JP2020017110A - Object detection system - Google Patents

Abstract

To provide an object detection system that can appropriately detect a target object even in a search target image in which the object is expanded due to halation or the like.SOLUTION: An object detection system according to the present invention compares a "processed template image" with a "search target image" to detect an object from the search target image, and is a system including processed template image creation means. The processed template image creation means is means for creating a processed template image in which a "specific color" is expanded or contracted from a standard template image. The system performs matching processing between the processed template image and the search target image to detect the object from the search target image.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a technology for detecting a target object (object) from an image, and more specifically, an object detection system for detecting an object using a template image obtained by expanding a shape of a specific color. It is about.

  When detecting a target object such as a surveying sign (for example, an anti-aircraft sign) from an image photographed over a relatively wide area such as an aerial photograph or a satellite photograph, detection by human eyes may cause omission of detection and requires considerable labor. For this reason, automatic detection using a computer has become mainstream. For example, an image obtained by photographing a general surveying sign from above is prepared as a template image, and the template image is compared with an aerial photograph or the like, whereby template matching for detecting an object (in this case, a sign for surveying) is known. Have been.

  Patent Literature 1 discloses an invention in which an existing solar panel (photovoltaic power generation device) is detected by collating a template with an aerial photograph or the like.

JP 2013-186719 A

  By the way, when an object having a white portion such as a survey mark is detected, it may not be possible to properly detect the object by pattern matching. 2. Description of the Related Art It is known that, in an image photographed from the sky such as an aerial photograph, a white portion causes halation of light, so that the region becomes larger than the actual region. On the other hand, in the template image prepared in advance for template matching, the white portion is not expanded since the object is generally photographed in a favorable environment. Therefore, even if the template image is compared with the aerial photograph in which the white portion is expanded, the object is not properly detected.

  Therefore, an aerial photograph, which is an image on the searched side (hereinafter, referred to as a "searched image"), may be subjected to image processing (for example, contraction processing for a white portion) and then pattern matching. However, not all objects in the searched image cause halation in the same manner. For example, in FIG. 10, many images having the same size and shape are contained in the image, but only the central one (encircled by an ellipse) is expanded by halation. Therefore, even if image processing is performed on the searched image, it is difficult to prevent omission of object extraction by pattern matching.

  An object of the present invention is to solve the conventional problem, that is, to provide an object detection system that can appropriately detect a target object even in a search target image in which an object is expanded due to halation or the like. is there.

  The present invention focuses on performing image processing on a template image instead of an image to be searched, and detecting an object using a template image obtained by expanding or contracting a specific color (for example, white). This is an invention based on an idea that has not existed in the past.

  The object detection system according to the present invention detects an object from a searched image by comparing a “processed template image (an image created based on a standard template image of an object)” with a “searched image”. And a processing template image creating means. The processing template image creating means is means for creating a processing template image in which a “specific color” is expanded or contracted from the standard template images. Then, an object is detected from the search image by performing a matching process between the processed template image and the search target image.

  The object detection system of the present invention may further include a candidate detection unit and an object selection unit. Among these, the candidate detecting means is means for detecting two or more “object candidates” from the searched image by performing a matching process between the processed template image and the searched image. This is a means for selecting a target object. In this case, the processing template image generating means generates two or more types of processing template images in which the specific color is gradually expanded or contracted. The candidate detecting means detects object candidates based on two or more types of processed template images.

  The object detection system of the present invention may be a system in which the processing template image generating means generates a “color-change processing template image”. In this case, the processing template image creating means creates two or more types of different color processing template images having different specific colors, and also expands or contracts the specific color step by step for each of the different color processing template images. To create a processing template image.

  The object detection system of the present invention may further include a binary processed template image creating unit and a binary searched image creating unit. Among these, the binary processed template image creating means is a means for creating a binary processed template image obtained by binarizing the processed template image, and the binary searched image creating means converts the searched image into a binary image. This is a means for creating a binary searched image. In this case, an object is detected from the binary searched image by performing a matching process between the binary processed template image and the binary searched image.

