WO2018036023A1 - 智能手表的文本输入方法及装置 - Google Patents

智能手表的文本输入方法及装置 Download PDF

Info

Publication number
WO2018036023A1
WO2018036023A1 PCT/CN2016/109509 CN2016109509W WO2018036023A1 WO 2018036023 A1 WO2018036023 A1 WO 2018036023A1 CN 2016109509 W CN2016109509 W CN 2016109509W WO 2018036023 A1 WO2018036023 A1 WO 2018036023A1
Authority
WO
WIPO (PCT)
Prior art keywords
smart watch
sensing data
axis
character
data sequence
Prior art date
Application number
PCT/CN2016/109509
Other languages
English (en)
French (fr)
Inventor
黄倩怡
王巍
张黔
Original Assignee
广州市香港科大霍英东研究院
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 广州市香港科大霍英东研究院 filed Critical 广州市香港科大霍英东研究院
Publication of WO2018036023A1 publication Critical patent/WO2018036023A1/zh

Links

Images

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
    • G06F3/02—Input arrangements using manually operated switches, e.g. using keyboards or dials
    • G06F3/023—Arrangements for converting discrete items of information into a coded form, e.g. arrangements for interpreting keyboard generated codes as alphanumeric codes, operand codes or instruction codes
    • G06F3/0233—Character input methods
    • G06F3/0234—Character input methods using switches operable in different directions
    • 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
    • G06F3/02—Input arrangements using manually operated switches, e.g. using keyboards or dials
    • G06F3/023—Arrangements for converting discrete items of information into a coded form, e.g. arrangements for interpreting keyboard generated codes as alphanumeric codes, operand codes or instruction codes
    • 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
    • G06F3/03—Arrangements for converting the position or the displacement of a member into a coded form
    • G06F3/033—Pointing devices displaced or positioned by the user, e.g. mice, trackballs, pens or joysticks; Accessories therefor
    • G06F3/0346—Pointing devices displaced or positioned by the user, e.g. mice, trackballs, pens or joysticks; Accessories therefor with detection of the device orientation or free movement in a 3D space, e.g. 3D mice, 6-DOF [six degrees of freedom] pointers using gyroscopes, accelerometers or tilt-sensors
    • 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
    • G06F3/0488—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 using a touch-screen or digitiser, e.g. input of commands through traced gestures
    • G06F3/04883—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 using a touch-screen or digitiser, e.g. input of commands through traced gestures for inputting data by handwriting, e.g. gesture or text

