CN107992792A - A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer - Google Patents
A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer Download PDFInfo
- Publication number
- CN107992792A CN107992792A CN201710960292.7A CN201710960292A CN107992792A CN 107992792 A CN107992792 A CN 107992792A CN 201710960292 A CN201710960292 A CN 201710960292A CN 107992792 A CN107992792 A CN 107992792A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- handwriting
- image
- aerial
- chinese character
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/28—Character recognition specially adapted to the type of the alphabet, e.g. Latin alphabet
- G06V30/287—Character recognition specially adapted to the type of the alphabet, e.g. Latin alphabet of Kanji, Hiragana or Katakana characters
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Multimedia (AREA)
- Character Discrimination (AREA)
Abstract
The present invention discloses a kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer.System includes electronic equipment and the identification equipment of information interchange is carried out with electronic equipment, and electronic equipment includes handwriting input instruction acquiring unit, the collection of aerial handwriting tracks and processing unit and transmitting element;Identification equipment includes receiving unit, pretreatment unit and aerial handwritten Kanji recognition unit.User's hand-hold electronic equipments carry out sky-writing, the electronic equipment is based on the three dimension acceleration sensor tracking skyborne handwriting input of user, three-dimensional handwriting input data are produced, 3 D motion trace projection is produced into the two dimensional image for handwriting recognition and is sent to identification equipment;After identification equipment receives two dimensional image, image is pre-processed, and based on pretreated two dimensional image, completes Chinese Character Recognition.The present invention can be widely applied in the man-machine interactive systems such as interactive entertainment scene, TV control, tutoring system from the limitation of any screen or plane.
Description
Technical field
The invention belongs to computer application and area of pattern recognition, more particularly to one kind to be based on the aerial hand of acceleration transducer
Writing of Chinese characters identifying system and method.
Background technology
In recent years, the popularization of smart machine such as smart mobile phone, tablet computer etc. so that the recreational and work(of portable set
Energy property is more and more stronger.In today that device intelligence is constantly lifted, people propose the comfort level of human-computer interaction wanting for higher
Ask.Handwriting recognition is also a kind of mode of human-computer interaction, it is one kind application of the pattern-recognition in field of human-computer interaction.Hand-written knowledge
It is not divided into aerial handwriting recognition and plane handwriting recognition.The application of wherein plane handwriting recognition is quite varied, disappears all kinds of
The application of plane handwriting recognition technology can be seen in expense electron-like such as mobile phone, tablet computer, PC.Plane handwriting recognition technology
Develop into more mature now, the substantially all built-in ripe plane handwriting recognition algorithm of present smart machine, uses this
Algorithm carries out handwriting input to control smart machine, its accuracy rate identified, adaptability can meet making for vast majority of people
With requiring.But plane handwriting recognition also possesses the limitations such as inconvenient, user experience is bad.In recent years, aerial handwriting recognition
The extensive concern of people is obtained.Aerial handwriting recognition has broken away from the constraint of single plane so that it is hand-written can be in three-dimensional sky
Between in freely carry out.User only needs with a hands movement, to make specific gesture, just can transmit specific information, as led to
Cross aerial hand-written opening television set, into computer game etc..The implementation of aerial handwriting recognition mainly has two kinds:One kind is to use
Camera catches aerial hand-written gesture, is contained the special of gesture by the location technology based on computer vision and special algorithm
Justice, which analyzes, to be come.For this method to being had higher requirements using the aerial hand-written environment of camera, camera can not be in dark surrounds
Lower work.In addition it is high to camera required precision, cost is added, is unfavorable for the popularization of aerial handwriting recognition technology.Aerial hand
Another for writing identification is achieved in that by the aerial hand-written exercise data of inertial sensor collection, is passed through and is analyzed movement number
According to extraction correlated characteristic vector simultaneously identifies classification by mode identification method.Inertial sensor refers to acceleration transducer, gyro
Instrument and magnetometric sensor, in current in the market, the smart machine of mainstream is all equipped with including micro- electricity of all inertial sensors
Sub- mechanical system.No matter for cost or ease for use, the aerial handwriting recognition based on inertial sensor is all aerial hand-written
The optimal implementation of identification.
The content of the invention
The invention discloses a kind of aerial handwritten Chinese character recognition system based on acceleration transducer, and provide a kind of sky
Middle Chinese handwriting identifying method, to improve the accuracy rate of aerial handwritten Kanji recognition.
