CN209486697U - A kind of gesture recognition identification authentication system based on deep learning - Google Patents
A kind of gesture recognition identification authentication system based on deep learning Download PDFInfo
- Publication number
- CN209486697U CN209486697U CN201920246394.7U CN201920246394U CN209486697U CN 209486697 U CN209486697 U CN 209486697U CN 201920246394 U CN201920246394 U CN 201920246394U CN 209486697 U CN209486697 U CN 209486697U
- Authority
- CN
- China
- Prior art keywords
- module
- data
- user
- identification authentication
- user terminal
- 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.)
- Expired - Fee Related
Links
Landscapes
- User Interface Of Digital Computer (AREA)
Abstract
The utility model discloses a kind of gesture recognition identification authentication system based on deep learning, including user terminal, sensor module, data preprocessing module, data interaction module, online Semi-supervised training module, identification authentication module, server end；Sensor module is collected for user behavior data；Data preprocessing module is used for user's noise reduction；Data interaction module is used for and arranges user data upload to remote server；Online Semi-supervised training module carries out user behavior characteristics extraction to the data come are collected, trains the model to identify user identity, mobile phone wrist-watch end is back to after the completion of training, realize identified off-line by the operational capability of server；Identification authentication module is used to the result of identification feeding back to user, decides whether further to open permission；The data that the utility model is acquired by user terminal realize that the gesture mode exclusive to everyone identifies, can achieve the effect that precisely to distinguish equipment owner and non-owner.
Description
Technical field
The utility model belongs to deep learning and human body attitude identification technology field, is related to a kind of gesture recognition authentication
Device, and in particular to a kind of method with deep learning to gyro sensor, acceleration transducer, gravity sensor number
According to model of mind training is carried out, to reach the device for realizing authentication.
Background technique
Existing identity recognizing technology such as fingerprint recognition now, iris recognition, face recognition etc. technology, although accurately
Spend it is high, but due to exist leakage the important privacy of user risk.Based on this, the utility model, which uses, passes through sensing data
To carry out identification.Since sensing data is compared to image, fingerprint, more there is concealment, while leakage not
Material risk can be brought, therefore this method has higher safety.Simultaneously compared to image recognition, the body based on sensing data
Part identification is low for equipment requirements, and real-time is higher.
The existing identity recognizing technology based on sensing data before, has that accuracy is not high.
Utility model content
In order to solve the above-mentioned technical problem, the utility model provides a kind of gesture recognition identity based on deep learning
Authentication device.
The technical scheme adopted by the utility model is a kind of gesture recognition identification authentication system based on deep learning,
It is characterized by comprising user terminal, sensor module, data preprocessing module, data interaction module, the training of online Semi-supervised
Module, identification authentication module, server end；
The sensor module is collected for user behavior data；The data preprocessing module is used for user's noise reduction；
The data interaction module is used for and arranges user data upload to remote server；The online Semi-supervised training module is borrowed
The operational capability for helping server carries out user behavior characteristics extraction to the data come are collected, and training is to identify user identity
Model is back to mobile phone wrist-watch end after the completion of training, realize identified off-line；The result that the identification authentication module is used to identify
User is fed back to, decides whether further to open permission；
The sensor module, data preprocessing module, data interaction module, identification authentication module are installed in the use
In the end of family, it is connect with the processor of user terminal；The online Semi-supervised training module is mounted in the server end, with clothes
The processor connection at business device end；
The user terminal is communicated with the server end by network connection.
The data that the utility model is acquired by user terminal realize that the gesture mode exclusive to everyone identifies, accurately
Degree reaches the degree for being enough to realize authentication, can achieve the effect that precisely to distinguish equipment owner and non-owner.
Detailed description of the invention
Fig. 1 is the frame principle of the utility model embodiment；
Fig. 2 is the on-line training module working principle diagram of the utility model embodiment.
Specific embodiment
The utility model is understood and implemented for the ease of those of ordinary skill in the art, it is right with reference to the accompanying drawings and embodiments
The utility model is described in further detail, it should be understood that implementation example described herein is only used for describing and explaining this
Utility model is not used to limit the utility model.
Referring to Fig.1, a kind of gesture recognition identification authentication system based on deep learning provided by the utility model, including with
Family end, sensor module, data preprocessing module, data interaction module, online Semi-supervised training module, identification certification mould
Block, server end；
Sensor module is collected for user behavior data；Data preprocessing module is used for user's noise reduction；Data interaction
Module is used for and arranges user data upload to remote server；Online Semi-supervised training module by server operation energy
Power carries out user behavior characteristics extraction to the data come are collected, trains the model to identify user identity, return after the completion of training
It is back to user terminal, realizes identified off-line；Identification authentication module is used to the result of identification feeding back to user, decides whether further
Open permission；
Sensor module, data preprocessing module, data interaction module, identification authentication module are installed in user terminal,
It is connect with the processor of user terminal；Online Semi-supervised training module is mounted in server end, the processor with server end
Connection；
User terminal is communicated with server end by network connection.