Definitions

  • the present invention relates to the field of smart wearable devices, and in particular to a text input method and apparatus for a smart watch.
  • Smart watches have ignited a wave of wearable devices. Smart watches play an important role in mobile payments, health tracking and event reminders. However, due to the limitations of portable design, the screen of a smart watch is usually small, which is not conducive to text input.
  • the object of the present invention is to provide a text input method and device for a smart watch, which can overcome the defects of the existing text input method by using the inertial sensor on the smart watch, so that the input is no longer limited by the occasion and the screen size.
  • the present invention provides a text input method for a smart watch, including:
  • each group of the character information includes M candidate characters
  • the inertial sensor of the smart watch comprises a three-axis acceleration sensor
  • the sensing data sequence comprises an x-axis sensing data sequence and a y-axis sensing data sequence
  • the x-axis and the y-axis are in contact with the smart watch dial Parallel planes.
  • the text input method of the smart watch when the user writes a character with a fingertip on the back of the hand, the force of the back of the hand can be transmitted to the wrist through the conduction of the skin and the bone, and the smart watch is in contact with the wrist, and the wrist can be driven.
  • Watch movement using the inertial sensor of the smart watch to capture the inertial sensing data when the watch is moving, ie the watch is parallel to the dial.
  • the motion data on the plane because the writing of different characters, the direction of the force of the back of the hand and the magnitude of the force are different, so the motion data captured by the sensor has a corresponding relationship with the characters handwritten by the user, and the inertial sensing data is analyzed and processed.
  • Alternative characters can be identified to identify the text entered by the user.
  • the invention does not need to set the keyboard on the touch screen, overcomes the limitation of the size of the touch screen, and also saves space for displaying the keyboard, can have more space to display other content, and overcomes the requirement of the voice input to the surrounding environment, and the text input is not Subject to the restrictions of the occasion.
  • Inertial sensors are widely used in smart watches on the market. Therefore, this design is directly applicable to existing equipment and does not require hardware modification of the equipment.
  • Inertial sensors are widely used in smart watches on the market. Therefore, this design is directly applicable to existing equipment and does not require hardware modification of the equipment.
  • character information corresponding to each group of sensing data sequences specifically includes:
  • the acquiring, by using the inertial sensor, detecting one or more sets of sensing data sequences caused by the user's handwriting specifically includes:
  • the current sensor data of the smart watch is detected by the inertial sensor as the sensing data of the stationary state;
  • the reading of the general inertial sensor will be affected by objective factors.
  • the reading of the acceleration sensor will be affected by gravity, and the data of the gyroscope will drift.
  • the user when the user enters the state in which the handwritten text is prepared, the user remains stationary, detects the sensory data, and subtracts the value of the sensory data collected after the text recognition is started, in a static state.
  • the effects of gravity and gyroscope drift can be eliminated, and the rest of the changes are caused by the user's handwriting on the back of the hand. By eliminating the effects of drift, the data entered into the machine learning model is more accurate and efficient.
  • the pre-processing includes a normalization process.
  • the positive direction of the x-axis is the direction in which the smart watch points to the back of the user's hand, or the direction of the user's hand points to the smart watch;
  • the x-axis energy sum is the sum of the motion energy of the smart watch on the x-axis
  • the y-axis The sum of energy is the sum of the kinetic energy of the smart watch in the y-axis direction
  • the x-axis sensing data sequence of the f-th sense data sequence and the correlation coefficient of the y-axis sensing data sequence are identical to each other.
  • the machine learning model is a random forest model
  • the configuration method of the random forest model includes:
  • the output prompts the user to handwrite information of the specified character on the back of the hand; wherein the specified character includes at least N different characters;
  • a random forest classifier is trained to obtain the random forest model.
  • the M candidate characters are taken as M different alternative texts
  • the smart watch detects a plurality of sets of sensor data sequences, combining the plurality of sets of the character information into a plurality of candidate texts; wherein each of the candidate texts includes a plurality of characters, each character is from the character
  • the information is extracted from an alternate character.
  • the present invention also provides a text input device for a smart watch, comprising:
  • a data detecting module configured to detect one or more sets of sensing data sequences caused by user handwriting by using an inertial sensor
  • a character recognition module configured to acquire, according to the sensing data sequence, character information corresponding to each group of sensing data sequences; wherein each group of the character information includes M candidate characters;
  • a text recognition module configured to obtain a plurality of different alternative texts according to the character information
  • a text confirmation module configured to select, from the plurality of the alternative texts, text matching the user's handwriting as the text recognized by the smart watch;
  • the inertial sensor of the smart watch comprises a three-axis acceleration sensor
  • the sensing data sequence comprises an x-axis sensing data sequence and a y-axis sensing data sequence
  • the x-axis and the y-axis are in contact with the smart watch dial Parallel planes.
  • the text input device of the smart watch when the user writes a character with a fingertip on the back of the hand, the force of the back of the hand can be transmitted to the wrist through the conduction of the skin and the bone, and the smart watch is in contact with the wrist, and the wrist can be driven.
  • Watch movement using the inertial sensor of the smart watch, captures the inertial sensing data of the watch when it is moving, that is, the movement data of the watch on the plane parallel to the dial.
  • the motion data captured by the sensor has a corresponding relationship with the characters handwritten by the user, and the analysis and processing of the inertial sensing data can identify the candidate characters, thereby identifying the text input by the user.
  • the invention does not need to set the keyboard on the touch screen, overcomes the limitation of the size of the touch screen, and also saves space for displaying the keyboard, can have more space to display other content, and overcomes the requirement of the voice input to the surrounding environment, and the text input is not Subject to the restrictions of the occasion.
  • Inertial sensors are widely used in smart watches on the market. Therefore, this design is directly applicable to existing equipment and does not require hardware modification of the equipment.
  • the character recognition module includes:
  • a data acquisition unit configured to acquire a f-th sense data sequence detected by an inertial sensor of the smart watch
  • a feature information extracting unit configured to extract, according to the f-th group sensing data sequence, feature information related to character recognition of the f-th group
  • a matching probability calculation unit configured to input the feature information of the f-th group into a pre-configured machine learning model, and obtain a matching probability of the f-th group of the feature information and the N different candidate characters output from the machine learning model ;
  • An candidate character obtaining unit configured to select, from the N different candidate characters, M candidate characters having the highest matching probability as character information corresponding to the f-th group sensing data sequence; wherein M ⁇ N.
  • the data detection module includes:
  • a static data detecting unit configured to detect, by the inertial sensor, the current sensing data of the smart watch as the sensing data of the stationary state when obtaining a text recognition start instruction for confirming that the user starts the handwritten text
  • a data acquisition unit configured to start text recognition, and continuously collect sensing data of the smart watch by using the inertial sensor
  • the anti-drift unit is configured to subtract the sensing data of the static state from all the sensing data collected after starting the text recognition, to obtain sensing data that eliminates the influence of drift;
  • a pre-processing unit configured to perform pre-processing on the sensing data for eliminating the drift effect, to obtain stable sensing data
  • a data grouping unit configured to monitor whether the sensing data continuously acquired by the inertial sensor is paused; and when the fth pause occurs, the stable between the (f-1)th pause and the fth pause Sensing data is used as the f-th group sensing data sequence.
  • the pre-processing includes a normalization process.
  • the positive direction of the x-axis is the direction in which the smart watch points to the back of the user's hand, or the direction of the user's hand points to the smart watch;
  • the x-axis energy sum is the sum of the motion energy of the smart watch on the x-axis
  • the y-axis The sum of energy is the sum of the kinetic energy of the smart watch in the y-axis direction
  • the x-axis sensing data sequence of the f-th sense data sequence and the correlation coefficient of the y-axis sensing data sequence are identical to each other.
  • the machine learning model is a random forest model
  • the configuration device of the random forest model includes:
  • the information prompting module includes: outputting information prompting the user to handwrite a specified character on the back of the hand; wherein the specified character includes at least N different characters;
  • the training data acquisition module is configured to collect a plurality of sets of sensor data sequences caused by the user's handwritten designated characters, and extract feature information corresponding to each group of sensor data sequences one by one as the plurality of sets of training data of the random forest learning model. Wherein each set of training data corresponds to a specified character;
  • a random forest classifier is trained to obtain the random forest model.
  • the text recognition module includes:
  • a single character recognition unit configured to use the M candidate characters as M different alternative texts if the smart watch detects only one set of sensing data sequences
  • a vocabulary identification unit configured to combine the plurality of sets of the sensor data into a plurality of candidate texts according to the plurality of sets of the character information; wherein each of the candidate texts includes a plurality of characters, each character Both are obtained by extracting an alternate character from the character information.
  • FIG. 1 is a flow chart of a first embodiment of a text input method provided by the present invention
  • FIG. 2 is a schematic diagram of the operation of the first embodiment of the text input method provided by the present invention.
  • FIG. 3 is a flow chart of a second embodiment of a text input method provided by the present invention.
  • FIG. 4 is a block diagram showing the structure of a text input device provided by the present invention.
  • FIG. 1 is a flowchart of a first embodiment of a text input method provided by the present invention
  • FIG. 2 is a schematic diagram of the operation of the first embodiment of the text input method provided by the present invention.
  • the text input method provided in this embodiment includes:
  • the embodiment detects one or more sets of sensing data sequences caused by the user's handwriting.
  • the inertial sensor of the smart watch includes a three-axis acceleration sensor and a gyroscope. Therefore, the data directly detected by the inertial sensor includes the x-axis of the three-axis acceleration sensor. Sensing data for the y-axis and z-axis, and sensing data for the x-axis, y-axis, and z-axis of the gyroscope.
  • the sensor data sequence for identifying the character information needs to include the sensing data sequence of the x-axis and the y-axis on a plane parallel to the dial of the smart watch.
  • the x-axis sensing data sequence and the y-axis sensing data sequence can be based on the x-axis, y-axis, and z-axis sensing data and gyroscope of the triaxial acceleration sensor described above. Sensing data for the x-axis, y-axis, and z-axis is obtained.
  • the working principle of the smart watch will be described below with reference to FIG. 2:
  • the x-axis and the y-axis are in a plane parallel to the dial of the smart watch, and the positive direction of the x-axis is the direction in which the smart watch points to the back of the user's hand.
  • the hand When the user wears the smart watch, when interacting with the smart watch, the hand is placed flat on the chest with the palm facing down, and the back of the hand and the watch are in the user's line of sight.
  • the size of the back of a normal person is above 3.5 inches by 3.5 inches, enough for the fingertip to be written on it.
  • the force on the back of the hand can be transmitted to the wrist through the conduction of the skin and bones, while the smart watch is in contact with the wrist, and the wrist can drive the watch to move, using the inertial sensor of the smart watch to capture
  • the inertial sensing data of the watch when moving that is, the motion data of the watch on the plane parallel to the dial, the force direction and the magnitude of the force of the back of the hand are different when writing different characters, so the motion data captured by the sensor and the user
  • the handwritten characters have a corresponding relationship, and the analysis and processing of the inertial sensing data can identify the candidate characters, thereby identifying the text input by the user.
  • the invention does not need to set the keyboard on the touch screen, overcomes the limitation of the size of the touch screen, and also saves space for displaying the keyboard, can have more space to display other content, and overcomes the requirement of the voice input to the surrounding environment, and the text input is not Subject to the restrictions of the occasion.
  • Inertial sensors are widely used in smart watches on the market. Therefore, this design is directly applicable to existing equipment and does not require hardware modification of the equipment.
  • the inertial sensor includes a triaxial acceleration sensor and a gyroscope, and the sensing data sequence is obtained by using the triaxial acceleration sensor and the gyroscope, which is a preferred embodiment.
  • the inertial sensor is selected according to the accuracy requirements of the identification, the cost, the use occasion, and the like. The selection of the inertial sensor is a technical means commonly used by those skilled in the art, and will not be described herein.
  • the reading of the general inertial sensor will be affected by objective factors.
  • the reading of the acceleration sensor will be affected by gravity, and the data of the gyroscope will drift.
  • the user when the user enters the state of preparing the handwritten text, the user remains stationary, detects the sensing data, and subtracts the value of the sensing data collected after the text recognition is started, thereby eliminating gravity.
  • the rest of the changes are caused by the user's handwriting on the back of the hand. That is, in the step S1 of the embodiment, the one or more sets of sensing data sequences caused by the user's handwriting are detected by the inertial sensor, and specifically includes:
  • the current sensor data of the smart watch is detected by the inertial sensor as the sensing data of the stationary state;
