KR20160031873A - Gesture recognizer setting method, gesture recognition method and apparatus - Google Patents

Abstract

Disclosed are a method for setting a gesture recognizer, and a method and apparatus for recognizing a gesture. The method for setting the gesture recognizer comprises the following steps: (a) converting input images into a plurality of binary images based on color information on an object in the input images including the object which performs a target gesture, wherein the input images are a plurality of continuous images; (b) extracting the center of the object from the converted binary images to individually obtain gesture data; (c) clustering the individually obtained gesture data, and obtaining a status sequence including a class of each cluster in each status by using a clustering result; (d) calculating a transition probability value between each cluster in the status sequence; and (e) setting a recognizer for the target gesture by using the status sequence and the calculated transition probability value.

Description

[0001] The present invention relates to a gesture recognizer setting method, a gesture recognizing method,

The present invention relates to a method and apparatus for recognizing a gesture using a computer vision.

Currently, research is being conducted on how to control using a hand gesture that can naturally interact with a computer. Most of the existing methods mainly use HMM (Hidden Markov Model).

When a gesture is recognized using an HMM, it is generally used because it is robust against noise and has good gesture recognition results. However, HMM has a large computational complexity and requires a lot of training data for training. In addition, when the HMM is used, only a pre-trained gesture can be recognized.

The present invention provides a gesture recognizer setting method, a gesture recognition method, and a gesture recognition method that can obtain a good gesture recognition result even with a relatively small amount of training data.

It is another object of the present invention to provide a gesture recognizer setting method, a gesture recognition method, and a gesture recognition method that allow a user to directly input a gesture to be used by himself / herself relatively fewer times to recognize a gesture.

Another object of the present invention is to provide a gesture recognizer setting method, a gesture recognition method, and a device for recognizing the gesture correctly without any separate training process even if the operation speed or time for the gesture is different for each user.

According to an aspect of the present invention, there is provided a method of recognizing a gesture by setting a recognizer for each gesture so as to obtain a good gesture recognition result with relatively little training data.

According to an embodiment of the present invention, there is provided a method for processing a target gesture, the method comprising the steps of: (a) converting an input image including an object performing a target gesture into a plurality of binary images based on color information about the object, A plurality of consecutive images; (b) extracting a center of the object from the transformed plurality of binary images to obtain gesture data; (c) clustering the obtained gesture data, and obtaining a state sequence including each class of each cluster in each state using the clustering result; (d) calculating a transition probability value between each cluster in the state sequence; And (e) setting a recognizer for the target gesture using the state sequence and the calculated transition probability value.

According to another embodiment of the present invention, there is provided a method of recognizing a gesture with a plurality of recognizers for a plurality of gestures set, the recognizer including a state sequence for each gesture and a transition probability value, Converting each of the plurality of input images into binary images based on color information of the object in a plurality of consecutive input images including the object; Extracting a center of the object from the plurality of converted binary images to obtain gesture data;

Identifying a cluster class for each gesture data; And recognizing the specific gesture with a gesture in which the cluster class coincides with the state sequence.

According to another aspect of the present invention, there is provided an apparatus for recognizing a gesture by setting a recognizer for each gesture so as to obtain a good gesture recognition result with relatively little training data.

According to an embodiment of the present invention, there is provided a binary image transform unit for transforming the input image into a plurality of binary images based on color information about the object in an input image including an object performing a target gesture, A plurality of consecutive images; A tracking unit for extracting gesture data by extracting a center of the object from the transformed plurality of binary images; A sequence obtaining unit for clustering the obtained gesture data and generating a state sequence for the target gesture including each cluster in each state using the clustering result; A calculation unit for calculating a transition probability value between each cluster in the state sequence; And a recognizer setting unit for setting a recognizer for the target gesture using the state sequence and the calculated transition probability value.

According to another embodiment of the present invention, there is provided an apparatus for recognizing a gesture with a plurality of recognizers for a plurality of gestures being set, the recognizer comprising a state sequence for each gesture and a transition probability value, A binary image transform unit for transforming the plurality of input images into binary images based on color information for the object in a plurality of consecutive input images including the object; A tracking unit for extracting a center of the object from the plurality of converted binary images to obtain gesture data; A classifying unit for classifying a cluster class for each gesture data; A gesture recognition apparatus may be provided that includes a recognition unit that recognizes the specific gesture as a gesture corresponding to a state sequence coinciding with the cluster class in consideration of a time sequence.

