JP2937098B2 - Area detection method and device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for detecting a region from a binary image, a method for detecting a moving object from a grayscale image, and an apparatus for realizing the method.

[0002]

2. Description of the Related Art First, a conventional technique for detecting an area from a binary image will be described. As an example, see JP-A-1-128.
There is a method of detecting a circumscribed frame related to a connected component of a region used by the image processing apparatus described in Japanese Patent Application Publication No. 171/171. The present invention
The purpose of the present invention is to provide an image processing device that efficiently extracts circumscribed frame information of a connected component from a given figure in a short time and efficiently using a small-capacity memory. The feature is that the circumscribed frame of each connected component can be detected even when the structure between them is complicated. Since the present invention does not aim at approximating the region shape by the circumscribed frame, a detailed region shape cannot always be obtained based on the detected circumscribed frame. In particular, for a connected component whose circumscribed rectangle does not overlap at all with other connected components, this circumscribed rectangle may be obtained as a circumscribed frame. I can only get it.

Next, a conventional technique for detecting a moving object from a grayscale image will be described. For example, Japanese Patent Application Laid-Open No. 5-2985
The object recognition device described in Japanese Patent Publication No. 91 aims to extract a moving object without being affected by environmental changes or hiding, and to accurately obtain a trajectory. In the present invention, the following processing is performed. 1. The characteristic part of the moving object is cut out from the input image as a plurality of shaded template images. 2. The characteristic portion is tracked by performing gray-scale pattern matching between the subsequent input image and the template image. 3. The template image is updated to correspond to the change in the shape of the moving object. 4. Integrate templates whose distances of the approximate curves of the trajectories are short. 5. Integrate templates using the knowledge to be recognized. 6. Separate templates based on the dynamics approximation. Therefore, when different moving objects are photographed overlapping and connected as an area in the input image, it is necessary that the dynamics in the input image be clearly different in order to individually detect these moving objects. .

[0004]

First, a problem relating to a technique for detecting a region from a binary image will be described. For example, as in a method of detecting a circumscribed frame related to a connected component of an area used by an image processing apparatus described in Japanese Patent Application Laid-Open No. 1-128171, the conventional technique detects an area by obtaining a circumscribed rectangle or a circumscribed frame of the area. Therefore, there is a problem that only ambiguous shape information of the area can be obtained.

[0005] Next, a problem relating to a technique for detecting a moving object from a grayscale image will be described. For example, Japanese Patent Application Laid-Open No. Hei 5-2
In the object recognition device described in Japanese Patent No. 98591, when different moving objects are photographed repeatedly and form a connected region in an image, they are separated based on a difference in dynamic state.
As described above, in the conventional moving object detection technique, when different moving objects are photographed in an overlapping manner and form a connected region in an image, the moving objects are individually detected unless the dynamics of these moving objects are clearly different. There is a problem that it is difficult.

[0006]

According to a first aspect of the present invention, there is provided a region detecting method for a binary input image composed of 0 and 1 pixel values, the position of each pixel belonging to the lower end of each region having a pixel value of 1; Is detected for each pixel, the length of a connected component including the pixel at the intersection of the vertical line passing through the pixel and the region is detected as the height of the region on the pixel, and each pixel belonging to the lower end of the region is detected. By generating the lower end detection result data composed of the position of the area and the height of the area on the pixel, by approximating the lower end of the area with a group of line segments with reference to the lower end detection result data,
Generate line segment approximation result data composed of the position of each pixel belonging to a line segment group obtained by approximating the lower end of the region and the height of the region on the pixel, and refer to the line segment approximation result data for each. By detecting a group of rectangles approximating the region, a group of rectangles approximating the shape of each region included in the binary input image is output.

In a moving object detection method according to a second aspect of the present invention, a background image consisting only of a background that does not include a moving object is prepared in advance in an input image, and each pixel corresponds to the input image and the background image. Generating a background difference image having an absolute value of a pixel value difference as a pixel value, generating a binary image including pixel values of 0 and 1 with reference to the background difference image, A rectangle group approximating the shape of the region constituted by the pixel values of 1 is detected, the rectangle group approximating the region shape is referred to, and the rectangle is erased and divided according to a predetermined condition.
To generate a new rectangle by fusion, or a new rectangle by fusion
By performing shape generation, the rectangle group is changed, and a rectangle corresponding to each moving object is generated and used as a moving object detection result, thereby individually detecting moving objects included in the input image. It is characterized by.

In a moving object detecting method according to a third aspect of the present invention, an input image to be inputted always has a background image consisting only of a background not including a moving object, and the background image is referred to by referring to the input image. Update, generate a background difference image having a pixel value of an absolute value of a difference between a corresponding pixel value of the input image and the background image at each pixel, and refer to the background difference image with pixel values of 0 and 1 generates a binary image composed, said detecting a rectangular group of approximating the shape of the region composed of the binary image pixel value 1, the reference to a rectangular group of approximating the area shape, predetermined Erase rectangle, minute according to conditions
Create a new rectangle by splitting, or a new one by fusing
By performing generation of a rectangle, the rectangle group is changed,
The method is characterized in that a moving object included in an input image is individually detected by generating a rectangle corresponding to each moving object and generating a moving object detection result.

The region detecting apparatus according to a fourth aspect of the present invention comprises:
Input image storage means for storing a binary input image composed of the pixel values of the following, control means for controlling the operations of a lower end detection means, a line segment approximation means and a rectangle approximation means which will be described later, and control of the control means. First, the position of each pixel belonging to the lower end of each area of the pixel value 1 of the binary image stored in the input image storage unit is detected, and secondly, the vertical position of each pixel passing through the pixel is detected. Detecting the length of the connected component including the pixel at the intersection of the line and the region as the height of the region on the pixel;
Thirdly, a lower end detection unit that stores information on the position of each pixel and the height of an area on each pixel obtained as a result of the first and second operations as a lower end detection result in a lower end detection result storage unit described below; A lower end detection result storage unit for storing information on a lower end detection result detected by the lower end detection unit; and a first lower end detection result stored in the lower end detection result storage unit under control of the control unit. Second, the lower end of the region is approximated by a line segment group with reference to the information, and secondly, the information on the position of each pixel belonging to the line group and the height of the region on the pixel is obtained as a line segment approximation result, which will be described later. A line segment approximating unit to be stored in a storage unit; a line segment approximation result storing unit that stores information of a line segment approximation result obtained by approximation by the line segment approximating unit; Near the line segment stored in the line approximation result storage means A rectangle approximating unit that detects a rectangle group approximating each region with reference to the result information and stores the detected rectangle group in a region detection result storage unit described below, and a region detection result storage that stores the rectangle group obtained by the rectangle approximation unit Means, and 2
It is characterized in that a group of rectangles approximating the shape of each area included in the input image of the value is output.