  • the text recognition revelation instruction confirming that the user starts the handwritten text may be the information that the user starts by the confirmation of the touch input on the touch screen of the smart watch, or may be after the data collection of the last text recognition is completed.
  • the text recognition revelation instruction for confirming the user's handwriting is automatically generated.
  • the determination condition of the end of the data collection of the text recognition may be that the time when the sensor data continuously acquired by the inertial sensor is paused is greater than a certain threshold, or the user finishes the confirmation by the touch input on the touch screen.
  • the continuous stable sensing data can be divided into multiple groups. Sensing data sequence.
  • the pre-processing includes a normalization process.
  • the acquiring the character information corresponding to each group of the sensing data sequence according to the sensing data sequence specifically includes:
  • the inertial sensor detects four sets of sensor data sequences; extracts feature information related to character recognition from the first set of sensor data sequences, for example, extracts the set of sensor data The similarity with the 26 sets of sensing data sequence samples stored in the smart watch, wherein the 26 sets of sensing data sequence samples respectively correspond to the letters a to the letter z; and the characteristic information is input into the pre-configured machine learning model.
  • the model will output the matching probability of the first group of sensor data sequences with 26 letters respectively; if the five letters with the highest matching probability are a, m, n, w and v, then the five letters will be used as the first group. Sensing the character information corresponding to the data sequence.
  • step S3 the obtaining a plurality of different alternative texts according to the character information
  • the M candidate characters are taken as M different alternative texts
  • the smart watch detects a plurality of sets of sensor data sequences, combining the plurality of sets of the character information into a plurality of candidate texts; wherein each of the candidate texts includes a plurality of characters, each character is from the character
  • the information is extracted from an alternate character.
  • the selecting, from the plurality of the alternative texts, the text matching the user's handwriting as the text recognized by the smart watch may include the following steps: outputting the candidate text to the user; receiving the user's Selecting information of the alternative text; confirming the text matching the user's handwriting as the text recognized by the smart watch according to the selection information. That is, a, m, n, w, and v are output, and the user selects the letter w matching the handwriting as the text recognized by the smart watch.
  • the text matching the user's handwriting according to the user selection information is only one embodiment of the design for improving the recognition accuracy in the embodiment.
  • the candidate text with the highest matching probability may also be considered. Text that matches the handwriting of the user.
  • the selecting, from the plurality of the alternative texts, the text matching the user's handwriting as the text recognized by the smart watch may include the following steps: sorting the candidate texts according to probability; Outputting the candidate text to the user in an order of low probability; receiving selection information of the candidate text by the user; confirming text matching the handwriting of the user according to the selection information, as text recognized by the smart watch; wherein The calculation method of the probability is the number of times the candidate text appears in the historical alternative text data recorded by the smart watch multiplied by the word frequency.
  • the feature information related to the character recognition includes the above similarity
  • the similarity calculation may be performed by a DTW (Dynamic Time Warping) algorithm to calculate the distance between the current sensing data sequence and the sample sequence.
  • the similarity is reflected by the distance, that is, the similarity is the similarity between the N sets of sensor data sequence samples stored in the smart watch calculated by the dynamic time warping method and the f-th group sensing data sequence respectively.
  • the N sets of sensor data sequence samples are respectively in one-to-one correspondence with N different candidate characters;
  • feature information may further include any one or more of the following feature information:
  • the time when the user writes a character that is, the duration of the f-th sense data sequence; the letters with more strokes, such as 'm', 'w' take more time to complete writing; and the strokes with fewer strokes, For example, 'c', 'l' can be completed faster, so the duration of the f-th sense data sequence can be used as a feature information for effectively identifying characters;
  • an energy ratio of the x-axis to the y-axis that is, a ratio of the sum of the x-axis energy calculated from the f-th sensor data sequence to the sum of the y-axis energy; wherein the x-axis energy sum is the smart watch
  • the sum of the kinetic energy on the x-axis which is the sum of the kinetic energy of the smart watch in the y-axis direction; we observe that when writing strokes parallel to the x-axis, the energy is mainly concentrated on the x-axis. Similarly, when writing strokes parallel to the y-axis, the energy is mainly concentrated on the y-axis.
  • the energy distribution of the letter 'e' on the x-axis is mainly concentrated in the first half of the signal, because the starting pen is parallel to the x-axis; and the letter 'f' is exactly the opposite, because the ending stroke is parallel to the x-axis; preferably,
  • the time point at which the energy reaches 25%, 50%, and 75% of the total energy of the uniaxial energy and the time point at which the energy reaches the maximum value is used as the characteristic information.
  • the four feature information are extracted separately.
  • the correlation between the x-axis and the y-axis energy that is, the x-axis sensing data sequence of the f-th sensing data sequence and the correlation coefficient of the y-axis sensing data sequence; we observe that when writing level Or vertical strokes, the x-axis and y-axis energy have a good correlation, they will reach the peak at the same time; however, when writing curved strokes, such as 'c', the x-axis and y-axis energy are poorly correlated. The time at which they reach the peak is not synchronized; therefore, the correlation coefficient between the x-axis sensing data sequence and the y-axis sensing data sequence can be used as a feature information.
  • the Random Forest model uses the above feature information to identify the correctness of characters. Compared with other machine learning models, it does not require complicated parameter adjustment process, and the model is intuitive and easy to understand. Therefore, the machine learning model in this embodiment is preferably a random forest model.
  • the configuration method of the random forest model includes:
  • the output prompts the user to handwrite the information of the specified character on the back of the hand; wherein the specified character includes at least N different characters; that is, the designated character includes 26 letters;
  • a random forest classifier is trained to obtain the random forest model.
  • the positive direction of the x-axis is the direction of the smart watch pointing to the back of the user's hand.
  • the positive direction of the x-axis may also be the direction of the user's hand pointing to the smart watch, or
  • the positive direction of the x-axis can also be other directions, in order to make the x-axis sensing data sequence and the y-axis sensing data sequence reflect the smart watch. In the case of a force-changing motion, it is necessary to keep the x-axis and the y-axis in a plane parallel to the dial.
  • the text input method provided in this embodiment includes:
  • the candidate text is used as the text recognized by the smart watch;
  • the configuration process of the random forest model is specifically:
  • the output prompts the user to handwrite 26 letters of information on the back of the hand; wherein each letter needs to be written 10 times;
  • the feature information is extracted from the new sample, and the random forest model is updated.
  • each letter user writes 10 times, taking 3 of the sensor data sequences to form the sensor data sequence sample set of the letter.
  • the present invention also provides a text input device for a smart watch.
  • a text input device for a smart watch.
  • FIG. 4 it is a structural block diagram of a text input device provided by the present invention, including:
  • the data detecting module 11 is configured to detect one or more sets of sensing data sequences caused by user handwriting by using an inertial sensor;
  • the character recognition module 12 is configured to acquire, according to the sensing data sequence, character information that is in one-to-one correspondence with each group of sensing data sequences; wherein each group of the character information includes M candidate characters;
  • a text recognition module 13 for obtaining a plurality of different alternative texts according to the character information
  • the text confirmation module 14 is configured to select, from the plurality of the alternative texts, text matching the user's handwriting as the text recognized by the smart watch;
  • the inertial sensor of the smart watch comprises a three-axis acceleration sensor
  • the sensing data sequence comprises an x-axis sensing data sequence and a y-axis sensing data sequence
  • the x-axis and the y-axis are in contact with the smart watch dial Parallel planes.
  • the text input device of the smart watch uses the inertial sensor of the smart watch to capture the inertial sensing data of the watch, that is, the motion data of the watch on a plane parallel to the dial, and identify the candidate characters according to the motion data, and further The text entered by the user is recognized.
  • the user can use the back of the hand as a writing screen, handwriting characters directly on the back of the hand, and the force of the back of the hand can be transmitted to the wrist through the conduction of the skin and bones, and the smart watch is in contact with the wrist, and the wrist can drive the watch to move. Therefore, the motion data captured by the sensor has a corresponding relationship with the characters handwritten by the user.
  • the characters written by the user can be inferred, and the keyboard is not required to be set on the touch screen, thereby overcoming the limitation of the size of the touch screen, and saving the need for display.
  • the space of the keyboard allows for more space to display other content, and overcomes the requirements of the voice input for the surrounding environment, and the text input is not limited by the occasion.
  • the character recognition module includes:
  • a data acquisition unit configured to acquire a f-th sense data sequence detected by an inertial sensor of the smart watch
  • a feature information extracting unit configured to extract, according to the f-th group sensing data sequence, feature information related to character recognition of the f-th group
  • a matching probability calculation unit configured to input the feature information of the f-th group into a pre-configured machine learning model, and obtain a matching probability of the f-th group of the feature information and the N different candidate characters output from the machine learning model ;
  • An candidate character obtaining unit configured to select, from the N different candidate characters, M candidate characters having the highest matching probability as character information corresponding to the f-th group sensing data sequence; wherein M ⁇ N.
  • the data detection module includes:
  • a static data detecting unit configured to detect, by the inertial sensor, the current sensing data of the smart watch as the sensing data of the stationary state when obtaining a text recognition start instruction for confirming that the user starts the handwritten text
  • a data acquisition unit configured to start text recognition, and continuously collect sensing data of the smart watch by using the inertial sensor
  • the anti-drift unit is configured to subtract the sensing data of the static state from all the sensing data collected after starting the text recognition, to obtain sensing data that eliminates the influence of drift;
  • a pre-processing unit configured to perform pre-processing on the sensing data for eliminating the drift effect, to obtain stable sensing data
  • a data grouping unit configured to monitor whether the sensing data continuously collected by the inertial sensor is paused; When f pauses, the stable sensing data between the (f-1)th pause and the fth pause is taken as the f-th group sensing data sequence.
  • the pre-processing includes a normalization process.
  • the positive direction of the x-axis is the direction in which the smart watch points to the back of the user's hand, or the direction of the user's hand points to the smart watch;
  • the x-axis energy sum is the sum of the motion energy of the smart watch on the x-axis
  • the y-axis The sum of energy is the sum of the kinetic energy of the smart watch in the y-axis direction
  • the x-axis sensing data sequence of the f-th sense data sequence and the correlation coefficient of the y-axis sensing data sequence are identical to each other.
  • the machine learning model is a random forest model
  • the configuration device of the random forest model includes:
  • the information prompting module includes: outputting information prompting the user to handwrite a specified character on the back of the hand; wherein the specified character includes at least N different characters;
  • the training data acquisition module is configured to collect a plurality of sets of sensor data sequences caused by the user's handwritten designated characters, and extract feature information corresponding to each group of sensor data sequences one by one as the plurality of sets of training data of the random forest learning model. Wherein each set of training data corresponds to a specified character;
  • a random forest classifier is trained to obtain the random forest model.
  • the text recognition module includes:
  • a single character recognition unit configured to use the M candidate characters as M different alternative texts if the smart watch detects only one set of sensing data sequences
  • a vocabulary identification unit configured to combine the plurality of sets of the sensor data into a plurality of candidate texts according to the plurality of sets of the character information; wherein each of the candidate texts includes a plurality of characters, each character Both are obtained by extracting an alternate character from the character information.
  • the text input method and device of the smart watch provided by the invention, when the user writes a character with a fingertip on the back of the hand, the hand
  • the force of the back can be transmitted to the wrist through the conduction of the skin and bones, while the smart watch is in contact with the wrist, the wrist can drive the watch to move, and the inertial sensor of the smart watch is used to capture the inertial sensing data of the watch during the movement, that is, the watch is in
  • the motion data on the plane parallel to the dial because the different directions of the hand and the magnitude of the force are different when writing different characters, the motion data captured by the sensor has a corresponding relationship with the characters handwritten by the user, and the inertial sensing data
  • the analysis process identifies the candidate characters and identifies the text entered by the user.
  • the invention does not need to set the keyboard on the touch screen, overcomes the limitation of the size of the touch screen, and also saves space for displaying the keyboard, can have more space to display other content, and overcomes the requirement of the voice input to the surrounding environment, and the text input is not Subject to the restrictions of the occasion.
  • Inertial sensors are widely used in smart watches on the market. Therefore, this design is directly applicable to existing equipment and does not require hardware modification of the equipment.