  The object detection system according to the present invention includes candidate detection means for detecting two or more object candidates from the binary searched image by performing a matching process between the binary processed template image and the binary searched image. You can also. In this case, the binary search target image creating means creates two or more types of binary search target images by changing the binary boundary. Then, the candidate detecting means detects an object candidate based on the two or more types of binary searched images, and the object selecting means selects a target object from these object candidates.

The object detection system of the present invention has the following effects.
(1) A target object can be appropriately detected even in a searched image in which an object is expanded (or contracted) due to halation or the like.
(2) By using a plurality of types of processed template images obtained by gradually expanding (or contracting) a shape having a specific color, a target object can be further appropriately detected.
(3) By using the processing template images of different colors having specific colors different from each other, various objects can be appropriately detected.

The model figure which shows the standard template image (upper part) of the anti-aircraft sign, and the processed template image (lower part) created based on this. FIG. 4 is a model diagram showing a standard template image (upper part) of a CD (Compact Disc) and a processed template image (lower part) created based on the standard template image. FIG. 1 is a block diagram showing a main configuration of a monitoring system according to the present invention. FIG. 3 is a step diagram showing main processing of the object detection system of the present invention. FIG. 5 is a flowchart showing details of a processing template image creation process in the object detection system. FIG. 6 is a flowchart showing details of a search target image creation process in the object detection system. FIG. 4 is a flowchart showing details of an object detection process in the object detection system. FIG. 3A is a model diagram illustrating a reference color pixel included in a binary processed template image, and FIG. 3B is a model diagram illustrating a special region. (A) is an image to be searched before binarization processing is performed; (b) is an image in which a special region is removed from a binary processed template image with a boundary value of 200 (gray scale); (c) FIG. 4 is an image diagram in which a special region is removed from a binary processed template image with a boundary value of 235 (gray scale). FIG. 9 is an image diagram showing a state where only a part of many similar objects has expanded due to halation.

  An example of the object detection system of the present invention will be described with reference to the drawings.

1. Definitions In describing an example of an embodiment of the present invention, first, definitions of terms used herein are shown.

(Template image and searched image)
When a target (search target) object is detected from an image by template matching, a standard image (a so-called model image) of the object is prepared. Here, the standard image of the object is referred to as a “template image”, and the image searched by the template image is referred to as a “searched image”. For example, when trying to detect an anti-aircraft sign from an aerial photograph, the anti-aircraft sign is an object, a standard image of the anti-aircraft sign is a template image, and an aerial photograph is a search target image.

(Standard template image and processed template image)
The present invention is characterized in that a matching process is performed using an image obtained by processing a template image. Specifically, a predetermined specific color (hereinafter, simply referred to as “specific color”) of a normal template image (a standard image of an object) is expanded or reduced, and an image obtained as a result is used. To search for the searched image. Here, a normal template image (a standard image of an object) is particularly referred to as a “standard template image”, and an image created based on this standard template image is referred to as a “processed template image”. I do.

  FIG. 1 is a model diagram showing a standard template image of an anti-aircraft sign (upper part) and a processed template image (lower part) created based on the standard template image. FIG. ) And a model diagram showing a processed template image (lower part) created based on this. In these drawings, an example is shown in which white is set as a specific color, and a processed template image is created by expanding the white of the standard template image. Of course, not limited to this example, the processed template image can be created based on a specific color other than white, such as black, or the processed template image can be created by contracting a specific color of the standard template image. . For processing for expanding or processing a specific color, a conventionally used technique can be used. For example, a processed template image can be created using white expansion processing of OpenCV.

  FIGS. 1 and 2 show three types of processed template images. These are created by expanding a specific color (white in the figure) in a stepwise manner. The processing template image at the left end is the smallest and the specific color is expanded (intensity 1). The image is a medium in which the specific color is expanded to a medium level (intensity 5), and the rightmost processed template image is the one in which the specific color is expanded to the maximum (intensity 10).

  Also, by changing the specific color, a plurality of types of processed template images can be created. In FIG. 1, the processed template image is created by expanding white (specific color). However, the processed template image is created by setting black as a specific color and expanding (or contracting) black. The processed template image obtained by changing the specific color with respect to the same standard template image in this manner is referred to as a “different-color processed template image” for convenience.

