JP6758263B2 - Object detection device, object detection method and object detection program - Google Patents

Description

The present invention relates to a technique for detecting a columnar or spherical object from an image.

Currently, in the field of sports competition and surveillance, the position of an object existing in real space is specified from an image taken by a camera, and the position of the object is tracked. Specifically, research and development is being conducted on a technique for detecting a cylindrical or spherical object existing in space from at least one or more images input from a camera that images real space, and outputting the information of the object to a screen. It is being developed.

For example, Non-Patent Document 1 discloses a technique for detecting a cylindrical object arranged so that the normal direction of the bottom surface is parallel to the normal line of the floor plane by using the Mean shift method of Non-Patent Document 2. ing. In the Mean shift method, a feature amount is extracted from a designated area that serves as a template, a region having a feature amount similar to this is detected from the image currently being processed, and a position corresponding to the template is set as a cylindrical object from the image being processed. Estimate as. A histogram created from the color information in the designated area is often used as the feature amount. However, erroneous detection may occur when a pattern of a similar color exists in the background or the like. Therefore, in Non-Patent Document 1, highly accurate detection of a cylindrical object is realized by using a histogram created by adding edge information to color information.

J. Kim, 3 outsiders, “Curling Stone Tracking by an Algorithm Using Appearance and Color Features”, Proceedings of the World Congress on Electrical Engineering and Computer Systems and Science (EECSS 2015), Paper No. 334, July 2015 D. Comaniciu, 2 outsiders, “Real-time Tracking of Non-rigid Objects Using Mean Shift”, Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2015, p.142-p.149

In the conventional method, a histogram created from the edge information in the designated area is used to detect a cylindrical object arranged so that the normal direction of the bottom surface is parallel to the normal direction of the floor plane.

However, when a shape pattern similar to the cylindrical object to be detected exists in the background, there is a problem that the detection target is erroneously detected by using only the edge information. In addition, since geometric information is lost in the histogram representation, the histogram generated from an object having a shape pattern other than the cylindrical shape existing around it can be detected by making it similar to the one created from a cylindrical object. There is also a problem that false detection occurs.

Another problem is that if the orientation of the camera that took the current image is different from the orientation of the camera when the template image was taken, the orientation and strength of the edges extracted from the cylindrical object will be between the two. Since they are different, there is a problem that detection omission of a cylindrical object occurs. When a cylindrical object is photographed so that the direction of the cylindrical axis and the optical axis of the camera match, it is observed as a perfect circle on the captured image, but it is observed as an ellipse on the image captured at an angle, and the edge The size is much different than a perfect circle. If the former is used as a template and the latter is used as an input image, erroneous detection of a cylindrical object will occur.

The present invention has been made in view of the above circumstances, and when there is a shape pattern similar to the shape of a cylinder or sphere as a subject in the background or surroundings, or when the normal direction of the bottom surface of the cylinder and the optical axis of the camera are different. Even in the case of an input image taken from a direction that does not match, a cylinder that exists on a plane in which the normal direction of the bottom surface existing in space is known and parallel to the bottom surface from at least one or more images of the subject. The purpose is to accurately detect a body or sphere and output information on that object.

In order to solve the above problems, the object detection device according to claim 1 is an object detection device that detects a cylindrical or spherical object from an image, and is an original image taken from an oblique angle with respect to a plane in space. Is designated from the foreground region, a projection conversion unit that casts and transforms the front image taken from an angle perpendicular to the plane, and a foreground extraction unit that extracts the foreground region located in front of the plane from the front image. A designated color extraction unit that extracts a designated color region that matches the color, a contour detection unit that approximates the contour surrounding the designated color region with an ellipse, and a determination whether or not the ellipse is circular, and the ellipse is circular. In the case of, it is characterized by including a shape verification unit that outputs the ellipse as a cylindrical or spherical object.

The object detection device according to claim 2 is the object detection device according to claim 1, wherein the position of the ellipse is inversely converted to the position of the coordinate system before the projection conversion, and the plane of the cylinder or sphere forming the ellipse. It is characterized by further including a detection position correction unit that returns to the position of the coordinate system after the projection conversion after adding a predetermined offset value based on the height from.