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)
  • Character Discrimination (AREA)

Abstract

一种智能手表的文本输入方法和输入装置，方法包括：通过惯性传感器检测由用户手写引起的一组或多组传感数据序列（S1）；根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符(S2)；根据所述字符信息，获得多个不同的备选文本(S3)；从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本(S4)。无需在触摸屏上设置键盘，克服了触摸屏大小的局限，同时也节省了需要显示键盘的空间，可以有更多的空间来显示其他内容，并且克服语音输入对周围环境的要求，文本输入不受场合的限制。

Description

智能手表的文本输入方法及装置 技术领域
本发明涉及智能穿戴设备领域，具体地，涉及一种智能手表的文本输入方法及装置。
背景技术
近年来，智能手表兴起了可穿戴设备的热潮。智能手表在移动支付、健康追踪和事件提醒等方面发挥着重要作用。然而，由于受到便携设计的限制，智能手表的屏幕通常很小，不利于文本输入。
目前，智能手表的文本输入方式主要有两种：最常见的方式是将语音转化为文本，比如在Apple Watch上的Siri。这种输入方式非常方便有效，但是，它不适用于在一些公共场所中使用，诸如图书馆、办公室等。而且，在嘈杂的环境中，错误率较高。第二种方式是将常用的全键盘或九宫格键盘重新布局设计，使之适用于智能手表的触摸屏。由于屏幕较小，每次只能显示若干个键，并且会挡住其他内容的显示，效率较低。
发明内容
本发明的目的在于提供一种智能手表的文本输入方法及装置，其能利用智能手表上的惯性传感器，克服现有文本输入法的缺陷，使得输入不再受到场合限制和屏幕大小的限制。
为了实现上述目的，本发明提供一种智能手表的文本输入方法，包括：
通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符；
根据所述字符信息，获得多个不同的备选文本；
从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本；
其中，本智能手表的惯性传感器包括三轴加速度传感器，所述传感数据序列包括x轴的传感数据序列和y轴的传感数据序列；所述x轴与y轴处于与本智能手表表盘平行的平面。
实施本发明，具有如下有益效果：
本发明提供的智能手表的文本输入方法，当用户在手背上用指尖手写字符时，手背的受力通过皮肤和骨骼的传导，可以传导至手腕部分，而智能手表与手腕接触，手腕可带动手表运动，利用智能手表的惯性传感器，捕获手表运动时的惯性传感数据，即手表在与表盘平行 的平面上的运动数据，由于书写不同的字符时，手背的受力方向和力度大小都有所不同，因此传感器捕获的运动数据与用户手写的字符具有对应关系，对惯性传感数据进行分析处理可以识别出备选字符，进而识别出用户输入的文本。本发明无需在触摸屏上设置键盘，克服了触摸屏大小的局限，同时也节省了需要显示键盘的空间，可以有更多的空间来显示其他内容，并且克服语音输入对周围环境的要求，文本输入不受场合的限制。而惯性传感器广泛存在于市面上的智能手表中，因此，这种设计直接适用于现有的设备，不需要对设备进行硬件上的改造。
惯性传感器广泛存在于市面上的智能手表中，因此，这种设计直接适用于现有的设备，不需要对设备进行硬件上的改造。
进一步地，所述根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息，具体包括：
获取本智能手表的惯性传感器检测到的第f组传感数据序列；
根据所述第f组传感数据序列，提取第f组与字符识别相关的特征信息；
将第f组所述特征信息输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与N个不同的备选字符的匹配概率；
从所述N个不同的备选字符中选择匹配概率最高的M个备选字符作为与所述第f组传感数据序列对应的字符信息；其中M＜N。
进一步地，所述字符为英文字母，N＝26，所述N个备选字符为字母a～字母z。
进一步地，所述获取通过惯性传感器检测由用户手写引起的一组或多组传感数据序列，具体包括：
在获得确认用户开始手写文本的文本识别启动指令时，通过惯性传感器检测本智能手表当前的传感数据，作为静止状态的传感数据；
启动文本识别，通过所述惯性传感器持续采集本智能手表的传感数据；
将启动文本识别后所采集到的所有传感数据减去所述静止状态的传感数据，得到消除漂移影响的传感数据；
对所述消除漂移影响的传感数据执行预处理，得到稳定的传感数据；
监控所述惯性传感器持续采集的传感数据是否发生停顿；在发生第f次停顿时，将第(f-1)次停顿与第f次停顿之间的所述稳定的传感数据作为所述第f组传感数据序列。
一般的惯性传感器的读数都会受到客观因素的影响，比如加速度传感器的读数会受到重力的影响，陀螺仪的数据会有漂移。在进一步方案中，在用户进入准备手写文本的状态时，用户保持静止，检测传感数据，将启动文本识别后采集到的传感数据减去静止状态时的值， 可以消除重力和陀螺仪漂移的影响，剩下的变化就是因为用户在手背上手写引起的。通过消除漂移影响，使输入机器学习模型的数据更准确有效。
进一步地，所述预处理包括归一化处理。
通过归一化处理将信号值投射到预定的范围，可以消除书写位置和力度的影响。
进一步地，所述x轴的正方向为本智能手表指向用户手背的方向，或者用户手背指向本智能手表的方向；
所述第f组与字符识别相关的特征信息，包括以下特征信息中的至少一项：
第f组传感数据序列的持续时间；
采用动态时间规整法计算得到的存储于本智能手表中的N组传感数据序列样本分别与所述第f组传感数据序列的相似度；其中，所述N组传感数据序列样本分别与N个不同的备选字符一一对应；
根据所述第f组传感数据序列计算得到的x轴能量总和与y轴能量总和的比值；其中，所述x轴能量总和为本智能手表在x轴上的运动能量总和，所述y轴能量总和为本智能手表在y轴方向上的运动能量总和；
本智能手表在x轴上的运动能量达到预设阈值的时间点，以及本智能手表在y轴上的运动能量达到预设阈值时的时间点；
所述第f组传感数据序列的x轴的传感数据序列以及y轴的传感数据序列的相关系数。
进一步地，所述机器学习模型为随机森林模型；
所述随机森林模型的配置方法包括：
输出提示用户在手背上手写指定字符的信息；其中，所述指定字符包括至少N个不同的字符；
采集多组由用户手写指定字符引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个指定字符对应；
采用所述训练数据，训练随机森林分类器，得到所述随机森林模型。
进一步地，所述根据所述字符信息，获得多个不同的备选文本；
若本智能手表仅检测到一组传感数据序列，则将所述M个备选字符作为M个不同的备选文本；
若本智能手表检测到多组传感数据序列，则根据多组所述字符信息，组合成多个备选文本；其中每个备选文本包括多个字符，每个字符都是从所述字符信息中抽取一个备选字符得到的。
相应地，本发明还提供一种智能手表的文本输入装置，包括：
数据检测模块，用于通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
字符识别模块，用于根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符；
文本识别模块，用于根据所述字符信息，获得多个不同的备选文本；
文本确认模块，用于从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本；
其中，本智能手表的惯性传感器包括三轴加速度传感器，所述传感数据序列包括x轴的传感数据序列和y轴的传感数据序列；所述x轴与y轴处于与本智能手表表盘平行的平面。
本发明提供的智能手表的文本输入装置，当用户在手背上用指尖手写字符时，手背的受力通过皮肤和骨骼的传导，可以传导至手腕部分，而智能手表与手腕接触，手腕可带动手表运动，利用智能手表的惯性传感器，捕获手表运动时的惯性传感数据，即手表在与表盘平行的平面上的运动数据，由于书写不同的字符时，手背的受力方向和力度大小都有所不同，因此传感器捕获的运动数据与用户手写的字符具有对应关系，对惯性传感数据进行分析处理可以识别出备选字符，进而识别出用户输入的文本。本发明无需在触摸屏上设置键盘，克服了触摸屏大小的局限，同时也节省了需要显示键盘的空间，可以有更多的空间来显示其他内容，并且克服语音输入对周围环境的要求，文本输入不受场合的限制。而惯性传感器广泛存在于市面上的智能手表中，因此，这种设计直接适用于现有的设备，不需要对设备进行硬件上的改造。
进一步地，所述字符识别模块包括：
数据获取单元，用于获取本智能手表的惯性传感器检测到的第f组传感数据序列；
特征信息提取单元，用于根据所述第f组传感数据序列，提取第f组与字符识别相关的特征信息；
匹配概率计算单元，用于将第f组所述特征信息输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与N个不同的备选字符的匹配概率；
备选字符获取单元，用于从所述N个不同的备选字符中选择匹配概率最高的M个备选字符作为与所述第f组传感数据序列对应的字符信息；其中M＜N。
进一步地，所述字符为英文字母，N＝26，所述N个备选字符为字母a～字母z。
进一步地，所述数据检测模块包括：
静止数据检测单元，用于在获得确认用户开始手写文本的文本识别启动指令时，通过惯性传感器检测本智能手表当前的传感数据，作为静止状态的传感数据；
数据采集单元，用于启动文本识别，通过所述惯性传感器持续采集本智能手表的传感数据；
消除漂移单元，用于将启动文本识别后所采集到的所有传感数据减去所述静止状态的传感数据，得到消除漂移影响的传感数据；
预处理单元，用于对所述消除漂移影响的传感数据执行预处理，得到稳定的传感数据；
数据分组单元，用于监控所述惯性传感器持续采集的传感数据是否发生停顿；在发生第f次停顿时，将第(f-1)次停顿与第f次停顿之间的所述稳定的传感数据作为所述第f组传感数据序列。
进一步地，所述预处理包括归一化处理。
进一步地，所述x轴的正方向为本智能手表指向用户手背的方向，或者用户手背指向本智能手表的方向；
所述第f组与字符识别相关的特征信息，包括以下特征信息中的至少一项：
第f组传感数据序列的持续时间；
