CN110196635A - A kind of gesture input method based on wearable device - Google Patents

A kind of gesture input method based on wearable device Download PDF

Info

Publication number
CN110196635A
CN110196635A CN201910351496.XA CN201910351496A CN110196635A CN 110196635 A CN110196635 A CN 110196635A CN 201910351496 A CN201910351496 A CN 201910351496A CN 110196635 A CN110196635 A CN 110196635A
Authority
CN
China
Prior art keywords
input
character string
character
user
hand
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
Application number
CN201910351496.XA
Other languages
Chinese (zh)
Other versions
CN110196635B (en
Inventor
董玮
高艺
曾思钰
刘汶鑫
张文照
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201910351496.XA priority Critical patent/CN110196635B/en
Publication of CN110196635A publication Critical patent/CN110196635A/en
Application granted granted Critical
Publication of CN110196635B publication Critical patent/CN110196635B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
    • G06F3/014—Hand-worn input/output arrangements, e.g. data gloves
    • 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
    • G06F3/017—Gesture based interaction, e.g. based on a set of recognized hand gestures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Machine Translation (AREA)
  • Character Discrimination (AREA)

Abstract

A kind of gesture input method based on wearable device, step are as follows: wrist data set when different letters hand-written by smartwatch acquisition user, and corresponding hand data set is acquired by optical sensor, by the corresponding alphabetical label as them of two datasets；Two groups of exercise datas corresponding to wrist and hand, are cut out compression and time unifying, Wrist-sport data that treated are as hand-written data collection；Extract hand data set in finger tip, refer to the positions such as root, wrist exercise data as benchmark dataset, then calculate separately the length of metacarpal bone and index finger；Neural network is built, by training the model of fit and disaggregated model alphabetical for identification that obtain wrist data to reference data；Word lookup library is established, the construction work of input method is completed.When user inputs, input method can acquire the wrist data of user's input, first it is fitted, classify again, provides corresponding input letter, and input character string is obtained to the alphabetical progress of input sequence combination, then it calculates and character string similar in input character string, go in dictionary to search qualified character string, finally return to user's input character string and with word similar in user inputs character string, guarantee to provide in the case where user's input error and user input similar prediction result.

