KR101636171B1 - Skeleton tracking method and keleton tracking system using the method - Google Patents

Abstract

A skeleton tracking method capable of increasing the accuracy of a skeleton model and a skeleton tracking system using the same. The skeleton tracking method includes the steps of generating N pieces of photographing information for an arbitrary person having a plurality of joints from N depth sensor cameras (N is an integer equal to or greater than 2) from the photographing information, Extracting N positional information from each of the joints, calculating integrated position information for each of the joints from N positional information in each of the joints, To form a skeleton model for the person. As described above, by calculating the integrated position information for each of the joints from the N positional information in each of the joints, a more accurate skeleton model can be formed.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a skeleton tracking method and a skeleton tracking method using the same,

The present invention relates to a skeleton tracking method and a skeleton tracking system using the method, and more particularly, to a skeleton tracking method and a skeleton tracking system using the method.

Recently, a depth sensor camera combining a RGB camera and a depth sensor has been popular. A representative product of the depth sensor camera is a Kinect of Microsoft. The depth sensor camera can photograph an arbitrary object and generate image and position information about the object.

For example, when the depth sensor camera photographs an arbitrary person to generate shot data, position information of each of the joints of the person can be extracted from the shot data, and the position of each of the joints One skeleton model can be formed from the information. Furthermore, it is also possible to use a user interface for controlling an arbitrary application program by recognizing a movement of a user through the movement of the skeleton model.

Such a user interface can generate relatively erroneous input values as compared to a unimodal interface that provides accurate input values to produce error inputs in the application. Therefore, it is important to extract a more accurate skeleton model in order to suppress such error input.

Meanwhile, as one method of increasing the accuracy of the user interface, a method of using two depth sensor cameras has been studied. That is, the accuracy of the user interface can be increased by integrating the two skeleton models extracted from the two depth sensor cameras.

Specifically, when two skeleton models extracted from two depth sensor cameras are extracted and integrated, the method of ignoring the remaining one position information based on the position information collected at a distance close to the actual user's body Can be used. However, this method also has limitations in increasing the accuracy of the user interface when the location information collected at a close distance includes its own error.

SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a skeleton tracking method capable of increasing the accuracy of a skeleton model.

Another object of the present invention is to provide a skeleton tracking system using the skeleton tracking method.

The skeleton tracking method according to an embodiment of the present invention includes the steps of generating N pieces of photographing information for an arbitrary person having a plurality of joints from N depth sensor cameras (N is an integer of 2 or more) Extracting N positional information in each of the joints from the imaging information, calculating integrated positional information for each of the joints from N positional information in each of the joints, And forming a skeleton model for the person using the integrated position information for each of the plurality of objects.

The combined position information for each of the joints may include a value of any one of a simple average, a weighted average, and a root mean square of N positional information in each of the joints.

The skeleton tracking method may further include calculating a skeleton similarity between the two models by comparing the skeleton model with a reference model.

The reference model may include positional information of corresponding reference joints of the joints.

Wherein the step of calculating the skeleton similarity includes calculating joint similarities by comparing position information of the joints and position information of the reference joints with each other, and calculating the skeleton similarity using the joint similarities . ≪ / RTI >

The skeleton similarity may include a simple average value of the joint similarities.

Wherein calculating the joint similarities comprises: calculating Euclidean distances representing distances between two positions corresponding to each other from position information of the joints and position information of the reference joints; And calculating the joint similarities using Cladian distances.

Each of the joint similarities may include a value obtained by subtracting 1 from a value obtained by dividing each of the euclidean distances by a set value k.

The skeleton tracking method may further include setting the set value k in consideration of at least one of the number of the depth sensor cameras and the angle of view of each of the depth sensor cameras.

The skeleton tracking method may further include displaying the skeleton similarity.

Meanwhile, the skeleton tracking system according to an embodiment of the present invention includes N depth sensor cameras and a skeleton tracking module (N is an integer of 2 or more).

The depth sensor cameras capture an arbitrary person having a plurality of joints and generate N photographing information for the person, respectively. Wherein the skeleton tracking module extracts N positional information in each of the joints from the photographing information, calculates integrated positional information on each of the joints from N positional information in each of the joints, A skeleton model for the person is formed using the integrated position information for each of the joints.