The gesture recognizer setting method, the gesture recognizing method, and the apparatus according to an embodiment of the present invention are advantageous in that even a relatively small amount of training data can provide excellent gesture recognition results.

In addition, the present invention has an advantage in that a user can directly input a gesture to be used by himself / herself to a gesture by a relatively small number of times.

In addition, the present invention has an advantage in that even if an operation speed or time for a gesture is different for each user, the user can accurately recognize the operation speed or the time without a separate training process.

FIG. 1 is a flowchart illustrating a method of setting a recognizer for each gesture for gesture recognition according to an exemplary embodiment of the present invention. Referring to FIG.
2 is an exemplary illustration of hand centered extraction in accordance with one embodiment of the present invention.
3 is an illustration of gesture data tracking in accordance with an embodiment of the present invention.
4 is a diagram illustrating normalization of gesture data according to an embodiment of the present invention;
5 is a diagram illustrating an orthogonal projection to a plane for gesture data according to an embodiment of the present invention;
6 is a diagram illustrating clustering of gesture data according to an embodiment of the present invention.
FIG. 7 is a diagram illustrating a state sequence and a transition probability value according to an embodiment of the present invention; FIG.
FIG. 8 is a flowchart for explaining a method of recognizing a gesture using a plurality of recognizers set by FIG. 1; FIG.
9 is a block diagram illustrating an internal configuration of a gesture recognition apparatus according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention is capable of various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the detailed description. It is to be understood, however, that the invention is not to be limited to the specific embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a flowchart illustrating a method of setting a recognizer for each gesture in order to recognize a gesture according to an embodiment of the present invention. FIG. 2 is an exemplary view for extracting a hand center according to an embodiment of the present invention, FIG. 3 is a diagram illustrating gesture data tracking according to an exemplary embodiment of the present invention. FIG. 4 is a diagram illustrating normalization of gesture data according to an exemplary embodiment of the present invention. 6 is a view for explaining clustering of gesture data according to an embodiment of the present invention, and Fig. 7 is a view for explaining clustering of gesture data according to an embodiment of the present invention. FIG. 6 is a diagram illustrating a state sequence and a transition probability value according to an embodiment of the present invention. FIG. Setting the recognizer for gesture recognition can be performed in advance in the gesture recognition device for gesture recognition. In FIG. 1, the subject performing each step will be collectively referred to as a gesture recognition device. However, according to the implementation method, the gesture recognition apparatus 100 may receive input of a recognizer set for each gesture. In this case, the gesture recognition apparatus 100 does not include a configuration for recognizer setting for recognition of the gesture It is of course also possible to have a separate configuration.

In step 110, the gesture recognition apparatus 100 acquires an input image by photographing an object (for example, a hand) performing a gesture, and converts the obtained input image into a binary image based on color information. Hereinafter, an embodiment of the present invention will be described mainly on the assumption that an object is a hand. Although it is assumed in FIG. 1 that the gesture recognition apparatus 100 captures an input image by photographing a hand performing a gesture, it is natural that the input image may be separately input from the outside.

The gesture recognition apparatus 100 according to an embodiment of the present invention recognizes a gesture of a user's hand and can convert an input image into a binary image based on color information of a user's hand.

That is, when the gesture recognition apparatus 100 is included in the color information corresponding to the hand based on the color information of each pixel such as RGB or HSV of the input image using general color information of the hand, The pixel value of the pixel may be converted to the first value and the pixel value of the remaining pixels may be binarized by converting the pixel value to the second value.

In an embodiment of the present invention, binarization based on color information of an input image using hand color information has been described in order to explain binarization of an input image based on color information. It should be understood that the present invention is not limited to the case where the hand can be extracted through other methods.

In step 115, the gesture recognition apparatus 100 performs filtering on the binary image to remove the error.

In general, human hands have continuous, smooth boundaries. Therefore, the gesture recognition apparatus 100 masks the point-type error that occurs intermittently and intermittently due to the error of the color information for each pixel corresponding to the hand in the binary image, thereby eliminating the error.

At this time, the filter can be applied in a simple form in the horizontal, vertical, and diagonal directions for real-time processing. In addition, a plurality of iterations of the filter can eliminate errors in the binary image.

More specifically, the gesture recognition apparatus 100 may match a pixel value of a specific pixel with neighboring pixels when a specific pixel among the pixels included in the hand region of the binary image is different from all surrounding pixels.