采用动态时间规整法计算得到的存储于本智能手表中的N组传感数据序列样本分别与所述第f组传感数据序列的相似度；其中，所述N组传感数据序列样本分别与N个不同的备选字符一一对应；
根据所述第f组传感数据序列计算得到的x轴能量总和与y轴能量总和的比值；其中，所述x轴能量总和为本智能手表在x轴上的运动能量总和，所述y轴能量总和为本智能手表在y轴方向上的运动能量总和；
本智能手表在x轴上的运动能量达到预设阈值的时间点，以及本智能手表在y轴上的运动能量达到预设阈值时的时间点；
所述第f组传感数据序列的x轴的传感数据序列以及y轴的传感数据序列的相关系数。
进一步地，所述机器学习模型为随机森林模型；
所述随机森林模型的配置装置包括：
信息提示模块，包括输出提示用户在手背上手写指定字符的信息；其中，所述指定字符包括至少N个不同的字符；
训练数据采集模块，用于采集多组由用户手写指定字符引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个指定字符对应；
采用所述训练数据，训练随机森林分类器，得到所述随机森林模型。
进一步地，所述文本识别模块包括；
单字符识别单元，用于若本智能手表仅检测到一组传感数据序列，则将所述M个备选字符作为M个不同的备选文本；
词汇识别单元，用于若本智能手表检测到多组传感数据序列，则根据多组所述字符信息，组合成多个备选文本；其中每个备选文本包括多个字符，每个字符都是从所述字符信息中抽取一个备选字符得到的。
附图说明
图1是本发明提供的文本输入方法的第一实施例的流程图；
图2是本发明提供的文本输入方法的第一实施例的操作示意图；
图3是本发明提供的文本输入方法的第二实施例的流程图；
图4是本发明提供的文本输入装置的结构框图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
参见图1和图2，其中图1是本发明提供的文本输入方法的第一实施例的流程图；图2是本发明提供的文本输入方法的第一实施例的操作示意图。
如图1所示，本实施例提供的文本输入方法，包括：
S1、通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
S2、根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符；
S3、根据所述字符信息，获得多个不同的备选文本；
S4、从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本。
本实施例检测由用户手写引起的一组或多组传感数据序列，智能手表的惯性传感器包括三轴加速度传感器和陀螺仪，因此，惯性传感器直接检测到的数据包括三轴加速度传感器的x轴、y轴和z轴的传感数据以及陀螺仪的x轴、y轴和z轴的传感数据。为了在用户佩戴本智能手表时，手表能按上述方法识别文本，用于识别字符信息的传感数据序列需要包括处于与本智能手表表盘平行的平面上的x轴和y轴的传感数据序列。x轴的传感数据序列和y轴的传感数据序列可以根据上述的三轴加速度传感器的x轴、y轴和z轴的传感数据和陀螺仪 的x轴、y轴和z轴的传感数据获得。下面结合图2说明本智能手表的工作原理：
如图2所示，x轴与y轴处于与本智能手表表盘平行的平面，x轴的正方向为本智能手表指向用户手背的方向。当用户佩戴本智能手表，要和智能手表交互时，会将手平放在胸前，手掌朝下，此时手背和手表同在用户的视线中。正常人的手背大小在3.5英寸*3.5英寸的面积以上，足够指尖在其上书写。当用户在手背上用指尖手写字符时，手背的受力通过皮肤和骨骼的传导，可以传导至手腕部分，而智能手表与手腕接触，手腕可带动手表运动，利用智能手表的惯性传感器，捕获手表运动时的惯性传感数据，即手表在与表盘平行的平面上的运动数据，由于书写不同的字符时，手背的受力方向和力度大小都有所不同，因此传感器捕获的运动数据与用户手写的字符具有对应关系，对惯性传感数据进行分析处理可以识别出备选字符，进而识别出用户输入的文本。本发明无需在触摸屏上设置键盘，克服了触摸屏大小的局限，同时也节省了需要显示键盘的空间，可以有更多的空间来显示其他内容，并且克服语音输入对周围环境的要求，文本输入不受场合的限制。而惯性传感器广泛存在于市面上的智能手表中，因此，这种设计直接适用于现有的设备，不需要对设备进行硬件上的改造。
需要说明的是，本实施例中，惯性传感器包括三轴加速度传感器和陀螺仪，利用三轴加速度传感器和陀螺仪获得传感数据序列，是一种优选实施方式，在其他实施例中，也可以根据识别的精确度需求和成本、使用场合等因素选用惯性传感器，选用惯性传感器是本领域技术人员常用的技术手段，在此不作赘述。
一般的惯性传感器的读数都会受到客观因素的影响，比如加速度传感器的读数会受到重力的影响，陀螺仪的数据会有漂移。在本实施例的进一步方案中，当用户进入准备手写文本的状态时，用户保持静止，检测传感数据，将启动文本识别后采集到的传感数据减去静止状态时的值，可以消除重力和陀螺仪漂移的影响，剩下的变化就是因为用户在手背上手写引起的。即在本实施例的步骤S1中，所述通过惯性传感器检测由用户手写引起的一组或多组传感数据序列，具体包括：
在获得确认用户开始手写文本的文本识别启动指令时，通过惯性传感器检测本智能手表当前的传感数据，作为静止状态的传感数据；
启动文本识别，通过所述惯性传感器持续采集本智能手表的传感数据；
将启动文本识别后所采集到的所有传感数据减去所述静止状态的传感数据，得到消除漂移影响的传感数据；
对所述消除漂移影响的传感数据执行预处理，得到稳定的传感数据；
监控所述惯性传感器持续采集的传感数据是否发生停顿；在发生第f次停顿时，将第(f-1)次停顿与第f次停顿之间的所述稳定的传感数据作为所述第f组传感数据序列。
在步骤S1的上述具体步骤中，确认用户开始手写文本的文本识别启示指令可以是用户在智能手表的触摸屏上通过触摸输入的确认开始的信息，也可以是在上一次文本识别的数据采集结束后，又检测到数值标准差在特定范围内的传感数据时自动生成确认用户开始手写的文本识别启示指令。而一次文本识别的数据采集结束的判定条件可以是惯性传感器持续采集的传感数据停顿的时间大于设定的某一阈值，或者用户在触摸屏上通过触摸输入的确认结束的信息。
在启动文本识别后，用户在手写文本的过程中，若写了多个字符，则字符与字符之间会有短暂的停顿，利用这个停顿，可以将连续的稳定的传感数据分割为多组传感数据序列。
进一步地，所述预处理包括归一化处理。
通过归一化处理将信号值投射到预定的范围，可以消除书写位置和力度的影响。
在本实施例的步骤S2中，所述根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息，具体包括：
获取本智能手表的惯性传感器检测到的第f组传感数据序列；
根据所述第f组传感数据序列，提取第f组与字符识别相关的特征信息；
将第f组所述特征信息输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与N个不同的备选字符的匹配概率；
从所述N个不同的备选字符中选择匹配概率最高的M个备选字符作为与所述第f组传感数据序列对应的字符信息；其中M＜N。
为了便于说明步骤S2的上述具体步骤，优选地，所述字符为英文字母，N＝26，所述N个备选字符为字母a～字母z，M＝5。假设用户在手背上写了一个单词“word”，惯性传感器检测到4组传感数据序列；从第1组传感数据序列中提取与字符识别相关的特征信息，比如，提取该组传感数据与存储于本智能手表中的26组传感数据序列样本的相似度，其中，所述的26组传感数据序列样本分别对应字母a～字母z；经特征信息输入预先配置的机器学习模型中，模型会输出第1组传感数据序列分别与26个字母的匹配概率；假设匹配概率最高的5个字母是a、m、n、w和v，则这5个字母则作为与第1组传感数据序列对应的字符信息。
进一步地，在步骤S3中，所述根据所述字符信息，获得多个不同的备选文本；
若本智能手表仅检测到一组传感数据序列，则将所述M个备选字符作为M个不同的备选文本；
若本智能手表检测到多组传感数据序列，则根据多组所述字符信息，组合成多个备选文本；其中每个备选文本包括多个字符，每个字符都是从所述字符信息中抽取一个备选字符得到的。
假设用户只写了一个字母“w”，则将传感数据序列对应的字符信息：a、m、n、w和v作为备选文本。在步骤S4中，所述从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本，可包括以下步骤：向用户输出所述备选文本；接收用户对所述备选文本的选择信息；根据所述选择信息确认与用户手写匹配的文本，作为本智能手表识别的文本。即输出a、m、n、w和v，用户选择与自己手写匹配的字母w，作为本智能手表识别的文本。需要说明的是，根据用户选择信息确认与用户手写匹配的文本，仅仅是本实施例为了提高识别准确度设计的一种实施方式，在其他实施方式中，也可以认为匹配概率最高的备选文本为与用户手写匹配的文本。
假设用户书写了单词“word”，则根据每一组传感数据序列对应的5个备选字符进行排列组合，可以得到一系列备选单词，将这些备选单词作为备选文本。在步骤S4中，所述从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本，可包括以下步骤：将所述备选文本按照概率进行排序；以概率高至概率低的次序向用户输出所述备选文本；接收用户对所述备选文本的选择信息；根据所述选择信息确认与用户手写匹配的文本，作为本智能手表识别的文本；其中，所述概率的计算方法为所述备选文本在本智能手表记录的历史备选文本数据中出现的次数乘以词频。
本实施例中，与字符识别相关的特征信息包括了上述的相似度，相似度的计算可以DTW(Dynamic Time Warping，动态时间规整法)算法计算出当前的传感数据序列与样本序列的距离，用距离来反映相似度，即所述相似度具体为采用动态时间规整法计算得到的存储于本智能手表中的N组传感数据序列样本分别与所述第f组传感数据序列的相似度；其中，所述N组传感数据序列样本分别与N个不同的备选字符一一对应；
此外，特征信息还可以包括以下特征信息中的任意一项或多项：
(1)用户书写一个字符的时间，即第f组传感数据序列的持续时间；笔画多的字母，诸如‘m’，‘w’需要较多的时间来完成书写；而笔画少的字母，诸如‘c’，‘l’则可以较快完成，因此第f组传感数据序列的持续时间可以作为一个有效识别字符的特征信息；
