KR101687175B1 - Target Analysis Apparatus and Method of the same - Google Patents

Abstract

The present invention provides a method and apparatus for target analysis based on intensity and depth images. The apparatus includes: a body detecting unit that detects a body of a target among intensity image of a target; The intensity threshold is calculated based on the detected intensity value of the body, the intensity image is converted into the binary image using the intensity threshold, and the mask image is acquired by masking the depth image of the target using the binary image as a mask. Installment; And an active portion detecting portion for detecting an active portion of the target from the masking depth image.

Description

[0001] The present invention relates to a target analysis apparatus and method,

The present invention relates to a video target analysis method and apparatus, and more particularly, to a method and apparatus for target analysis based on depth image and intensity image, Thereby obtaining the operation information of the target.

Extensive research is currently underway on video target analysis techniques in the areas of computer video and model identification, depending on a wide range of applications, such as 3D animation, games, and human interfaces. In order to proceed with the animation simulation, it is first necessary to detect each part of the target of the video and acquire the operation information of the target.

The bottom-up method and the up-bottom method are generally used to detect the constituent parts of the target. In the above method, the target portion is first detected using the feature of the target portion, for example, the detection of the skin color, the detection of the line shape of the limb, and the like, and then the matching is performed on the detected candidate portion according to the detection model . This method is highly dependent on the development of typical features of the target portion and the suppression of background noise. In the latter method, iterative search is performed on the target shape space, a hypothesis is proposed, and evaluation is conducted for each hypothesis. This method requires that the search space be very small in order to speed up the calculation.

During the video target analysis process, it is generally necessary to divide the foreground among the raw images, that is, to remove the confusing background area, thereby accurately detecting the target. Conventional foreground segmentation methods are based primarily on color images of CCD cameras. The background of a color image is generally very complicated, so the amount of division computation along the foreground of a color image is very large and inaccurate. Although we have proposed a foreground segmentation method based on the current depth image, the above method is inaccurate.

Once the foreground segmentation is inaccurate, subsequent successive target analysis becomes very difficult and the target analysis results are subject to interference of similar background noise. In addition, abundant edge features often appear in confused backgrounds, and this edge feature becomes noise and affects the edge feature analysis results during detection or hypothesis evaluation of the target portion.

Therefore, there is a need for a target analysis method and apparatus that improves target analysis performance by precisely dividing the foreground.

According to an embodiment of the present invention, it is possible to provide a target analysis method and apparatus that improves target analysis performance by precisely dividing the foreground.

According to another embodiment of the present invention, a target analysis method and apparatus capable of realizing accurate ridge verification can be provided.

A target analyzing apparatus and method according to an embodiment of the present invention is not limited to the use of a color image, and can use, for example, a strength image and a depth image of a Time of Flight (TOF) camera.

In order to achieve the object of the present invention, there is provided a target analyzing apparatus for inputting a depth image and an intensity image of a target, the apparatus comprising: a main body detecting unit for detecting a body of a target among intensity image of the target; Calculating a intensity threshold according to the detected intensity value of the body, switching the intensity image to a binary image using the intensity threshold, and masking the depth image of the target using the binary image as a mask, A foreground division for acquiring an image; And an active portion detecting portion that detects an active portion of the target among the masked depth images.

According to the present invention, the body detecting section trains the statistical learning method to detect the body of the target among the intensity images. Also, candidates matching the geometric constraint among the plurality of candidates for the component of the body are selected as the components of the body.

In addition, the foreground partitions use a value obtained by statistically averaging the intensity values of the detected bodies as intensity threshold values.

According to the present invention, the active portion detecting portion may include: a ridge detecting portion that detects a plurality of ridges among the masked intensity images; And a ridge verifying unit for verifying the plurality of ridges and selecting an optimum ridge.

A ridge detecting section distinguishes between an active portion and an inactive portion of the body among the depth images of the mask, and calculates the plurality of ridge candidates from the candidates of the active portion. The active portion and the inactive portion are distinguished from each other by setting the average depth value of the inactive portion detected by the body detecting portion as a threshold value.

The ridge verification unit determines a connection point between the active portion and the inactive portion according to the body detected by the body detection unit, forms a virtual ridge by connecting one end of the predetermined connection point and one of the plurality of ridges, A plurality of ridges and virtual ridges are verified.

