face images;
comparing one pixel point in each region with a pixel point in a neighborhood, if the gray value of the pixel point in the neighborhood is larger than that of the central pixel point, marking the position of the pixel point in the neighborhood as 1, and otherwise, marking the position of the pixel point in the neighborhood as 0, thereby obtaining the LBP value of the central pixel point;
calculating histograms of all neighborhoods, namely calculating the frequency of each number, and then normalizing the histograms;
and connecting the calculated histograms of each neighborhood into a feature vector, wherein the feature vector is a vector for describing LBP texture features of the whole image.
10. An intelligent processing terminal, characterized in that the intelligent processing terminal is equipped with the artificial intelligence recognition device based on big data of any claim 1-4; the intelligent processing terminal comprises: computer, panel, cell-phone.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111303955.0A CN114038035A (en) | 2021-11-05 | 2021-11-05 | Artificial intelligence recognition device based on big data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111303955.0A CN114038035A (en) | 2021-11-05 | 2021-11-05 | Artificial intelligence recognition device based on big data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114038035A true CN114038035A (en) | 2022-02-11 |
Family
ID=80142890
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111303955.0A Pending CN114038035A (en) | 2021-11-05 | 2021-11-05 | Artificial intelligence recognition device based on big data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114038035A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115830762A (en) * | 2023-01-17 | 2023-03-21 | 四川三思德科技有限公司 | Safety community access control platform, control method and control terminal |
-
2021
- 2021-11-05 CN CN202111303955.0A patent/CN114038035A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115830762A (en) * | 2023-01-17 | 2023-03-21 | 四川三思德科技有限公司 | Safety community access control platform, control method and control terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106682598B (en) | Multi-pose face feature point detection method based on cascade regression | |
| JP5121506B2 (en) | Image processing apparatus, image processing method, program, and storage medium | |
| US7912253B2 (en) | Object recognition method and apparatus therefor | |
| CN112784763B (en) | Expression recognition method and system based on local and overall feature adaptive fusion | |
| Sonkusare et al. | A review on hand gesture recognition system | |
| Yuan et al. | Facial expression feature extraction using hybrid PCA and LBP | |
| CN111339990A (en) | Face recognition system and method based on dynamic update of face features | |
| Kadhim et al. | A Real-Time American Sign Language Recognition System using Convolutional Neural Network for Real Datasets. | |
| CN109919077B (en) | Gesture recognition method, device, medium and computing equipment | |
| CN111126240A (en) | Three-channel feature fusion face recognition method | |
| CN111444488A (en) | Identity authentication method based on dynamic gesture | |
| CN110942110A (en) | Feature extraction method and device of three-dimensional model | |
| Angadi et al. | Face recognition through symbolic modeling of face graphs and texture | |
| CN110188646B (en) | Human ear identification method based on fusion of gradient direction histogram and local binary pattern | |
| CN114359917A (en) | Handwritten Chinese character detection and recognition and font evaluation method | |
| CN110610131A (en) | Method and device for detecting face motion unit, electronic equipment and storage medium | |
| CN114038035A (en) | Artificial intelligence recognition device based on big data | |
| CN103942572A (en) | Method and device for extracting facial expression features based on bidirectional compressed data space dimension reduction | |
| CN111881261A (en) | Internet of things multipoint response interactive intelligent robot system | |
| Muthukumar et al. | Vision based hand gesture recognition for Indian sign languages using local binary patterns with support vector machine classifier | |
| CN111898454A (en) | Weight binarization neural network and transfer learning human eye state detection method and device | |
| Srininvas et al. | A framework to recognize the sign language system for deaf and dumb using mining techniques | |
| Manolova et al. | Facial expression classification using supervised descent method combined with PCA and SVM | |
| Jaiswal et al. | Brief description of image based 3D face recognition methods | |
| CN113688864A (en) | Human-object interaction relation classification method based on split attention |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20220211 |
|
| WD01 | Invention patent application deemed withdrawn after publication |