(2)x轴与y轴的能量比，即根据所述第f组传感数据序列计算得到的x轴能量总和与y轴能量总和的比值；其中，所述x轴能量总和为本智能手表在x轴上的运动能量总和，所述y轴能量总和为本智能手表在y轴方向上的运动能量总和；我们观察到，当书写和x-轴平行的笔画时，能量主要集中在x轴；类似，当书写和y轴平行的笔画时，能量主要集中在y轴。通过计算x轴与y轴的能量比，可以推测出笔画的大致走向。
(3)能量的时域分布，即本智能手表在x轴上的运动能量达到预设阈值的时间点，以及本智能手表在y轴上的运动能量达到预设阈值时的时间点；无论x轴或y轴，能量在时域上 的分布和字母的笔画顺序相关。比如，字母‘e’在x轴上的能量分布主要集中在信号的前半部分，因为起始笔与x轴平行；而字母‘f’恰恰相反，因为结束笔画与x轴平行；优选地，可以用能量达到单轴总能量的25％、50％和75％的时间点和能量达到最大值对应的时间点作为特征信息。对于x轴和y轴，分别提取这四个特征信息。
(4)x轴与y轴能量的相关性，即所述第f组传感数据序列的x轴的传感数据序列以及y轴的传感数据序列的相关系数；我们观察到，当书写水平或竖直笔画时，x轴与y轴能量有较好的相关性，它们会在同时达到峰值；然而，书写弧形笔画时，如‘c’，x轴与y轴能量的相关性较差，它们达到峰值的时间不同步；因此，可以将x轴的传感数据序列与y轴的传感数据序列的相关系数作为一个特征信息。
通过大量的实验发现，Random Forest(随机森林)模型利用上述特征信息识别字符的正确性最好，与其他机器学习模型相比，它不需要复杂的参数调整过程，模型直观，易于理解。因此，本实施例中的机器学习模型优选为随机森林模型。
所述随机森林模型的配置方法包括：
输出提示用户在手背上手写指定字符的信息；其中，所述指定字符包括至少N个不同的字符；即指定字符包括了26个字母在内；
采集多组由用户手写指定字符引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个指定字符对应；
采用所述训练数据，训练随机森林分类器，得到所述随机森林模型。
需要说明的是，本实施例中，x轴的正方向为本智能手表指向用户手背的方向，在其他实施例中，x轴的正方向也可以是用户手背指向本智能手表的方向，或者，当采用其他特征信息作为与字符识别相关的特征信息时，x轴的正方向还可以是其他方向，为了使x轴的传感数据序列和y轴的传感数据序列能够反映智能手表随手背受力变化的运动情况，需要保持x轴和y轴处于与表盘平行的平面。
参见图3，是本发明提供的文本输入方法的第二实施例的流程图；本实施例提供的文本输入方法包括：
S21、提示用户输入文本；
S22、在获得用户确认开始输入文本的信息后，通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
S23、通过动态时间规整法计算第f组传感数据序列与本智能手表预存的26个字母的传感数据序列样本集比对，获得第f组传感数据序列与26个字母的相似度；其中，每个字母的 传感数据序列样本集包含3个样本，使用动态时间规整法计算样本与第f组传感数据序列的距离大小以反映相似度的过程，具体为用动态时间规整法计算出第f组传感数据序列与每个字母的3个样本的距离，取其平均值作为第f组传感数据序列与该字母的距离，反映第f组传感数据序列与该字母的相似度；
S24、提取包含所述相似度在内的第f组所述特征信息，输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与26个字母的匹配概率；选取匹配概率最高的5个字母作为备选字符；
S25、将得到的备选字符进行排列组合，得到多个备选文本；输出备选文本以及备选文本中没有正确字符的选项供用户选择；
S26、若用户选择了其中一个备选文本，则将该备选文本作为本智能手表识别的文本；
S27、若用户选择了备选文本中没有正确字符的选项，则搜索编辑距离为1的单词作为备选文本供用户选择；依次扩大搜索范围，直到用户选择了其中一个备选文本作为本智能手表识别的文本为止；其中，编辑距离指两个文本之间，由一个文本转成另一个文本所需的最少编辑操作次数；
S28、从本智能手表识别的文本中提取与每一组传感数据序列对应的正确字母；并将所述正确字母的样本集中的一个样本更新为本次文本输入过程中该正确字母对应的传感数据序列。
本实施例中，随机森林模型的配置过程具体为：
输出提示用户在手背上手写26个字母的信息；其中，每个字母需要书写10次；
采集用户手写26个字母引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个字母对应；每个字母对应10组训练数据；
采用所述训练数据，训练随机森林分类器，得到初始的随机森林模型；
每次完成文本识别后，从新的样本中提取特征信息，更新所述随机森林模型。
在配置随机森林模型的过程中，每个字母用户都书写了10次，取其中3次的传感数据序列组成该字母的传感数据序列样本集。
相应地，本发明还提供一种智能手表的文本输入装置，参见图4，是本发明提供的文本输入装置的结构框图，包括：
数据检测模块11，用于通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
字符识别模块12，用于根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符；
文本识别模块13，用于根据所述字符信息，获得多个不同的备选文本；
文本确认模块14，用于从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本；
其中，本智能手表的惯性传感器包括三轴加速度传感器，所述传感数据序列包括x轴的传感数据序列和y轴的传感数据序列；所述x轴与y轴处于与本智能手表表盘平行的平面。
本发明提供的智能手表的文本输入装置，利用智能手表的惯性传感器，捕获与手表的惯性传感数据，即手表在与表盘平行的平面上的运动数据，根据运动数据识别出备选字符，进而识别出用户输入的文本。采用这种方法，用户可以把手背当做书写屏幕，直接在手背上手写字符，手背的受力通过皮肤和骨骼的传导，可以传导至手腕部分，而智能手表与手腕接触，手腕可带动手表运动，因此，传感器捕获的运动数据与用户手写的字符具有对应关系，通过分析传感数据可以推测出用户所书写的字符，无需在触摸屏上设置键盘，克服了触摸屏大小的局限，同时也节省了需要显示键盘的空间，可以有更多的空间来显示其他内容，并且克服语音输入对周围环境的要求，文本输入不受场合的限制。
进一步地，所述字符识别模块包括：
数据获取单元，用于获取本智能手表的惯性传感器检测到的第f组传感数据序列；
特征信息提取单元，用于根据所述第f组传感数据序列，提取第f组与字符识别相关的特征信息；
匹配概率计算单元，用于将第f组所述特征信息输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与N个不同的备选字符的匹配概率；
备选字符获取单元，用于从所述N个不同的备选字符中选择匹配概率最高的M个备选字符作为与所述第f组传感数据序列对应的字符信息；其中M＜N。
进一步地，所述字符为英文字母，N＝26，所述N个备选字符为字母a～字母z。
进一步地，所述数据检测模块包括：
静止数据检测单元，用于在获得确认用户开始手写文本的文本识别启动指令时，通过惯性传感器检测本智能手表当前的传感数据，作为静止状态的传感数据；
数据采集单元，用于启动文本识别，通过所述惯性传感器持续采集本智能手表的传感数据；
消除漂移单元，用于将启动文本识别后所采集到的所有传感数据减去所述静止状态的传感数据，得到消除漂移影响的传感数据；
预处理单元，用于对所述消除漂移影响的传感数据执行预处理，得到稳定的传感数据；
数据分组单元，用于监控所述惯性传感器持续采集的传感数据是否发生停顿；在发生第 f次停顿时，将第(f-1)次停顿与第f次停顿之间的所述稳定的传感数据作为所述第f组传感数据序列。
进一步地，所述预处理包括归一化处理。
进一步地，所述x轴的正方向为本智能手表指向用户手背的方向，或者用户手背指向本智能手表的方向；
所述第f组与字符识别相关的特征信息，包括以下特征信息中的至少一项：
第f组传感数据序列的持续时间；
采用动态时间规整法计算得到的存储于本智能手表中的N组传感数据序列样本分别与所述第f组传感数据序列的相似度；其中，所述N组传感数据序列样本分别与N个不同的备选字符一一对应；
根据所述第f组传感数据序列计算得到的x轴能量总和与y轴能量总和的比值；其中，所述x轴能量总和为本智能手表在x轴上的运动能量总和，所述y轴能量总和为本智能手表在y轴方向上的运动能量总和；
本智能手表在x轴上的运动能量达到预设阈值的时间点，以及本智能手表在y轴上的运动能量达到预设阈值时的时间点；
所述第f组传感数据序列的x轴的传感数据序列以及y轴的传感数据序列的相关系数。
进一步地，所述机器学习模型为随机森林模型；
所述随机森林模型的配置装置包括：
信息提示模块，包括输出提示用户在手背上手写指定字符的信息；其中，所述指定字符包括至少N个不同的字符；
训练数据采集模块，用于采集多组由用户手写指定字符引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个指定字符对应；
采用所述训练数据，训练随机森林分类器，得到所述随机森林模型。
进一步地，所述文本识别模块包括；
单字符识别单元，用于若本智能手表仅检测到一组传感数据序列，则将所述M个备选字符作为M个不同的备选文本；
词汇识别单元，用于若本智能手表检测到多组传感数据序列，则根据多组所述字符信息，组合成多个备选文本；其中每个备选文本包括多个字符，每个字符都是从所述字符信息中抽取一个备选字符得到的。
本发明提供的智能手表的文本输入方法及装置，当用户在手背上用指尖手写字符时，手 背的受力通过皮肤和骨骼的传导，可以传导至手腕部分，而智能手表与手腕接触，手腕可带动手表运动，利用智能手表的惯性传感器，捕获手表运动时的惯性传感数据，即手表在与表盘平行的平面上的运动数据，由于书写不同的字符时，手背的受力方向和力度大小都有所不同，因此传感器捕获的运动数据与用户手写的字符具有对应关系，对惯性传感数据进行分析处理可以识别出备选字符，进而识别出用户输入的文本。本发明无需在触摸屏上设置键盘，克服了触摸屏大小的局限，同时也节省了需要显示键盘的空间，可以有更多的空间来显示其他内容，并且克服语音输入对周围环境的要求，文本输入不受场合的限制。而惯性传感器广泛存在于市面上的智能手表中，因此，这种设计直接适用于现有的设备，不需要对设备进行硬件上的改造。
以上是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和变形，这些改进和变形也视为本发明的保护范围。