The object detection method according to claim 3 is an object detection method for detecting a cylindrical or spherical object from an image, in which a computer captures an original image taken from an oblique angle with respect to a plane in space on the plane. A step of projecting and converting to a front image taken from a vertical angle, a step of extracting a foreground region located in front of the plane from the front image, and an extraction of a designated color region matching a designated color from the foreground region. Steps to be performed, a step of approximating the contour surrounding the designated color region with an ellipse, and determining whether or not the ellipse is circular, and if the ellipse is circular, the ellipse is output as a cylindrical or spherical object. It is characterized by performing steps and performing.

The object detection program according to claim 4 is characterized in that the computer functions as the object detection device according to claim 1 or 2.

According to the present invention, a cylinder or a sphere can be accurately detected from an image.

It is a figure which shows the example of the real space and the input image used in each embodiment. It is a figure which shows the functional block composition of the object detection apparatus. It is a figure which shows the flowchart of the object detection process. It is a figure which shows the flowchart of the projection conversion processing. It is a figure which shows the example of the input image. It is a figure which shows the example of the background image. It is a figure which shows the example of the background geometric transformation image. It is a figure which shows the example of the input geometric transformation image. It is a figure which shows the flowchart of the foreground extraction processing. It is a figure which shows the example of the foreground area image. It is a figure which shows the flowchart of the designated color extraction process. It is a figure which shows the example of the designated color extraction image. It is a figure which shows the flowchart of the contour detection process. It is a figure which shows the example of the contour extraction image. It is a figure which shows the example of the ellipse approximation image. It is a figure which shows the flowchart of the shape verification process. It is a figure which shows the example of the deviation of the center position between the coordinate systems by the height of a cylindrical object. It is a figure which shows the functional block composition (second embodiment) of the object detection apparatus. It is a figure which shows the flowchart (second embodiment) of a projective transformation process. It is a figure which shows the flowchart of the detection position correction processing. It is a figure which shows the example of the offset value. It is a figure which shows the conversion example of the position coordinate from a projective conversion image coordinate system to an input image coordinate system. It is a figure which shows the conversion example of the position coordinate from the input image coordinate system to the projective conversion image coordinate system.

As described above, an object of the present invention is to improve the detection accuracy of a cylinder or a sphere from an image. Therefore, in the present invention, an original image taken from an angle oblique to a plane in space is projected and converted into a front image taken from an angle perpendicular to the plane, and the front image is projected to the front of the plane. The foreground area located in is extracted, the specified color area matching the specified color is extracted from the foreground area, the outline surrounding the specified color area is approximated by an ellipse, and it is determined whether or not the ellipse is circular. When the ellipse is circular, the ellipse is output as a cylindrical or spherical object.

That is, in the present invention, since the original image taken from an oblique angle with respect to the plane in the space is projected and converted into the front image, the shape of the subject can be grasped in the original shape, and the cylinder to be detected. It is possible to suppress omission of detection of a body or a sphere and reliably detect it. Further, in the present invention, a foreground region is extracted from the front image, an object is detected from the foreground region using color and contour information, and it is verified whether or not the shape of the detected object is circular. It is possible to suppress excessive detection in which an object other than a body or a sphere is also detected. As a result, the accuracy of detecting a cylinder or a sphere from an image can be improved.

Hereinafter, the object detection device, the object detection method, and the object detection program according to the embodiment will be described with reference to the drawings. In each embodiment described later, in the real space of the coordinate system having the x'-y'plane as the floor surface (x', y', z') as shown in FIG. 1, the real space is the real space. From the input image taken by the camera 3 in which the z'axis and the optical axis A do not match, the normal direction of the bottom surface of the cylinder in the real space is the positive direction of the z'axis vertically upward with respect to the floor surface, and the cylinder. Cylindrical objects 5a, 5b, and 5c existing on the floor surface, which is a plane parallel to the bottom surface, are detected as target objects.

