CN107300971B - The intelligent input method and system propagated based on osteoacusis vibration signal - Google Patents

The intelligent input method and system propagated based on osteoacusis vibration signal Download PDF

Info

Publication number
CN107300971B
CN107300971B CN201710433231.5A CN201710433231A CN107300971B CN 107300971 B CN107300971 B CN 107300971B CN 201710433231 A CN201710433231 A CN 201710433231A CN 107300971 B CN107300971 B CN 107300971B
Authority
CN
China
Prior art keywords
signal
vibration signal
training sample
central point
class
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.)
Active
Application number
CN201710433231.5A
Other languages
Chinese (zh)
Other versions
CN107300971A (en
Inventor
伍楷舜
陈文强
王璐
蔡素到
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen University filed Critical Shenzhen University
Priority to CN201710433231.5A priority Critical patent/CN107300971B/en
Priority to PCT/CN2017/092769 priority patent/WO2018223489A1/en
Publication of CN107300971A publication Critical patent/CN107300971A/en
Application granted granted Critical
Publication of CN107300971B publication Critical patent/CN107300971B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/011Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
    • G06F3/014Hand-worn input/output arrangements, e.g. data gloves
    • GPHYSICS
    • G04HOROLOGY
    • G04GELECTRONIC TIME-PIECES
    • G04G21/00Input or output devices integrated in time-pieces
    • G04G21/02Detectors of external physical values, e.g. temperature
    • G04G21/025Detectors of external physical values, e.g. temperature for measuring physiological data
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133Distances to prototypes
    • G06F18/24143Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Physiology (AREA)
  • Human Computer Interaction (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
  • User Interface Of Digital Computer (AREA)

Abstract

The present invention provides a kind of intelligent input method propagated based on osteoacusis vibration signal and system, which includes: S1. receiving step, and smart machine vibrating sensor receives vibration signal, and carries out noise reduction process to vibration signal;S2. extraction step is detected using double threshold end-point detection method and extracts the vibration signal segment generated due to tapping human body appointed part;S3. processing step extracts signal characteristic, based on RNM algorithm come category signal position.The beneficial effects of the present invention are: the present invention not only solves the problems, such as smartwatch text input mode, moreover it is possible to reach high discrimination, and can quickly be inputted, the cruise duration of smartwatch is not lost yet.

