JPS62269287A - Visual recognizing device - Google Patents

Abstract

PURPOSE:To accurately recognize an object even when both end points of a constituent segment is not obtained exactly but obtained as interrupted segments by making an area determined by positional relation of a segment and corresponding indeterminate method as the base of degree of collation. CONSTITUTION:The degree of collation arithmetic section D consists of the length degree of collation arithmetic unit 12, an attitude candidate arithmetic unit 13, an attitude deciding device 14, an existence possible area arithmetic unit 16, the degree of collation arithmetic unit 17, a collation success deciding device 18, a constituent straight line set storing memory 109, a model straight line set storing memory 110, an attitude possible storing memory 111 and an existence possible area storing memory 112. When cumulative degree of collation exceeds a preset threshold level, it is recognized as the success of collation, and a starting signal is outputted to a collation after processing section. If cumulative degree of collation does not exceed the threshold level, the obtained correspondent set is additionally stored in a correspondent set list memory 107 together with the cumulative degree of collation. A starting signal S3 is outputted to a constituent straight line selecting device 11, and next correspondent set is formed.

Description

[Detailed description of the invention] [Industrial application field]

The present invention relates to a device that accurately identifies the outline of a desired object from an image containing the detected object. Since the present invention can easily determine the shape, position, orientation, etc. of a detected object from the identified contour, it can be applied to, for example, a visual recognition device for a robot.

[Prior art]

Conventionally, one of the methods of recognizing a predetermined object by capturing an image of the object to be detected and processing the obtained image data.
One known method is to compare feature points of an object with those of a model (Japanese Unexamined Patent Publication No. 111575/1982). In this method, corners of the outline of the image are selected as feature points, a distance map between the feature points is calculated, and the distance map between the feature points is compared with the distance map between the feature points in the model. Based on the degree of agreement, detection is performed. This method detects feature points on an object and recognizes the desired object. Other methods include extracting contour lines from images, approximating the contour lines with primitives such as straight lines, circles, and ellipses, calculating distance maps that represent the positional relationships between these primitives, and comparing this with the distance map of the model. A method (Japanese Unexamined Patent Publication No. 60-204086) is known in which a desired object is recognized based on the degree of coincidence.

[Problems to be solved by the invention]

The first method has a problem in that it is difficult to accurately detect feature points such as corners, the outline of a circle, and the center of a circle, and therefore the recognition accuracy is not good. In the second method, in order to accurately approximate the contour line with primitives such as straight lines, circles, and ellipses, it is necessary to extract the contour line accurately. There is a problem in that maintenance must be carried out strictly. For example, even if the illumination and background are devised and image processing is performed with high accuracy, the contour line will not be continuous, but will be cut at several points. For this reason, feature points such as the ends of the line segment cannot be accurately grasped, and the accuracy of object recognition cannot be improved. Furthermore, in robot control, in order to accurately recognize the position and orientation of an object, a part of the object may be zoomed and imaged. In such a case, it is difficult to apply a distance map because only a part of the outline of the object can be obtained as an image. In this way, the conventional method is to extract feature points representing a part of the contour, obtain a distance map between the feature points, and compare it with that of the model. Therefore, the accuracy of the object fJLt depends on how accurately the feature points are grasped. However, since feature points are understood as points, they are directly affected by the accuracy of contour line extraction. Therefore, in reality, when a contour line can only be understood as a broken line segment, there is a problem in the accuracy of object recognition. The present invention has been made to solve the above problems, and its purpose is to ensure that the obtained contour line is
If the object is a broken object and the points at both ends of the line segment that make up the contour cannot be accurately grasped, or if only a partial image of the object to be detected is obtained, the object to be detected may be different from other objects. It is necessary to accurately recognize the location, orientation, etc. of a target object even when the objects are superimposed.