For example, assume that the pixel value corresponding to the hand region in the binary image is the first value and the other pixel value is the second value. Here, it is assumed that the pixel value of a specific pixel among the pixels included in the hand region is a second value, and that the pixel values of all the surrounding pixels of the specific pixel are the first values. In this case, since an error occurs in the binary image conversion due to the color error in the hand region, the gesture recognition apparatus 100 can remove the error by changing the pixel value of the specific pixel to the first value by applying the filter.

In step 120, the gesture recognition apparatus 100 acquires the hand center using the binary image from which the error has been eliminated, and extracts the hand center in each image to obtain the gesture data.

The gesture recognition apparatus 100 can calculate the average center for all the pixels corresponding to the hand in the binary image from which the error has been eliminated,

In this way, the hand center can be extracted from a plurality of successively obtained images, and data (gesture data) about the extracted hand center can be extracted.

FIG. 2 shows an example in which the center of the hand is extracted from the image.

FIG. 3 is a diagram showing data (hereinafter, referred to as gesture data) for each hand center in a continuous image. FIG. 3 shows an example in which gesture data is obtained by extracting a hand center in units of frames in successive images.

In step 125, the gesture recognition apparatus 100 normalizes and stores the obtained gesture data for positional correction in space.

FIG. 4A shows gesture data obtained from a plurality of consecutive images, and FIG. 4B shows an example of the gesture data normalized for positional correction in space.

In step 130, the gesture recognition apparatus 100 orthogonalizes the normalized gesture data in a plane regardless of the time.

Generally, gesture data obtained is stored in two dimensions, but considering the input time of the gesture, the data becomes three-dimensional.

5 (a) shows gesture data in consideration of time. In order to cluster the three-dimensional gesture data as shown in Fig. 5 (a) in space, the gesture data is orthogonalized regardless of time.

If the gesture data as shown in FIG. 5 (a) is orthogonalized on a two-dimensional plane regardless of the time, it appears as shown in FIG. 5 (b).

In step 135, the gesture recognition apparatus 100 clusters orthogonal gesture data on a two-dimensional plane regardless of time.

At this time, the gesture recognition apparatus 100 may cluster data using a Gaussian Mixture Modeling (GMM). Here, the Gaussian mixture model itself is a matter that is obvious to those skilled in the art, so a detailed description thereof will be omitted.

In addition, the gesture recognition apparatus 100 may set the number of blending to 1 and use the Gaussian blend model to set the shape of the covariance as Full to cluster the vertically running gesture data.

FIG. 6 shows clustering results for vertically running gesture data as shown in FIG. 5 (b).

As shown in FIG. 6, when the clustering is completed, all data of the orthogonal gesture data are classified into clusters closest to the clusters. Here, the class of the cluster may be a cluster index for distinguishing each cluster. At this time, the gesture recognition apparatus 100 can use the distance value between the orthographic gesture data. Distance values can be used for mahalanobis's distance.

In step 140, the gesture recognition apparatus 100 arranges the clustered result values in time order.

If the clustering results of FIG. 6 are sorted in time order, for example, they may appear as follows.

333 ... ... 33332222 ... .2222111 ... 1111555 ... 555333 ... 333444 ... 444

As such, if the class according to the clustering result sorted in chronological order is summarized according to the time order using only the first class, the state sequence for the "a" gesture can be summarized as 3 → 2 → 1 → 5 → 3 → 4 (Step 145).

In step 150, the gesture recognition apparatus 100 calculates a transition probability value between each state (cluster) of the state sequence for the gesture. At this time, the gesture recognition apparatus 100 can calculate a transition probability value between each state (for example, each cluster) in a similar manner to the Baum-Welch algorithm.

This can be expressed by the following equation (1).

는

이거나

일 확률값을 나타낸다.Here, S (t) represents a state (cluster) closest to the gesture data at time t. Also,

The

Or

Represents a probability value.

는 모든 시간에 대해

,

에 있는 제스처 데이터의 개수를 의미한다.In other words,

For every hour

,

The number of the gesture data in the < / RTI >

는 모든 시간에 대해

에 있는 제스처 데이터의 개수를 의미한다.

For every hour

The number of the gesture data in the < / RTI >

)를 계산할 수 있다.Therefore, by using the number 1, a transition probability value (

) Can be calculated.