Description

A kind of gesture input method based on wearable device
Technical field
The present invention relates to a kind of gesture input methods based on wearable device.
Background technique
In recent years, intelligent input method becomes the hot spot studied both at home and abroad, information age convenient and fast input method technology must It is indispensable.Smart phone, smartwatch, smart television of today etc. all support installation intelligent input method, and good input method must Surely efficient interactive experience can be brought to user.Input method will not only provide corresponding defeated in the case where user inputs correct situation Out as a result, also to correct in the case where user's input error to the input of user, correctly output result is provided.Separately On the one hand, associating input function also increasingly has been favored by people, and when user inputs a word, input method can According to the content that user current input prediction user inputs next time, and the selection of prediction result can constantly be adjusted according to user The algorithm of prediction enables the result of prediction to conform better to the expectation of user.In conclusion intelligent input method is people's Increasingly important role is played in life, studying a intelligent input method suitable for public users has potentially using valence Value.
Input method on existing smart machine needs user and inputs letter with the key that hand is gone on a beating keyboard, then Corresponding output result is provided according to the input of user.During user's input, the input of text generally requires two hands It completes, a hand is used to fix input equipment, and another hand carries out input operation in equipment.Under certain special scenes, User is difficult to vacate both hands to carry out input operation, such as user needs to be opened an umbrella with a hand when rainy day, takes express delivery When need to take express delivery with a hand.In this way, user cannot carry out input operation using both hands, and only just with one hand The intelligent input method that can complete input operation can satisfy corresponding user demand.
As a whole, under certain special scenes, singlehanded input method has good application prospect.
Summary of the invention
The present invention will overcome the disadvantages mentioned above of the prior art, provide a kind of gesture input side based on wearable device Method.
Wrist data set of the present invention by smartwatch acquisition user when hand-written different alphabetical, and pass through optical sensor Corresponding hand data set is acquired, the characteristic of these data is extracted, builds neural network, wrist data are obtained to base by training The model of fit of quasi- data and disaggregated model alphabetical for identification, then establish word lookup library, in user's input, adopt The wrist data for collecting user's input, are first fitted, then classify, and provide corresponding input letter, and to the letter of input Progress sequence combination obtains input character string, provides the word to match with user inputs character string finally by dictionary is searched, User completes corresponding input by sentence by screening correct word.
In order to achieve the above object, the technical solution used in the present invention is: a kind of gesture input based on wearable device Method, comprising the following steps:
Step 1, the hand-written data collection and benchmark dataset of user are obtained, comprising:
Wrist-sport data are acquired with intelligence wearing wrist-watch, optical body sense controller acquires hand exercise data；
Step 2, the training stage, in step (1.1) Wrist-sport data and hand exercise data pre-process, instruct Practice neural network and obtain model of fit and classifier, comprising:
(2.1) wrist and hand exercise data when different letters hand-written to user in step (1.1), records corresponding word Mother is used as label C；
(2.2) to the wrist and the corresponding two groups of exercise datas of hand obtained in step (1.1), be cut out compression and when Between be aligned, Wrist-sport data that treated are as hand-written data collection A；
(2.3) to the hand exercise data obtained in step (1.1), the movement number for extracting finger tip, referring to the positions such as root, wrist According to as benchmark dataset B；Calculate separately the length of metacarpal bone and index finger；
(2.4) neural network is built, the neural network model that training obtains is as model of fit Mfitting, MfittingFor Hand-written data collection is fitted to benchmark dataset D；
(2.5) another neural network is constructed, the neural network model that training obtains is as classifier Mclassify, MclassifyThe probability that current input may be each letter can be exported；
(2.6) by MfittingAnd MclassifySolidify and be converted into the model M that mobile terminal usesfitting-liteWith Mclassify-lite；
Step 3, forecast period identifies the input of user and carries out input prediction, comprising:
(3.1) corresponding word lookup library is established；
(3.2) the handwriting input data set of user is acquired；
(3.3) the hand-written data collection in step (3.2) is compressed, obtains compressed hand-written data collection；
(3.4) the compressed hand-written data collection for obtaining step (3.3) as input, first pass through data model of fit into Row fitting, then classified by word identification classifier, obtain the letter of user's input；
(3.5) combined according to the sequence that the current input character of user string, that is, user inputs letter, calculating edit therewith away from From the character string for 1, a candidate characters set of strings is constituted；
(3.6) the candidate characters set of strings obtained using step (3.5) removes inquiry character in the dictionary of step (3.1) foundation The word of conjunction condition；