According to a fifth aspect of the present invention, there is provided a moving object detecting apparatus, comprising: an input image storing means for storing an input image; a background image storing means for storing a background image consisting only of a background not including a moving object; Control means for controlling the operation of the difference means, the binarization means, the area detection means and the area shaping means; and the input image stored in the input image storage means for each pixel under the control of the control means; A background difference means for generating a background difference result image having an absolute value of a difference between corresponding pixel values of the background image stored in the background image storage means as a corresponding pixel value and storing the image in a background difference result storage means described later; A background difference result storage unit for storing the background difference image obtained by the background difference unit, and a background difference image stored in the background difference result storage unit under the control of the control unit. And binarizing means for and storing to generate a binary image composed of 0 and 1 of the pixel value to the binary image storage unit described later, the generated in binarizing means 2
Binary image storage means for storing a binarized image, and a shape of an area composed of pixel values 1 of the binarized image stored in the binarized image storage means under the control of the control means Area detection means for detecting a group of rectangles approximating and storing them in an area detection result storage means described later; area detection result storage means for storing a group of rectangles approximating the area shape obtained by the area detection means; Under the control of the control means, reference is made to the group of rectangles stored in the area detection result storage means, and a predetermined condition is satisfied.
Therefore, erasure of a rectangle, generation of a new rectangle by division, and
Or by generating a new rectangle by fusing,
A group of rectangles is changed, and a rectangle corresponding to each moving object is generated as a moving object detection result, which is obtained by a moving object detection means to be stored in a moving object detection result storage means described later, and the moving object detection means. Moving object detection result storage means for storing the group of rectangles, wherein the moving objects included in the input image are individually detected.

A moving object detecting apparatus according to a sixth aspect of the present invention includes an input image storing means for storing an input image, and operation of a background image updating means, a background subtracting means, a binarizing means, an area detecting means and an area shaping means which will be described later. Generating a new background image by referring to an input image stored in the input image storage unit and a background image stored in a background image storage unit described below under the control unit. A background image generating means to be stored in a background image storing means described later;
A background image storage unit for storing a background image output by the background image generation unit; and an input image and the background image storage unit stored in the input image storage unit for each pixel under the control of the control unit. A background difference means for generating a background difference result image having an absolute value of a difference between corresponding pixel values of the stored background image as a corresponding pixel value and storing the image in a background difference result storage means described later; A background difference result storage means for storing the background difference image obtained in step 1; and a pixel value of 0 and 1 with reference to the background difference image stored in the background difference result storage means under the control of the control means. Binarization means for generating a binarized image composed of the following and storing the same in a binarized image storage means described later, and a binarized image storage for storing the binarized image generated by the binarization means Means and the control means Under the control, a group of rectangles approximating the shape of the area formed by the pixel value 1 of the binarized image stored in the binarized image storage is detected and stored in the area detection result storage described below. Area detecting means for storing, area result storing means for storing a group of rectangles approximating the area shape obtained by the area detecting means, and area detecting result storing means under the control of the control means. Referring to a group of rectangles approximating the region shape , erasing and dividing rectangles according to predetermined conditions.
Create a new rectangle by splitting, or a new one by fusing
By performing generation of a rectangle, the rectangle group is changed,
A moving object detecting unit that generates a rectangle corresponding to each moving object to be a moving object detection result and stores the moving object detection result in a moving object detection result storage unit described below; and a moving unit that stores a rectangle group obtained by the moving object detecting unit. Object detection result storage means for individually detecting moving objects included in the input image.

As a result of area detection by a conventional technique relating to area detection, only ambiguous area shape information such as a circumscribed rectangle can be obtained. However, the area detection method according to the first invention and the area detection method according to the fourth invention have been described. When the apparatus is used, a group of rectangles approximating the region shape is detected, so that a fine region shape can be obtained from the result of the region detection.

In addition, according to the result of the moving object detection by the conventional technique relating to the detection of a moving object, for example, when different moving objects are photographed repeatedly and their corresponding regions are connected, these moving objects are individually identified. Detection was difficult without a clear difference in kinetics, but the second or third
Utilizing the moving object detection method of the invention and the moving object detection device of the fifth or sixth invention, if the shape characteristics of these objects appear in the region shape without investigating the dynamics of the moving objects , Can be detected by approximating individual rectangles.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The first invention relates to a method for detecting an area from a binary image. The second and third inventions relate to a method for detecting a moving object using the area detection method of the first invention. The fourth, fifth, and sixth inventions are apparatuses for realizing the first, second, and third inventions, respectively. In the second and third inventions, by applying the region detection method of the first invention to a binary image obtained from an image obtained by background subtraction, an area different from the background image of the input image is detected. doing. In particular, the second invention is characterized by using a constant background image throughout.
The third invention is characterized in that background images are sequentially updated. Hereinafter, embodiments of each invention will be described.

A first object of the present invention is to approximate a region shape of a region having a pixel value of 1 by a group of rectangles from a binary image having pixel values of 0 and 1.

An embodiment of the area detecting method according to the first invention will be described with reference to FIGS. 1, 7, 8 and 9.
Here, FIG. 1 is a flowchart showing the present embodiment, wherein 101 is a lower end detection process, 102 is a line segment approximation process, and 103 is a rectangle approximation process. FIGS. 7, 8, and 9 are explanatory diagrams of the lower end detection processing 101, the line segment approximation processing 102, and the rectangle approximation processing 103, respectively.

In the description of the present embodiment, the input image is I,
U represents the lower end image with height output in the lower end detection process, L represents underline data with height output in the line segment approximation process, and R represents rectangle approximate data output in the rectangle approximation process. Also,
The pixel and pixel value at the coordinates (x, y) of the image Φ are Φ
(X, y) and {Φ (x, y) | x = i} of image Φ
Will be referred to as the i-th column of Φ. Further, the image size of the input image I is X × Y, t is an approximate allowable width in the line segment approximation processing, and k is a minimum integer equal to or more than Y / t. At this time, in this embodiment, the height-added lower end image U
Is X × Y, the image size of the underlined data L with height is X × 2k, the size of the approximate rectangle data R in the y-axis direction is 4, and the number of detected rectangles in the x-axis direction. here,
Since the size of the input image I is finite, it is possible to estimate the upper limit of the number of detected rectangles. In this embodiment, FIG.
As shown in the figure, the lower end detection processing 101 and the line segment approximation processing 10
2. Processing is performed in the order of the rectangular approximation processing 103.