The user terminal of the present embodiment is connect with smartwatch, smart phone or tablet computer by bluetooth or USB interface logical
Letter.
The sensor module of the present embodiment includes gravity sensor, acceleration transducer, gyro sensor, is respectively used to
The acceleration of gravity on the corresponding direction XYZ of human body particular pose is acquired, the acceleration and user terminal generated in motion process is inclined
The angle information turned.
The data preprocessing module of the present embodiment uses TMS320F2812 chip, to data by the way of threshold value is arranged
Carry out basic noise reduction operation；Made when data preprocessing module uploads data using 9 data of three sensors X YZ axial directions
For a vector, it is packaged in the form of the duplicate Vector Groups in front and back 50% and transmits the front and back continuity for guaranteeing data.
The data interaction module of the present embodiment uses bluetooth 4.0.
The processor of the user terminal of the present embodiment uses millet ADXL362.
The processor of the server end of the present embodiment uses GTX 1080TI Hybrid.
The user terminal of the present embodiment is also configured with data-interface, charge port and indicator light.
The online Semi-supervised training module of the present embodiment carries out feature extraction firstly the need of to sensing data, utilizes x
Axis, y-axis, z-axis a series of numerical characteristics of sensing data carry out subsequent operation.Then data set is randomly divided into
Training set and test set, ratio are about 8:2.Finally data are trained and are verified using algorithm.
Low latitudes problem is mapped to high latitude using kernel function by the online Semi-supervised training module of the present embodiment, thus
The case where making original unclassified, becomes to classify, and data can be divided by finding best hyperplane in high latitude space
Class, i.e., point to hyperplane back gauge and to reach maximum, can guarantee that error is smaller during the test in this way, can achieve more
Good effect.
The online Semi-supervised training module of the present embodiment has also used the DeepConvLSTM algorithm based on deep learning,
Algorithm fusion convolution sum LSTM operation, both can be with the space attribute of learning sample, can also be with learning time attribute.The algorithm will
3 reference axis (XYZ) data of every class sensor are merged into a data matrix, i.e., (128,3) are tieed up, as input data, often
Class sensor creates 1 DeepConvLSTM model, totally 3 models.By 3 convolution operations and 3 LSTM operations, will count
According to the LSTM output vector for being abstracted as 128 dimensions.In the convolution unit of CNN, by convolution (1x1 convolution kernel), BN,
The combination operation of MaxPooling (2 Wei Chiization), Dropout, continuous 3 groups, last group executes Dropout.Pass through
The dimensionality reduction of MaxPooling operates (2^3=8), and the data of 128 dimensions are switched to for the high-level characteristic of 16 dimensions.In the timing list of RNN
It in member, is operated by LSTM, hidden layer neuron number is set as 128, and continuously three times, the convolution Feature Conversion by 16 dimensions is
The temporal aspect of 128 dimensions, then execute Dropout operation.Finally, 3 models of 3 sensors are exported, merge into one it is defeated
Enter, i.e. 128*3=384, then execute the operations such as Dropout, full connection, BN, finally uses activation primitive, calculate user and be in
Certain appearance probability of state, and determine whether user is me from according to the highest corresponding gesture mode of probability.
In the present invention, after model training success, it will open identification authentication module, and feed back to user authentication
Result.Assuming that authentification failure, it will this section of sensing data is labeled as non-user data, authenticating successfully can then be marked
For user data, on-line training module is fed back to.
It is the on-line training module working principle diagram of the present embodiment see Fig. 2.After behavioral data is obtained from client, into
Row Data processing (data mart modeling), Feature generation&selction (generate and select feature), Semi-
Supervised online learning (online semi-supervised learning), finally obtains Classfier (classifier), submits to knowledge
Other module.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of utility model patent protection scope, those skilled in the art are not departing under the enlightenment of the utility model
Under ambit protected by the claims of this utility model, replacement or deformation can also be made, the utility model is each fallen within
Within protection scope, the utility model is claimed range and should be determined by the appended claims.
Claims (5)
1. a kind of gesture recognition identification authentication system based on deep learning, it is characterised in that: including user terminal, sensor die
Block, data preprocessing module, data interaction module, online Semi-supervised training module, identification authentication module, server end；
The sensor module is collected for user behavior data；The data preprocessing module is used for user's noise reduction；It is described
Data interaction module is used for and arranges user data upload to remote server；The online Semi-supervised training module is by clothes
The operational capability of business device carries out user behavior characteristics extraction to the data come are collected, trains the model to identify user identity,
It is back to user terminal after the completion of training, realizes identified off-line；The identification authentication module is used to the result of identification feeding back to use
Family decides whether further to open permission；
The sensor module, data preprocessing module, data interaction module, identification authentication module are installed in the user terminal
It is interior, it is connect with the processor of user terminal；The online Semi-supervised training module is mounted in the server end, with server
The processor at end connects；
The user terminal is communicated with the server end by network connection；
The sensor module includes gravity sensor, acceleration transducer, gyro sensor, and it is special to be respectively used to acquisition human body
Determine the acceleration of gravity on the corresponding direction XYZ of posture, the angle letter of the acceleration generated in motion process and user terminal deflection
Breath；
The data preprocessing module uses TMS320F2812 chip；
The data interaction module uses bluetooth 4.0.