(3.7) if the result that step (3.6) inquiry obtains is not empty, return query result；If query result is sky, into Row step (3.8), (3.9) and (3.10)；
(3.8) a similar character string is asked again to the candidate characters set of strings in step (3.5), i.e., editing distance is therewith 1 character string obtains the candidate characters set of strings for being 2 with user inputs character string editing distance；
(3.9) the candidate characters set of strings for being 2 with user inputs character string editing distance obtained using step (3.8), Qualified word is inquired in the dictionary for going to step (3.1) to establish；
(3.10) if the result that step (3.9) inquiry obtains is not empty, the character string of return query result and user's input； If query result is sky, the character string of user's input is only returned；
(3.11) user is up or down or left or right direction is waved, that is, corresponding upper and lower, left and right side on dial plate may be selected To word.
The beneficial effects of the present invention are: in terms of input mode, it can be complete in such a way that finger writes letter in the sky At input, when a word input is completed, rotation wrist may be selected by desired word, and input mode is novel, only with single Hand can be completed to input.In terms of Letter identification, 80% is reached to the accuracy rate of alphabetic sort, error of fitting is small, the finger of fitting Accumulated error between tongue mark and actual track is 3.3cm, and the speed of classification is fast, pretreatment and two-stage model on wrist-watch The temporal summation of operation is only 0.28s or so, model it is small in size, solidify compressed two-stage model be respectively 30kb and 90kb or so.In terms of Word prediction, by the similitude between editing distance calculating character string, it can provide and input word with user Accord with the correct word that string editing distance is less than or equal to 2.In terms of input speed, the average used time of one seven alphabetic word of input is The speed of 12s, input are fast.
Detailed description of the invention
Fig. 1 is the work flow diagram of the data acquisition of the method for the present invention, alphabetic sort and word selection.
Fig. 2 is the model training flow chart of the method for the present invention.
Fig. 3 is the Word prediction flow chart of the method for the present invention.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.A kind of gesture input based on wearable device of the invention Method, specific embodiment are as follows:
Step 1, hand-written data collection and benchmark dataset are obtained, comprising:
With intelligence wearing wrist-watch LG Watch Urbane, the Wrist-sport of the hand-written different letters of TicWatch acquisition user Data m group, optical body sense controller Leap Motion acquire the hand exercise data m group of the hand-written different letters of user simultaneously；
Step 2, the training stage, in step (1.1) Wrist-sport data and hand exercise data pre-process, mention It takes data characteristics and establishes property data base and classifier, comprising:
(2.1) exercise data of m group wrist when different letters hand-written to user in step (1.1) and hand, record pair The letter answered is as label C { C1, C2..., Cm}；
(2.2) it to the wrist and the corresponding two groups of exercise datas of hand obtained in step (1.1), is cut out and the time pair Together, and using linear interpolation method it is compressed to identical length l, Wrist-sport data that treated are as hand-written data collection A { A1, A2..., Am, AiSize is l × 6 × 1, i ∈ (1,2 ..., m)；
(2.3) to the hand exercise data obtained in step (1.1), the movement number for extracting finger tip, referring to the positions such as root, wrist According to, and by its space coordinates center translation to the center of metacarpophalangeal joints, as benchmark dataset B { B1, B2..., Bm, BiGreatly It is small for l × 3 × 3, i ∈ (1,2 ..., m)；Calculate separately the length l of metacarpal bonemetacarpalWith the length l of index fingerfinger；
(2.4) convolutional neural networks M is built1, it is convolutional layer 1~6 respectively that each layer successively links from top to bottom；Optimizer For Adam optimizer；Activation primitive is ReLU function；The loss function of model is designed as L1(θ), as shown in formula (1),
Wherein, y is the output of neural network as a result, k is penalty coefficient, for reducing the mistake of sensor acquisition phalanges length Difference；
Using the A obtained in step (2.2) as input, the middle B obtained of step (2.3) is trained as reference data, The neural network model that training obtains is as model of fit Mfitting；MfittingFor hand-written data collection to be fitted to reference data Collection, the data set being fitted are D { D1, D2..., Dm, DiSize is l × 3 × 3, i ∈ (1,2 ..., m)；
(2.5) convolutional neural networks M is constructed2, it is convolutional layer 1~6, flatten respectively that each layer successively links from top to bottom Layer 7, full articulamentum 8, softmax classifier layer 9；Optimizer is Adam optimizer；Activation primitive is ReLU function；The damage of model It loses function and is designed as L2(θ), as shown in formula (2),
Y is the output result of neural network；The D obtained in step (2.4) makees as input, the middle C obtained of step (2.1) For benchmark data, the neural network model that training obtains is as classifier Mclassify, MclassifyCurrent input can be exported may For the probability of each letter；
(2.6) M is readfittingAnd MclassifyThe member figure meta graph and check point file saved after training Checkpoint file specifies output node output node, all necessary nodes is all preserved to obtain curing mold Type M 'fittingWith M 'classify；Using TensorFlow Lite converter by M 'fittingWith M 'classifyIt is converted into movement Hold the model M usedfitting-liteAnd Mclassify-lite；
Step 3, cognitive phase, data prediction and integrated classification device result, comprising:
(3.1) corresponding word lookup library is established by dictionary tree, each letter of each word is inserted into word one by one In allusion quotation tree；It whether there is before insertion referring initially to prefix, if it does, just sharing the prefix character string；Otherwise corresponding node is created The side and；Dictionary tree establish after the completion of, the side between adjacent node represents a character, from root node to the path of a certain node on All Connection operators passed through get up, and represent the corresponding character string of the node, and the corresponding character string of each node not phase Together；
(3.2) the handwriting input data set W of user is acquired；
(3.3) the hand-written data collection in step (3.2) is compressed, obtains compressed hand-written data collection Wcompressed；
(3.4) the compressed hand-written data collection W for obtaining step (3.3)compressedAs input, it is quasi- to first pass through data Close model Mfitting-liteRow fitting, then pass through word identification classifier Mclassify-liteClassify, obtains the word of user's input Female λ；
(3.5) according to the current input character string S of userinputThe sequence that i.e. user inputs letter λ combines, and calculating is compiled therewith Collect the character string S that distance is 1edit, i.e. Edit [Sinput][Sedit]=1 constitutes a candidate characters set of strings Setedit；With there are four types of meet the S of conditioneditSet:
The first can become to input character string S to increase a characterinput'sEspecially by time Go through SinputAnd delete a character and obtain, as shown in formula (3),
Wherein, DeleteChar () is to delete function, for deleting the character in character string on some position, SinputIt is Character string to be deleted, i are SinputThe position of the middle character for needing to delete, n is the length of character string to be deleted；
Second is to delete a character to become to input character string Sinput'sEspecially by Traverse SinputAnd be inserted into a character and obtain, as shown in formula (4),
Wherein, InsertChar () is insertion function, for being inserted into corresponding character on some position in character string, SinputIt is to be inserted into character string, j is SinputThe middle position for needing to be inserted into character, c be insertion character, value range be [a, Z], n is the length for being inserted into character string；
The third can become to input character string S for one character of replacementinput'sEspecially by time Go through SinputAnd replace a character and obtain, as shown in formula (5),
Wherein, AlterChar () is replacement function, for replacing corresponding character on some position in character string, SinputIt is character string to be replaced, k is SinputThe position of the middle character for needing to replace, α are used to replace SinputWord on middle position k Symbol, value range are [a, z], and n is the length of character string to be replaced；
4th kind is that can become S by two adjacent characters of transpositioninput'sIt is specific logical Cross traversal SinputAnd two characters of transposition obtain, as shown in formula (6),
Wherein, TransposeChars () is transposition function, for the character on two neighboring position in transposition character string, SinputIt is to transposition character string, p, p+1 are SinputThe position of the middle character for needing transposition, n are the length to transposition character string；
(3.6) the candidate characters set of strings obtained using step (3.5)In the dictionary for going to step (3.1) to establish Inquire qualified word；If candidate characters set of stringsIn word not in dictionary, then directly give up；Instead Then according to the frequency of occurrences of respective word in dictionary, return to the most top n word of frequency of occurrence as prediction result, specifically As shown in formula (7), (8),
Wherein, InDictionary () function is used to find out candidate characters set of stringsIn be present in dictionary Set of lettersMaxFrequence () function is used to find outN before middle frequency of occurrence Word；
(3.7) if step (3.6) inquires obtained resultIt is not sky, returns to query result；IfFor sky, then step (3.8), (3.9) and (3.10) are carried out；
(3.8) to the candidate characters set of strings in step (3.5)Seek a similar character string again, i.e., withEditing distance is 1, Edit [Sinput][SeditThe character string of]=2 obtains and user inputs character string SinputEditor The candidate characters set of strings that distance is 2
(3.9) the candidate characters set of strings for being 2 with user inputs character string editing distance obtained using step (3.5)Qualified word is inquired in the dictionary for going to step (3.1) to establish；If candidate characters set of strings In word not in dictionary, then directly give up；The frequency of occurrences on the contrary then according to respective word in dictionary returns to frequency of occurrence For most top n words as prediction result, specific such as formula (9), (10) are shown,
Wherein, InDictionary () function is used to find out candidate characters set of stringsIn be present in dictionary and work as In set of lettersMaxFrequence () function is used to find outMiddle frequency of occurrence The word of preceding N；
(3.10) if step (3.9) inquires obtained resultIt is not sky, returns to query resultWith the character string S of user's inputinput；IfFor sky, then the character string of user's input is only returned；
(3.11) user is up or down or left or right direction is waved, that is, corresponding upper and lower, left and right side on dial plate may be selected To word.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention Range should not be construed as being limited to the specific forms stated in the embodiments, and protection scope of the present invention is also and in art technology Personnel conceive according to the present invention it is conceivable that equivalent technologies mean.

