JP2011210183A - Mobile object position detection system and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a mobile object position detection system in which reliability of detection is secured by making it difficult to generate poor collation when pole star algorithm is used.SOLUTION: The mobile object position detection system 10 detects a position of a mobile object 16 which moves on a plane 12, and includes: a pole star collation means 24 which performs pole star collation using pole star feature quantity in which likelihood for absorbing an error between dots in the image obtained by the imaging means 20 and dots in the pole star database 34 corresponding to the dots is set; a narrowing down means 26 for narrowing down dots from collation dot candidates collated by the pole star collation means 24 to a predetermined collation dot; and a detection means 28 for detecting the current position and the direction of the mobile object 16 based on position relation of dot pattern consisting of the predetermined collation dot narrowed down by the narrowing down means 26.

Description

  The present invention relates to a moving body position detection system and method.

  Conventionally, a method of detecting the position of a moving body moving on a plane using a pattern (pattern) arranged on the plane is known (for example, see Patent Document 1).

  On the other hand, the inventors including a part of the present inventors have already proposed moving body position detection systems of Japanese Patent Application Nos. 2008-274680 and 2008-274681.

  As shown in FIGS. 17 and 18, the moving body position detection system images the floor surface 2 from the moving body 4 arranged on the floor surface 2 on which the known dot pattern 3 is drawn. By comparing the feature amount of the dot pattern 3 of the dot 5 with the database 6 in which the feature amount of the dot pattern 3 of the entire floor surface 2 is stored, which dot on the floor surface 2 The position of the moving body 4 on the floor surface 2 is detected.

  On the other hand, a digital pen technique developed by Anoto, Sweden, is known as a technique for discriminating the position and direction of an image taken by a camera (Patent Document 1 is a related document). As shown in FIG. 19, a paper on which a unique dot pattern 80 is printed and a pen 82 built in the camera 81 are used to detect the position on the paper surface of the pen tip 83 when writing a character. It is recorded as digital data. The dot pattern 80 is obtained by converting the XY position coordinate value into an encoded pattern. Conversely, the XY coordinate value can be obtained by restoring the encoded pattern. However, this technique has a problem that it is necessary to accurately draw the dot pattern 80 expressing the specified code, and position detection using a naturally occurring dot pattern cannot be performed.

  As shown in FIG. 20, a technique is known in which a pattern 84 in which two-dimensional information is encoded is drawn on a plane 85 and is read by a camera 86 to detect the position (for example, Patent Document 2). reference). This technique is based on the premise that a pattern is drawn with several types of figures, and has design restrictions. For example, a pattern cannot be configured with only dots.

  In addition, a pole star algorithm is known as a matching algorithm to a star arrangement diagram 88 (star map) of an imaging star point group used for a device (star sensor) used for attitude identification of the spacecraft 87 in outer space (for example, Non-Patent Document 1). As shown in FIG. 21, the star sensor captures the sky 89 with a camera mounted on the spacecraft 87, and the orientation of the camera by matching the arrangement of star points in the imaging region with a known star map 88. It is a sensor that identifies Stellar placement is naturally occurring and does not necessarily distribute points uniformly, but the Polestar algorithm is effective for matching such objects.

  This Polestar algorithm is a method for specifying the position of a part on the plane from the point distribution of the part when the points are widely distributed randomly on the plane. As shown in FIG. 22, the outline processing procedure of this algorithm is to acquire an image, label the points in the image (step S1), perform pole star matching using the image point group (step S2), This is a procedure of performing polygon matching on the star candidate group (step S3) and calculating the camera posture (step S4).

  Here, in the pole star collation in step S2, there are a plurality of straight lines from an arbitrary point on the plane to a peripheral point existing within a circle with a specified radius, but the combination of the lengths of the plurality of straight lines Makes use of what is unique about its central point. A feature amount representing the point distribution around each point is defined as a pole star feature amount. The database relating to the feature amount is previously created offline in advance by the processing of steps S5 to S8 shown in FIG.

  And the collation star point (database point) candidate is calculated | required by collating the pole star feature-value of each image point in an acquired image with a database. In the polygon collation in step S3, candidates having a geometric relationship with the image point group are selected from the collation candidate point groups by collation using polygons formed by connecting image points. In step S4, the imaging camera posture / position is calculated from the combination of the coordinate values of the remaining image point group and collation point group.

Next, the processing contents of each step in FIG. 23 will be specifically described with reference to FIGS.
(1) Calculation of pole star feature quantity (step S6)
As shown in FIG. 24, the length of a straight line (hereinafter referred to as an arm length) from a certain point (Pn) to a peripheral point existing within a specified distribution radius R is obtained, and the arm length is determined as a unit length D. To be indexed (hereinafter referred to as an arm length index).

ai = fix (ri / D) +1
amax = fix (R / D) +1
Here, ri: arm length, ai: arm length index, D: index unit length, R: distribution radius, amax: maximum arm length index. Also, fix is an operator that returns an integer value by truncating the decimal part.

  A combination of arm length indexes is a feature amount representing a point distribution around the point (hereinafter referred to as a pole star feature amount). Here, even if the same arm length index is calculated for the center point, the arm length index string as the pole star feature quantity does not change.