One aspect of the present invention discloses a kind of aerial handwritten Chinese character recognition system based on acceleration transducer, the system bag
An electronic equipment and an identification equipment that information interchange is carried out with the electronic equipment are included, electronic equipment refers to including handwriting input
Make acquiring unit, the collection of aerial handwriting tracks and processing unit and transmitting element;Identification equipment includes receiving unit, pretreatment
Unit and aerial handwritten Kanji recognition unit.Wherein:
The handwriting input instruction acquiring unit is used to receive the instruction for starting to write and terminating writing.
The aerial handwriting tracks collection and processing unit pass through the three dimension acceleration sensor being disposed therein, collection electricity
The skyborne three-dimensional handwriting input data of sub- equipment, and three-dimensional data is changed and is projected on a two dimensional surface, produce use
In the two dimensional image of handwriting recognition.
The transmitting element is used to send two-dimentional handwriting tracks image to identification equipment.
The receiving unit is used for the two-dimentional handwriting tracks image for receiving transmitting element transmission.
The pretreatment unit is used to pre-process the two dimensional image of reception, reduces the redundancy letter in image path
Breath.
The aerial handwritten Kanji recognition unit is used to carry out Chinese Character Recognition to the two dimensional image by pretreatment.
On the other hand, the present invention provides a kind of aerial Chinese handwriting identifying method, mainly include the following steps that:
Step S1, detects whether to get the instruction for starting writing;
Step S2, if getting the instruction for starting writing, goes to step S3, otherwise return to step S1;
Step S3, the three-dimensional acceleration data of electronic equipment is obtained by three dimension acceleration sensor, calculates three-dimensional motion
Coordinate, is configured to the two dimensional image of handwriting recognition based on three-dimensional motion coordinate;
Step S4, detects whether to get the instruction for terminating writing, if got, preserves the X-Y scheme that projection obtains
Picture, goes to step S5, otherwise return to step S3;
Step S5, identification equipment is sent to by the two-dimentional handwriting tracks image of preservation;
Step S6, identification equipment receive two-dimentional handwriting tracks image;
Step S7, docks received two-dimentional handwriting tracks image and is pre-processed;
Step S8, extracts Hanzi features, and Chinese character is identified using convolutional neural networks, exports recognition result;
Step S9, if getting the instruction for starting writing again, return to step S1, otherwise terminates.
Aerial hand-written character recognition method proposed by the present invention, has the following advantages:
(1) identification is convenient and efficient, can meet the individual demand of different groups;
(2) track based on three dimension acceleration sensor rather than camera and gather three-dimensional handwriting input data, and by three-dimensional
Movement locus floor projection obtains two-dimentional hand-written image, prepares for follow-up identification, can adapt to the requirement of varying environment;
(3) image, is cut into the characteristic area of series of stable, to these area sizes by the method for using elastic cutting
A series of SIFT extracted after normalization describe subcharacter and are spliced, and feature of the composition with superperformance, can adapt to not
With the requirement of writing style, there is good robustness;
(4) Chinese Character Recognition grader of the invention is had more outstanding using the higher convolutional neural networks of accuracy of identification
Performance.
Aerial hand-written discrimination system proposed by the present invention and method, a kind of novel easily writing side is provided for people
Formula, more hommization and intelligence.The invention can be widely applied to interactive entertainment scene, TV control, tutoring system et al.
In machine interactive system.
Brief description of the drawings
Fig. 1 is the structure diagram of the aerial handwritten Chinese character recognition system based on acceleration transducer;
Fig. 2 is the flow chart of aerial Chinese handwriting identifying method;
Fig. 3 is the schematic diagram that electronic equipment provided in an embodiment of the present invention produces the movement of third dimension direction;
Fig. 4 is the aerial handwritten Chinese character track schematic diagram of generation, which is sampled A, B, C, D, E, F, G totally 7
A, its midpoint A is the starting point of sky-writing track, and point G is the terminal of sky-writing track.
Fig. 5 is the basic structure schematic diagram for identifying the convolutional neural networks of Chinese character.
Embodiment
With reference to the accompanying drawings, and in conjunction with specific embodiments, technical scheme is described in further detail, but this
The embodiment not limited to this of invention.
Fig. 1 is the structure diagram of the present invention, as shown in Figure 1, aerial handwritten Chinese character recognition system is by an electronic equipment
Formed with the identification equipment of one and electronic equipment progress information interchange, electronic equipment includes handwriting input instruction and obtains list
First, aerial handwriting tracks collection and processing unit and transmitting element;Identification equipment includes receiving unit, pretreatment unit and sky
Middle handwritten Kanji recognition unit.Wherein:
The handwriting input instruction acquiring unit is used to receive the instruction for starting to write and terminating writing, the one of the present invention
In embodiment, the instruction for starting or terminating handwriting input can be used as by pressing the specific keys of electronic equipment, when electronics is set
It is standby when receiving instruction, entrance or terminate handwriting input state.