First, the lower end detection processing 101 will be described. In the input image I, the pixel value in a region to be detected is 1 and the pixel value in other regions is 0. In this processing, for each pixel corresponding to the lower end of each area of the pixel value 1 in the binary input image I, the height in the vertical direction of the pixel in the area is obtained, and the pixel value corresponding to the pixel is calculated as the height. Satoshi,
Except for the lower end of the region, a lower end image U with height having a pixel value of 0
The goal is to generate Here, for a pixel belonging to the lower end of the region, the height of the pixel in the vertical direction of the region indicates the length of a connected component including the pixel at the intersection of a straight line parallel to the y-axis and the region. ing. The processing will be specifically described below.

All pixels of the height-added lower end image U are cleared in advance with a pixel value of 0. Since this process is performed independently for each column of the input image I, the process for the i-th column will be described below as an example. Let the i-th column of the input image I be the column of interest,
The column of interest is scanned from its upper end in the y-axis direction, that is, downward, and the height of the region is counted. The initial value of the height count is set to 0, and the following processing is performed. (1) When the height count value is 0 and the target pixel value is 0,
do nothing. (2) When the height count value is 0 and the pixel value of interest is 1,
Interpret that the top of the area was found and set the height count value of the area to 1
And (3) If the height count is not 0 and the target pixel value is 1, the height count value is increased by 1. (4) When the height count is not 0 and the target pixel value is 0, it is interpreted that the pixel immediately above the target pixel corresponds to the lower end of the region, and the upper end with height corresponding to the lower end pixel of the region. The height count value is written to the pixels of the image U, and thereafter, the height count value is cleared to zero. (5) When the lower end of the image has not been reached, the process proceeds to the next pixel and the operations (1) to (4) are performed. If it has reached the lower end of the image, the processing for the column of interest ends. FIG. 7 shows an example of the above processing. In FIG. 7, only the target columns of the input image I and the lower end image U with height are displayed in (a) and (c), respectively. Also, the input image I
The pixel value of the column of interest is 0, 3 ≦ y ≦ for 0 ≦ y ≦ 2.
The value of 1 is 8 for 8 and 0 for 9 ≦ y ≦ 11, and the height of the pixel value 1 portion is 6 pixels. FIG. 7 (b)
The value of the counter indicates the value before and after the processing of the height counter when scanning each pixel. Therefore, the pixel value in the column of interest of the output lower end image U with height is 6 only for the pixel of y = 8 and 0 for the other pixels.

By performing the above-described processing for each column, in each pixel of the lower end image U with height, the pixel corresponding to the lower end of the area of the input image having the pixel value 1 is set to the pixel value of the area. , And the other pixel values are 0.

Next, the line segment approximation processing 102 will be described. This processing divides the lower end image U with height into predetermined blocks, and obtains the height of the lower end and the height of the area obtained by approximating the lower end with height for each block by a line segment. The purpose is to generate underline data L. The lower end detected by the lower end detection processing 101 includes not a precise line segment but a vertical fluctuation. The purpose of this processing is to absorb this fluctuation and recognize the lower end as a line segment. The contents of this processing will be described with reference to FIG.

As described above, the size of the lower end image U with height is X × Y, the approximate allowable width is t, and the minimum integer equal to or larger than Y / t is k. As shown in FIG.
Is divided into X × t blocks. That is, the j-th block B _j is a set of pixels satisfying {U (x, y) | jt ≦ y <(j + 1) t}. Where B _k-1 = ｛U (x, y)
| (K−1) t ≦ y <Y｝. At this time, the number of blocks is k.

Each successive lower end piece with height included in the same block is approximated by one line segment with height. The process of approximating the lower end piece with height in B _j by a line segment with height will be described. First, height data h _j , position data l _j and average position data a _j of size X × 1 are prepared, and the values of the height data, the position data and the average position data corresponding to the i-th column of B _j are calculated. , Each h
_j [i], l _j [i] and a _j [i]. In the search window, for each i, scan the ith column of B _j from top to bottom, write the sum of pixel values in h _j [i], and write a line of non-zero pixel values in l _j [i] Write the maximum value of the address. Accordingly, the height of the height information with lower piece is recorded in the height data h _j, the position is recorded in the position data l _j. When the above processing is performed for the search block in FIG. 8A, the result is as shown in FIG. 8B.

Next, for i where h _j [i]> 0, h
The average value of the position data in the connected component having a positive value of the height data h _j to which _j [i] belongs is written to a _j [i].
Further, in order to correct a height shift caused by approximating the position of the lower end with height by a line segment, a _j [i] −1
_j [i] is added to _hj [i]. The above processing is shown in FIG.
When applied to the height data and the position data of (b), the result is as shown in FIG. Then, L [i] [2j] = h
By setting _j [i] and L [i] [2j + 1] = _aj [i], lower end data with height as shown in FIG. 8D is generated.

Next, the rectangle approximation processing 103 will be described. From the underlined data L with height obtained in the line segment detection processing 102, a rectangle approximation is performed by determining a substantially equal segment on the same underline as the lower side of one rectangle to generate rectangle approximate data R. . The contents of this processing will be described with reference to FIG.

First, the underlined data L with height is subdivided according to the height information. Each underline with height stored in the underline data with height L is sequentially referred to. In the height data in L, the difference in height between adjacent pixels is referred to as a height step, and a threshold T relating to the height step is prepared. Each underline with height is subdivided into underlined pieces with height where the height step is greater than or equal to T.

Next, rectangular approximation data R is generated based on each underlined piece with height. The underlined piece l _k with height obtained at the k-th is L [i] in the underlined data L with height.
[J] (α _k ≦ i ≦ β _k , 2n _k ≦ j ≦ 2n _k +1), x _k = α _k , y _k = L [ _ak ]
[2n _k +1], w _k = β _k −α _k +1,

[0028]

(Equation 1)

Then, (x _k , y _k ) represents the coordinates of the lower left vertex of the rectangle, w _k represents the width of the rectangle, and h _k represents the height of the rectangle. The above processing is performed for each underline with height, and the rectangle approximate data R is obtained as R [k] [0] = x _k , R [k]
[1] = y _k , R [k] [2] = w _k , R [k] [3]
= It is generated by the h _k.