For example, when the state sequence for the "a" gesture is equal to 3 → 2 → 1 → 5 → 3 → 4, the transition probability values between the respective states (clusters) can be summarized as follows.

The transition probability value of each state (between the classes) in the state sequence for the gesture computed as in the above equation can be expressed as shown in FIG. 7 by applying the transition probability value to each state sequence.

In step 155, the gesture recognition apparatus 100 sets a recognizer for the gesture using the transition probability value and the state sequence between the state sequences.

In Fig. 1, a description has been given mainly of generating a recognizer for the " a " gesture. 1, the gesture recognition apparatus 100 derives a state sequence for each of a plurality of gestures, calculates a transition probability value for the state sequence, and uses them to design a recognizer for each gesture .

The recognizer can be implemented in the same manner as described with reference to FIG. 1 for other gestures, and a duplicate description will be omitted.

FIG. 8 is a flowchart illustrating a method of recognizing a gesture using a plurality of recognizers set by FIG.

That is, in FIG. 8, it will be assumed that a state sequence and a corresponding transition probability value are stored in association with each recognizer in order to recognize each gesture. Hereinafter, the following process will be described assuming that the user performs a gesture to recognize a specific gesture.

In step 810, the gesture recognition apparatus 100 acquires an input image including a hand (object) performing a specific gesture, and converts the obtained input image into a binary image based on hand color information. Accordingly, the gesture recognition apparatus 100 can extract an object region such as a hand.

Next, in step 815, the gesture recognition apparatus 100 extracts the hand center from the binary image and obtains the gesture data, respectively.

Steps 810 to 815 may be repeatedly performed on a frame-by-frame basis for a plurality of consecutive input images or videos to acquire successive gesture data, respectively.

In step 820, each of the gesture recognition apparatuses 100 clusters the obtained gesture data to classify each of the cluster classes. Here, the cluster class represents an index of a class of each gesture data. As described above in Fig. 1, the gesture data may be clustered into a plurality of clusters. At this time, each cluster can be indexed as a class. Therefore, in this specification, the cluster class should be understood as an index of each cluster.

At this time, the gesture recognition apparatus 100 can classify the gesture data into a cluster closest to the gesture data.

Then, in step 825, the gesture recognition apparatus 100 recognizes the specific gesture as a gesture corresponding to a state sequence corresponding to each classified cluster class considering the time sequence.

That is, a specific gesture can be identified by a gesture corresponding to a state sequence in which cluster classes are classified in the same order of states in consideration of each state sequence and time sequence set in a plurality of recognizers.

Here, it is assumed that there are a plurality of state sequences coinciding with the corresponding cluster class.

That is, when it is confirmed that the cluster class is inputted in accordance with the sequence of the training state sequence, when the plurality of state sequences are inputted in order, the gesture recognition apparatus 100 uses each cluster class to generate a state sequence of a plurality of recognizers A distance value for each state may be derived, an average distance may be determined, and a specific gesture may be recognized by a gesture corresponding to a recognizer having a minimum value.

More specifically, the gesture recognition apparatus 100 initializes the distance value for each state of the state sequence set in the plurality of recognizers.

Next, when the cluster class according to the time sequence transitions, the gesture recognition apparatus 100 derives a distance value for the state of the state sequence corresponding to the previous cluster class. At this time, the distance value can update the distance value of the state corresponding to the corresponding cluster class by the distance value between the gesture data classified into the cluster class and the center of the cluster class.

For example, assume that the gesture data is classified as a third cluster class and then transferred to a second cluster class. At this time, the gesture recognition apparatus 100 can calculate and update the distance value of the third state of the plurality of recognizers corresponding to the third cluster class.

When all the processes are completed and the transition is made to the last cluster class, the gesture recognition apparatus 100 may calculate the average distance by summing the distance values of the states of the respective recognizers and then calculating an average value.

The gesture recognition apparatus 100 can recognize the specific gesture as a gesture corresponding to the recognizer having the minimum average distance.

FIG. 9 is a block diagram illustrating an internal configuration of a gesture recognition apparatus according to an embodiment of the present invention.

9, the gesture recognition apparatus 100 according to an exemplary embodiment of the present invention includes an image acquisition unit 910, a binary image conversion unit 915, a tracking unit 920, a classification unit 925, 930, a recognition unit 935, a memory 940, and a control unit 945.

