KR20110060315A - A method of recognizing self-localization for a road-driving robot - Google Patents

Abstract

PURPOSE: A self-positioning method of a road-driving robot is provided to stably recognize the accurate position by calculating a virtual position according to a driving distance measured by an odometer and increasing the accuracy of the position using landmarks in a road image. CONSTITUTION: A self-positioning method of a road-driving robot(10) is as follows. A driving distance from the previous position to a moved position is measured through an odometer(12). The positions of road landmarks are extracted from a road image captured by a camera(11) at the moved position. A plurality of virtual positions moved from the previous position by the driving distance are created. The moved position is estimated from the average of virtual positions, wherein the weighted values are calculated as the accuracy of the landmark positions on the image, the virtual images are weighted and sampled again. The position of the robot is calculated through repetition of the above steps.

Description

A method of recognizing self-localization for a road-driving robot}

The present invention relates to a magnetic position recognition method of a road driving robot that detects its own position while driving on a road using a road image captured by a camera and a travel distance measured by a travel system.

In particular, the present invention relates to a method for recognizing a magnetic position of a road driving robot that obtains and averages virtual positions of a moving position from a driving distance measured by a traveling system, and estimates the moving position by resampling the virtual positions using a road image. will be.

The present invention also relates to a method for recognizing a magnetic position of a road driving robot that obtains a vertex from an image of a road image and divides the area by a triangulation method and extracts a point where representative colors contact other areas as a landmark.

In general, a riding robot has a function of moving a person or object to a desired destination. That is, the riding-type mobile robot can be easily used by anyone by simply pressing the destination button without moving any further to find the optimum route. When an unexpected obstacle appears while driving, the robot stops safely.

For this purpose, the mobile robot should have the following functions.

First, as a technology for recognizing walls and curbs, obstacles, lanes, signs, current positions, and the like, the boarding mobile robot uses a laser range finder, a camera, a GPS, an encoder, and the like. In addition, by using the integrated analysis of the recognized signal to determine the position of the robot, the relative movement of the feature and the adjacent obstacle, and to generate the appropriate movement path and motion pattern toward the destination, it uses probability judgment, learning, optimization algorithm, etc. do.

In particular, a technique for accurately identifying its own location, that is, precise localization is essential.

An example of a technology for recognizing a magnetic position of a mobile robot is [Korean Patent Laid-Open Publication 2005-0020614 (2005.03.04 publication), "Self-Location Recognition Apparatus and Method of Intelligent System Using Artificial Mark and Its Intelligent System"] The technique is presented in 1). The prior art 1 constitutes a detectable artificial mark from an image captured by the camera regardless of the position where the camera looks at the mark, attaches the artificial mark to the surrounding environment, and detects the artificial mark by detecting the artificial mark. It is about.

The prior art 1 can recognize its position by an artificial mark, but it is practically impossible to install such an artificial mark on all roads.

In order to solve the above problems, a technique for determining a magnetic position by recognizing a surrounding environment with a camera has been proposed. Method for estimating the magnetic position of the robot based on the surrounding environment information including the "" (hereinafter referred to as prior art 2). In the prior art 2, the mobile robot recognizes individual objects in the environment by using a vision sensor, and analyzes the surrounding environment (three-dimensional) information including the recognized object and the recognized object based on a given map (map). A technique for estimating one's position is proposed. At this time, the camera recognizes the individual objects in the image obtained by the camera unit, generates the camera coordinate system position values of the local feature points of the individual objects and the local feature points of the surrounding environment including the individual objects, and compares the position values with the map to A technique for estimating location.

Since the prior art 2 estimates the magnetic position only by the vision of the camera, it has no choice but to rely solely on the degree of recognition of the individual object. However, the camera's vision differs in accuracy depending on the surrounding environment (such as very bright sunlight, dark at night, or shade). It is not practical to apply magnetic location recognition on a road running at high speed.