As an example, the operation of this processing will be described using underlined data L with height in FIG. Note that the threshold value for the height step is T = 5. First, the first underlined data with height is 20 ≦ x ≦ 21, 0 ≦ y ≦ 1
, And there is no place where the height step is 5 or more, so it is approximated by one rectangle, and the underlined left end coordinate x ₀ = 20
And y ₀ = 2, width w ₀ = 2, and height h ₀ = 3. Next, the data of the second underline with height is 6 ≦ x ≦ 16, 3 ≦
The height step exceeds T at only one location between x = 10 and x = 11 at y ≦ 4. Therefore, the data of the second underline with height is 6 ≦ x ≦ 10 and 11 ≦ x ≦ 16
, And the underlined left end coordinates x ₁ = 6 and y ₁ = 6, width w ₁ = 5, height h with respect to the left underlined piece with height
₁ = 10 is obtained, and the underlined left end coordinates x ₂ = 11 and y ₂ = 6, width w ₂ = 6, and height h ₂ = 21 are obtained for the right underlined piece with height. Similarly, for the data of the third underline with height, the underline left end coordinates x ₃ = 0 and y ₃ =
9, a width w ₃ = 4 and a height h ₃ = 9 are obtained. From these results, rectangular approximation data R shown in FIG. 9B is generated.

A second object of the present invention is to detect a moving object faithfully in an area shape by binarizing a fluctuation area from the background and applying the area detection method of the first invention. And For one embodiment of the present invention,
This will be described with reference to FIGS. Here, FIG. 2 is a flowchart showing one embodiment of the moving object detection method according to the second invention, wherein 201 is a background difference process, 202 is a binarization process,
03 is an area detection process, and 204 is a moving object detection process. FIG. 10 is an explanatory diagram for explaining the operation of this embodiment, taking a case where a vehicle is detected from a road image as an example.

In this embodiment, the input image is represented by I, the background image is represented by B, the background difference image is represented by S, the binarized image is represented by C, and the area detection data is represented by R.

In this embodiment, a background image B which does not include a moving object and includes only a background is prepared in advance. FIG.
10A and 10B show examples of the input image I and the background image B, respectively.

First, the background difference processing 201 will be described. In the background subtraction processing, the input image I
And a background difference image S having an absolute value of a difference between corresponding pixel values of the background image B and a corresponding pixel value. FIG.
When the input image of (a) and the background image of FIG. 10B are used, the background difference image is as shown in FIG. 10C.

Next, in the binarization process 202, a binary image comprising 0 and 1 pixel values is referred to the background difference image S.
Generate a digitized image C. For example, a threshold value T for a pixel value
Is prepared, and the corresponding pixel value is set to 1 or 0 in accordance with whether each pixel value of the background difference image S is equal to or greater than T or less than T, so that the binarized image C can be generated. FIG.
FIG. 10D shows a binarized image obtained by binarizing the background difference image of FIG.

Next, in the area detection processing 203, the area constituted by the pixel value 1 of the binarized image is detected as a group of rectangles approximating the area shape by the area detection method of the first invention. Therefore, the embodiment of the area detecting method of the first invention can be adopted as one embodiment of the present processing. 2 in FIG. 10 (d)
FIG. 10E shows a rectangle group obtained by applying the area detection method of the first invention to the quantified image in an image space.

In the moving object detecting process 204, the group of rectangles is changed according to a predetermined operation with reference to a group of rectangles approximating the region shape, and a region corresponding to each moving object is generated to detect the moving object. Result. An example of a change operation of a rectangle group is described below. (1) A rectangle whose height is less than the threshold is deleted. (2) When the rectangles A and B satisfy the following three conditions,
The rectangle A and the rectangle B are deleted, and the circumscribed rectangle C is generated.

(A) The difference between the x coordinate of the right end of rectangle A and the x coordinate of the left end of rectangle B is less than the threshold value.

(B) The difference between the y-coordinates of the lower ends of the rectangles A and B is less than the threshold value.

(C) The difference between the heights of the rectangles A and B is less than the threshold value. (3) When rectangles A, B, and C satisfy the following conditions, upper rectangle B obtained by bisecting rectangle B up and down
₁ and generates a rectangular B ₂ lower, erasing the rectangle B, performed rectangle A and the rectangle B _1, and the rectangular B ₂ and with respect to the rectangular C fusion operations (2).

(A) The difference between the x coordinate of the right end of rectangle A and the x coordinate of the left end of rectangle B is less than the threshold value.

(B) The difference between the x coordinate of the right end of the rectangle B and the x coordinate of the left end of the rectangle C is less than the threshold value.

(C) The difference between the y-coordinates of the upper ends of the rectangles A and B is less than the threshold value.

(D) The difference between the y-coordinates of the lower ends of the rectangles B and C is less than the threshold value. A moving object detection process is performed on the rectangle group of FIG. 10E based on the above three operations, and the detected moving object is
(F).

In the moving object detection method according to the third aspect of the present invention, in the detection of a moving object from a moving image, even if the background is not stable, the moving object is detected while generating the background image so as to obtain the same effect as in the second aspect. This is an invention aimed at detecting an object. Further, similarly to the second aspect, by using the area detection method of the first aspect, it is possible to detect a moving object faithfully according to the area shape.

An embodiment of the moving object detecting method according to the third invention will be described with reference to FIG. Here, FIG.
3 is a flowchart showing an embodiment of a moving object detection method according to the present invention, wherein 301 is a background generation process, 302 is a background difference process,
03 is a binarization process, 304 is an area detection process, and 305 is a moving object detection process.

First, in the background generation processing of 301 in FIG.
As an example, the background image generation method described in “Background Image Generation Method and Apparatus” described in Japanese Patent Application No. 6-66950 can be used.

Further, 302, 303, 304, 30 in FIG.
Regarding the background subtraction processing, binarization processing, area detection processing, and moving object detection processing of 5, the same method as that of the embodiment of the second invention can be used as an example.

An embodiment of the region detecting apparatus according to the fourth invention will be described with reference to FIGS. 4, 7, 8 and 9.
Here, FIG. 4 is a configuration diagram of the present embodiment, where 401 is input image storage means, 402 is control means, 403 is lower end detection means, 404 is lower end detection result storage means, 405 is line segment approximation means, and 406 is Line segment approximation result storage means, 407 is a rectangle approximation means, and 408 is an area detection result storage means. Also,
FIGS. 7, 8 and 9 are explanatory diagrams relating to the operations of the lower end detecting means, the line segment approximating means and the rectangular approximating means shown in FIG. 4, respectively.

