CN110069199A - A kind of skin-type finger gesture recognition methods based on smartwatch - Google Patents
A kind of skin-type finger gesture recognition methods based on smartwatch Download PDFInfo
- Publication number
- CN110069199A CN110069199A CN201910248707.7A CN201910248707A CN110069199A CN 110069199 A CN110069199 A CN 110069199A CN 201910248707 A CN201910248707 A CN 201910248707A CN 110069199 A CN110069199 A CN 110069199A
- Authority
- CN
- China
- Prior art keywords
- gesture
- finger
- skin
- acoustic signal
- smartwatch
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input 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
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
- G06F3/0487—Interaction techniques based on graphical user interfaces [GUI] using specific features provided by the input device, e.g. functions controlled by the rotation of a mouse with dual sensing arrangements, or of the nature of the input device, e.g. tap gestures based on pressure sensed by a digitiser
Landscapes
- Engineering & Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- User Interface Of Digital Computer (AREA)
Abstract
The skin-type finger gesture recognition methods based on smartwatch that the invention discloses a kind of, comprising: step 1, the passive acoustic signal for the generation that rubs between finger and skin of dorsum of hand signal acquisition: is acquired by smartwatch;Step 2, data prediction: noise signal is removed by filtering processing, obtains acoustic signal to be detected;Step 3, gesture motion detects: being handled by gestures detection acoustic signal to be detected being divided into multiple independent gesture acoustic signals;Step 4, multiple independent gesture acoustic signals feature extraction: are converted into the time-frequency spectrum of gray level image expression acoustic signal and the image of mel cepstrum coefficients as characteristic value;Step 5, gesture motion identifies: using the characteristic value of each independent finger movement of extraction as input data, carrying out finger gesture action recognition to input data using convolutional neural networks model, obtains corresponding finger gesture movement.This method extends the input area of wearable device, without being equipped with other equipment, has simple, the good advantage of real-time.
Description
Technical field
The present invention relates to be based on smartwatch application field more particularly to a kind of skin-type finger hand based on smartwatch
Gesture recognition methods.
Background technique
Currently, thering are numerous studies to develop much new softwares based on smartwatch and proposing many new
Method is used for the interaction of smartwatch.Research before many is exactly that some special hardware or sensor are utilized.Than
Such as, Google develops a special chip in project Soli, utilizes 60GHz radar moving normal input hand
Make to replace;WatchIt provides the equipment prototype of an extension Intelligent bracelet input;Mole propose using in wrist-watch from
The motion sensor of band goes to speculate what content user is writing.
The skin of people can be used as the input face that can be used at any time, and also have many technical research this respect
Content, such as the hardware of novel skin-worn is devised in the existing scheme of public technology.Or utilize some differences
Ordinary technology, such as electric signal, sound, even optical projection etc..ISkin, which is proposed, has plated biofacies using one
The small sensor of capacitive metal is inputted on human body.SkinTrack reflects RF signal using a finger ring, and measures
Receive the phase offset of signal out to track finger.Skinput carries out the transmission of biological sound using the body of people, so that skin
Skin can become the input face for having one group of sensor on arm.SkinButton proposes small-sized projector being embedded into wrist-watch
In, portrait is projected on skin.
But in these above-mentioned existing methods using human skin input, exists and specific hardware or system is needed to constitute again
The problems such as miscellaneous.
Summary of the invention
Based on the problems of prior art, the object of the present invention is to provide a kind of skin-type hand based on smartwatch
Refer to gesture identification method, can realize handwriting input using the back of the hand of human body as input face using the smartwatch worn.
The purpose of the present invention is what is be achieved through the following technical solutions:
Embodiment of the present invention provides a kind of skin-type finger gesture recognition methods based on smartwatch, comprising:
Step 1, signal acquisition: finger passes through Mike's elegance of smartwatch when friction marks gesture on skin of dorsum of hand
Collect the passive acoustic signal for the generation that rubs between finger and skin of dorsum of hand;
Step 2, the noise in the passive acoustic signal that the step 1 acquires data prediction: is removed by filtering processing
Signal obtains acoustic signal to be detected;
Step 3, gesture motion detects: handling the sound to be detected that will be obtained after the step 2 pretreatment by gestures detection
It learns signal and is divided into multiple independent gesture acoustic signals;
Step 4, multiple independent gesture acoustic signals feature extraction: are converted into the acoustic signal indicated with gray level image
Characteristic value of the image of time-frequency spectrum and mel cepstrum coefficients as each independent finger movement;
Step 5, gesture motion identifies: using the characteristic value of each independent finger movement of the step 4 extraction as input
Data carry out finger gesture action recognition to input data using convolutional neural networks model, it is dynamic to obtain corresponding finger gesture
Make.
