KR20110111662A - Background modeling method for detecting target object silhouette - Google Patents

Abstract

A background modeling method capable of modeling a background image from an image is provided. In the background modeling method, detecting the motion generating area of the object and the whole body area of the object from the obtained image and outputting the object area, generating and outputting a background image excluding the object area from the image and the difference between the image and the background image Detecting and outputting an object silhouette based on the;

Description

Background modeling method for detecting target object silhouette}

The present invention relates to a background modeling technique in an indoor robot environment. More specifically, the present invention can quickly adapt to environmental changes such as robot movement and lighting conversion in a room, and stably model a background regardless of a person's movement and stoppage. A background modeling method that can be done.

The present invention is derived from the research conducted as part of the IT source technology development project of the Ministry of Knowledge Economy [Task Management Number: 2008-F-037-01, Task name: u-robot HRI solution and core device technology development].

Techniques for extracting a target object from an image have been studied for a long time in many fields, and the types of techniques are also very diverse. Among these technologies, the background subtraction through the difference between the background image without an object and the current image with an object can quickly and stably separate the silhouette (shape) of the object. It is a widely used technique.

In the robot field, a background separation technique that detects a person's silhouette is used as a base technology for locating a person to interact with or recognizing a user's gestures and actions.

In order to use the background separation technique in the robot field, a background modeling technique is required to generate a background image without an object, and to continuously update and maintain it.

However, the conventional background modeling technique approaches the basic assumption that the pixel value corresponding to the background region appears longer than the pixel value corresponding to the object region at a specific pixel position in the image. The problem arises due to the difference in the speed at which it moves.

The problem to be solved by the present invention is to provide a background modeling method that can quickly cope with changes in the indoor environment, and can model the background stably.

In the background modeling method according to an embodiment of the present invention for solving the above problems, the step of detecting the motion generating region and the whole body region of the object in the obtained image outputting the object region, the background excluding the object region from the image Generating and outputting an image; and detecting and outputting an object silhouette based on a difference between the image and the background image.

The background modeling method of the present invention has an effect of rapidly modeling a background from an image regardless of the movement or stationary of an object while quickly coping with an environmental change caused by a rapid environmental change, for example, a robot movement or an object movement in an indoor robot environment. have.

BRIEF DESCRIPTION OF THE DRAWINGS In order to better understand the drawings cited in the detailed description of the invention, a brief description of each drawing is provided.
1 is a schematic diagram of a background modeling apparatus according to an embodiment of the present invention.
FIG. 2 is a schematic configuration diagram of the object detector illustrated in FIG. 1.
3 is a flowchart of a background modeling operation of the background modeling apparatus illustrated in FIG. 1.
4 to 10 are diagrams illustrating an embodiment according to a background modeling operation.

In order to fully understand the present invention, the operational advantages of the present invention, and the objects achieved by the embodiments of the present invention, reference should be made to the accompanying drawings that illustrate embodiments of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Like reference numerals in the drawings denote like elements.

FIG. 1 is a schematic diagram of a background modeling apparatus according to an embodiment of the present invention, and FIG. 2 is a schematic diagram of an object detector shown in FIG. 1.

Referring to FIG. 1, the background modeling apparatus 1 according to the present invention may include an image input unit 10, a background modeling unit 20, and a silhouette (form) detector 30.

The image input unit 10 may obtain an image IM photographed from a camera provided in a robot (not shown) moving indoors and output the image IM to the background modeling unit 20.

The background modeling unit 20 may model and output a background from an image IM provided from the image input unit 10.

1 and 2, the background modeling unit 20 may include an object detector 110 and a background generator 150.

The object detector 110 may estimate and output an area in which an object, for example, a person exists, that is, an object area TA, from the image IM provided from the image input unit 10.

The object detector 110 may include a motion detector 120, a whole body region detector 130, and a mixer 140.

The motion detector 120 may detect a motion of a person from the image IM provided from the image input unit 10 and estimate and output the first object area Ta.

Here, the motion detector 120 may detect the motion of the person from a plurality of images in the same space provided from the image input unit 10 to stably estimate the object region even when there is little motion between the adjacent images. .

