KR100961616B1 - Method and system for calibrating of omnidirectional camera based on contour matching - Google Patents

Abstract

The present invention relates to a method and system for omnidirectional camera correction using contour matching, wherein an epipolar geometric estimation is performed on at least two omnidirectional images, and a contour is performed on the at least two omnidirectional images. Detect and use the epipolar geometry estimation to match the detected contours, minimize angular errors of epipolar planes and three-dimensional vectors of the matched contours, and minimize the epipolar planes and three-dimensional vectors. Rotation and movement information between the cameras photographing the at least two omnidirectional images are estimated based on the angle error of.

Contour registration, omni-directional camera, correction

Description

Method and system for omnidirectional camera correction using contour matching {METHOD AND SYSTEM FOR CALIBRATING OF OMNIDIRECTIONAL CAMERA BASED ON CONTOUR MATCHING}

The present invention relates to a omnidirectional camera correction method and system using contour matching, and more particularly, to a omnidirectional camera correction method and system using contour matching that can estimate the rotation and movement information of the camera.

The omnidirectional camera system having a wide viewing angle has been widely used in surveillance and robot fields, and researches for analyzing 3D information or camera motion information from omnidirectional video sequences are being actively conducted. In this process, further consideration is needed to apply a technique for accurately detecting and matching feature components such as corners, straight lines, curves, and the like to the omnidirectional image. In particular, straight lines and the like, which are commonly found in artificial structures such as buildings, are projected as contours in omnidirectional images, but there are relatively few studies for automatically matching them and analyzing camera movements.

In the conventional omnidirectional image and related imaging system, the analysis of the projection model of the target camera system, the estimation of the external parameters including the motion information of the camera and the analysis of the three-dimensional space, the method of processing the omnidirectional image for this, and then It is divided into various uses.

The study of mathematically and experimentally interpreting the projective relationship of various designed omnidirectional imaging systems is divided according to the use of multi-view geometry. There are techniques such as a calibration pattern in which linearity or three-dimensional position of an object is defined in advance, and a technique using geometric conditions in an image sequence acquired according to the movement of a camera.

However, in the related art, a user must set a correlation between scenes in advance, and the contour information cannot be automatically matched in the omnidirectional image.

The present invention has been made to improve the prior art as described above, and automatically detects the corresponding contour existing in the image of the projection model of the omnidirectional camera previously analyzed, and then through the process of minimizing the angle error function An object of the present invention is to make it possible to estimate the rotational and movement information of.

In addition, an object of the present invention is to provide a method for automatically setting contours and efficiently setting correspondences.

In order to achieve the above object and solve the problems of the prior art, a method for estimating rotation and movement information between cameras using contour matching according to one aspect of the present invention, selecting a reference image and a reference image from the omnidirectional image, Performing Epipolar geometric estimation on the image and the reference image; Detecting contours from each of a reference image and a reference image using the epipolar geometric estimation; Matching contours detected from each of the reference image and the reference image using a brightness distribution in a region normalized to a window having a predetermined size with respect to the reference image and the reference image; Obtaining an epipolar plane for both endpoints constituting each of the matched contours; Minimizing an angular error of a three-dimensional vector that is an intersection of the epipolar plane and the plane formed by the epipolar plane and the matched contour; And estimating rotation and movement information between cameras photographing the reference image and the reference image, based on the angle error of the minimized epipolar plane and the 3D vector.

According to another aspect of the present invention, the step of performing epipolar geometric estimation on the at least two omnidirectional images, detecting a feature point using a Harris corner operator, and the specific corresponding of the at least two omnidirectional images Determining a correspondence point of the at least two omnidirectional images by examining a correlation of an area.

According to another aspect of the invention, the step of determining the corresponding point of the at least two omnidirectional image, selecting an inlier set by the RANSAC algorithm by at least eight or more corresponding points, the at least two In this step, the corresponding point of the omnidirectional image is determined.

According to another aspect of the present invention, the detecting of the contour of the at least two omnidirectional images is a step of detecting the contour of the at least two omnidirectional images using a Kenny operator and an edge connection algorithm. .

According to another aspect of the present invention, detecting the contour of the at least two omnidirectional images is detecting a contour of one pixel thickness.