For example, it can be used to detect stones on ice used in curling competitions. In addition to the cylindrical object, a sphere used in billiard competition or gateball competition may be the target object.

The input image can be rephrased as an original image taken from an oblique angle with respect to the x'-y'plane in the real space.

<First Embodiment>
FIG. 2 is a diagram showing a functional block configuration of the object detection device 1 according to the first embodiment. The object detection device 1 is a device that detects a cylindrical or spherical object from an image, and as shown in FIG. 2, has a projection conversion unit 11, a foreground extraction unit 12, a designated color extraction unit 13, and a contour. It is configured to include a detection unit 14, a shape verification unit 15, and a memory 16.

The projective transformation unit 11 inputs at least one or more input images, calculates a projective transformation matrix, geometrically transforms a predetermined background image using the calculated projective transformation matrix, and stores the background geometrically transformed image in the memory 16. It is a functional unit that records and geometrically transforms an input image using the projective transformation matrix and outputs it as an input geometrically transformed image.

The foreground extraction unit 12 inputs the input geometric transformation image and the background geometric transformation image, and is located in front of the x'-y'plane by background subtraction processing using the input geometry transformation image and the background geometry transformation image. This is a functional unit that extracts the foreground area and outputs the extracted foreground area as a foreground area image.

The designated color extraction unit 13 inputs a foreground region image, inputs predetermined color information and a color range, extracts a designated color region similar to the predetermined color within the color range from the foreground region image, and specifies a designated color. This is a functional unit that outputs as an extracted image.

The contour detection unit 14 inputs a designated color extracted image, extracts a contour surrounding each designated color region over all the designated color regions included in the designated color extracted image, and approximates the extracted contour with an ellipse. It is a functional part that outputs as ellipse information.

The shape verification unit 15 inputs all the ellipse information, inputs a predetermined roundness threshold value, and performs roundness threshold value processing on each ellipse information to determine whether or not the ellipse being processed is close to a circle. Is determined, and when the ellipse being processed is close to a circle, it is a functional unit that outputs ellipse information of the ellipse as object information.

Next, the object detection process performed by the object detection device 1 will be described with reference to the flowchart of FIG.

Step S101;
First, the projective conversion unit 11 is activated. The projective conversion process performed by the projective conversion unit 11 will be described with reference to the flowchart of FIG.

Step S101a;
At least one or more images are input to the projective conversion unit 11. In this embodiment, for example, the input image shown in FIG. 5 is input. The projective conversion unit 11 inputs predetermined four corresponding points when the activation is the first time. In the present embodiment, as corresponding points, for example, (x ₁ , y ₁ ) → (x ′ ₁ , y ′ ₁ ), (x ₂ , y ₂ ) → (x ′ ₂ , y ′ ₂ ), (x ₃ ). _{, y 3) → (x '} 3, y' 3), are designated _{(x 4, y 4) →} (x '4, y' 4). The symbol “→” indicates the correspondence between the real space and the input image. _{(X '1, y' 1} ), (x '2, y' 2), (x '3, y' 3), (x '4, y' 4) , _{respectively,} (x _{1, y} 1) , (X ₂ , y ₂ ), (x ₃ , y ₃ ), (x ₄ , y ₄ ) in real space, and points on a plane parallel to the bottom surface of the cylinder.

Step S101b;
Next, the projective transformation unit 11 calculates the projective transformation matrix H. The projective transformation matrix is also called homography, and is a matrix representing the geometric transformation from the reference image to the subject in the input image when the subject is a plane, and is represented by, for example, a 3 × 3 matrix. The projective transformation matrix can be obtained if there are four or more corresponding points (points that are the same) in the recognition target and the reference image in the input image. The projective transformation unit 11 calculates the projective transformation matrix H using the above-mentioned four corresponding points.