(Binary processing template image and binary searched image)
In the present invention, the search target image can be searched for using the processing template image in its original state (without further processing), and the search target image can be searched for using an image obtained by binarizing the processing template image as described later. You can also search for For convenience, here, the processed template image subjected to the binarization processing is particularly referred to as “binary processed template image”, and the processed template image before the binarization processing is particularly referred to as “primary processed template image”. And

  Further, in the case of searching with a binary processed template image, a binarization process can be performed on the searched image. In other words, the search target image subjected to the binarization processing is searched for using the binary processed template image. For convenience, the image to be searched that has been binarized is referred to as a “binary image to be searched”. Further, a threshold value set for separating into binary values in the binarization process is referred to as a “boundary value” here. Here, the binarization processing indicates that an image is represented by two predetermined gradations, and is represented by various combinations of two gradations including white and black.

2. Object Detection System FIG. 3 is a block diagram showing a main configuration of the object detection system 100 of the present invention. As shown in this figure, the object detection system 100 of the present invention includes a modified template image creating means 101, and further includes a standard template image creating means 102, a standard template image storage means 103, a modified template image resizing means 104, Value processed template image creating means 105, binary processed template image storing means 106, searched image storing means 107, binary searched image creating means 108, special area removing means 109, binary searched image storing means 110, candidate detection It can also be configured to include means 111, an object selecting means 112, and output means such as a display and a printer.

  In the object detection system 100, the modified template image creating means 101 and the standard template image creating means 102, the modified template image resizing means 104, the binary modified template image creating means 105, the binary searched image creating means 108, the special area removing means 109 , The candidate detecting means 111 and the object selecting means 112 can be manufactured as dedicated ones, or a general-purpose computer device can be used. This computer device can be configured by a personal computer (PC), a tablet PC such as an iPad (registered trademark), a portable terminal including a smartphone, a PDA (Personal Data Assistance), or the like. The computer device includes a processor such as a CPU, a memory such as a ROM and a RAM, and further includes an input unit such as a mouse and a keyboard and a display. Further, the standard template image storage unit 103, the binary processed template image storage unit 106, the search target image storage unit 107, and the binary search target image storage unit 110 can be constructed in, for example, a database server, and can be connected to a local network : Local Area Network) or a cloud server that stores the data via the Internet (that is, wireless communication).

  As shown in FIG. 4, the object detection system 100 creates a processing template image (including a primary processing template image, a binary processing template image, and a different color processing template image) from the standard template image (Step 100). , A process of creating a searched image (including a binary searched image) (Step 200), and a process of detecting an object from the processed template image and the searched image (Step 300). Hereinafter, each process will be described in detail. 4 is a step diagram showing main processing of the object detection system 100 according to the present invention, FIG. 5 is a flowchart showing details of a processing template image creation process (Step 100) in the object detection system 100, and FIG. FIG. 7 is a flowchart showing details of the image creation processing (Step 200), and FIG. 7 is a flowchart showing details of the object detection processing (Step 300).

(Creating a processing template image)
In creating the processed template image, first, the standard template image creating means 102 creates a standard template image. In general, a template image is created by various methods such as a scanner and trimming after photographing. Therefore, in many cases, the size of a pixel (pixel) is not determined, and as a result, the influence of an alias is not constant. Therefore, in order to unify the influence of the alias and the like, it is preferable that an image once resized to a predetermined pixel size (for example, 100 × 100 pixels) is used as a standard template image. The standard template image created by the standard template image creation unit 102 is stored in the standard template image storage unit 103.

  Next, the processing template image creating means 101 reads the standard template image from the standard template image storing means 103, and creates a primary processing template image from the standard template image (Step 101 in FIG. 5). At this time, when a plurality of types (three types in FIGS. 1 and 2) of the processed template image are created by expanding or contracting a specific color stepwise (while changing the intensity), it is necessary to repeat the process for all the predetermined intensities. The next processing template image creation processing (Step 101) is performed (Step 102). Similarly, when a plurality of types of differently-colored processed template images are created while changing the specific color, the primary processed template image creation processing (Step 101) is repeatedly performed with all the planned specific colors (Step 103).

