CN103442114B - A kind of identity identifying method based on dynamic gesture - Google Patents

Abstract

The invention discloses a kind of identity identifying method based on dynamic gesture, multidate information when the method utilizes smart mobile phone acceleration transducer acquisition gesture to perform, the DTW high efficiency method adopting combination to relax end points restriction and premature termination carries out coupling certification to gesture multidate information.Relax the restriction of coupling path end points in conjunction with the DTW high efficiency method utilization of relaxing end points restriction and premature termination and solve the authentification failure problem caused because end points does not line up between gesture sequence, also utilize the restriction of bending slope and premature termination strategy to decrease amount of calculation, experiment shows that this method has good result in the precision and efficiency of authentication simultaneously.

Description

A kind of identity identifying method based on dynamic gesture

Technical field

The invention belongs to pattern recognition and identity identifying technology field, relate to a kind of dynamic gesture identity identifying method based on Android platform.

Background technology

Authentication is the process whether true identity of system validation operator conforms to its alleged identity, and in today that mobile phone is universal, the authenticating user identification on mobile phone also becomes a pith of information security.Current mobile phone identity authentication is mainly divided into the authentication based on password and the authentication based on biological characteristic.The conventional authentication based on password has user's pin mode and nine grids unlocking manner, and the common feature based on the authentication of password is that password is easily revealed, and in order to the fail safe of password frequently changes password, makes again password be difficult to safeguard.Authentication based on biological characteristic can as of a user password good alternative method, and biological characteristic is the natural attribute of people, comprises physiological characteristic or the behavioural characteristic of people.Physiological characteristic is inborn feature, comprises the static natures such as face phase, fingerprint, palm shape, sound, iris, retina; Behavioural characteristic is formed by posteriori study or development, comprises the behavioral characteristics such as signature, keystroke, gait, dynamic gesture.Biological characteristic easily as password can not be guessed and forgotten, also can not easily as having be stolen, so, utilize biological character for identity authentication will be a kind of safer reliably, conveniently popular authentication means.

The identity identifying technology based on biological characteristic conventional at present comprises following several:

1. finger print identifying

Finger print identifying is a kind of biometric identity authentication techniques the most ancient and conventional, occupies the share exceeding half in biological characteristic authentication market.Fingerprint is the lines on the finger tips surface of people, the minutias such as abundant breakpoint, crosspoint, binding site are contained in these rough skin lines, these features are unique, are also constancies, can be determined the identity of a people by the comparison of fingerprint.Finger print identifying is exactly utilize image processing techniques to mate the fingerprint gathered, thus differentiates the identity of user.

2. iris authentication

Iris authentication is most convenient in current all biometric authentication technology, the most accurate a kind of, is also 21st century biometrics most with prospects.Iris is annular section between sclera and pupil, and it comprises abundant textural characteristics, and structure is random, is that gene determine, is not easily forged.Contactless iris image acquisition health is easy-to-use, and not by the unexpected environmental impact of light during acquisition, stability is high.

3. face authentication

Face authentication is one of the most difficult research topic in biometric authentication technology field, the extraction of face characteristic is more difficult, the expression that same people is different, position, direction, illumination all can produce larger impact to the extraction of face characteristic, so at present face authentication accuracy than finger print identifying and iris authentication low, but contactless face characteristic information obtains relatively natural and not easily discovers, and good Consumer's Experience makes face authentication become the easiest received biological characteristic authentication mode.

4. signature authentication

Signature authentication is a kind of behavioural characteristic authentication techniques, and signature authentication is divided into static signature certification and on-line signature certification according to data acquiring mode difference.Static signature certification is the accessible image of computer by scanner the character conversion on paper, and extract the features such as texture information and carry out certification.On-line signature certification gathers the written information of user by special board, and signature sequence is converted into image, and records the information such as pressure, acceleration, speed of writing, and the writing style according to user carries out certification to user.

5. vena identification

Vena identification is a kind of biometric authentication technology of by the vein distribution patterns on people's finger, the back of the hand, palm, people's identity being carried out to certification.Vein pattern is a kind of physiological characteristic, and the vein pattern of different people is all different, even if the vein pattern of the right-hand man of same person is all different, is difficult to forge, the very high contactless data acquisition of fail safe, also makes user be very easy to accept.Vena identification utilizes infrared C CD camera to obtain vein image data, uses that binaryzation of surprise attack, refinement means to digital image zooming-out feature, then mates with the vein pattern stored in main frame, thus reach the effect of authentication.