As seen from the above technical solution provided by the invention, the skin provided in an embodiment of the present invention based on smartwatch
Skin formula finger gesture recognition methods, it has the advantage that:
Finger is acquired in the sound of skin of dorsum of hand hand writing gesture stroke friction by the microphone of smartwatch, and to acquisition
Sound handled, therefrom identify the finger gesture of writing, wearable device can be carried out on skin of dorsum of hand by finger
It interactively enters, extends the input area of wearable device, without being equipped with other heavy equipments.With simple, real-time is good
The advantages of.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those skilled in the art, without creative efforts, it can also be obtained according to these attached drawings
His attached drawing.
Fig. 1 is the flow chart of the skin-type finger gesture recognition methods provided in an embodiment of the present invention based on smartwatch;
Fig. 2 is the gesture motion collection schematic diagram of recognition methods provided in an embodiment of the present invention;
Fig. 3 is the application system configuration diagram of recognition methods provided in an embodiment of the present invention;
Fig. 4 is the acoustic signal schematic diagram handled in recognition methods provided in an embodiment of the present invention, wherein (1) is this hair
The original signal schematic diagram handled in the recognition methods that bright embodiment provides;It (2) is recognition methods provided in an embodiment of the present invention
Signal schematic representation after the bandpass filtering of middle processing;
Fig. 5 is the acoustic signal schematic diagram in recognition methods provided in an embodiment of the present invention after wavelet transformation;
Fig. 6 is figure, frequency spectrum after the data processing of four kinds of standard gesture motions in recognition methods provided in an embodiment of the present invention
Figure and mel cepstrum figure.
Specific embodiment
Below with reference to particular content of the invention, technical solution in the embodiment of the present invention is clearly and completely retouched
It states, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on the present invention
Embodiment, every other embodiment obtained by those of ordinary skill in the art without making creative efforts,
Belong to protection scope of the present invention.The content being not described in detail in the embodiment of the present invention belongs to this field professional technique people
The prior art well known to member.
As described in Figure 1, the embodiment of the present invention provides a kind of skin-type finger gesture recognition methods based on smartwatch, packet
It includes:
Step 1, signal acquisition: finger passes through Mike's elegance of smartwatch when friction marks gesture on skin of dorsum of hand
Collect the passive acoustic signal for the generation that rubs between finger and skin of dorsum of hand;
Step 2, the noise in the passive acoustic signal that the step 1 acquires data prediction: is removed by filtering processing
Signal obtains acoustic signal to be detected;
Step 3, gesture motion detects: handling the sound to be detected that will be obtained after the step 2 pretreatment by gestures detection
It learns signal and is divided into multiple independent gesture acoustic signals;
Step 4, multiple independent gesture acoustic signals feature extraction: are converted into the acoustic signal indicated with gray level image
Characteristic value of the image of time-frequency spectrum and mel cepstrum coefficients as each independent finger movement;
Step 5, gesture motion identifies: using the characteristic value of each independent finger movement of the step 4 extraction as input
Data carry out finger gesture action recognition to input data using convolutional neural networks model, it is dynamic to obtain corresponding finger gesture
Make.
In the step 2 of the above method, filtering processing used includes:
Low frequency in passive acoustic signal collected is removed by FIR filter filtering (i.e. bandpass filtering) and high frequency is made an uproar
Sound obtains the acoustic signal to be detected only comprising finger gesture movement acoustic signal.
The low frequency and high-frequency noise in passive acoustic signal collected are removed above by FIR filter filtering are as follows: benefit
To be detected acoustic signal y of the filtered signal for using multiple FIR filter filtering outputs to obtain as removal out-of-band noise
[n], y [n] are as follows:
bk=bn+2-k, k=1,2 ..., n+1;
Wherein, N is the quantity of FIR filter, sets N=112;biIt is n-th layer of i-th of example in FIR filter
Frequency response values are the coefficient of this layer of FIR filter;The parameter of two FIR filters is respectively set to two passband cut-offs
Frequency is 6000 and 14000Hz and two stopband cutoff frequency is 5000 and 15000Hz, and the sample rate of original signal is Fs
=44100Hz.
In the step 3 of the above method, the acoustics to be detected that will be obtained after the step 2 pretreatment is handled by gestures detection
Signal is divided into multiple independent gesture acoustic signals
The beginning and end that each independent finger gesture is detected from acoustic signal to be detected, by each independent
The beginning and end of finger gesture extracts effective segment, confirms multiple independent gesture acoustic signals according to effective segment.