In order to solve the above problems, a technique for recognizing the position of a mobile robot using dead-reckining and range sensing has been proposed, and an example of the technique is disclosed in Korea Patent Laid-Open Publication No. 2007-0054557. (Published May 29, 2007), "Method and Apparatus for Recognizing a Magnetic Position of a Mobile Robot"]. The prior art 3 is a dead-reckining step of detecting a state amount changed according to the movement of the mobile robot by using a gyroscope and the like, and an accessory form or a charging station or the like to at least Sensing a distance from at least one fixed position and calculating an absolute position of the mobile robot, and applying a Kalman filter to the changed amount of state and the calculated absolute position to estimate an optimal position of the mobile robot at present. The method is divided into a step and a step of determining and correcting whether the determined current optimum position falls within a predetermined effective area.

In addition, an example of a technique for estimating a magnetic position using a mileage and an image taken by a camera is disclosed in Korean Patent Laid-Open Publication No. 2007-0113939 (Nov. 29, 2007), "Mobile robot, a control method and a program recorded with a computer. Possible recording medium "] (hereinafter, referred to as prior art 4). The prior art 4 accumulates the driving distance detected by the wheel rotation angle sensor at the estimated current driving position to predict the current driving position, and virtually mounts the camera at the estimated current driving position and the candidate position around the estimated driving position. A plurality of predicted edge images composed of the edge information to be arranged and imaged are generated based on the environment layout information, and an actual edge image is generated from the actual images photographed by the camera. The running position is updated by comparing the edge image with the plurality of predictive edge images, estimating the candidate position of the predicted edge image having the maximum similarity as the traveling position.

The prior art 3 has a problem in that dead recording using a gyroscope or a charging station for distance sensing is difficult to apply to a technology for recognizing a position when driving on a road.

In addition, the prior art 4 has a problem that must have accurate environmental information enough to generate a surrounding environment image. That is, a plurality of predicted edge images including edge information captured by virtually arranging a camera at a current driving position predicted based on the environmental layout information such as the position and height of a column or a wall stored in advance, and a candidate position in the vicinity thereof are obtained. Should be created. That is, the prior art 4 can be applied in a limited space, such as inside the house, but because it is not possible to store all the environment image around the road, it is difficult to apply to a robot driving on the road.

An object of the present invention is to solve the problems described above, using the road image captured by the camera and the driving distance measured by the odometer, the magnetic position recognition of the road driving robot to determine its position while driving the road To provide a way.

In addition, an object of the present invention is to obtain the virtual position of the moving position from the mileage, weighted average of the virtual position according to the accuracy matched with the position of the landmark obtained from the road image to recognize the magnetic position of the road running robot To provide a way.

In addition, an object of the present invention is to obtain a vertex in the road image captured by the method to segment the area by a triangulation (Delaunary Triangulation), and to extract the point where the representative color is in contact with the other areas as a landmark as a self-recognition method of the road driving robot To provide.

In order to achieve the above object, the present invention relates to a method for recognizing a magnetic position of a road driving robot equipped with a odometer for measuring a mileage and a camera for photographing a front surface of the vehicle. Measuring the distance traveled from the position to the movement position; (b) extracting a position of a landmark on a road (hereinafter referred to as a landmark position on an image) from a road image at a moving position captured by the camera; (c) generate a plurality of virtual positions that would have moved by the distance from the previous position and estimate the movement position by the average of the virtual positions, obtain the weight with the accuracy of the landmark position on the image, weight the virtual position and sample again Making; And (d) obtaining a magnetic position (or a moving position) of the driving by repeating the steps (a) to (c).

In another aspect, the present invention provides a method for recognizing a magnetic position of a road driving robot, wherein step (b) includes: (b1) photographing a road image of a moving position through the camera; (b2) recognizing a lane from the road image; (b3) extracting a landmark from the road image; (b4) projecting the extracted landmark onto a plane; And (b5) extracting the location of the landmark by matching the projected landmark with the landmark on the map.

