• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    An interaction method and an interactive system of a smart watch The present invention describes a method of interacting with a smart watch comprising the following steps: S1. transmitting vibration signals based on a human body, and collecting the vibration signals of an accelerometer and a gyroscope of a smart watch; S2. identification of vibration signals with an anomaly detection algorithm; S3. preprocessing the vibration signals and applying an improved K nearest neighbor algorithm to classify and identify the vibration signals; S4. analyzes user feedback, and applies timely adjustment to maintain identification accuracy. Further, the invention describes an interaction system of a smart watch, comprising a signal detection module, an identification and classification module, and a real-time feedback module. The vibration signals are transmitted according to the human body, the body parts are used as a virtual screen, and the improved machine learning algorithm is combined, which expands the ways of interaction of watches and improves the experience of l 'user. The interaction method of the present invention is new and interesting, and capable of meeting user demands and being widely used in text input and video games on watches, ect. (Fig. 1)
    公开号:FR3091606A1
    申请号:FR1904179
    申请日:2019-04-19
    公开日:2020-07-10
    发明作者:Kaishun Wu；Wenqiang CHEN；Lu Wang；Minghui Qiu
    申请人:Shenzhen University；
    IPC主号:
    专利说明:

    Description
    Title of the invention: An interaction method and an interactive system of a smart watch
    Technical area
    The invention relates to the field of interaction mode of a smart device, in particular an interaction method and an interaction system based on a smart watch.
    Prior art
    [0002] Currently, portable intelligent detection devices are developing rapidly, in particular smart watches. However, since smart watches are worn on the wrist without being able to have a sufficiently large screen, people cannot enter as they do on mobile phones. There are three main modes for entering smart watches, namely single touch, finger tracking and voice recognition. The simple touch and finger tracking method is limited by the screen, and voice recognition is even more limited due to the sensitivity of the information. In order to remedy the difficulty of entering smart watches, numerous research teams have also carried out research. Most of the time, additional peripherals are required for this purpose and, due to the cost of purchasing and learning additional peripherals, they are not widely accepted.
    Statement of the invention
    Given the technical problems mentioned above, the present invention provides an interaction method and an interactive system of a smart watch transmitting a vibration signal based on the human body. Under the condition of adapting the usage habits of users, a new means of interaction for a smart watch is developed, which solves the problem of the lack of means of interaction of a smart watch. The invention adopts the following technical solutions:
    A method of interacting with a smart watch, characterized in that it comprises the following steps:
    SI: transmit vibration signals based on a human body, and collect vibration signals from an accelerometer and a gyroscope from a smart watch;
    S2: identification of the vibration signals with an anomaly detection algorithm;
    S3: preprocessing of vibration signals and application of an improved algorithm by the method of K nearest neighbors to classify and identify the vibration signals;
    S4: analyzes the feedback from a user, and application of timely adjustment to maintain identification accuracy.
    In addition, the vibration signals on the X, Y and Z axes of the accelerometer and the gyroscope are respectively collected.
    In addition, with application of an anomaly detection algorithm, step S2 of identification of the vibration signals comprises:
    S21: collecting data from the Z axis of the accelerometer;
    S22: filtering the data of the Z axis of the accelerometer in application of a high-pass filter;
    S23: setting of an effective tap signal threshold value and a noise signal threshold value;
    S24: reading of a signal segment whose amplitude is less than the noise signal threshold value as a first state;
    S25: continuation of monitoring, waiting to read a signal whose amplitude is greater than the threshold value of actual tap signal, the position whose amplitude is greater than the threshold value of actual tap signal being X, and the signal start position being fixed at (XL), position of L before position X;
    S26: continuation of monitoring while waiting to read a continuous signal whose amplitude is less than the noise signal threshold value, when the continuous amplitude is less than the noise signal threshold value, defines the final position of the signal a current position;
    S27: acquisition of the signal data by the signal start and stop position, determining whether the signal length respects a length range, if not, return to step S25, if yes, go to next step;
    S28: application of high-pass filtering on the data, respectively calculation of the energies of the first m signal points and the energies of the signal points after the m points, determining whether the signal is greater than the threshold value of a noise signal, if yes, determining the signal as an effective signal, otherwise, determining the signal as a noise signal and returning to S25.
