JP2006185271A - Image processing apparatus, image processing method, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing apparatus, an image processing method detecting individual objects with good precision and with ease when two or more objects exist in an image, a program and a recording medium thereof. <P>SOLUTION: The image forming apparatus is provided with an image input part 1 to which a processing subject image is inputted, a pre-processing part 2 extracting an object region occupied by the object and an outline of the object from the processing subject image, a template storing part 3 storing beforehand the subject to be detected in a processing subject image as a template, a poll part 4 carrying out polling by using the template for the outline of the object and mask processing all the poll result except the poll result matching the object region, and an analyzing part 5 detecting the object contained in the processing subject image based on the mask processed poll result. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image processing apparatus that detects a specific object from an image, and in particular, an image processing apparatus and an image that can accurately detect individual objects when a plurality of objects to be detected exist in the image. The present invention relates to a processing method, a program for causing a computer to execute the method, and a recording medium on which the program is recorded.

As a conventional image processing apparatus, for example, an apparatus provided with a moving body detection apparatus capable of detecting a moving body is known (for example, see Patent Document 1).
As shown in FIG. 16, the moving body detection apparatus includes a change area extraction unit that extracts a luminance value change area based on the luminance values of the input image from the image input unit 101 and the background image from the background image database unit 102. 103, a voting unit 104 for voting the weight of each point of the voting pattern to the parameter space corresponding to each pixel of the luminance change region based on the Hough transform, and the maximum point of the vote frequency in the parameter space A maximum point detection unit 105 that detects a candidate point, a model matching unit 106 that calculates the degree of coincidence between a luminance change region around each moving object candidate point and each moving object model, and determines the presence or absence of the moving object; The mobile unit is detected by configuring the mobile unit with the tracking unit 107 that tracks each mobile unit over time.
JP 2001-118182 A

However, in the conventional image processing apparatus, when a plurality of objects to be detected exist in an image, the accuracy of detecting individual objects may be reduced.
For example, the template 203 registered in the voting pattern database 108 stored in the image processing apparatus shown in FIG. 16 is detected by the conventional image processing apparatus from the outline 202 indicated by the dotted line in the parameter space image 201 of FIG. A method will be described.
Here, the voting unit 104 scans the image 201 and performs voting processing by Hough transform using the coordinates where the pixels of the contour 202 exist as a reference point. For example, the voting process by the Hough transform using the coordinates 204 as the reference point is performed as a voting result 205. A state in which the vote result voted on the parameter space is projected onto the parameter image 201 is indicated by a solid line in the parameter image 201.
Next, the local maximum point detection unit 105 of the image processing apparatus detects the local maximum point (coordinates most voted) on the parameter space. At this time, the coordinates of the object originally expected by the template 203 (here, the head of the moving body (person)) is the coordinates 206, and the object area to be detected is the real space image 211 in FIG. In contrast, since there are a plurality of objects in the real space image 211, the vote values of neighboring objects interfere with each other, and the maximum point becomes the coordinate 207, and the background portion The region 213 is detected.
When the moving body 214 is registered in the moving body database 109 of the image processing apparatus of FIG. 16, the model matching unit 106 calculates the degree of coincidence with the template using the coordinates 207 as a reference point, and the degree of coincidence is low. Therefore, it is determined that the object cannot be detected. Therefore, when a plurality of objects exist in the image, the voting result interferes and the voting accuracy is lowered, so that it is difficult to detect the object.

  The present invention has been made in view of the above circumstances, and an image processing apparatus, an image processing method, and a program capable of accurately and easily detecting individual objects when a plurality of objects are present in an image. It is another object of the present invention to provide a recording medium.

  The image processing apparatus of the present invention includes an image input unit that inputs a processing target image including a specific object to be detected, an object region occupied by the object, and an outline of the object from the processing target image. A processing unit, a template storage unit that preliminarily stores a target to be detected in the processing target image as a template, and voting using the template with respect to the contour of the object; A voting unit that performs mask processing on a part other than the voting result that matches the region, and an analysis unit that detects the object included in the processing target image based on the voting result subjected to the masking process. With this configuration, when a plurality of objects are included in the target image, voting is performed on the outline of the object using a template, and in this voting result, masking processing is performed in addition to the voting result that matches the object region. Detect the object. This eliminates erroneous detection of the background portion of the target image as an object, and the object can be detected with high accuracy.