[Means and operations for solving the problems] The structure of the invention to solve the above problems is as follows. That is,
The present invention provides a constituent line segment extraction means for extracting a plurality of constituent line segments constituting a contour line from a two-dimensional image and obtaining information regarding the position of the constituent line segments, and a model to be a target of matching of a detected object. model line segment storage means that stores information regarding the positions of model line segments that constitute at least the main part of the contour line; and extracted component line segments;
A visual recognition device comprising a matching means for matching a model line segment with a model line segment and identifying the component line segment corresponding to the model line segment based on the degree of matching; When calculating the degree of matching between a line segment and the component line segment 1, it is determined based on the positional relationship between the specific model line segment, which is the model line segment 1 that is the target of matching, and other model line segments. , when the model line segment of the above-mentioned gender is made to correspond to the corresponding other constituent components, a range in which the specific model line segment can be located due to the uncertainty of the correspondence is determined, and from this range, the specific model line segment is found. Existence possibility region calculation means for calculating the existence possibility region of component line segments corresponding to the model line segment, and existence possibility region calculating means for calculating the degree of matching in matching according to the extent to which the component line segments exist in the existence possibility region. The apparatus is characterized in that it has a position determination means. Here, constituent line segments and model line segments include not only straight lines but also curves such as circular arcs, elliptic curves, hyperbolas, and parabolas.
It can also be constructed simply from straight line segments. Information regarding the position includes, in addition to the position coordinates of both ends, if it is a straight line segment.
It can be expressed as a set of position coordinates of one end or midpoint, length, and inclination angle. Further, if it is a curved line segment, it can be expressed by a set of the type of curve, position coordinates of both ends, and arbitrary position coordinates on the curve. The possible existence area of a specific model line segment determined by the existence possibility area calculation means is conceptually illustrated in FIG. 2. (a) As shown in the figure, the model line segment is (ff1l
, mztm3. m4. m5), and the constituent line segments (e+ to e, 4) are extracted. Now, it is assumed that the model line segment m+ and the component line segment e1 correspond to each other with an error of length ΔL as shown in FIG. Generally, there are four ways to correspond between the model line segment m+ and the component line segment e. That is, the method is to make the end point A of the model line segment correspond to the end point C of the constituent line segment, and similarly, the method is to make the end point B correspond to the end point, the end point A to the end point, and the end point B to the end point C. This inner cutting 2
are illustrated in (c) and (d). As is clear from the figure, the model line segments m2 to m are Δ
This means that there is uncertainty in the position of L. Therefore, if we pay attention to the model line segment m3 (the model line segment that is the target of matching and is focused on is referred to as a "specific line segment"), the range in which the specific model line segment m can be located is the constituent line segment e1.
When fixed as a reference, the range shown by 81 and 82 is obtained as shown in FIG. Then, the constituent line segments corresponding to the specific model line segment m must exist at least within this range. Therefore, the constituent line segments that exist in this area can correspond to this specific model line segment, and conversely, the constituent line segments that do not exist in this range are considered not to correspond to this specific model line segment, and are considered to be the specific model line segments. It is possible to calculate the degree of coincidence of the constituent line segments with respect to the segment. The above explanation assumes that the model line segment m1 and the component line segment e1 correspond, and calculates the degree of coincidence between the model line segment m and the corresponding component line segment. Assuming that the component line segment (mI + m 2 + m 31 m 4) corresponds to the component line segment (e l + e 21 e 3. e 4), the degree of coincidence of the component line segment that is matched with the specific model line segment m. The possible range S of the above-mentioned specific model line segment m6 when calculating is calculated as follows. However, N (m+, e+,; ms
) is the possible range of the specific model line segment m when the model line segment m is made to correspond to the constituent line segment e. Therefore, S becomes their common part. The combination of model line segments and component line segments is created by arbitrarily matching the number of component line segments corresponding to the number of model line segments from the beginning to generate correspondence sets, and then for each set of specific model line segments. Find the above range, calculate the degree of matching of the matching for that specific model line segment, and calculate the matching of the entire set from the sum of the matching degrees of matching when changing the specific model line segment sequentially. There is a method to calculate the degree of In addition, as explained in the example, while sequentially selecting model line segments and calculating the degree of matching in matching, further model line segments are The present invention can also be applied to an efficient method of expanding correspondence sets in a tree-like manner by selecting . The range in which the above specific model line segment can exist can be used as it is as the existence possibility area of the corresponding component line segment, but it is possible to add a tolerance range to the above range and make this the existence possibility area. You can also do it. Further, the method for determining the possible range of the above-mentioned specific model is explained in a somewhat limited manner to facilitate understanding, but the method is not limited to the method of determining based on the above-mentioned correspondence relationship. The point is, when one model line segment is associated with one component line segment with a certain degree of uncertainty, the range in which a specific model line segment can exist is determined by the relative positional relationship between the model lines. . In addition to the determination based on the above-mentioned existence possibility region, based on the relative posture between model line segments, when one model line segment is made to correspond to one constituent line segment, the possible posture of a specific model line segment is determined, By taking into consideration the degree of coincidence with this orientation of the constituent line segments to be verified, the reliability and speed of verification are further improved. Furthermore, if the degree of agreement between the lengths of the model line segment and the constituent line segments is calculated, and this is added to the degree of agreement in the above-mentioned comparison for determination, the recognition accuracy can be further improved. As described above, in the device of the present invention, the region in which the constituent line segments corresponding to the specific model line segment should exist is calculated by the action of the existence possibility region calculating means of the matching means, and the region in which the constituent line segments that should correspond to the specific model line segment should exist is calculated by the action of the existence position determining means. , the degree of matching between the specific model line segment and the corresponding constituent line segment is calculated according to the extent to which the constituent line segment exists in the area. Therefore, we introduce the concept of existence possibility region to determine the degree of matching,
In other words, since the determination is made based on a two-dimensional feature called area, it is not necessarily necessary to accurately know one-dimensional features such as both end points of the constituent line segments. Therefore, even if the constituent line segments can only be obtained in pieces, or only constituent line segments related to a part of the contour of the detected object can be obtained, the relative positional relationship between the constituent line segments may not be included as information. if,
The above existence possibility region can be calculated, and the detected object can be accurately recognized aF1.