The motion detector 120 may include a motion map generator 121 and a motion region setting unit 123.

The motion map generator 121 may generate a motion map by displaying a region where a human motion is generated in units of pixels from one or more images IM provided from the image input unit 10, and output the motion map.

The motion region setting unit 123 detects a pixel unit motion as a block unit region based on the motion map output from the motion map generator 121, and sets and outputs a motion region from the detected block unit region. Can be.

The motion detector 120 may estimate and output the first object area Ta from the set motion area.

The whole body area detector 130 may detect the whole body area of the person from the image IM provided from the image input unit 10, and estimate and output the second object area Tb.

The whole body area detection unit 130 may include a face-based whole body area setting unit 131 and a shape-based whole body area setting unit 133.

The face-based whole body area setting unit 131 may detect a face area of the person from the image IM, and set and output the whole body area of the person based on the detected face area.

The shape-based whole body area setting unit 133 detects an omega shape of a person from the image IM, for example, a shape area according to a shape of an outline connecting the head and shoulders of the person and based on the detected omega shape area. You can print the whole body area of the person.

The whole body area detection unit 130 may estimate and output the second object area Tb based on the set whole body area of the person.

The mixer 140 mixes the first object area Ta and the second object area Tb output from the motion detector 120 and the full-body area detector 130, respectively, and then the object area TA in the image IM. You can output

The background generator 150 may generate and output a remaining region excluding the object region TA from the image IM as the background image IM based on the object region TA output from the object detector 110. .

Referring back to FIG. 1, the silhouette detector 30 may be a real object, that is, a real silhouette of a person, in an image IM provided from the image input unit 10 based on the background image IM output from the background modeling unit 20. (TG) can be isolated and detected.

3 is a flowchart illustrating a background modeling operation of the background modeling apparatus illustrated in FIG. 1, and FIGS. 4 to 10 are views illustrating an embodiment according to the background modeling operation.

1 to 3, the image input unit 10 of the background modeling apparatus 1 may acquire an image IM photographed from an external device, that is, a camera of a robot (S10).

Here, the image input unit 10 may obtain a plurality of images continuously photographed in the same space.

The motion map generator 121 of the background modeling apparatus 1 may generate a motion map by displaying, in units of pixels, an area in which motion is generated by an object, that is, a person, from one or more images IM provided from the image input unit 10. Can be.

In addition, the motion region setting unit 123 of the background modeling apparatus 1 may set and output a motion region in units of blocks from the motion map output from the motion map generator 121 (S30).

For example, referring to FIGS. 2 and 4, the motion map generator 121 may perform a motion as shown in FIG. 4F from a plurality of images IM as shown in FIGS. 4A through 4E. You can create and print a map. 4E is an image at the present time.

In detail, the motion map generator 121 calculates the pixel value M (x, y) having the largest difference between the current image (e) and the previous image (a) to (d). It can be calculated using.

Subsequently, when the calculated pixel value M (x, y) is larger than a preset threshold, it may be regarded as motion, and if smaller than the threshold, it may be regarded as a background.

[Equation 1]

Here, I _t (x, y) means a pixel value of the current image, and I _{t - 1} (x, y) means a pixel value of the previous image.

2 and 5, after the motion map generator 121 generates a motion map from the image IM as shown in FIG. 5A, the motion region setting unit 123 generates the generated motion map. In order to separate the motion region from the map, connection element analysis may be performed as shown in FIG.

Here, the motion region setting unit 123 may merge connection elements at close distances from each of the connection elements when performing the connection element analysis from the motion map.

Subsequently, as shown in FIG. 5C, the motion region may be set and output. Here, the motion region setting unit 123 may set the lower boundary of the motion region to the height of the image in order to include the shadow region that can be generated by a person in the set motion region.

Referring back to FIGS. 1 to 3, the motion detector 120 may output the first object area Ta by estimating the position of the person from the set motion area.

Meanwhile, the face-based whole body region setting unit 131 of the background modeling apparatus 1 detects a face region of the person from the image IM provided from the image input unit 10 and based on the detected face region, the whole body region of the person. Can be set and output (S40).