  In the image processing apparatus according to the aspect of the invention, the preprocessing unit includes a unit that extracts the object region and specifies a range of the processing target image, and the voting unit masks the processing target image that specifies the range. It has a means for processing. With this configuration, the processing amount can be reduced by specifying a range from the target image and performing mask processing.

  The image processing method according to the present invention includes an image input step of inputting a processing target image including a specific object to be detected, and a preprocessing for extracting an object region occupied by the object and the contour of the object from the processing target image Voting is performed on the contour of the object using a step and a template stored as a detection target to be detected in the processing target image, and masking is performed on the voting result except for the voting result that matches the object region The method includes a voting step and an analysis step of detecting the object included in the processing target image based on the voting result subjected to mask processing. With this configuration, when a plurality of objects are included in the target image, voting is performed using the template for the outline of the object, and each object is subjected to mask processing other than the voting result that matches the object region in the voting result. Is detected. This eliminates erroneous detection of the background portion of the target image as an object, and the object can be detected with high accuracy.

  In the image processing method of the present invention, the preprocessing step extracts the object region and specifies a range of the processing target image, and the voting step masks the processing target image that specifies the range. It is a feature. With this configuration, the processing amount can be reduced by specifying a range from the target image and performing mask processing.

  The program of the present invention is characterized by causing a computer to execute each step in the above image processing method.

  The recording medium of the present invention is characterized by recording a program for causing a computer to execute each step in the above image processing method.

  According to the present invention, an area extraction unit that extracts an object region, an outline extraction unit that extracts an outline of an object, a template storage unit that stores a detection target, and the template for the outline of the object are used. The voting unit that performs voting and masks other than the voting result that matches the object region in the voting result eliminates erroneously detecting the background portion of the target image as an object, and can detect the object with high accuracy. An image processing apparatus that can be used is provided.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
[First Embodiment]
FIG. 1 shows an overall block configuration of an image processing apparatus according to the first embodiment of the present invention. The image processing apparatus according to the present embodiment stores an image input unit 1 that inputs a target image, a preprocessing unit 2 that extracts an object region and an outline of an object from an image input from the image input unit 1, and a detection target. Template storage unit 3, voting unit 4 for voting based on the template stored in template storage unit 3 from the object region and the contour of the object extracted by preprocessing unit 2, and input from this voting unit 4 And an analysis unit 5 that detects a detection target from the image using the voted result.

First, the preprocessing unit 2 of the image processing apparatus according to the present embodiment will be described with reference to FIGS.
FIG. 2 particularly shows a detailed configuration of the preprocessing unit 2. As shown in FIG. 2, the preprocessing unit 2 includes at least a frame memory 2A that holds an image input from the image input unit 1. An object region extraction unit 2B for extracting an object region from the target image input from the image input unit 1 and the image held in the frame memory 2A, and the contour of the object from the target image input from the image input unit 1 An outline extracting unit 2C for extracting is provided. In the following embodiment, it is assumed that an image 10A shown in FIG. 3A is held in the frame memory 2A, and a case where image data is input from the image input unit 1 will be described.
The object region extraction unit 2B is, for example, the background image 10A shown in FIG. 3A input from the image input unit 1 and held in the frame memory 2A and the background image 10A input from the image input unit 1 shown in FIG. An object region is extracted from the target image 10B by the background subtraction method. That is, by this background difference method, the difference in luminance value between the background image 10A and the target image 10B is calculated for each pixel, and the difference is binarized with a predetermined threshold value, and only pixels whose difference is larger than the threshold value are extracted. To do. When an object appears in the image, a large difference is generated in the pixel corresponding to the object by comparing the target image 10B and the background image 10A, and the portion shown in black in the binarized image 20 shown in FIG. An area other than the object area 20A of the binarized image 20 extracted as the object area 20A is a background area (hereinafter, this image is referred to as “object area image 20”).
The contour extraction unit 2C extracts the contour of the object from the target image 10B input from the image input unit 1 by filtering. As a method for extracting the contour of an object from an image, for example, a method using a Sobel filter is widely known, and for example, it is extracted as a contour 30A of an object shown in a binarized image 30 as shown in FIG. Is referred to as “contour image 30”).