Biological characteristic authentication mode solves the various limitation of traditional password authentication mode, but on intelligent mobile phone platform, the use amount of biological characteristic authentication mode but can not show a candle to password authentication mode, and main cause has following 2 points: be first because the resource on cell phone platform, equipment limit.Current mobile phone all there is no substantially the equipment that can obtain fingerprint, use the words of finger print identifying just to need external equipment, use inconvenience; And iris authentication requires very high to camera, such camera lowest price is 7000 dollars, and mobile phone is also difficult to realize, and on-line signature certification also needs external equipment; The collecting device of vena identification also has particular/special requirement in addition, and design is complicated, and manufacturing cost is high, and product is difficult to miniaturization, portable inapplicable.Next is some drawbacks limit of authentication mode itself.Iris authentication mode is extremely difficult reads Black Eyes feature, and face authentication and sound authentication are all very easy to by the external world's even self impact, and static signature certification is easily stolen and shifts.

Therefore, be necessary to design a kind of novel identity identifying method.

Summary of the invention

Technical problem to be solved by this invention is to provide a kind of identity identifying method based on dynamic gesture, the method adopts the DTW high efficiency method combining and relax end points restriction and premature termination when mating certification, on DTW method basis, limit bending slope, relax coupling path end points and to limit and in conjunction with premature termination strategy, effectively can improve computational efficiency and the gesture coupling authentication precision of DTW method, desirable certification effect can be obtained.

The technical solution of invention is as follows:

Based on an identity identifying method for dynamic gesture, comprise the following steps:

1) gather x when gesture performs, the acceleration information in y, z tri-directions is as test sample book;

2) preliminary treatment is carried out to test data; Described preliminary treatment comprises smoothing denoising and quantification;

3) dynamic time warping (DTW) algorithm improved is adopted to mate with masterplate data test sample book; [time shaft of test pattern, based on the thoery of dynamic programming, bends by DTW algorithm unevenly, until test data feature is alignd with template characteristic.Because smart mobile phone has some time delays when obtaining acceleration information from transducer, the head of the gesture data therefore got and afterbody may comprise some insignificant data, in order to these nonsignificant datas of place to go are to avoid comparatively Iarge-scale system error, just do these to improve] cancel DTW method the 2nd) test sample book that obtains of step aligns with end points when mating between optimum template and limits, allow the starting point in dynamic programming matching path at line segment [(1, 1), (1, ] or [(1 L), 1), (L, 1)], and allow terminal at line segment [((M-L+1), N), (M, ] or [(M N), (N-L+1)), (M, N)], that is the first frame of certain gesture can mate with any frame in L frame before another gesture, and last frame can mate with any frame in another gesture end L frame, and limit Dynamic Programming bending slope between 0.5-2, the amount L utilizing end points to relax and the slope meter of bending calculate the boundary condition of Dynamic Programming,

Wherein, M is optimum template length, and N is test data length, and L is the amount that end points relaxes;

4) the 3rd) carry out Dynamic Programming within the boundary condition that obtains of step, do not need to preserve all Cumulative Distances and frame matching distance, statement column vector D preserves the Cumulative Distance of previous column, and statement column vector d preserves the Cumulative Distance that current column is calculated;

5) calculate in Dynamic Programming and judge whether the Cumulative Distance when prostatitis is all greater than threshold value τ, if be all greater than threshold value τ, authentification failure, premature termination; If be also not all greater than threshold value τ, then proceed to step 6);

6) judge that whether current data frame is the last frame of test data, if not last frame, then next frame assignment to current data frame, and proceed to step 5); If last frame, then get D [M-L+1 ..., M] in minimum value min, and to compare with threshold value τ, if min is less than threshold value, then authentication success, ending method; If min is greater than threshold value τ, then authentification failure, terminates this verification process.

Described optimum template and the defining method of threshold value are:

For user gathers the sample of 15 certain gestures, the DTW efficient matchings method loosening end points restriction is adopted to calculate DTW Cumulative Distance between two between sample, select the sample minimum with other sample Distance geometry to be optimum sample and optimum template, and select the ultimate range between optimum sample and other samples to be threshold value.

In order to adapt to the change that may occur when people's different times performs gesture, adopting template adaptive strategy, replacing original optimum template from by again choosing new optimum template the gesture of certification at set intervals.

