KR101394473B1 - Method for detecting moving object and surveillance system thereof - Google Patents

Abstract

The present invention relates to a moving object detection method for discriminating an image as foreground and background in a surveillance system in which an image is input at regular time intervals, the method comprising the steps of: (a) Generating a transform matrix from the input image and generating an n-th background image using the transformation matrix; (b) obtaining a probability that a pixel of the n-th input image belongs to the n-th background image, (C) setting an area adjacent to the initial foreground map as an additional foreground candidate area and setting an additional foreground map by obtaining a probability that a pixel belonging to the additional foreground candidate area belongs to the foreground A moving object detection method and a monitoring system thereof are provided.

Description

TECHNICAL FIELD [0001] The present invention relates to a moving object detection method,

BACKGROUND OF THE INVENTION 1. Field of the Invention [0002] The present invention relates to a moving object detection method and a monitoring system thereof, and more particularly, to a moving object detection method and a monitoring system thereof capable of detecting moving objects in an image photographed by a pan / tilt camera.

The present invention is derived from a research carried out as a part of the social safety robot development project using the cluster intelligence of the Ministry of Knowledge Economy [assignment number: 0148-2007011, title: Robust adaptive video surveillance algorithm].

Conventional moving object detection techniques have been applied to most fixed cameras and have been widely used. When the camera is fixed, the moving object detection algorithm is mainly used to model the fixed background and remove the background area from the input image using the characteristics of the camera. Typically, the most commonly used method for background modeling is the Gaussian Mixture Model, which models the probability of a background with multiple Gaussian distributions per pixel. Fixed cameras use a pixel-based background modeling method with fast object detection speed and relatively accurate pixel-unit background.

In addition, a pixel - by - pixel background modeling method is adopted to detect objects in input images. However, the pixel-based background modeling method assumes that the position of the background illuminated by the camera does not change. Therefore, there is a problem that the edge area of the background is detected as a motion in a practical application surveillance environment where the camera itself is slightly shaken by the wind.

In particular, in the environment where the camera is not fixed, such pixel-based background modeling may cause the following three problems due to errors that may occur in the image conversion process. First, it is a problem caused by mismatch between background image and current image as background edge detection problem. The inside of a background object having the same color does not have a great influence, but a corner portion that becomes a boundary surface of a background object can change the model itself. In fact, it is necessary to assume that the image is a two-dimensional image. In fact, since an input image through a camera is a two-dimensional image of a three-dimensional model, errors due to parallax may occur. Such a problem causes a problem that the edge area of the background is detected as a moving object.

Second, it is a matter of noise generation in dynamic background. This is a dynamic background, even though it is a real background, such as shaking branches or rotating fans in the background. In the Gaussian mixture model, a number of background models are generated to solve this problem. However, since the number of models required for each pixel is not known, many cases are not satisfied. Therefore, in an environment with many dynamic elements such as fractions, a dynamic background may be detected as a moving object.

Third, similar color object detection error problem. When a background model is created in black and white, the average of the color values on the screen is obtained, and the color intensity is represented by the intensity. Therefore, even different colors can be displayed with the same intensity. Therefore, even if an object of a different color than the actual background appears, it may not be detected. Even if the background model is a color model, when an object of a similar color appears, the object may not be detected.

The main object of the present invention is to provide a moving object detection method and a monitoring system thereof capable of detecting moving objects in an input image of a surveillance system such as a pan / tilt camera.

A moving object detection method according to an embodiment of the present invention is a moving object detection method for discriminating the image as foreground and background in a surveillance system in which an image is input at regular time intervals, the method comprising: (a) Generating an n-th background image using the transform matrix, and (b) generating a transform matrix from the n-th input image and the n-1-th input image, (C) setting a region adjacent to the initial foreground map as an additional foreground candidate region, and determining a probability that a pixel belonging to the additional foreground candidate region belongs to the foreground region And setting an additional foreground map.

In the present invention, the step (a) includes the steps of: (a1) calculating a difference between the n-th input image and the (n-1) -th input image in the n- (A2) converting an (n-1) -th background image to fit the coordinate system of the n-th input image by the transformation matrix, and (a3) And generating the n-th background image by restoring the undetermined region in the n-th input image to the n-th input image.

In the present invention, the step (a) extracts a local feature point from the n-th input image and the (n-1) -th input image, To generate the transformation matrix.