According to another aspect of the invention, the step of matching the detected contour using the epipolar geometric estimation, checking whether the candidate points constituting the detected contour match, the at least two omnidirectional An end point of the first image, which is one of the images, is set as a first endpoint candidate, and an intersection of the epipolar curve and the extension line of the contour in the second image, which is one of the at least two omnidirectional images, is obtained. Setting the second endpoint candidate of the corresponding contour, and matching the endpoints constituting the detected contour by calculating a three-dimensional vector using the first endpoint candidate and the second endpoint candidate.

According to another aspect of the invention, minimizing the angular error of the epipolar plane and the three-dimensional vector, the camera parameters for minimizing the angular error of the epipolar plane and three-dimensional vector of the matched contour And estimating rotation and movement information of the camera photographing the at least two omnidirectional images, based on the estimated parameter, rotation and movement of the camera photographing the at least two omnidirectional images It is a step of estimating information.

An omnidirectional camera correction system using contour matching according to an aspect of the present invention includes an epipolar geometry estimator that performs epipolar geometry estimation on at least two omnidirectional images, and a contour on the at least two omnidirectional images. a contour detection unit for detecting a contour, a contour matching unit for matching the detected contours using the epipolar geometric estimation, and an error adjusting unit for minimizing an angular error of an epipolar plane and a three-dimensional vector of the matched contours And a camera correction unit for estimating rotation and movement information between cameras photographing the at least two omnidirectional images based on the angle error of the minimized epipolar plane and the 3D vector.

According to another aspect of the invention, the epipolar geometry estimator, a feature point detector for detecting a feature point using a Harris corner operator, and by examining the correlation of the particular corresponding region of the at least two omnidirectional image, And a correspondence point determiner for determining corresponding points of two omnidirectional images.

According to another aspect of the present invention, the correspondence point determination unit selects an inlier set by a RANSAC algorithm based on at least eight correspondence points, and determines corresponding points of the at least two omnidirectional images.

According to another aspect of the present invention, the contour detection unit detects a contour on the at least two omnidirectional images by using a Kenny operator and an edge connection algorithm.

According to another aspect of the invention, the contour detector detects a contour of one pixel thickness.

According to another aspect of the present invention, the contour matching unit checks whether the candidate points constituting the detected contour match, and determines an end point of the first image, which is one of the at least two omnidirectional images, of the first image. An intersection point between an epipolar curve and an extension line of a contour in a second image, which is one of the at least two omnidirectional images, is set as an endpoint candidate, and the first endpoint By calculating a three-dimensional vector using the candidate and the second endpoint candidate, the endpoints constituting the detected contour are matched.

According to another aspect of the invention, the error adjusting unit, the camera parameters for minimizing the angular error of the epipolar plane and the three-dimensional vector of the matched contour, the camera correction unit, the estimated parameters The rotation and movement information of the camera photographing the at least two omnidirectional images is estimated.

According to the present invention, it is possible to automatically detect the corresponding contour existing in the image of the projected model of the omnidirectional camera analyzed in advance, and then estimate the rotation and movement information of the camera by minimizing the angle error function.

In addition, according to the present invention, it is possible to provide a method for efficiently setting the correspondence by automatically matching the contour.

Further, according to the present invention, the position of the camera can be automatically tracked from the image sequence, or the software can be useful for restoring the three-dimensional structure of the target scene and using this to synthesize virtual objects three-dimensionally. In particular, it can be applied to the core technologies in the realization of ubiquitous environment such as security, surveillance, and autonomous driving of robots, and the production of high-tech digital image contents such as virtual reality, computer animation, and visual effects of film, and services related to photography and interior architecture. It is widely applicable to industry.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings and the contents described in the accompanying drawings, but the present invention is not limited or limited to the embodiments. Like reference numerals in the drawings denote like elements.

1 is a block diagram of a omni-directional camera correction system using contour matching according to an example of the present invention. An omnidirectional camera correction system using contour matching according to an example of the present invention will be described with reference to FIG. 1.

As shown in FIG. 1, the omnidirectional camera correction system using contour matching according to an example of the present invention includes an epipolar geometry estimation unit 110, a contour detection unit 120, a contour matching unit 130, and an error adjustment unit ( 140, and a camera corrector 150.

For reference, FIG. 1 illustrates a module directly or indirectly related to the present invention. Of course, other drawings may apply as well. However, blocks shown in the present specification can be designed as a module, and the module means one unit for processing a specific function or operation. The module may be implemented in hardware or software, or may be implemented in a combination of hardware and software.