In addition, the present invention is a method for recognizing a magnetic position of a road driving robot, step (b2) is characterized in that the lane is recognized using a pair of hyperbola-pair lane detection algorithm (Hyperbola-pair lane detection algorithm).

In another aspect, the present invention provides a method for recognizing a magnetic position of a road driving robot, wherein step (b3) includes: (b31) detecting an edge from the road image; (b32) extracting a corner image from the detected edges; (b33) dividing the region by applying Delaunary Triangulation from the vertex image; (b34) determining a representative color of the region as the dominant color in each region; And (b35) comparing the representative colors of the respective areas and extracting the points in contact with the changed areas as the points of the landmark.

In addition, the present invention is characterized in that in the step (b4) in the method for recognizing the magnetic position of the road running robot, the planar position of the extracted landmark is obtained by [Equation 1].

[Equation 1]

Where h _c and β are the camera's height and tilt angle, respectively, and α _h and α _v are the camera's horizontal and vertical angles, respectively.

In addition, the present invention is a magnetic position recognition method of the road running robot, the step (c), (c1) the virtual position of the moving position from the virtual position of the previous position (hereinafter referred to as the virtual position) (hereinafter referred to as the virtual position) Randomly generating them by an error; (c2) estimating a moving position by averaging the moving virtual position; (c3) calculating weights for the moving virtual positions using a probability that the landmark position in the moving virtual position matches the landmark position on the image; And (c4) sampling the moving virtual position again by the weight.

In addition, the present invention is a magnetic position recognition method of the road running robot, the step (c1) is characterized in that to generate a moving virtual position by the formula (2).

[Equation 2]

Where P _t-1 ^[m] is the mth virtual position at time t-1, Δl is the travel distance, V is the error covariance constant, and gauss_rand () is a random variable with normal distribution.

In addition, the present invention is a method for recognizing a magnetic position of a road driving robot, the step (c3) is characterized in that to obtain a weight by [Equation 3].

[Equation 3]

Where N is the number of landmarks in the image,

As described above, according to the magnetic position recognition method of the road driving robot according to the present invention, by using the road image captured by the camera and the traveling distance measured by the odometer, the position of the self is detected, thereby increasing the accuracy of the position estimation. The effect of lowering the installation cost is obtained.

In addition, according to the magnetic position recognition method of the road driving robot according to the present invention, the virtual position is first obtained according to the traveling distance of the driving system, and the accuracy of the position is improved by using the landmark of the road image to compensate for the large deviation of the image sensing. The more stable and accurate position can be obtained.

That is, the measurement of the mileage by the odometer has an error range in a certain range, the landmark recognition by the camera may be very large according to the environment (strong sunlight, dark shade, etc.). By first calculating the virtual distance based on the driving distance, and applying the technique of image sensing only for accuracy, the magnetic position can be recognized within the error range of the odometer even when the camera error is large. If the error is very small, it can recognize the more accurate magnetic position than the position recognition of the odometer only. Therefore, more stable and accurate magnetic position recognition is possible.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the drawings.

In addition, in describing this invention, the same code | symbol is attached | subjected and the repeated description is abbreviate | omitted.

First, a configuration of a road driving robot according to an embodiment of the present invention will be described with reference to FIGS. 1 and 2.

As shown in FIG. 1, the road running robot 10 is equipped with a camera 11 and a traveling system 12.

Preferably, the camera 11 is installed on the roof of the road driving robot 10. Camera 11 is preferably installed slightly inclined downward so as to photograph the surface of the road. However, the installation location may be anywhere. However, the camera 11 may be installed at any position as long as it can capture the front surface of the road driving robot 10 or the front of the driving direction. The camera 11 is preferably capable of imaging at a resolution of 30 frames per second (30 fps) and 320 * 240.

On the other hand, odometer (12) is preferably installed on the drive wheel of the road running robot (10). The installed odometer 12 preferably has an accuracy capable of measuring 100,000 pulses per meter.