【Example】

The present invention will be described below based on specific examples. As shown in FIG. 11, the device R of this embodiment is configured in advance to a robot control device T that detects the position and orientation of a detection object Q and controls the orientation of a robot S that uses the object as a work target. The present invention relates to a visual recognition device that outputs displacement data of the position and orientation of a detected object with respect to the position and orientation of a model, and is used to correct the taught path of a robot. (1) Overall configuration The present device is roughly divided into five parts, as shown in FIG. That is, by capturing an image of the detected object and processing its video signal, straight line segments that constitute the outline of all objects including the detected object (hereinafter, constituent line segments composed of straight line segments are referred to as " A constituent straight line extraction unit A extracts constituent straight lines (hereinafter referred to as "constituent straight lines") and stores information regarding their positions; A modeling section B creates and stores information regarding the position of a "model straight line", which is composed of straight line segments, and selects the model straight line and the constituent straight lines sequentially according to a predetermined logic. A correspondence set generation unit C generates a pair with a constituent straight line (hereinafter simply referred to as a "correspondence set") and a degree of matching between the newly selected model straight line and the corresponding constituent straight line (hereinafter simply referred to as a "matching degree"). ), and a post-verification processing section E that calculates the position and orientation of the detected object based on the results of the verification. In this embodiment U, all the model line segments and constituent line segments are straight lines. Furthermore, the generation of correspondence pairs follows a sequential logic in which the branch with the highest past matching degree is selected along a search tree that sequentially increases the number of model straight lines to be matched, as shown in FIG. Then, when the degree of matching exceeds a predetermined value, it is assumed that the detected object has been recognized. (2) The logical model straight line of the corresponding coarse generation is (ml + m2rm
s, m +) is selected, and the constituent line is (e+
−eth) is extracted. The model straight line is given a weight value (2ω3ω4 of ω8) depending on the importance of recognition. In the first step, the first model straight line is selected. The one with the largest weight value is selected. This weight value of 172 is the accumulated matching degree (in this embodiment, since the matching degrees for the specific model straight line obtained at each step are accumulated and used as the matching degree for the entire corresponding set, this accumulated matching degree is When this is particularly distinguished, it is used as the initial value of the "cumulative matching degree"). In the figure, the cumulative matching degree is shown as a value within 0. Next, in the second step,
The constituent lines (e+ to e14) are made to correspond to the selected model straight line m, to form a correspondence set (hereinafter, when the correspondence set generated in the n-th step is particularly distinguished, the "n-th correspondence set") Then, for all of these quadratic correspondence sets, the matching degree for the specific model straight line is calculated by the matching degree calculation unit, and the matching degree is added to the cumulative matching degree, and a new cumulative matching degree is generated. is calculated.Then, among the correspondence sets generated in the past (primary to secondary correspondence sets), the correspondence set with the highest cumulative matching degree (m In 814>) is selected.Next, in the third step A cubic correspondence set incorporating the second model straight line is generated. At this time, 1/2 of the weight value of the incorporated model straight line is added to the cumulative matching degree. The correspondence set with the maximum cumulative matching degree (m l+
e l 4 + m 2) is selected. Next, in the fourth step, similarly to the second step, a quaternary correspondence set incorporating the constituent straight lines is generated, and the cumulative matching degree is calculated. Thereafter, in the same way, new model straight lines are incorporated in odd-numbered steps, constituent straight lines are incorporated in even-numbered steps, and the corresponding set with the maximum cumulative matching degree is selected, in the direction that all model straight lines are selected. Progress is made along the search tree path. (3) Constituent straight line extraction unit The constituent straight line extraction unit A includes a camera, an image input device 1, a differentiator 2, a contour detection device 3, a contour line broken line approximation device 4, and a memory 101 to 101 for storing output data of each device. It consists of 104 pieces. (a) An image input device scans an electronic image of an object captured by a camera, extracts a video signal, samples it corresponding to the pixel coordinates of the light-receiving surface, digitizes it to 256 brightness levels, and converts it into pixels in the image memory 101. This is a device that creates a grayscale image by storing it in an address corresponding to the coordinates. (b) Differentiator This is a device that calculates the brightness gradient (gradient end) at each pixel coordinate of the grayscale image stored in the image memory 101 and stores the result in the image memory 102 to create a differential image. The brightness gradient calculation method is 3× including the moving point P on the pixel coordinates.
The brightness of each pixel divided into three meshes is used for calculation using a Sobel operator. Here, the Sobel operator means the brightness gradient (fX
, fy) are calculated as follows. f, = c + 2 f + i − (a + 2
d + g ) −(2) f y= g + 2 h
+ i −(a + 2 b + c) −(3
) This kind of differential operation using the Sobel operator is