Claims (1)

1. a kind of gesture input method based on wearable device, comprising the following steps:
Step 1, hand-written data collection and benchmark dataset are obtained, comprising:
With intelligence wearing wrist-watch LG Watch Urbane, the Wrist-sport data m of the hand-written different letters of TicWatch acquisition user Group, optical body sense controller Leap Motion acquire the hand exercise data m group of the hand-written different letters of user simultaneously；
Step 2, the training stage, in step (1.1) Wrist-sport data and hand exercise data pre-process, extract number According to feature and establish property data base and classifier, comprising:
(2.1) exercise data of m group wrist when different letters hand-written to user in step (1.1) and hand, records corresponding Letter is used as label C { C1,C2,…,Cm}；
(2.2) it to the wrist and the corresponding two groups of exercise datas of hand obtained in step (1.1), is cut out and time unifying, And it is compressed to identical length l using linear interpolation method, Wrist-sport data that treated are as hand-written data collection A { A1, A2,…,Am, AiSize is l × 6 × 1, i ∈ (1,2 ..., m)；
(2.3) to the hand exercise data obtained in step (1.1), the exercise data for extracting finger tip, referring to the positions such as root, wrist, And by its space coordinates center translation to the center of metacarpophalangeal joints, as benchmark dataset B { B1,B2,…,Bm, BiSize is L × 3 × 3, i ∈ (1,2 ..., m)；Calculate separately the length l of metacarpal bonemetacarpalWith the length l of index fingerfinger；
(2.4) convolutional neural networks M is built1, it is convolutional layer 1~6 respectively that each layer successively links from top to bottom；Optimizer is Adam Optimizer；Activation primitive is ReLU function；The loss function of model is designed as L1(θ), as shown in formula (1),
Wherein, y is the output of neural network as a result, k is penalty coefficient, for reducing the error of sensor acquisition phalanges length；
Using the A obtained in step (2.2) as input, the middle B obtained of step (2.3) is trained as reference data, training Obtained neural network model is as model of fit Mfitting；MfittingFor hand-written data collection to be fitted to benchmark dataset, Being fitted obtained data set is D { D1,D2,…,Dm, Di size is l × 3 × 3, i ∈ (1,2 ..., m)；
(2.5) convolutional neural networks M is constructed2, it is convolutional layer 1~6 respectively that each layer successively links from top to bottom, flatten layer 7, Full articulamentum 8, softmax classifier layer 9；Optimizer is Adam optimizer；Activation primitive is ReLU function；The loss letter of model Number is designed as L2(θ), as shown in formula (2),
Y is the output result of neural network；The D obtained in step (2.4) is as input, and the middle C obtained of step (2.1) is as base Quasi- data, the neural network model that training obtains is as classifier Mclassify, MclassifyIt may be each that current input, which can be exported, The probability of a letter；
(2.6) M is readfittingAnd MclassifyThe member figure meta graph and check point file checkpoint saved after training File specifies output node output node, all necessary nodes is all preserved to obtain curing model M 'fittingWith M′classify；Using TensorFlow Lite converter by M 'fittingWith M 'classifyIt is converted into the model that mobile terminal uses Mfitting-liteAnd Mclassify-lite；
Step 3, cognitive phase, data prediction and integrated classification device result, comprising:
(3.1) corresponding word lookup library is established by dictionary tree, each letter of each word is inserted into dictionary tree one by one In；It whether there is before insertion referring initially to prefix, if it does, just sharing the prefix character string；Otherwise corresponding node and side are created； After the completion of dictionary tree is established, the side between adjacent node represents a character, from root node to passing through on the path of a certain node All Connection operators get up, represent the corresponding character string of the node, and the corresponding character string of each node is different from；
(3.2) the handwriting input data set W of user is acquired；
(3.3) the hand-written data collection in step (3.2) is compressed, obtains compressed hand-written data collection Wcompressed；
(3.4) the compressed hand-written data collection W for obtaining step (3.3)compressedAs input, data fitting mould is first passed through Type Mfitting-liteRow fitting, then pass through word identification classifier Mclassify-liteClassify, obtains the alphabetical λ of user's input；
(3.5) according to the current input character string S of userinputI.e. user input letter λ sequence combination, calculate edit therewith away from From the character string S for 1edit, i.e. Edit [Sinput][Sedit]=1 constitutes a candidate characters set of strings Setedit； In, there are four types of the S for the condition that meetseditSet:
The first can become to input character string S to increase a characterinput'sEspecially by traversal SinputAnd delete a character and obtain, as shown in formula (3),
Wherein, DeletChar () is to delete function, for deleting the character in character string on some position, SinputIt is to be deleted Character string, i are SinputThe position of the middle character for needing to delete, n is the length of character string to be deleted；
Second is to delete a character to become to input character string Sinput'sEspecially by traversal SinputAnd be inserted into a character and obtain, as shown in formula (4),
Wherein, InsertChar () is insertion function, for being inserted into corresponding character, S on some position in character stringinput It is to be inserted into character string, j is SinputThe middle position for needing to be inserted into character, c are the characters of insertion, and value range is [a, z], and n is It is inserted into the length of character string；
The third can become to input character string S for one character of replacementinput'sEspecially by traversal SinputAnd replace a character and obtain, as shown in formula (5),
Wherein, AlterChar () is replacement function, for replacing corresponding character, S on some position in character stringinput It is character string to be replaced, k is SinputThe position of the middle character for needing to replace, α are used to replace SinputCharacter on middle position k, takes Being worth range is [a, z], and n is the length of character string to be replaced；
4th kind is that can become S by two adjacent characters of transpositioninput'sEspecially by time Go through SinputAnd two characters of transposition obtain, as shown in formula (6),
Wherein, TransposeChars () is transposition function, for the character on two neighboring position in transposition character string, Sinput It is to transposition character string, p, p+1 are SinputThe position of the middle character for needing transposition, n are the length to transposition character string；
(3.6) the candidate characters set of strings obtained using step (3.5)It is inquired in the dictionary for going to step (3.1) to establish Qualified word；If candidate characters set of stringsIn word not in dictionary, then directly give up；It is on the contrary then According to the frequency of occurrences of respective word in dictionary, the most top n word of frequency of occurrence is returned as prediction result, specific such as public affairs Shown in formula (7), (8),
Wherein, InDictionary () function is used to find out candidate characters set of stringsIn be present in the list in dictionary Set of wordsMaxFrequence () function is used to find outThe list of N before middle frequency of occurrence Word；
(3.7) if step (3.6) inquires obtained resultIt is not sky, returns to query result；If For sky, then step (3.8), (3.9) and (3.10) are carried out；
(3.8) to the candidate characters set of strings in step (3.5)Seek a similar character string again, i.e., withIt compiles Collecting distance is 1, Edit [Sinput][SeditThe character string of]=2 obtains and user inputs character string SinputThe time that editing distance is 2 Select string assemble
(3.9) the candidate characters set of strings for being 2 with user inputs character string editing distance obtained using step (3.5)Qualified word is inquired in the dictionary for going to step (3.1) to establish；If candidate characters set of strings In word not in dictionary, then directly give up；The frequency of occurrences on the contrary then according to respective word in dictionary returns to frequency of occurrence For most top n words as prediction result, specific such as formula (9), (10) are shown,
Wherein, InDictionary () function is used to find out candidate characters set of stringsIn be present in the list in dictionary Set of wordsMaxFrequence () function is used to find outThe list of N before middle frequency of occurrence Word；
(3.10) if step (3.9) inquires obtained resultIt is not sky, returns to query resultWith The character string S of user's inputinput；IfFor sky, then the character string of user's input is only returned；
(3.11) user is up or down or left or right direction is waved, that is, corresponding upper and lower, left and right direction on dial plate may be selected Word.
CN201910351496.XA 2019-04-28 2019-04-28 Gesture input method based on wearable equipment Active CN110196635B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910351496.XA CN110196635B (en) 2019-04-28 2019-04-28 Gesture input method based on wearable equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910351496.XA CN110196635B (en) 2019-04-28 2019-04-28 Gesture input method based on wearable equipment