The image acquiring unit 910 acquires an input image by photographing an object performing a gesture. For example, the image acquisition unit 910 may be a camera.

The binary image converting unit 915 performs a function of converting a plurality of consecutive input images including an object (hand) performing a gesture into a binary image based on the color information of the object. At this time, the binary image converting unit 915 may continuously receive the input image according to the time order and convert it into a binary image.

The tracking unit 920 performs a function of extracting the object center recognized in the binary image and tracking the object center to obtain the gesture data.

For example, assuming that the object is a hand, the gesture data can be extracted by extracting the center of the hand from the binary image and tracking the center of the hand in a subsequent input image or frame of video.

For example, when the user uses the gesture using his / her hand, the input image including the hand due to the execution of the gesture can be continuously input, and the tracking unit 920 can continuously input the center of the hand The gesture data can be extracted by tracing.

The classifying unit 925 performs a function of clustering the gesture data into each cluster. The gesture data is input in time sequence. Accordingly, when the time order is considered, the gesture data becomes three-dimensional data. Accordingly, the classifying unit 925 performs orthogonal projection of the gesture data on the two-dimensional plane without considering the time order, and then clusters the gesture data into the respective clusters. Since this is the same as described above, redundant description will be omitted.

This is described in detail in FIG. 1, and redundant description will be omitted.

The recognizer setting unit 930 sets a recognizer for each gesture.

For example, assuming a gesture in which the user inputs "a" using his / her hand, the recognizer setting unit 930 can set the recognizer according to the "a" gesture using the gesture data.

For this, the recognizer setting unit 930 may further include a sequence acquiring unit and a calculating unit.

Here, the sequence acquiring unit functions to generate a state sequence for the target gesture including each cluster in each state using the clustering result.

In addition, the calculation unit calculates a transition probability value between each state (cluster) in the state sequence.

The recognizer setting unit 930 may finally set the recognizer for the gesture using the state sequence and the calculated transition probability value.

This is the same as that described in detail with reference to FIG. 1, so that redundant description will be omitted.

1, the gesture recognition apparatus 100 may include a recognizer setting unit 930. [ In this case, it is natural that the gesture recognition apparatus 100 can receive the recognizers for the gestures from the apparatus for setting the recognizers for the respective gestures.

The recognition unit 935 performs a function of recognizing a specific gesture using a recognizer set for each gesture.

For example, the recognition unit 935 receives the gesture data for the specific gesture continuously through the tracking unit 920, identifies the cluster class for the input gesture data, and determines the state of each recognizer It is possible to recognize a specific gesture by confirming whether or not the sequence is input in the same state (cluster) sequence.

That is, the recognizing unit 935 can recognize the specific gesture as a gesture corresponding to a state sequence in which the cluster classes are inputted in consideration of the time sequence among the state sequences set in the plurality of recognizers.

It is natural that the recognition unit 935 and the recognizer setting unit 930 can share the binary image conversion unit 915 and the tracking unit 920, respectively.

The memory 940 stores various algorithms necessary for operating the gesture recognition apparatus 100 according to an embodiment of the present invention, various data derived from the gesture recognition apparatus 100, and the like.

The control unit 945 controls the internal components of the gesture recognition apparatus 100 according to an embodiment of the present invention such as the binary image conversion unit 915, A setting unit 930, a recognizing unit 935, a memory 940, and the like).

Meanwhile, the gesture recognition method according to an embodiment of the present invention may be implemented in a form of a program command that can be executed through a variety of means for processing information electronically and recorded in a storage medium. The storage medium may include program instructions, data files, data structures, and the like, alone or in combination.

Program instructions to be recorded on the storage medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of software. Examples of storage media include magnetic media such as hard disks, floppy disks and magnetic tape, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, magneto-optical media and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as devices for processing information electronically using an interpreter or the like, for example, a high-level language code that can be executed by a computer.

The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention as defined in the appended claims. It will be understood that the invention may be varied and varied without departing from the scope of the invention.