Steps S101c, S101d;
After that, the projective transformation unit 11 performs geometric transformation of a predetermined background image using the calculated projective transformation matrix H, and records it in the memory 16 as a background geometric transformation image. In the present embodiment, the background image shown in FIG. 6 is given in advance, and it is assumed that the background geometric transformation image shown in FIG. 7 is obtained by using the projective transformation matrix H.

Steps S101e, S101f;
Finally, the projective transformation unit 11 geometrically transforms the input image using the projective transformation matrix H in the same manner as in step S101c, and records the input geometrically transformed image in the memory 16 to finish the process. In this embodiment, it is assumed that the input geometric transformation image shown in FIG. 8 is obtained by using the projective transformation matrix H.

As a result, the input image of the cylindrical object taken from the direction in which the normal direction of the bottom of the cylinder and the optical axis of the camera do not match is projected onto the image taken from the direction in which the normal direction of the bottom of the cylinder and the optical axis of the camera match. It will be converted. In other words, the original image taken from an angle oblique to the x'-y'plane in space is projected and transformed into a front image taken from an angle perpendicular to the plane.

Step S102 of FIG. 3;
Next, the foreground extraction unit 12 is activated. The foreground extraction process performed by the foreground extraction unit 12 will be described with reference to the flowchart of FIG.

Step S102a;
The foreground extraction unit 12 inputs the background geometric transformation image read from the memory 16 and the input geometric transformation image. In the present embodiment, the background geometric transformation image shown in FIG. 7 and the input geometric transformation image shown in FIG. 8 are input.

Step S102b;
Next, the foreground extraction unit 12 performs background subtraction processing using the two input converted images. The background subtraction processing can be realized by obtaining the absolute value of the difference between the pixels corresponding to the same positions in the background geometric transformation image and the input geometric transformation image.

Step S102c;
Next, the foreground extraction unit 12 performs smoothing processing on the difference image because the difference image obtained by the background subtraction processing often contains fine noise. The smoothing process can be realized, for example, by applying a Gaussian filter to the difference image. If the noise contained in the difference image is small, the smoothing process may be skipped with priority given to the processing speed.

Step S102d;
Next, the foreground extraction unit 12 obtains a mask image indicating the position of the foreground pixel by performing a binarization process on the difference image according to a predetermined threshold value.

Step S102e;
Next, the foreground extraction unit 12 extracts foreground pixels from the input geometric transformation image and the mask image. This process can be realized by, for example, an AND operation process of the input geometric transformation image and the mask image.

Step S102f;
Finally, the foreground extraction unit 12 records the image composed of the extracted foreground pixels as a foreground area image in the memory 16 and finishes the process. In this embodiment, it is assumed that the foreground region image shown in FIG. 10 can be obtained. As shown in FIG. 10, the background pattern that may cause erroneous detection is removed in advance by step S102.

Step S103 of FIG. 3;
Next, the designated color extraction unit 13 is activated. The designated color extraction process performed by the designated color extraction unit 13 will be described with reference to the flowchart of FIG.

Step S103a;
The designated color extraction unit 13 inputs predetermined color information and a foreground area image. The predetermined color information is a value indicating the color of the object to be detected, and has two values, a minimum value and a maximum value, for each predetermined color. For example, when the object to be detected is blue, the minimum value (0,0,200) and the maximum value (0,0,255) are set in RGB notation. In the case of red, the lowest value (200,0,0) and the highest value (255,0,0) are also expressed in RGB. Although it has been described in RGB notation for simplification of the explanation, it may be expressed in other color spaces such as HSV, HLS, and LAB. In that case, the color space of the foreground area image is also converted.

Step S103b;
Next, the designated color extraction unit 13 extracts a region corresponding to the designated color from the foreground region image and the color information. This process can be realized by performing extraction on all the pixels of a pixel in the foreground region image when the pixel value falls between the maximum value and the minimum value of the color information. The image obtained as a result of performing for all pixels is called a designated color extraction image.