In addition, the shape-based whole body region setting unit 133 of the background modeling apparatus 1 detects an omega shape region of a person from an image IM provided from the image input unit 10, and based on the detected omega shape region, The whole body region may be set and output (S40).

For example, referring to FIGS. 2 and 6, the face-based whole body region setting unit 131 may have a face of a person as shown in FIG. 6B in an image IM in which a face of a person is shown as shown in FIG. 6A. The area can be detected.

In this case, the face-based whole body area setting unit 131 may detect a face area of a person using a face detector (not shown).

Here, the face detector may use the AdaBoost algorithm, which is one of appearance-based pattern recognition methods.

The face-based whole body area setting unit 131 may set the whole body area of the person as shown in FIG. 6B based on the detected position and size of the face area of the person.

Also, referring to FIGS. 2 and 7, when the shape-based whole body area setting unit 133 is an image (IM) in which a face of the human is invisible or is not located at a distance as shown in FIG. In FIG. 7B, an omega-shaped region can be detected from the image IM.

To this end, the shape-based whole body area setting unit 133 may include an omega shape detector (not shown), and the omega shape detector may use an AdaBoost algorithm. Here, the omega shape is a shape defined from the fact that the contour connecting the head and the shoulder of a person resembles an omega (Ω) shape.

The shape-based whole body area setting unit 133 may set the whole body area of the person as shown in FIG. 7B based on the detected position and size of the omega shape area.

Referring back to FIGS. 1 to 3, the whole-body area detector 130 may output the second object area Tb by estimating the position and size of the person from the set whole-body area.

The mixer 140 finally mixes the first object area Ta output from the motion detector 120 and the second object area Tb output from the whole body area detector 130 as shown in FIG. 8B. In the image IM, an area in which a person exists, that is, an object area TA may be output.

The background generator 120 may generate and output the background image IM from which the object area TA is excluded from the image IM based on the object area TA output from the object detector 110 (S50). ).

Meanwhile, the background generator 120 may update and output the background image IM from Equations 2 and 3 below.

[Equation 2] is an equation for updating the background image IM for the pixel determined as the background of the input image, and [Equation 3] is applied to the pixel determined as the object area TA for the input image. A formula for updating the background image IM.

[Equation 2]

 &Quot; (3) "

Here, B _{t + 1} (x, y) is the pixel value of the newly updated background image, B _t (x, y) is the pixel value of the current background image, and I _t (x, y) is the pixel value of the current input image. And α denotes a constant value for adjusting the background update rate.

The silhouette detector 30 may detect and output the silhouette TG of the real object from the image IM based on the background image IM output from the background generator 120 (S60).

2 and 9, the silhouette detector 30 may detect a silhouette TG of a person in a moving state from the image IM.

For example, in the moving image IM as shown in FIG. 9A, the object detector 110 of the background modeling unit 20 may output the object area TA as shown in FIG. 9B. The background generator 120 may generate a background image IM as shown in FIG. 9C.

At this time, the silhouette detector 30 may detect and output the actual silhouette TG of the person in the image IM from the difference between FIGS. 9A and 9C as shown in FIG. 9D.

2 and 10, the silhouette detector 30 may detect a silhouette TG of a person in a stationary state from the image IM.

For example, in an image IM in which a person is stationary as shown in FIG. 10A, the object detector 110 of the background modeling unit 20 may output the object area TA as shown in FIG. 10B. In addition, the background generator 120 may generate a background image IM as illustrated in FIG. 10C.

In this case, the silhouette detector 30 may detect and output the actual silhouette TG of the person in the image IM from the difference between FIGS. 10A and 10C as shown in FIG. 10D.

Although the contents of the present invention have been described with reference to one embodiment shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible therefrom. will be. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1: background modeling apparatus 10: image input unit
20: background modeling unit 30: silhouette detection unit
110: object detection unit 120: motion detection unit
130: whole body area detection unit 150: background generation unit

Claims (1)

Detecting the motion generating region and the whole body region of the object from the obtained image and outputting the object region;
Generating and outputting a background image excluding the object area from the image; And
And detecting and outputting an object silhouette based on a difference between the image and the background image.

Images