In the above method, the beginning and end of each independent finger gesture is detected from acoustic signal to be detected, is led to
Cross each independent finger gesture beginning and end extract effective segment mode it is as follows:
Acoustic signal y [n] to be detected is divided into multiple segment data by sliding window, calculates the short-term averaging energy of every segment data
AmountWherein, W is window size, and W is set as 882, and step-length is set as 750;
By judge average energy E [n] withBetween first difference valueIt whether is more than warp
Threshold gamma is tested, confirms the starting point n of finger gesture, i.e.,
In the two sides of the gesture input sound of estimation, two protection intervals are setWithIt calculatesObtain hand
Gesture inputs the candidate of starting point, and it is { n that candidate gesture, which plays point set,1,n2,...,nm};These points subtractAs starting point, in additionAs end point, i.e. Candidate Set is changed to
In the step 4 of the above method, multiple independent gesture acoustic signals are converted into gray level image to obtain finger movement
Characteristic value mode are as follows:
Time-frequency spectrum is calculated by carrying out Short Time Fourier Transform to each independent gesture acoustic signal, and to each independence
Gesture voice signal carry out mel cepstrum the image that mel cepstrum coefficients are calculated, Short Time Fourier Transform is obtained
Time-frequency spectrum and the image of mel cepstrum coefficients that is calculated combine after be converted to gray level image, pass through the ash
Degree image obtains the characteristic value of gesture acoustic signal;
Wherein, Short Time Fourier Transform STFT are as follows:
Wherein, w [t] is window function, and Y [m, ω] is the Fourier transformation of y [n] w [n-m], and being sized is 512
Hamming window, the FFT that length is 512, overlap length 256.
Wherein, the calculating of mel cepstrum coefficients are as follows:
Wherein, f is frequency, setpoint frequency 100Hz.
In the step 5 of the above method, convolutional neural networks model used are as follows:
Using the structure of LeNet as main structure, with the convolutional layer of AlexNet;
Including four convolutional layers and four pond layers, followed by two full articulamentums and an output layer;Wherein,
The size of convolution kernel is two 11 × 11,5 × 5 and 3 × 3, pond size is 3 × 3, stride 2.
In the convolutional neural networks model of above method step 5,
It willIt is added in the error function of the convolutional neural networks model, is used in the layer being fully connected;
Each layer of the convolutional neural networks model all uses degradation mechanism, and it is fixed that its probability value is set in training process
Value p=0.8.
Method of the invention by the way that the back of the hand is extended to input face, using finger skin of dorsum of hand slip gesture friction sound
Sound signal identifies the finger gesture of the input on the back of the hand, and it is smaller to can solve existing intelligent wearable device screen,
Touch screen power consumption, and input inconvenient problem.
The embodiment of the present invention is specifically described in further detail below.
The skin-type finger gesture recognition methods based on smartwatch of the embodiment of the present invention is in commercial smartwatch
On the basis of, a virtual handwriting input keyboard is constructed on the back of the hand, the identification to finger gesture realizes skin-type
Handwriting input.The recognition methods collects finger in the acoustic signal of the back of the hand sliding friction using the microphone that smartwatch carries;
Then the feature extraction for the acoustic signal being collected into is come out, as the input of machine learning, and then is realized to each movement
Identification.The places different from existing recognition methods are that the recognition methods carries out the input mode of intelligent wearable device
Extension, but do not need to wear on finger or equipment outside retained amount on hand.
Skin-type finger gesture recognition methods based on smartwatch proposed by the invention, using being embedded in intelligent wearing
Microphone in equipment (smartwatch) extracts the grating between finger and the back of the hand, and Lai Shixian identifies finger hand in real time
Gesture.The microphone that recognition methods of the invention is mainly based upon commercial intelligent wearable device can capture faint skin
Grating, therefore it is feasible that the back of the hand, which is used as the gesture input surface of extension,.In view of extension the small screen and few button can
The input of wearable device, the present invention execute more useful operations using the multi-finger gesture on the back of the hand.As shown in Fig. 2, can determine
Adopted 4 kinds of basic finger gestures, including slide to the left, it slides to the right, mediates and extend, these are user and intelligent wearable device
Common gesture when interaction.In order to realize more useful action controls, the gesture of these four types is extended to for two
Or more finger 12 pairs of finger gestures.Multi-finger gesture provides flexibility for intelligent wearable device developer, so as to from specific
Multi-finger gesture concentrate selection be most suitable for its application gesture.These gestures are only related with finger and hand motion, and not
It is related to body action.The unique motion of each finger gesture introduces the difference of acoustic signal, can use these differences to identify
These finger gestures, and then realize input face of the extension skin of dorsum of hand as intelligence wearing the back of the hand.
