KR100824744B1 - Localization System and Method for Mobile Robot Using Corner's Type - Google Patents

Abstract

According to the present invention, when a map of an indoor environment is given, a corner shape is extracted from an image obtained from a single camera, a corner shape and a two-dimensional map are matched according to possible corner shape and geometric constraints, and between the matched edges. The method of detecting the position of a mobile robot is derived by deriving a nonlinear system of equations from the rigid and perspective transformations and solving the solution by numerical methods.

Mobile Robot, Robot Vision, Magnetic Position Detection, 3D Recognition, Image Processing

Description

Magnetic Position Detection System and Method for Mobile Robot Using Corner Shape {Localization System and Method for Mobile Robot Using Corner's Type}

1 is a system configuration diagram of a magnetic position detector in the present invention

2 is an example of a two-dimensional map in the present invention.

3 is a table relating to the type of corner shape in the present invention

4 is an example of a result of defining the input image and the edge shape in the present invention

5 is a rigid body and perspective transformation for position calculation in the present invention

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

DESCRIPTION OF SYMBOLS 1 Camera part 2 Computer part 3 Robot drive part 4 Corner part 5 Matching part 6 Position calculation part 7 Robot drive motor 8 Convex edge 9 Concave edge 10 Vertical boundary 11 Branch boundary 12 Corner shape

In order to detect the magnetic position of the mobile robot indoors, the present invention extracts the corner shape from the image obtained from a single camera when a map of the indoor environment is given, and the corner shape and the two-dimensional shape according to the possible corner shape and geometric constraints. Magnetic position detection system of mobile robot using edge shape which detects magnetic position by conducting matching between maps, deriving nonlinear system of equations from rigid and perspective transformation between matched edges, and solving the solution by numerical method It's about how.

The present invention belongs to the field of detecting the magnetic position of the mobile robot, the conventional magnetic position detection method using a relative position detection method for tracking the position using an encoder or an acceleration sensor, using a laser distance detector or computer vision There is an absolute location detection method that estimates the actual location within a map of the environment. The magnetic position estimation method using a camera includes a stereo imaging method, an omnidirectional camera method, and a single camera method, and the stereo imaging method uses a parallax information from two images for a distance to a point of an object. In detecting the information, the pixels of the two images need to be matched with each other, which is relatively time consuming and the system is complicated because two cameras are used. In the omnidirectional visual method using a conical mirror, it is possible to obtain the front, side, and rear view at once, but the image information is geometrically distorted, takes a lot of processing time, and it is difficult to accurately estimate the magnetic position. Among the methods using a single camera, the method of using an artificial fixed marker is a method of using a certain pattern or figure including information such as position and direction, which is a relatively simple and effective method, but has a disadvantage of having to install the marker directly. In the indoor environment, there is a method of detecting a magnetic position using a vertical boundary as a camera, but there is a problem in that registration between the two-dimensional map and the vertical line is almost impossible using only the position of the vertical line.

The present invention relates to a method of detecting a magnetic position using a single camera under indoor conditions. The present invention relates to detecting a shape of an edge corresponding to a lower end portion of a vertical line in an image, matching it with a map, and then matching the nonlinearity from the matched information. It is an object of the present invention to provide a magnetic position detection system that detects the magnetic position of a mobile robot with only one camera by deriving a system of equations and solving the solution by a numerical method.

Referring to the configuration of the present invention for achieving the above object is as follows.

The configuration of the present invention comprises the steps of extracting the edge shape from the hardware consisting of the mobile robot, the camera and the computer, the image obtained from the camera, and matching the edge shape and the two-dimensional map by geometrically possible edge shape and geometric constraints. The implementation is divided into software, which consists of detecting the position of the mobile robot by deriving a nonlinear system of equations from the rigid and perspective transformations between the matched edges and solving the solution numerically.

Referring to the hardware configuration of the system of the present invention in detail with the accompanying drawings as follows.