2, the arithmetic unit 20 is a processing apparatus for grasping the traveling position of the road running robot 10. As shown in FIG. The computing device 20 receives a road image captured by the camera 11, and receives a driving distance measured from the odometer 12. In addition, the calculation device 20 is a device for calculating the current position of the road running robot 10 by using the input road image and the travel distance.

The computing device 20 may be implemented as a microprocessor or a small computer. In this case, an arithmetic operation of the arithmetic unit 20 is created by a program processed by a processing unit such as a CPU. In addition, the arithmetic operation 20 may be implemented as a circuit by an application-specific semiconductor (ASIC).

The memory 30 is a storage space for storing the data processed by the arithmetic unit 20 during operation. The memory 30 may store a program for calculating a current position.

The memory 30 also stores map information on roads that drive. The map information is information in which a road location, a landmark on the road, and the like are displayed. The road location information is information on a lane, and preferably includes map information having an accuracy of distinguishing the inside and the outside of the lane.

A landmark is an indicator displayed on a road such as an arrow indicating a driving direction, a crosswalk, a stop line, and the like. That is, information about the type of landmark and its location is included on the map.

Next, the magnetic position recognition method of the road running robot according to an embodiment of the present invention will be described with reference to FIG.

As shown in Figure 3a, the magnetic position recognition method of the road running robot 10 (a) measuring the traveling distance traveled from the previous position to the moving position through the odometer 12 (S10); (b) extracting a position of a landmark on the road (hereinafter referred to as a landmark position on the image) from the road image at the moving position captured by the camera 11 (S20); (c) generate a plurality of virtual positions that would have moved by the distance from the previous position and estimate the movement position by the average of the virtual positions, obtain the weight with the accuracy of the landmark position on the image, weight the virtual position and sample again Step S30; And (d) obtaining a magnetic position (or a moving position) by repeating steps (a) to (c) (S40).

First, the distance traveled from the previous position to the movement position through the odometer 12 is measured (S10). The traveling system 12 is a measuring mechanism which measures a traveling distance using the rotation angle of the drive wheel of the road running robot 10. For example, the distance traveled from the previous position to the movement position can be measured as Δl. Δl has an error with respect to the actual mileage. That is, an error generated by the driving system 12 itself or an error caused by slipping of the driving wheel, an error caused by the road driving robot 10 not traveling in the correct driving path, and the like are accumulated. The more distance you travel, the greater the error. Therefore, it is necessary to correct the cumulative error even in the middle to determine the exact moving position.

Next, the arithmetic unit 20 extracts the position of the landmark on the road (the landmark position on the image hereinafter) from the road image at the moving position picked up by the camera 11 (S20).

As shown in FIG. 4A, a landmark refers to a road marking such as a lane, an arrow, a pedestrian crossing, a speed bump, and the like. The landmarks are very important information for locating the location in the road environment.

A method of extracting a landmark position on an image will be described in more detail with reference to FIG. 2. As shown in FIG. 3B, a method of extracting a landmark position on an image includes (b1) photographing a road image of a moving position through a camera 11 (S21); (b2) recognizing a lane from the road image (S22); (b3) extracting a landmark from the road image (S23); (b4) projecting the extracted landmark on a plane (S24); And (b5) extracting the location of the landmark by matching the projected landmark with the landmark on the map (S25).

When the odometer 12 measures the travel distance from the previous position to the movement position, the road image at the movement position is captured by the camera 11 (S21). That is, the captured road image is an image captured at the moving position.

The lane is recognized from the road image (S22). At this time, a lane is extracted by applying a pair of hyperbola-pair lane detection algorithms (see Document 1) to the road image. The lane detection algorithm has an advantage that it can be calculated in real time by calculating at a high speed. In addition, the algorithm is resistant to noise so that even when a part of the lane is obscured, it can be properly obtained.