Next, the voting unit 4 of the image processing apparatus according to the present embodiment will be described with reference to FIGS.
The voting unit 4 reads at least a template registered in the template storage unit 3 and performs voting by Hough transform on the contour image 30 input from the preprocessing unit 2 as shown in FIG. 4A, and a mask processing unit 4B that performs mask processing on the voting result obtained from the voting processing unit 4A using the object region image 20 input from the preprocessing unit 2.
Among these, the voting processing unit 4 </ b> A performs voting processing by Hough transform using the templates stored in the contour image 30 and the template storage unit 5. For example, as illustrated in FIG. 7, it is assumed that the template 3A is registered in the template storage unit 3. The x mark of the template 3A represents the center of gravity of the template, and is used as a reference point when voting using the template in the parameter space. When the voting processing unit 4A scans the contour image 30 and finds a pixel having a contour, the voting processing unit 4A votes the parameter space 40 using the coordinate value of the contour of the contour image 30 as a reference coordinate for voting. Do. For example, a vote for the reference coordinate 40A is expressed as a vote result 40B. Votes at each coordinate in the parameter space 40 are accumulated, and the more accumulated the number of votes, the higher the possibility of being the center of gravity of the template 3A. These processes are performed on all the coordinates of the contour image 30 to obtain a voting result by Hough transform.
On the other hand, the mask processing unit 4B performs mask processing on the result of voting by the voting processing unit 4A using the object region image 20 obtained by the preprocessing unit 2, and removes the voting result corresponding to the background region. For example, when only the voting result corresponding to the object region 20A (see FIG. 4) of the object region image 20 is masked from the voting result of the parameter space 40 shown in FIG. 7, it corresponds to the object region 20A in the parameter space 50 shown in FIG. The voting data is accumulated only in the area 50A.

Next, the analysis unit 5 will be described.
As shown in FIG. 9, the analysis unit 5 includes at least a local maximum point detection unit 5A and a determination unit 5B.
Among these, the local maximum point detection unit 5 </ b> A detects coordinates (hereinafter simply referred to as “coordinates”) 50 </ b> B where the number of votes is the largest from the voting results obtained by the voting unit 4. This coordinate 50B is an object candidate point to be detected.
On the other hand, the determination unit 5B determines whether or not the coordinate 50B obtained by the local maximum point detection unit 5A is a candidate point of an object to be detected. For example, only when the number of votes at the coordinate 50B is equal to or larger than a certain threshold value, it is determined as a candidate point of the object, and as shown in FIG. 10, from the target image 60 (the same as the target image 10A in FIG. 3), the object 60A Is detected.

  Therefore, according to the present embodiment, when a plurality of objects are included in the target image, each object is detected by performing mask processing with the object region image on the result of voting by the Hough transform on the contour image. This eliminates erroneous detection of the background portion of the target image as an object, and the object can be detected with high accuracy.

  In the present embodiment, the object region extraction method in the preprocessing unit 2 is not particularly limited. For example, in the present embodiment, the object region is extracted by calculating the difference in luminance value between the background image and the target image by the background difference method, but the present invention is not particularly limited thereto, and for example, other general An object region may be extracted by calculating a difference in brightness between target images that are temporally different by an inter-frame difference method that is widely known as a typical image processing method. In that case, the frame memory 2 </ b> A may hold an image that is temporally different from the target image input from the image input unit 1. In addition, the object region is extracted by dividing the target image into block units composed of a plurality of pixels by using moving image analysis by optical flow, which is widely known as another general image processing method. Also good. In the present embodiment, the target image input from the image input unit 1 and an image that is temporally different from the target image are held in the frame memory 2A, and an object region is extracted from the two images. However, the object region may be extracted from three or more images. In that case, it can be realized by preparing a plurality of frame memories 2A which are components of the preprocessing unit 2.

  In the present embodiment, the contour extraction method in the preprocessing unit 2 is not particularly limited. For example, in the present embodiment, the contour image is extracted by the contour extraction method using the Sobel filter. However, the present invention is not particularly limited to this, and the Roberts filter widely known as a general image processing method. Etc. may be used.