Claims (16)

  1. 一种智能手表的文本输入方法，其特征在于，包括：
    通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
    根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符；
    根据所述字符信息，获得多个不同的备选文本；
    从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本；
    其中，本智能手表的惯性传感器包括三轴加速度传感器，所述传感数据序列包括x轴的传感数据序列和y轴的传感数据序列；所述x轴与y轴处于与本智能手表表盘平行的平面。
  2. 如权利要求1所述的智能手表的文本输入方法，其特征在于，所述根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息，具体包括：
    获取本智能手表的惯性传感器检测到的第f组传感数据序列；
    根据所述第f组传感数据序列，提取第f组与字符识别相关的特征信息；
    将第f组所述特征信息输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与N个不同的备选字符的匹配概率；
    从所述N个不同的备选字符中选择匹配概率最高的M个备选字符作为与所述第f组传感数据序列对应的字符信息；其中M＜N。
  3. 如权利要求2所述的智能手表的文本输入方法，其特征在于，所述字符为英文字母，N＝26，所述N个备选字符为字母a～字母z。
  4. 如权利要求2所述的智能手表的文本输入方法，其特征在于，所述通过惯性传感器检测由用户手写引起的一组或多组传感数据序列，具体包括：
    在获得确认用户开始手写文本的文本识别启动指令时，通过惯性传感器检测本智能手表当前的传感数据，作为静止状态的传感数据；
    启动文本识别，通过所述惯性传感器持续采集本智能手表的传感数据；
    将启动文本识别后所采集到的所有传感数据减去所述静止状态的传感数据，得到消除漂移影响的传感数据；
    对所述消除漂移影响的传感数据执行预处理，得到稳定的传感数据；
    监控所述惯性传感器持续采集的传感数据是否发生停顿；在发生第f次停顿时，将第(f-1) 次停顿与第f次停顿之间的所述稳定的传感数据作为所述第f组传感数据序列。
  5. 如权利要求4所述的智能手表的文本输入方法，其特征在于，所述预处理包括归一化处理。
  6. 如权利要求2至5任一项所述的智能手表的文本输入方法，其特征在于，所述x轴的正方向为本智能手表指向用户手背的方向，或者用户手背指向本智能手表的方向；
    所述第f组与字符识别相关的特征信息，包括以下特征信息中的至少一项：
    第f组传感数据序列的持续时间；
    采用动态时间规整法计算得到的存储于本智能手表中的N组传感数据序列样本分别与所述第f组传感数据序列的相似度；其中，所述N组传感数据序列样本分别与N个不同的备选字符一一对应；
    根据所述第f组传感数据序列计算得到的x轴能量总和与y轴能量总和的比值；其中，所述x轴能量总和为本智能手表在x轴上的运动能量总和，所述y轴能量总和为本智能手表在y轴方向上的运动能量总和；
    本智能手表在x轴上的运动能量达到预设阈值的时间点，以及本智能手表在y轴上的运动能量达到预设阈值时的时间点；
    所述第f组传感数据序列的x轴的传感数据序列以及y轴的传感数据序列的相关系数。
  7. 如权利要求2至5任一项所述的智能手表的文本输入方法，其特征在于，所述机器学习模型为随机森林模型；
    所述随机森林模型的配置方法包括：
    输出提示用户在手背上手写指定字符的信息；其中，所述指定字符包括至少N个不同的字符；
    采集多组由用户手写指定字符引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个指定字符对应；
    采用所述训练数据，训练随机森林分类器，得到所述随机森林模型。
  8. 如权利要求1至5任一项所述的智能手表的文本输入方法，其特征在于，所述根据所述字符信息，获得多个不同的备选文本；
    若本智能手表仅检测到一组传感数据序列，则将所述M个备选字符作为M个不同的备选文本；
    若本智能手表检测到多组传感数据序列，则根据多组所述字符信息，组合成多个备选文本；其中每个备选文本包括多个字符，每个字符都是从所述字符信息中抽取一个备选字符得到的。
  9. 一种智能手表的文本输入装置，其特征在于，包括：
    数据检测模块，用于通过惯性传感器检测由用户手写引起的一组或多组传感数据序列；
    字符识别模块，用于根据所述传感数据序列，获取与每一组传感数据序列一一对应的字符信息；其中，每组所述字符信息包括M个备选字符；
    文本识别模块，用于根据所述字符信息，获得多个不同的备选文本；
    文本确认模块，用于从多个所述备选文本中选择与用户手写匹配的文本，作为本智能手表识别的文本；
    其中，本智能手表的惯性传感器包括三轴加速度传感器，所述传感数据序列包括x轴的传感数据序列和y轴的传感数据序列；所述x轴与y轴处于与本智能手表表盘平行的平面。
  10. 如权利要求9所述的智能手表的文本输入装置，其特征在于，所述字符识别模块包括：
    数据获取单元，用于获取本智能手表的惯性传感器检测到的第f组传感数据序列；
    特征信息提取单元，用于根据所述第f组传感数据序列，提取第f组与字符识别相关的特征信息；
    匹配概率计算单元，用于将第f组所述特征信息输入预先配置的机器学习模型，获取从所述机器学习模型输出的第f组所述特征信息与N个不同的备选字符的匹配概率；
    备选字符获取单元，用于从所述N个不同的备选字符中选择匹配概率最高的M个备选字符作为与所述第f组传感数据序列对应的字符信息；其中M＜N。
  11. 如权利要求10所述的智能手表的文本输入装置，其特征在于，所述字符为英文字母，N＝26，所述N个备选字符为字母a～字母z。
  12. 如权利要求10所述的智能手表的文本输入装置，其特征在于，所述数据检测模块包括：
    静止数据检测单元，用于在获得确认用户开始手写文本的文本识别启动指令时，通过惯性传感器检测本智能手表当前的传感数据，作为静止状态的传感数据；
    数据采集单元，用于启动文本识别，通过所述惯性传感器持续采集本智能手表的传感数据；
    消除漂移单元，用于将启动文本识别后所采集到的所有传感数据减去所述静止状态的传感数据，得到消除漂移影响的传感数据；
    预处理单元，用于对所述消除漂移影响的传感数据执行预处理，得到稳定的传感数据；
    数据分组单元，用于监控所述惯性传感器持续采集的传感数据是否发生停顿；在发生第f次停顿时，将第(f-1)次停顿与第f次停顿之间的所述稳定的传感数据作为所述第f组传感数据序列。
  13. 如权利要求12所述的智能手表的文本输入装置，其特征在于，所述预处理包括归一化处理。
  14. 如权利要求10至13任一项所述的智能手表的文本输入装置，其特征在于，所述x轴的正方向为本智能手表指向用户手背的方向，或者用户手背指向本智能手表的方向；
    所述第f组与字符识别相关的特征信息，包括以下特征信息中的至少一项：
    第f组传感数据序列的持续时间；
    采用动态时间规整法计算得到的存储于本智能手表中的N组传感数据序列样本分别与所述第f组传感数据序列的相似度；其中，所述N组传感数据序列样本分别与N个不同的备选字符一一对应；
    根据所述第f组传感数据序列计算得到的x轴能量总和与y轴能量总和的比值；其中，所述x轴能量总和为本智能手表在x轴上的运动能量总和，所述y轴能量总和为本智能手表在y轴方向上的运动能量总和；
    本智能手表在x轴上的运动能量达到预设阈值的时间点，以及本智能手表在y轴上的运动能量达到预设阈值时的时间点；
    所述第f组传感数据序列的x轴的传感数据序列以及y轴的传感数据序列的相关系数。
  15. 如权利要求10至13任一项所述的智能手表的文本输入方法，其特征在于，所述机器学习模型为随机森林模型；
    所述随机森林模型的配置装置包括：
    信息提示模块，包括输出提示用户在手背上手写指定字符的信息；其中，所述指定字符包括至少N个不同的字符；
    训练数据采集模块，用于采集多组由用户手写指定字符引起的传感数据序列，提取与每一组传感数据序列一一对应的特征信息，作为所述随机森林学习模型的多组训练数据；其中，每一组训练数据分别与一个指定字符对应；
    采用所述训练数据，训练随机森林分类器，得到所述随机森林模型。
  16. 如权利要求9至13任一项所述的智能手表的文本输入方法，其特征在于，所述文本识别模块包括；
    单字符识别单元，用于若本智能手表仅检测到一组传感数据序列，则将所述M个备选字符作为M个不同的备选文本；
    词汇识别单元，用于若本智能手表检测到多组传感数据序列，则根据多组所述字符信息，组合成多个备选文本；其中每个备选文本包括多个字符，每个字符都是从所述字符信息中抽取一个备选字符得到的。
PCT/CN2016/109509 2016-08-24 2016-12-12 智能手表的文本输入方法及装置 WO2018036023A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610719040.0A CN106339104B (zh) 2016-08-24 2016-08-24 智能手表的文本输入方法及装置
CN201610719040.0 2016-08-24