[1] Qiang Chen and Hong Wang, "A Real-time Lane Detection Algorithm Based on a Hyperbola-Pair Model", Intelligent Vehicles Symposium, pp. 510-515, 2006

Next, a landmark is extracted from the road image (S23). Most landmarks are detected in white and yellow. Since scanning all road spaces may be insufficient, an algorithm is used to extract the points of the landmark by Delaunary Triangulation.

That is, as shown in Figure 4c, (b31) detecting the edge from the road image (S23a); (b32) extracting a corner image from the detected edge (S23b); (b33) dividing the region by applying a triangle method from a vertex image (S23c); (b34) determining a representative color of the region as the dominant color in each region (S23d); And (b35) comparing the representative colors of the respective areas to extract a point in contact with the changed area as a landmark (S23e).

As shown in (b) of FIG. 4B, an edge is detected from the road image (S23a).

4B, a corner image is extracted from the detected edge (S23b).

In addition, as shown in (d) of FIG. 4B, a region is divided by applying a triangulation method (see [2]) from the vertex image (S23c).

[2] Dominique Attali and Jean-Daniel Boissonnat, "A Linear Bound on the Complexity of the Delaunay Triangulation of Points on Polyhedral Surfaces", Discrete and Computational Geometry, Vol. 31, No. 3, pp. 369-384, 2004

Next, as shown in (e) of FIG. 4B, the representative color of the region is determined as the dominant color in each region. Since roads are mainly divided into white, yellow and black, the color most widely distributed in one area is defined as the representative color of the area.

The representative color of each area is compared to extract points that are in contact with the change area as the points of the landmark. An example of the result is shown in (f) in FIG. 4B.

When the landmark is extracted from the road image, the extracted landmark is projected on the plane (S24). The image extracted from the road image cannot be directly compared with the landmark displayed on the map. Since the captured road image is from the camera's perspective, it must be converted into an image from the sky. Therefore, by comparing the landmarks on the image from the perspective of the sky with the landmarks on the map to determine the moving position of the robot (10).

The planar position of the landmark extracted from the road image is obtained by [Equation 1].

[Equation 1]

Where h _c and β are the height and tilt angle of the camera, respectively, and α _h and α _v are the horizontal and vertical angles of the camera, respectively.

d _v (k) and w _v (k) are in-plane coordinates of the points of the landmark.

5 shows an example of converting the view of the camera from the view of the sky.

After obtaining the landmark position from the point of view of the sky, the position of the landmark is extracted by matching the projected landmark with the landmark on the map (S25).

Next, generate a plurality of virtual positions that would have moved by the distance from the previous position and estimate the moving position by the average of the virtual positions, obtain the weight with the accuracy of the landmark position on the image, weight the virtual position and sample again (S30).

Specifically, as shown in FIG. 3D, the method for estimating the moving position includes: (c1) randomly generating virtual positions of the moving position (hereinafter, referred to as virtual positions) from the virtual positions of the previous position (hereinafter, referred to as virtual positions). (S31); (c2) estimating a moving position by averaging the moving virtual position; (c2) determining a probability that the landmark position in the moving virtual position matches the landmark position on the image as a weight for the moving virtual positions (S32); And (c3) step S33 of sampling the moving virtual position again based on the weight.

As shown in FIG. 6A, the case of traveling from position P1 to P2 and from P2 to P3 will be described as an example. In this case, it is assumed that the virtual position (previous virtual position) of the previous position is P ₁ ¹ , P ₁ ² ,..., P ₁ ⁵ with respect to the previous position.

As shown in FIG. 6C, moving virtual positions are randomly generated from previous virtual positions (S31). For example, suppose you move from P1 (previous position) to P2. At this time, the virtual positions (previous virtual positions) at the P1 position are P ₁ ¹ , P ₁ ² , ..., P ₁ ⁵ . From these previous virtual positions, the moving virtual positions P ₂ ¹ , P ₂ ² , ..., P ₂ ⁵ are obtained.