In addition, the inactive portion of the target includes a human head and body, and the active portion of the target includes a human limb.

To achieve the above object, a target analysis method based on a depth image and an intensity image of a target includes the steps of: a) detecting a body of a target among intensity image of the target; b) calculating a strength threshold according to an intensity value of the body of the detected target; And c) converting the intensity image to a binary image using the intensity threshold; d) masking the depth image using the binary image and obtaining a mask depth image; And e) detecting an active portion of the target of the mask depth image.

Thus, unlike the foreground segmentation of color images in the prior art, the present invention first uses masking to the simple body of the target according to the intensity image using a simple and fast method. Then divides the foreground of the depth image according to the masking and further detects ridges of the active portion of the target. The partial detector obtained by training using the statistical learning method can be used to detect the body of the target and can measure the position and the ratio of the target roughly. This can significantly reduce the parameter search space. In addition, the ridge verification can verify the ridge accurately using the virtual part.

1 is a block diagram showing a structure of a target analyzing apparatus according to the present invention.
2 is an illustration showing an example of a strength image and a depth image.
3 is a view showing the main body of the target.
4 (a) to 4 (d) are views showing examples of intensity image binary image, depth image and masking depth image.
5 is a view showing the structure of the active portion detecting unit.
6 is a diagram showing the structure of the ridge detection unit.
7 is a diagram showing the detection process of the ridge detection unit.
8A and 8B are diagrams showing examples of search tree and ridge verification results of a target, respectively.
9 is a flowchart of a target analysis method according to the present invention.

Hereinafter, a target analyzing apparatus and method of the present invention will be described in detail with reference to the drawings. The target analyzing apparatus and method of the present invention is to conduct analysis on a target of interest in a video stream to obtain information of each part of the video target.

In general, for analysis-pending video targets, most are divided into two parts: one part is a relatively fixed inactive part such as a shape, a ratio, for example, a person's head and body, and the base of a particular device; The other part is the part of the activity that can be active and whose shape is not relatively fixed, such as the limbs of a person, the arm of a particular device, and so on. In the present invention, detection is performed using a classifying learning device or a detector that has been trained on a simple main body through a sample statistical learning method training, and background noise is removed through foreground division for a relatively complex activity part, and detection can be performed.

It should be noted that, for convenience of explanation, the target analyzing apparatus and method of the present invention will be described below by taking a human target as an example. However, the present invention is not limited to the treatment of the human body target.

Referring to FIG. 1, FIG. 1 is a block diagram showing the structure of a target analysis apparatus 100 of the present invention. The target analyzing apparatus 100 of the present invention includes a body detecting unit 101, a foreground dividing unit 102, and an active portion detecting unit 103.

The target analysis apparatus 100 receives the intensity image and the depth image of the target from the outside and inputs it as an input of the system. Alternatively, the target analyzing apparatus 100 of the present invention may further include a device for capturing the intensity image and the depth image of the target, for example, a time of flight (TOF) camera.

In an input system, a TOF camera can functionally include two cameras, one is a high resolution color CCD camera and the other is a low resolution TOF camera. The output of a color CCD camera is a color video stream, and the output of a TOF camera is an intensity video stream and a depth video stream. The color camera and the TOF camera share the same optical axis and are stored in pixels between the actual color (color image) and the TOF camera (intensity and depth image). At the manufacturing stage, the camera has already been calculated and set. The output of the camera is a six component video stream, which is associated with color components (red, green, blue) and coordinates (x, y, z) in three dimensional space. For non-integrated camera sets, color cameras and TOF cameras are used in combination with different locations or baselines. The video images from both cameras (color and intensity) are first calculated using the standard calculation method. The alignment between the color camera image and the TOF camera image is then accomplished by switching and optimizing the image matching of the feature points from both images. In this way, a six-component video can be obtained as an input.

The intensity image reflects the reflective capability of the target surface material, and each pixel value represents the reflection intensity value of the target surface material. The depth image reflects the distance between the camera and the target surface, and each pixel value represents the depth value from the camera to the target surface.

2 (a) and 2 (b) show examples of the intensity image and the depth image, respectively.

As shown in FIG. 3, the body detecting unit 101 detects the body of the target from the intensity image of the target, and in particular detects the inactive part including the human head and the body in the case of a human body target among the bodies. Conventionally, a target body detection method uses a color image, for example, a background difference detection method, a color classification method, and the like. However, the present invention uses the intensity image as a basis for detecting the target main body. The intensity image can be acquired, for example, by a TOF camera.

