CN103455794A - Dynamic gesture recognition method based on frame fusion technology - Google Patents

Abstract

The invention discloses a dynamic gesture recognition method based on the frame fusion technology. The method is characterized by comprising the following steps that an established group of dynamic gestures are divided into a group of gesture sets according to the density distribution feature, for each gesture set, the density distribution feature parameter of a frame fusion image of each gesture is obtained, then the mean value of the density distribution feature parameters in each set is obtained, the mean value is used as the template feature vector H of the gesture set; for a dynamic gesture image to be recognized, a mapped static combination image Q is obtained; the density distribution feature of the static combination image Q is computed, according to the density distribution feature, the gesture set where the dynamic gesture is placed is recognized; gestures in the gesture set are determined according to the range of the density distribution feature, and a Hausdorff distance method or a fingertip feature point method is further selected for use.

Description

A kind of dynamic gesture identification method based on the frame integration technology

Technical field

The present invention relates to dynamic gesture identification field, specifically, relate to a kind of dynamic gesture identification method based on the frame integration technology.

Background technology

Along with developing rapidly of computer technology, human-computer interaction technology becomes current one of the most popular research topic gradually.And the importance of staff has determined its great researching value in field of human-computer interaction.Staff, as a kind of the most natural, directly perceived and be easy to the man-machine interaction means of study, is applied by people's broad research, and the gesture identification based on vision becomes one of content of people's broad research.

According to the movement characteristic of gesture, gesture can be divided into to static gesture and dynamic gesture.Static gesture relies on shape and the profile transmission of information of hand, and dynamic gesture is along with the variation of time, and position and the shape of hand also change, thereby can transmit more accurately detailed information.Current gesture identification method has template matching method (Model/Template Matching), hidden Markov model (HMM, Hidden Markov Model), dynamic time programming (DTW) etc.

The method of template matches is multiplex in static gesture identification.2002, the gesture identification based on the hausdorff distance that the people such as Zhang Liangguo, Wujiang qin propose, utilized the thought of hausdorff apart from template matches, realized the gesture identification that robustness is stronger.Huang Guofan and Li Ying propose an Alphabet Gesture recognition methods, at first, to all Alphabet Gesture image pre-service, then utilize the method for template matches to be identified.

2003, Ahmed Elgammal proposed a kind of HMM model of imparametrization, and the method means dynamic gesture, according to the probability volume architecture, carries out gesture identification by a series of attitudes of both having learned.Yan Yan etc. are used HMM gesture instruction model, adopt k_means to obtain the vector quantization of gesture sequence.Thereby improved the performance of gesture identification.

The DTW method is a kind of pattern matching algorithm with Nonlinear Time normalization effect.The people such as Trevor J. proposed a kind of DTW method in 1996,

The DTW method is simply effective, allows sufficient elasticity between test mould and reference model, thereby realizes correct classification.The people such as warp is of heap of stone, Marvin's army use the method for dynamic time programming effectively to improve the dynamic gesture recognition efficiency based on acceleration.

Summary of the invention

The technical problem to be solved in the present invention is to provide a kind of dynamic gesture identification method based on the frame integration technology, has increased the accuracy of dynamic gesture identification.

The present invention adopts following technical scheme to realize goal of the invention:

A kind of dynamic gesture identification method based on the frame integration technology, is characterized in that, comprises the steps:

(1) one group of set dynamic gesture is divided into to one group of gesture set according to Density Distribution Feature, each gesture set is asked to the Density Distribution Feature parameter of its each gesture frame fused images, then ask respectively the mean value of its Density Distribution Feature parameter in each set, the template characteristic vector H using mean value as this gesture set;

(2) static constitutional diagram Q dynamic gesture Image Acquisition to be identified shone upon;

(3) calculate the Density Distribution Feature of static constitutional diagram Q, identify the gesture set at dynamic gesture place according to Density Distribution Feature;

(4) determine that according to the scope of Density Distribution Feature the gesture in the gesture set further selects to adopt Haus dorff distance method or finger tip unique point method.