(2) Creation of the pole star database (steps S7 to S8: offline)
For all the target points (hereinafter referred to as database points), the above-mentioned pole star feature values of (1) are calculated, organized and stored as a lookup table (step S7).

  As shown in FIG. 25, the look-up table is a table organized by listing database points having arm length indexes in the pole star feature amount for each arm length index of 1 to amax.

(3) Pole star verification (Step S2: Online)
An image is acquired with a camera, and the position of each point (hereinafter referred to as an image point) in the acquired image is extracted. As shown in FIG. 26, the pole star feature is calculated for the image points of the acquired image.

  With respect to each arm length index in the pole star feature quantity of the image point, the database point (plurality) having the arm length index is examined as shown in FIG. 27 using the lookup table of FIG.

  Further, as shown in FIG. 28, a counter corresponding to each database point (P1 to Pmax) is prepared, and the counter of the corresponding point is incremented every time a database point appears in the above processing.

  The point having the maximum counter value is a matching point candidate corresponding to the image point. By performing the same processing on the pole star feature amount of each image point, a matching point candidate for each image point is obtained. A lower limit is set for the counter, and when the maximum counter value is less than or equal to this lower limit, the image point is not verified, and the image point is excluded from subsequent verification targets.

(4) Polygon matching (Step S3: Online)
For the database points corresponding to the image points obtained in (3) above, as shown in FIG. 29, a polygon formed by connecting several image points and a database corresponding to these image points. By selecting a combination of points that match a polygon created by connecting the points, a combination of image points and database points with high matching accuracy is obtained. In FIG. 29, a set of points 1, 3, 4, and 7 are finally collated.

(5) Position / direction calculation by camera posture calculation (Step S4: Online)
From the combination of the three or more image points obtained in (4) above and the XY coordinate values of the matching points, the posture (position / direction) of the camera is obtained.

JP 2006-141061 A JP 2008-52403 A

E. SILANI, M. LOVERA: Ploitecnico di Milano Italy, “Star Identification Algorithms: Novel Approach & Comparison Study”, IEEE Transactions on Aerospace and Electronic Systems, Vol.42, No.4 October 2006

  By the way, if the above-mentioned Pole Star algorithm is applied as it is as an algorithm for point group collation on the plane of the above-mentioned conventional mobile object position detection system, there is a problem of collating with an incorrect point (hereinafter referred to as erroneous collation), and collation cannot be performed. There is a possibility that a problem (hereinafter referred to as “unmatched”) and a problem (hereinafter referred to as “multiple matching”) that cannot be identified by checking many points may occur. Such a collation failure is considered to be caused by a difference in the preconditions of the star map collation and the point cloud collation on the plane. The prerequisites for these verifications and their differences are as follows.

[Prerequisites for star map matching]
(1) The relative positional relationship between the star points on the imaging surface due to lens distortion and camera posture is extremely small.
(2) According to the above (1), the position of the star point can be calculated in the image with extremely fine resolution.
(3) The number of star points (database points) in the star map can be selected by grade (representing luminance) and is approximately several thousand.

[Preconditions for point cloud matching on a plane]
(1) Since a wide-angle lens is used, the lens distortion is large, and the objective distance to a plane with a point cloud is short. Therefore, the change in the relative positional relationship between the points due to the camera posture change is so large that it cannot be ignored.
(2) Since the processing speed is improved, the imaging resolution cannot be increased, and the point position cannot be calculated with a fine resolution according to the above (1).
(3) The number of point groups (database points) depends on the size of the target plane, but is about 20 square meters, which is tens of thousands, more than the star map.

  Due to the difference in the condition (1), there is a possibility that a problem of incorrect collation and inability to collate occurs, and due to the difference in the conditions (2) and (3), a problem of multiple matching may occur.

  The present invention has been made in view of the above, and provides a mobile body position detection system and method that ensure detection reliability by making it difficult to cause a verification failure when using the Polestar algorithm. Objective.

  In order to solve the above-described problems and achieve the object, a mobile body position detection system according to claim 1 of the present invention is a mobile body position detection system that detects the position of a mobile body that moves on a plane, A dot pattern composed of a plurality of dots arranged on the plane; an imaging means provided in the moving body for imaging an area on the plane on which the moving body is located; and dots in an image acquired by the imaging means And a pole star matching means for performing a pole star matching using a pole star feature that sets a likelihood to absorb an error between the dots in the pole star database of the pole star algorithm corresponding to this dot, Narrowing means for narrowing the matching dot candidates collated by the pole star matching means to a predetermined matching dot, and narrowing by the narrowing means Based on the positional relationship between the dot pattern consisting of predetermined matching dots, characterized in that it comprises a detection means for detecting a current position and direction of the moving body.

  According to a second aspect of the present invention, there is provided the mobile object position detection system according to the first aspect, wherein the narrowing-down means has a predetermined matching dot candidate for the matching dot candidate verified by the pole star verification means. The method includes a process of obtaining a common area and narrowing down to collation dots existing in the common area.