At this time, the moving virtual position is generated by the following [Equation 2].

[Equation 2]

Where P _t-1 ^[m] is the mth virtual position at time t-1, Δl is the travel distance, V is an error covariance constant, and gauss_rand () is a random variable with normal distribution.

Therefore, it can be seen that the mileage error when the road driving robot 10 travels is expressed by multiplying a covariance constant with a random variable having a normal distribution.

The moving position estimation estimates the moving virtual positions as shown in Equation 3 below, and estimates the moving positions by averaging the moving virtual positions as shown in Equation 4 (S32).

&Quot; (3) "

for m = 1 to M

end

&Quot; (4) "

That is, the M virtual positions are moved by the traveling distance measured from the traveling system 12. As shown in FIG. 6C, P ₁ ¹ , P ₁ ² ,..., P ₁ ⁵ are moved to P ₂ ¹ , P ₂ ² , ..., P ₂ ⁵ . The virtual positions moved at this time are obtained by Equation 2. Also, the moving position is estimated by averaging P ₂ ¹ , P ₂ ² ,..., P ₂ ⁵ .

Next, the probability that the landmark position in the moving virtual position matches the landmark position on the image is determined as a weight for the moving virtual positions (S32).

That is, the weights for the moving virtual positions are obtained by the following [Equation 5] (S33).

[Equation 5]

for m = 1 to M

end

는 확률밀도함수(PDF, Probability Density Function)를 나타낸다. N개의 이미지상 랜드마크가 M개의 이동 가상위치와 매치된다.

Denotes the probability density function (PDF). The landmarks on the N images match the M moving virtual positions.

Thus, the weight W _t ^[m] is obtained by comparing the position on the map with the position on the landmark of each virtual position. The weights are then normalized so that the sum of the total weights is one. Thus, the number of virtual positions can be maintained at M according to each step.

Next, the moving virtual position is sampled again by the weight (S34).

Sampling again from the moving virtual positions P ₂ ¹ , P ₂ ² , ..., P ₂ ⁵ of FIG. 6C to the moving virtual positions P ' ₂ ¹ , P' ₂ ² , ..., P ' ₂ ⁵ of FIG. 6D. do. In this case, the moving virtual location having a higher weight may be sampled into more moving virtual locations, and the moving virtual location having a lower weight may disappear. For example, in FIG. 6C, P ₂ ⁵ has the highest weight and P ₂ ⁴ has the lowest weight. In particular, since P ₂ ⁴ is 0.1 or less (reference value), it disappears without being sampled. Instead, P ₂ ⁵ is sampled twice. As a result, P ' ₂ ⁴ and P' ₂ ⁵ are sampled from P ₂ ⁵ .

When the road driving robot 10 moves from P2 to P3 again, the moving position is estimated again at the point P3. At this time, P2 becomes the previous position and P3 becomes the movement position. Step (c1) to (c4) is repeated in the process of repeating step (a) to (c).

Therefore, by the repetition as described above, since the moving position of the road driving robot 10 is obtained by the average of the virtual positions having a large weight, the error may be gradually converged to decrease.

Next, the experimental results by the magnetic position recognition method according to an embodiment of the present invention will be described with reference to FIG.

FIG. 7 shows the results obtained by comparing the results of the estimation of the 30th movement position on the circular track of 100m and the actual movement position. FIG. 7A shows the average error of the estimated movement position, and FIG. 7B shows the track driving. The error of each case is plotted.

As shown in FIG. 7B, it can be seen that the position of the y-axis is more accurately estimated than the x-axis. This is because the lane landmark of the road, the most important characteristic, exists on both sides of Robo.

As mentioned above, although the invention made by this inventor was demonstrated concretely according to the said Example, this invention is not limited to the said Example, Of course, a various change is possible in the range which does not deviate from the summary.