Detailed process is as follows for recognition methods of the present invention:
A. it firstly the need of the smartwatch or Intelligent bracelet that one piece of operation has recognition methods of the invention to apply is prepared, wears
On hand;
B. using wear wrist-watch hand the back of the hand as extension input face, finger gesture movement shown in Fig. 2 is carried out above;
C. the start button in smartwatch or Intelligent bracelet interface is clicked, sound-recording function is opened, is then done on the back of the hand
Finger gesture movement, smartwatch or Intelligent bracelet will provide corresponding reaction, that is, carry out corresponding operation.
Recognition methods of the invention is applied in acoustic sensing system, be can recognize finger gesture, is utilized commercial smart machine
In microphone realize human-computer interaction.Fig. 3 depicts the system architecture of the system, mainly includes five processing modules, it may be assumed that letter
Number acquisition, data prediction, gesture motion detection, feature extraction and gesture motion identification;Wherein,
In first signal acquisition module, smartwatch collect finger gesture issue passive acoustic signal, then by
In its limited computing capability, it sends smart phone or other calculating treatmenting equipments (such as the signal of record by bluetooth
Tablet computer, computer etc.) to be further processed.
In second data preprocessing module, ring is minimized by the way that bandpass filter is applied to original audio signal
The interference of border noise;
Then in gesture motion detection module use gesture detecting method, with extract it is pretreated after voice signal
It is middle that there are the parts of gesture motion;
In characteristic extracting module, time-frequency spectrum and mel cepstrum coefficients figure are converted acoustical signals into, as each only
The feature of vertical finger movement;
In the last one gesture motion identification module, finger gesture motion is carried out using convolutional neural networks (CNN)
Identification, wherein time-frequency spectrum and mel cepstrum coefficients are converted into input of the visual pattern as CNN.Finally, smartwatch or
Smart phone calls the corresponding function being each used alone to interact with user based on the output of CNN.
When recognition methods work, smartwatch can continue to monitor the gesture input on the back of the hand and pass the sound of recording
It is defeated to arrive smart phone.The sample rate for being embedded in the microphone in smartwatch is 44100Hz, it is sufficient to collect the sound of surrounding.Figure
The single finger gesture of 5 display right hand slidings.
The original sound of commercial microphones capture inherently carry certain noise, and ambient enviroment usually have it is different
Noise level, therefore using wavelet time-frequency analysis to obtain the frequency range for generating sound by finger and skin of dorsum of hand friction.It is logical
It crosses a series of filters and calculates wavelet transform (DWT) to analyze the reception acoustical signal x [n] with time-varying length n, i.e.,
Wherein, xα,L[n] and xα,H[n] is the output of low-pass filter g and high-pass filter h respectively.Such as Fig. 4 (1) and figure
Shown in 4 (2), nearby there is a highlighted vertical line at 0.4 second, finger gesture has occurred in expression.Bright line occupies 500Hz
To the frequency of 20000Hz, the normal noise of this frequency is unable to reach.
In order to optimize the audio signal for finger gesture identification, system proposed by the present invention is first filtered them by band logical
Wave device is to remove low frequency and high-frequency noise.FIR filter is a kind of naturally selection, inherently stable, can be designed
At generation linear frequency response.The output signal of FIR filter are as follows:
bk=bn+2-k, k=1,2 ..., n+1;
Wherein, N is the quantity of FIR filter, biIt is frequency response values of i-th of example in the n-th layer of FIR filter,
It that is to say the coefficient of this layer of FIR filter.Therefore, the out-of-band interference of voice signal is eliminated using filter.As previously mentioned, by
The frequency of the variation of the voice signal caused by skin friction is usually within 5000~15000Hz.By two FIR filters
It is 6000 and 14000Hz that parameter, which is respectively set to two cut-off frequecy of passband, and two stopband cutoff frequencies are 5000 Hes
15000Hz, wherein the sample rate of original sound signal is Fs=44100Hz.Rule of thumb setting N=112 is required to obtain
Denoising result.It is obvious that FIR filter almost eliminates shown in out-of-band noise such as Fig. 4 (2), Fig. 4 (1) is original sound signal
Schematic diagram.