In step 1) in, utilize (SensorManager) context.getSystemService (context.SENSOR_SERVICE) to obtain the packaged sensor management object SensorManager of Android system, then obtain acceleration transducer Sensor.TYPE_ACCELEROMETER by SensorManager; The set of frequency of sensors for data is SENSOR_DELAY_GAME;

In step 2) in, adopt simple Moving Average (SMA) filter to the smoothing denoising of test data; Computing formula is: the value after denoising wherein X _ibe i-th test data; N gets a value in 5-10;

In step 2) in, the acceleration information of the floating type collected is converted to the integer data of 33 grades;

In step 3) in, axle match time is divided into 5 sections, (1, L), (L+1, X _a), (X _a+ 1, X _b), (X _b+ 1, N-L), (N-L+1, N), wherein X _a=(2M-N-L)/3, X _b=(2 (2N-M)+L)/3 are all got the most close integer, and must be met X _a>=1, X _b>=L, show that the pass of optimum template length M and test data length N is thus:

$\left\{\begin{array}{c}2M-N\≥3+L\\ 2N-M\≥L\end{array};\right.$

The test data not meeting above formula relation is considered to differ with optimum template too large, cannot carry out dynamic bending, L value 8; 5 sections that bending is divided into, y boundary value is expressed as follows by x:

$y_{\min }=\left\{\begin{array}{cc}0& 1\≤x\≤L\\ \frac{1}{2}(x-L)& L+1\≤x\≤X_b\\ 2(x-N)+M-L& X_b+1\≤x\≤N\end{array}\right.;$

$y_{\max }=\left[\begin{array}{cc}2x+L& 1\≤x\≤X_z\\ \frac{1}{2}(x-N+L)+M& X_a+1\≤x\≤N-L\\ M& N-L+1\≤x\≤N\end{array}\right.;$

When Dynamic Programming is carried out to test gesture every column data, calculate its corresponding boundary value, and only mate the lattice point in border, thus reduce amount of calculation; The matching primitives of each row lattice point has only used 3 lattice points (do not need to preserve the distance matrix and Cumulative Distance matrix that mate between all frames when realizing in method, reduce memory space) of previous column; L value 8.[value of M and N depends on the duration gathering gesture data.After user clicks " beginning gesture " button, smart mobile phone can gather the data of acceleration transducer with fixed frequency, after user clicks " completing gesture " button, terminate image data.The data acquisition number of times carried out in this process is exactly the value of M or N.】

The present invention can utilize existing equipment on mobile phone, facilitates easy-to-use, and effectively can carry out certification to user again.Authentication based on dynamic gesture is the principle according to on-line signature certification, utilize the acceleration transducer of intelligent mobile phone platform, obtain acceleration information when user performs gesture as feature, the gesture sequence prestored and gesture sequence to be certified are mated thus reached the effect of authentication.Matching process adopts dynamic time bending (DTW) principle, object solves gesture sequence in time and spatially inconsistent problem, on the basis of the efficient matchings method of DTW, relax the end points restriction in coupling path simultaneously, and once Cumulative Distance exceeds threshold value, premature termination mates, core of the present invention relaxes the efficient identity identifying method of DTW of end points restriction and premature termination.

Design of the present invention is:

The present invention proposes a kind of gesture authentication method based on intelligent mobile phone sensor equipment, comprising: the data sample first gathering the some gestures of user, preliminary treatment is carried out to this data sample; Then on the basis of DTW method, limit bending slope, relax end points restriction, calculate the boundary condition asking the lattice point of Euclidean distance needed for mating between this data sample with optimum template, best accumulated distance between test data sample and optimum template is asked in last Dynamic Programming, and adopt premature termination method, once Cumulative Distance exceedes threshold value, then gesture authentication failure, if the Cumulative Distance of final optimal path is less than threshold value, then gesture authentication success.

Beneficial effect:

The present invention proposes a kind of identity identifying method based on dynamic gesture.Multidate information when the method utilizes smart mobile phone acceleration transducer acquisition gesture to perform, the DTW high efficiency method adopting combination to relax end points restriction and premature termination carries out coupling certification to gesture multidate information.This matching authentication method is a kind of method raising authentication precision and authentication efficiency combined, relax end points and limit the end points alignment restriction eliminated in DTW method, allow the starting point of dynamic programming path at line segment [(1, 1), (1, ] or [(1 L), 1), (L, 1)], and terminal can at line segment [((M-L+1), N), (M, ] or [(M N), (N-L+1)), (M, N)], require that (0.5-2) can calculate the boundary condition of dynamic programming path according to the slope of dynamic programming path bending again, Dynamic Programming is carried out within boundary condition, once certain arranges all DTW Cumulative Distances be all greater than threshold value, then premature termination.Relax the restriction of coupling path end points in conjunction with the DTW high efficiency method utilization of relaxing end points restriction and premature termination and solve the authentification failure problem caused because end points does not line up between gesture sequence, also utilize the restriction of bending slope and premature termination strategy to decrease amount of calculation, experiment shows that this method has good result in the precision and efficiency of authentication simultaneously.