  In the mobile body position detection system according to claim 3 of the present invention, in the above-described claim 1 or 2, the pole star collating unit divides the collation range in the pole star database into a plurality of regions. Pole star matching is performed in the area, and the area where the matching rate is maximized is obtained.

  A moving body position detecting method according to claim 4 of the present invention is a moving body position detecting method for detecting the position of a moving body moving on a plane, and a dot pattern comprising a plurality of dots on the plane. The moving body is provided with imaging means for imaging a region on a plane on which the moving body is located. A dot in an image acquired by the imaging means and a pole of the Polestar algorithm corresponding to the dot Pole star matching is performed using the pole star feature value that sets the likelihood to absorb the error with the dots in the star database, and the specified matching dots are narrowed down from the verified matching dot candidates The present invention is characterized in that the current position and direction of the moving body are detected based on the positional relationship of the dot pattern consisting of the matching dots.

  According to the present invention, there is provided a moving body position detection system that detects a position of a moving body that moves on a plane, the dot pattern including a plurality of dots arranged on the plane, and the moving body, An error between an imaging unit that captures an area on a plane on which the moving body is located, a dot in the image acquired by the imaging unit, and a dot in the pole star database of the Pole Star algorithm corresponding to the dot is calculated. Pole star collation means that performs pole star collation using the pole star feature amount for which the likelihood for absorption is set, and a narrowing means for narrowing down to a predetermined collation dot from the collation dot candidates collated by the pole star collation means, Based on the positional relationship of the dot pattern composed of the predetermined collation dots narrowed down by the narrowing-down means, Comprising a detection means for detecting the position and direction. For this reason, it is difficult to cause a verification failure when the Polestar algorithm is used, and the detection reliability can be ensured.

FIG. 1 is a configuration diagram showing an embodiment of a moving body position detection system according to the present invention. FIG. 2 is a diagram illustrating an example of a dot pattern and an acquired image. FIG. 3 is a diagram for explaining the concept of detecting a position from an acquired image. FIG. 4 is a flowchart showing the first embodiment of the present invention. FIG. 5 is a flowchart showing the second embodiment of the present invention. FIG. 6 is a diagram for explaining the collation point existence region. FIG. 7 is a flowchart showing the third embodiment of the present invention. FIG. 8 is a diagram for explaining the division of the verification floor area. FIG. 9 is a flowchart showing Embodiment 4 of the present invention. FIG. 10 is a plan view showing conventional inventory location management (1). FIG. 11 is a side view showing conventional inventory location management (2), (3). FIG. 12 is a plan view showing conventional inventory location management (2), (3). FIG. 13 is a plan view showing conventional inventory location management (2), (3). FIG. 14 is a perspective view when the mobile body position detection system according to the present invention is applied to inventory location management. FIG. 15 is a plan view when the mobile body position detection system according to the present invention is applied to inventory location management. FIG. 16 is a perspective view when the mobile body position detection system according to the present invention is applied to inventory location management. FIG. 17 is a schematic configuration diagram of a conventional mobile body position detection system. FIG. 18 is a diagram illustrating an example of a dot pattern and an acquired image of a conventional moving body position detection system. FIG. 19 is a diagram showing a conventional digital pen technique. FIG. 20 is a diagram illustrating a conventional position detection technique based on a coded pattern. FIG. 21 is a diagram showing a star sensor technique used for attitude identification of a conventional spacecraft. FIG. 22 is a main part flowchart of the Polestar algorithm used in the conventional star sensor technology. FIG. 23 is a flowchart of the Polestar algorithm used in the conventional star sensor technology. FIG. 24 is a diagram illustrating the pole star feature amount. FIG. 25 is a diagram illustrating a lookup table. FIG. 26 is a diagram illustrating the pole star feature amount of an image point. FIG. 27 is a diagram showing the correspondence between arm length indexes and database points. FIG. 28 is a diagram illustrating collation using a lookup table. FIG. 29 is a diagram for explaining the concept of polygon matching. FIG. 30 is a diagram illustrating the pole star bit. FIG. 31 is a diagram for explaining the creation of a pole star database matrix. FIG. 32 is a diagram for explaining the pole star feature quantity and the pole star bit. FIG. 33 is a diagram illustrating an arithmetic expression using a Polestar database matrix. FIG. 34 is a diagram for explaining a pole star database when the likelihood is set to ± 1. FIG. 35 is a diagram for explaining matching point candidates corresponding to image points. FIG. 36 is a diagram showing imgArmLsMtx. FIG. 37 is a diagram showing flrArmLsMtx. FIG. 38 is a diagram illustrating an example of source code. FIG. 39 is a diagram showing mchArmLsMtx. FIG. 40 is a diagram illustrating repetition of summation in the column direction. FIG. 41 is a diagram showing the remaining image point group-database point group pairs. FIG. 42 is a diagram showing a matrix operation expression for obtaining a position / direction. FIG. 43 is a diagram for explaining the creation of an address matrix. FIG. 44 is a diagram for explaining the calculation of the collation rate.

  Embodiments of a moving body position detection system and method according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments.