(B1) calculating a probability that each pixel of the n-th input image belongs to the n-th background image; (b2) calculating a probability that each pixel of the n-th input image belongs to the n-th background image; And setting the pixels having low probability of belonging to the pixels of the n-th background image as the initial foreground map.

In the present invention, the step (b1) includes the steps of: (b11) setting a part of the n-th input image as a first comparison area; and (b12) (b13) setting a second comparison area corresponding to the first comparison area among the n-th background image, (b14) setting a second comparison area corresponding to the first comparison area, Generating a second vector having r, g, b, x, and y as components of pixels belonging to the second comparison region; and (b15) generating a background vector space composed of the second vectors (B16) Kernel density estimation is performed on the background vector space of the first vector to determine a probability that a pixel belonging to the first comparison area belongs to the n-th background image . (Where r, g, and b are the color values of the pixel, and x and y are the coordinates of the pixel).

In the present invention, the step (c) further includes the steps of: (c1) setting an area adjacent to the initial foreground map as the additional foreground candidate area; (c2) (C3) setting a pixel having a high probability of belonging to the foreground among the pixels of the additional foreground candidate region as an additional foreground map.

In the present invention, the step (c2) includes the steps of: (c21) generating a foreground vector having r, g, b, x, and y as components of pixels belonging to the additional foreground candidate area; Generating a foreground vector space including the foreground vectors; and (c23) performing a kernel density estimation on the foreground vector space to obtain a probability that the pixel belongs to the foreground.

In the present invention, the step (c3) compares the probability that the pixels of the additional foreground candidate area belong to the foreground and the probability of belonging to the background, and sets pixels having a high probability of belonging to the foreground as an additional foreground map The method comprising the steps of:

The surveillance system according to an embodiment of the present invention is a surveillance system for distinguishing foreground and background from images input at regular time intervals. The surveillance system includes an image input unit for receiving an input image, A background image generation unit that generates a transformation matrix from the input image and the (n-1) th input image and generates an n-th background image using the transformation matrix; An initial foreground map generator for setting an initial foreground map and a region adjacent to the initial foreground map as an additional foreground candidate region and determining a probability that a pixel belonging to the additional foreground candidate region belongs to the foreground, And an additional foreground map generator for setting a foreground map.

In the present invention, the background image generation unit may generate the transformation matrix between the n-th input image and the (n-1) -th input image in the n-th input image and the (n-1) -th input image input by the image input unit Th background image so as to fit the coordinate system of the n-th input image by the transformation matrix, restoring the unspecified area in the transformed n-1-th background image into the n-th input image, an n-th background image can be generated.

In the present invention, the background image generation unit may extract a local feature point from the n-th input image and the (n-1) -th input image, and output a local feature point corresponding to the n-th input image and the To generate the transformation matrix.

In the present invention, the initial foreground map generator may calculate a probability that each pixel of the n-th input image belongs to the n-th background image, Can be set as the initial foreground map.

In the present invention, the additional foreground map generator sets an area adjacent to the initial foreground map as the additional foreground candidate area, calculates a probability that each of the pixels belonging to the additional foreground candidate area belongs to the foreground, Pixels having a high probability of belonging to the foreground among the pixels in the additional foreground candidate region can be set as an additional foreground map.

In the present invention, the additional foreground map generator compares the probability that the pixels of the additional foreground candidate area belong to the foreground and the probability that the pixels of the additional foreground candidate area belong to the background, Can be set.

The moving object detection method and the monitoring system of the present invention as described above reduce noise in a dynamic background and can prevent false detection of an object of a similar color.

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

FIG. 1 shows a flowchart of a moving object detection method according to an embodiment of the present invention. The moving object detection method can be implemented by the monitoring system 200 of FIG.

Referring to FIG. 1, a moving object detection method according to an embodiment of the present invention includes an input image input step S110, a background image generation step S120, an initial foreground map generation step S130, S140).

The input image input step S110 receives the input image for the surveillance region to be monitored. The input image may be a frame image input according to the set time interval.

The background image generation step S120 may generate a transformation matrix from the n-th input image and the (n-1) -th input image input by the monitoring system, and generate an n-th background image using the transformation matrix. The background image generation step S120 may include a transformation matrix generation step S121, a background image transformation step S122 using a transformation matrix, and a background image restoration step S123.

In the transformation matrix generation step S121, the transformation matrix between the n-th input image and the (n-1) -th input image is generated from the n-th input image and the (n-1) -th input image input by the monitoring system. In more detail, in the transform matrix generation step S121, a local feature point is extracted from the n-th input image and the (n-1) -th input image, and the local feature point is extracted from the n-th input image and the And the transformation matrix is generated by obtaining corresponding points.