The epipolar geometry estimator 110 performs epipolar geometry estimation on an omnidirectional image including a reference image and a reference image.

The epipolar geometry estimator 110 detects a feature point by using a Harris edge operator, and examines a correlation between a specific corresponding region of the at least two omnidirectional images, thereby detecting the at least two omnidirectional images. It may be configured to include a corresponding point determiner for determining the corresponding point of the.

In particular, the correspondence point determiner may select an inlier set by a RANSAC algorithm based on at least eight correspondence points to determine corresponding points of the at least two omnidirectional images.

In general, epipolar constraints exist between omnidirectional images obtained from two viewpoints, and at least eight corresponding points are required to estimate the epipolar geometry from the obtained element matrix.

Feature points are detected using the Harris edge operator, and the corresponding points of the two images are determined by checking the correlation of the corresponding area. Based on the correspondence of the feature points, we obtain the element matrix and epipolar geometry. In this process, because an incorrect element matrix is estimated due to mismatched feature points between images, an inlier set is selected using a robust algorithm, 8 point RANSAC method. In addition, by using singular value decomposition, the element matrix is decomposed into unit rotation vectors of relative rotation and movement, and used as an initial parameter for matching the corresponding contour.

The contour detector 120 detects a contour on at least two omnidirectional images.

The contour detector 120 may detect a contour of the at least two omnidirectional images by using a Kenny operator and an edge connection algorithm. In particular, the contour detector 120 may detect a contour of one pixel thickness. That is, the contour detector 120 may detect contours from each of the reference image and the reference image by using epipolar geometry estimation.

The contour matching unit 130 matches the detected contours using the epipolar geometric estimation.

In more detail, the contour matching unit 130 checks whether candidate points constituting the detected contour coincide with each other and checks an endpoint of the first image, which is one of the at least two omnidirectional images. Can be set as one endpoint candidate. The intersection of the epipolar curve and the extension line of the contour in the second image, which is one of the at least two omnidirectional images, may be set as the second endpoint candidate of the corresponding contour. Accordingly, the contour matching unit 130 may match the endpoints constituting the detected contour by calculating a three-dimensional vector using the first and second endpoint candidates. That is, the contour matching unit 130 may match the contours detected from each of the reference image and the reference image by using a brightness distribution in a region normalized to a window having a predetermined size with respect to the reference image and the reference image.

The error adjusting unit 140 minimizes the angle error of the epipolar plane and the three-dimensional vector of the matched contour.

The camera corrector 150 estimates rotation and movement information between the cameras photographing the at least two omnidirectional images based on the angle error of the minimized epipolar plane and the 3D vector.

In this case, the error adjusting unit 140 estimates a camera parameter that minimizes an angular error of an epipolar plane and a three-dimensional vector of the matched contour, and the camera corrector 150 calculates the estimated parameter. The rotation and movement information of the camera photographing the at least two omnidirectional images may be estimated based on.

FIG. 2 is a diagram illustrating contour separation using a three-dimensional plane normal vector according to an example of the present invention, and FIG. 3 illustrates a merge using a three-dimensional plane normal vector formed by a contour according to an example of the present invention. It is for the drawing.

A contour detection method according to an example of the present invention will be described with reference to FIG. 2, and merging using a normal vector of a three-dimensional plane formed by the contour according to an example of the present invention will be described with reference to FIG. 3.

For each image, Kenny's operator and edge concatenation algorithm are used to detect contours one pixel thick. FIG. 2 illustrates a method of dividing into two different contours by using a normal vector of planes composed of a camera center and three-dimensional segments when two line segments are projected as one contour in the omnidirectional image in three-dimensional space. It is also shown.

As shown in FIG. 2, Cs 210 and Ce 220 are both endpoints of the contour and Ps 230 and Pe 240 are incident on camera center Co 250 and Cs 210 and Ce ( Is a three-dimensional vector projected to 220).

In order to separate the contour to which one or more three-dimensional line segments are connected, two three-dimensional vectors are cross-marked (Ps × Pk) while tracking a trajectory from the starting point Cs 210 to Ce 220. If the calculated normal vector changes larger than a preset threshold, it is separated into separate contours based on Ck 260. When the direction of the normal vector formed by the target contour is within a certain error range, merged into one connected contour.