In the description of the present embodiment, the input image stored in the input image storage means 401 is I, the lower end image with height stored in the lower end detection result storage means is U, and the height stored in the line segment approximation result storage means. L represents the underlined data, and R represents the rectangular approximation data stored in the rectangular approximation result storage means. Also,
The pixel and pixel value at the coordinates (x, y) of the image Φ are Φ
(X, y) and {Φ (x, y) | x = i} of image Φ
Will be referred to as the i-th column of Φ. In this embodiment, the image size of the input image I is X × Y, t is an approximate allowable width used in the operation of the line segment approximating means, and k is a minimum integer equal to or more than Y / t. At this time, the image size of the underlined image with height U is X × Y, the image size of the underlined data L with height is X × 2k, the size of the approximate rectangle data R in the y-axis direction is 4, and the x-axis direction is detected. Number of rectangles. Input image I
Is finite, it is possible to estimate the upper limit of the number of detected rectangles.

First, the operation of the control means 402 will be described. In this embodiment, as shown in FIG.
2 controls the lower end detection means 403, the line segment approximation means 405, and the rectangle approximation means 407, and determines the order of each processing. That is, the control means 402 first starts the lower end detection means 4
03, the start signal is transmitted to the line segment approximation unit 405 when the end signal is received from the lower end detection unit 403, and the start signal is transmitted to the rectangle approximation unit 407 when the end signal is received from the line segment approximation unit 405. Is transmitted, and when the end signal is finally received from the rectangle approximating means 407, the entire process is terminated.

Next, the operation of the lower end detecting means 403 will be described. This unit starts processing in response to a start signal from the control unit 402, and refers to the input image I stored in the input image storage unit 401. In the input image I, it is assumed that the pixel value in an area to be detected is 1 and the pixel value in other areas is 0. The processing in the lower end detecting means 403 is performed by using the pixel value 1 in the binary input image I.
For each pixel corresponding to the lower end of each area, the height of the pixel in the vertical direction of the area is determined, and for the pixel corresponding to the lower end, the height of the area is used as the pixel value.
It is an object to generate a height-added lower end image.
Here, for a pixel belonging to the lower end of the region, the height of the pixel in the vertical direction of the region indicates the length of a connected component including the pixel at the intersection of a straight line parallel to the y-axis and the region. ing. The processing will be specifically described below.

All the pixels of the lower end image U with height, which is the contents of the lower end detection result storage means 404, are cleared with a pixel value of 0 in advance. Since this process is performed independently for each column of the input image I, the process for the i-th column will be described below as an example. The i-th column of the input image I is set as the column of interest, and the column of interest is scanned from its upper end in the y-axis direction, that is, downward, and the height of the region is counted. The initial value of the height count is set to 0, and the following processing is performed. (1) When the height count value is 0 and the target pixel value is 0,
do nothing. (2) When the height count value is 0 and the pixel value of interest is 1,
Interpret that the top edge of the area was found and set the area height count value to 1.
And (3) If the height count value is not 0 and the pixel value of interest is 1, the height count value is increased by one. (4) If the height count value is not 0 and the pixel value of interest is 0, it is interpreted that the pixel immediately above the pixel of interest corresponds to the lower end of the region, and a height corresponding to the lower end pixel of the region is added. The height count value is written to the pixels of the lower end image U, and thereafter, the height count value is cleared to zero. (5) When the lower end of the image has not been reached, the process proceeds to the next pixel and the operations (1) to (4) are performed. If it has reached the lower end of the image, the processing for the column of interest ends.

FIG. 7 shows an example of the above processing. In FIG.
Only the input image as the content of the input image storage means 401 and the column of interest of the height-added lower image U as the content of the lower edge detection result storage means 404 are displayed in (a) and (c), respectively. . The pixel value is 0 for 0 ≦ y ≦ 2, 1 for 3 ≦ y ≦ 8, and 0 for 9 ≦ y ≦ 11, and the height of the pixel value 1 portion is 6 pixels. The value of the counter in FIG. 7B indicates the value before and after the processing of the height counter when scanning each pixel. Accordingly, the pixel value in the i-th column of the output lower end image U with height is 6 for only the pixel of y = 8 and 0 for the other pixels.

By performing the above processing for each column, in each pixel of the lower end image U with height, the pixel corresponding to the lower end of the area of the input image having the pixel value 1 is set to the pixel value of the area. , And the other pixel values are 0. When the above processing is completed, the lower end detecting means 403
2 is sent an end signal.

In the above operation of the lower end detecting means 403, the processing for each column is performed independently, so that it is possible to adopt a configuration in which these processings are performed in parallel.

Next, the operation of the line segment approximating means 405 will be described. This unit starts processing upon receiving a start signal from the control unit 402. The processing in this means is obtained by dividing the lower end image U with height, which is the content of the lower end detection result storage means 404, into predetermined blocks, and approximating the lower end with height for each block by a line segment. It is an object to generate underlined data L with height expressing the position of the underline and the height of the area. The lower end in the lower end image U with height stored by the lower end detection result storage unit 404 includes not a precise line segment but a vertical fluctuation. The purpose of this processing is to absorb this fluctuation and recognize the lower end as a line segment. The contents of this processing will be described with reference to FIG.

As described above, the size of the lower end image U with height is X × Y, the approximate allowable width is t, and the minimum integer equal to or larger than Y / t is k. As shown in FIG.
Is divided into X × t blocks. That is, the j-th block B _j is a set of pixels satisfying {U (x, y) | jt ≦ y <(j + 1) t}. Where B _k-1 = ｛U (x, y)
| (K-1) t ≦ y ≦ Y｝. At this time, the number of blocks is k.

Each successive underlined piece with height included in the same block is approximated by one line segment with height. The process of approximating the underlined piece with height in B _j with a line segment with height will be described. First, height data h _j , position data l _j and average position data a _j of size X × 1 are prepared, and height data h _j corresponding to the ith column of B _j are prepared.
And the values of the position data l _j and the average position data a _j are given by h _j [i], l _j [i] and a _j , respectively.
_j Represented by [i]. In the search window, for each i,
Scan the ith column of B _j from top to bottom, h _j [i]
And the maximum value of the line address of the pixel value other than 0 is written in l _j [i]. Accordingly, the height of the height information with lower piece is recorded in the height data h _j, the position is recorded in the position data l _j. When the above processing is performed for the search block in FIG.
(B).

Next, for i where h _j [i]> 0, h
The average value of the position data in the positive connected component of the height data to which _j [i] belongs is written to a _j [i]. Further, in order to correct a height shift caused by approximating the position of the lower end with height by a line segment, a _j [i] −1
_j [i] is added to _hj [i]. The above processing is shown in FIG.
When applied to the height data and the position data of (b), the result is as shown in FIG. Then, L [i] [2n] = h
By setting _j [i] and L [i] [2n + 1] = a _j [i], lower end data L with height as shown in FIG. 8D is generated.