  As shown in FIG. 1, a moving body position detection system 10 according to the present invention is provided in a dot pattern 14 disposed on a floor surface (plane) 12 in a building and a moving body 16. The moving object position detecting device 18 moves on the surface 12. Here, the moving body 16 is, for example, a transport cart, a robot, a human being, or the like. The dot pattern 14 is randomly arranged on the floor 12 by means such as printing or drawing.

  The moving body position detection device 18 includes an imaging unit 20 that captures an area on the floor 12 where the moving unit 16 is located, a pole star feature amount storage unit 22, a pole star collating unit 24, a narrowing unit 26, and an imaging unit. Detecting means 28 for detecting the position of the moving body 16 based on the image captured by the means 20.

  The imaging means 20 is an image acquisition device such as a CCD camera. The imaging means 20 is fixed to the moving body 16, and while moving with the moving body 16, the area on the floor 12 where the moving body 16 is located is imaged at predetermined time intervals, and an image as shown in FIG. To get. In the present embodiment, images are acquired at a time interval of about 10 ms, for example, using a high-speed camera. As shown in FIG. 1, the distance between the imaging unit 20 and the floor surface 12 is several centimeters, and by using the illumination 30, an image with little influence of disturbance light can be acquired.

  As shown in FIG. 3, the pole star feature quantity storage means 22 calculates in advance the pole star feature quantity of all dots 32 in the dot pattern 14 on the floor 12 before detecting the moving body 16, Is stored in the pole star database 34.

The pole star collating unit 24 selects a predetermined dot 32a from a plurality of dots 32 in the acquired image, and previously calculates a pole star feature amount obtained from the dot 32a, a plurality of dots 32b _{1 to} 32b _{5 and the} like located in the vicinity. This is for collation with the created pole star database 34.

  Here, the pole star collating means 24 performs pole star collation using the pole star feature quantity for which the likelihood is set. Here, this likelihood is for absorbing an error between the dot in the image acquired by the imaging means 20 and the dot in the pole star database 34 of the pole star algorithm corresponding to this dot.

  The narrowing means 26 is for narrowing down the matching dot candidates collated by the pole star matching means 24 to a predetermined matching dot.

  The detecting means 28 detects the current position and direction of the moving body 16 on the floor surface 12 based on the positional relationship of the dot pattern consisting of the predetermined collation dots narrowed down by the narrowing means 26.

  Examples 1-4 of the present invention will be described below.

[Example 1]
As shown in FIG. 4, in the first embodiment of the present invention, an image is acquired and a point in the image is labeled (step S <b> 11), and the likelihood is introduced by the pole star matching means 24 using this image point group. Pole star collation is performed (step S12), the collation floor point candidate group is collated with a combination of distances between point groups by the narrowing means 26 (step S13), and the detection means 28 calculates the camera position and direction ( Step S14) is a procedure of confirming by the collation rate (Step S15).

Next, a specific processing procedure of the first embodiment will be described with reference to FIGS.
(1) Bit Representation of Pole Star Feature Amount As shown in FIG. 30, a pole star bit that represents a pole star feature is represented by a bit.

(2) Creation of Pole Star Database Matrix (hereinafter referred to as PSD Matrix) As shown in FIG. 31, the Pole Star feature amount of each database point (P1 to Pmax) is calculated. A PSD matrix is created by expressing each bit and arranging them vertically.

(3) Pole Star Collation of Image Points As shown in FIG. 32, pole star feature values are calculated for a certain image point.
Next, an operation as shown in FIG. 33 is performed using the PSD matrix described above. The number of matches between each database point and the arm length index of the image point of interest is calculated as a match arm counter. The element that takes the maximum value in the coincidence arm counter is the collation point.

(4) Addition of likelihood Since each actual image point has an error, the arm length index of the pole star feature amount is also different from the corresponding database point in the pole star database. In order to absorb this error, a likelihood (width) is given to the pole star database.

  For example, as shown in FIG. 34, it is realized by setting the width 1 corresponding to the likelihood to the left and right bits adjacent to 1 for the pole star bits of each database point of the PSD matrix.

  As a result, the error is absorbed, and even if the position of 1 of the pole star bit of the image point is shifted within the likelihood range due to the error, the position of the corresponding database point of the matching arm counter is the same as that at the time of positive matching. The number of arm length index matches is output. However, since the match of the arm length index increases as a whole, it is conceivable that a plurality of database points having the maximum number of matches appear in addition to the correct corresponding database points.

(5) Narrowing after Pole Star verification At the time of verification by the database, there is a possibility that a plurality of database points will be verified against one image point. In addition, there is a possibility that mistakes are included in the database points that are collated when there is dust or missing points. Therefore, between the image point group and the database point group that correspond correctly, the distance between any two image points and the distance between the two database points corresponding to the image point coincide with each other. Narrow down the correct matching points from the matching point candidates.

(5-1) Preparation of Image Point Cloud Data and Database Point Cloud Data As shown in FIG. 35, image point cloud data and database point cloud data are prepared so that the database point candidates and the image points are in a one-to-one relationship.

  Image points are set corresponding to database point candidates. That is, when there are a plurality of database point candidates for one image point, a plurality of the same image points are provided corresponding to the database point candidates.