It consists of a product-sum calculator. (C) Contour extraction device This device inputs a differential image from the image memo IJ 102, detects a contour, and stores the pixel coordinates of points on the contour in the contour list memory 103. It is known that the absolute value of the brightness gradient of a point on the contour line is larger than that of the surrounding points. Therefore, the sequence of points constituting the contour line is determined by detecting a point where the absolute value of the brightness gradient at the moving point P on the pixel coordinates is larger than a predetermined value and not smaller than that of the surrounding points, and starting from that point. Then, the adjacent pixels in the direction perpendicular to the direction of the brightness gradient are selected as points on the contour line adjacent to that point, and the next point on the contour line is selected based on the brightness gradient of that point based on the selected point. It can be obtained by sequentially executing operations. (d) Contour line broken line approximation device Input a sequence of points on the contour line from the contour line list memory 103, trace the contour line from the point sequence data, approximate it to a broken line, and configure pixel coordinates at both ends of each broken line. This is a device for creating a straight line list memory 104. The polygonal line approximation is about points in the contour list memory,
This is done by defining a series of points that constitute the same straight line as a series of points whose direction vectors along the contour line of the points are equal within a tolerance range. Here, the direction of the direction vector along the contour line means that any point in the contour list is PI(Pi", Pi')
Then, in the contour list, the points that are before and after PL by a are P,, -, (P, -♂, Pt-&tsu) and P, and a (
To Plea. Plea', γ), it is given as Plea'Pi-a'Plea' - Pt-a''. A broken line node is detected as a point where the direction vector at a point on the contour line changes by more than a predetermined value. The straight lines (constituent straight lines) obtained in this way as shown in Fig. 6 are specified by the pixel coordinates of both end points, and are stored in the constituent straight line list memory as follows. × 11ylI×2blindy2'X+ko'Y+2X22yx'
---,,,,-xl″y1″×21′y2゜Here, (X+' * ”!+' )+ (X+' +
y+') is the coordinate of the end point of the i@th constituent straight line e1. (4) Modeling unit This is a device that creates and stores information regarding the position of a straight line (model straight line) that constitutes the main part of the contour line in a model that is a target of matching of detected objects. This data is input by moving a cursor on the CR7 screen to generate a model straight line m, and specifying the position of the end point of the straight line. When an end point is designated, the pixel coordinates of that point are automatically generated and stored in the model straight line list memory 105. Further, a weight value ω1 is set for the model straight line, and it is stored in the model straight line weight value list memory 106. This weight value indicates the degree of matching for one model straight line. Therefore, the weight value is set larger for a straight line that better represents the form of the model and can identify an object if the straight line is recognized. In addition, when it is particularly important to recognize the ridgeline of an object to be grasped by a robot,
The weight value is set larger for a model line with higher recognition importance. Furthermore, when a constituent straight line of good quality is stably detected in an image of an object, the weight value of the model straight line corresponding to this straight line is set to a large value. By setting the weight values in this way, matching accuracy can be improved. In addition, the model straight lines are selected in order of increasing weight value,
The matching set with the highest cumulative matching degree is quickly selected, and when the cumulative matching degree reaches a predetermined value, the matching is considered successful and the matching operation is aborted, so by setting the weight value as above, , it is possible to improve the matching speed. Due to the action of the modeling unit B, information regarding the position of the model straight line as shown in FIG. 5 is inputted with respect to the contour line of the model shown in FIG. (5) The corresponding group generation unit C includes a corresponding group list initialization device 6, a model straight line selection and list creation device 9, and a constituent straight line selection device 11.
, maximum matching degree matching pair selection device 8, matching pair list memory 1
07. It consists of a maximum matching degree correspondence group storage memory 108 and a status register 7.10. The correspondence set generation section C is a device that generates and selects correspondence sets according to the logic shown in FIG. 3, and outputs them to the matching degree calculation section. (a) Corresponding group initialization device Inputs a signal from the contour line broken line approximation device 4 to the effect that the generation of the constituent straight lines has been completed, clears the corresponding group list memory 107, and turns on the bit of the status register 7. This is the device to be configured. (b) Correspondence set list memory This memory 107 stores the correspondence sets generated by the model straight line selection and list creation device 9 and the matching degree (processing results of odd-numbered steps in FIG. 3), and the matching degree calculating device 17. The degree (the degree of matching calculated by the even-numbered step processing in FIG. 3) is written. It is also a storage device accessed by the maximum matching degree correspondence group selection device 8. (C) Maximum matching degree correspondence set storage memory This memory 108 stores the selected matching set having the maximum matching degree and its matching degree in order to select the next search branch in FIG. It is a device. (d) Model straight line selection and list creation device This device 9 includes:
This is a device that generates a corresponding set of odd-numbered steps shown in FIG. 3 and calculates the degree of matching. That is, the maximum matching degree corresponding group selection device 8
, inputs the activation signal S1 from , selects an unselected model straight line from the model straight line list memory 105 for the corresponding set (parent node) stored in the maximum matching degree corresponding set storage memory 108, and selects the unselected model straight line from the model straight line list memory 105. This is a device that adds and creates a new matching degree to the correspondence set list memory. In the first step, all model straight lines m, are selected,
The cumulative matching degree ρ is initially set as 172 for each weight value ω1. In general, the weight values are ω
1~(DH, and the corresponding model straight line is city,~i
? i, 4, and the model straight line selected so far is Nae,
~Seedling 3, assuming that the constituent straight line of the selected correspondence set with the largest cumulative matching degree is 61~, then the next correspondence group and the cumulative matching degree are stored in the maximum matching degree correspondence group storage memory 108. There is. (City 1 Nae 2. calligraphy, calligraphy; ρ81.)
-(4) Therefore, the newly generated correspondence set and matching degree are as follows. (Seedling, - Seedling 2. Seedling 2. δ1-・δ,; ρ., 1)
-(5) However, i=P+1, which is a bar. Then, the new odd-numbered step cumulative matching degree ρ. , I are calculated using the following equation using the cumulative matching degree ρ61 of the previous even-numbered step and the weight value f. ρ0・1: ρi, p+necessary 1/2 −(
6). When a correspondence set including the cumulative matching degrees of odd-numbered steps is generated in this way, the bit of the status register 10 is reset to indicate that the processing of the apparatus 9 is completed. (e) Constituent line selection device This device 11 receives the activation signal S2 from the maximum matching degree corresponding group selection device 8, and selects the maximum matching degree corresponding group storage memory 10.
This is a device for generating the next even-numbered step correspondence set for the correspondence set stored in 8 (the correspondence set with the maximum matching degree for the odd-numbered steps in FIG. 3). Memory 108.104
The constituent straight lines and model straight lines selected from are output to memories 109 and 110, respectively. The following correspondence sets are stored in the R high matching degree correspondence set storage list memory 108. (ilil+, seedling 1. seedling P+l+E+ ・・δP
) - (7) Therefore, the next correspondence set is sequentially generated from this correspondence set. (Flute, City 1. City P+1.6.-δ1.δt)
-(S) However, i=P+1, -M. When calculating the matching degree in the second step, since there is only one model straight line selected, the posture candidate and existence possibility region cannot be calculated, so the length matching degree calculation device 1
・A signal to activate only step 2 is output, and in the case of other steps, a signal to activate length matching degree calculation device 12,
Posture candidate calculation device 13 when calculating the matching degree of the first correspondence set.
A signal for activating the existence possibility region calculation device 15 is output. The matching degree calculating section outputs the corresponding pairs generated in the matching group list memory 107 and the cumulative matching degree every time the matching degree is calculated. When all matching degrees have been calculated, a bit in the status register 7 is set to indicate that the processing of the constituent straight line selection device 11 has been completed. (f) Maximum Matching Degree Corresponding Group Selection Device This device 8 is a device that selects the corresponding pair having the maximum cumulative matching degree from among the corresponding pairs generated so far in each step. That is, from among the correspondence sets stored in the correspondence set list memory 107, the correspondence set with the largest cumulative matching skin is selected, and that correspondence set is stored in the maximum matching degree correspondence phase storage memory 108.
is memorized. Furthermore, the selected correspondence group is deleted from the correspondence group list memory to prevent the same correspondence group from being selected. When the correspondence list memory is empty in a state other than the initial state, even though the cumulative matching degree has been calculated for all correspondence pairs,
This means that the cumulative matching degree has not reached a predetermined value, and in this case, a matching failure signal is output. Further, this device distinguishes between even-numbered steps and odd-numbered step processing according to the state of the bits in the status register 7.8, and starts up the device 9 and the device 11 alternately. (6) Matching degree calculation unit The matching degree calculation unit includes a length matching degree calculating device 12, a posture candidate calculating device 13, a posture determining device 14, an existence possibility region calculating device 15, an existence position determining device 16, and a matching degree calculating device. Device 17
, matching success determination device 18 , constituent line set storage memory 109
, a model straight line set storage memory 110, a posture candidate storage memory 111, and a possibility region storage memory 112. This device calculates the matching degree for a specific model straight line for the combination of the constituent straight line and model straight line outputted by the constituent straight line selection device 11 to the constituent straight line group storage memory 109 and the model straight line group storage memory 110, and adds it to the cumulative matching degree. It is a device that does If the cumulative matching degree exceeds a preset threshold, it is determined that the matching is successful, and an activation signal is output to the post-matching processing section. If the cumulative matching degree does not exceed the threshold value, the obtained corresponding pair is additionally stored in the corresponding group list memory 107 together with the cumulative matching degree, and the starting signal S3 is output to the constituent straight line selection device 11, and the next corresponding group is selected. generated. (a) Length matching degree calculation device This device 12 is a device that calculates the length matching degree Δρ from the length of the model straight line and the length of the constituent straight lines using the following equation. However, Len(rii p), Ien(g p)