Publications (2)

Publication Number Publication Date
CN110196635A true CN110196635A (en) 2019-09-03
CN110196635B CN110196635B (en) 2020-07-31

Family

ID=67752241

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910351496.XA Active CN110196635B (en) 2019-04-28 2019-04-28 Gesture input method based on wearable equipment

Country Status (1)

Country Link
CN (1) CN110196635B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110569800A (en) * 2019-09-10 2019-12-13 武汉大学 detection method of handwriting signal
CN115118688A (en) * 2022-08-29 2022-09-27 中航信移动科技有限公司 Method for sending message based on message template, storage medium and electronic equipment

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101930545A (en) * 2009-06-24 2010-12-29 夏普株式会社 Handwriting recognition method and device
CN103226388A (en) * 2013-04-07 2013-07-31 华南理工大学 Kinect-based handwriting method
CN103577843A (en) * 2013-11-22 2014-02-12 中国科学院自动化研究所 Identification method for handwritten character strings in air
CN104156491A (en) * 2014-09-01 2014-11-19 携程计算机技术（上海）有限公司 Mobile terminal and list information retrieval method thereof
CN106598227A (en) * 2016-11-15 2017-04-26 电子科技大学 Hand gesture identification method based on Leap Motion and Kinect
CN106874874A (en) * 2017-02-16 2017-06-20 南方科技大学 Motion state identification method and device
CN107330480A (en) * 2017-07-03 2017-11-07 贵州大学 Hand-written character Computer Identification
CN108664877A (en) * 2018-03-09 2018-10-16 北京理工大学 A kind of dynamic gesture identification method based on range data
CN109145793A (en) * 2018-08-09 2019-01-04 东软集团股份有限公司 Establish method, apparatus, storage medium and the electronic equipment of gesture identification model
KR20190008168A (en) * 2018-12-28 2019-01-23 정우곤 An artificial neural network application technology for converting hand-written Korean sample text into its own font.