Description

The intelligent input method and system propagated based on osteoacusis vibration signal
Technical field
The present invention relates to intelligent wearable device technical fields, more particularly to the intelligence propagated based on osteoacusis vibration signal Input method and system.
Background technique
In recent years, we witnessed the fast development of intelligent wearable device, Intelligent bracelet, intelligent earphone, Brilliant Eyes The wearable devices such as mirror, intelligent helmet and smartwatch increased popularity in people's daily life, is received by people.In intelligence In the case that the innovative space of mobile phone constantly reduces and the market growth approaches saturation, intelligent wearable device is as mobile terminal Next hot spot of industry is accepted by market extensively, is foretold invention for that will replace mobile phone.
Smartwatch is popular as a kind of portable device.But it small and exquisite and light also faces and can not evade Technical problem.Since screen is smaller, it is only able to display several keys every time, and the display of other content can be blocked, efficiency is very low. There are mainly three types of the modes of the text input of smartwatch realization at present: conventional keyboard, text prediction and speech recognition.But with top Formula or not convenient, flexible enough or not safe enough.In the case where there is ambient noise conditions, phonetic recognization rate is extremely difficult to ideal effect, It the use of voice input is in public that significant discomfort is closed and in terms of for protection user password and other privacies.Although Multinational Research Team develops finger tracking identification technology, but the experience that user uses keyboard always can not be with large screen intelligence The effectively quick comfortable text input comparison of energy mobile phone, such as the Research Team of Washington, DC university in 2016 realize millimeter The accurate finger tracking technology of grade, allows user to realize the handwriting input based on acoustic location, but handwriting input on the mobile apparatus Always still too slow, it is not able to satisfy the demand of people.Want to expand the market demand of smartwatch, just must depth excavate it and answer With and solve the problems, such as text input.
Summary of the invention
The present invention provides a kind of intelligent input methods propagated based on osteoacusis vibration signal, include the following steps:
S1. receiving step, smart machine vibrating sensor receives vibration signal, and carries out at noise reduction to vibration signal Reason;
S2. extraction step is detected using double threshold end-point detection method and is extracted due to tapping the generation of human body appointed part Vibration signal segment;
S3. processing step extracts signal characteristic, based on RNM algorithm come category signal position.
As a further improvement of the present invention, include: in the S2. extraction step
High and low two thresholdings are arranged to the signal of processing in step S21;
Step S22 primarily determines knocking starting point, and works as signal when the energy or zero-crossing rate of signal surmount low threshold Energy or zero-crossing rate break through high threshold, just determination the real starting point of knocking;When signal energy and zero-crossing rate are lower than simultaneously When low threshold, signaling destination point is determined;
Step S23 retains the data of starting point to the end, obtains the vibration signal segment by tapping the back of the hand generation.
As a further improvement of the present invention, include: in the S3. processing step
Dissection signal is normalized in step S31, extracts Mel-frequency Cepstral Coefficients, obtains signal characteristic;
Step S32 divides signal characteristic using the RNM algorithm based on stochastic subspace and nearest central point algorithm Class, and then judge beating position;
The step S32 includes:
Step S321, according to the difference of beating position, the signal characteristic that acquisition step S31 is obtained makees training sample, goes forward side by side Row classification;
Step S322 calculates every a kind of central point of training sample using nearest central point algorithm;
Step S323, based on stochastic subspace, the class central point of test sample and training sample is repeatedly compared in subspace, Obtain multiple classification results;
Step S324 uses simple majority Voting principle to classification results, and certain poll ratio is arranged, and obtains most Ticket obtains final classification result in the case where reaching certain poll simultaneously;
Step S325, the successful sample that will classify are included into new training sample, recalculate in the new class of new training sample Heart point.
As a further improvement of the present invention, in the step S321, the back of the hand different location is tapped, step is collected The signal characteristic of rapid S31 is divided into n class as training sample, mark position and according to the difference of position, and every class percussion m times, wherein n It is greater than or equal to 1 with m;
In the step S323, T attribute of random sampling, weight are distinguished into the class center generated each time in step S322 Multiple above operation Q times, obtains Q sub-spaces, each class center of contrast test sample and training sample one by one in subspace The Euclidean distance of point finds nearest central point to get to the classification of Q sub-spaces as a result, wherein T, Q are greater than 1.