When the above processing is completed, the line segment approximating means 4
05 transmits an end signal to the control means 402.

In the processing by the line segment approximating means 405, the recording area for storing the data of h _j , l _j , and a _j can be configured to secure an independent area for each block. Are performed in order, not in parallel, between blocks, it is also possible to use the same area regardless of the block number j.

Next, the operation of the rectangular approximation means 407 will be described. This unit starts processing upon receiving a start signal from the control unit 402. From the underlined data L with height stored in the line segment detection result recording unit 406, a rectangle approximation process is performed by determining that a segment having substantially the same height on the same underline is the lower side of one rectangle. Generate R. The details of the rectangle approximation processing will be described with reference to FIG.

First, the underlined data L with height is subdivided according to the height information. Each underline with height stored in the underline data with height L is sequentially referred to. In the height data in L, the difference in height between adjacent pixels is referred to as a height step, and a threshold T relating to the height step is prepared. Each underline with height is subdivided into underlined pieces with height where the height step is greater than or equal to T.

Next, approximate rectangular data R is generated based on each underlined piece with height. The underlined piece l _k with height obtained at the k-th is L [i] in the underlined data L with height.
[J] (α _k ≦ i ≦ β _k , 2n _k ≦ j ≦ 2n _k +1), x _k = α _k , y _k = L [α _k ]
[2n _k +1], w _k = β _k −α _k +1,

[0066]

(Equation 2)

Then, (x _k , y _k ) represents the coordinates of the lower left vertex of the rectangle, w _k represents the width of the rectangle, and h _k represents the height of the rectangle. The above processing is performed for each underline with height, and the rectangle approximate data R is obtained as R [k] [0] = x _k , R [k]
[1] = y _k , R [k] [2] = w _k , R [k] [3]
= It is generated by the h _k.

As an example, the operation of this processing will be described using underlined data L with height in FIG. Note that the threshold value for the height step is T = 5. First, the first underlined data with height is 20 ≦ x ≦ 21, 0 ≦ y ≦ 1
, And there is no place where the height step is 5 or more, so it is approximated by one rectangle, and the underlined left end coordinate x ₀ = 20
And y ₀ = 2, width w ₀ = 2, and height h ₀ = 3. Next, the data of the second underline with height is 6 ≦ x ≦ 16, 3 ≦
The height step exceeds T at only one location between x = 10 and x = 11 at y ≦ 4. Therefore, the data of the second underline with height is 6 ≦ x ≦ 10 and 11 ≦ y ≦ 16
, And the underlined left end coordinates x ₁ = 6 and y ₁ = 6, width w ₁ = 5, height h with respect to the left underlined piece with height
₁ = 10 is obtained, and the underlined left end coordinates x ₂ = 11 and y ₂ = 6, width w ₂ = 6, and height h ₂ = 21 are obtained for the right underlined piece with height. Similarly, for the data of the third underline with height, the underline left end coordinates x ₃ = 0 and y ₃ =
9, a width w ₃ = 4 and a height h ₃ = 9 are obtained. From these results, rectangular approximation data R shown in FIG. 9B is generated.

The rectangular data R generated as described above is stored in the area detection result storage means 408. When all of the above processes have been completed, the rectangle approximating unit 407 returns to the control unit 402.
To send an end signal.

An embodiment of the moving object detecting apparatus according to the fifth invention will be described with reference to FIGS. Here, FIG. 5 is a configuration diagram showing an embodiment of the moving object detection device according to the fifth invention, wherein 501 is an input image storage means, 502
Is a background image storage means, 503 is a control means, 504 is a background difference means, 505 is a background difference result storage means, and 506 is 2
A binarizing unit, 507 is a binarized image storing unit, 508 is a region detecting unit, 509 is a region detecting result storing unit, 510 is a moving object detecting unit, and 511 is a moving object detecting result storing unit. FIG. 10 is an explanatory diagram for explaining the operation of the moving object detection device according to the present embodiment, taking as an example a case where a vehicle is detected from a road image.

In the description of this embodiment, the input image stored in the input image storage means 501 is I, and the background image storage means 5
02 is B, the background difference result storage means 5
S, the background difference result image stored in the area 05 is C, the binarized image stored in the binarized image storage means 507 is C, the area detection result data stored in the area detection result storage means 509 is R, and the moving object detection result storage means. The moving object detection result data stored in 511 is represented by M. Further, a pixel and a pixel value at the coordinates (x, y) of the image Φ are represented by Φ (x, y), and in this embodiment, the image size of the input image I is X × Y. At this time, the image sizes of the background image B, the background difference result image S, and the binarized image C are all X × Y. The contents of the region detection result data R and the moving object detection result data M are the position and size of a rectangle expressing these results. The size of these data in the y-axis direction is 4 and the data in the x-axis direction is detected. Number of rectangles.

In this embodiment, an image composed of only the background without any moving object is prepared in advance, and stored in the background image storage means 502 as the background image B. FIG.
10A and 10B show an example of an input image I as the content of the input image storage means 501 and an example of a background image B as the content of the background image storage means 502, respectively.

First, the operation of the control means 503 will be described. In the present embodiment, as shown in FIG.
3 controls the background subtraction unit 504, the binarization unit 506, the area detection unit 508, and the moving object detection unit 510, and determines the order of each processing. That is, the control means 503
Transmits an activation signal to the background subtraction means 504 first, transmits an activation signal to the binarization means 506 when receiving an end signal from the background subtraction means 504, and transmits an activation signal to the binarization means 506 when receiving an end signal from the binarization means 506. An activation signal is transmitted to the area detection means 508, and when an end signal is received from the area detection means 508, an activation signal is transmitted to the moving object detection means 510. When an end signal is finally received from the moving object detection means 510, all processes are performed. To end.

Next, the operation of the background subtraction means 504 will be described. This means starts processing in response to a start signal from the control means 503, and in each pixel (i, j), the pixel value I (i, j) of the input image I stored in the input image storage means 501 and Absolute value | of difference from pixel value B (i, j) of background image B stored in background image storage unit 502
I (i, j) -B (i, j) |
A background difference image S is generated as (i, j) and stored in the background difference result storage unit 505. When the input image I of FIG. 10A and the background image B of FIG. 10B are used, the background difference image S is as shown in FIG. When the above processing is completed, the background difference means 504 sends an end signal to the control means 50.
Send to 3.