The handwriting input instruction acquiring unit receives the instruction for starting writing, described aerial into after writing state
Handwriting tracks gather and processing unit measures and gathers the three-dimensional handwriting input data of input equipment, including input equipment is in X, Y
When the acceleration information moved with Z-direction, wherein reference frame are set as and receive the beginning handwriting input instruction
Input equipment it is perpendicular or parallel.Sample point is gathered using sampling rate set in advance, by calculate each sample point in X, Y and
The three-dimensional coordinate of Z axis, forms corresponding 3 D motion trace, and 3 D motion trace is projected on suitable two dimensional surface,
Produce the two-dimentional hand-written image for being sent to identification equipment.When handwriting input instruction acquiring unit receives the instruction for terminating to write
When, represent that a Chinese-character writing finishes, current writing terminates, and the handwriting input electronic equipment preserves the two dimension currently extracted
Hand-written image.The aerial handwriting tracks collection and processing unit further by three dimension acceleration sensor, control module and on
Position machine processing module composition.
The three dimension acceleration sensor is used for measuring X, Y that electronic equipment moves in the air and the acceleration of Z-direction;
Fig. 3 is the schematic diagram that electronic equipment provided in an embodiment of the present invention produces the movement of third dimension direction;Fig. 4 is the aerial hand-written of generation
Chinese character track schematic diagram, the aerial track are sampled A, B, C, D, E, F, G totally 7 points, its midpoint A is sky-writing track
Starting point, point G are the terminal of sky-writing track.
The control module gathers three-dimensional handwriting input data and control circuit, the number that will be collected with certain sample frequency
According to being transferred to host computer;
The host computer processing module is responsible for realizing the Core Feature of aerial handwriting tracks collection and processing unit, including by mistake
Difference processing, the calculating of three-dimensional motion coordinate and the formation of 3 D motion trace, projection produce two-dimentional hand-written image, and preserve aerial
The two dimensional image of handwriting trace projection generation.
The transmitting element is used to send two-dimentional handwriting tracks image to identification equipment.
The receiving unit is used for the two-dimentional handwriting tracks image for receiving transmitting element transmission.
The pretreatment unit is used to pre-process the two dimensional image of reception, reduces the redundancy in image path and makes an uproar
Sound.
The aerial handwritten Kanji recognition unit is used to carry out Chinese Character Recognition to the two dimensional image by pretreatment.The present invention
An embodiment in, using the method for elastic cutting, image is cut into the characteristic area of series of stable, it is big to these regions
A series of SIFT extracted after small normalization describe subcharacter and are spliced, feature of the composition with superperformance;Finally by
Nearest prototype grader determines candidate Chinese character collection, and obtains recognition result using convolutional neural networks.The tool of wherein elastic cutting
Body step is as follows:
First to image, Nonhomogeneous Elastic divides three rectangles in the horizontal direction, and the principle of division is to allow each rectangle
Inside there is equal Chinese character track pixel as far as possible, Nonhomogeneous Elastic then is carried out to each rectangle segment in vertical direction
Division so that each rectangle segment is divided into three equal small rectangle segments of Chinese character track pixel as far as possible, finally may be used
Obtain 9 subimage blocks;Using the first vertical rear horizontal cutting handwriting tracks image again of identical method, other 9 sons are obtained
Image block.
Fig. 2 is the flow chart of aerial Chinese handwriting identifying method proposed by the present invention, and the step is described as follows:
Step S1, detects whether to get the instruction for starting writing, can be by pressing electricity in one embodiment of the invention
The specific keys of sub- equipment are as the instruction for starting handwriting input, when electronic equipment receives instruction, into handwriting input shape
State;
Step S2, when getting the instruction for starting writing, goes to step S3, otherwise return to step S1;
Step S3, gathers three-dimensional handwriting input data, and is configured to based on corresponding three-dimensional handwriting input data
The two dimensional image of handwriting recognition, mainly including following steps:
Step S31, control module gather three-dimensional handwriting input data according to sampling rate set in advance, including acceleration passes
The electronic equipment that sensor obtains is set as and receives the beginning in X, Y and the acceleration of Z-direction, wherein reference frame
Electronic equipment when handwriting input instructs is perpendicular or parallel;
Step S32, host computer processing module calculate three-dimensional motion coordinate, form 3 D motion trace;
Step S33, host computer processing module project to three-dimensional track on two dimensional surface, obtain two-dimentional hand-written image.This
In one most preferred embodiment of invention, following methods are used to derive suitable two-dimensional projection respectively for the three-dimensional track of each Chinese character
Plane:
According to geometrical principle, a suitable two-dimensional projection plane is a plane, each sample point to the plane away from
From square summation it is minimum.Assuming that the coordinate of n sample point is as follows:(x1,y1,z1),(x2,y2,z2)...(xn,yn,zn), plane
Equation is Ax+By+Cz+D=0, wherein A2+B2+C2≠0.Arbitrary 3 D coordinate (xi,yi,zi) arrive plane distanceMake square distance and
In A2+B2+C2Under=1 constraint, A, B, C, the value of D, equationof structure can be asked by the method for Lagrange multiplier
G (A, B, C, D)=F'(A, B, C, D)+λ (A2+B2+C2+D2- 1), wherein λ is Lagrange multiplier, it is a constant.G(A,
B, C, D) on A, B, C and D partial differential equation it is as follows:
Following equation can be obtained from aforementioned four equation:
A2+B2+C2+D2=1 (5)
Wherein equation (4) can be rewritten as:
Equation (6) is substituted into (1), (2), (3) can be in the hope of the value of A, B, C, D.