The body detecting unit 101 uses a partial classifier or a detector for training based on the statistical learning method of the sample. For example, the main body detection unit 101 may use an open statistical learning method. Since the input is a century image, the detection speed is faster and saves about half the time compared to a relatively confusing color image of the background. Of course, the present invention is not limited to this, and the body detection unit 101 can detect the body of the target from the intensity image using another target detection method.

The body detection unit 101 scans the input intensity image to determine the position, size, and ratio of the body of the target. In general, the inactive portion of the body of the target is constrained to a relatively fixed geometric constraint relationship, for example, the head and the body are in an up-and-down position in a linear state.

Based on this, you can search the body of the target in two strategies. One strategy is to scan all the windows of an input intensity image to detect one body, and then scan the adjacent image region of the body according to the geometrical relationship between the components forming the body to detect other body components. This strategy can significantly reduce the scan window. Another strategy is to scan all windows in the input intensity image to detect all possible body components. The strategy does not take into account the known geometric constraint between the body components of the target. After detecting candidates for all body components, if there are a plurality of candidates for each body component, the most desirable combination of body components is selected according to the geometric constraint between the components.

The inactive part of the body of the target detected by the body detection unit 101 is used for calculation of a threshold during the foreground segmentation that follows and determination of the active part area upon detection of the active part.

A foreground part 102 is used to remove a background area (noise area) in the depth image, and divides the background including the foreground of the target. As described above, both the foreground division method based on the conventional color image and the foreground division method based on the depth image are all inaccurate and thus have great difficulty in target analysis. Therefore, in the present invention, it is possible to realize a simple and accurate foreground division using the intensity image and the depth image. The operation of the foreground division unit 102 will be described in detail below.

First, according to the body of the target detected by the body detection unit 101, the foreground division unit 102 calculates one intensity threshold value of the intensity image. First, statistical analysis is performed on the intensity values of all the pixels indicated by the body by the body detection unit 101 in the intensity image to obtain the intensity threshold value.

Then, the foreground division unit 102 converts the intensity image into a binary image by binarizing each pixel value of the intensity image by comparing each pixel value of the input intensity image with the calculated intensity threshold value. For example, a pixel whose intensity value is greater than the intensity threshold value is marked as a target, and a pixel whose intensity value is less than the intensity threshold value is marked as a background. Here, noise is removed from the binary image by performing value filtering, erosion, and dilate operation on the generated binary image.

FIG. 4A shows an example of an intensity image, and FIG. 4B shows an example of a binary image formed based on the intensity threshold.

Then, the input depth image is masked using the binary image, and a masking depth image is obtained. A binary image can divide the depth image into two parts: one part is the foreground area where the target exists, and the other part is the background area. Therefore, the binary image can be used to mask the depth image and remove the background area. For example, an AND operation is performed between the binary image and the depth image to obtain a masking depth image.

FIG. 4C shows an example of a depth image input in FIG. 4C, and FIG. 4D shows an example in which a mask depth image is acquired using a binary image.

After acquiring the mask depth image through the foreground division unit 102, the active portion detection unit 103 detects an active portion (a human limb in the case of a human target) of the target according to the mask depth image. Since the background area of the mask depth image has already been removed as shown in Fig. 4 (d), the confused edge has already been significantly reduced, so that the active part detection part 103 can accurately detect the active part of the target.

Currently, the method of detecting active parts, for example, the method of detecting limbs is roughly divided into two categories: one is a method according to a low-level feature, and the other is a method according to a high-level feature.

The method according to the low-level feature detects the active part of the target using the characteristics of the turning point, the edge line, and the like.

The method according to the high-layer feature is a method using a positive sample and a negative sample learning classifier or detector as a model identification method. The active portion detecting unit 103 of the present invention can employ the above method to detect the active portion of the target from the mask depth image.

The active portion detecting unit 103 will be described below as an example of a method of detecting an edge feature based on the Hough transform below. It is to be understood that the present invention is not limited to this.

The currently illustrated active portion detection unit 103 may include two lower units: a ridge detection unit 501 and a ridge verification unit 502. The ridge detecting unit 501 detects a possible ridge of the target active portion. The ridge verification unit 502 verifies the ridge candidate detected by the ridge detection unit 501 and selects the most preferable ridge. Here, the ridge means the center line of each part of the target.