    In addition, step S3 specifically includes:
    S31: preprocessing the signal with normalization, deducting an average value of the signal and dividing it by a variance;
    S32: during a step of initializing a learning model, storing the data processed in step S31 as a learning sample in a database; in the real use phase, application of the algorithm improved by the method of K nearest neighbors to classify and identify the vibration signals.
    In addition, the algorithm improved by the method of K nearest neighbors is specifically: based on the dynamic time distortion algorithm, the real signal and the training signal are paired in frame units and the distance from The shortest Manhattan between them is calculated and applied as the basis for the classification and recognition of the method of K nearest neighbors.
    In addition, step S4 includes:
    S41: after collecting the classification result obtained in step 3, correcting the result of user input;
    S42: after the correction, makes a correction to a certain extent on the training sample, thereby maintaining the stability of the precision.
    In addition, step S41 corrects the actual entry by providing a candidate key or by a correlation result in the entry method.
    In addition, step S42 specifically includes:
    S421: if the result of the correction is consistent with the result of the classification, performing no operation;
    S422: if the result of the correction is incompatible with the result of the classification, for the training sample of the same category as the result of the classification, delete from the sample with the greatest calculated distance by the algorithm improved by the method K nearest neighbors, then replace the deleted sample by the current sample.An interactive system for a smart watch, comprising:
    A signal detection module, transmitting vibration signals based on the human body, and collecting the vibration signals from an accelerometer and a gyroscope from a smart watch;
    A classification identification module which applies an anomaly detection algorithm to identify the vibration signals, preprocesses the vibration signals and applies an algorithm improved by the method of K nearest neighbors to classify and identify the signals of vibration;
    A real-time feedback module that analyzes the feedback from a user, and applies adjustments at the right time to maintain identification accuracy.
    Program for executing the interaction process of said smart watch.
    Compared to the prior art, the advantages of the present invention are as follows:
    vibration signals are transmitted according to the human body, parts of the body (like the back of the hand) are used as a virtual screen and the modified machine learning algorithm is combined, which widens the means of interaction of watches and improves the user experience. The interaction method of the present invention is new and interesting, and capable of responding to user requests and of being widely used in text entry and video games on watches, ect.
    Brief description of the drawings
    Figure 1 is a flowchart of the interaction method of the present invention;
    Figure 2 is a flowchart showing the operation of a signal detection module of the present invention;
    Figure 3 is a flow chart showing the operation of a return system of the present invention;
    Figure 4 is a structural diagram of the present invention;
    Figure 5 is the corresponding signal result of the original dynamic time distortion algorithm of the present invention;
    FIG. 6 is the corresponding signal result of the dynamic time distortion algorithm executed in frame units after the improvement of the present invention (the frame offset is equal to 1 and the frame length is equal to 3);
    Detailed description of exemplary embodiments of the invention
    The present invention will be described in more detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described here are merely illustrative of the invention and are not intended to limit the present invention.
    The preferred embodiment of the present invention will be described in more detail below with reference to the accompanying drawings.
    The invention relates to an interaction method and an interactive system of a smart watch based on vibration signals transmitted by the human body and machine learning. As described in FIG. 1, the interaction method of the present invention comprises the following steps:
    IF: transmitting vibration signals based on a human body, the vibration signals, the program controls the accelerometer and the gyroscope of a smart watch to collect the vibration signals from the accelerometer and the gyroscope a smart watch.
    Specifically, the vibration signals of the X, Y and Z axes of the accelerometer and the gyroscope are respectively collected;
    S2: identification of the vibration signals with an anomaly detection algorithm;
    S3: pretreatment of the vibration signals and application of an improved algorithm by the method of K nearest neighbors to classify and identify the vibration signals;
    S4: analyzes the feedback from a user, and application of timely adjustment to maintain identification accuracy.