  When the primary processing template image can be created, the primary processing template image may be resized by the processing template image resizing means 104 (Step 104 in FIG. 5). That is, the size of the object in the primary processing template image is adjusted to be somewhat the same as the size of the object in the search target image. Specifically, “the actual size value (mm) / resolution (mm / pixel) of the primary processing template image” may be calculated, and the primary processing template image may be resized based on the calculated value.

  As described above, according to the present invention, the matching process can be performed using the primary processed template image, or the matching process can be performed using the binary processed template image. When performing the matching process using the binary processed template image, the binary processed template image generating unit 105 generates the binary processed template image (Step 105 in FIG. 5). Specifically, the binary-processed template image creating unit 105 creates a binary-processed template image by performing a binarization process on the resized primary processed template image. The binary processing template image created here is stored in the binary processing template image storage unit 106.

(Creating an image to be searched)
The search target image is created by cutting out a predetermined range (a range in which an object is to be detected) from an aerial photograph taken in a wide range (Step 201 in FIG. 6). Of course, if the prepared aerial photograph or the like is in a range in which an object is to be detected or is in an equivalent range, it can be used without cutting out a predetermined range.

  When performing the matching process using the binary processed template image, it is preferable to use the binary searched image as the image on the searched side. The binary searched image is created by the binary searched image creating means 108. Specifically, the binary searched image creating means 108 converts the searched image read out from the searched image storing means 107 into a binary image. A binary search target image is created by performing the processing (Step 202 in FIG. 6). At this time, a plurality of types of binary processed template images can be created using different boundary values. In this case, binarization of the image to be searched (Step 202) is repeatedly performed with all the planned boundary values (Step 203).

  By the way, an area (hereinafter, referred to as “special area”) that is clearly different from the binary processed template image may be included in the binary searched image. It is preferable to perform the matching process after removing the special region in advance, because the detection accuracy is improved and the calculation time is shortened. The process of removing the special region from the binary searched image (Step 204 in FIG. 6) is performed by the special region removing unit 109, and the binary searched image from which the special region has been removed is stored in the binary searched image storage unit 110. Is stored. Hereinafter, the process of removing the special area from the binary search target image will be described in detail.

  FIGS. 8A and 8B are model diagrams for explaining a special region. FIG. 8A is a model diagram showing a binary processing template image, and FIG. 8B is a model diagram showing a special region. The special region is a region having the same image size as the binary processed template image and including a “reference color pixel” exceeding the “upper limit number”. Here, the reference color pixel is one of the predetermined colors (gradations) of the binary (two gradations) constituting the binary processed template image, and the upper limit number is the binary processed template image. Is a threshold value set based on the total number of reference color pixels included in. For example, the binary processed template image shown in FIG. 8A has a pixel size defined by 10 × 10 pixels, and has 60 reference color pixels (here, white pixels). On the other hand, the image shown in FIG. 8B has the same image size (10 × 10 pixels) as the binary-processed template image, and has 90 images exceeding the upper limit of 72 (60 × 1.2). (Here, white pixels). Therefore, the image shown in FIG. 8B is recognized as a special area, and is removed from the binary search target image by the special area removing unit 109. In this example, the upper limit number of the reference color pixels is set to 1.2 times (ie, 72) the total number (60) of the reference color pixels included in the binary processed template image. The upper limit number of reference color pixels can be appropriately designed, for example, twice as large as the total number of reference color pixels included in the binary processed template image.

  When a plurality of types of binary processed template images are created using different boundary values, a special region removal process (Step 204) may be performed on each of the binary processed template images. FIG. 9A shows a search target image before binarization processing (for convenience, the image is represented by binary values, but is actually an image with a predetermined color), and FIG. An image obtained by removing a special region from a binary processed template image (white and black binary image) set to (gray scale), and (c) is a binary processed template image (235 scaled to gray scale) This is an image obtained by removing a special area from a binary image (white and black). As shown in this figure, all of the binary processed template images from which the special regions have been removed have a limited number of reference color pixels (here, white pixels), which facilitates the matching process using the binary processed template images. You can see that there is.