The present invention proposes a kind of gesture authentication method based on intelligent mobile phone sensor equipment, the method is on the basis of DTW method, limit bending slope, relax the end points alignment restriction of two matching sequences, simultaneously during coupling once DTW Cumulative Distance is beyond the threshold value of certification, then matching process premature termination.Adopt optimum template strategy in gesture authentication process, cycle tests is not needed and all templates compare, thus decrease amount of calculation, threshold value during certification is the ultimate range between other templates of optimum template and same people's same gesture.Utilize original DTW method, existing DTW high efficiency method (FastDTW) and this method (EIDTW) to comprise " a " and " in " 3000 gesture samples of two kinds of gestures carry out certification, the FAR of the present invention's three kinds of method validations is 0, the FRR of original DTW method is 8.6%, the FRR of FastDTW method is 5.8%, and the FRR of this method certification is 2.1%.Also can find out from Fig. 7 and Fig. 8 in addition, the authentication efficiency of this method is 2-3 times of original DTW, and one needs 5-10 millisecond to carry out certification to single gesture, and certification real-time is very high.

Accompanying drawing explanation

Fig. 1 is flow chart of the present invention.

Fig. 2 is classical DTW Method And Principle figure.The schematic diagram of DTW method Dynamic Programming search optimal path, tilt not too much to make coupling path, local restriction is carried out to coupling path, constraints is as shown in right side block diagram in this figure, black color dots refers to current lattice point, three white points show the way the position that the previous lattice point in footpath may occur, namely current lattice point (n is arrived, m) the previous lattice point before must be (n-1, m), (n-1, m-1), minimum one of Cumulative Distance in (n-1, m-2) three lattice points.

Fig. 3 is the dynamic programming matching region that existing DTW high efficiency method calculates by limiting bending slope.

Fig. 4 is the schematic diagram that this method dynamic programming path end points loosens.Eliminate the restriction of dynamic programming matching path end points alignment, namely allow the starting point of dynamic programming path in line segment [(1,1), (1, L)] or [(1,1), (L, 1)] on, and terminal can at line segment [(M-L+1, N), (M, N)] or [(M, N-L+1), (M, N)] on.

Fig. 5 be this method relax end points limit and limit Dynamic Programming bending slope be after 0.5-2, the dynamic programming matching region calculated, solid line hexagon is dynamic programming matching zone boundary.

Fig. 6 is this method premature termination schematic diagram.After representing two gesture sequence dynamic programming matching parts, the Cumulative Distance in coupling path just exceedes certification threshold value, so the calculating of terminator sequence further part Euclidean distance, and method premature termination.

Fig. 7 is test set when being positive sample, the results contrast of the present invention and original methods experiment.Test set is my 60 real gestures.A () figure adopts original DTW matching process to carry out False Rejects 4 samples in result figure: 60 samples of certification, 1706ms consuming time; B () adopts existing efficient DTW method to carry out False Rejects 2 samples in result figure: 60 samples of certification, 438ms consuming time; C () adopts the present invention to carry out False Rejects 0 sample in result figure: 60 samples of certification, 653ms consuming time;

Fig. 8 is test set when being negative sample, the results contrast of the present invention and original methods experiment.Test set is the gesture of 33 other people imitations.A () figure adopts original DTW matching process to carry out mistake in result figure: 33 samples of certification to accept 0,941ms consuming time; B () is that result figure: 33 the sample mistakes adopting existing efficient DTW method to carry out certification accept 0,316ms consuming time; C () is that result figure: 33 the sample mistakes adopting the present invention to carry out certification accept 0,190ms consuming time;

Embodiment

Below with reference to the drawings and specific embodiments, the present invention is described in further details:

Identity identifying method based on dynamic gesture of the present invention, has following characteristics:

1. utilize mobile phone acceleration sensor to obtain the acceleration information of gesture to be certified; For the restriction of intelligent mobile phone battery flying power, beginning and the end of gesture control with button respectively, only monitor mobile phone sensor when needs obtain data, consume battery of mobile phone resource as little as possible, also reduce unnecessary amount of calculation;

2. pair gesture data to be certified carries out the preliminary treatment such as denoising, quantification; The impact of shake and sensor accuracy is subject in data acquisition, unavoidably have noise jamming, in addition, all floating type by the initial data in acceleration transducer past, real-coded GA computation complexity is large, lose time, and use integer data to be the same with the effect using real-coded GA to reach during gesture authentication, so need to data de-noising, quantification;

3. to relax the restriction and the DTW efficient matchings method of premature termination calculates DTW distance between test sample book and optimum template with combining, this distance compared with threshold value, be less than or equal to threshold value then certification pass through, otherwise certification is not passed through.

In above-mentioned step, relax the restrictive condition that end points restriction refers to cancel the alignment of DTW highly effective algorithm coupling path end points, as shown in Figure 4, the starting point in permission dynamic programming matching path can in line segment [(1,1), (1, ] or [(1,1), (L L), 1)], and terminal also can at line segment [(M-L+1, N), (M, N)] or [(M, N-L+1), (M, N)] on.That is the first frame of certain gesture can mate with any frame in L frame before another gesture, and last frame can mate with any frame in another gesture end L frame.During algorithm realization, only the Euclidean distance of lattice point on starting point line segment and terminal line segment all just can need be realized the restriction of coupling path end points loosen for being set to 0.Limit the slope requirement of scope and the bending of relaxing according to path end points, the zone boundary of the lattice point of compute euclidian distances needed for Dynamic Programming can be calculated, as shown in Figure 5.

Premature termination refers to that DTW mates the Cumulative Distance in path once exceed threshold value, just stops calculating, as shown in Figure 6.X-axis often takes a step forward, all only use the Cumulative Distance of previous column, column vector D and d preserves the Cumulative Distance that the Cumulative Distance of previous column and current column calculate respectively, even if calculate the Euclidean distance of the just required node when prostatitis Cumulative Distance, DTW Cumulative Distance is a kind of laddering calculating, once the pointwise of previous section sequence is to computational process middle distance with exceeded threshold value, then without the need at the sequence node-by-node algorithm continuing aft section, thus saving calculation procedure, improve computational efficiency.

Select 10 samples that optimum template gathers when using first from user, the DTW high efficiency method relaxing end points restriction is utilized to calculate DTW distance between two between sample, suppose gesture sample Normal Distribution, to select with other samples apart from minimum sample as optimum template, distance maximum between this sample and other samples is threshold value.In order to adapt to the change that may occur when people's different times performs gesture, adopt template adaptive strategy, when the test sample book number of certain user's gesture has exceeded n (when this module realizes, n value is 50), just from by again choosing new optimum template the gesture of certification, the method chosen is the same with when choosing first.

Embodiment 1:

As shown in Figure 1, that now introduces each step realizes details to idiographic flow.

1, utilize the acceleration transducer on smart mobile phone, obtain user and perform three directional acceleration (x, y, z) data sequences in gesture process.First utilize (SensorManager) context.getSystemService (context.SENSOR_SERVICE) to obtain the packaged sensor management object SensorManager of Android system, then obtain acceleration transducer Sensor.TYPE_ACCELEROMETER by SensorManager; The set of frequency of sensors for data is SENSOR_DELAY_GAME, adopt the words data redundancy amount of higher frequency SENSOR_DELAY_FASTEST too large, the waste processing time, adopt more low frequency SENSOR_DELAY_UI and SENSOR_DELAY_NORMAL time, the gesture acceleration information got again very little, be unfavorable for the coupling of gesture, experiment proof selects SENSOR_DELAY_GAME proper.When gathering gesture data, need monitor acceleration transducer, and power consumption is compared in the operation of monitoring acceleration transducer, millet 1 generation mobile phone is used to test the power consumption that this operates, close application except system program during test, the result of test is: when not monitoring acceleration transducer, the electricity of 1% can use 16 points 10 seconds, and when monitoring acceleration transducer, the electricity of 1% can only use 6 points 35 seconds.For the restriction of intelligent mobile phone battery flying power, beginning and the end of gesture control with button respectively, acceleration transducer is monitored while gesture starts, remove acceleration transducer at the end of gesture immediately to monitor, so, only when needs obtain data, mobile phone sensor is monitored, ensure that battery life; Adopt another benefit of beginning and conclusion button to be the beginning and the end position that do not need to calculate according to the data obtained again gesture, thus decrease some unnecessary amounts of calculation.