The background image transformation step S122 of the transformation matrix transforms the (n-1) th background image to fit the coordinate system of the n-th input image by the transformation matrix.

As in the embodiment of the present invention, the surveillance system equipped with the pan / tilt camera captures an image while moving the pan / tilt camera, so that the coordinate system of the n-th input image and the n-1th input image are different from each other, The n-1th background image and the n-th input image generated based on the input image are not mutually comparable because their coordinate systems are different from each other. Therefore, in the background image conversion step S122 by the transformation matrix, the (n-1) th background image is transformed according to the coordinate system of the n-th input image by the transformation matrix.

The background image reconstruction step S123 reconstructs the n-th input image of the transformed n-1th background image to generate the n-th background image.

5 is a photograph showing the process of generating a background image step by step. FIG. 5A shows an n-1th background image, and FIG. 5B shows an nth input image. Since the surveillance system of the present invention captures an image while moving from right to left, there is a difference between the n-th input image and the (n-1) -th background image as shown in FIGS. 5A and 5B. FIG. 5C shows the (n-1) th background image converted by the transformation matrix. The change matrix transforms the image to fit the coordinate system of the n-th input image, so that the unregulated region 110 occurs as shown in FIG. 5 (c). The undefined area 110 is generated as the monitoring system of the present invention moves. The background image reconstruction step S123 reconstructs the unidentified area 10 into the n-th input image. 5D is a background image in which the undefined area 110 is restored. In the background image restoring step S123, the converted n-1th background image of FIG. 5C and the nth input image of FIG. 5B are compared, and the nth input image is restored to the unknown region 110 (N-1) th background image to restore the background image as shown in FIG. 5 (d). The restored background image of FIG. 5 (d) becomes the n-th background image which is compared with the n-th input image.

The initial foreground map generation step S130 sets the initial foreground map by calculating the probability that the pixel of the n-th input image belongs to the n-th background image. The initial foreground map generation step S130 includes a first comparison area setting step S131, a first vector generating step S132, a second comparison area setting step S133, a second vector generating step S134, A generation step S135, a probability calculation step S136 to belong to the background image, and an initial foreground map generation step S137.

The first comparison area setting step S131 sets a part of the n-th input image as a first comparison area. 6 shows the first comparison area 103 set in the n-th input image 101. In FIG. Referring to FIG. 6, the first comparison area 103 includes an arbitrary pixel 102 of the n-th input image 101 and is a region including pixels around the arbitrary pixel 102.

)를 생성한다. 상기 제1 벡터(

)는 다음 수학식 1과 같이 정의된다. The first vector generating step (S132) may include generating a first vector having r, g, b, x, and y as components of the pixels belonging to the first comparison area

). The first vector (

) Is defined by the following equation (1).

Here, r, g, and b are the color values of the respective pixels, and x and y are the positions of the pixels.

The second comparison area setting step S133 sets a second comparison area corresponding to the first comparison area among the n-th background images. That is, the second comparison area setting step (S133) searches the n-th background image for the area having the same position as the first comparison area to set the comparison area. Since the n-th background image and the n-th input image have the same coordinate system, the second comparison region corresponding to the first comparison region can be set in the n-th background image.

)를 생성한다. 상기 제2 벡터(

)는 다음 수학식 2과 같이 정의된다 The second vector generation step S134 includes a second vector having r, g, b, x, and y as components of the pixels belonging to the second comparison area

). The second vector (

) Is defined as the following equation (2)

)로 이루어진 배경 벡터 공간(

)을 생성한다. 배경 벡터 공간(

)은 다음 수학식 2와 같이 정의된다 The background vector space generation step S135 may include generating second vectors (e.g.,

) ≪ / RTI >

). Background Vector Space (

) Is defined by the following equation (2)

)를 배경 벡터 공간(

)에 대하여 커널 밀도 추정(Kernel density estimation)을 수행하여 상기 제1 비교 영역에 속하는 픽셀들이 상기 n번째 배경 영상에 속하게 될 확률을 구한다. 상기 픽셀이 상기 n번째 배경 영상에 속하게 될 확률은 다음 수학식 4에 의해 결정된다. The probability calculation step (S136) of belonging to the background image is performed using the first vector

) As background vector space (

) To determine the probability that pixels belonging to the first comparison area belong to the n-th background image. The probability that the pixel belongs to the n-th background image is determined by the following equation (4).