  By calculating the mutual distance between the image points, a matrix imgArmLsMtx as shown in FIG. 36 is obtained. Also, the mutual distance between corresponding matching database points is calculated to obtain a matrix flrArmLsMtx as shown in FIG. Here, each element represents a distance between a point corresponding to a row and a point corresponding to a column. However, the distance between database points corresponding to the same image point is NA.

  Next, the distance difference is obtained from imgArmLsMtx and flrArmLsMtx, and the element is set to 1 when the distance is within the likelihood range. This calculation can be performed using, for example, a source code as shown in FIG. Note that the NA element is 0. By this calculation, a matrix MchArmLsMtx as shown in FIG. 39 is obtained.

  As shown in FIG. 40, when MchArmLsMtx is summed in the column direction, the number of the distance from each database point to the other database point starting from it corresponds to the distance from the corresponding image point to the other image point. Is calculated. Delete the rows and columns with the smallest number of distance matches in the matrix.

  When summing up again in the column direction, the same number of matches is calculated for the remaining database points. This operation is repeated until all the matching numbers are equal.

  When the mutual distance between the image points and the corresponding mutual distance between the database points all match, the number of matches is −1, and the set of the remaining image point group and database point group is the relative position of each point group. The relationship is considered to be geometrically consistent. This includes the case of mirroring.

(6) Calculation of Camera Position / Direction As shown in FIG. 41, the camera position / direction is calculated using the coordinate values of the image point group and the floor point group. The position / direction of the image coordinate origin as seen from the floor coordinate system is obtained from an arithmetic expression as shown in FIG.

The position and direction of the moving body from the position and direction of the image point group can be calculated by the following arithmetic expression.
Position = [x, y]
Direction θ = tan ⁻¹ ((−c1 × c2 / s1 × s2) ^1/2 )

(7) Calculation of collation rate (a) Creation of address matrix (offline)
As shown in FIG. 43, the floor is divided into square blocks in advance and addresses are assigned in units of blocks. The presence or absence of a point at each address is examined, a matrix corresponding to the address (hereinafter referred to as an address matrix) is prepared, and 1 is substituted if there is a point at the address, and 0 is substituted if there is no point. .

(A) Collation rate As shown in FIG. 44, all image points in the image are coordinate-converted to the floor coordinate system based on the obtained moving body position and direction. The ratio at which points exist in the address matrix corresponding to the converted coordinates is obtained and used as the matching rate.

  For this reason, according to the first embodiment of the present invention, it is possible to prevent the correct collation from being eliminated in the collation process by giving a wide range of the point distribution collation, and it is possible to prevent the erroneous collation or collation that may occur in the conventional algorithm. It is possible to make it difficult to cause a verification failure. In addition, by using a method to find a combination in which the distance between the point clouds in the acquired image and the distance between the floor matching point candidate groups all match, the image point cloud and floor matching point cloud pairs are extracted, and matching failure such as multiple matching Can be made difficult.

  Also, using the camera position direction detection value as the matching result, all the points (image coordinate system) in the image are coordinate-converted to the floor coordinate system, and the address matrix element corresponding to the coordinate position of each point is 1. A ratio (= collation rate) is calculated. If this ratio is less than the specified value, the reliability of the output value can be ensured by eliminating it as an uncertain detection.

  Further, in the present invention, since the spontaneous random point cloud pattern can be used for position detection, the following effects (1) and (2) are expected.

(1) Application to floor pattern by mixing chip material Chips are mixed into the flooring material even if it is difficult or costly to print on the floor, such as a painted floor that is frequently used in factories. Thus, by creating a random point cloud pattern, it can be used as a floor for position detection. Also, some floor tiles are of a type in which a chip material can be mixed in the manufacturing process and a random point cloud pattern can be created on the surface, but this is also applicable to such types of floor tiles.

(2) Correspondence to a wide floor Generally, in the case of tile printing, since the types of printing plates are limited, it is difficult to realize different point distribution patterns in a wide range. In a naturally occurring random point cloud pattern, a point distribution with no similarity can be realized in a wide range, so that position detection on a wide floor can be supported.

[Example 2]
The second embodiment of the present invention narrows down the floor area in the database having the correct candidate points by utilizing the fact that the correct candidate point group is distributed within the range of the point distribution being imaged.

  As a specific processing procedure, as shown in FIG. 5, in the second embodiment of the present invention, an image is acquired, and a point in the image is labeled (step S21). Pole star matching in which likelihood is introduced is performed by means 24 (step S22), and floor spot distributable areas are calculated by means of narrowing means 26 for collating floor point candidate group 1 (step S23) to obtain each distributable area ( In step S24), floor points in the distributable area are extracted (step S25).

  Then, the collation floor point candidate group 2 is collated with the combination of the distances between the point groups by the narrowing means 26 (step S26), and the camera position / direction is calculated by the detecting means 28 (step S27). The calculation and extraction of the maximum verification rate area are performed (step S28), the camera position / direction of the maximum verification rate area is obtained (step S29), and the maximum verification rate is confirmed (step S30).

  That is, the narrowing-down means 26 includes processing for obtaining a predetermined common area where the matching dot candidates exist for the matching dot candidates collated by the pole star matching means 24 and narrowing down to the matching dots existing in the common area. It is.