2, preliminary treatment is carried out to test data, comprise denoising, quantification.The impact of shake and sensor accuracy is subject in data acquisition, unavoidably have noise jamming, adopt simple Moving Average (SMA) filter to the smoothing denoising of test data, elimination random noise on the basis of response fast, derivation formula is as follows:

SMA _now＝(X _i+X _i-1+...+X _i-n+1)/n n＝1，2，3，... (1)

N in formula (1) is smoothing denoising parameter, and when n value is too small, denoising effect is not obvious, and the excessive gesture information that can make again of n value is lost, and sums up and learn in experimentation, during n value 5-10, proper.

The initial data that mobile phone sensor obtains is all floating type, in view of the resource of mobile phone limits, should reduce amount of calculation as much as possible, improves computational efficiency.So in order to avoid floating type calculating, real-coded GA is converted to the shape data of 33 grades, by studying a large amount of gesture datas, finding that the value of gesture acceleration mainly concentrates between-g to g, seldom having data on 2g or under-2g, so to the initial data memory nonlinear quantization collected, below-2g is quantified as-10 to 10 between being quantified as-15 to-11 ,-g to g between being quantified as-16 ,-2g to-g, be quantified as 11-15 between g to 2g, more than 2g is quantified as 16.

3, the object of dynamic time warping DTW method is between reference template and test data, find the coupling path of a time calibration optimized.The method is applied in speech recognition the earliest, solves the difficult problem that speech rate is uneven.And in gesture authentication, also there is the uneven problem of speed, the gesture of execution may be not of uniform size, so DTW method is also applicable to gesture authentication.As shown in Figure 2, R represents optimum template, and m is the sequential label of optimum template, and m=1 is start frame, and m=M is terminal gesture frame; T represents gesture test data, and n is the sequential label of optimum template, and n=1 is start frame, and n=N is terminal gesture frame.DTW method is exactly hunting time warping function j=w (i), and the time shaft of test data is non-linearly mapped on optimum template time axle j, and meets:

$\underset{w\left(i\right)}{\min }\left(i\right),R\left(w\left(i\right)\right)]---\left(2\right)$

D [T (i) in formula (2), R (w (i))] be distance metric between cycle tests i-th frame T (i) and template sequence jth frame template vector R (j), be Euclidean distance, the required Euclidean distance calculated of classical DTW method is all lattice points in rectangle frame.Dist is then the Cumulative Distance between two data under optimal situation.Meanwhile, DTW method limits for coupling path with the addition of some, and first is Experience about Monotonicity of Functions restriction, because data have timing, so warping function must meet monotonicity restriction w (i+1) >=w (i); Second is continuity restriction, and some special direction vector plays critical effect to matching effect sometimes, in order to ensure correctness, requires that Time alignment function does not allow any one match point of effect.3rd is end points alignment restriction, and can find out in Fig. 2, during DTW method coupling, the beginning end points of test data and reference template and end caps align all respectively.

Existing DTW high efficiency method is the slope limiting bending edge circle on the basis of classical DTW method is 0.5-2, and as shown in Figure 3, high efficiency method only need calculate the Euclidean distance of the lattice point within parallelogram, dynamic bending is divided into 3 sections, (1, X _a), (X _a+ 1, X _b), (X _b+ 1, N).Wherein X _a=(2M-N)/3, X _ball the most close integer is got in=2 (2N-M)/3, thus the pass obtained between the most optimum template length M and test template length N is

$\left\{\begin{array}{c}2M-N\≥3\\ 2N-M\≥2\end{array}\right.---\left(3\right)$

The test data not meeting formula 3 relation can be thought and differs too large with optimum template, cannot carry out dynamic bending.