The hardware of the detection system of the present invention includes a robot driving unit 3 which can move arbitrarily, a computer unit 2 which controls the input, inputs an image of the front of the robot, detects a magnetic position, and a camera unit 1 inputs an image. It is composed.

The robot driving unit 3 is a part for adjusting the movement of the robot, and is generally composed of wheels composed of two or more robot driving motors 7, and attaches a camera unit to an upper end, and any position in an indoor environment. It can be moved to the coordinate (a, b, θ) corresponding to. Where a is the X-axis coordinate, b is the Y-axis coordinate, and θ represents the rotation. The camera unit 1 is a camera that inputs an image of the front while moving indoors, and includes an image input unit capable of inputting an image to the computer unit 2. The computer unit 2 extracts a corner shape from an image input from the camera unit 1, performs matching between the corner shape and the two-dimensional map, and positions the mobile robot from the transform using the matched information. , θ). The computer unit 2 specifically comprises an edge extracting unit 4, a matching unit 5, and a position calculating unit 6, wherein the corner extracting unit 4 is an edge from an image obtained from the camera unit 1. The shape is extracted, and the matching part 5 performs the matching between the corner shape and the two-dimensional map by geometrically possible edge shape and geometric constraints, and the position calculating part 6 is a rigid body transformation and perspective between the matched edges. The position of the mobile robot is detected by deriving a nonlinear system of equations from the transformation and solving the solution by numerical method.

The corner extractor 4 first inputs an image, obtains a vertical boundary 10 from the input image, and then obtains a horizontal boundary again. Horizontal boundary components exist between the vertical boundaries, which are defined as branch boundaries 11. Between one vertical boundary 10 has a branch boundary 11 on both left and right sides, and the edge shape 12 is determined by the angle of the vertical boundary 10 and the branch boundary 11. The obtained corner shape 12 has one of four types of FIG. 2, and the process is as follows.

With respect to one vertical boundary 10,

If the left and right branch boundaries 11 are connected to each other,

If the left branch boundary is greater than 90 degrees and the postal branch boundary is less than 90 degrees, the edge is arrow.

If the left edge is less than 90 degrees and the postal border is greater than 90 degrees, the edge is Fork.

If the left edge is less than 90 degrees and the postal border is greater than 90 degrees, the edge is Fork.

If the left and right branch boundaries 11 are not connected,

If the lower branch boundary is right, the edge is Right-L.

If the lower branch boundary is left, the edge is Left-L.

If the occluded branch boundary is to the right, the edge is Left-L.

 If the occluded branch boundary is left, the edge is Right-L.

Taking the corner shape detection result as described above, in Fig. 4, b01, b12, b23, b34, and b45 all have branch boundaries, and the detected edge shapes are u1, u2, u3, and u4.

The matching part 5 performs matching between the corner shape and the two-dimensional map by the geometrically possible corner shape and geometric constraints. 3 shows, as an example of a two-dimensional map, all the edges included on the map consisting of convex edges 8 and concave edges 9. The two-dimensional map is represented as shown in FIG. 2, and the edge is composed of a convex edge 8 and a concave corner 9. First, all the edges of a two-dimensional map are listed in order. The shape is one of convex and concave. Next, the two-dimensional shape according to the edge shape in the image is defined according to FIG. 3, and the convex and concave are arranged in order. By comparing the order of the corner shapes on the two-dimensional map with the order of the corner shapes on the two-dimensional map of the edges in the image to find the best matching pair. Obtain the coordinates of the u-axis coordinate u _i on the horizontal line in the matched pair image and the coordinates (x _i , y _i ) on the map.

The position calculating unit 6 performs matching using the edges in the matched map and the image. The position calculating unit 6 derives a nonlinear system of equations from the rigid body transformation and the perspective transformation for the matched edge point pairs. See FIG. 5 to describe this. In FIG. 5, (x, y, z) is a map coordinate system, (x _i , y _i ) is a corner coordinate of the map coordinate system, (X _c , Y _c , Z _c ) is a camera coordinate system, (Cx _i , Cy _i ) Is the corner coordinate of the camera coordinate system, u _i is the U-axis coordinate of the corner point of the camera coordinate system, (a, b, θ) is the position of the robot in the map coordinate system, and n is the number of matched corners, respectively. . First, the coordinate transformation of the corner point of the map coordinate system to the camera coordinate system is expressed by Equation 1 when the rigid body transformation T is applied. Here, the robot is limited to the movement in the X and Y axes and the rotation about the Z axis. The feature point is located on the horizontal axis, so it is set to 0.