As a further improvement of the present invention, in the S2. extraction step, human body appointed part is the back of the hand;
Vibrating sensor is piezoelectric ceramic vibration sensor;
In the S1. receiving step, carrying out noise reduction process to vibration signal includes:
Step S11 filters DC component and low-frequency noise using the butterworth high pass filter of 20Hz;
Step S12 filters high-frequency noise using 800Hz low-pass filtering.
The present invention also provides a kind of intelligent input systems propagated based on osteoacusis vibration signal, comprising:
Receiving module, smart machine vibrating sensor receives vibration signal, and carries out noise reduction process to vibration signal;
Extraction module is detected using double threshold end-point detection method and extracts the vibration generated due to tapping human body appointed part Signal segment;
Processing module extracts signal characteristic, based on RNM algorithm come category signal position.
As a further improvement of the present invention, include: in the extraction module
First extraction module, for high and low two thresholdings to be arranged to the signal of processing;
Second extraction module surmounts low threshold for the energy or zero-crossing rate when signal, primarily determines knocking starting point, And when the energy of signal or zero-crossing rate breakthrough high threshold, just determine the real starting point of knocking;When signal energy and zero-crossing rate When being lower than low threshold simultaneously, signaling destination point is determined;
Third extraction module obtains the vibration signal piece by tapping the back of the hand generation for retaining the data of starting point to the end Section.
As a further improvement of the present invention, include: in the processing module
First processing module extracts Mel-frequency Cepstral Coefficients, obtains signal for dissection signal to be normalized Feature;
Second processing module is calculated for using signal characteristic based on stochastic subspace and the RNM of nearest central point algorithm Method is classified, and then judges beating position;
The Second processing module includes:
First processing units, for the difference according to beating position, the signal characteristic that acquisition step S31 is obtained makees training sample This, and classify;
The second processing unit, for using nearest central point algorithm to calculate every a kind of central point of training sample;
Third processing unit, for being based on stochastic subspace, the class central point of test sample and training sample is repeatedly in son Spatial contrast obtains multiple classification results;
For using simple majority Voting principle to classification results, and certain poll ratio is arranged in fourth processing unit, It gains the majority while obtaining final classification result in the case where reaching certain poll;
5th processing unit, for that will classify, successful sample is included into new training sample, recalculates new training sample New class central point.
As a further improvement of the present invention, in the first processing units, the back of the hand different location is tapped, is acquired Signal characteristic to step S31 is divided into n class as training sample, mark position and according to the difference of position, and every class taps m times, Wherein n and m is greater than or equal to 1;
In the third processing unit, random sampling T is distinguished at the class center generated each time in the second processing unit A attribute, repeats above operation Q times, obtains Q sub-spaces, and contrast test sample and training sample one by one is each in subspace The Euclidean distance of a class central point finds nearest central point to get to the classification of Q sub-spaces as a result, wherein T, Q are greater than 1.
As a further improvement of the present invention,
In the extraction module, human body appointed part is the back of the hand;
Vibrating sensor is piezoelectric ceramic vibration sensor;
In the receiving module, carrying out noise reduction process to vibration signal includes:
First receiving module, for filtering DC component and low-frequency noise using the butterworth high pass filter of 20Hz;
Second receiving module, for filtering high-frequency noise using 800Hz low-pass filtering.
The beneficial effects of the present invention are: the present invention not only solves the problems, such as smartwatch text input mode, moreover it is possible to reach High discrimination, and can quickly be inputted, the cruise duration of smartwatch is not lost yet.
Detailed description of the invention
Fig. 1 is user in the virtual nine grids typewriting schematic diagram of the back of the hand percussion;
Fig. 2 is the schematic diagram of piezoelectric ceramic vibration sensor;
Fig. 3 is the structure chart of piezoelectric ceramic vibration sensor;
Fig. 4 is original signal waveform figure;
Fig. 5 is adaptive-filtering figure;
Fig. 6 is low-pass filtering figure;
Fig. 7 is RNM algorithm flow chart.
Specific embodiment
As shown in Figure 1, the invention discloses a kind of intelligent input methods propagated based on osteoacusis vibration signal, including such as Lower step:
S1. receiving step, smart machine vibrating sensor receives vibration signal, and carries out at noise reduction to vibration signal Reason;
S2. extraction step is detected using double threshold end-point detection method and is extracted due to tapping the generation of human body appointed part Vibration signal segment;