The combined position information for each of the joints may include a value of any one of a simple average, a weighted average, and a root mean square of N positional information in each of the joints.

The skeleton tracking system may further include a similarity evaluation module that compares the skeleton model with a reference model and calculates a skeleton similarity between the two models.

The reference model may include positional information of corresponding reference joints of the joints.

The similarity evaluating module may calculate joint similarities by comparing the position information of the joints and the position information of the reference joints with each other, and calculate the skeleton similarity using the joint similarities.

The skeleton similarity may include a simple average value of the joint similarities.

The similarity-evaluating module calculates Euclidean distances representing distances between two positions corresponding to each other based on the position information of the joints and the position information of the reference joints, and uses the Euclidean distance The joint similarities can be calculated.

Each of the joint similarities may include a value obtained by subtracting 1 from a value obtained by dividing each of the euclidean distances by a set value k.

The similarity evaluation module may set the set value k in consideration of at least one of the number of the depth sensor cameras and the angle of view of each of the depth sensor cameras.

The skeleton tracking system may further include a display device for displaying the skeleton similarity.

According to the skeleton tracking method and the skeleton tracking system using the same according to the present invention, N position information in each of the joints is extracted from N photographing information generated from the N depth sensor cameras, As the integrated position information for each of the joints is calculated from the N positional information in each of the joints, a more accurate skeleton model can be formed.

That is, the skeleton tracking method or system does not select the nearest one of the N positional information items in each of the joints to form a skeleton model, but rather a simple average of N positional information items in each of the joints, Calculating a value of any one of weighted average and root-mean-square roots, extracting integrated position information for each of the joints, and forming a skeleton model through the extracted values, The error may be reduced to 1 / N, and the probability of forming a more accurate skeleton model may be increased.

Also, the skeleton tracking method or system may evaluate the similarity of the skeleton model with the reference model by calculating the skeleton similarity between the two models by comparing the skeleton model with the reference model.

Also, when the set value k indicating the lowest value of the skeleton similarity is adjusted, the rate of change in similarity or the sensitivity can be adjusted. Accordingly, when the skeleton tracking method or system is applied to a rehabilitation service that provides rehabilitation training, the set value k may be a value that determines a training difficulty in the rehabilitation service.

1 is a flowchart for explaining a skeleton tracking method according to an embodiment of the present invention.
2 is a view for explaining an example of a depth sensor camera applied to the skeleton tracking method of FIG.
3 is a view for explaining an example of photographing using four depth sensor cameras of FIG.
FIG. 4 is a view showing an example of joints of a person to be imaged in the skeleton tracking method of FIG. 1. FIG.
FIG. 5 is a flowchart for explaining a process of calculating the skeleton similarity in the skeleton tracking method of FIG.
6 is a block diagram for explaining an example of a skeleton tracking system using the skeleton tracking method of FIG.
7 is a table for explaining evaluation variables used in the skeleton tracking method of FIG.
8 is a graph showing changes in the skeleton similarity according to the logarithmic change of the depth sensor camera.
9 is a graph showing a change in skeleton similarity according to a change in the set value k.

The present invention is capable of various modifications and various forms, and specific embodiments are illustrated in the drawings and described in detail in the text.

It should be understood, however, that the invention is not intended to be limited to the particular forms disclosed, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. The terms first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprising" or "having ", and the like, are intended to specify the presence of stated features, integers, steps, operations, elements, parts, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, parts, or combinations thereof.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings.

FIG. 1 is a flowchart for explaining a skeleton tracking method according to an embodiment of the present invention. FIG. 2 is a view for explaining an example of a depth sensor camera applied to the skeleton tracking method of FIG. 1, FIG. 4 is a view showing an example of joints of a person to be photographed in the skeleton tracking method of FIG. 1; FIG. 4 is a view for explaining an example of photographing using four depth sensor cameras of FIG.

Referring to FIGS. 1 to 4, a skeleton tracking method according to an embodiment of the present invention relates to a tracking method capable of measuring and tracking human movements, and can be performed by an arbitrary computer system.

In the skeleton tracking method, N photographing information for an arbitrary person having a plurality of joints is first generated from N depth sensor cameras 100 (S100). Here, N is an integer of 2 or more.