Step S103c;
Next, in the designated color extraction unit 13, the color information on the image may change due to the influence of lighting in the actual environment, and the extracted pixels are often missing or excessively extracted. Performs smoothing processing on the specified color extracted image. The smoothing process can be realized, for example, by applying a Gaussian filter or a median filter, or by applying a morphology operation. Similar to step S102c, the smoothing process may be skipped with priority given to the processing speed.

Step S103d;
Finally, the designated color extraction unit 13 records the designated color extraction image in the memory 16 and finishes the process. In the present embodiment, there are two colors of an object to be detected, blue and red, and the above-mentioned RGB values are used as the color information. As a result, the images shown in FIGS. 12A and 12B can be obtained as the designated color extraction images.

Step S104 of FIG. 3;
Next, the contour detection unit 14 is activated. The contour detection process performed by the contour detection unit 14 will be described with reference to the flowchart of FIG.

Step S104a;
The contour detection unit 14 inputs a designated color extraction image.

Step S104b;
Next, the contour detection unit 14 converts the input designated color extraction image into grayscale and binarizes it with a predetermined threshold value.

Step S104c;
Next, the contour detection unit 14 extracts contours surrounding each of the designated color regions included in the binarized image. This process can be achieved, for example, by the method of reference (Suzuki, S, 1 outsider, “Topological Structural Analysis of Digitized Binary Images by Border Following”, CVGIP 30 1, 1985, p.32-p.46). .. In the present embodiment, the contours shown in FIGS. 14A and 14B can be obtained for each of the two designated color extraction images.

Step S104d;
Finally, the contour detection unit 14 approximates all the extracted contours with ellipses, outputs ellipse information of the approximated ellipses, and finishes the process. Here, the ellipse information is information including at least the center coordinates of the ellipse, the major axis and the minor axis of the ellipse, and the rotation angle of the ellipse. For example, the process of approximating the contour with an ellipse is described in References (Andrew W, 1 outsider, "A Buyer's Guide to Conic Fitting", Proc.5th British Machine Vision Conference, Birmingham, 1995, p.513- p.522. ) Can be achieved. In the present embodiment, the ellipses shown in FIGS. 15A and 15B are obtained for each of the two designated color extraction images.

Step S105 of FIG.
Next, the shape verification unit 15 is activated. The shape verification process performed by the shape verification unit 15 will be described with reference to the flowchart of FIG.

Step S105a;
The shape verification unit 15 inputs all ellipse information and a predetermined roundness threshold value.

Steps S105b to S105d;
Next, the shape verification unit 15 performs roundness threshold processing on each ellipse information to determine whether or not the ellipse being processed is close to a circle. The roundness threshold processing can be measured, for example, by whether the ratio of the minor axis to the major axis of the ellipse is close to 1.0. In this case, the roundness threshold processing determines that it is close to a circle if the minor axis ÷ major axis> threshold value. All the ellipses judged to be close to a circle are output as object information and the process is completed. In the present embodiment, the center coordinates, the major axis and the minor axis of the ellipse, and the rotation angle of the ellipse are output as circular object information for 5a, 5b, and 5c shown in FIGS. 15A and 15B, respectively. Since 5d, which is an approximation of the human shape with an ellipse, is not close to a circle, its object information is not output.

The first embodiment has been described above.

According to the present embodiment, an input image of a cylindrical object taken from a direction in which the normal direction of the bottom surface of the cylinder and the optical axis of the camera do not match is taken from a direction in which the normal direction of the bottom surface of the cylinder and the optical axis of the camera match. Since the projection is converted into an image, the shape of the subject can be grasped in the original shape, the omission of detection of the cylinder or sphere to be detected can be suppressed, and the detection can be performed reliably.

Further, according to the present embodiment, a foreground region is extracted from the projected image, an object is detected from the foreground region using color and contour information, and it is verified whether the shape of the detected object is circular. Therefore, it is possible to suppress excessive detection in which objects other than cylindrical objects are also detected.

As a result, the accuracy of detecting a cylindrical object from the input image can be improved. Even if it is a sphere, the shape after the projective conversion is circular like a cylindrical object, so even if the input image contains a sphere, the object of the sphere can be detected accurately. ..