3 shows a process of merging the short contours obtained in the initial stage into one contour.

4 is a diagram for describing positions of candidate end points of each tourer of a reference image with respect to a contour of the reference image according to an embodiment of the present invention. Referring to FIG. 4, positions of candidate end points of respective tourers of the reference image with respect to the contour of the reference image according to an example of the present invention will be described.

The epipolar geometry estimated at the beginning is used to determine the correspondence of the contours between the images and to match the corresponding points of both contours.

When the contour is described as two end points 440 and 450, the candidate points constituting the target contour may be used to determine the contour of the second image (reference image) corresponding to the contour of the first image (reference image). Check for match. For the contour end point 420 of the reference image, the intersection point of the epipolar curve 450 and the extension line 430 of each contour in the reference image is set as candidates for both end points 440 of the corresponding contour. At this time, since the three-dimensional vector projected at both ends is an intersection between the epipolar plane and the plane formed by the contour, it can be calculated as the external of the normal vectors of the two planes.

In order to determine the correspondence of the contours in the two images, the end points constituting each contour must be accurately matched. Even if the robust method is applied, the initial epipolar geometry obtained by the matching of the feature points causes some error. Therefore, since the end point candidate position of the contour in the reference image is not accurate, the rectangular window used for the general image is difficult to obtain an accurate result in the matching process of the omnidirectional image.

Accordingly, an active matching method of variably setting a region considered according to a position in the omnidirectional image and normalizing it to a window having a predetermined size is used.

Equation 1 is used to compare the brightness distributions within the normalized area and determine whether they match.

및

는 윈도우 내 밝기의 평균임 .) Where I and J are the brightness of the image in the window,

And

Is the average of the brightness within the window.)

Through the calculation of Equation 1, the contour having the highest value is determined as the corresponding contour in the second image (reference image), and at the same time, the corresponding position corresponds to the end point of the contour in the first image (reference image). Is the point.

Finally, the parameters of the camera are estimated by minimizing the angular error of the corresponding plane from the corresponding contour that coincides both ends. The epipolar plane is obtained from the 2m end points that make up the m contours corresponding to each other in the two images, and the angle error is calculated based on the rotation and movement information by considering the normal vector of each plane and the 3D vector of the corresponding point. do.

Using the omnidirectional camera model, three-dimensional vectors Pi0 and Pi1 passing from the center of each of the first image (reference image) and the second image (reference image) passing through the i-th corresponding point existing in the image are obtained. The baseline vector connecting the centers of the two images, the normal vector of the epipolar plane of the reference image formed by Pi0 (ni1), and the planar normal vector of the reference image (ni0) in the same manner are expressed by the following equation (2). Obtain

('^' Represents the unit vector of the vector, and Θ and Φ are rotation angles about the y-axis of rotation and movement, respectively.)

When the estimation of the rotation and movement of the camera is accurate or not, the result of Equation 2 shows values of 0 and 1, respectively, and calculates the angle error at the reference point and the reference point for both end points of the corresponding contour. We divide by 4m to normalize between 0 and 1.

5 is a diagram illustrating feature points detected by a Harris corner detector according to an example of the present invention.

According to an example of the present invention, a corresponding contour is automatically determined from two omnidirectional images, and a geometric vector can be used to estimate a direction vector for rotation and movement of a camera.

In order to verify the accuracy of the proposed algorithm, we can use the POV-Ray rendering software to use the composite image acquired by the virtual camera with equidistant projection model and the image acquired by the digital camera equipped with fisheye lens.

FIG. 5 illustrates the results of obtaining feature points using a Harris edge detector to estimate an initial epipolar geometry in an omnidirectional image having an equidistant projection model.

Since the boundary part around the image may not overlap with the movement of the camera, only the feature points within 95% of the radius of the omnidirectional image are selected. In the case of the first image (reference image) 510, 1193 feature points may be detected, and in the case of the second image (reference image) 520, 1169 feature points may be detected.

FIG. 6 is a diagram illustrating contour components separated by three-dimensional line segments for merging contours according to an example of the present invention, and FIG. 7 is a diagram illustrating two or more contours adjacent feature points according to an example of the present invention.

As illustrated in FIG. 6, 89 contours are distributed in the first image (reference image) 610, and 104 contours are distributed in the second image (reference image).