As the further restriction to the technical program, described step (2) comprises the steps:

(2.1) obtain the continuous dynamic gesture sequence image of N frame { P _i, i=0,1,2 ..., N;

(2.2) pixel in static constitutional diagram Q is with { P _iin the mapping relations of pixel as follows

$\left[\begin{array}{c}{x}_i\\ y_i\end{array}\right]=\left[\begin{array}{ccc}\frac{a}{2}+\left(\begin{array}{cc}X-\mathrm{\γ}& \sin \mathrm{\β}-\frac{A}{2}\end{array}\right)& \cos \mathrm{\α}+\left(\begin{array}{cc}Y+\mathrm{\γ}& \cos \mathrm{\β}-\frac{B}{2}\end{array}\right)& \sin \mathrm{\α}\\ \frac{b}{2}-\left(\begin{array}{cc}X-\mathrm{\γ}& \sin \mathrm{\β}-\frac{A}{2}\end{array}\right)& \sin \mathrm{\α}-\left(\begin{array}{cc}Y+\mathrm{\γ}& \cos \mathrm{\β}-\frac{B}{2}\end{array}\right)& \cos \mathrm{\α}\end{array}\right]---\left(1\right)$

Wherein, x _i, y _irepresent respectively little figure P _ithe transverse and longitudinal coordinate of a certain pixel, X, Y means respectively the transverse and longitudinal coordinate of respective pixel in constitutional diagram Q, a, b means respectively P _iwidth and the height, A, B mean respectively Q width and the height, r is each little figure P _ithe radius that rotation is arranged in Q, α presentation graphic P _ithe angle that self rotates, β means the P after over-rotation _ithe radius at center with the angle of the rectangular coordinate system ordinate of setting up in Q, and:

$\mathrm{\α}=\frac{3}{2}\mathrm{\π}+\frac{2}{N}\mathrm{\πi}---\left(2\right)$

$\mathrm{\β}=\frac{2}{N}\mathrm{\πi}---\left(3\right)$

As the further restriction to the technical program, described step (3) comprising:

(3.1) plane picture that static constitutional diagram Q forms is f (x, y), the centre of form of computed image f (x, y), i.e. center of gravity

(3.2) in computed image f (x, y), target pixel points is to the ultimate range D of the centre of form _max, minor increment D _min;

(3.3) in difference computed image f (x, y), take the centre of form as the center of circle, with D _maxfor the maximum circumscribed circle in the target area of radius with D _minfor the target area minimum circumscribed circle of radius, in the zone formed at maximum circumscribed circle and minimum circumscribed circle, use equidistant regional partitioning that image is divided into to M number of sub images zone (M>0);

(3.4) each sub-image area is added up respectively, calculated the total S of each sub-image area internal object pixel _i(i=1 ..., M), and find out S _imaximal value.

$S_{\max }=\underset{i=1,\·\·\·,M}{\max }\left(S_i\right)---\left(6\right)$

(3.5) calculate the Density Distribution Feature d of static constitutional diagram:

r _i=S _i/S _max(i=1,…,M) (7)

${\mathrm{dr}}_i=\left\{\begin{array}{cc}|r_1-r_2|& i=1\\ |2r_i-r_{i-1}-r_{i+1}|& 1<i<M\\ |r_M-r_{M-1}|& i=M\end{array}\right.---\left(8\right)$

d=(r ₁,…,r _M；dr ₁,…,dr _M) (9)

(3.6) according to formula DDF '=(r ₁..., r ₁₀, ar ₁₁..., ar ₁₅, br ₁₆..., br ₂₀; Dr ₁..., dr ₁₀, cdr ₁₁..., cdr _m) obtain amended proper vector , wherein, a, b, c is customized parameter;

(3.7) by the proper vector obtained

with the template characteristic vector H in each gesture set, compare respectively, the difference compute euclidian distances, the gesture set of Euclidean distance minimum is selected gesture set.

As the further restriction to the technical program, described step (4) comprises the steps:

(4.1) for gesture in the 35th the gesture set of parameter vector between scope 5 to 6 of Density Distribution Feature, adopt the Hausdorff distance further to identify the last frame of dynamic gesture, thereby reach gesture identification effect accurately;

(4.2) for gesture in the 35th the gesture set of parameter vector between scope 3 to 4 of Density Distribution Feature, adopt the method for finger tip unique point further to identify the last frame of dynamic gesture, thereby reach gesture identification effect accurately.

As the further restriction to the technical program, the Hausdorff distance method of described step (4.1) carries out further identifying and comprising the steps: to the last frame of dynamic gesture

(4.1.1) the frontier point set L of gesture in the set of first trained gesture;

(4.1.2) then the last frame of gesture to be identified is asked the set e of its frontier point;

(4.1.3) calculate respectively the hausdorff distance of e with frontier point set L;

(4.1.4) output hausdorff is apart from minimum template gesture sequence i, and this sequence is identified correct gesture sequence.