is the length of the model straight line and the constituent straight line. (b) As shown in FIG. 8, in the posture candidate arithmetic unit correspondence group (帛,・Sapling 1. Air P+1.6.-61.δ,), the model line front 1. Hanging 2.1
The angle formed by is θmp+, and the plane coordinate of the constituent line δ1 is
The angle formed with the axis is θ1, and the expected value of the angle formed with the X axis of the constituent straight line 6I to be compared with the specific model straight line pa1 is defined as shi. Then, the expected value θ is obtained by the following formula. However, the α0 or aI) formula is selected so as to satisfy −π/2≦shio≦π/2. Shio=01+θmp++ ”-αQ
i=θ1+θmp++±π−αυThis) IJI waiting value δ is stored in the posture candidate storage memory 111. (C) Attitude determining device The angle θ1 actually formed by the selected constituent straight line δi with the X-axis is determined, and the attitude matching degree Δρ2 is calculated from the value and the expected value θ using the following equation. △ρ, −(1,21 氐−0,1/π)Xo, 2 ゛・
(d) Existence possibility region arithmetic unit model of -α Straight line Nae l, City, , -, ,19i2゜1 and constituent line δ1゜ε2. The possible existence area of the constituent straight line δpa+ is calculated from δ2, and the coordinates of the corners of the area are stored in the possible existence area storage memory 112. The existence possibility area Aδpal of δpa+ is given by the following equation. Here, N (seedling 5. δ1; mp+1) is before the model line, and when the constituent line δ is associated with the line before the specific model line, p+
This is the range in which the specific model straight line front pa1 can exist, which is determined by the mutual positional relationship between the model straight line front pa1 and the other model straight line front 2. Therefore, the area Δδpa+ becomes a common part of the individual areas when other model straight lines are changed. (e) Existence position determination device configuration It is verified whether the straight line δ, +1 is included in the existence possibility area, and the area matching degree Δρ3 is calculated according to the following equation based on the degree of inclusion. Δρ, = 0.5 (when all are included) = 0.25 (when some are included) 04) 20 (
(f) Matching degree calculation device Wooden U is a length matching degree calculation device 12, posture determination device 14,
Output Δρ1 of the existence position determining device 16. By summing Δρ2 and Δρ, the matching degree ρ′ between the specific model straight line and the selected constituent straight line is calculated, and the previous cumulative matching degree ρ is calculated. , P+, and the matching degree ρ' is added to the new even-numbered step cumulative matching degree ρ.
5. l is found. ρ11. is calculated using the following formula. Δρ1+Δρ2+Δρ, if <THI then ρ,,,=-1
・-・09 If not, ρ1. I=ρO+P+1” (Δρ1+Δρ2
+Δρ, )ω, /2-OED where ρ. , □1 is the correspondence set (m
,m,--"'m,. 4.This is the cumulative matching degree of δ112゛-...δp). In other words, it is the cumulative matching degree of the parent node of the corresponding set whose matching degree is currently being calculated. The result of this operation, along with its corresponding set, is
It is additionally stored in the corresponding group list memory 107. (g) Matching success determination device The cumulative matching degree calculated by the matching degree calculating device 17 is
When it is larger than a preset threshold TH2, it is determined that the matching is successful, and the corresponding set stored in the constituent straight line extractor memory 109 and the model straight line set storage memory 110 is transferred to the position/rotation parameter calculation device 19. (7) Post-verification processing unit (a) Position/rotation parameter calculation unit Corresponding group that was successfully matched (Ichi, Nae2-・- Nae, +1゜e
This is a device that calculates the posture and position displacement of the detected object with respect to the model from the + 82 """ e pal) and outputs it to the robot control device. How to determine the position and rotation parameters is described below. - Select two non-parallel straight lines m1°m2* from the seedling pal, and draw the corresponding constituent straight lines e
, *e, *. For example, m+'m2" e
Assume that +"e2" is in a positional relationship as shown in FIG. The displacement of the posture is the angle θ3 formed between mdoe1' and the coordinate X axis.
．． θ. using θ. It can be obtained as -01. The displacement of the position is, , * m2 * and e , I e 2 *
It can be determined by knowing the deviation of the intersection points. Intersection C,,C. Let the coordinates of (Cm', Cm'), (Ce”
, Ce'), and the positional displacement is (CeX-Cm')
, Cey-Cm'). In this way, the amount of displacement in the position and orientation of the detected object with respect to the model is taught to the robot, and the robot can use these amounts of displacement to correct the gripping position of the object. (8) Effect Next, the effect will be explained according to FIG. When the component straight line extraction unit A completes the extraction of the component straight lines, the corresponding group list initialization device 6 stores the corresponding group list memory 1.
Clears 07 and turns on status register 7. When the status register is on, the maximum matching degree correspondence set selection device 8 outputs the activation signal S1 to the model straight line and list creation device 9, and the device 9 generates the first correspondence set from the model straight line list memory 105. , selects weight values from the model straight line weight value list 106, stores them together with the cumulative matching degree in the correspondence set list memory 107, and turns off the status register 10. Then, the maximum matching degree correspondence group selection device 8 selects the correspondence group with the largest cumulative matching degree from among the primary correspondence sets, and stores this in the maximum matching degree storage memo IJ 108.