은 배경 벡터 공간이며,

는 시간 t에 확률을 계산할 i번째 픽셀의 배경 벡터이며, N은 상기 비교 영역에 속하는 픽셀들의 개수이다.

는 배경 벡터 공간이 주어진 경우 픽셀이 배경에 속하게 될 확률을 의미한다. here,

Is a background vector space,

Is the background vector of the ith pixel to calculate the probability at time t, and N is the number of pixels belonging to the comparison area.

Means the probability that a pixel will belong to the background if the background vector space is given.

In the initial foreground map generation step S137, pixels having low probability of belonging to the pixel of the n-th background image among the pixels belonging to the n-th input image are set as the initial foreground map. In the initial foreground map generation step S137, when the probability obtained in the probability calculation step S136 to be included in the background image becomes less than or equal to the threshold value, the pixel determines the foreground, not the background, As an initial foreground map.

)은 다음 수학식 5에 의해 결정된다. Initial foreground map (

) Is determined by the following equation (5).

)이 임계치(τ_B) 이하인 경우에는 상기 픽셀은 초기 전경 맵에 해당하며, 그렇지 않은 경우에는 상기 픽셀은 초기 전경 맵에 해당하지 않다. Here, τ _B is a threshold at which the pixel belongs to the background image. The probability that the pixel will belong to the background (

Is less than or equal to the threshold value? _B , the pixel corresponds to the initial foreground map, otherwise the pixel does not correspond to the initial foreground map.

After generating the initial foreground map, an additional foreground map is generated (S140). The additional foreground map generation step S140 sets an area adjacent to the initial foreground map as an additional foreground candidate area and sets an additional foreground map by obtaining a probability that a pixel belonging to the additional foreground candidate area belongs to the foreground. The additional foreground map generation step S140 includes an additional foreground candidate area setting step S141, a foreground vector generating step S142, a foreground vector space generating step S143, a probability calculation step S144 belonging to foreground, Step S145 may be included.

The additional foreground candidate area setting step S141 sets an area adjacent to the initial foreground map as the additional foreground candidate area in the n-th input image. 7 shows an input image in which an initial foreground map and an additional foreground candidate area are set. Referring to FIG. 7, the initial foreground maps 105a and 105b can be set in steps S110 to S130 of the input image 101. FIG. On the other hand, in the additional foreground candidate area setting step S141, the additional foreground candidate areas 106, 107, and 108 are set for the area adjacent to the initial foreground maps 105a and 105b.

)를 생성한다. 상기 전경 벡터(

)는 다음 수학식 6과 같이 정의된다. The foreground vector generating step S142 is a step of generating foreground vectors (r, g, b, x, y) for the pixels belonging to the additional foreground candidate region

). The foreground vector (

) Is defined by the following equation (6).

Here, r, g, and b are the color values of the respective pixels, and x and y are the positions of the pixels.

)로 구성된 전경 벡터 공간(

)을 생성한다. 전경 벡터 공간(

)은 다음 수학식 7에 의해 결정된다. The foreground vector space generation step S143 may include generating foreground vectors (e.g.,

) In the foreground vector space (

). Foreground vector space (

) Is determined by the following equation (7).

)에 대하여 커널 밀도 추정을 수행하여 상기 추가 전경 후보 영역에 속하는 픽셀들이 전경에 속하게 될 확률을 구한다. 상기 픽셀들이 전경에 속하게 될 확률은 다음 수학식 8에 의해 결정된다. The probability calculation step (S144) of belonging to the foreground is a step of calculating the foreground vector of the pixel belonging to the additional foreground candidate area as the foreground vector space

) To obtain the probability that the pixels belonging to the additional foreground candidate region will belong to the foreground. The probability that the pixels belong to the foreground is determined by the following equation (8).

는 시간 t에 확률을 계산할 i번째 픽셀의 전경 벡터이며,

는 전경 벡터 공간이 주어진 경우 픽셀이 전경에 속하게 될 확률을 의미한다. here,

Is the foreground vector of the ith pixel to calculate the probability at time t,

Represents the probability that a pixel will belong to the foreground if the foreground vector space is given.