2. the gesture recognition identification authentication system according to claim 1 based on deep learning, it is characterised in that: the use
Family end and smartwatch, smart phone or tablet computer pass through bluetooth or USB interface connection communication.
3. the gesture recognition identification authentication system described in -2 any one based on deep learning according to claim 1, feature
Be: the processor of the user terminal uses millet ADXL362.
4. the gesture recognition identification authentication system described in -2 any one based on deep learning according to claim 1, feature
Be: the processor of the server end uses 1080 TI Hybrid of GTX.
5. the gesture recognition identification authentication system described in -2 any one based on deep learning according to claim 1, feature
Be: the user terminal is also configured with data-interface, charge port and indicator light.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201920246394.7U CN209486697U (en) | 2019-02-27 | 2019-02-27 | A kind of gesture recognition identification authentication system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201920246394.7U CN209486697U (en) | 2019-02-27 | 2019-02-27 | A kind of gesture recognition identification authentication system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN209486697U true CN209486697U (en) | 2019-10-11 |
Family
ID=68134065
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201920246394.7U Expired - Fee Related CN209486697U (en) | 2019-02-27 | 2019-02-27 | A kind of gesture recognition identification authentication system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN209486697U (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111750919A (en) * | 2020-07-02 | 2020-10-09 | 陕西师范大学 | Identity authentication method and apparatus using multi-axis sensor and accelerometer |
| CN113626785A (en) * | 2021-07-27 | 2021-11-09 | 武汉大学 | Fingerprint authentication security enhancement method and system based on user fingerprint pressing behavior |
-
2019
- 2019-02-27 CN CN201920246394.7U patent/CN209486697U/en not_active Expired - Fee Related
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111750919A (en) * | 2020-07-02 | 2020-10-09 | 陕西师范大学 | Identity authentication method and apparatus using multi-axis sensor and accelerometer |
| CN113626785A (en) * | 2021-07-27 | 2021-11-09 | 武汉大学 | Fingerprint authentication security enhancement method and system based on user fingerprint pressing behavior |
| CN113626785B (en) * | 2021-07-27 | 2023-10-27 | 武汉大学 | Fingerprint authentication security enhancement method and system based on user fingerprint pressing behavior |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Deep Fisher discriminant learning for mobile hand gesture recognition | |
| CN112528259B (en) | Identity verification method, device, computer equipment and storage medium | |
| Wang et al. | A hierarchical generative model for eye image synthesis and eye gaze estimation | |
| CN108154075A (en) | The population analysis method learnt via single | |
| Kundu et al. | Advancements and recent trends in emotion recognition using facial image analysis and machine learning models | |
| Jaswanth et al. | A novel based 3D facial expression detection using recurrent neural network | |
| CN109829362A (en) | Safety check aided analysis method, device, computer equipment and storage medium | |
| Zhang et al. | Robust multimodal recognition via multitask multivariate low-rank representations | |
| Soyel et al. | 3D facial expression recognition with geometrically localized facial features | |
| CN209486697U (en) | A kind of gesture recognition identification authentication system based on deep learning | |
| CN111444488A (en) | Identity authentication method based on dynamic gesture | |
| CN113779643B (en) | Signature handwriting recognition system and method based on pre-training technology and storage medium | |
| Li et al. | Adaptive deep feature fusion for continuous authentication with data augmentation | |
| Malik et al. | DeepAirSig: End-to-end deep learning based in-air signature verification | |
| Samanta et al. | On the role of head motion in affective expression | |
| Ma et al. | Landmark‐Based Facial Feature Construction and Action Unit Intensity Prediction | |
| CN113536262A (en) | Unlocking method and device based on facial expression, computer equipment and storage medium | |
| Ahmed et al. | A state of art for survey of combined iris and fingerprint recognition systems | |
| CN113742669B (en) | User authentication method based on twin network | |
| Aleluya et al. | Faceture ID: face and hand gesture multi-factor authentication using deep learning | |
| Boncolmo et al. | Gender Identification Using Keras Model Through Detection of Face | |
| Jamdar et al. | Multimodal biometric identification system using fusion level of matching score level in single modal to multi-modal biometric system | |
| Gonzalez-Garcia et al. | Introduction to presentation attacks in signature biometrics and recent advances | |
| Khoh et al. | In-air hand gesture signature recognition: An iHGS database acquisition protocol | |
| Li et al. | Memory-augmented autoencoder based continuous authentication on smartphones with conditional transformer gans |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191011 Termination date: 20200227 |