As illustrated in FIG. 7, 213 and 217 feature points may be selected for the first image (reference image) 710 and the second image (reference image), respectively.

8 illustrates a merged contour image according to an example of the present invention.

As shown in FIG. 8, components within an error range of 2 ° in the direction of the normal vector of the plane formed by the contour are merged to form a long contour.

According to an example of the present invention, 16 and 23 contours are determined in the first image (the reference image) 810 and the second image (the reference image) 820, respectively, and the contours that are incorrectly merged are also present because only the angle is considered. However, this information is corrected during matching of the corresponding contour.

9 illustrates eight pairs of corresponding contours whose contour endpoints coincide in accordance with an example of the present invention.

According to an example of the present invention, although there is an error at some end points even when the correlation of brightness is used in the active window, the contour of the contour in the first image (the reference image) 910 and the second image (the reference image) 920 is used. Correspondence is set correctly.

10 is a flowchart illustrating a omnidirectional camera correction method using contour matching according to an example of the present invention. An omnidirectional camera correction method using contour matching according to an example of the present invention will be described with reference to FIG. 10.

After estimating the projection model of the omnidirectional camera (S1010), epipolar geometric estimation is performed on at least two omnidirectional images (S1020).

In performing the epipolar geometry estimation, a feature point is detected using a Harris corner operator, and a correlation point of a specific corresponding region of the at least two omnidirectional images is determined to determine corresponding points of the at least two omnidirectional images. Can be.

In addition, when determining a corresponding point of at least two omnidirectional images as described above, an inlier set is selected by a RANSAC algorithm by at least eight corresponding points, and the corresponding points of the at least two omnidirectional images are selected. You can decide.

Thereafter, a contour is detected with respect to the at least two omnidirectional images (S1030).

In this case, a contour may be detected on the at least two omnidirectional images using a Kenny operator and an edge connection algorithm, and a contour having a thickness of one pixel may be detected.

Thereafter, the detected contour is matched using the epipolar geometry estimation (S1040).

In this case, the candidate points constituting the detected contour are matched to each other, and an end point of a first image, which is one of the at least two omnidirectional images, is set as a first endpoint candidate, and the at least two omnidirectional directions An intersection of the epipolar curve and the extension line of the contour in the second image, which is one of the images, may be set as the second endpoint candidate of the corresponding contour. The endpoints constituting the detected contour may be matched by calculating a three-dimensional vector using the first and second endpoint candidates.

Then, the angle error of the epipolar plane and the three-dimensional vector of the matched contour is minimized (S1050), and based on the angle error of the minimized epipolar plane and the three-dimensional vector, the at least two omnidirectional images are taken. Rotation and movement information between the photographed cameras may be estimated (S1060).

At this time, when minimizing the angular error of the epipolar plane and the three-dimensional vector, it is possible to estimate a camera parameter that minimizes the angular error of the epipolar plane and the three-dimensional vector of the matched contour.

In addition, when estimating rotation and movement information of the camera photographing the at least two omnidirectional images, the rotation and movement information of the camera photographing the at least two omnidirectional images is estimated based on the estimated parameter. It can be configured to.

Therefore, according to an example of the present invention, a corresponding contour existing in the image is automatically detected for a projected model of the omnidirectional camera previously analyzed, and then the rotation and movement information of the camera is estimated by minimizing the angle error function. can do.

Embodiments of the invention also include computer-readable media containing program instructions for performing various computer-implemented operations. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions may be those specially designed and constructed for the purposes of the present invention, or they may be known and available to those skilled in the art. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the claims below but also by the equivalents of the claims.

1 is a block diagram of a omni-directional camera correction system using contour matching according to an example of the present invention.

2 is a view for explaining contour separation using a normal vector of a three-dimensional plane according to an example of the present invention.

3 is a diagram for explaining merging using a normal vector of a three-dimensional plane formed by a contour according to an example of the present invention.

4 is a diagram for describing positions of candidate end points of each tourer of a reference image with respect to a contour of the reference image according to an embodiment of the present invention.

5 is a diagram illustrating feature points detected by a Harris corner detector according to an example of the present invention.

6 is a diagram illustrating contour components separated by three-dimensional line segments for contour merging according to an example of the present invention.

7 is a diagram illustrating feature points where two or more contours are adjacent according to an example of the present invention.

8 illustrates a merged contour image according to an example of the present invention.