)은 다음 수학식 9에 의해 결정된다. The additional foreground map generation step S145 sets pixels having a high probability of belonging to the foreground among the pixels of the additional foreground candidate region as an additional foreground map. More specifically, the additional foreground map generation step S145 compares the probability that the pixels of the additional foreground candidate area belong to the foreground and the probability that the pixels of the additional foreground candidate area belong to the background, Setting. Additional Foreground Maps (

) Is determined by the following equation (9).

) 보다 전경에 속하는 확 률(

)에 가중치(ω_F)를 곱한 값(

)보다 작은 경우 전경을 판별하며 전경으로 판별된 픽셀들을 추가 전경 맵으로 생성한다. That is, in the additional foreground map generation step S145, the probability that a pixel belonging to the additional foreground candidate area belongs to the background

) Probability that belongs to the foreground (

) By the weight (? _F ) (

), The foreground is discriminated and pixels discriminated as foreground are generated as an additional foreground map.

As described above, an initial foreground map is generated for a moving object and an additional foreground map is generated for an area adjacent to the initial foreground map, thereby eliminating errors of background corner detection due to inconsistency between the background image and the current image. Can be resolved by detecting a moving background such as a moving background.

In addition, the background image is transformed by the transform matrix, and then the background image is restored as the input image, thereby minimizing errors in detection errors due to errors of the transformation matrix and image noise.

8 is a photograph showing an object detection of a surveillance system according to an embodiment of the present invention. 8 (a), 8 (b), and 8 (c), the right image is the input image, the intermediate image is the background image, and the left image is the detection result image. 8A, 8B, and 8C are the n-th image, the n + 1-th image, and the n + 2-th image, respectively. That is, in FIG. 8A, the right image is the n-th input image, the intermediate image is the n-th background image, and the left image is an image showing the initial foreground map and the additional foreground map detected in the n-th input image. In FIG. 8B, the right image is the (n + 1) th input image, the intermediate image is the (n + 1) th background image, and the left image is the initial foreground map and the additional foreground map detected in the It is a video. 8 (c), the right image is the (n + 2) th input image, the intermediate image is the (n + 2) th background image, and the left image is the initial foreground map and the additional foreground map detected in the . 8 (a), 8 (b), and 8 (c) are images taken while moving the surveillance system according to the embodiment of the present invention from left to right.

Referring to FIG. 8 (a), unlike the n-th background image, the object 201 is captured in the n-th input image. In this case, in the object detection method according to the embodiment of the present invention, as described above, in the initial foreground map generation step (S130), the probability that the pixels belonging to the nth input image belongs to the nth background image is obtained, Pixels with probability lower than the threshold value to the initial foreground map 202. Thereafter, in the additional foreground map generation step S140, an area adjacent to the initial foreground map 202 is set as an additional foreground candidate area, the probability that the pixels belonging to the additional foreground candidate area belongs to the foreground, Pixels with a larger probability of belonging to the foreground are set to the additional foreground map 203. [ 8 (a), it can be seen that the initial foreground map 202 and the additional foreground map 203 are displayed with respect to the moving object 201. FIG.

Referring to FIG. 8B, the (n + 1) -th background image generates a transformation matrix from the n-th input image and the n-th input image, transforms the transformation matrix to the n-th background image, Is transformed by the (n + 1) -th input image to generate the (n + 1) -th background image. After the (m + 1) th background image is generated, the initial foreground map 202 and the foreground map 203 are set as described with reference to Fig. 8 (a).

Figure 9 is a schematic block diagram of a surveillance system in accordance with an embodiment of the present invention.

Referring to FIG. 9, a monitoring system according to an embodiment of the present invention can be controlled by the object detection method shown in FIG. Therefore, the same items as those in the object detection method shown in FIG. 1 will be referred to, and a detailed description may be omitted.

The surveillance system 400 according to an exemplary embodiment of the present invention may include an image input unit 401, a background image generation unit 402, an initial foreground map generation unit 403, and an additional foreground map generation unit 404 have.

The video input unit 401 receives an input video for the surveillance region to be monitored. The input image may be a frame image input according to the set time interval.

The background image generation unit 402 generates a transformation matrix from the n-th input image and the (n-1) -th input image input by the image input unit 101 and generates an n-th background image using the transformation matrix . The background image generation unit 402 generates the transformation matrix between the n-th input image and the (n-1) -th input image in the n-th input image and the (n-1) -th input image by the image input unit 401, Th background image so as to fit the coordinate system of the n-th input image by the transformation matrix, restores the unspecified area in the converted n-1th background image into the n-th input image, Images can be generated. In more detail, the background image generating unit 402 extracts local feature points from the n-th input image and the (n-1) -th input image, and outputs a local feature point corresponding to the n-th input image and the The transformation matrix can be generated.