  By the way, in the processing of the above-described first embodiment, there is a possibility that the calculation memory required for the multiple matching may increase depending on the case. That is, when the number of image points in the acquired image is small or the point distribution of the image points is almost uniform, there may be a case where the number of arm length indexes constituting the Polestar feature amount is small. In this case, since many points in the database are candidates for verification, a large amount of calculation memory is required, and it may be difficult to continue the calculation substantially.

  Therefore, in the process of the second embodiment, the floor area in the database having the correct candidate points is narrowed down by using the fact that the correct candidate point group is distributed within the range of the point distribution being imaged. Other image points exist within a circle centered at an arbitrary image point and having a radius of a straight line to the image point farthest from the arbitrary image point. At this time, using the fact that the same relationship holds for the corresponding database points, the region where the corresponding database points exist is obtained, and the collation candidate database points are limited to those in that region, and narrowing down is performed. By doing in this way, the subject of said Example 1 can be solved.

[Common area for matching points]
For example, as shown in FIG. 6, if the distance from the image point C1 as the center is C4 and the distance is R, the database point f1 corresponding to C1 is centered within the circle of radius R. There are database points corresponding to other image points. Therefore, if the same other point group existence range radius is obtained for each image point, a circle centered on each corresponding matching candidate database point is drawn, and the sum area is set as the matching point existence area, the corresponding point correctly The group is included in the region where the overlapping of the matching point existence regions with respect to the respective image points is overlapped most (= the matching point existence common region).

  By obtaining this common region for the matching point candidates after the processing of the pole star matching + likelihood in step S22 and narrowing down to points existing in the common region, the matching point candidates can be greatly reduced.

[Setting the number of overlaps]
In addition, when dust points are mixed in with image points, there is an erroneous matching point existence area, and in the largest overlapping area, a correct corresponding database point may be excluded. Is set to be smaller than the number obtained by subtracting the number of dust that will be mixed from the number of image points.

[Handling multiple common areas]
There may be a plurality of collation point existence common regions. In that case, a collation candidate point group is individually obtained for each region, and a collation rate is calculated. Finally, based on the calculation result of the matching rate in each region, the matching point group in the region with the highest matching rate is set as the answer.

According to the second embodiment, an increase in necessary calculation memory can be suppressed due to multiple matching, and therefore the following effects can be obtained.
(1) Position detection for a wide floor range is possible.
(2) Position detection is possible even with image point input with a relatively large position error.
(3) Position detection is possible even with a camera with low image resolution.

[Example 3]
In the third embodiment of the present invention, the floor area of the database to be collated is divided in advance, and collation is performed within the divided area. After collating all the divided areas, the result of the area with the highest collation rate is output as a correct answer based on the calculated collation rate in each divided area.

  As a specific processing procedure, as shown in FIG. 7, an image is acquired and labeled at a point in the image (step S31), and the search floor area is divided by the pole star matching means 24 (step S32). Pole star matching in which likelihood is introduced for each search area is performed (steps S33 and S34).

  Then, the collation floor point candidate group 2 is collated with the combination of the distances between the point groups by the narrowing means 26 (step S35), the collation rate is calculated, and the maximum collation rate area is extracted (step S36). The procedure is to obtain the camera position / direction in the area of the maximum verification rate by means 28 (step S37) and confirm the maximum verification rate (step S38).

  In other words, the pole star collation means 24 divides the collation range in the pole star database 34 into a plurality of areas, performs pole star collation in each area, and obtains an area where the collation rate is maximized.

  By the way, in the processing of the above-described first embodiment, there is a possibility that the calculation memory required for the multiple matching may increase depending on the case. That is, when the number of image points in the acquired image is small or the point distribution of the image points is almost uniform, there may be a case where the number of arm length indexes constituting the Polestar feature amount is small. In this case, since many points in the database are candidates for verification, a large amount of calculation memory is required, and it may be difficult to continue the calculation substantially.

  However, in the processing of the third embodiment, the floor area of the database to be collated is divided in advance, collation is performed in the divided area, and after the collation of all the divided areas, the calculated collation rate in each divided area is calculated. Originally, the result of the area with the highest collation rate is output as a correct answer, so that the problem of the first embodiment can be solved.

[Split floor area division]
As shown in FIG. 8, the target verification floor area is divided into an appropriate size. For example, the size is about 25 times the image field of view.

[Verification and results]
In each verification floor division area, pole star verification + likelihood and inter-point distance combination verification are performed, and a verification rate is calculated. The detected value of the area where the maximum matching rate is obtained is taken as the answer.

According to the third embodiment, an increase in necessary calculation memory can be suppressed due to multiple matching, and therefore the following effects can be achieved.
(1) Position detection for a wide floor range is possible.
(2) Position detection is possible even with image point input with a relatively large position error.
(3) Position detection is possible even with a camera with low image resolution.

[Example 4]
The fourth embodiment of the present invention is a method that combines the division of the verification floor area of the third embodiment and the narrowing down of the point cloud presence common area of the second embodiment, and solves the problem of the first embodiment. Can do.