After obtaining two-dimensional projection plane equation Ax+By+Cz+D=0, with reference to the equation of the straight line perpendicular to projection planeIt can obtain the corresponding two-dimensional coordinate (x, y) of each three-dimensional sample point.Wherein:
Step S4, detects whether to get the instruction for terminating writing, if got, preserves the X-Y scheme that projection obtains
Picture, goes to step S5, otherwise return to step S3, in one embodiment of the invention, can by press electronic equipment it is specific by
Key is as the instruction for terminating current handwriting input;
Step S5, identification equipment is sent to by the two-dimentional handwriting tracks image of preservation;
Step S6, identification equipment receive two-dimentional handwriting tracks image;
Step S7, docks received two-dimentional handwriting tracks image and is pre-processed, specifically included:
(1) redundancy is removed.When tracing point is overlapped or is within close proximity, can be removed in the suitable threshold value of local setting superfluous
Yu Dian;
(2) interpolation processing.When the sampling rate of the collection of aerial handwriting tracks and processing unit setting is relatively low, it will usually make
The points of one Chinese character are very few.Interpolation processing can make Chinese character track have enough information;
(3) smoothing filter denoising.Two dimensional image is generating and is being frequently subjected to various noise sources in transmitting procedure
Interference and influence, aerial handwriting tracks gather and the setting of processing unit sampling frequency is unreasonable can similarly make an uproar to picture strip
Sound, can use linear/non-linear wave filter to eliminate miscellaneous image interference, strengthen image appearance feature;
(4) normalize.By the normalization of size and position, the Chinese character in image is transformed into unified size, and to rotation
The Chinese character track turned is corrected.
Step S8, is identified the two-dimentional handwriting tracks image by pretreatment, and output Chinese Character Recognition is as a result, main bag
Include following several steps:
Step S81, extracts the main feature of two-dimentional handwriting tracks image.Step S81 is further made of following steps:
Vertical spring cutting two dimensional image is Local Subgraphs picture block after step S811 is first horizontal.Step elasticity cutting two dimension
The process of image into subimage block is:First to image, Nonhomogeneous Elastic divides three rectangles, the original of division in the horizontal direction
It is then to allow in each rectangle that there is equal Chinese character track pixel as far as possible, then in vertical direction to each rectangle diagram
Block carries out Nonhomogeneous Elastic division so that each rectangle segment is divided into three equal small length of Chinese character track pixel as far as possible
Square segment, finally can obtain 9 subimage blocks.
Step S812 describes subcharacter to subimage block extraction SIFT.Detailed process is:The segment size being first syncopated as
Linear normalization is carried out, then to every block diagram as extracted region SIFT describes subcharacter, the SIFT description son ginsengs that are used in extraction
Number is:Cutting number of regions 2 × 2, statistics direction are 8, and the feature vector dimension that each subgraph block obtains is 32.In the step
SIFT describes subcharacter by the gradient vector in subgraph block calculating all pixels, then to the two dimension centered on segment center
Gaussian function is weighted the amplitude of all gradient vectors, then the subregion the cutting of regional area rule for N × N,
The gradient orientation histogram in D direction is counted on per sub-regions, each gradient direction accumulated value is calculated and forms a seed point,
Each seed point has the vector information in 8 directions, finally combines each seed dot into the feature of output.
Horizontal cutting handwriting tracks image again is into topography's block, repeat step S812 after step S813 is first vertical.Tool
Body cutting method is:First to image in vertical direction Nonhomogeneous Elastic divide three rectangles, the principle of division be allow it is each
There is equal Chinese character track pixel as far as possible, then in the horizontal direction in the same way to each rectangle in rectangle
Cutting is carried out, can finally obtain other 9 subimage blocks.