Publications (1)

Publication Number Publication Date
WO2018036023A1 true WO2018036023A1 (zh) 2018-03-01

Family

ID=57824534

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/109509 WO2018036023A1 (zh) 2016-08-24 2016-12-12 智能手表的文本输入方法及装置

Country Status (2)

Country Link
CN (1) CN106339104B (zh)
WO (1) WO2018036023A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113033539A (zh) * 2021-03-30 2021-06-25 北京有竹居网络技术有限公司 练字格检测方法、装置、可读介质及电子设备

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107300971B (zh) * 2017-06-09 2019-04-02 深圳大学 基于骨传导振动信号传播的智能输入方法及系统
CN108469912A (zh) * 2018-02-01 2018-08-31 华北电力大学 一种文字输入方法及系统
CN110866499B (zh) * 2019-11-15 2022-12-13 爱驰汽车有限公司 手写文本识别方法、系统、设备及介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101853380A (zh) * 2010-06-18 2010-10-06 华南理工大学 一种基于加速度传感器的手势识别的旋转特征提取方法
US20120323521A1 (en) * 2009-09-29 2012-12-20 Commissariat A L'energie Atomique Et Aux Energies Al Ternatives System and method for recognizing gestures
CN103345627A (zh) * 2013-07-23 2013-10-09 清华大学 动作识别方法和装置
CN103543826A (zh) * 2013-07-30 2014-01-29 广东工业大学 一种基于加速度传感器的手势识别方法
CN104615275A (zh) * 2015-03-02 2015-05-13 三星电子（中国）研发中心 文字输入方法和装置
CN104793747A (zh) * 2015-04-24 2015-07-22 百度在线网络技术（北京）有限公司 通过可穿戴式设备进行输入的方法、装置及系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120323521A1 (en) * 2009-09-29 2012-12-20 Commissariat A L'energie Atomique Et Aux Energies Al Ternatives System and method for recognizing gestures
CN101853380A (zh) * 2010-06-18 2010-10-06 华南理工大学 一种基于加速度传感器的手势识别的旋转特征提取方法
CN103345627A (zh) * 2013-07-23 2013-10-09 清华大学 动作识别方法和装置
CN103543826A (zh) * 2013-07-30 2014-01-29 广东工业大学 一种基于加速度传感器的手势识别方法
CN104615275A (zh) * 2015-03-02 2015-05-13 三星电子（中国）研发中心 文字输入方法和装置
CN104793747A (zh) * 2015-04-24 2015-07-22 百度在线网络技术（北京）有限公司 通过可穿戴式设备进行输入的方法、装置及系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113033539A (zh) * 2021-03-30 2021-06-25 北京有竹居网络技术有限公司 练字格检测方法、装置、可读介质及电子设备
CN113033539B (zh) * 2021-03-30 2022-12-06 北京有竹居网络技术有限公司 练字格检测方法、装置、可读介质及电子设备

Also Published As

Publication number Publication date
CN106339104B (zh) 2019-02-15
CN106339104A (zh) 2017-01-18

Similar Documents

Publication Publication Date Title
US10353476B2 (en) Efficient gesture processing
US11762474B2 (en) Systems, methods and devices for gesture recognition
JP6064280B2 (ja) ジェスチャを認識するためのシステムおよび方法
Ward et al. Activity recognition of assembly tasks using body-worn microphones and accelerometers
Ardüser et al. Recognizing text using motion data from a smartwatch
WO2018036023A1 (zh) 智能手表的文本输入方法及装置
CN100403235C (zh) 信息处理方法和信息处理设备
WO2008069519A1 (en) Gesture/speech integrated recognition system and method
JP2017516176A (ja) 手跡及びジェスチャを認識するための方法とシステム
WO2010011923A1 (en) Enhanced detection of circular engagement gesture
US20200118540A1 (en) Information processing method, information processing device, and computer-readable recording medium recording information processing program
Lu et al. Gesture on: Enabling always-on touch gestures for fast mobile access from the device standby mode
CN109840480B (zh) 一种智能手表的交互方法及交互系统
Sharma et al. Numeral gesture recognition using leap motion sensor
JP6225612B2 (ja) プログラム、情報処理装置、および方法
KR101567154B1 (ko) 다중 사용자 기반의 대화 처리 방법 및 이를 수행하는 장치
KR101875472B1 (ko) 지화 인식 시스템 및 방법
US10228905B2 (en) Pointing support apparatus and pointing support method
US11270109B2 (en) Interactive method and interactive system for smart watch
KR20120127000A (ko) 문자인식을 이용한 이미지 관리 장치 및 그 방법
Darbar et al. Magitext: Around device magnetic interaction for 3d space text entry in smartphone
Zaidi et al. Machine learning aided minimal sensor based hand gesture character recognition
KR101499044B1 (ko) 사용자의 손동작 및 음성에 기초하여 사용자가 의도한 텍스트를 취득하는 웨어러블 컴퓨터 및 사용자가 의도한 텍스트를 취득하는 방법
Huang et al. Grib: Gesture recognition interaction with mobile devices for blind people
Van Zaen et al. Gesture recognition for interest detection in mobile applications

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16914043

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16914043

Country of ref document: EP

Kind code of ref document: A1