A starting signal S2 is output to the constituent straight line selection device 11, and the device 1
1 selects a constituent line from the constituent line list memo U 104, generates the first quadratic correspondence set, and stores it in the memory 109.1.
10, and outputs an activation signal to the length matching degree calculation device 12. This calculation device calculates the degree of matching between the model straight line and the constituent straight lines according to the above equation (9). The result of this calculation is additionally stored by the matching degree calculation device 17 together with the cumulative matching degree in the corresponding group list memory. Thereafter, the device sequentially repeats the above processing, expands all the secondary correspondence sets, and determines the cumulative matching degree in the memo IJ.
Obtained at 107. When all the operations of the second step are completed, the device 11 turns on the status register 7. Then, the device 8 selects the correspondence group having the maximum cumulative matching degree from among the secondary correspondence groups, and outputs an activation signal to the device 9. As in the first step, the device 9 incorporates the model straight line, calculates the cumulative degree of matching, develops the cubic correspondence set in the memory 107, and turns off the status register 10. After that, the device 8 outputs an activation signal to the device 11, and the device 11 outputs the activation signal to the second device 11.
Similar to step 1, a first quaternary correspondence set is generated and a start signal is output to the arithmetic units 12, 13, and 15. Each arithmetic device calculates each matching degree using equations (9), α, and αa, and based on these results, a cumulative matching degree is calculated by the matching degree calculating device 17. These results are stored in memory 107 along with the selected correspondence set. The device 11 generates the next correspondence set, and the devices 12, 14, and 16 are activated to perform the computation of the degree of matching regarding the other correspondence sets in the fourth step in the same manner as above. Next, the odd-numbered step calculations are performed in the same manner as the third step calculation processing described above, and the even-numbered step calculations are performed in the same manner as the fourth step calculation processing. In this manner, as the calculation progresses, the matching success determination device 18 completes the matching process when the cumulative matching degree becomes equal to or greater than a predetermined value. (9) As shown in FIG. 10, the calculation of the possibility of the existence of a modified example is performed by calculating the distance tL, dz from the selected model straight line m l + m 20 intersection 0 to both end points of the specific model straight line seedling 2. Calculate the distance d+, d2 from the intersection O1 of the corresponding constituent straight line δ and δ2 in the direction of the constituent straight line δ2 to be compared, and set the range with a certain width ΔW as the possible existence area. You can also do it. (10 Other Embodiments The present apparatus can also be configured by a computer system as shown in FIG. 41. The processing procedure of the CPU 40 is shown in FIG.
Corresponding sets of steps are generated and initial values of their cumulative matching degrees are set, and the corresponding sets are stored in the storage area A in descending order of size. In steps 114 to 118, a corresponding set of even steps is generated, each matching degree is calculated, and a cumulative matching degree is determined. In contrast, steps 122-12.6
Then, the corresponding set of odd steps excluding the il step and the cumulative matching degree are calculated. Then, in step 120, the correspondence sets that have been generated so far are sorted in descending order of cumulative matching degree. The arithmetic processing is completed in step 102 or step 106. The former means that generation of all correspondence pairs is completed before the cumulative matching degree reaches a predetermined value, indicating a matching failure. On the other hand, the latter ends when the cumulative matching degree reaches a predetermined value, which means that matching has been performed sufficient to recognize the detected object. Effects of the Invention The present invention provides a visual recognition device in which one model line segment and one
When calculating the degree of matching between the constituent line segments of When a model line segment is made to correspond to other constituent components corresponding to it, a range in which the specific model line segment can be located is determined due to uncertainty in the correspondence, and from this range, the specific model line segment is The present invention is characterized by comprising a means for determining a region of possibility of the existence of constituent line segments corresponding to the above, and a means for calculating the degree of matching in matching according to the degree to which the constituent line segments exist in the region of possibility of existence. That is. Therefore, in the present invention, since the matching degree is based on the region determined by the positional relationship of line segments and the uncertainty of correspondence, both end points of the constituent line segments cannot be accurately determined, but are instead determined as broken line segments. Even in cases where objects are identified, information on the positional relationships of line segments can be grasped, allowing accurate object recognition. In addition, even when a part of the detected object is enlarged and imaged, such as in visual recognition devices used in robots, the amount of information about the positional relationship between line segments is included, making it easy to recognize the detected object. I can do it.