    As shown in Figure 2, step S2 applies the anomaly detection algorithm to identify the signal. The specific steps are as follows:
    First, the data on the Z axis of the accelerometer are collected, the 40hz high-pass signal is selected, and a signal of amplitude lower than the threshold value of the noise signal is read at the same time, a signal length is preferably 10 signal points, and the threshold value of the noise signal is 0.015, at this time, a first signal detection state is obtained; continuation of monitoring, waiting to read a signal whose amplitude is greater than the actual tap signal threshold value, the position whose amplitude is greater than the actual tap signal threshold value being X, and the signal start position being fixed at (XL), position of L before position X, and the threshold value of the actual tap signal is preferably 2. After receiving the target signal, continue monitoring while waiting to read a continuous signal whose amplitude is less than the noise signal threshold value, when the continuous amplitude is less than the noise signal threshold value, defines the final position of the signal a current position, and preferably the signal length is 10 points and the threshold value of the noise signal is 0.015. After detecting the signal segment obtained between the threshold value of the noise signal and the threshold value of the actual tap signal, the signal length and the signal to noise ratio are limited. Preferably, the length of the signal L satisfies 37 <L <60. When the length of the signal satisfies the constraint condition, calculate respectively the energies of the m first signal points and the energies of the signal points after the m points, determine if the signal is greater than the threshold value of a noise signal, if so determining the signal as an effective signal, otherwise determining the signal as a noise signal and the signal ratio threshold value on noise is S10, so far the signal is detected.
    In this embodiment, the vibration signals is preprocessed in step S3 and is further classified and identified by the algorithm improved by the method of K nearest neighbors. The specific steps are as follows:
    First of all, the vibration signals on the three axes X, Y and Z of each sample of the accelerometer and gyroscopes are spliced by type of sensor and the global normalization processing is performed on the data at 3 axes of the corresponding sensor. Specifically, an average value is subtracted from the data and divided by a variance of the data. Then, during a step of initializing a training model, storing the processed data as a training sample in a database; in the phase of real use, application of the algorithm improved by the method of K nearest neighbors to classify and identify the vibration signals, in particular on the basis of the algorithm of the dynamic temporal deformation, calculates the distance between the sample to be tested / the input and the training sample, a classification result is obtained according to the distance.
    Among these, the algorithm of dynamic time distortion is based on the principle of dynamic programming. The object of dynamic temporal distortion is extended from the original one-dimensional point to the three-dimensional frame (three axes) and the distance between them is calculated, which allows to measure more precisely the degree of similarity between the two signals. The algorithm allows at the same time to adjust the length and the offset of the frame according to the actual sampling frequency and the demand, in order to reduce the energy consumption of the algorithm, thus offering the desired performances . The problem that two signals cannot be compared due to the alignment offset is resolved and the difference between the two is quantified. The distance is not limited to the distance from Manhattan or the distance from Euler.
    FIG. 5 is a result of signal pairing of the original dynamic time distortion algorithm; FIG. 6 is a result of pairing of the algorithm of the dynamic temporal distortion in units of frame after the improvement of the present invention (the frame offset is equal to 1 and the frame length is equal to 3); It can be seen that after the increase in the frame length constraint and the frame offset in the dynamic time distortion algorithm, the signal matching mode was changed.
    As shown in Figure 3, in step S4 of the embodiment, after obtaining the result of the classification in step S2, the result is sent to the application, and at the same time, the new sample X and the distance of the training sample obtained in the algorithm of step S3 are recorded and the return of the application is monitored. After receiving feedback on the classification result, an operation is performed on the training sample in accordance with the predetermined sample replacement strategy, thereby obtaining superior robustness. More precisely, after having collected the classification result obtained in step 3, the user input result is corrected and the actual input is corrected by providing the candidate key or by the correlation result in the input method; after the correction, the training sample is corrected to some extent, therefore, maintaining the stability of the accuracy. More precisely, if the result of the correction is consistent with the result of the classification, performing no operation; if the result of the correction is incompatible with the result of the classification, for the training sample of the same category as the result of the classification, delete from the sample with the greatest distance calculated by the improved algorithm by the K nearest neighbor method, then replace the deleted sample with the current sample.
    As shown in FIG. 4, the structure of the specific implementation of this example comprises three modules, namely, a signal detection module, an identification and classification module, and a feedback module. real time. The signal detection module detects the signal, then normalizes it, the average value is subtracted from the data and divided by the variance of the data, used as input to the identification and classification module; the learning phase (initialization) of the identification and classification module is a simple signal storage operation, which is implemented after the training is completed, will implement the improved classification algorithm. The classification results will be transmitted to the feedback module in real time.