Here, the rigid transformation matrix T is a movement and rotation matrix as shown in [Equation 2].

Here, the corner points (x _i , y _i ) of the global coordinate system are given in the map and the parameters to be estimated are (a, b, θ). The perspective conversion equation from the camera coordinate system to the image coordinate system is shown in [Equation 3].

From above [Equation 2] and [Equation 3], nonlinear simultaneous equations with (a, b, θ) as variables are derived. If the number of features is three or more, the nonlinear system of equations can be solved using the Newton method, one of the numerical methods. If the nonlinear simultaneous equation is expressed as a vector function F, it has n different component functions f _n and can be expressed as shown in [Equation 4].

If the variable (a, b, θ) is expressed as a vector P, the nonlinear simultaneous equation can be expressed as F (a, b, θ) = 0, and the Jacobian matrix J (a, b, θ) is expressed as [Equation 5]. .

The Newton's method for finding the solution P of the nonlinear simultaneous equation represented by the vector equation F (a, b, θ) = 0 is given by Equation 6 given the initial approximation P ⁽⁰⁾ .

Repeat the above procedure to find a solution and exit if the value does not change within a certain range. The solution P obtained here is the position and direction of the mobile robot with respect to the reference coordinate.

According to the present invention, when a two-dimensional map of an indoor environment is given, the corner shape is extracted from an image obtained from a single camera, and the corner shape and the two-dimensional map are matched according to possible corner shape and geometric constraints, and the matched edge By providing a method of detecting the position of a mobile robot that derives a nonlinear system of equations from the rigid and perspective transformation between them and obtains the solution numerically, unlike the conventional method, it is possible to use a single camera, Since the matching is possible, the magnetic position detection of the mobile robot moving indoors is easier, faster, and lower cost magnetic position detector can be provided.

Claims (5)

Given a map of the indoor environment, the edge extraction unit 4 extracts the edge shape from the image obtained from one camera unit 1, and the edge shape and the two-dimensional map are matched by possible edge shape and geometric constraints. A matching part 5 for detecting the position of the mobile robot by deriving a nonlinear system of equations from the rigid and perspective transformations between the matched edges and calculating the solution in a numerical method. Magnetic position detection providing system for a robot, characterized in that used. In the method of claim 1, wherein the edge extracting portion in the magnetic position detection providing system of the robot to determine the corner shape, Method for determining the edge shape (12) by the angle of the vertical boundary (10) and the branch boundary (11) after defining the vertical boundary (10) and the branch boundary (11) from the image inputted by the corner extractor (4). The method of claim 2, Based on one vertical boundary 10 to obtain the corner shape 12, if the left and right branch boundaries 11 are connected to each other, the corner shape is determined with the angles of the post branch boundary and the left branch boundary, If the left and right branch boundary (11) is not connected to determine the corner shape (12) according to the left and right positions of the lower branch or occluded branch. In the method of matching part in the magnetic position detection providing system of the robot of claim 1,  The matching unit 5 lists all the edges of the two-dimensional map in order of one of the convex and concave corners, and determines the corner shapes in the image as the convex and concave corners according to the two-dimensional shape and arranges them in order. And comparing the order of the edge shapes on the 2D map of the edges in the image with the order of the edge shapes on the 2D map to find and match the best matching pair. In the method for detecting the position of a mobile robot in the magnetic position detection providing system of claim 1,  The position calculating section (6) detects the position of a mobile robot by deriving a nonlinear system of equations from the rigid and perspective transformations between the matched edges and finding the solution by numerical method.

Images