In the background subtraction processing by the background subtraction means 504, since independent processing is performed for each pixel,
While a configuration in which processing is performed sequentially while scanning each pixel is possible, a configuration in which processing for each pixel is performed in parallel is also possible.

Next, the operation of the binarizing means 506 will be described. This means starts the processing in response to the activation signal from the control means 503, and refers to the background difference image S stored in the background difference image storage means 505 to perform binarization comprising 0 and 1 pixel values. An image is generated and stored in the binarized image storage unit 507. For example, a threshold value T for a pixel value is prepared, and the corresponding pixel value C (i,
By setting j) to 1 or 0, a binary image C can be generated. That is, in the binarized image C, the pixel value is 1 in an area where the background difference value is large and the moving object is considered to be imaged, and the area where the background difference value is small and the background is considered to be imaged. Pixel value is 0 at
The image is FIG. 10D shows a binarized image C obtained by binarizing the background difference image S in FIG. 10C. When the above processing is completed, the binarizing unit 506 transmits an end signal to the control unit 503.

In the above-described binarization processing by the binarization means 506, since independent processing is performed for each pixel, a configuration in which processing is performed sequentially while scanning each pixel is possible.
A configuration in which processing for each pixel is performed in parallel is also possible.

Next, the operation of the area detecting means 508 will be described. The present means starts processing in response to a start signal from the control means 503, and stores the area constituted by the pixel value 1 of the binarized image C stored in the binarized image storage means in the fourth invention. A group of rectangles approximating the region shape is detected using the region detection device of (1), and is stored in the region detection result storage means 509 as region detection result data R. As one embodiment of the present means, the embodiment of the area detecting device of the fourth invention can be adopted.
FIG. 10E shows a rectangle group obtained by applying the region detection device of the fourth invention to the binarized image of FIG. 10D in an image space. When the above processing is completed, the area detecting means 508 sends an end signal to the control means 503.

Next, the operation of the moving object detecting means 510 will be described. The present means starts processing in response to a start signal from the control means 503, and refers to a group of rectangles approximating the area shape described in the area detection result data R stored in the area detection result storage means 509, The rectangle group is changed in accordance with a predetermined operation, an area corresponding to each moving object is generated and stored as moving object detection result data M in the moving object detection result storage unit 511. An example of a change operation of a rectangle group is given below. (1) A rectangle whose height is less than the threshold is deleted. (2) When the rectangles A and B satisfy the following three conditions,
The rectangle A and the rectangle B are deleted, and the circumscribed rectangle C is generated.

(A) The difference between the x coordinate at the right end of rectangle A and the x coordinate at the left end of rectangle B is less than the threshold value.

(B) The difference between the y coordinates of the lower ends of the rectangles A and B is less than the threshold value.

(C) The difference between the heights of the rectangles A and B is less than the threshold value. (3) When rectangles A, B, and C satisfy the following conditions, upper rectangle B obtained by bisecting rectangle B up and down
₁ and generates a rectangular B ₂ lower, erasing the rectangle B, performed rectangle A and the rectangle B _1, and the rectangular B ₂ and with respect to the rectangular C fusion operations (2).

(A) The difference between the x coordinate at the right end of rectangle A and the x coordinate at the left end of rectangle B is less than the threshold value.

(B) The difference between the x coordinate of the right end of rectangle B and the x coordinate of the left end of rectangle C is less than the threshold value.

(C) The difference between the y-coordinates of the upper ends of the rectangles A and B is less than the threshold value.

(D) The difference between the lower y coordinates of the rectangle B and the rectangle C is smaller than the threshold value. A rectangle group corresponding to the moving object detection data M obtained by the present means based on the above three operations is compared with a rectangle group corresponding to the area detection result data R shown in FIG.
(F). When the above processing is completed, the moving object detection unit 510 transmits an end signal to the control unit 503.

An embodiment of the moving object detecting apparatus according to the sixth invention will be described with reference to FIG. Here, FIG.
1 is a configuration diagram showing an embodiment of a moving object detection device according to the present invention, wherein 601 is an input image storage unit, 602 is a control unit, and 6
03 is a background image generation means, 604 is a background image storage means,
605 is a background difference means, 606 is a background difference result storage means, 607 is a binarization means, 608 is a binarized image storage means, 609 is an area detection means, 610 is an area detection result storage means, and 611 is a moving object detection means , 612 are moving object detection result storage means.

First, the operation of the control means 602 will be described. In the present embodiment, as shown in FIG.
2 is a background image generating means 603, a background subtracting means 605, 2
It controls the value conversion means 607, the area detection means 609, and the moving object detection means 611, and determines the order of each processing.
That is, the control unit 602 first sets the background image generation unit 6
03, the start signal is transmitted to the background subtraction unit 605 when the end signal is received from the background image generation unit 603, and the start signal is transmitted to the binarization unit 607 when the end signal is received from the background difference unit 605. And the binarizing means 60
7 receives an end signal from the area detecting means 609, receives an end signal from the area detecting means 609, transmits an activation signal to the moving object detecting means 611, and finally transmits a starting signal from the moving object detecting means 611. When the end signal is received, the entire process is terminated.

Next, the operation of the background image generating means 603 will be described. The present means starts processing in response to a start signal from the control means 602, refers to the input image stored in the input image storage means 601 and the background image stored in the background image storage means 604, and creates a new background image. An image is generated and stored in the background image storage unit 604. Here, as an example of a background image generation method, a background image generation method described in “Background Image Generation Method and Apparatus” described in Japanese Patent Application No. 6-66950 can be used. When the above processing is completed, the background image generation unit 603 transmits an end signal to the control unit 602.

The background difference means 605 and the binarization means 6
07, the operation of the area detecting means 609, the operation of the moving object detecting means 611, and the roles of the background difference result storing means 606, the binarized image storing means 608, the area detecting result storing means 610, and the moving object detecting result storing means 612, The same method as that of the fifth embodiment can be used.

[0091]

By using the area detection method of the first invention and the device of the fourth invention, two pixels consisting of pixel values 0 and 1 are used.
Since it is possible to detect, from the value image, a group of rectangles approximating the shape of the area formed by the pixel value 1, it is possible to detect the area while expressing the shape.

Further, when the moving object detecting method of the second invention and the moving object detecting device of the fifth invention are used, the moving object overlaps the region shape corresponding to a plurality of moving objects that can exist in the grayscale image. Even when the image is taken with the camera, it can be detected by approximating it with an individual rectangle.