5 and 6, the ridge detecting unit 501 includes a depth slicing unit 601, an edge detecting unit 602, a rectangular designing unit 603, and a ridge extracting unit 604.

The depth slicing unit 601 determines an area in which the active portion of the target exists among the mask depth images. Here, the region excluding the inactive portion of the mask depth image is determined as an active portion extending from the inactive portion of the target, for example, as shown in Fig. 7A, which the body detecting portion 101 has determined.

There may also be the following situation: the active and inactive portions of the target of the intensity image, such as the left arm of the person shown in Figure 3, may overlap. For this situation, the depth slicing unit 601 calculates the average depth value of the inactive portion and determines the value of the depth slicing unit 601 to determine whether the pixel of the inactive portion actually belongs to the inactive portion, As shown in FIG. This is because when the active part and the inactive part of the target are overlapped, the active part and the inactive part are located in different depth planes. If the absolute value of the difference between the pixel and the threshold of depth slicing is greater than the predetermined constant, the pixel belongs to the active portion of the target, otherwise the pixel belongs to the inactive portion.

7B shows the limb area of the person determined by the depth slicing unit 601. [

The edge detection unit 602 advances the Hope transform for the depth image to obtain a hops line representing the edge feature of the image. In the active portion determined by the depth slicing unit 601, the hop line indicates the edge of the target active portion.

Fig. 7 (c) shows an example of the detected hop line.

As shown in Fig. 7 (d), the rectangular design unit 603 forms a rectangle using the hop line and covers the active portion of the target. The rectangular design process is roughly as follows: first, align the hop lines according to the length of the detected hop lines; A rectangle is formed by scanning a depth value transition point of a depth image along a long side of a relatively long hop line and a direction perpendicular to a long side to confirm the short side of the rectangle; If the rectangles overlap each other, remove the extra rectangles according to the principle of eliminating large rectangles and overlapping small rectangles.

As shown in Fig. 7 (e), the ridge extracting unit 604 extracts the central line by connecting the two centers of the short sides in the rectangle formed from the rectangular design unit 603, and makes the ridge possible.

The ridge detecting unit 501 detects all possible active partial ridges from the mask depth image. With the ridge as a candidate, the ridge verifying unit 502 proceeds the verification to determine the final skeleton structure of the active portion.

The ridge verifying unit 502 selects the most desirable ridge by verifying the ridge candidate detected by the ridge detecting unit 501 through the pre-verified model, and finally selects the ridge of the active portion of the original target. Of all the ridge candidates detected by the ridge detecting unit 501, the rearranger verifying unit 502 must verify whether or not to match the actual activity part after combining certain ridge candidates. To this end, the bridge verification unit 502 may employ a hypothesis generation evaluation method. That is, a plurality of ridges among all the ridge candidates are selected and connected according to a specific method, and then the probability of the hypothesis is calculated by using the constraint and calculation method of the pre-verified model of the target. The final ridge structure of the target is detected by selecting the most probable hypothesis as the best path.

The ridge verifier 502 of the present invention can carry out the ridge verification with a currently validated verification model. For example, the pictorial structure method of Felzenszwalb (PF Felzenszwalb. DP Huttenlocher, "Efficient Matching of Pictorial Structures" CVPR 2000). The operation of the ridge verification unit 502 will be briefly described below as an example of the above-mentioned image forming method. Note that the present invention is not limited to this.

First, as shown in Fig. 8 (a), the ridge verification unit 502 establishes a tree structure of a target. The tree structure represents the search space of the target, and displays the position and the connection relationship of each part of the target, and the variable of each node may include a position, a partial ratio, a rotation angle, and the like of the 3D space.

Here, the body detecting unit 101 distributes the connection position, the ratio, and the angle of the inactive part of the detected target to the search tree, and sets this as a constant. However, the ridge verification unit 502 generates a hypothesis by distributing the ridge candidate to the nodes. The ridge verification unit 502 calculates an observation similarity between the hypothesis and the actual image through the following equation (1): < EMI ID =

 Here, "MatchedPt" is the ratio of the size of the non-zero pixel covered by the hypothetical portion to the portion of the depth image. "NoMatchedPt" is the ratio of the size of the portion to the zero pixel covered by the hypothetical portion of the depth image. "nS" indicates the proportional factor of the hypothetical rectangle. "w" is the weight coefficient.