As the further restriction to the technical program, the method for the finger tip unique point of described step (4.2) comprises the steps:

(4.2.1) first trained gesture set, detect the finger tip dot information of various gestures in the gesture set, the vector that each finger tip point is formed with the gesture center of gravity

characteristic information charge to vectorial template G;

(4.2.2) then detect the finger tip vector of gesture to be identified

, with the finger tip vector in vectorial template G, compare, the template gesture sequence i of output similarity maximum, this sequence is identified correct gesture sequence.

Compared with prior art, advantage of the present invention and good effect are: the present invention adopts the method for constitutional diagram, to be converted into to the identification of dynamic gesture the identification to static constitutional diagram, effectively improved the discrimination of dynamic gesture, in addition, the time loss of the method is not high, within the scope allowed at computing machine.

The accompanying drawing explanation

Fig. 1 is the static constitutional diagram of correspondence that dynamic gesture of the present invention captures.

10 kinds of initial state and various final states images of gestures thereof that dynamically capture that Fig. 2 is the preferred embodiments of the present invention.

The process flow diagram that Fig. 3 is the preferred embodiment of the present invention.

Fig. 4 is that the preferred embodiment of the present invention adopts the Density Distribution Feature method to carry out the Classification and Identification schematic diagram to static constitutional diagram.

Fig. 5 is that preferred embodiment of the present invention finger tip is surveyed schematic diagram.

Embodiment

Below in conjunction with accompanying drawing and preferred embodiment, the present invention is further described in detail.

1, constitutional diagram

How the image stream of dynamic gesture being processed, is a major issue in the dynamic gesture identifying.This paper is mapped as a static constitutional diagram by each two field picture in a dynamic gesture process.By the analyzing and processing to static constitutional diagram, complete the identification to dynamic gesture.In this constitutional diagram, rounded being arranged in wherein of dynamic image stream.This paper completes the first step of gesture identification by the identification to constitutional diagram: rough gesture identification, constitutional diagram as shown in Figure 1:

The mapping relations of dynamic gesture image stream and constitutional diagram are as follows:

Suppose to obtain the continuous dynamic gesture sequence image of N frame { P _i, i=0,1,2 ..., N.Pixel in constitutional diagram is with { P _iin the mapping relations of pixel as follows $\left[\begin{array}{c}{x}_i\\ y_i\end{array}\right]=\left[\begin{array}{ccc}\frac{a}{2}+\left(\begin{array}{cc}X-\mathrm{\&gamma;}& \sin \mathrm{\&beta;}-\frac{A}{2}\end{array}\right)& \cos \mathrm{\&alpha;}+\left(\begin{array}{cc}Y+\mathrm{\&gamma;}& \cos \mathrm{\&beta;}-\frac{B}{2}\end{array}\right)& \sin \mathrm{\&alpha;}\\ \frac{b}{2}-\left(\begin{array}{cc}X-\mathrm{\&gamma;}& \sin \mathrm{\&beta;}-\frac{A}{2}\end{array}\right)& \sin \mathrm{\&alpha;}-\left(\begin{array}{cc}Y+\mathrm{\&gamma;}& \cos \mathrm{\&beta;}-\frac{B}{2}\end{array}\right)& \cos \mathrm{\&alpha;}\end{array}\right]$ $\left(1\right)$

Wherein, x _i, y _irepresent respectively little figure P _ithe transverse and longitudinal coordinate of a certain pixel, X, Y means respectively the transverse and longitudinal coordinate of respective pixel in constitutional diagram Q, a, b means respectively P _iwidth and the height, A, B mean respectively Q width and the height.R is each little figure P _ithe radius that rotation is arranged in Q.α presentation graphic P _ithe angle that self rotates, β means the P after over-rotation _ithe radius at center with the angle of the rectangular coordinate system ordinate of setting up in Q.And:

$\mathrm{\&alpha;}=\frac{3}{2}\mathrm{\&pi;}+\frac{2}{N}\mathrm{\&pi;i}---\left(2\right)$

$\mathrm{\&beta;}=\frac{2}{N}\mathrm{\&pi;i}---\left(3\right)$

2, Density Distribution Feature (DDF)

After the frame integration technology, the static constitutional diagram obtained is asked to its Density Distribution Feature.