    The above descriptions are detailed illustrations of the present invention in combination with specific / preferred embodiments, and it should not be understood that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several variants or modifications can be made to the embodiments described without departing from the inventive concept, which must all be considered to fall within the scope of protection of the present invention.
    权利要求:
    Claims (1)
    [1" id="c-fr-0001]
    Claims [Claim 1] A method of interacting with a smart watch, characterized in that it comprises the following steps:SI: transmit vibration signals based on a human body, and collect vibration signals from an accelerometer and a gyroscope from a smart watch;S2: identification of vibration signals with an anomaly detection algorithm;S3: preprocessing of vibration signals and application of an improved algorithm by the method of K nearest neighbors to classify and identify the vibration signals;S4: analyzes user feedback, and application of timely adjustment to maintain identification accuracy. [Claim 2] Method for interacting with a smart watch according to claim 1, characterized in that the vibration signals on the X, Y and Z axes of the accelerometer and the gyroscope are respectively collected. [Claim 3] Method for interacting with a smart watch according to claim 2, characterized in that, with the application of an anomaly detection algorithm, step S2 of identification of the vibration signals comprises:S21: data collection of the Z axis of the accelerometer;S22: filtering the Z axis data of the accelerometer using a high pass filter;S23: setting of an effective tap signal threshold value and a noise signal threshold value;S24: reading of a signal segment whose amplitude is less than the noise signal threshold value as a first state;S25: continuation of monitoring, waiting to read a signal whose amplitude is greater than the actual tap signal threshold value, the position whose amplitude is greater than the actual tap signal threshold value being X, and the signal start position being fixed at (XL), position L before position X;S26: continuation of monitoring while waiting to read a continuous signal whose amplitude is less than the noise signal threshold value, when the continuous amplitude is less than the noise signal threshold value, defines the final position of the signals a current position;S27: acquisition of signal data by start and stop position
    signal, determine if the signal length is within a length range, if not, go back to step S25, if yes, go to the next step; S28: application of high-pass filtering on the data, respectively calculation of the energies of the first m signal points and the energies of the signal points after the m points, determining whether the signal is greater than the threshold value of a noise ratio signal, if yes, determining the signal as an effective signal, otherwise determining the signal as a noise signal and returning to S25. [Claim 4] Method for interacting with a smart watch according to claim 1, in which step S3 specifically comprises:S31: preprocessing the signal with normalization, deducting an average value of the signal and dividing it by a variance;S32: during a step of initializing a training model, storing the data processed in step S31 as a training sample in a database; in the real use phase, application of the algorithm improved by the method of K nearest neighbors to classify and identify the vibration signals. [Claim 5] Method for interaction of a smart watch according to claim 1, in which the algorithm improved by the method of K nearest neighbors is specifically: based on the algorithm of dynamic time distortion, the real signal and the training signal are matched in frame units, and the shortest distance between them is calculated and applied as the basis for classification and recognition of the nearest K neighbor method. [Claim 6] A method of interacting with a smart watch according to claim 5, wherein the distance is the distance from Manhattan or the distance from Euler. [Claim 7] Method for interacting with a smart watch according to claim 1, in which step S4 comprises:S41: after collecting the classification result obtained in step 3, correcting the result of user input;S42: after the correction, corrects the training sample to a certain extent, thus maintaining the stability of the precision. [Claim 8] A method of interacting with a smart watch according to claim 7, wherein step S41 corrects the actual entry by providing a candidate key or by a correlation result in the entry method. [Claim 9] Method of interaction of a smart watch according to claim 8,
    in which step S42 specifically comprises:
    S421: if the result of the correction is consistent with the result of the classification, performing no operation;
    S422: if the result of the correction is incompatible with the result of the classification, for the training sample of the same category as the result of the classification, delete from the sample with the greatest distance calculated by the algorithm improved by the K nearest neighbor method, then replace the deleted sample with the current sample.