The model matching between the hypothesis and the geometric model is calculated according to the following equation:

Where alpha represents the angle between adjacent parts, that is, the angle between the parent node and the child node of the tree structure; alpha _ij is the ideal angle between node i and node j; log nS _j ^k means the ratio factor of the j part of the k dimension; log nS _ij ^k means the anomaly rate factor of the model; p _ij is the coordinate difference of the connecting line between the i and j parts.

However, the post-hypothesis verification probability is approximated through Equation 3:

The ridge verification unit 502 detects the ridge structure of the final target by searching for the best path having the maximum probability. Fig. 8 (b) shows an example of the ridge verification result.

As described above, in the present invention, the bridge verification unit 502 distributes the connection position, the ratio, and the angle of the body component of the target detected by the body detection unit 1010 to the search tree, thereby significantly reducing the search space, .

The number of candidate ridges can be large and there are many corresponding connection methods, so the search space is very large and the computation amount can be very large. The body detection unit 101 has already detected the body of the target, in particular, the inactive part, for example, the human head and body, and confirmed the position and the ratio of the inactive part.

For a particular target type, the ratio between each part meets certain conditions and the change is relatively small. For example, statistical methods can be used to deduce the percentage of a person's arm or head from a person's body. Therefore, the search space can be significantly reduced and the amount of calculation can be reduced without involving in every possible calculation of each possible combination, with the previously detected inactive part being already known constants in the search space.

There may also be a case where the portion of the target is shielded or missing, such as the right arm of the person of Fig. In this situation, it is possible to introduce the concept of hypothesis in the ridge verifier of the present invention to recover missing or rejected parts, and to improve verification speed and accuracy.

First, the body detecting unit 101 can determine the connection point between the active part and the inactive part of the target, for example, the position of the shoulder point of the human body. In general, as shown in Fig. 3, the connection point is a jump point of the curve on the vertical projection and the horizontal projection of the main body of the target.

After confirming the connection point, one virtual ridge is formed from the connection point, and one hypothesis is formed to carry out the evaluation verification. For example, it connects with the end point of the connection point and the kth ridge candidate, and forms a virtual ridge candidate. The virtual ridge candidate and the kth ridge candidate form a hypothesis by forming one shoulder of the target.

Here, the second equation (2) can be simplified to form the following equation (4).

Hereinafter, the target analysis method of the present invention will be described with reference to FIG.

In step 901, the body of the target is detected from the intensity image of the target. In order to detect the body of the target, it is possible to mainly detect inactive portions, for example, the head and body components of a person. The intensity image of the target can be acquired with a TOF camera. Further, as described above, the body of the target among the intensity images is detected by scanning the components of the inactive portion using the partial detector according to the statistical learning method training and applying the constraint.

In step 902, the intensity threshold value of the intensity image is calculated according to the body of the target detected in step S901. Specifically, the intensity threshold is obtained by statistically averaging the intensity values of all the pixels displayed in the body in step S901 through the intensity image.

Subsequently, the intensity image is converted into a binary image using the intensity threshold value calculated in step 903. [ The mask depth image is acquired by masking the depth image of the target using the binary image formed in step 903 in step 904. [

In step 905, the active part of the target, for example, the limb of the person, is detected among the masking depth images. Through the above-described method, the active portion of the target is detected among the mask depth images.

The target analyzing apparatus and method of the present invention have been described above. The target analyzing apparatus and method can detect each portion of the target by analyzing the target according to the intensity image and the depth image of the target, and obtain the operation information of the target by extracting the ridge. The target analyzing apparatus and method of the present invention can achieve the following effects.

First, according to the present invention, accurate foreground segmentation can be realized by performing foreground segmentation using intensity and depth video streams that are waiting for analysis in place of a complicated color image.

Next, each target component can be detected by rapidly extracting the target edge according to the good foreground division result.

In addition, in the present invention, a statistical model detector of a sample can be applied to an intensity image, and a body of a target among intensity images can be successfully detected. Also, the inactive part of the detected body can be applied to the subsequent ridge verification system, thereby reducing the search space of the verification system and significantly reducing the amount of computation.

In addition, in the ridge verification system, it is possible to improve the verification speed and accuracy by restoring missing or shielded parts using the virtual part.