Step S814 splices all Local Subgraphs picture block SIFT features and forms final feature vector.By by step S812 and
18 segment features that step S813 is obtained carry out sequential concatenation, one 576 (18 × 32) dimensional feature vector are obtained, then to this
Each element carries out x in a vector0.4Conversion, convert gained feature vector be final output feature vector.
Step S82, aerial handwritten Kanji recognition.The step is further made of following steps:
Step S821, the spy of the Hanzi features vector and all Chinese characters in template that are extracted by Nearest prototype classifier calculated
The Euclidean distance of vector is levied, candidate Chinese character collection of 10 Chinese characters as identification before selection.
Step S822, inputs convolutional neural networks by the feature vector of extraction, carries out Chinese Character Recognition, and exports identification knot
Fruit, the basic structure of convolutional neural networks that the present invention uses for:First layer convolutional layer, using the convolution kernel of 3*3 sizes
(conv1~conv4), padding 1, activation primitive are ReLu functions (relu1~relu6);First layer max_pooling
Layer (pooling1), using the core of 2*2;Second and third layer of convolutional layer, using the convolution kernel of 3*3, padding 1, activates letter
Number is ReLu functions;4th layer of convolutional layer, using the convolution kernel of 2*2, activation primitive is ReLu functions;The full articulamentum of first layer
(fc1~fc2), shares 4096 neurons, and activation primitive is ReLu functions, and increase dropout layers in the training process with
Machine abandons some weights;The full articulamentum of the second layer, shares 4096 neurons, and activation primitive is ReLu functions;It is finally
Softmax layers, export the 10 Chinese character classifications concentrated for candidate Chinese character.
Examples detailed above is preferred embodiments of the present invention, but embodiments of the present invention and from the limit of examples detailed above
System, other any spirit without departing from the present invention and change, modification or the replacement made under technology, should be equivalent displacement,
It is included within protection scope of the present invention.
Claims (10)
- A kind of 1. aerial handwritten Chinese character recognition system based on acceleration transducer, it is characterised in that including an electronic equipment and One identification equipment that information interchange is carried out with electronic equipment;The electronic equipment receives input instruction, gathers three-dimensional handwriting input data, and three-dimensional handwriting input data are corresponding The two-dimentional hand-written image of 3 D motion trace projection generation, specifically includes:Handwriting input instructs acquiring unit, for receiving the instruction for starting to write and terminating writing;Aerial handwriting tracks collection and processing unit, for tracking and gathering the exercise data of electronic equipment in three dimensions, And project to 3 D motion trace on one two dimensional surface, produce the two dimensional image for handwriting recognition;Transmitting element, for sending two-dimentional handwriting tracks image to identification equipment;The identification equipment is exchanged with electronic equipment, and the two dimensional image of reception is pre-processed, and using two dimensional image as Two-dimentional handwriting recognition is completed on basis, is specifically included:Receiving unit, for receiving the two-dimentional handwriting tracks image of transmitting element transmission;Pretreatment unit, for being pre-processed to the two dimensional image of reception, reduces the redundancy in image path;Aerial handwritten Kanji recognition unit, for identifying the Chinese character in the two dimensional image by pretreatment.
- 2. a kind of aerial handwritten Chinese character recognition system based on acceleration transducer according to claim 1, its feature exist In,The aerial handwriting tracks collection and processing unit include:Three dimension acceleration sensor, for measuring the three-dimensional handwriting input data of electronic equipment, three-dimensional handwriting input data include The acceleration in electronic equipment X, Y, Z axis direction in rectangular coordinate system in space;Control module, with the three-dimensional handwriting input data of sample frequency collection of setting, by the data sending collected to host computer Processing module;Host computer processing module, is responsible for realizing the Core Feature of aerial handwriting tracks collection and processing unit, including three-dimensional motion The calculating of coordinate and the formation of 3 D motion trace, projection produce two-dimentional hand-written image, and preserve the projection life of sky-writing track Into two dimensional image.
- 3. a kind of aerial handwritten Chinese character recognition system based on acceleration transducer according to claim 2, its feature exist In host computer processing module calculates the three-dimensional motion coordinate of each sample point according to three-dimensional handwriting input data;Host computer processing Module is derived for projection with the minimum criterion of the summation of the square distance of sample point in 3 D motion trace to projection plane Two dimensional surface.