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101930545A (en) * 2009-06-24 2010-12-29 夏普株式会社 Handwriting recognition method and device
CN103226388A (en) * 2013-04-07 2013-07-31 华南理工大学 Kinect-based handwriting method
CN103577843A (en) * 2013-11-22 2014-02-12 中国科学院自动化研究所 Identification method for handwritten character strings in air
CN104156491A (en) * 2014-09-01 2014-11-19 携程计算机技术（上海）有限公司 Mobile terminal and list information retrieval method thereof
CN106598227A (en) * 2016-11-15 2017-04-26 电子科技大学 Hand gesture identification method based on Leap Motion and Kinect
CN106874874A (en) * 2017-02-16 2017-06-20 南方科技大学 Motion state identification method and device
CN107330480A (en) * 2017-07-03 2017-11-07 贵州大学 Hand-written character Computer Identification
CN108664877A (en) * 2018-03-09 2018-10-16 北京理工大学 A kind of dynamic gesture identification method based on range data
CN109145793A (en) * 2018-08-09 2019-01-04 东软集团股份有限公司 Establish method, apparatus, storage medium and the electronic equipment of gesture identification model
KR20190008168A (en) * 2018-12-28 2019-01-23 정우곤 An artificial neural network application technology for converting hand-written Korean sample text into its own font.

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110569800A (en) * 2019-09-10 2019-12-13 武汉大学 detection method of handwriting signal
CN110569800B (en) * 2019-09-10 2022-04-12 武汉大学 Detection method of handwriting signal
CN115118688A (en) * 2022-08-29 2022-09-27 中航信移动科技有限公司 Method for sending message based on message template, storage medium and electronic equipment
CN115118688B (en) * 2022-08-29 2022-11-04 中航信移动科技有限公司 Method for sending message based on message template, storage medium and electronic equipment

Also Published As

Publication number Publication date
CN110196635B (en) 2020-07-31

Similar Documents

Publication Publication Date Title
CN111488426B (en) Query intention determining method, device and processing equipment
CN109271529B (en) Method for constructing bilingual knowledge graph of Xilier Mongolian and traditional Mongolian
CN111274365B (en) Intelligent inquiry method and device based on semantic understanding, storage medium and server
CN101404015B (en) Automatically generating a hierarchy of terms
CN112559556A (en) Language model pre-training method and system for table mode analysis and sequence mask
CN103810998B (en) Based on the off-line audio recognition method of mobile terminal device and realize method
CN113065003B (en) Knowledge graph generation method based on multiple indexes
CN104933113A (en) Expression input method and device based on semantic understanding
CN111259987A (en) Method for extracting event main body based on BERT (belief-based regression analysis) multi-model fusion
CN103605665A (en) Keyword based evaluation expert intelligent search and recommendation method
CN101206673A (en) Intelligent error correcting system and method in network searching process
CN109063159A (en) A kind of entity relation extraction method neural network based
CN110196635A (en) A kind of gesture input method based on wearable device
CN110502640A (en) A kind of extracting method of the concept meaning of a word development grain based on construction
CN105678244B (en) A kind of near video search method based on improved edit-distance
CN105740227A (en) Genetic simulated annealing method for solving new words in Chinese segmentation
CN103810162A (en) Method and system for recommending network information
CN104199606A (en) Sliding input method and device
CN112328859B (en) False news detection method based on knowledge-aware attention network
CN112084373A (en) Multi-source heterogeneous network user alignment method based on graph embedding
CN112559747A (en) Event classification processing method and device, electronic equipment and storage medium
CN110427404A (en) A kind of across chain data retrieval system of block chain
CN115422220A (en) Method for converting natural language into SQL (structured query language) based on deep learning model
CN113590818B (en) Government text data classification method based on integration of CNN (carbon fiber network), GRU (grid-like network) and KNN (K-nearest neighbor network)
CN106933380A (en) The update method and device of a kind of dictionary

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