  As a specific processing procedure, as shown in FIG. 9, the image is acquired and labeled at a point in the image (step S41), and the search floor area is divided by the pole star matching means 24 (step S42). Pole star matching in which likelihood is introduced for each search area is performed (steps S43 and S44). The floor spot distribution area is calculated by the narrowing means 26 for the verification floor point candidate group 1 (step S45), each distribution area is obtained (step S46), and the floor points in the distribution area are extracted (step S47).

  Then, the combination of the distances between the point groups is collated for the collation floor point candidate group 2 (step S48), the camera position / direction is calculated by the detecting means 28 (step S49), and the collation rate calculation and the maximum collation are performed. The rate area is extracted (step S50), the camera position / direction of the area with the maximum matching rate is obtained (step S51), and the maximum matching rate in the search area and the camera position / direction at that time are stored (step S52). , S53), and the maximum collation rate is confirmed (step S54).

  That is, the collation floor area is divided by an appropriate size, and for the collation of each collation floor divided area, the point cloud existence common region (plurality) is calculated after the pole star collation considering the likelihood. The point group mutual distance combination matching is performed on the matching candidate point groups in each point group existing common area, the position / direction of the camera is calculated, and then the matching rate is calculated. The calculation is performed for all point cloud existence areas, and the maximum collation rate in the corresponding collation division area and the camera position / direction at that time are obtained. This is performed for all the collation division areas, and finally, the position / direction of the camera having the highest collation rate in all the collation division areas is set as the detection result.

According to the fourth embodiment, an increase in necessary calculation memory can be suppressed due to multiple matching, and therefore the following effects can be achieved.
(1) Position detection for a wide floor range is possible.
(2) Position detection is possible even with image point input with a relatively large position error.
(3) Position detection is possible even with a camera with low image resolution.

  Next, application examples 1 and 2 of the present invention will be described. The moving body position detection system 10 according to the present invention can be applied to a transport system in a hospital, a factory, or the like.

[Application 1 of the present invention]
First, as application example 1 of the present invention, a case of applying to a hospital delivery service using a robot will be described.

  First, the problems of the prior art will be described. Conventionally, when a hospital inpatient wants to purchase a product such as a magazine, the patient usually moves to a store in the hospital and purchases it. However, in a large hospital, the distance from the hospital room where the patient is located to the store is large. It may be far away and is not convenient for the patient. In some cases, the patient cannot move on his own. There are hospitals that have introduced delivery services on behalf of purchases and deliveries at retail stores to deal with patients in such an environment. However, in such hospitals, there have been problems such as incurring labor costs for services and the limited number of people who can enjoy services, and being unable to deliver quickly to patients.

  Then, the mobile body position detection system 10 which concerns on this invention is applied to this delivery service, and the new service with respect to a patient is provided. Specifically, the mobile body 16 in the mobile body position detection system 10 is used as a commodity delivery robot, and the patient orders a product to be delivered from a terminal device provided on the bedside of the hospital room. Then, the robot of the store moves to deliver the ordered product to the bedside where the terminal device is placed. This movement is performed when, for example, a movement route from a store to a terminal device is linked to each terminal device using a dot pattern drawn on the floor, and an order signal is received from a predetermined terminal device. This is possible by setting the robot to automatically move to the predetermined terminal device. This allows the robot to enter the bedside of the hospital room. In addition, you may set freely the rule regarding the movement route and movement of a robot suitably.

  Thus, when the present invention is applied to a delivery service in a hospital, the robot can be guided with a dot pattern drawn on the floor surface. For this reason, a large-scale construction for laying a guide line for guiding a delivery robot, which has been necessary in the past, in a corridor or a hospital room is not necessary. In addition, since guidance is performed using a dot pattern drawn on the floor surface, it is possible to flexibly cope with future layout changes in the building. In addition, since the robot is delivered instead of the person, the labor cost is not incurred compared to the conventional hospital delivery service, and the running cost can be suppressed. Furthermore, by shortening the delivery distance by the robot, the lead time can be shortened and the patient's satisfaction can be improved.

[Application 2 of the present invention]
Next, as application example 2 of the present invention, a case where it is applied to flat warehouse free location management will be described.

  First, the problems of the prior art will be described. Conventionally, the following techniques are mainly known as inventory location management methods in the case of storing pallets or the like without using a fixed shelf in a flat warehouse.

(1) As shown in FIG. 10, the line 50 is drawn on the floor, the area is divided and the storage area ID is recognized by the pillar number or the like matched with the pillar 52, and is linked to the luggage.

(2) As shown in FIG. 11, a grid is set in accordance with a package 60 (such as a pallet) in a predetermined packing form, and an ultrasonic transmitter 62 that transmits a unique ID is installed on the ceiling. The ID transmitted from the ultrasonic transmitter 62 on the ceiling when the load 60 is placed on the grid by the forklift 70 is read by the reader 64 installed on the forklift 70, and linked to the load.
(3) Similar to (2) above, but as shown in FIG. 11, the passive RFID 66 is embedded in the floor instead of the ultrasonic transmitter 62, and the ID is read by the reader 68 installed on the forklift 70.