- 4. a kind of aerial handwritten Chinese character recognition system based on acceleration transducer according to claim 1, its feature exist In the aerial handwritten Kanji recognition unit completes the feature extraction and identification of two-dimensional character image;The two-dimensional character image Feature extraction using the method for elastic cutting, image is cut into the characteristic area of series of stable, to these area sizes A series of SIFT (Scale-invariant feature transform) the description subcharacters extracted after normalization are spelled Connect, obtain the Hanzi features with superperformance;Calculate the HCL2000 hands that the Hanzi features is collected with National 863 plan The Euclidean distance of the feature vector of all Chinese characters in writing of Chinese characters sample database, preceding 10 Chinese characters of selected distance minimum are as time Select Chinese Character Set;The identification of two-dimensional character image is completed using convolutional neural networks.
- 5. a kind of aerial handwritten Chinese character recognition system based on acceleration transducer according to claim 4, its feature exist In the basic structure of convolutional neural networks is:First layer convolutional layer, using the convolution kernel of 3*3 sizes, padding 1, activation primitive is ReLu functions;First layer max_ Pooling layers, using the core of 2*2;Second and third layer of convolutional layer, is using the convolution kernel of 3*3, padding 1, activation primitive ReLu functions;4th layer of convolutional layer, using the convolution kernel of 2*2, activation primitive is ReLu functions;The full articulamentum of first layer, shares 4096 neurons, activation primitive is ReLu functions, and increases dropout layers of some weights of random drop in the training process; The full articulamentum of the second layer, shares 4096 neurons, and activation primitive is ReLu functions;It is finally Softmax layers, exports as candidate 10 Chinese character classifications in Chinese Character Set.
- 6. utilize one kind of the aerial handwritten Chinese character recognition system based on acceleration transducer described in 5 any one of Claims 1 to 5 Aerial Chinese handwriting identifying method, it is characterised in that this method comprises the following steps:Step S1, handwriting input instruction acquiring unit detect whether to get the instruction for starting writing;Step S2, if getting the instruction for starting writing, goes to step S3, otherwise return to step S1;Step S3, the three-dimensional handwriting input data of aerial handwriting tracks collection and processing unit collection, and with corresponding three-dimensional hand-written The two dimensional image of handwriting recognition is configured to based on input data;Step S4, detects whether to get the instruction for terminating writing, if got, preserves the two dimensional image that projection obtains, turns To step S5, otherwise return to step S3;The two-dimentional handwriting tracks image of preservation is sent to identification equipment by step S5, transmitting element;Step S6, the receiving unit of identification equipment receive two-dimentional handwriting tracks image;Step S7, the received two-dimentional handwriting tracks image of pretreatment unit docking are pre-processed;Step S8, is identified pretreated two dimensional image, and exports Chinese Character Recognition result;Step S9, if getting the instruction for starting writing again, return to step S1, otherwise terminates.
- 7. a kind of aerial Chinese handwriting identifying method according to claim 6, it is characterised in that step S3 further comprises Following steps:Step S31, three dimension acceleration sensor measure electronic equipment in X, Y and the acceleration of Z-direction, wherein reference frame The electronic equipment for being set as and receiving when the beginning handwriting input instructs is perpendicular or parallel;Step S32, host computer processing module calculate three-dimensional motion coordinate, form 3 D motion trace;Step S33, host computer processing module project to three-dimensional track on two dimensional surface, with the three-dimensional coordinate of sample point to two dimension The minimum criterion of total sum of squares of the distance of plane, the two dimensional image for handwriting recognition is derived from three-dimensional track.
- A kind of 8. aerial Chinese handwriting identifying method according to claim 6, it is characterised in that pretreatment described in step S7 Including removing redundancy, interpolation processing, smoothing filter denoising, normalization and image thinning.
- 9. a kind of aerial Chinese handwriting identifying method according to claim 6, it is characterised in that step S8 further comprises Following steps:Step S81, extracts the main feature of two-dimentional handwriting tracks image;Step S82, aerial handwritten Kanji recognition.