The initial foreground map generation unit 403 calculates the probability that each pixel of the n-th input image belongs to the n-th background image and belongs to the pixel of the n-th background image among the pixels of the n-th input image Pixels with low probability can be set as the initial foreground map.

The additional foreground map generating unit 404 sets an area adjacent to the initial foreground map as the additional foreground candidate area and obtains the probability that each of the pixels belonging to the additional foreground candidate area will belong to the foreground, Pixels having a high probability of belonging to the foreground among the pixels of the area can be set as an additional foreground map. More specifically, the additional foreground map generator 404 compares the probability that the pixels of the additional foreground candidate area belong to the foreground and the probability that the pixels of the additional foreground candidate area belong to the background, .

 It is intended that the present invention not be limited by the foregoing embodiments and the accompanying drawings, but rather by the appended claims, and that various changes, substitutions, and alterations can be made hereto without departing from the spirit of the invention as defined in the appended claims. It will be apparent to those skilled in the art that changes may be made.

FIG. 1 shows a flowchart of a moving object detection method according to an embodiment of the present invention.

FIG. 2 shows a detailed flowchart of the background image generation step of FIG.

FIG. 3 shows a detailed flowchart of the initial foreground map generation step of FIG.

4 shows a detailed flowchart of the additional foreground map generation step of FIG.

5 is a photograph showing the process of generating a background image step by step.

6 shows a first comparison area set in the n-th input image.

7 shows an input image in which an initial foreground map and an additional foreground candidate area are set.

8 is a photograph showing an object detection of a surveillance system according to an embodiment of the present invention.

9 is a block diagram illustrating a surveillance system in accordance with an embodiment of the present invention.

Description of the Related Art

101: input image 102: pixel

103: first comparison area 105a, 105b: initial foreground map

106, 107, 108: additional foreground map 110: unspecified area

Claims (14)

A moving object detection method for discriminating the image as foreground and background in a surveillance system in which an image is input at regular time intervals, (a) generating a transformation matrix from an n-th input image and an (n-1) -th input image input by the monitoring system, and generating an n-th background image using the transformation matrix; (b) setting an initial foreground map by obtaining a probability that a pixel of the n-th input image belongs to the n-th background image; (c) setting an area adjacent to the initial foreground map as an additional foreground candidate area, and determining a probability that a pixel belonging to the additional foreground candidate area belongs to the foreground, and setting an additional foreground map Object detection method. Claim 2 has been abandoned due to the setting registration fee. The method according to claim 1, The step (a) (a1) generating the transformation matrix between the n-th input image and the (n-1) -th input image in the n-th input image and the (n-1) -th input image input by the monitoring system; (a2) transforming the (n-1) th background image to fit the coordinate system of the n-th input image by the transformation matrix; And Claim 3 has been abandoned due to the setting registration fee. The method according to claim 1, (A) extracting local feature points from the n-th input image and the (n-1) -th input image to obtain corresponding points between the n-th input image and the (n-1) -th input image, And a matrix is generated. delete delete delete delete delete A surveillance system for discriminating foreground and background from images input at regular time intervals, A video input unit receiving an input video; A background image generation unit for generating a transformation matrix from an n-th input image and an (n-1) -th input image input by the image input unit and generating an n-th background image using the transformation matrix; An initial foreground map generator configured to determine a probability that a pixel of the n-th input image belongs to the n-th background image to set an initial foreground map; And And an additional foreground map generator configured to set an area adjacent to the initial foreground map as an additional foreground candidate area and obtain a probability that a pixel belonging to the additional foreground candidate area belongs to the foreground to set an additional foreground map. Surveillance system. Claim 10 has been abandoned due to the setting registration fee. 10. The method of claim 9, The background image generation unit generates the transformation matrix between the n-th input image and the (n-1) -th input image in the n-th input image and the (n-1) -th input image input by the image input unit, Th background image so as to correspond to the coordinate system of the n-th input image, and restores the unspecified area in the converted n-1th background image into the n-th input image, The monitoring system comprising: Claim 11 has been abandoned due to the set registration fee. The background image generation unit extracts a local feature point from the n-th input image and the (n-1) -th input image to obtain a corresponding point between the n-th input image and the n-1-th input image, And generates a matrix. delete delete delete

Images