  Moreover, in this embodiment, the detection method of the object in the analysis part 5 is not specifically limited. For example, in this embodiment, the coordinates of the place where the number of votes is large is detected by the local maximum point detection unit 5A, and the object candidate only when the number of votes at the place where the number of votes is large is equal to or greater than a certain threshold value by the determination unit 5B. In addition to this, the following configuration may be used. That is, for example, the coordinates where the number of votes is large is detected by the local maximum point detection unit 5A, and the matching rate between the template and the target image stored in the template storage unit 5 at the coordinates where the number of votes is large is determined by the determination unit 5B. May be determined as an object candidate point only when the matching rate is equal to or greater than a certain threshold.

[Second Embodiment]
Next, a second embodiment of the present invention will be described with reference to the drawings. In the present embodiment, the same parts as those in the first embodiment are denoted by the same reference numerals to avoid redundant description.
As in the first embodiment shown in FIG. 1, the image processing apparatus according to the present embodiment includes an image input unit 1, a preprocessing unit 2, a template storage unit 3, a voting unit 4, an analysis unit 5, and the like. However, the configuration of the preprocessing unit 2 is different from that of the first embodiment.

Here, the pre-processing unit 2 will be described in detail with reference to FIG.
As shown in the figure, the preprocessing unit 2 has at least a frame memory 2A that holds the target image input from the image input unit 1, and a target image and frame memory 2A that are input from the image input unit 1. An object region extracting unit 2D that extracts an object region from an existing image, an object region dividing unit 2E that divides and extracts the object region extracted by the object region extracting unit 2D from the target image, and an object region extracting unit 2D A contour extraction unit 2F that extracts a contour of the object from the partial image of the object region by filtering. Hereinafter, in the present embodiment, it is assumed that the background image 10A illustrated in FIG. 3 is held in the frame memory 2A, and a case where the target image 10B is input from the image input unit 1 will be described.
Among these, the object region extraction unit 2D is the same as the object region extraction unit 2B shown in FIG. 2 described in the first embodiment of the present invention, and thus detailed description thereof is omitted. That is, the object region extraction unit 2D processes the background image 10A and the target image 10B to obtain the object region image 70, similarly to the object region extraction unit 2B in FIG.
On the other hand, the object area dividing unit 2E extracts the object area partial images 70A and 70B corresponding to the areas of the individual objects from the object area image 70 obtained by the object area extracting unit 2D by labeling processing. This labeling process is a process in which adjacent points or regions having the same feature are grouped with the same label, and is a widely known technique. Hereinafter, an example of processing in the object region partial image 70A extracted by the object region extraction unit 2D will be described. The same processing is performed for the object region partial image 70B.
The contour extraction unit 2F extracts the contour of the object by filtering from the object region partial image 80A (see FIG. 13) input from the object region extraction unit 2D. Since the method for extracting the object region from the image is the same as that in the first embodiment of the present invention, the description thereof is omitted. By this process, the contour partial image 80B (see FIG. 14) is extracted.

Next, the voting unit 4 according to the present embodiment will be described with reference to FIGS. 6, 14, and 15.
As in the first embodiment shown in FIG. 6, the voting unit 4 of the present embodiment applies at least the template storage of FIG. 6 to the contour partial image 80B shown in FIG. 14 input from the preprocessing unit 2. A voting processing unit 4A for reading a template registered in the unit 3 and voting by Hough transform, and an object region shown in FIG. 13 inputted from the preprocessing unit 2 with respect to a voting result obtained from the voting processing unit 4A And a mask processing unit 4B that performs mask processing using the partial image 80A. However, unlike the first embodiment of the present invention, the voting unit 4 in the present embodiment masks the object region partial image 80A cut out from the processing target image and the contour partial image 80B shown in FIG. .
The voting processing unit 4A performs voting processing by the Hough transform using the template stored in the template storage unit 3 for the contour partial image 80B. For example, it is assumed that a template 3B shown in FIG. Here, the cross mark of the template 3B represents the center of gravity of the template and is used as a reference point when voting using the template in the parameter space. When the voting processing unit 4A scans the contour partial image 80B and finds a pixel having a contour, the voting processing unit 4A votes the parameter space 90 using the coordinate value of the contour of the contour partial image 80B as a reference coordinate for voting. . For example, a vote for the reference coordinate 90A is expressed as a vote result 90B. Votes at each coordinate in the parameter space 90 are accumulated, and the more accumulated the number of votes, the higher the possibility of being the center of gravity of the template 3B. These processes are performed on all coordinates of the contour partial image 80B, and the result is a voting result by Hough transform.
The mask processing unit 4B performs mask processing on the result of voting by the voting processing unit 4A using the object region partial image 80A, and removes the voting result corresponding to the background region. For example, if only the voting result shown in black in the object region partial image 80A is masked from the voting result in the parameter space 90 shown in FIG. 15A, it corresponds to the object partial image 80A in the parameter space 100 shown in FIG. Votes are accumulated only in the parameter space 100A.