For example, each of the depth sensor cameras 100 may include an RGB camera, an infrared depth sensor, and an infrared projector. The infrared projector emits infrared rays of a number of points at regular intervals toward the person, and the infrared depth sensor senses the infrared rays reflected by the person and acquires distance information between the camera and the object at each point. The RGB camera acquires image information of the person. As described above, each of the photographing information may include distance information between the camera and the object at each point and image information of the person.

In the present embodiment, the joints may be composed of 20 as shown in FIG. For example, the twenty joints include a head, a shoulder center, a left shoulder, a shoulder right, a spine, a left elbow, a right elbow, WristLeft, WristRight, HandLeft, HandRight, HipCenter, HipLeft, HipRight, KneeLeft, KneeRight, , AnkleLeft, AnkleRight, FootLeft, and FootRight joints.

Next, N pieces of position information in each of the joints can be extracted from the photographing information (S200).

Here, the position information of an arbitrary joint extracted from the depth sensor camera n can be expressed by the following equation.

CAMn.PositionInfo _Joint = (x _n , y _n , z _n )

Then, integrated location information for each of the joints is calculated from N positional information in each of the joints (S300).

For example, the joint position information for each of the joints can be extracted by calculating one of the simple average, the weighted average, and the root mean square of the N positional information in each of the joints. At this time, the weighted average value can be extracted by a calculation method of giving a higher weight to a distance between a camera and an object and obtaining an average.

Assuming that the joint position information for each of the joints is calculated as a simple average value of N position information in each of the joints, the joint position information for an arbitrary joint can be expressed by the following equation.

Calculated.PositionInfo _Joint = (x _c , y _c , z _c )

x _c = (x ₁ + x ₂ + x ₃ + ... + x _n ) / n

y _c = (y ₁ + y ₂ + y ₃ + ... + y _n ) / n

z _c = (z ₁ + z ₂ + z ₃ + ... + z _n ) / n

Next, a skeleton model for the person is formed using the integrated position information for each of the joints (S400). That is, the coordinate values of the joints may be connected to each other to form a single skeleton model.

Next, the skeleton similarity between the two models is calculated by comparing the skeleton model with a reference model (S500). Here, the reference model refers to a reference object for evaluating the similarity with the skeleton model, and may include position information of corresponding reference joints. On the other hand, before calculating the skeleton similarity, the step of generating the reference model may be further performed.

Then, the skeleton similarity is displayed externally (S600). At this time, the skeleton similarity may be graphically displayed through a monitor, but may also be printed and displayed on paper. Meanwhile, the skeleton similarity may be transmitted to another external system through a wired or wireless communication method.

FIG. 5 is a flowchart for explaining a process of calculating the skeleton similarity in the skeleton tracking method of FIG.

Referring to FIG. 5, the process of calculating the skeleton similarity according to step S500 may first calculate joint similarities by comparing positions of the joints and position information of the reference joints corresponding to each other ).

Specifically, in step S510, a set value k necessary for calculating each of the joint similarities is set (S512). At this time, the set value k may be a value freely input by a user, but may be a previously stored arbitrary value or a value automatically calculated by an arbitrary condition.

For example, the set value k may be automatically calculated using at least one of the number of the depth sensor cameras and the angle of view of each of the depth sensor cameras. Alternatively, a range in which the user can set the set value k may be determined by at least one of the number of the depth sensor cameras and the angle of view of each of the depth sensor cameras.

Next, the Euclidean distances representing the distances between two positions corresponding to each other from the position information of the joints and the position information of the reference joints may be calculated (S512).

For example, any one of the joint B are standard and correspond to each other and the connection point A, the position information of the reference connection point A (x _a, y _a, z _a), and the location information of the joint B (x _b, y _b , z _b ), the euclidean distance between the joint B and the reference joint A can be expressed by the following equation.

Then, the joint similarities may be calculated using the euclidean distance (S514). In this case, each of the joint similarities may be a value obtained by subtracting 1 from a value obtained by dividing each of the euclidean distances by a set value k.

Here, the joint similarity according to the euclidean distance between the joint B and the reference joint A can be expressed by the following equation.

Then, when the joint similarities are calculated, the skeleton similarity can be calculated using the joint similarities (S520). At this time, the skeleton similarity may be a simple average value of the joint similarities.