Further, when the moving object detecting method of the third invention and the moving object detecting device of the sixth invention are used, a moving object can be detected in the same manner as in the second and fifth inventions while generating a background image. In order to perform detection, a region shape corresponding to a plurality of moving objects that may exist in a grayscale image input under unstable shooting conditions can be separated into individual rectangles even when the moving object is overlapped. Can be approximated and detected.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an embodiment of a region detection method according to the first invention.

FIG. 2 is a flowchart showing an embodiment of a moving object detection method according to the second invention.

FIG. 3 is a flowchart showing an embodiment of a moving object detection method according to the third invention.

FIG. 4 is a block diagram showing an embodiment of an area detecting device according to the fourth invention.

FIG. 5 is a block diagram showing an embodiment of a moving object detection device according to the fifth invention.

FIG. 6 is a block diagram showing an embodiment of a moving object detection device according to the sixth invention.

FIG. 7 is an explanatory diagram of an embodiment of a region detection method according to the first invention and a region detection device according to a fourth invention;

FIG. 8 is an explanatory view of an embodiment of a moving object detection method according to the second invention and a moving object detection device according to the fifth invention;

FIG. 9 is a view for explaining a rectangle approximation process 103;

FIG. 10 is a diagram illustrating an operation of detecting a vehicle from a road image.

[Explanation of symbols]

 101 Lower edge detection processing 102 Line segment approximation processing 103 Rectangle approximation processing 201, 302 Background difference processing 202, 303 Binarization processing 203, 304 Area detection processing 204, 305 Moving object detection processing 301 Background generation processing 401, 501, 601 Input image Storage means 402, 503, 602 Control means 403 Lower edge detection means 404 Lower edge detection result storage means 405 Line segment approximation means 406 Line segment approximation result storage means 407 Rectangular approximation means 408 Area detection result storage means 502, 604 Background image storage means 504, 605 Background difference means 505,606 Background difference result storage means 506,607 Binarization means 507,608 Binary image storage means 508,609 Area detection means 509,610 Area detection result storage means 510,611 Moving object detection means 511 , 612 Moving object Body detection result storage means 603 Background image update means

Claims (6)

(57) [Claims] A binary input image composed of pixel values of 0 and 1 detects a position of each pixel belonging to a lower end of each area of a pixel value of 1, and a vertical line passing through the pixel for each pixel. And the length of the connected component including the pixel at the intersection of the region is detected as the height of the region on the pixel, and is constituted by the position of each pixel belonging to the lower end of the region and the height of the region on the pixel. The lower end detection result data is generated, and the lower end of the region is approximated by a line segment group with reference to the lower end detection result data, whereby the position of each pixel belonging to the line segment group obtained by approximating the lower end of the region and the pixel By generating line segment approximation result data composed of the height of the upper region and detecting a group of rectangles approximating each region with reference to the line segment approximation result data, the data is included in the binary input image. Output a group of rectangles approximating the shape of each area. Region detection method according to claim. 2. A background image comprising only a background not including a moving object with respect to an input image is prepared in advance, and an absolute value of a difference between a corresponding pixel value of the input image and a corresponding pixel value of the background image is determined for each pixel. Generating a background difference image as a value, generating a binarized image composed of pixel values of 0 and 1 with reference to the background difference image, and generating an area composed of a pixel value of 1 of the binarized image. A group of rectangles approximating the shape of the region is detected, and the group of rectangles approximating the shape of the region is referred to, and according to predetermined conditions.
To erase a rectangle, create a new rectangle by dividing,
Generates a new rectangle by fusing,
A moving object detection method, wherein a moving object included in an input image is individually detected by changing a group and generating a rectangle corresponding to each moving object to obtain a moving object detection result. 3. An input image that is input always has a background image composed of only a background that does not include a moving object, updates the background image with reference to the input image, and updates the input image at each pixel. And a background difference image having the absolute value of the difference between the corresponding pixel values of the background image as the pixel value, and generating a binarized image composed of 0 and 1 pixel values with reference to the background difference image. Then, a group of rectangles approximating the shape of the region formed by the pixel value 1 of the binarized image is detected, and the group of rectangles approximating the shape of the region is referred to, according to a predetermined condition.
To erase a rectangle, create a new rectangle by dividing,
Generates a new rectangle by fusing,
A moving object detection method, wherein a moving object included in an input image is individually detected by changing a group and generating a rectangle corresponding to each moving object to obtain a moving object detection result. 4. First, a pixel having a pixel value of 0 and 1
Second, the position of each pixel belonging to the lower end of each area of the pixel value 1 of the input image of the value is detected, and secondly, for each of the pixels, a connected component including the pixel at the intersection of a vertical line passing through the pixel and the area The length is detected as the height of the area on the pixel,
And bottom edge detecting means for outputting information on the position of each pixel and the height of the area on each pixel obtained as a result of the second operation as a bottom edge detection result. First, referring to the information on the bottom edge detection result, Secondly, a line segment approximating means for approximating the lower end of the region with a line segment group and outputting information on the position of each pixel belonging to the line segment group and the height of the region on the pixel as a line segment approximation result; Rectangle approximation means for detecting a group of rectangles approximating each region with reference to the information of the approximation result, and outputting a group of rectangles approximating the shape of each region included in the binary input image. Area detection device. 5. A background difference generating unit which generates a background difference result image in which an absolute value of a difference between corresponding pixel values of a background image including only an input image and a background not including a moving object in each pixel is a corresponding pixel value. Means for generating a binary image composed of pixel values of 0 and 1 with reference to the background difference result image; An area detecting means for detecting a group of rectangles having an approximate shape, and referring to the group of rectangles, according to predetermined conditions,
Erasing shapes, creating new rectangles by splitting, or merging
The generation of a new rectangle by
Performs further, comprising a moving object detecting means for outputting a moving object detection result to generate a rectangle corresponding to each moving object, and detects separately the moving object included in the input image moves Object detection device. 6. A background image generating means for generating a new background image by referring to an input image and a background image, wherein each pixel corresponds to an absolute value of a difference between a corresponding pixel value between the input image and the background image. A background subtraction unit for generating a background difference result image having pixel values, a binarization unit for generating a binarized image composed of pixel values of 0 and 1 with reference to the background difference result image; a region detecting means for detecting a rectangular group of approximating the shape of the region formed by the pixel value 1 of binarized image, by referring to the rectangular group, according to a predetermined condition, rectangular
Create new rectangles by erasing or dividing, or by merging
By generating a new rectangle, the group of rectangles can be changed.
Deeds, comprising a moving object detecting means for outputting a moving object detection result to generate a rectangle corresponding to each moving object, the moving object detection and detecting separately the moving object included in the input image apparatus.