Skin-type input system proposed by the invention is that the method that each finger gesture extracts audio signal is based on time domain
In signal processing.Grating between finger and the back of the hand is rising edge or decline on the main influence of received voice signal
Edge.These variations are vital for detection finger gesture, and using the uniqueness of different changing patteries come to finger
Movement is classified.For the starting point of detection gesture, effective segment is extracted from the acoustic signal of processing.
By the inspiration of constant false alarm rate (CFAR), a kind of method of the beginning and end of detection input gesture is proposed.This hair
Voice signal y [n] is divided into sectional data by sliding window by bright mentioned system.The every segment data of the system-computed
Short-term averaging energyWherein W is window size.882 are set by window size, step-length setting
It is 750, i.e., each segment includes the audio signal of 0.02s, sample rate 44100Hz.Since signal is via bandpass filter
Processing, if short-term averaging energy is slightly different with neighbouring part without gesture.However, being put down when gesture input occurs
The outburst suddenly of equal ENERGY E [n] and its first differenceIt becomes much larger.In addition, gesture input
Appearance cause E [n] andBetween difference it is bigger, it is contemplated that can be more than empirical value γ.Then, it is indicated that the starting of finger gesture
Point n, i.e.,
During gestures detection, which is arranged two protection intervals in the two sides of the gesture input sound of estimationWithThe system-computedTo obtain the candidate of gesture input starting point.For example, candidate gesture point set is
{n1,n2,...,nm}.These points should subtractAs starting point, in additionAs end point, i.e. Candidate Set is changed toThe purpose of the operation is preferably to extract complete gesture input letter
Number.
After detecting gesture input, which obtains effective voice signal of each gesture.Fig. 6 the first row, which is shown, to be mentioned
The gesture taken is slided to the left, is slided to the right, the voice signal mediated and extended.Due to only be used only time-domain signal be it is inadequate,
Therefore the system is based primarily upon time frequency analysis and extracts feature.
However, classical Fourier transformation is not providing resolution ratio over time and frequency.On the contrary, using Fourier in short-term
This limitation can be improved by the way that long period signal is divided into shorter section by converting (STFT).STFT is defined by following formula:
Wherein w [t] is window function, and Y [m, ω] is substantially the Fourier transformation of y [n] w [n-m].The effect of STFT, is depended on
In selection variable appropriate.Within the system, the Hamming window that a size is 512, the FFT that length is 512, weight are set
Folded length is 256.The second row of Fig. 6 is drawn gesture and is slided to the left, slides to the right, the sound time-frequency spectrum for sliding to the right and mediating.
Then, 12 MFCC features are extracted.Mel cepstrum image is shown in Fig. 6 the third line.In gesture extraction process, the system
Simultaneously combination S TFT and MFCC coefficient is calculated, gray level image is then converted into.
Skin-type input system proposed by the invention identifies different gesture inputs using CNN model, passes through calculating
STFT and MFCC coefficient simultaneously converts the result to input of the gray level image as CNN model, thus obtain image appropriate for
Guarantee that the performance of CNN classification is very important, the present invention devises the above method to obtain input picture.
In recognition methods of the invention, one is designed with reference to two popular CNN structures (i.e. LeNet-5 and AlexNet)
It is a to be suitble to the CNN structure run on the mobile apparatus as CNN model of the invention.CNN models coupling LeNet-5 and
The advantages of CNN structure of AlexNet.The CNN model is specifically: selecting the structure of LeNet as main structure, and utilizes
The convolutional layer of AlexNet.Using four convolutional layers and four pond layers, followed by two full articulamentums and an output layer.Volume
The size of product core is two 11 × 11,5 × 5 and 3 × 3, pond size is 3 × 3, stride 2.Meanwhile in identification side of the invention
The regularization method and degradation mechanism for being usually used in handling over-fitting are introduced in method.L2 regularization can pass through byAddition
Error function into neural network is realized, is only used in the layer being fully connected.We all employ degeneration in each layer
Mechanism sets its probability as a fixed value p=0.8 in the training process.
Those of ordinary skill in the art will appreciate that: realizing that all or part of the process in above-described embodiment method is can be with
Relevant hardware is instructed to complete by program, the program can be stored in a computer-readable storage medium, should
Program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic disk, light
Disk, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..Method i.e. of the invention can be operated in smartwatch, bracelet and mobile phone in a manner of application program.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Within the technical scope of the present disclosure, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the protection model of claims
Subject to enclosing.