Next, the analysis unit 5 will be described.
The analysis unit 5 includes at least a local maximum point detection unit 5A and a determination unit 5B, similar to the analysis unit 5 of the first embodiment shown in FIG. Note that the analysis unit 5 of the present embodiment is different from the first embodiment of the present invention in that the parameter space 100 is processed.
The local maximum point detection unit 5 </ b> A detects the coordinates 100 </ b> B having the largest number of votes from the voting result obtained by the voting unit 4. The coordinates 100B are object candidate points to be detected in the cut out object region partial image 80A.
The determination unit 5B determines whether or not the coordinate 100B obtained by the local maximum point detection unit 5A is a candidate point of an object to be detected. For example, only when the number of votes at the coordinates 100B is equal to or larger than a certain threshold value, it is determined that the point is an object candidate point, and the coordinate system in the object region partial image 80A is converted to the coordinate system of the target image 60 shown in FIG. Then, the object 60A is detected from the target image 60 (same as the target image 10A in FIG. 3).

  Therefore, according to the present embodiment, when a plurality of objects are included in the target image, the interference of voting due to the influence of the edge of the object that is not the processing target is prevented, and the result of voting by the Hough transform on the contour image In addition, by performing mask processing on the object region image to detect individual objects, it is possible to eliminate erroneous detection of the background portion of the target image as an object, and to detect the object with high accuracy. Further, by extracting and processing the object region partial image from the target image, it is possible to reduce the processing amount of the voting process by the Hough transform as compared with the conventional image processing apparatus.

In the present embodiment, the object region extraction method in the preprocessing unit 2 is not particularly limited, as in the first embodiment. For example, in the present embodiment, the preprocessing unit 2 calculates the difference in luminance value between the background image and the target image by the background subtraction method and extracts the object region, but is not particularly limited. That is, an object region may be extracted by calculating a difference in luminance value between temporally different target images by an inter-frame difference method that is widely known as another general image processing method. In that case, an image that is temporally different from the target image input from the image input unit 1 may be held in the frame memory 2A. In addition, the object region is extracted by dividing the target image into block units composed of a plurality of pixels by using moving image analysis by optical flow, which is widely known as another general image processing method. Also good.
In the present embodiment, similarly to the first embodiment, the target image input from the image input unit 1 and an image temporally different from the target image are held in the frame memory 2A, and two images are stored. Although the object region is extracted from the object region, the object region may be extracted from three or more images. In this case, it is possible to realize this by preparing a plurality of frame memories 2A that are components of the preprocessing unit 2.
In the present embodiment, as in the first embodiment, the contour extracting method in the contour extracting unit 2F is not particularly limited. For example, in the present embodiment, the contour image is extracted by the contour extraction method using the Sobel filter. However, a Roberts filter that is widely known as a general image processing method may be used.
In the present embodiment, the object detection method in the analysis unit 5 is not particularly limited, as in the first embodiment. For example, in the present embodiment, the maximum point detection unit 5A detects the coordinates where the number of votes is large, and the determination unit 5B detects the coordinates of the object only when the number of votes at the coordinates where the number of votes is large is equal to or greater than a certain threshold. Although it is determined as a candidate point, for example, a coordinate and a target stored in the template storage unit 5 at a coordinate where the vote count is large is detected by the maximum point detection unit 5A, and the number of votes is determined by the determination unit 5B. The coincidence rate with the image may be calculated, and the object candidate point may be determined only when the coincidence rate is equal to or greater than a predetermined threshold.