[Brief explanation of drawings]

FIG. 1 shows visual perception J y according to a specific embodiment of the present invention.
A block diagram showing the B4 configuration of the t device, FIG. 2 is an explanatory diagram for explaining the existence possibility region, and FIG.
An explanatory diagram explaining the logic of generating correspondence sets, Fig. 4 is an outline diagram showing the contour line of the model, Fig. 5 is an explanatory diagram showing the model straight line, and Fig. 6 is an explanatory diagram showing the contour extracted from the image. Fig. 7 is an explanatory diagram explaining the differential operator, Fig. 8 is an explanatory diagram explaining the calculation of the posture matching degree, and Fig. 9 is the model. FIG. 4 is an explanatory diagram showing the attitude and position displacement of a detected object relative to the detected object. Figure 10 shows
An explanatory diagram showing another possible existence area, FIG. 11 is a block diagram showing the configuration of the entire robot system using the example device, and FIG. 12 is a block diagram showing the configuration of another example device. The diagram, FIG. 13, is a flowchart showing the processing procedure of the CPU used in the device of the same embodiment.

Claims (5)

[Claims] (1) Obtain the image outline from the two-dimensional image, extract multiple constituent line segments that make up the outline from the contour, and obtain information regarding the positions of the extracted constituent line segments. an extraction means, a model line storage means storing information regarding the positions of model line segments constituting at least the main part of the contour of the model to be compared with the detected object; The constituent line segments are compared with the model line segments stored in the model line segment storage means, the degree of coincidence is calculated, and the constituent line segments corresponding to the model line segments are determined based on the degree of coincidence. A visual recognition device comprising a matching means for specifying, the matching means being a target of matching when calculating the degree of matching between one model line segment and one constituent line segment. Based on the positional relationship between the specific model line segment, which is the model line segment 1, and other model line segments, when the other model line segment is made to correspond to it based on the other corresponding component, the corresponding Existence possibility region calculating means for determining a possible range in which the specific model line segment may be located due to uncertainty, and calculating an existence possibility region for a constituent line segment corresponding to the specific model line segment from the range; 1. A visual recognition device, comprising: an existence position determining means for calculating the degree of matching according to the extent to which the object exists in the existence possibility region. (2) When calculating the degree of matching between one model line segment and one constituent line segment, the matching means may be configured to match a specific model line segment that is one model line segment that is the object of matching. When the other model line segment is made to correspond to its corresponding other component based on the relative attitude with respect to the other model line segment, the configuration corresponding to the specific model line segment is determined from the possible postures of the specific model line segment. A posture candidate calculating means for calculating possible postures of a line segment, and a posture determining means for calculating the degree to which the constituent line segments match the possible postures, taking into account the degree of coincidence according to the existence possibility region. The visual recognition device according to claim 1, wherein the visual recognition device calculates the degree of matching in the comparison. (3) When calculating the degree of matching between the model line segment 1 and the component line segment 1, the matching means calculates the degree of matching of the lengths of the line segments. 3. The visual recognition device according to claim 2, further comprising means for calculating the degree of matching in verification by further taking into consideration the degree of matching determined by the means. (4) The constituent line segments and the model line segments are straight line segments, and the constituent line segment extracting means includes a broken line approximation means that approximates the contour of the detected object as a broken line. A visual recognition device according to scope 1. (5) The matching means sequentially increases the number of model line segments to be matched while sequentially selecting pairs of model line segments and constituent line segments that have the highest degree of matching calculated up to that point. 2. The visual recognition device according to claim 1, further comprising an order processing means for calculating the degree of matching.