Combination of the present invention is loosened end points restriction and the DTW high efficiency method that terminates in advance limiting on the basis that bending slope in coupling path is 0.5-2 and is loosened end points restriction, and adopts premature termination strategy in matching process.Loosening end points restriction is exactly the end points alignment restriction cancelled in DTW algorithm.Allow the starting point of dynamic programming path in line segment [(1,1), (1, ] or [(1,1), (L L), 1)], and terminal can at line segment [(M-L+1, N), (M, ] or [(M N), N-L+1), (M, N)] on.As shown in Figure 4, the DTW highly effective algorithm loosening end points only need calculate the lattice point Euclidean distance realized in polygon, the amount of calculation of relative DTW highly effective algorithm is some more a little, but can solve the start frame of gesture, situation that end frame does not line up, can improve the precision of certification.Dynamic bending is divided into 5 sections, (1, L), (L+1, X _a), (X _a+ 1, X _b), (X _b+ 1, N-L), (N-L+1, N), wherein X _a=(2M-N-L)/3, X _b=(2 (2N-M)+L)/3 are all got the most close integer, and must be met X _a>=1, X _b>=L, show that the pass of optimum template length M and test data length N is thus:

$\left\{\begin{array}{c}2M+N\≥3+L\\ 2N-M\≥L\end{array}\right.---\left(4\right)$

The test data not meeting formula 4 relation can be thought and differs too large with optimum template, cannot carry out dynamic bending.L value is too little, DeGrain, L value too conference lead to errors percent of pass (FAR) increase, L value 8 in the present invention.5 sections that bending is divided into, according to the x coordinate put in the slope of every section of up-and-down boundary line segment and section, can obtain the y boundary value of every section, y boundary value is expressed as follows by x:

$y_{\min }=\left\{\begin{array}{cc}0& 1\≤x\≤L\\ \frac{1}{2}(x-L)& L+1\≤x\≤X_b\\ 2(x-N)+M-L& X_b+1\≤x\≤N\end{array}\right.---\left(5\right)$

$y_{\max }=\left[\begin{array}{cc}2x+L& 1\≤x\≤X_z\\ \frac{1}{2}(x-N+L)+M& X_a+1\≤x\≤N-L\\ M& N-L+1\≤x\≤N\end{array}\right.---\left(6\right)$

When Dynamic Programming is carried out to test gesture every column data, calculate its corresponding boundary value, and only mate the lattice point in border, thus reduce amount of calculation.Matching primitives due to each row lattice point has only used 3 lattice points of previous column, does not need to preserve the distance matrix and Cumulative Distance matrix that mate between all frames, reduce memory space when realizing in method.

In method Reusability to optimum template utilize the DTW high efficiency method loosening end points restriction to select.First, user gathers the data sample of 10 certain gestures, the DTW high efficiency method loosening end points restriction is utilized to calculate DTW Cumulative Distance between two between data sample, select one with the minimum sample of other sample Distance geometry as optimum template, the ultimate range between this sample and other samples is threshold value.In order to adapt to the change that may occur when people's different times performs gesture, adopt template adaptive strategy, at set intervals just from by again choosing new optimum template the gesture of certification, the method chosen is the same with when choosing first.

4, loosen end points restriction DTW highly effective algorithm basis on again in conjunction with premature termination strategy, premature termination is a kind of method used in restricted distance calculates, as shown in Figure 6, once all DTW Cumulative Distances arranged above exceed threshold value, then without the need to asking the Euclidean distance of further part lattice point in border again, thus saving calculation procedure, premature termination strategy can reduce a large amount of amounts of calculation when testing gesture and differing larger with optimum template.Dynamic programming step based on premature termination is as follows:

State two column vectors D, d, D preserves the Cumulative Distance of previous column, and d preserves the Cumulative Distance that current column calculates.Second frame data of the Cumulative Distance of L frame to be 0, data i be test sample book before initialization test data first frame and optimum template;

5, the Cumulative Distance d [] of optimum template data frame within i and boundary condition is calculated according to dynamic programming principle; Then vector d assignment to vector D; Judge whether data all in vector D are all greater than threshold value, if be all greater than threshold value, terminate coupling in advance, authentification failure; If not being all greater than threshold value, then enter step 6;

6, judge that whether data i is the last frame of test data, if i is not last frame, then the next frame assignment of i to i, proceed to step 5; If i is last frame, then obtain D [M-L+1 ... M] between minimum value min, and compare with threshold value, if threshold value is greater than min, authentication success, terminates coupling; If threshold value is less than min, authentification failure, terminates coupling.