100: Gesture recognition device
910:
915:
920:
925; Classification section
930:
935:
940: Memory
945:

Claims (17)

(a) converting the input image into a plurality of binary images based on color information of the object in an input image including an object performing a target gesture, the input image being a plurality of consecutive images;
(b) extracting a center of the object from the transformed plurality of binary images to obtain gesture data;
(c) clustering the obtained gesture data, and using the clustering result, generating a state sequence including each class of each cluster in each state;
(d) calculating a transition probability value between each cluster in the state sequence; And
(e) setting a recognizer for the target gesture using the state sequence and the calculated transition probability value. The method according to claim 1,
Before the step (b)
Applying a filter to each of the plurality of binary images to remove an error,
Wherein the step (b) acquires the gesture data using a binary image from which an error has been eliminated. The method according to claim 1,
Prior to step (c)
And normalizing the gesture data based on the received gesture data. The method according to claim 1,
The step (c)
Orthogonalizing the obtained gesture data to a plane regardless of a time sequence;
Clustering the orthographic gesture data;
Arranging classes of each cluster according to the time sequence according to the clustering result; And
And generating the state sequence including only the first class of each of the classes sorted according to the time order.

5. The method of claim 4,
Wherein the clustering comprises:
Calculating a distance value between the orthographic gesture data and clustering the distance value; 6. The method of claim 5,
Wherein the distance value is calculated as a distance of Mahalanobis between the orthographic gesture data.

The method according to claim 1,
Wherein the transition probability value
And a probability that a second cluster appears after the first cluster in the status sequence. A method of recognizing a gesture with a plurality of recognizers for a plurality of gestures being set, the recognizer comprising a state sequence for each gesture and a transition probability value,
Converting each of the plurality of input images into binary images based on color information of the object in a plurality of consecutive input images including an object performing a specific gesture;
Extracting a center of the object from the plurality of converted binary images to obtain gesture data;
Identifying a cluster class for each gesture data; And
Wherein the cluster class recognizes the specific gesture with a gesture that matches the state sequence.

9. The method of claim 8,
Wherein identifying the cluster class of each gesture data comprises:
And classifying cluster classes into clusters having a distance closest to each of the gesture data. 9. The method of claim 8,
When recognizing the gesture, if there are a plurality of cluster classes matching the state sequence,
Updating a distance value of each state of the state sequence of the plurality of recognizers according to a transition between the cluster classes; And
And recognizing the specific gesture with a gesture corresponding to a recognizer whose average distance of the updated distance value is the smallest. 11. The method of claim 10,
Wherein updating the distance of each state of the state sequence comprises:
Calculating a distance value between the center of the cluster class and the gesture data included in the cluster class before the transition according to the transition of the cluster class and updating the distance value for the state of the state sequence according to the cluster class; Gesture recognition method. 12. A recording medium product on which a program code for performing the method according to any one of claims 1 to 12 is recorded.

A binary image converting unit for converting the input image into a plurality of binary images based on color information of the object in an input image including an object performing a target gesture, the input image being a plurality of consecutive images;
A tracking unit for extracting gesture data by extracting a center of the object from the transformed plurality of binary images;
A sequence obtaining unit for clustering the obtained gesture data and generating a state sequence for the target gesture including each cluster in each state using the clustering result;
A calculation unit for calculating a transition probability value between each cluster in the state sequence; And
And a recognizer setting unit for setting a recognizer for the target gesture using the state sequence and the calculated transition probability value. 14. The method of claim 13,
The sequence acquiring unit,
The gesture data obtained by vertically aligning the obtained gesture data on a plane regardless of a time order, clustering the orthogonalized gesture data, sorting classes of each cluster according to the time order according to the clustering result, Wherein the gesture recognition unit acquires the state sequence including only the first class of each class among the classes aligned to the gesture recognition unit.

14. The method of claim 13,
Wherein the sequence acquiring unit calculates and clusters a distance value between the vertically running gesture data. An apparatus for recognizing a gesture in which a plurality of recognizers for a plurality of gestures are set up, the apparatus comprising a state sequence for each gesture and a transition probability value,
A binary image transform unit for transforming the plurality of input images into binary images based on color information about the object in a plurality of consecutive input images including an object performing a specific gesture;
A tracking unit for extracting a center of the object from the plurality of converted binary images to obtain gesture data;
A classifying unit for classifying a cluster class for each gesture data;
And a recognizing section which recognizes the specific gesture with a gesture corresponding to a state sequence coinciding with the cluster class in consideration of a time sequence.

17. The method of claim 16,
Wherein,
Updates the distance value of each state of the state sequence of the plurality of recognizers according to the transition between the cluster classes when the number of state sequences coinciding with the cluster class according to the time order is larger than the average distance of the updated distance values And recognizes the specific gesture with a gesture corresponding to the recognizer that is the minimum.

Images