S3. processing step extracts signal characteristic, based on RNM algorithm come category signal position.
In the S2. extraction step, human body appointed part is the back of the hand;
Vibrating sensor is piezoelectric ceramic vibration sensor, and smart machine includes smartwatch, piezoelectric ceramic vibration sensing Device is built in smartwatch, Fig. 2,3 be piezoelectric ceramic vibration sensor schematic diagram and structure chart.Because piezoelectric effect makes inside Polarity generates variation, externally shows the variation of voltage, and operator is allowed to tap the back of the hand, and acquisition taps generated vibration signal.
Fig. 4 is original signal waveform figure, it can be seen that original signal collected to extraneous strong antijamming capability, noise compared with It is few.Fig. 5 uses the waveform diagram after low-pass filtering using adaptive-filtering and Fig. 6.It is more special that low-pass filtering remains signal Sign, it is more preferable to the effect of the vibration signal for minute differences of classifying later.
In the S1. receiving step, carrying out noise reduction process to vibration signal includes:
Step S11 filters DC component and low-frequency noise using the butterworth high pass filter of 20Hz;
Step S12 filters high-frequency noise using 800Hz low-pass filtering.
Include: in the S2. extraction step
High and low two thresholdings are arranged to the signal of processing in step S21;
Step S22 primarily determines knocking starting point, and works as signal when the energy or zero-crossing rate of signal surmount low threshold Energy or zero-crossing rate break through high threshold, just determination the real starting point of knocking;When signal energy and zero-crossing rate are lower than simultaneously When low threshold, signaling destination point is determined;
Step S23 only retains the data of starting point to the end, obtains the vibration signal segment by tapping the back of the hand generation.
In S3. processing step, extracting signal characteristic includes: that starting point and terminal in initialization training sample is longest The length unified as dissection signal is normalized signal consistent in length after dissection, uses formula are as follows:Its Middle x is vibration signal, and n is the dimension of signal.Feature extraction is carried out to the signal after normalization, reduces calculation amount, and retain original Signal most information, feature extraction is the Mel-frequency Cepstral Coefficients of signal, while remaining the spy of time domain and frequency domain Sign.
Include: in the S3. processing step
Dissection signal is normalized in step S31, extracts Mel-frequency Cepstral Coefficients, obtains signal characteristic;
Step S32 divides signal characteristic using the RNM algorithm based on stochastic subspace and nearest central point algorithm Class, and then judge beating position;
It come category signal position include: to include: based on RNM algorithm as shown in fig. 7, in the step S32
Step S321, according to the difference of beating position, the signal characteristic that acquisition step S31 is obtained makees training sample, goes forward side by side Row classification (being divided into 9 classes according to nine grids position);
Step S322 calculates every a kind of central point of training sample using nearest central point algorithm;
Step S323, based on stochastic subspace, the class central point of test sample and training sample is repeatedly compared in subspace, Obtain multiple classification results;
Step S324 uses simple majority Voting principle to classification results, and certain poll ratio is arranged, and obtains most Ticket obtains final classification result in the case where reaching certain poll simultaneously;
Step S325, the successful sample that will classify are included into new training sample, recalculate in the new class of new training sample Heart point.
In the step S321, the back of the hand different location is tapped, collects the signal characteristic of step S31 as training Sample, mark position are simultaneously divided into n class according to the difference of position, and every class taps m times, and wherein n and m is greater than or equal to 1;
In the step S323, T attribute of random sampling, weight are distinguished into the class center generated each time in step S322 Multiple above operation Q times, obtains Q sub-spaces, each class center of contrast test sample and training sample one by one in subspace The Euclidean distance of point finds nearest central point to get to the classification of Q sub-spaces as a result, wherein T, Q are greater than 1.
The invention also discloses a kind of intelligent input systems propagated based on osteoacusis vibration signal, comprising:
Receiving module, smart machine vibrating sensor receives vibration signal, and carries out noise reduction process to vibration signal;
Extraction module is detected using double threshold end-point detection method and extracts the vibration generated due to tapping human body appointed part Signal segment;
Processing module extracts signal characteristic, based on RNM algorithm come category signal position.
Include: in the extraction module
First extraction module, for high and low two thresholdings to be arranged to the signal of processing;
Second extraction module surmounts low threshold for the energy or zero-crossing rate when signal, primarily determines knocking starting point, And when the energy of signal or zero-crossing rate breakthrough high threshold, just determine the real starting point of knocking;When signal energy and zero-crossing rate When being lower than low threshold simultaneously, signaling destination point is determined;