(Object detection)
When the processed template image and the search target image are prepared, a target object is detected by performing template matching processing using these. When a plurality of types of processed template images are prepared or a plurality of types of searched images are prepared, a plurality of object candidates (hereinafter, simply referred to as “object candidates”) are detected by the candidate detecting unit 111 and objects are selected. The means 112 may select the most suitable one from these object candidates as the object. Hereinafter, the process of detecting an object candidate by the candidate detecting unit 111 and the process of selecting an object by the object selecting unit 112 will be described in detail with reference to FIG.

  First, the candidate detecting unit 111 performs a matching process using a combination of all prepared template images and all prepared images to be searched (Step 301 to Step 303). Then, the candidate detecting unit 111 detects a matching object in each combination (the processed template image and the searched image) as an object candidate (Step 304). When object candidates are detected for all combinations, the object selecting means 112 selects the most suitable object from these object candidates as an object (Step 305). At this time, the object having the largest number of object candidates indicating the same place is selected as an object, the object having the highest degree of matching is selected as an object, and the number indicating the same place and the degree of matching are comprehensively determined. An object can be selected from the object candidates by various methods such as selecting an object.

  The object detection system of the present invention can be used to detect a specific facility or a specific structure such as a manhole cover in addition to a survey sign including an anti-aircraft sign. Can be used to detect various objects from photos from

REFERENCE SIGNS LIST 100 object detection system 101 processing template image creation means (of object detection system) 102 standard template image creation means (of object detection system) 103 standard template image storage means (of object detection system) 104 processing template image (of object detection system) Resize means 105 Binary processing template image creation means (of the object detection system) 106 Binary processing template image storage means 107 (of the object detection system) Searched image storage means 108 (of the object detection system) 108 (of the object detection system) Value searched image creating means 109 Special area removing means (of the object detection system) 110 Binary searched image storage means (of the object detection system) 11 1 candidate detection means (of the object detection system) 112 object selection means (of the object detection system)

Claims (6)

In an object detection system that detects the object from the searched image by comparing the template image of the object with the searched image,
The template image includes a processed template image created based on a standard template image,
Processing template image creating means for creating the processing template image obtained by expanding a specific color of the standard template image,
By performing a matching process between the processed template image and the searched image, the object is detected from the searched image,
An object detection system, characterized in that: Candidate detecting means for detecting two or more object candidates from the searched image by performing a matching process between the processed template image and the searched image;
An object selection unit that selects the target object from the object candidates,
The processing template image creating means creates two or more types of the processing template images obtained by expanding the specific color stepwise,
The candidate detecting means detects object candidates based on two or more types of the processed template images, respectively.
2. The object detection system according to claim 1, wherein: The processing template image creating means creates two or more types of different color processing template images different in the specific color, and expands the specific color step by step for each of the different color processing template images. Create a processing template image,
3. The object detection system according to claim 2, wherein: A binary processing template image creating means for creating a binary processing template image by binarizing the processing template image;
A binary searchable image creating means for creating a binary searchable image by binarizing the searchable image,
Performing a matching process between the binary processed template image and the binary searched image to detect the object from the binary searched image;
The object detection system according to claim 1, wherein: A special area removing unit that removes, from the binary search target image, a special area that has the same image size as the binary processed template image and includes a reference color pixel exceeding an upper limit number;
The reference color pixel is a pixel having a color selected from two colors constituting the binary image, and the upper limit number is equal to the total number of the reference color pixels included in the binary processed template image. Set based on
By performing a matching process between the binary processed template image and the binary searched image from which the special area has been removed, the object is detected from the binary searched image.
The object detection system according to claim 4, wherein: Candidate detecting means for detecting two or more object candidates from the binary searched image by performing a matching process between the binary processed template image and the binary searched image;
An object selection unit that selects the target object from the object candidates,
The binary searched image creating means creates two or more types of the binary searched image by changing a binary boundary,
The candidate detecting means detects an object candidate based on two or more types of the binary searched images, respectively.
The object detection system according to claim 4 or 5, wherein

Images