Claims (4)

1. based on an identity identifying method for dynamic gesture, it is characterized in that, comprise the following steps:

1) gather x when gesture performs, the acceleration information in y, z tri-directions is as test data;

2) preliminary treatment is carried out to test data; Described preliminary treatment comprises smoothing denoising and quantification;

3) dynamic time warping (DTW) algorithm improved is adopted to mate with template data test data; Cancel DTW method the 2nd) test data that obtains of step aligns with end points when mating between optimum template and limit, and the starting point in permission dynamic programming matching path is in line segment [(1,1), (1, L)] or [(1,1), (L, 1)] on, and allow terminal at line segment [((M-L+1), N), (M, N)] or [(M, (N-L+1)), (M, N)] on; First frame of certain gesture that is in test data can mate with any frame in L frame before another gesture in template data, and the last frame in test data can mate with any frame in another gesture end L frame in template data; And limit Dynamic Programming bending slope between 0.5-2, the amount L utilizing end points to relax and the slope meter of bending calculate the boundary condition of Dynamic Programming;

Wherein, M is optimum template length, and N is test data length, and L is the amount that end points relaxes;

4) the 3rd) carry out Dynamic Programming within the boundary condition that obtains of step, do not need to preserve all Cumulative Distances and frame matching distance, statement column vector D preserves the Cumulative Distance of previous column, and statement column vector d preserves the Cumulative Distance that current column is calculated;

5) calculate in Dynamic Programming and judge whether the Cumulative Distance when prostatitis is all greater than threshold tau, if be all greater than threshold tau, authentification failure, premature termination; If be also not all greater than threshold tau, then proceed to step 6);

6) judge that whether current data frame is the last frame of test data, if not last frame, then next frame assignment to current data frame, and proceed to step 5); If last frame, then get D [M-L+1 ..., M] in minimum value min, and to compare with threshold tau, if min is less than threshold value, then authentication success, ending method; If min is greater than threshold tau, then authentification failure, terminates this verification process.

2. a kind of identity identifying method based on dynamic gesture according to claim 1, is characterized in that, described optimum template and threshold value determination method are:

For user gathers the sample of 15 certain gestures, the DTW efficient matchings method loosening end points restriction is adopted to calculate DTW Cumulative Distance between two between sample, select the sample minimum with other sample Distance geometry to be optimum sample and optimum template, and select the ultimate range between optimum sample and other samples to be threshold value.

3. a kind of identity identifying method based on dynamic gesture according to claim 1, it is characterized in that, in order to adapt to the change that may occur when people's different times performs gesture, adopting template adaptive strategy, replacing original optimum template from by again choosing new optimum template the gesture of certification at set intervals.

4. a kind of identity identifying method based on dynamic gesture according to any one of claim 1-3, is characterized in that:

In step 1) in, utilize (SensorManager) context.getSystemService (context.SENSOR_SERVICE) to obtain the packaged sensor management object SensorManager of Android system, then obtain acceleration transducer Sensor.TYPE_ACCELEROMETER by SensorManager; The set of frequency of sensors for data is SENSOR_DELAY_GAME;

In step 2) in, adopt simple Moving Average (SMA) filter to the smoothing denoising of test data; Computing formula is: the value after denoising wherein X _ibe i-th test data; N gets a value in 5-10;

In step 2) in, the acceleration information of the floating type collected is converted to the integer data of 33 grades;

In step 3) in, axle match time is divided into 5 sections, (1, L), (L+1, X _a), (X _a+ 1, X _b), (X _b+ 1, N-L), (N-L+1, N), wherein X _a=(2M-N-L)/3, X _b=(2 (2N-M)+L)/3 are all got the most close integer, and must be met X _a>=1, X _b>=L, show that the pass of optimum template length M and test data length N is thus:

$\left\{\begin{array}{c}2M-N\≥3+L\\ 2N-M\≥L\end{array}\right.;$

The test data not meeting above formula relation is considered to differ with optimum template too large, cannot carry out dynamic bending, L value 8; 5 sections that bending is divided into, y boundary value is expressed as follows by x:

$y_{\min }=\{\begin{array}{cc}0& 1\≤x\≤L\\ \frac{1}{2}(x-L)& L+1\≤x\≤X_b\\ 2(x-N)+M-L& X_b+1\≤x\≤N\end{array};$

$y_{\max }=\{\begin{array}{cc}2x+L& 1\≤x\≤X_a\\ \frac{1}{2}(x-N+L)+M& X_a+1\≤x\≤N-L\\ M& N-L+1\≤x\≤N\end{array};$

When Dynamic Programming is carried out to test gesture every column data, calculate its corresponding boundary value, and only mate the lattice point in border, thus reduce amount of calculation; The matching primitives of each row lattice point has only used 3 lattice points of previous column; L value 8.