For example, the skeleton similarity may be represented by the joint similarities corresponding to the twenty joints shown in FIG. 4 as follows.

Meanwhile, in this embodiment, the step S512 of setting the set value k may be performed at any time point before the step S514 of calculating the joint similarities. For example, the step S512 of setting the set value k may be performed before the step S100 of generating the shooting information. In this case, when the setting value k is set (S512) before the step of generating the photographing information (S100), the setting value k is calculated in consideration of the value of the skeleton similarity derived by the previously performed process May be set.

Hereinafter, an example of a skeleton tracking system using the skeleton tracking method described above will be described in detail.

6 is a block diagram for explaining an example of a skeleton tracking system using the skeleton tracking method of FIG.

Referring to FIG. 6, the skeleton tracking system according to the present embodiment may include the depth sensor cameras 100, the skeleton tracking module 200, the similarity evaluation module 300, and the display device 400.

The depth sensor cameras 100 photograph an arbitrary person having a plurality of joints and generate N photographing information for the person, respectively. That is, the depth sensor cameras 100 may perform step S100.

The skeleton tracking module 200 extracts N positional information from each of the joints from the photographing information and calculates integrated positional information on each of the joints from N positional information in each of the joints And forms a skeleton model for the person using the integrated position information for each of the joints. That is, the skeleton tracking module 200 may perform steps S200, S300, and S400.

Specifically, the skeleton tracking module 200 includes a joint position information extracting unit 210 for extracting N position information from each of the joints from the photographing information, A skeleton model forming unit 230 for forming a skeleton model for the human using the combined position information for each of the joints, . ≪ / RTI > At this time, the joint position information extraction unit 210, the integrated position information calculation unit 220, and the skeleton model formation unit 230 may be physically separated process elements, Lt; RTI ID = 0.0 > process elements. ≪ / RTI >

The joint position information extraction unit 210 extracts N position information from each of the joints from the photographing information by the joint position information extraction unit 210 performing various processing on the shooting information A concept including not only extracting or generating new N positional information in each of the joints but also using N positional information in each of the joints directly from the depth sensor cameras 100 Lt; / RTI >

For example, the photographing information provided to the joint position information extracting unit 210 in the depth sensor cameras 100 may directly include N position information in each of the joints. That is, each of the depth sensor cameras 100 performs self-processing of the photographed image and data to generate N position information in each of the joints and provides the generated position information to the joint position information extracting unit 210 .

The similarity evaluation module 300 may compute the skeleton similarity between the two models by comparing the skeleton model with the reference model.

Specifically, the similarity evaluating module 300 calculates the Euclidean distances, which are distances between two positions corresponding to each other, from the position information of the joints and the position information of the reference connecting points, The joint similarities are calculated using the distance, and the skeleton similarity can be calculated using the joint similarities. That is, the similarity evaluation module 300 may perform the operation S500.

For example, the similarity evaluation module 300 may include a K value setting unit 310, a reference model storage unit 320, and a similarity calculation unit 330. The K value setting unit 310, the reference model storage unit 320, and the similarity calculation unit 330 may be processes or memory elements physically separated from each other. However, the K value setting unit 310, Or may be programs executed in a process element.

The K value setting unit 310 may set the set value k necessary for calculating each of the joint similarities. That is, the K value setting unit 310 may perform step S512. Here, if the set value k is a value freely input by the user, the K value setting unit 310 may include input means for receiving an input value from the user.

The reference model storage unit 320 stores information on the reference model for calculating the euclidean distances. Here, the information about the reference model should be stored in the reference model storage 320 before performing the process according to the present embodiment.

The similarity calculation unit 330 may include information on the skeleton model provided from the skeleton model formation unit 230, information on the reference model provided from the reference model storage unit 320, and information on the K value setting unit 310 The skeleton similarity degree may be calculated using the set value k provided from the skeleton similarity degree calculating unit. That is, the similarity calculation unit 330 may perform the steps S514, S516, and S520.

In the present embodiment, the skeleton tracking module 200 and the similarity evaluation module 300 may be process elements or computer devices that are physically separated from each other, but may be a program element or a computer device Or the like.