- 10. a kind of aerial Chinese handwriting identifying method according to claim 9, it is characterised in that step S81 is further wrapped Include following steps:Step S811, it is first horizontal after vertical spring cutting two dimensional image into Local Subgraphs picture block;Step S812, subcharacter is described to subimage block extraction SIFT;Step S813, horizontal cutting handwriting tracks image again is into topography's block, repeat step S812 after first vertical;Step S82 further comprises the steps:Step S821, using the feature vector of all Chinese characters in the feature vector and template of the extraction of Nearest prototype classifier calculated Euclidean distance, candidate Chinese character collection of 10 Chinese characters as identification before selection;Step S822, convolutional neural networks are inputted by feature vector, complete the identification of Chinese character, and export recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710960292.7A CN107992792A (en) | 2017-10-16 | 2017-10-16 | A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710960292.7A CN107992792A (en) | 2017-10-16 | 2017-10-16 | A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107992792A true CN107992792A (en) | 2018-05-04 |
Family
ID=62029773
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710960292.7A Pending CN107992792A (en) | 2017-10-16 | 2017-10-16 | A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107992792A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108551575A (en) * | 2018-06-25 | 2018-09-18 | 汉字造梦(武汉)文化传播有限公司 | A kind of glyphomancy interactive experience system and method |
CN109344793A (en) * | 2018-10-19 | 2019-02-15 | 北京百度网讯科技有限公司 | Aerial hand-written method, apparatus, equipment and computer readable storage medium for identification |
CN109389173A (en) * | 2018-10-13 | 2019-02-26 | 河南理工大学 | A kind of paper score automatic statistical analysis method and apparatus based on M-CNN |
CN109558830A (en) * | 2018-11-27 | 2019-04-02 | 钟祥博谦信息科技有限公司 | Hand-written recognition method, device and equipment |
CN109871849A (en) * | 2019-01-11 | 2019-06-11 | 西安艾润物联网技术服务有限责任公司 | A kind of method and system of vehicle identification |
CN110119226A (en) * | 2019-05-22 | 2019-08-13 | 深圳市绘王动漫科技有限公司 | Hand input device, hand-written input system and hand-written inputting method |
CN110633673A (en) * | 2019-09-17 | 2019-12-31 | 江苏科技大学 | Handwriting track modification method |
CN110633666A (en) * | 2019-09-10 | 2019-12-31 | 江南大学 | Gesture track recognition method based on finger color patches |
CN110781734A (en) * | 2019-09-18 | 2020-02-11 | 长安大学 | Children cognitive game system based on paper-pen interaction |
CN112162635A (en) * | 2020-09-27 | 2021-01-01 | 华南理工大学 | Method for generating and synthesizing model of in-air handwritten word sequence |
CN112533040A (en) * | 2020-12-14 | 2021-03-19 | 北京百度网讯科技有限公司 | Method and device for remotely controlling television application process and electronic equipment |
CN113012800A (en) * | 2021-04-19 | 2021-06-22 | 联仁健康医疗大数据科技股份有限公司 | Multipurpose clamping method, multipurpose clamping system, medical robot and storage medium |
WO2021134795A1 (en) * | 2020-01-03 | 2021-07-08 | Byton Limited | Handwriting recognition of hand motion without physical media |
CN114842146A (en) * | 2022-05-10 | 2022-08-02 | 中国民用航空飞行学院 | Civil aviation engine maintenance manual and work card modeling method and storable medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1485711A (en) * | 2002-09-28 | 2004-03-31 | �ʼҷ����ֵ��ӹɷ�����˾ | Three-dimensional hand-written identification process and system thereof |
CN101488182A (en) * | 2008-12-24 | 2009-07-22 | 华南理工大学 | Image characteristics extraction method used for handwritten Chinese character recognition |
CN101853380A (en) * | 2010-06-18 | 2010-10-06 | 华南理工大学 | Rotary characteristic extraction method based on gesture identification of acceleration sensor |
CN102810008A (en) * | 2012-05-16 | 2012-12-05 | 北京捷通华声语音技术有限公司 | Air input system, method and air input acquisition equipment |
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 |
-
2017
- 2017-10-16 CN CN201710960292.7A patent/CN107992792A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1485711A (en) * | 2002-09-28 | 2004-03-31 | �ʼҷ����ֵ��ӹɷ�����˾ | Three-dimensional hand-written identification process and system thereof |
CN101488182A (en) * | 2008-12-24 | 2009-07-22 | 华南理工大学 | Image characteristics extraction method used for handwritten Chinese character recognition |
CN101853380A (en) * | 2010-06-18 | 2010-10-06 | 华南理工大学 | Rotary characteristic extraction method based on gesture identification of acceleration sensor |
CN102810008A (en) * | 2012-05-16 | 2012-12-05 | 北京捷通华声语音技术有限公司 | Air input system, method and air input acquisition equipment |
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 |
Non-Patent Citations (3)
Title |
---|
刘排排: "空中手写字符串识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
王玮等: "空中手写笔笔迹检测***", 《光学精密工程》 * |