Third extraction module, for only retaining the data of starting point to the end, remaining makees dissection processing, obtains by tapping the back of the hand The vibration signal segment of generation.
Include: in the processing module
First processing module extracts Mel-frequency Cepstral Coefficients, obtains signal for dissection signal to be normalized Feature;
Second processing module is calculated for using signal characteristic based on stochastic subspace and the RNM of nearest central point algorithm Method is classified, and then judges beating position;
The Second processing module includes:
First processing units, for the difference according to beating position, the signal characteristic that acquisition step S31 is obtained makees training sample This, and classify;
The second processing unit, for using nearest central point algorithm to calculate every a kind of central point of training sample;
Third processing unit, for being based on stochastic subspace, the class central point of test sample and training sample is repeatedly in son Spatial contrast obtains multiple classification results;
For using simple majority Voting principle to classification results, and certain poll ratio is arranged in fourth processing unit, It gains the majority while obtaining final classification result in the case where reaching certain poll;
5th processing unit, for that will classify, successful sample is included into new training sample, recalculates new training sample New class central point.
In the first processing units, the back of the hand different location is tapped, the signal characteristic conduct of step S31 is collected Training sample, mark position are simultaneously divided into n class according to the difference of position, and every class taps m times, and wherein n and m is greater than or equal to 1;
In the third processing unit, random sampling T is distinguished at the class center generated each time in the second processing unit A attribute, repeats above operation Q times, obtains Q sub-spaces, and contrast test sample and training sample one by one is each in subspace The Euclidean distance of a class central point finds nearest central point to get to the classification of Q sub-spaces as a result, wherein T, Q are greater than 1.
In the extraction module, human body appointed part is the back of the hand;Vibrating sensor is piezoelectric ceramic vibration sensor.
In the receiving module, carrying out noise reduction process to vibration signal includes:
First receiving module, for filtering DC component and low-frequency noise using the butterworth high pass filter of 20Hz;
Second receiving module, for filtering high-frequency noise using 800Hz low-pass filtering.
The present invention is embedded in small thin thin piezoelectric transducer on smartwatch, and machinery may be implemented in this piezoelectric transducer The conversion of electric energy can be arrived.A virtual nine grids keyboard on the back of the hand, when the grid of finger tapping different location, mechanical wave meeting It blazes abroad from all directions, encounters object back reflection and return.So the broadcast nature based on mechanical wave, piezoelectric transducer one It is secondary to receive the mechanical wave signals with different multipath transmisstions.On the one hand this mechanical wave signals are disseminated to by voice signal In air, on the other hand in the internal communication of hand and so-called osteoacusis.This some mechanical wave signal is not by the shadow of ambient noise It rings, can preferably be received by piezoelectric transducer, be converted to electric signal and handled by the controller of smartwatch.Because different grid produce Raw mechanical wave multipath effect is different, and the signal that smartwatch receives is with regard to different, using this species diversity, in conjunction with engineering The sorting algorithm of habit can sort out each key of nine grids.Thus, it is possible to realize the intelligence based on the back of the hand bone conduction technology It can wrist-watch text entry method and system.
The present invention for the first time makees the collected vibration for tapping the back of the hand in piezoelectric ceramic vibration sensor built in smartwatch It is defeated to be easy to implement text using the back of the hand as the virtual big screen of smartwatch the small screen for the text input mode of smartwatch Enter;Collected is to tap vibration signal on manpower after multipath transmisstion after the back of the hand, strong interference immunity, and is carried out to signal Noise reduction, dissection, normalization, after extracting the processing such as Mel-frequency Cepstral Coefficients, the RNM algorithm for reusing invention is classified, Its discrimination reaches 92%.Algorithm complexity as used herein is also linear rank, it is possible to realize the quick of text Input.In addition, piezoelectric ceramic vibration sensor power consumption pole bottom, will not be greatly decreased the cruise duration of smartwatch.
The present invention not only solves the problems, such as smartwatch text input mode, moreover it is possible to reach high discrimination, and can be fast Speed is inputted, and the cruise duration of smartwatch is not lost yet.
For hardware of the invention at low, system is simple, easy to use, osteoacusis of the realization based on the back of the hand that can be simple and quick The input for the smartwatch that vibration signal is propagated.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to of the invention Protection scope.