<Second embodiment>
In the first embodiment, even in the case of an input image taken from a direction in which the normal direction of the bottom surface of the cylinder and the optical axis of the camera do not match, the normal direction of the bottom surface of the cylinder and the optical axis of the camera are set by projection conversion processing. The subject is a cylindrical object by projecting conversion from the matching direction to the captured image, detecting the object using the color and contour information from the projected image, and performing shape verification on the detected result. Also realized that the detection was performed accurately.

However, the projective conversion process used in the first embodiment is based on the premise that the subject is a flat body having no height. For example, when the real space shown in FIG. 5 is targeted, it is surrounded by (x ₁ , y ₁ ), (x ₂ , y ₂ ), (x ₃ , y ₃ ), and (x ₄ , y ₄ ). Although accurate projective conversion processing can be performed on a flat surface portion, a cylindrical object, a person, or the like existing on the flat surface will be distorted in the image after the projective conversion.

Therefore, if the upper surface of the cylindrical object exists at a position higher than the plane of the floor composed of four points specified for calculating the projective transformation matrix, the detection position and the position on the plane may be deviated. become.

On the other hand, for example, when it is desired to measure the exact position of a stone in a curling competition, it may be necessary to specify the position of a cylindrical object on a plane serving as a game field. Therefore, in the present embodiment, as shown in FIG. 17, the deviation of the detection position due to the height of the cylindrical object or the sphere is eliminated, and the center position of the detected cylindrical object is transferred to the plane targeted by the projective conversion process. Estimate the center position.

FIG. 18 is a diagram showing a configuration block configuration of the object detection device 1 according to the second embodiment. As shown in FIG. 18, the object detection device 1 is configured by adding the detection position correction unit 17 to the configuration of the first embodiment.

The detection position correction unit 17 is a functional unit that inputs the object information output from the shape verification unit 15, corrects the position of the detected object, and outputs the object information after the position correction.

Next, the object detection process performed by the object detection device 1 will be described. In this embodiment, each process of the projective conversion unit 11 that performs processing different from that of the first embodiment and the detection position correction unit 17 added in the second configuration will be described.

First, the projection conversion process performed by the projection conversion unit 11 will be described with reference to the flowchart of FIG.

At least one or more images are input to the projective conversion unit 11. In this embodiment, for example, the input image shown in FIG. 5 is input. Further, in the present embodiment, steps S205 and S206 are added to the process (FIG. 4) of the first embodiment. Hereinafter, the description of the same steps (steps S201 to S204, S207 to S208) as in the first embodiment will be omitted, and only steps S205 and S206 will be described.

Step S205;
The projective transformation unit 11 calculates the inverse matrix H ^-1 from the projective transformation matrix H calculated in step S202. The inverse matrix ^{H -1} is a matrix representing the geometric transformation to the input image from the projection transformation image, for _{^{example, H -1 × (x '1}} , y' 1, 1) ^ T = (x 1, y ₁ , 1).

Step S206;
Next, the projective transformation unit 11 records the calculated projective transformation matrix H and the inverse matrix H ^-1 of the projective transformation matrix H in the memory 16.

Next, the process performed by the detection position correction unit 17 will be described with reference to the flowchart of FIG.

Step S301;
The detection position correction unit 17 inputs all the object information for which the shape verification process has been completed and a predetermined offset. Here, as shown in FIG. 21, the offset means, for example, from the center position of the object on the input image to the center position on the plane targeted by the projection conversion process in consideration of the height of the cylindrical object. It is an offset value up to (a predetermined offset value based on the height of the cylindrical object from the x'-y'plane), and can be expressed as, for example, a vector d passing through both points.

Step S302;
Next, the detection position correction unit 17 performs the subsequent processing on a certain object.

Step S303;
First, the detection position correction unit 17 inversely converts the coordinates of the center position of the object to the coordinates of the input image coordinate system before the projection conversion using the inverse matrix H- ¹ . In the present embodiment, as shown in FIG. 22, the coordinates of the center position of the detected object are (c _x ', _cy '), and are converted into (c _x , _cy ) by the inverse matrix H- ^1. It shall be.