9 illustrates eight pairs of corresponding contours whose contour endpoints coincide in accordance with an example of the present invention.

<Explanation of symbols for the main parts of the drawings>

110: Epipolar Geometry

120: contour detection unit

130: contour matching unit

140: error adjustment unit

150: camera correction unit

Claims (15)

Selecting a reference image and a reference image from an omnidirectional image, and performing epipolar geometric estimation on the reference image and the reference image; Detecting contours from each of a reference image and a reference image using the epipolar geometric estimation; Matching contours detected from each of the reference image and the reference image using a brightness distribution in a region normalized to a window having a predetermined size with respect to the reference image and the reference image; Obtaining an epipolar plane for both endpoints constituting each of the matched contours; Minimizing an angular error of a three-dimensional vector that is an intersection of the epipolar plane and the plane formed by the epipolar plane and the matched contour; And Estimating rotation and movement information between cameras photographing the reference image and the reference image based on the angle error of the minimized epipolar plane and the 3D vector Method for estimating rotation and movement information between cameras using contour matching, comprising a. delete The method of claim 1, The epipolar geometric estimation method for estimating rotation and movement information between cameras using contour matching, characterized in that it uses a RANSAC algorithm with at least eight corresponding points. The method of claim 1, Detecting the contours (contours), And detecting a contour of the reference image and the reference image using a Kenny operator and an edge connection algorithm. The method of claim 1, Detecting the contours (contours), Detecting a contour of one pixel thickness, wherein the rotation and movement information between the cameras using the contour matching is estimated. The method of claim 1, Matching the contours, For each of the contours extracted from the reference image, merging the contours based on a normal vector direction of a plane formed by each contour component to generate merged contours for the reference image; For each of the contours extracted from the reference image, merging the contours based on a normal vector direction of a plane formed by each contour component to generate merged contours for the reference image; And Comparing brightness distributions within a region normalized to a window of a predetermined size with respect to the reference image and the reference image, and determining corresponding contours between the merged contours of the reference image and the contours of the reference image; Method for estimating rotation and movement information between cameras using contour matching, comprising a. The method of claim 6, Determining the corresponding contours, A brightness distribution is compared within a region normalized to a window of a predetermined size with respect to the reference image and the reference image by the following equation, [Equation]

Where I and J are the brightness of the image in the window,

And

Is the average brightness in the window) A method of estimating rotation and movement information between cameras using contour matching, wherein the merged contour having the highest value is determined as a corresponding contour in the reference image through the calculation of the equation. A computer-readable recording medium in which a program for executing the method of any one of claims 3 to 7 is recorded. An epipolar geometry estimator for performing epipolar geometry estimation in an omnidirectional image including a reference image and a reference image; A contour detector which detects contours from each of a reference image and a reference image using the epipolar geometric estimation; A contour matching unit matching contours detected from each of the reference image and the reference image by using a brightness distribution in a region normalized to a window having a predetermined size with respect to the reference image and the reference image; Obtaining an epipolar plane for both end points constituting each of the matched contours, and minimizing the angular error of the three-dimensional vector that is the intersection of the epipolar plane and the plane formed by the epipolar plane and the matched contour An error adjusting unit; And Camera correction unit for estimating the rotation and movement information between the camera taking the reference image and the reference image based on the angle error of the minimized epipolar plane and the three-dimensional vector Apparatus for estimating rotation and movement information between the camera comprising a. delete delete delete delete 10. The method of claim 9, The contour matching unit, For each of the contours extracted from the reference image, merge the contours based on the normal vector direction of the plane formed by each contour component to generate merged contours for the reference image, For each of the contours extracted from the reference image, merge the contours based on the direction of the normal vector of the plane formed by each contour component to generate merged contours for the reference image, Comparing brightness distributions within a region normalized to a window of a predetermined size with respect to the reference image and the reference image, to determine corresponding contours between the merged contours for the reference image and the contours for the reference image. Apparatus for estimating rotation and movement information between the camera, characterized in that. The method of claim 14, The determination of the corresponding contours, A brightness distribution is compared within a region normalized to a window of a predetermined size with respect to the reference image and the reference image by the following equation, [Equation]

Where I and J are the brightness of the image in the window,

And

Is the average brightness in the window) And determining the merged contour having the highest value as the corresponding contour in the reference image through the calculation of the equation.

Images