Claims (10)

1. a kind of intelligent input method propagated based on osteoacusis vibration signal, which comprises the steps of:
S1. receiving step, smart machine vibrating sensor receives vibration signal, and carries out noise reduction process to vibration signal;
S2. extraction step is detected using double threshold end-point detection method and extracts the vibration generated due to tapping human body appointed part Signal segment;
S3. processing step extracts signal characteristic, based on RNM algorithm come category signal position.
2. intelligent input method according to claim 1, which is characterized in that include: in the S2. extraction step
High and low two thresholdings are arranged to the signal of processing in step S21;
Step S22 primarily determines knocking starting point when the energy or zero-crossing rate of signal surmount low threshold, and works as the energy of signal Amount or zero-crossing rate break through high threshold, just determine the real starting point of knocking;When signal energy and zero-crossing rate are lower than low door simultaneously In limited time, signaling destination point is determined;
Step S23 retains the data of starting point to the end, obtains the vibration signal segment by tapping the back of the hand generation.
3. intelligent input method according to claim 1, which is characterized in that include: in the S3. processing step
The dissection signal in vibration signal segment is normalized in step S31, extracts Mel-frequency Cepstral Coefficients, obtains Signal characteristic;
Step S32 classifies to signal characteristic using the RNM algorithm based on stochastic subspace and nearest central point algorithm, into And judge beating position;
The step S32 includes:
Step S321, according to the difference of beating position, the signal characteristic that acquisition step S31 is obtained makees training sample, and is divided Class;
Step S322 calculates every a kind of central point of training sample using nearest central point algorithm;
Step S323, based on stochastic subspace, the class central point of test sample and training sample is repeatedly compared in subspace, is obtained Multiple classification results;
Step S324 uses simple majority Voting principle to classification results, and certain poll ratio is arranged, and gains the majority same When reach certain poll in the case where obtain final classification result;
Step S325, the successful sample that will classify are included into new training sample, recalculate the new class central point of new training sample.
4. intelligent input method according to claim 3, it is characterised in that:
In the step S321, the back of the hand different location is tapped, collects the signal characteristic of step S31 as training sample This, mark position is simultaneously divided into n class according to the difference of position, and every class taps m times, and wherein n and m is greater than or equal to 1;
In the step S323, T attribute of random sampling is distinguished at the class center generated each time in step S322, is repeated S322, S323 step operation Q times obtains Q sub-spaces, and contrast test sample and training sample one by one is each in subspace The Euclidean distance of class central point finds nearest central point to get to the classification of Q sub-spaces as a result, wherein T, Q are greater than 1.
5. intelligent input method according to claim 1, it is characterised in that:
In the S2. extraction step, human body appointed part is the back of the hand;
Vibrating sensor is piezoelectric ceramic vibration sensor;
In the S1. receiving step, carrying out noise reduction process to vibration signal includes:
Step S11 filters DC component and low-frequency noise using the butterworth high pass filter of 20Hz;
Step S12 filters high-frequency noise using 800Hz low-pass filtering.
6. a kind of intelligent input system propagated based on osteoacusis vibration signal characterized by comprising
Receiving module, smart machine vibrating sensor receives vibration signal, and carries out noise reduction process to vibration signal;
Extraction module is detected using double threshold end-point detection method and extracts the vibration signal generated due to tapping human body appointed part Segment;
Processing module extracts signal characteristic, based on RNM algorithm come category signal position.
7. intelligent input system according to claim 6, which is characterized in that include: in the extraction module
First extraction module, for high and low two thresholdings to be arranged to the signal of processing;
Second extraction module surmounts low threshold for the energy or zero-crossing rate when signal, primarily determines knocking starting point, and works as The energy or zero-crossing rate of signal break through high threshold, just determine the real starting point of knocking;When signal energy and zero-crossing rate simultaneously When lower than low threshold, signaling destination point is determined;
Third extraction module obtains the vibration signal segment by tapping the back of the hand generation for retaining the data of starting point to the end.
8. intelligent input system according to claim 6, which is characterized in that include: in the processing module
First processing module extracts mel-frequency cepstral for the dissection signal in vibration signal segment to be normalized Coefficient obtains signal characteristic;
Second processing module, for signal characteristic using the RNM algorithm based on stochastic subspace and nearest central point algorithm into Row classification, and then judge beating position;
The Second processing module includes:
First processing units acquire the signal characteristic that first processing module obtains and train for the difference according to beating position Sample, and classify;
The second processing unit, for using nearest central point algorithm to calculate every a kind of central point of training sample;
Third processing unit, for being based on stochastic subspace, the class central point of test sample and training sample is repeatedly in subspace Comparison, obtains multiple classification results;
Fourth processing unit for using simple majority Voting principle to classification results, and is arranged certain poll ratio, obtains Majority vote obtains final classification result in the case where reaching certain poll simultaneously;
5th processing unit, for that will classify, successful sample is included into new training sample, recalculates the new of new training sample Class central point.
9. intelligent input system according to claim 8, it is characterised in that:
In the first processing units, the back of the hand different location is tapped, the signal characteristic for collecting first processing module is made It is divided into n class for training sample, mark position and according to the difference of position, every class taps m times, and wherein n and m is greater than or equal to 1;
In the third processing unit, random sampling T category is distinguished at the class center generated each time in the second processing unit Property, the second processing unit, third processing unit operation Q times are repeated, Q sub-spaces are obtained, the contrast test one by one in subspace The Euclidean distance of each class central point of sample and training sample finds nearest central point to get the knot classified to Q sub-spaces Fruit, wherein T, Q are greater than 1.
10. intelligent input system according to claim 6, it is characterised in that:
In the extraction module, human body appointed part is the back of the hand;
Vibrating sensor is piezoelectric ceramic vibration sensor;
In the receiving module, carrying out noise reduction process to vibration signal includes:
First receiving module, for filtering DC component and low-frequency noise using the butterworth high pass filter of 20Hz;
Second receiving module, for filtering high-frequency noise using 800Hz low-pass filtering.
CN201710433231.5A 2017-06-09 2017-06-09 The intelligent input method and system propagated based on osteoacusis vibration signal Active CN107300971B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201710433231.5A CN107300971B (en) 2017-06-09 2017-06-09 The intelligent input method and system propagated based on osteoacusis vibration signal
PCT/CN2017/092769 WO2018223489A1 (en) 2017-06-09 2017-07-13 Intelligent input method and system based on bone conduction vibration signal propagation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710433231.5A CN107300971B (en) 2017-06-09 2017-06-09 The intelligent input method and system propagated based on osteoacusis vibration signal