After image binaryzation, the arrangement of object pixel in image is depended in the information transmission of image.The basic goal of Density Distribution Feature is to obtain the pixel distribution information of image in the distribution situation in zones of different space by the statistics object pixel, thereby reaches the purpose of expressing this bianry image.Classify and can carry out the identification of different images by the Density Distribution Feature to image.

This paper, according to the variation characteristic of gesture in constitutional diagram, has made certain modification to Density Distribution Feature.

Former DDF character representation is as follows:

DDF=(r ₁,…,r _M；dr ₁，…,dr _M) (4)

In static constitutional diagram, gesture changes the overwhelming majority and is present in finger part, and the variation of centre of the palm part is less.Thereby the DDF weight of increase finger part, can effectively reduce the DDF similarity between the different virtual constitutional diagram, improves recognition efficiency.Through great many of experiments, new DDF feature is as follows:

DDF′=(r ₁,…,r ₁₀,ar ₁₁,…,ar ₁₅,br ₁₆,…,br ₂₀；dr ₁,…,dr ₁₀,cdr ₁₁,…,cdr _M) (5)

Wherein, a, b, c is customized parameter, this paper operation parameter is: a=3, b=6, c=3.

3. gesture identification

The present invention intends 10 kinds of gestures of dynamically grabbing are identified.The initial state of dynamic gesture and various final states images of gestures thereof are as Fig. 4.

Identifying is divided into two stages.First stage adopts the method for Density Distribution Feature to be identified to constitutional diagram, identify the gesture set of dynamic gesture place, subordinate phase adopts Hausdorff distance or the method for unique point further to identify the last frame of dynamic gesture in the gesture set, thereby reach gesture identification effect accurately, identifying as shown in Figure 3.

3.1 rough gesture identification

The identification of this stage adopts the Density Distribution Feature method to carry out Classification and Identification (Fig. 4) to static constitutional diagram.10 kinds of gestures to be identified are divided into to A, B, C, tetra-set of D according to the DDF feature, gesture to be identified is carried out to sets classification.Method step is as follows:

(3.1) plane picture that static constitutional diagram Q forms is f (x, y), the centre of form of computed image f (x, y), i.e. center of gravity

(3.2) in computed image f (x, y), target pixel points is to the ultimate range D of the centre of form _max, minor increment D _min;

(3.3) in difference computed image f (x, y), take the centre of form as the center of circle, with D _maxfor the maximum circumscribed circle in the target area of radius with D _minfor the target area minimum circumscribed circle of radius, in the zone formed at maximum circumscribed circle and minimum circumscribed circle, use equidistant regional partitioning that image is divided into to M number of sub images zone (M>0);

(3.4) each sub-image area is added up respectively, calculated the total S of each sub-image area internal object pixel _i(i=1 ..., M), and find out S _imaximal value.

$S_{\max }=\underset{i=1,\&CenterDot;\&CenterDot;\&CenterDot;,M}{\max }\left(S_i\right)---\left(6\right)$

(3.5) calculate the Density Distribution Feature d of static constitutional diagram:

r _i=S _i/S _max(i=1,…,M) (7)

${\mathrm{dr}}_i=\left\{\begin{array}{cc}|r_1-r_2|& i=1\\ |2r_i-r_{i-1}-r_{i+1}|& 1<i<M\\ |r_M-r_{M-1}|& i=M\end{array}\right.---\left(8\right)$

d=(r ₁,…,r _M；dr ₁,…,dr _M) (9)

(3.6) according to formula DDF '=(r ₁..., r ₁₀, ar ₁₁..., ar ₁₅, br ₁₆..., br ₂₀; Dr ₁..., dr ₁₀, cdr ₁₁..., cdr _m) obtain amended proper vector

, wherein, a, b, c is customized parameter;

(3.7) the proper vector d obtained is compared with the template characteristic vector H in each gesture set respectively, the difference compute euclidian distances, the gesture set of Euclidean distance minimum is selected gesture set.

3.2 accurate gesture identification

In the subordinate phase identifying, different characteristic according to dynamic gesture last frame images of gestures, last frame is identified, thereby complete the accurate identification of gesture, learn by analysis, the 35th parameter vector of the gesture Density Distribution Feature of C set, between scope 5 to 6, therefore adopts the hausdorff range formula to be identified, the 35th parameter vector of the gesture Density Distribution Feature in the D set, between scope 3 to 4, adopts finger tip unique point method.Concrete recognition methods is as follows.