Claims (8)
1. a kind of skin-type finger gesture recognition methods based on smartwatch characterized by comprising
Step 1, signal acquisition: finger acquires hand when friction marks gesture on skin of dorsum of hand, through the microphone of smartwatch
Refer to the passive acoustic signal for the generation that rubs between skin of dorsum of hand;
Step 2, data prediction: removing the noise signal in the passive acoustic signal that the step 1 acquires by filtering processing,
Obtain acoustic signal to be detected;
Step 3, gesture motion detects: being handled by gestures detection and believes the acoustics to be detected obtained after the step 2 pretreatment
Number it is divided into multiple independent gesture acoustic signals;
Step 4, multiple independent gesture acoustic signals feature extraction: are converted to the time-frequency of the acoustic signal indicated with gray level image
Characteristic value of the image of spectrogram and mel cepstrum coefficients as each independent finger movement;
Step 5, gesture motion identifies: using the characteristic value of each independent finger movement of the step 4 extraction as input data,
Finger gesture action recognition is carried out to input data using convolutional neural networks model, obtains corresponding finger gesture movement.
2. the skin-type finger gesture recognition methods according to claim 1 based on smartwatch, which is characterized in that described
In the step 2 of method, filtering processing used includes:
Low frequency and high-frequency noise in passive acoustic signal collected are removed by FIR filter filtering, is obtained only comprising hand
Refer to the acoustic signal to be detected of gesture motion acoustic signal.
3. the skin-type finger gesture recognition methods according to claim 2 based on smartwatch, which is characterized in that described
In method, the low frequency and high-frequency noise in passive acoustic signal collected are eliminated by FIR filter filtering are as follows:
Believed using the filtered signal that the filtering output of multiple FIR filters obtains as the acoustics to be detected of removal out-of-band noise
Number y [n], y [n] are as follows:
bk=bn+2-k, k=1,2 ..., n+1;
Wherein, N is the quantity of FIR filter, sets N=112;biIt is that i-th of example is rung in the frequency of the n-th layer of FIR filter
It should be worth, be the coefficient of this layer of FIR filter;The parameter of two FIR filters is respectively set to two cut-off frequecy of passband
6000 and 14000Hz and two stopband cutoff frequency be 5000 and 15000Hz, and the sample rate of original signal is Fs=
44100Hz。
4. the skin-type finger gesture recognition methods according to any one of claims 1 to 3 based on smartwatch, feature
It is, in the step 3 of the method, is handled by gestures detection and believe the acoustics to be detected obtained after the step 2 pretreatment
Number being divided into multiple independent gesture acoustic signals includes:
The beginning and end that each independent finger gesture is detected from acoustic signal to be detected passes through each independent finger
The beginning and end of gesture extracts effective segment, confirms multiple independent gesture acoustic signals according to effective segment.
5. the skin-type finger gesture recognition methods according to claim 4 based on smartwatch, which is characterized in that described
In method, the beginning and end of each independent finger gesture is detected from acoustic signal to be detected, by each independent
The mode that the beginning and end of finger gesture extracts effective segment is as follows:
Acoustic signal y [n] to be detected is divided into multiple segment data by sliding window, calculates the short-term averaging energy of every segment dataWherein, W is window size, and W is set as 882, and step-length is set as 750;
By judge average energy E [n] withBetween first difference valueIt whether is more than experience threshold
Value γ confirms the starting point n of finger gesture, i.e.,
In the two sides of the gesture input sound of estimation, two protection intervals are setWithIt calculatesIt is defeated to obtain gesture
Enter the candidate of starting point, it is { n that candidate gesture, which plays point set,1,n2,...,nm};These points subtractAs starting point, in additionMake
For end point, i.e. Candidate Set is changed to
6. the skin-type finger gesture recognition methods according to any one of claims 1 to 3 based on smartwatch, feature
It is, in the step 4 of the method, multiple independent gesture acoustic signals is converted into gray level image to obtain the spy of finger movement
The mode of value indicative are as follows:
Time-frequency spectrum is calculated by carrying out Short Time Fourier Transform to each independent gesture acoustic signal, and to each independent hand
Gesture voice signal carries out the image that mel cepstrum coefficients are calculated of mel cepstrum, the time-frequency that Short Time Fourier Transform is obtained
The image of spectrogram and the mel cepstrum coefficients being calculated is converted to gray level image after combining, and is obtained by the gray level image
To the characteristic value of gesture acoustic signal;
Wherein, Short Time Fourier Transform STFT are as follows:
Wherein, w [t] is window function, and Y [m, ω] is the Fourier transformation of y [n] w [n-m], is sized the Hamming for 512
Window, the FFT that length is 512, overlap length 256.