The display device 400 may receive the skeleton similarity from the similarity calculation unit 330 and display the same. In addition, the display device 400 receives the skeleton model from the skeleton model forming unit 230, receives the reference model from the reference model storing unit 320, and stores the skeleton model and the reference model at the same time Or may display only one of them.

Meanwhile, in the present embodiment, when the set value k indicating the lowest value of the skeleton similarity is adjusted, the rate of change or sensitivity to the skeleton similarity can be adjusted. Specifically, as the set value k increases, the width of the similarity change decreases, and as the set value k decreases, the width of the similarity change may become larger. At this time, by setting the rate of change or sensitivity to the skeleton similarity determined by the set value k, the difficulty level setting function can be utilized when the patient performs rehabilitation training following the trainer's operation in the rehabilitation service.

Hereinafter, an experimental result to confirm that the accuracy of the skeleton model increases as the number of the depth sensor cameras increases.

First, the hardware specifications of the experimental environment are Windows 7 64bit Operating System, Intel Core i5-750 Lynnfield Quad-Core 2.66GHz CPU, and 8 Gigabyte RAM. The DeepSensor camera uses four Microsoft Kinect for Windows products, which are among the most commercially available products. The reason why the number of depth sensor cameras is four is because the skeleton model is based on the front view of the user's body, so it is most effective when four cameras are considered to be able to extract all parts of the front face.

The goal of the experiment is to check whether the accuracy of the integrated skeleton model is improved by using the proposed method as the number of depth sensor cameras increases. Therefore, the depth sensor camera should be increased from 1 to 4, but the experimental method should be the same.

Then, a setting value k that determines the sensitivity of the change in similarity degree is set. As the set value k (range is 0.00m to 1.00m) is set to a larger value, the degree of similarity change is small to the error of the skeleton model. When the setting value k is small, the sensitivity is also sensitive to small errors of the skeleton model. The threshold value k is 0.25m, which is very sensitive, while the change of similarity can be confirmed. Therefore, in this experiment, the threshold value k is set to 0.25m and used for the similarity measurement.

The generation of the reference model is a process of generating a reference value for the similarity measurement. The proposed method is used to integrate and record the user 's skeleton model for 10 seconds (30 frames per second). Depth Sensor Repeat the camera from 1 to 4, and create a reference model based on the number of depth sensor cameras n.

The similarity measure between the reference model and the integrated skeleton model measured in real time, ie, skeleton similarity, is calculated every second. Since the reproduction time of the reference model is 10 seconds, 10 skeleton similarities are output for each depth sensor camera.

7 is a table for explaining evaluation variables used in the skeleton tracking method of FIG.

The evaluation parameters of this experiment are set value k, joint position information, time, similarity, and the number of depth sensor cameras. The concept of each evaluation variable is shown in FIG.

8 is a graph showing changes in the skeleton similarity according to the logarithmic change of the depth sensor camera.

Referring to FIG. 8, it can be seen that the degree of similarity increases as the number of depth sensor cameras increases. That is, using this proposed scheme means that the accuracy of skeleton tracking is increased. In similarity graphs using one or two Kinect, the degree of similarity drops sharply when there is motion, but the similarity graph using 3 or 4 Kinect does not change much. Compared with the use of three Kinect and four Kinect, the use of four Kinect is able to extract a good quality skeleton model because the fluctuation range of similarity variation is narrow.

9 is a graph showing a change in skeleton similarity according to a change in the set value k.

Referring to FIG. 9, the change in similarity degree according to the setting of the set value k can be known. The set value k is set to 0.25 (m), 0.50 (m), 0.75 (m), and 1.00 (m) from FIGS. 9A to 9D, and as the set value k increases, And as the set value k becomes smaller, the variation range of the similarity degree increases.

Meanwhile, the skeleton tracking method and system according to the present embodiment can be applied to a rehabilitation service that compares a rehabilitation training posture of a user and a rehabilitation trainer to determine a degree of similarity and support a correct attitude.

The reference skeleton data of the trainer that has been stored in advance is reproduced and the reference model is reproduced. When the user follows the rehabilitation operation, the correct posture is corrected. The user's rehabilitation behavior is compared based on the trainer's reference skeleton data, and the similarity is expressed as a percentage. The degree of difficulty adjustment function can be given by setting the setting value k that determines the width of the degree of similarity change. The difficulty of rehabilitation training can be controlled by continuous posture correction and patient's improvement status. If the condition of the patient is much improved, the set value k can be set small to raise the difficulty level, thereby enhancing the effect of the rehabilitation training.