Publications (2)

Publication Number Publication Date
CN107300971A CN107300971A (en) 2017-10-27
CN107300971B true CN107300971B (en) 2019-04-02

Family

ID=60134750

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710433231.5A Active CN107300971B (en) 2017-06-09 2017-06-09 The intelligent input method and system propagated based on osteoacusis vibration signal

Country Status (2)

Country Link
CN (1) CN107300971B (en)
WO (1) WO2018223489A1 (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108681709B (en) * 2018-05-16 2020-01-17 深圳大学 Intelligent input method and system based on bone conduction vibration and machine learning
US20210255147A1 (en) * 2018-06-22 2021-08-19 iNDTact GmbH Sensor arrangement, use of the sensor arrangement and method for detecting structure-borne noise
CN109840480B (en) * 2019-01-04 2021-08-13 深圳大学 Interaction method and interaction system of smart watch
CN109933202B (en) * 2019-03-20 2021-11-30 深圳大学 Intelligent input method and system based on bone conduction
CN110058689A (en) * 2019-04-08 2019-07-26 深圳大学 A kind of smart machine input method based on face's vibration
CN110414196B (en) * 2019-07-29 2021-09-17 深圳大学 Smart watch identity verification method based on vibration signal
CN110931031A (en) * 2019-10-09 2020-03-27 大象声科(深圳)科技有限公司 Deep learning voice extraction and noise reduction method fusing bone vibration sensor and microphone signals
CN111752388A (en) * 2020-06-19 2020-10-09 深圳振科智能科技有限公司 Application control method, device, equipment and storage medium
CN111741419B (en) * 2020-08-21 2020-12-04 瑶芯微电子科技(上海)有限公司 Bone conduction sound processing system, bone conduction microphone and signal processing method thereof
CN112272019A (en) * 2020-10-22 2021-01-26 广东美的制冷设备有限公司 Control method and device for sound-control knocking switch, household appliance and storage medium
CN113342159A (en) * 2021-05-07 2021-09-03 哈尔滨工业大学 Wrist wearable system identified through wrist vibration