7. the skin-type finger gesture recognition methods according to any one of claims 1 to 3 based on smartwatch, feature
It is, in the step 5 of the method, convolutional neural networks model used are as follows:
Using the structure of LeNet as main structure, with the convolutional layer of AlexNet;
Including four convolutional layers and four pond layers, followed by two full articulamentums and an output layer;Wherein,
The size of convolution kernel is two 11 × 11,5 × 5 and 3 × 3, pond size is 3 × 3, stride 2.
8. the skin-type finger gesture recognition methods according to claim 7 based on smartwatch, which is characterized in that described
In the convolutional neural networks model of the step 5 of method,
It willIt is added in the error function of the convolutional neural networks model, is used in the layer being fully connected;
Each layer of the convolutional neural networks model all uses degradation mechanism, and probability is set in training process as a fixed value
P=0.8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910248707.7A CN110069199B (en) | 2019-03-29 | 2019-03-29 | Skin type finger gesture recognition method based on smart watch |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910248707.7A CN110069199B (en) | 2019-03-29 | 2019-03-29 | Skin type finger gesture recognition method based on smart watch |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110069199A true CN110069199A (en) | 2019-07-30 |
CN110069199B CN110069199B (en) | 2022-01-11 |
Family
ID=67366749
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910248707.7A Active CN110069199B (en) | 2019-03-29 | 2019-03-29 | Skin type finger gesture recognition method based on smart watch |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110069199B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110751105A (en) * | 2019-10-22 | 2020-02-04 | 珠海格力电器股份有限公司 | Finger image acquisition method and device and storage medium |
CN110784788A (en) * | 2019-09-18 | 2020-02-11 | 广东思派康电子科技有限公司 | Gesture recognition method based on microphone |
CN111158487A (en) * | 2019-12-31 | 2020-05-15 | 清华大学 | Man-machine interaction method for interacting with intelligent terminal by using wireless earphone |
CN111929689A (en) * | 2020-07-22 | 2020-11-13 | 杭州电子科技大学 | Object imaging method based on sensor of mobile phone |
CN112364779A (en) * | 2020-11-12 | 2021-02-12 | 中国电子科技集团公司第五十四研究所 | Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion |
CN112966662A (en) * | 2021-03-31 | 2021-06-15 | 安徽大学 | Short-range capacitive dynamic gesture recognition system and method |
CN113126764A (en) * | 2021-04-22 | 2021-07-16 | 中国水利水电科学研究院 | Personal water volume detection method based on smart watch |
CN113197569A (en) * | 2021-04-23 | 2021-08-03 | 华中科技大学 | Human body intention recognition sensor based on friction power generation and recognition method thereof |
CN113849068A (en) * | 2021-09-28 | 2021-12-28 | 中国科学技术大学 | Gesture multi-mode information fusion understanding and interacting method and system |
US20240037529A1 (en) * | 2022-07-27 | 2024-02-01 | Bank Of America Corporation | System and methods for detecting and implementing resource allocation in an electronic network based on non-contact instructions |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103885744A (en) * | 2013-05-30 | 2014-06-25 | 美声克(成都)科技有限公司 | Sound based gesture recognition method |
US20140296935A1 (en) * | 2013-03-29 | 2014-10-02 | Neurometrix, Inc. | Transcutaneous electrical nerve stimulator with user gesture detector and electrode-skin contact detector, with transient motion detector for increasing the accuracy of the same |
CN106095203A (en) * | 2016-07-21 | 2016-11-09 | 范小刚 | Sensing touches the calculating Apparatus and method for that sound inputs as user's gesture |
CN106919958A (en) * | 2017-03-21 | 2017-07-04 | 电子科技大学 | A kind of human finger action identification method based on intelligent watch |
-
2019
- 2019-03-29 CN CN201910248707.7A patent/CN110069199B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140296935A1 (en) * | 2013-03-29 | 2014-10-02 | Neurometrix, Inc. | Transcutaneous electrical nerve stimulator with user gesture detector and electrode-skin contact detector, with transient motion detector for increasing the accuracy of the same |
CN103885744A (en) * | 2013-05-30 | 2014-06-25 | 美声克(成都)科技有限公司 | Sound based gesture recognition method |
CN106095203A (en) * | 2016-07-21 | 2016-11-09 | 范小刚 | Sensing touches the calculating Apparatus and method for that sound inputs as user's gesture |
CN106919958A (en) * | 2017-03-21 | 2017-07-04 | 电子科技大学 | A kind of human finger action identification method based on intelligent watch |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110784788A (en) * | 2019-09-18 | 2020-02-11 | 广东思派康电子科技有限公司 | Gesture recognition method based on microphone |