高学等: "基于CNN和随机弹性形变的相似手写汉字识别", 《华南理工大学学报(自然科学版)》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108551575A (en) * | 2018-06-25 | 2018-09-18 | 汉字造梦(武汉)文化传播有限公司 | A kind of glyphomancy interactive experience system and method |
CN109389173A (en) * | 2018-10-13 | 2019-02-26 | 河南理工大学 | A kind of paper score automatic statistical analysis method and apparatus based on M-CNN |
CN109344793A (en) * | 2018-10-19 | 2019-02-15 | 北京百度网讯科技有限公司 | Aerial hand-written method, apparatus, equipment and computer readable storage medium for identification |
CN109558830A (en) * | 2018-11-27 | 2019-04-02 | 钟祥博谦信息科技有限公司 | Hand-written recognition method, device and equipment |
CN109871849A (en) * | 2019-01-11 | 2019-06-11 | 西安艾润物联网技术服务有限责任公司 | A kind of method and system of vehicle identification |
CN110119226B (en) * | 2019-05-22 | 2023-03-28 | 深圳市绘王动漫科技有限公司 | Handwriting input device, handwriting input system and handwriting input method |
CN110119226A (en) * | 2019-05-22 | 2019-08-13 | 深圳市绘王动漫科技有限公司 | Hand input device, hand-written input system and hand-written inputting method |
CN110633666A (en) * | 2019-09-10 | 2019-12-31 | 江南大学 | Gesture track recognition method based on finger color patches |
CN110633673A (en) * | 2019-09-17 | 2019-12-31 | 江苏科技大学 | Handwriting track modification method |
CN110633673B (en) * | 2019-09-17 | 2023-04-18 | 江苏科技大学 | Handwriting track modification method |
CN110781734A (en) * | 2019-09-18 | 2020-02-11 | 长安大学 | Children cognitive game system based on paper-pen interaction |
CN110781734B (en) * | 2019-09-18 | 2023-04-07 | 长安大学 | Child cognitive game system based on paper-pen interaction |
WO2021134795A1 (en) * | 2020-01-03 | 2021-07-08 | Byton Limited | Handwriting recognition of hand motion without physical media |
CN112162635A (en) * | 2020-09-27 | 2021-01-01 | 华南理工大学 | Method for generating and synthesizing model of in-air handwritten word sequence |
CN112533040A (en) * | 2020-12-14 | 2021-03-19 | 北京百度网讯科技有限公司 | Method and device for remotely controlling television application process and electronic equipment |
CN113012800A (en) * | 2021-04-19 | 2021-06-22 | 联仁健康医疗大数据科技股份有限公司 | Multipurpose clamping method, multipurpose clamping system, medical robot and storage medium |
CN114842146A (en) * | 2022-05-10 | 2022-08-02 | 中国民用航空飞行学院 | Civil aviation engine maintenance manual and work card modeling method and storable medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107992792A (en) | A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer | |
Ren et al. | Robust part-based hand gesture recognition using kinect sensor | |
CN106682598B (en) | Multi-pose face feature point detection method based on cascade regression | |
Li et al. | Deep Fisher discriminant learning for mobile hand gesture recognition | |
Ma et al. | Kinect Sensor‐Based Long‐Distance Hand Gesture Recognition and Fingertip Detection with Depth Information | |
CN105739702B (en) | Multi-pose finger tip tracking for natural human-computer interaction | |
D’Orazio et al. | Recent trends in gesture recognition: how depth data has improved classical approaches | |
CN102810008B (en) | A kind of air input, method and input collecting device in the air | |
CN103839040A (en) | Gesture identification method and device based on depth images | |
CN104616028B (en) | Human body limb gesture actions recognition methods based on space segmentation study | |
CN103150019A (en) | Handwriting input system and method | |
CN102024151B (en) | Training method of gesture motion recognition model and gesture motion recognition method | |
CN103136986A (en) | Sign language identification method and sign language identification system | |
Qi et al. | Computer vision-based hand gesture recognition for human-robot interaction: a review | |
CN105975934A (en) | Dynamic gesture identification method and system for augmented reality auxiliary maintenance | |
CN102567703A (en) | Hand motion identification information processing method based on classification characteristic | |
Schomaker et al. | A handwriting recognition system based on properties of the human motor system | |
Misra et al. | Development of a hierarchical dynamic keyboard character recognition system using trajectory features and scale-invariant holistic modeling of characters | |
CN110866468A (en) | Gesture recognition system and method based on passive RFID | |
Xu et al. | Robust hand gesture recognition based on RGB-D Data for natural human–computer interaction | |
CN111444488A (en) | Identity authentication method based on dynamic gesture | |
Subrahmonia et al. | Pen computing: Challenges and applications | |
Wang et al. | Gesture Feature Extraction and Recognition Based on Image Processing. | |
CN103927555A (en) | Static sign language letter recognition system and method based on Kinect sensor | |
Ahmed et al. | Joint-triplet motion image and local binary pattern for 3d action recognition using kinect |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180504 |
|
RJ01 | Rejection of invention patent application after publication |