The present invention is applicable to the development of a road running robot that can grasp its position while driving on the road by using the road image captured by the camera and the mileage measured by the odometer. In particular, the present invention is applicable to the development of a road driving robot that obtains the virtual positions of the moving position from the driving distance and estimates the moving position by weighting the virtual positions according to the accuracy matched with the position of the landmark obtained from the road image. Do.

1 is a perspective view of a road driving robot according to an embodiment of the present invention.

2 is a block diagram of a configuration of a road driving robot according to an embodiment of the present invention.

3 is a flow chart illustrating a magnetic position recognition method of a road driving robot according to an embodiment of the present invention.

4 is a diagram illustrating an example of extracting a landmark from a road image according to the present invention.

5 is an exemplary view of projecting a road image according to the present invention on a plane.

6 is a diagram illustrating an example of estimating a moving position from a virtual position according to the present invention.

7 is a table and a diagram showing the experimental results by the magnetic position recognition method according to an embodiment of the present invention.

Explanation of symbols on the main parts of the drawings

10: road driving robot 11: camera

12: odometer 20: computing device

30: memory

Claims (8)

In a magnetic position recognition method of a road running robot equipped with a odometer for measuring the mileage and a camera for imaging the front surface of the driving, (a) measuring the distance traveled from the previous position to the movement position through the odometer; (b) extracting a position of a landmark on a road (hereinafter referred to as a landmark position on an image) from a road image at a moving position captured by the camera; (c) generate a plurality of virtual positions that would have moved by the distance from the previous position and estimate the movement position by the average of the virtual positions, obtain the weight with the accuracy of the landmark position on the image, weight the virtual position and sample again Making; And and (d) obtaining a magnetic position (or a moving position) of the traveling by repeating steps (a) to (c). According to claim 1, wherein step (b), (b1) photographing a road image of a moving position through the camera; (b2) recognizing a lane from the road image; (b3) extracting a landmark from the road image; (b4) projecting the extracted landmark onto a plane; And and (b5) extracting the location of the landmark by matching the projected landmark with the landmark on the map. The method of claim 2, The step (b2) is a magnetic position recognition method of a road running robot, characterized in that for detecting the lane using a pair of hyperbola-pair lane detection algorithm (Hyperbola-pair lane detection algorithm). The method of claim 2, wherein step (b3), (b31) detecting an edge from the road image; (b32) extracting a corner image from the detected edges; (b33) dividing the region by applying Delaunary Triangulation from the vertex image; (b34) determining a representative color of the region as the dominant color in each region; And (b35) comparing the representative color of each area, and extracting the points in contact with the change area as the points of the landmark, the magnetic position recognition method of the road robot. The method of claim 2, In the step (b4), the magnetic position recognition method of the road driving robot, characterized in that to obtain the planar position of the extracted landmark by [Equation 1]. [Equation 1]

Where h _c and β are the camera's height and tilt angle, respectively, and α _h and α _v are the camera's horizontal and vertical angles, respectively. The method of claim 2, wherein step (c) comprises: (c1) randomly generating virtual positions (hereinafter referred to as moving virtual positions) of the moving position from the virtual positions of the previous position (hereinafter referred to as virtual positions) by an error; (c2) estimating a moving position by averaging the moving virtual position; (c3) obtaining weights for the moving virtual positions using the probability that the landmark position in the moving virtual position and the landmark position on the image are matched; And and (c4) re-sampling the moving virtual position by the weight. The method of claim 6, The step (c1) is a magnetic position recognition method of the road running robot, characterized in that for generating a moving virtual position by [Equation 2]. [Equation 2]

Where P _t-1 ^[m] is the mth virtual position at time t-1, Δl is the travel distance, V is the error covariance constant, and gauss_rand () is a random variable with normal distribution. The method of claim 6, The step (c3) is a method for recognizing the magnetic position of the road running robot, characterized in that to obtain a weight according to [Equation 3]. [Equation 3]

Where N is the number of landmarks in the image,

Images