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104461004A (en) * 2014-12-12 2015-03-25 北京奇虎科技有限公司 Wearable intelligent equipment
CN106339104A (en) * 2016-08-24 2017-01-18 广州市香港科大霍英东研究院 Text input method and device for smart watch

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2011115035A1 (en) * 2010-03-15 2011-09-22 日本電気株式会社 Input device, input method and program
CN102436345A (en) * 2011-10-31 2012-05-02 广州市动景计算机科技有限公司 Method for interface operation and control of mobile device and implement device
US8798292B2 (en) * 2012-06-06 2014-08-05 Google Inc. External vibration reduction in bone-conduction speaker
US9199098B2 (en) * 2012-09-05 2015-12-01 Olympus Corporation Ultrasonic treatment device
CN106128452A (en) * 2016-07-05 2016-11-16 深圳大学 Acoustical signal detection keyboard is utilized to tap the system and method for content

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104461004A (en) * 2014-12-12 2015-03-25 北京奇虎科技有限公司 Wearable intelligent equipment
CN106339104A (en) * 2016-08-24 2017-01-18 广州市香港科大霍英东研究院 Text input method and device for smart watch

Also Published As

Publication number Publication date
WO2018223489A1 (en) 2018-12-13
CN107300971A (en) 2017-10-27

Similar Documents

Publication Publication Date Title
CN107300971B (en) The intelligent input method and system propagated based on osteoacusis vibration signal
Zhang et al. Data augmentation and dense-LSTM for human activity recognition using WiFi signal
US10061389B2 (en) Gesture recognition system and gesture recognition method
CN106658590B (en) Design and implementation of multi-person indoor environment state monitoring system based on WiFi channel state information
Qifan et al. Dolphin: Ultrasonic-based gesture recognition on smartphone platform
CN106128452A (en) Acoustical signal detection keyboard is utilized to tap the system and method for content
Du et al. Wordrecorder: Accurate acoustic-based handwriting recognition using deep learning
WO2019218725A1 (en) Intelligent input method and system based on bone-conduction vibration and machine learning
CN106919958B (en) Human body finger action recognition method based on smart watch
CN109508728B (en) Novel identity authentication method for wearable equipment
CN105938399B (en) The text input recognition methods of smart machine based on acoustics
CN110176226A (en) A kind of speech recognition and speech recognition modeling training method and device
CN106899968A (en) A kind of active noncontact identity identifying method based on WiFi channel condition informations
CN106237604A (en) Wearable device and the method utilizing its monitoring kinestate
CN105956558B (en) One kind being based on 3-axis acceleration sensor human motion recognition method
CN103948398B (en) Be applicable to the heart sound location segmentation method of android system
CN110069199A (en) A kind of skin-type finger gesture recognition methods based on smartwatch
CN109271901A (en) A kind of sign Language Recognition Method based on Multi-source Information Fusion
CN110059612B (en) Position-independent gesture recognition method and system based on channel state information
Xu et al. Attention-based gait recognition and walking direction estimation in wi-fi networks
CN108182418A (en) A kind of thump recognition methods based on multidimensional acoustic characteristic
Zou et al. AcouDigits: Enabling users to input digits in the air
CN103070682A (en) Extraction method and extraction device for direction of eye movement based on electro-oculogram signals
CN107491254A (en) A kind of gesture operation method, device and mobile terminal
CN108108015A (en) A kind of action gesture recognition methods based on mobile phone gyroscope and dynamic time warping

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
GR01 Patent grant
GR01 Patent grant