CN110751105B (en) * | 2019-10-22 | 2022-04-08 | 珠海格力电器股份有限公司 | Finger image acquisition method and device and storage medium |
CN110751105A (en) * | 2019-10-22 | 2020-02-04 | 珠海格力电器股份有限公司 | Finger image acquisition method and device and storage medium |
CN111158487A (en) * | 2019-12-31 | 2020-05-15 | 清华大学 | Man-machine interaction method for interacting with intelligent terminal by using wireless earphone |
CN111929689A (en) * | 2020-07-22 | 2020-11-13 | 杭州电子科技大学 | Object imaging method based on sensor of mobile phone |
CN111929689B (en) * | 2020-07-22 | 2023-04-07 | 杭州电子科技大学 | Object imaging method based on sensor of mobile phone |
CN112364779A (en) * | 2020-11-12 | 2021-02-12 | 中国电子科技集团公司第五十四研究所 | Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion |
CN112966662A (en) * | 2021-03-31 | 2021-06-15 | 安徽大学 | Short-range capacitive dynamic gesture recognition system and method |
CN113126764A (en) * | 2021-04-22 | 2021-07-16 | 中国水利水电科学研究院 | Personal water volume detection method based on smart watch |
CN113197569A (en) * | 2021-04-23 | 2021-08-03 | 华中科技大学 | Human body intention recognition sensor based on friction power generation and recognition method thereof |
CN113849068A (en) * | 2021-09-28 | 2021-12-28 | 中国科学技术大学 | Gesture multi-mode information fusion understanding and interacting method and system |
CN113849068B (en) * | 2021-09-28 | 2024-03-29 | 中国科学技术大学 | Understanding and interaction method and system for multi-modal information fusion of gestures |
US20240037529A1 (en) * | 2022-07-27 | 2024-02-01 | Bank Of America Corporation | System and methods for detecting and implementing resource allocation in an electronic network based on non-contact instructions |
US11983691B2 (en) * | 2022-07-27 | 2024-05-14 | Bank Of America Corporation | System and methods for detecting and implementing resource allocation in an electronic network based on non-contact instructions |
Also Published As
Publication number | Publication date |
---|---|
CN110069199B (en) | 2022-01-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110069199A (en) | A kind of skin-type finger gesture recognition methods based on smartwatch | |
Mouawad et al. | Robust detection of COVID-19 in cough sounds: using recurrence dynamics and variable Markov model | |
Bhat et al. | A real-time convolutional neural network based speech enhancement for hearing impaired listeners using smartphone | |
CN101599127B (en) | Method for extracting and identifying characteristics of electro-ocular signal | |
Zhao et al. | Towards low-cost sign language gesture recognition leveraging wearables | |
CN106919958B (en) | Human body finger action recognition method based on smart watch | |
WO2017152531A1 (en) | Ultrasonic wave-based air gesture recognition method and system | |
CN113349752B (en) | Wearable device real-time heart rate monitoring method based on sensing fusion | |
CN107928673A (en) | Acoustic signal processing method, device, storage medium and computer equipment | |
CN103294199B (en) | A kind of unvoiced information identifying system based on face's muscle signals | |
CN103413113A (en) | Intelligent emotional interaction method for service robot | |
JP2003255993A (en) | System, method, and program for speech recognition, and system, method, and program for speech synthesis | |
CN107300971A (en) | The intelligent input method and system propagated based on osteoacusis vibration signal | |
Kim et al. | Finger language recognition based on ensemble artificial neural network learning using armband EMG sensors | |
CN111643098A (en) | Gait recognition and emotion perception method and system based on intelligent acoustic equipment | |
WO2011092549A1 (en) | Method and apparatus for assigning a feature class value | |
EP4098182A1 (en) | Machine-learning based gesture recognition with framework for adding user-customized gestures | |
WO2017036147A1 (en) | Bioelectricity-based control method, device and controller | |
CN114707562A (en) | Electromyographic signal sampling frequency control method and device and storage medium | |
Zakaria et al. | VGG16, ResNet-50, and GoogLeNet deep learning architecture for breathing sound classification: a comparative study | |
Lu et al. | Detection of smoking events from confounding activities of daily living | |
CN107495939A (en) | Live biometric monitoring method, device and system | |
Casaseca-de-la-Higuera et al. | Effect of downsampling and compressive sensing on audio-based continuous cough monitoring | |
CN106323330A (en) | Non-contact-type step count method based on WiFi motion recognition system | |
CN116027911B (en) | Non-contact handwriting input recognition method based on audio signal |
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 |