Also, the target analysis method according to an embodiment of the present invention may be implemented in a form of a program command that can be executed through various computer means and recorded in a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy 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 high-level language code that can be executed by a computer using an interpreter or the like. 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.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those of ordinary skill 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. This is possible. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

100 target analyzer
101 main body detection unit
102 Foreground split unit
103 Activity part detection unit

Claims (19)

A target analyzing apparatus for inputting a depth image and an intensity image of a target,
A body detecting unit for detecting a body of the target among the intensity images of the target;
Calculating an intensity threshold according to the detected intensity value of the body, switching the intensity image to a binary image using the intensity threshold, and masking the depth image of the target using the binary image as a mask, A foreground division for acquiring an image; And
An active portion detecting portion for detecting, as an active portion, a portion of the mask depth image having a relatively large motion relative to the body of the target;
Lt; / RTI >
Wherein the inactive portion of the body of the target comprises a human head and body, and wherein the active portion comprises a human limb. The method according to claim 1,
And a TOF camera set for capturing and outputting a depth image and intensity image of the target. The method according to claim 1,
Wherein the body detecting unit detects a component constituting a body of the target among the intensity images trained by a statistical learning method and selects a candidate that matches a geometric constraint among a plurality of candidates for a component of the body as a component of the body Analysis device. The method according to claim 1,
Wherein the foreground parting section uses a value obtained by statistically averaging the intensity values of the detected bodies as an intensity threshold value. The method according to claim 1,
The active portion detection unit
A ridge detection unit for detecting a plurality of ridges among the mask depth images; And
A ridge verifying unit for verifying the plurality of ridges and selecting an optimum ridge;
The target analyzer. 6. The method of claim 5,
Wherein the ridge detection section distinguishes the inactive portion from the inactive portion of the body of the depth image of the mask and calculates a candidate for the plurality of ridges from the candidate of the active portion. The method according to claim 6,
Wherein the ridge detection unit distinguishes the active portion from the inactive portion with the average depth value of the inactive portion as a threshold value. The method according to claim 6,
Wherein the bridge verification unit determines a connection point between the active part and the inactive part according to the body detected by the body detection unit, forms a virtual ridge by connecting one end of the predetermined connection point and one of the plurality of ridges, And verifies the plurality of ridges and the virtual ridge. delete The target analysis method based on the depth image and the intensity image of the target,
comprising the steps of: a) detecting a target body of a target intensity image;
b) calculating a strength threshold according to an intensity value of the body of the detected target;
c) converting the intensity image to a binary image using the intensity threshold;
d) masking the depth image using the binary image and obtaining a mask depth image; And
e) detecting, as an active portion, a portion of the mask depth image that is relatively moving relative to the body of the target;
Lt; / RTI >
Wherein the inactive portion of the body of the target comprises a human head and body, and wherein the active portion comprises a human limb. 11. The method of claim 10,
Wherein the method further comprises capturing a depth image and intensity image of the target using a TOF camera set. 11. The method of claim 10,
Using a partial detector trained by the statistical learning method in the step a), detecting a candidate as a component of the body corresponding to a geometrical constraint among a plurality of candidates for the component of the body . 11. The method of claim 10,
Wherein the statistical average of the intensity values in the body of the target detected in the step b) is advanced and the statistical average value of the obtained intensity values is set as the intensity threshold value. 11. The method of claim 10,
The step e)
E1) detecting a plurality of ridges in the masked depth image; And
E2) checking the plurality of ridges to select an optimal ridge;
Lt; / RTI > 15. The method of claim 14,
Wherein said step (e1) comprises: dividing the active part and the inactive part of the body among the masked depth images obtained in step (d) and calculating a candidate for the plurality of ridges from the candidate of the active part . 16. The method of claim 15,
Wherein the active portion and the inactive portion are distinguished from each other by setting an average depth value of the inactive portion as a threshold value in the step e1). 17. The method of claim 16,
Wherein step e2) comprises:
e21) establishing a connection point between the active portion and the inactive portion of the body;
e22) connecting the determined connection point and one end of the plurality of ridges to form a virtual ridge; And
e23) verifying the plurality of ridges and the virtual ridge;
Lt; / RTI > delete 17. A computer-readable recording medium recording a program for performing the method of any one of claims 10 to 17.

Images