  INDUSTRIAL APPLICABILITY The image processing device and the image processing method of the present invention have an effect of preventing the background portion of the target image from being erroneously detected as an object, and can be useful as a security device for an urban area or an important facility. .

1 is a block diagram showing the overall configuration of an image processing apparatus according to a first embodiment of the present invention. The block diagram which shows the structure of the pre-processing part in the image processing apparatus (A) And (B) is explanatory drawing which respectively shows an example of the target image and background image in the 1st Embodiment of this invention. Explanatory drawing which shows the object area | region image in the 1st Embodiment of this invention. Explanatory drawing which shows the outline image in the 1st Embodiment of this invention. 1 is a block diagram showing a configuration of a voting unit of an image processing apparatus according to a first embodiment of the present invention. Explanatory drawing which shows a state when voting by the voting process part of the voting part of the image processing apparatus which concerns on the 1st Embodiment of this invention. Explanatory drawing which shows the state at the time of masking with the mask processor of the voting part of the image processing apparatus which concerns on the 1st Embodiment of this invention. 1 is a block diagram showing a configuration of an analysis unit of an image processing apparatus according to a first embodiment of the present invention. Explanatory drawing which shows the state after detecting an object from the target image of the image processing apparatus which concerns on the 1st Embodiment of this invention. The block diagram which shows the structure of the pre-processing part of the image processing apparatus which concerns on the 2nd Embodiment of this invention. Explanatory drawing which shows the state of a process of the object area division part of the image processing apparatus which concerns on the 2nd Embodiment of this invention. Explanatory drawing which shows the object area | region partial image of the image processing apparatus which concerns on the 2nd Embodiment of this invention. Explanatory drawing which shows the outline partial image of the image processing apparatus which concerns on the 2nd Embodiment of this invention. (A) And (B) is explanatory drawing which shows the outline | summary of the process of the voting part of the image processing apparatus which concerns on the 2nd Embodiment of this invention, and an analysis part. The block diagram which shows the whole structure of the conventional image processing apparatus (A) And (B) is a parameter space image figure and a real space image figure which show the outline | summary of the process of the conventional image processing apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image input part 10A Background image 10B Target image 2 Preprocessing part 2A Frame memory 2B, 2D Object area extraction part 2C, 2F Contour extraction part 2E Object area division part 20 Binary image 20A Object area 3 Template storage part 30 2 Binary image (contour image)
30A Object outline 4 Voting unit 4A Voting processing unit 4B Mask processing unit 40, 50 Parameter space 40A Normal coordinate 40B Voting result 5A Local maximum point detection unit 5B Determination unit 50A Area corresponding to object region 50B Coordinate with the largest number of votes 60 Target image 60A Object

Claims (6)

An image input unit for inputting a processing target image including a specific object to be detected;
A pre-processing unit that extracts an object region occupied by the object and a contour of the object from the processing target image;
A template storage unit that pre-stores a target to be detected in the processing target image as a template;
A voting unit that performs voting on the outline of the object using the template, masks other than the voting result that matches the object region in the voting result, and
An image processing apparatus comprising: an analysis unit that detects the object included in the processing target image based on the voting result subjected to the mask processing. The pre-processing unit has means for extracting the object region and specifying a range of the processing target image,
The image processing apparatus according to claim 1, wherein the voting unit includes means for performing mask processing on the processing target image whose range is specified. An image input step for inputting a processing target image including a specific object to be detected;
A preprocessing step of extracting an object region occupied by the object and an outline of the object from the processing target image;
A voting step that performs voting on the outline of the object using a template stored as a detection target to be detected in the processing target image, and masks other than the voting result that matches the object region in the voting result; ,
And an analysis step of detecting the object included in the processing target image based on the voting result subjected to mask processing. The preprocessing step identifies the range of the processing target image by extracting the object region,
The image processing method according to claim 3, wherein the voting step masks the processing target image whose range is specified.   The program which makes a computer perform each step in the image processing method of Claim 3 or 4.   A recording medium on which a program for causing a computer to execute each step in the image processing method according to claim 3 or 4 is recorded.

Images