As described above, according to the present embodiment, N pieces of position information in each of the joints are extracted from the N pieces of photographing information generated from the N depth sensor cameras, N pieces of position information in each of the extracted joints A more accurate skeleton model can be formed as the integrated positional information for each of the joints is calculated from the joint position information.

That is, the skeleton tracking method or system does not select the nearest one of the N positional information items in each of the joints to form a skeleton model, but rather a simple average of N positional information items in each of the joints, Calculating a value of any one of weighted average and root-mean-square roots, extracting integrated position information for each of the joints, and forming a skeleton model through the extracted values, The error may be reduced to 1 / N, and the probability of forming a more accurate skeleton model may be increased.

Also, the skeleton tracking method or system may evaluate the similarity of the skeleton model with the reference model by calculating the skeleton similarity between the two models by comparing the skeleton model with the reference model.

Also, when the set value k indicating the lowest value of the skeleton similarity is adjusted, the rate of change in similarity or the sensitivity can be adjusted. Accordingly, when the skeleton tracking method or system is applied to a rehabilitation service that provides rehabilitation training, the set value k may be a value that determines a training difficulty in the rehabilitation service.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

100: Depth sensor camera 200: Skeleton tracking module
210: joint position information extraction unit 220: integrated position information calculation unit
230: Skeleton model forming unit 300:
310: K value setting unit 320: Reference model storage unit
330: similarity calculation unit 400: display device

Claims (20)

Generating N photographic information for any person having a plurality of joints from N depth sensor cameras (where N is an integer greater than or equal to 2);
Extracting N positional information from each of the joints from the photographing information;
Calculating integrated position information for each of the joints from N positional information in each of the joints;
Forming a skeleton model for the human using joint position information for each of the joints; And
And comparing the skeleton model with a reference model to calculate a skeleton similarity between the two models,
The reference model
Each of the joints including position information of corresponding reference joints,
The step of calculating the skeleton similarity
Calculating Euclidean distances representing distances between two positions corresponding to each other from position information of the joints and position information of the reference connection points;
Calculating joint similarities using the euclidean distance; And
And calculating the skeleton similarity using the joint similarities,
Each of the joint similarities
1 by a value obtained by dividing each of the euclidian distances by a set value k. The method of claim 1, wherein the combined position information for each of the joints
A weighted average, and a root-mean-square root of N positional information in each of the joints. delete delete delete The method of claim 1, wherein the skeleton similarity is
And a simple average value of the joint similarities. delete delete The method according to claim 1, further comprising setting the set value k in consideration of at least one of the number of the depth sensor cameras and the angle of view of each of the depth sensor cameras. The method of claim 1, further comprising displaying the skeleton similarity. N depth sensor cameras (N is an integer of 2 or more) for capturing an arbitrary person having a plurality of joints to generate N photographing information for the person, respectively;
Extracting N pieces of position information in each of the joints from the photographing information, calculating integrated position information on each of the joints from N pieces of position information in each of the joints, A skeleton trekking module for forming a skeleton model for the person using the integrated position information; And
And a similarity evaluating module for comparing the skeleton model with a reference model and calculating a skeleton similarity between the two models,
The reference model
Each of the joints including position information of corresponding reference joints,
The similarity-
Calculating Euclidean distances representing distances between two positions corresponding to each other from the position information of the joints and the position information of the reference joints, and calculating joint similarities using the Euclidean distances Calculating the skeleton similarity using the joint similarities,
Each of the joint similarities
1 by a value obtained by dividing each of the Euclidean distances by a set value k. 12. The method of claim 11, wherein the combined position information for each of the joints
A weighted average and a root mean square of the N positional information in each of the joints. delete delete delete 12. The method of claim 11, wherein the skeleton similarity is
Wherein the skeleton tracking system includes a simple average value of the joint similarities. delete delete 12. The method of claim 11, wherein the similarity-
Wherein the set value k is set in consideration of at least one of the number of the depth sensor cameras and the angle of view of each of the depth sensor cameras. 12. The skeleton tracking system of claim 11, further comprising a display device for displaying the skeleton similarity.

Images