  However, in the above (1), since there are a plurality of packages in the area, the detailed position of the packages cannot be quickly grasped. Further, since manual work is involved, there is a risk that a location error due to human error may occur.

  Further, in the above (2) and (3), as shown in FIG. 12 (a), it is necessary to arrange the luggage in a certain direction. This is because there is a difference in the positions of the luggage 60 and the leader 64 (68) attached to the forklift 70. As shown in FIG. 12B, when placed from the side, the positions of the ultrasonic transmitter 62 and RFID 66 and the readers 64 and 68 are displaced and cannot be handled. Furthermore, as shown in FIG. 13 (a), the package must have the same form, and as shown in FIG. 13 (b), the dimensions differ with respect to the arrangement of the ultrasonic transmitter 62 and the RFID 66. It is not possible to handle different types of luggage such as pallets and pallets and basket cars.

  Therefore, the moving body position detection system 10 according to the present invention is applied to this flat warehouse free location management. Specifically, as shown in FIGS. 14 and 15, the moving body 16 in the moving body position detection system 10 is a forklift 70 or a person, and a camera (imaging means 20) is attached thereto. On the other hand, the dot pattern 32 is drawn on the floor surface or the ceiling surface, and the position and orientation of the forklift 70 and the person are detected from the image captured by the camera 20 to automatically recognize where the luggage is placed in the flat warehouse.

  Also, the distance between the plane center of the luggage 60 and the camera 20 is set to a predetermined value in advance, and the center of the luggage is recognized by the captured image of the camera 20, as shown in FIGS. In addition, the position of the load can be recognized from any direction, and loads having different forms can be placed in the flat warehouse from various directions. For this reason, the freedom degree of the utilization mode of a flat warehouse increases, and the efficiency of warehouse operation can be achieved. Further, in the above-described conventional technology, it is necessary to install ultrasonic transmitters or RFIDs corresponding to the number of pallets at the time of construction and register IDs and locations, but this need not be performed.

  As described above, according to the present invention, there is provided a moving body position detection system that detects the position of a moving body that moves on a plane, the dot pattern including a plurality of dots arranged on the plane, and An imaging unit that is provided in the moving body and images a region on a plane on which the moving body is located, a dot in an image acquired by the imaging unit, and a dot in a pole star database of a pole star algorithm corresponding to the dot A pole star matching unit that performs pole star matching using a pole star feature amount that is set with a likelihood for absorbing an error between and a matching dot from a matching dot candidate verified by the pole star matching unit Based on the positional relationship between the narrowing-down means narrowed down to a predetermined pattern and the dot pattern composed of the predetermined collation dots narrowed down by the narrowing-down means. Te, and a detection means for detecting a current position and direction of the moving body. For this reason, it is difficult to cause a verification failure when the Polestar algorithm is used, and the reliability of detection can be ensured.

  As described above, the mobile object position detection system and method according to the present invention are useful for detecting the position of a mobile object placed on a plane such as a floor surface in a building, and in particular, the pole star algorithm as a position detection algorithm. It is suitable for ensuring the reliability of detection by making it difficult to cause a verification failure when using.

10 Moving body position detection system 12 Floor (plane)
DESCRIPTION OF SYMBOLS 14 Dot pattern 16 Moving body 18 Moving body position detection apparatus 20 Imaging means 22 Pole star feature-value storage means 24 Pole star collation means 26 Narrowing means 28 Detection means 30 Illumination 32 Dots 34 Pole star database

Claims (4)

A moving body position detection system for detecting a position of a moving body moving on a plane,
A dot pattern consisting of a plurality of dots arranged on the plane;
An imaging means provided in the moving body for imaging a region on a plane on which the moving body is located;
Pole using a pole star feature that sets the likelihood to absorb the error between the dot in the image acquired by the imaging means and the dot in the pole star database of the pole star algorithm corresponding to this dot Pole star verification means for performing star verification,
Narrowing means for narrowing down the matching dot candidates collated by the pole star matching means to a predetermined matching dot;
Detecting means for detecting a current position and direction of the moving body based on a positional relationship of a dot pattern made up of predetermined matching dots narrowed down by the narrowing means;
A moving body position detection system comprising:   The narrowing-down means includes a process of obtaining a predetermined common area where the matching dot candidates exist for the matching dot candidates collated by the pole star matching means, and narrowing down to the matching dots existing in the common area. The moving body position detection system according to claim 1.   2. The pole star collation means divides a collation range in the pole star database into a plurality of areas, performs pole star collation in each area, and obtains an area where a collation rate is maximized. The moving body position detection system according to 2. A moving body position detection method for detecting a position of a moving body moving on a plane,
A dot pattern consisting of a plurality of dots is arranged on the plane,
The moving body is provided with imaging means for imaging a region on a plane where the moving body is located,
Pole using a pole star feature that sets the likelihood to absorb the error between the dot in the image acquired by the imaging means and the dot in the pole star database of the pole star algorithm corresponding to this dot Star matching is performed, the collation dot candidates that have been collated are narrowed down to a predetermined collation dot, and the current position and direction of the moving body are detected based on the positional relationship of the dot pattern that includes the narrowed down collation dots. A moving body position detection method.

Images