In the identification of C set, to last frame, adopt the hausdorff range formula to be identified.The frontier point set L of gesture in the set of first trained gesture; Then the last frame of gesture to be identified is asked the set e of its frontier point; Calculate respectively the hausdorff distance of e with frontier point set L; Hausdorff is apart from minimum template gesture sequence i in output, and this sequence is identified correct gesture sequence.

In the identification of D set, to last frame, adopt finger tip to survey (Fig. 5) and identified.

The set of first trained gesture, detect the finger tip dot information of various gestures in the gesture set, the vector that each finger tip point is formed with the gesture center of gravity

characteristic information charge to vectorial template G;

Then detect the finger tip vector of gesture to be identified

, with the finger tip vector in vectorial template G, compare, the template gesture sequence i of output similarity maximum, this sequence is identified correct gesture sequence.

4. beneficial effect

The present invention uses common camera, under constant illumination condition, to 10 kinds of dynamic gestures, adopt the method for constitutional diagram to carry out ground floor identification, then the method that adopts hausdorff distance and finger tip to survey is carried out second layer identification, and identification has improved recognition efficiency greatly by different level.Table 1 is depicted as the discrimination that the discrimination that adopts this paper recognition methods to obtain carries out the identification of DDF single frames with single frames gesture figure and is contrasted.

The single frames recognition methods refers to its DDF parameter of each frame recording to continuous dynamic gesture, and by the DDF Characteristics creation template of each gesture, the DDF feature of gesture to be identified is compared with the DDF feature in template, thereby carries out gesture identification.

Table 1: discrimination contrast

From form, for set gesture, the discrimination of this paper method is significantly improved than the discrimination of single-frame images DDF identification.

Certainly, above-mentioned explanation is not limitation of the present invention, and the present invention also is not limited only to above-mentioned giving an example, and the variation that those skilled in the art make in essential scope of the present invention, remodeling, interpolation or replacement, also belong to protection scope of the present invention.

Claims (6)

1. the dynamic gesture identification method based on the frame integration technology, is characterized in that, comprises the steps:

(1) one group of set dynamic gesture is divided into to one group of gesture set according to Density Distribution Feature, each gesture set is asked to the Density Distribution Feature parameter of its each gesture frame fused images, then ask respectively the mean value of its Density Distribution Feature parameter in each set, the template characteristic vector H using mean value as this gesture set;

(2) static constitutional diagram Q dynamic gesture Image Acquisition to be identified shone upon;

(3) calculate the Density Distribution Feature of static constitutional diagram Q, identify the gesture set at dynamic gesture place according to Density Distribution Feature;

(4) determine that according to the scope of Density Distribution Feature the gesture in the gesture set further selects to adopt Hausdorff distance method or finger tip unique point method.

2. the dynamic gesture identification method based on the frame integration technology according to claim 1, is characterized in that, described step (2) comprises the steps:

(2.1) obtain the continuous dynamic gesture sequence image of N frame { P _i, i=0,1,2 ..., N;

(2.2) pixel in static constitutional diagram Q is with { P _iin the mapping relations of pixel as follows

$\left[\begin{array}{c}{x}_i\\ y_i\end{array}\right]=\left[\begin{array}{ccc}\frac{a}{2}+\left(\begin{array}{cc}X-\mathrm{\&gamma;}& \sin \mathrm{\&beta;}-\frac{A}{2}\end{array}\right)& \cos \mathrm{\&alpha;}+\left(\begin{array}{cc}Y+\mathrm{\&gamma;}& \cos \mathrm{\&beta;}-\frac{B}{2}\end{array}\right)& \sin \mathrm{\&alpha;}\\ \frac{b}{2}-\left(\begin{array}{cc}X-\mathrm{\&gamma;}& \sin \mathrm{\&beta;}-\frac{A}{2}\end{array}\right)& \sin \mathrm{\&alpha;}-\left(\begin{array}{cc}Y+\mathrm{\&gamma;}& \cos \mathrm{\&beta;}-\frac{B}{2}\end{array}\right)& \cos \mathrm{\&alpha;}\end{array}\right]---\left(1\right)$