Step S304;
Next, the detection position correction unit 17 adds a predetermined offset to the converted coordinates. In the present embodiment, it is assumed that (d _x , _dy ) is given in advance as the offset d, and the coordinates after the addition are (c _x + d _x , _cy + _dy ).

Step S305;
Next, the detection position correction unit 17 reconverts the coordinates (c _x + d _x , _cy + _dy ) into the projective transformation image coordinate system using the projective transformation matrix H. In the present embodiment, as shown in FIG. 23, ((c _x + d _x )', ( _cy + d _y )') is calculated as the estimation result of the center position on the target plane in the projective transformation process. become.

Step S306;
Next, the detection position correction unit 17 updates the reconverted coordinate values as new coordinates of the center position of the object to be processed and stores them in the memory 16.

Step S307;
The detection position correction unit 17 performs the above processing on all the input objects, and finally outputs all the object information stored in the memory 16 to finish the processing.

The second embodiment has been described above.

According to this embodiment, the coordinates of the center position of the object are inversely converted to the coordinates of the input image coordinate system, and from the center position of the object on the input image to the center position on the target plane in the projection conversion process. After adding the offset values of, the coordinates after adding the offset values are returned to the coordinates of the projection conversion image coordinate system, so the upper surface of the cylindrical object is higher than the plane composed of the four points specified for calculating the projection conversion matrix. Even if it exists in a place, the center position on the plane can be estimated, and for example, in a curling competition, the exact position of the stone on the plane serving as the game field can be calculated.

Although the present invention has been specifically described above based on the example of the embodiment, the above description of the embodiment is for explaining the present invention, and limits the invention described in the claims. Or it should not be understood to reduce the range. Further, each means configuration of the present invention is not limited to the above-described embodiment, and it goes without saying that various modifications can be made within the technical scope described in the claims. For example, the foreground extraction process of step S102 shown in FIG. 3 may be performed before the projection conversion process of step S101.

Finally, the object detection device 1 described in each embodiment can be realized by a computer or the like having the above functions. It is also possible to create an object detection program for operating the computer as the object detection device 1 and a storage medium for the object detection program.

1 ... Object detection device 11 ... Projection conversion unit 12 ... Foreground extraction unit 13 ... Designated color extraction unit 14 ... Contour detection unit 15 ... Shape verification unit 16 ... Memory 17 ... Detection position correction unit

Claims (4)

In an object detection device that detects a columnar or spherical object from an image
A projective converter that transforms an original image taken from an oblique angle with respect to a plane in space into a front image taken from an angle perpendicular to the plane.
A foreground extraction unit that extracts a foreground region located in front of the plane from the front image,
A designated color extraction unit that extracts a designated color region that matches the designated color from the foreground region,
A contour detection unit that approximates the contour surrounding the designated color area with an ellipse,
A shape verification unit that determines whether or not the ellipse is circular, and if the ellipse is circular, outputs the ellipse as a cylindrical or spherical object.
An object detection device comprising. The position of the ellipse is inversely converted to the position of the coordinate system before the projective transformation, and a predetermined offset value based on the height of the cylinder or sphere forming the ellipse from the plane is added, and then the coordinate system after the projective transformation. The object detection device according to claim 1, further comprising a detection position correction unit for returning to the position of. In an object detection method that detects a columnar or spherical object from an image,
The computer
A step of projecting and transforming an original image taken from an angle oblique to a plane in space into a front image taken from an angle perpendicular to the plane.
A step of extracting a foreground region located in front of the plane from the front image,
A step of extracting a designated color area matching the designated color from the foreground area, and
A step of approximating the contour surrounding the specified color area with an ellipse,
A step of determining whether or not the ellipse is circular, and if the ellipse is circular, a step of outputting the ellipse as a cylindrical or spherical object.
An object detection method characterized by performing. An object detection program according to claim 1 or 2, wherein the computer functions as the object detection device.

Images