    [Claim 10] An interactive system for a smart watch, comprising:
    - a signal detection module, transmitting vibration signals based on the human body, and collecting vibration signals from an accelerometer and a gyroscope from a smart watch; a classification identification module that applies a anomaly detection algorithm for identifying vibration signals, preprocessing vibration signals and applying an improved algorithm by K nearest neighbors method to classify and identify vibration signals;
    - a real-time feedback module that analyzes a user's feedback, and applies adjustments at the right time to maintain identification accuracy.
    类似技术:
    公开号 | 公开日 | 专利标题
    FR3091606A1|2020-07-10|An interaction process and an interactive system of a smart watch
    EP2483758B1|2019-06-26|System and method for recognizing gestures
    US10353476B2|2019-07-16|Efficient gesture processing
    US9819677B2|2017-11-14|Supplementing biometric identification with device identification
    CN103345627B|2016-03-30|Action identification method and device
    CN109815795A|2019-05-28|Classroom student&#39;s state analysis method and device based on face monitoring
    US11127416B2|2021-09-21|Method and apparatus for voice activity detection
    EP2946311A2|2015-11-25|Accumulation of real-time crowd sourced data for inferring metadata about entities
    US8635181B2|2014-01-21|System and method for tagging signals of interest in time variant data
    CN106778477B|2020-04-03|Tennis racket action recognition method and device
    FR2915301A1|2008-10-24|PROCESS FOR COMPARISON OF IMAGES OF A BIOMETRY BETWEEN A REFERENCE IMAGE AND AT LEAST ONE TEST IMAGE WHICH WE SEEK TO EVALUATE A DEGREE OF CORRELATION WITH THE REFERENCE IMAGE
    CN104134440B|2018-05-08|Speech detection method and speech detection device for portable terminal
    Fernandez-Lopez et al.2017|Optimizing resources on smartphone gait recognition
    Killian et al.2019|Learning to detect heavy drinking episodes using smartphone accelerometer data
    Lu et al.2019|Detection of smoking events from confounding activities of daily living
    Zhou et al.2020|Xhar: Deep domain adaptation for human activity recognition with smart devices
    US10192118B2|2019-01-29|Analysis device, recording medium, and analysis method
    WO2021022795A1|2021-02-11|Method, apparatus, and device for detecting fraudulent behavior during facial recognition process
    Hazan et al.2015|Noise reduction of mobile sensors data in the prediction of demographic attributes
    Yang et al.2016|Person authentication using finger snapping—a new biometric trait
    US11270109B2|2022-03-08|Interactive method and interactive system for smart watch
    KR20190113252A|2019-10-08|Method for eye-tracking and terminal for executing the same
    CN106326672B|2019-04-09|Sleep detection method and system
    CN107170447B|2021-01-05|Sound processing system and sound processing method
    FR3051931A1|2017-12-01|DETERMINING A MOBILITY CONTEXT OF A CARRIER OF EQUIPMENT PROVIDED WITH INERTIAL SENSORS
    同族专利:
    公开号 | 公开日
    CN109840480B|2021-08-13|
    NL2022916B1|2020-01-31|
    FR3091606B1|2021-09-24|
    CN109840480A|2019-06-04|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    CN105810214B|2014-12-31|2019-11-05|展讯通信（上海）有限公司|Voice-activation detecting method and device|
    EP3065043A1|2015-03-02|2016-09-07|Nxp B.V.|Mobile device|
    CN107300971B|2017-06-09|2019-04-02|深圳大学|The intelligent input method and system propagated based on osteoacusis vibration signal|CN110169723A|2019-06-27|2019-08-27|九阳股份有限公司|A kind of food processor|
    CN111221420B|2020-01-13|2021-07-30|深圳大学|2D movement track identification method and system based on smart watch|
    CN113741703A|2021-11-08|2021-12-03|广东粤港澳大湾区硬科技创新研究院|Non-contact intelligent earphone or glasses interaction method|
    法律状态:
    2020-03-05| PLFP| Fee payment|Year of fee payment: 2 |
    2020-07-10| PLSC| Search report ready|Effective date: 20200710 |
    2021-04-06| PLFP| Fee payment|Year of fee payment: 3 |
    优先权:
    申请号 | 申请日 | 专利标题
    CN201910013634.3|2019-01-04|
    CN201910013634.3A|CN109840480B|2019-01-04|2019-01-04|Interaction method and interaction system of smart watch|
    [返回顶部]