Wherein, x _i, y _irepresent respectively little figure P _ithe transverse and longitudinal coordinate of a certain pixel, X, Y means respectively the transverse and longitudinal coordinate of respective pixel in constitutional diagram Q, a, b means respectively P _iwidth and the height, A, B mean respectively Q width and the height, r is each little figure P _ithe radius that rotation is arranged in Q, α presentation graphic P _ithe angle that self rotates, β means the P after over-rotation _ithe radius at center with the angle of the rectangular coordinate system ordinate of setting up in Q, and:

$\mathrm{\&alpha;}=\frac{3}{2}\mathrm{\&pi;}+\frac{2}{N}\mathrm{\&pi;i}---\left(2\right)$

$\mathrm{\&beta;}=\frac{2}{N}\mathrm{\&pi;i}---\left(3\right)$

3. the dynamic gesture identification method based on the frame integration technology according to claim 1, is characterized in that, described step (3) comprising:

(3.1) plane picture that static constitutional diagram Q forms is f (x, y), the centre of form of computed image f (x, y), i.e. center of gravity

(3.2) in computed image f (x, y), target pixel points is to the ultimate range D of the centre of form _max, minor increment D _min;

(3.3) in difference computed image f (x, y), take the centre of form as the center of circle, with D _maxfor the maximum circumscribed circle in the target area of radius with D _minfor the target area minimum circumscribed circle of radius, in the zone formed at maximum circumscribed circle and minimum circumscribed circle, use equidistant regional partitioning that image is divided into to M number of sub images zone (M>0);

(3.4) each sub-image area is added up respectively, calculated the total S of each sub-image area internal object pixel _i(i=1 ..., M), and find out S _imaximal value.

$S_{\max }=\underset{i=1,\&CenterDot;\&CenterDot;\&CenterDot;,M}{\max }\left(S_i\right)---\left(6\right)$

(3.5) calculate the Density Distribution Feature d of static constitutional diagram:

r _i＝S _i/S _max(i＝1,…,M) (7)

${\mathrm{dr}}_i=\left\{\begin{array}{cc}|r_1-r_2|& i=1\\ |2r_i-r_{i-1}-r_{i+1}|& 1<i<M\\ |r_M-r_{M-1}|& i=M\end{array}\right.---\left(8\right)$

d＝(r ₁,…,r _M;dr ₁,…,dr _M) (9)

(3.6) according to formula DDF '=(r ₁..., r ₁₀, ar ₁₁..., ar ₁₅, br ₁₆..., br ₂₀; Dr ₁..., dr ₁₀, cdr ₁₁..., cdr _m) obtain amended proper vector , wherein, a, b, c is customized parameter;

(3.7) by the proper vector obtained

with the template characteristic vector H in each gesture set, compare respectively, the difference compute euclidian distances, the gesture set of Euclidean distance minimum is selected gesture set.

4. the dynamic gesture identification method based on the frame integration technology according to claim 1, is characterized in that, described step (4) comprises the steps:

(4.1) for gesture in the 35th the gesture set of parameter vector between scope 5 to 6 of Density Distribution Feature, adopt the Hausdorff distance further to identify the last frame of dynamic gesture, thereby reach gesture identification effect accurately;

(4.2) for gesture in the 35th the gesture set of parameter vector between scope 3 to 4 of Density Distribution Feature, adopt the method for finger tip unique point further to identify the last frame of dynamic gesture, thereby reach gesture identification effect accurately.

5. the dynamic gesture identification method based on the frame integration technology according to claim 4, is characterized in that, the Hausdorff distance method of described step (4.1) carries out further identifying and comprising the steps: to the last frame of dynamic gesture

(4.1.1) the frontier point set L of gesture in the set of first trained gesture;

(4.1.2) then the last frame of gesture to be identified is asked the set e of its frontier point;

(4.1.3) calculate respectively the hausdorff distance of e with frontier point set L;

(4.1.4) output hausdorff is apart from minimum template gesture sequence i, and this sequence is identified correct gesture sequence.

6. the dynamic gesture identification method based on the frame integration technology according to claim 4, is characterized in that, the method for the finger tip unique point of described step (4.2) comprises the steps:

(4.2.1) first trained gesture set, detect the finger tip dot information of various gestures in the gesture set, the vector that each finger tip point is formed with the gesture center of gravity

characteristic information charge to vectorial template G;

(4.2.2) then detect the finger tip vector of gesture to be identified

, with the finger tip vector in vectorial template G, compare, the template gesture sequence i of output similarity maximum, this sequence is identified correct gesture sequence.

Images