JP3501647B2 - Apparatus and method for recognizing shape and recording medium storing program for recognizing shape - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a shape recognition apparatus for performing shape recognition by shape matching using shape data of an object acquired by a range sensor or the like used for detecting a copy point in an arc welding robot, and an implementation thereof. And a recording medium on which the shape recognition program is recorded.

[0002]

2. Description of the Related Art Shape recognition of an object is realized by comparing shape data obtained by measuring an object to be recognized with a shape model already registered.
The function of species recognition is known. The first function of recognition is to compare the shape data and the shape model and recognize that the recognition target object is an object in the same category as the model object.

The second function of recognition is to accurately detect the position and orientation of the object to be recognized. Regarding the detection of position and orientation, when there is a slight shape deviation between the recognition target object and its shape model, there is a need for overall alignment such as seal matching, and welding contour point detection. There are some that are required to accurately detect only the vicinity of a certain feature point of interest.

The third recognition function is to recognize the deformation and movement of the recognition target object by associating the shape data before and after the deformation and movement of the recognition target object. Conventionally, the recognition method of the second function is continuously applied in time series by applying the recognition method of the second function of extracting feature points from the shape data of the position and orientation detection technology and comparing with the characteristics of the shape model. By doing the third
The function was realized. The present invention is directed to the case of the third function.

The first measurement method and the second measurement method, which are the two types of measurement processing procedures in the third function, will be described below in order. First, the first measuring method is a measuring and recognizing method in the case where the overall alignment is required, targeting the case where the object to be recognized is deformed or moved. At each time point, the feature points of the recognition target object are extracted, and the overall alignment is performed from the correspondence with the feature points of the shape model. However, this is limited to the case where the characteristics of the feature points do not change.

Next, the measurement processing procedure of the first measurement method will be described with reference to FIG. FIG. 11 is a block flow chart showing the steps of processing of the first measuring method. First, at time 0, which is the initial time, shape data at time 0 is extracted from the input data unit 17, feature points of the shape data are extracted, a shape model is constructed from the feature points at time 0, and the shape model data unit 18 Save to (step ST
A).

At time t, the shape data at time t is fetched from the input data section 17, the characteristic points of the shape data are extracted (step STB), the shape model is fetched from the shape model data section 18, and the time t The correspondence between the feature points and the feature points of the shape model is obtained and the position of the entire recognition target object is adjusted (step STC), and the shape model in the shape model data unit 18 is set to the time t.
To the shape model from the feature points of
D) If the measurement process is not ended, the time is advanced and the process returns to step STB. If the measurement process is ended, the process is ended (step STE). The step STD may not be executed.

A specific example of the first measuring method will be described with reference to FIG. FIG. 12 is an explanatory diagram of a measurement processing procedure for tracking a moving ship in a two-dimensional moving image. The characteristic points of the ship 19 at time 0 of the initial time are represented by characteristic points D0, E0, F0 as shown in (a), and the D0, E0, F0 are constructed as a shape model as shown in (b). (Corresponding to step STA).

Next, at time t, the characteristic points are represented by characteristic points Dt, Et, Ft, Mt (corresponding to step STB) as shown in (c), and the characteristic points Mt are excluded. The characteristic points Dt, Et, Ft and the D0 at (b),
(D) shows the correspondence between the shape model of E0 and F0, and the overall alignment (step ST
From (corresponding to C), how the ship 19 moved was measured, and measurement was performed by repeating this procedure (corresponding to step STE).

Next, the second measurement method requires accurate detection and local alignment only in the vicinity of the feature point of interest of the recognition target object when the recognition target object deforms or moves. This is the measurement recognition method for the case. At each time, the feature points of interest are extracted temporarily, and the local feature points are described centering on that.

The local feature point mentioned here means a point having a feature within a certain limited range. Local alignment is performed based on the correspondence between the local feature points and the feature points of the shape model, and the exact position of the feature point of interest is extracted. However, like the first measurement method, it is limited to the case where the characteristics of the feature points do not change.

The measurement processing procedure of the second measurement method will be described with reference to FIG. FIG. 13 is a block flow diagram showing steps of the measurement processing procedure of the second measurement method. First, at time 0, which is the initial time, the input data section 17a
The shape data at time 0 is taken out from, the feature points of the shape data are extracted, a shape model is constructed from the feature points at time 0 and stored in the shape model data unit 18a (step STA ').

At time t, the shape data at time t is taken out from the input data section 17a, the shape model is taken out from the shape model data section 18a, and the feature points of the shape data at time t are compared with the feature points of the shape model. Then, the provisional feature point of interest of the shape data is specified (step STB '), and the local feature point of the shape data is described centering on the provisional feature point of interest (step STC').

The correspondence between the local feature points and the shape model is obtained, and local alignment is performed to obtain an accurate feature point of interest (step STD '). The shape model is updated to the shape model from the feature point at the time t (step STE ′),
If the measurement process is not ended, the time is advanced and the process returns to step STB ′, and if the measurement process is ended, the process is ended (step STF ′). The step STE 'does not have to be executed.

A specific example of the second measuring method will be described with reference to FIG. FIG. 14 is an explanatory diagram of a measurement processing procedure of tracking a copying point of welding in a two-dimensional moving image. The characteristic point of the welding plate cross-sectional image at time 0 of the initial time is
Represented by the characteristic points G0, H0, and I0 as shown in (a), the characteristic point of interest is the bending point G0.
H0 and I0 are constructed as a shape model as shown in (b) (corresponding to step STA ').

Next, at time t, the characteristic points are represented by the characteristic points Gt, Ht, It, and Jt as shown in (c),
By comparing with the feature points G0, H0, and I0 of the shape model of (b), the Gt is specified as a provisional feature point of interest (corresponding to the step STB ′), and the Gt is localized around the Gt. When the characteristic points are described, local characteristic points Kt and Lt are determined as shown in (d) (corresponding to step STC ').

The shape model and the local characteristic points Kt, L
The relationship with t is shown by (e), an accurate feature point of interest is obtained (corresponding to step STD '), how the welding copy point has moved is measured, and this procedure is repeated ( The measurement was performed in step STF ').

[0018]

However, the conventional method cannot deal with the case where the characteristics of the characteristic points change. A specific example in which the characteristics of the feature points change is shown in FIG.
This will be described with reference to (a). FIG. 10A is a diagram showing welding points A, B, and C which are copying points of welding of the joint of the cross elbow pipe 16 including two pipes.

From the welding point A through the welding point B,
Since the cross-sectional shape up to the welding point C changes from a concave shape to a convex shape, the condition around the feature point, which is the property of the feature point, also changes. Therefore, the feature point of the shape data and the feature point of the shape model It is impossible to compare the two because they cannot be associated with each other, and it is impossible to specify an accurate feature point of interest.

Here, the main objects to be solved by the present invention are as follows. An object of the present invention is to provide a shape recognition device and method capable of recognizing a shape even when the characteristics of feature points of a recognition target object change, and a recording medium recording the shape recognition program. is there.

Other objects of the present invention include the specification, drawings,
Especially, it will be apparent from the description of each claim.

[0022]

In order to solve the above problems, the present invention provides an input / output path which is an input / output path for data,
A shape measurement unit that forms shape data from measurement data, a shape data storage unit that stores the shape data, a preprocessing unit that erases noise, a profile storage unit that stores the noise-erased shape data, and a shape A geometric model construction unit that constructs a model, a geometric model storage unit that stores the geometric model, segment matching units that collate the current geometric data with the geometric model, output of feature points, and update the geometric model. And a device having a feature point detecting unit that operates.

Further, the measurement data of the object to be recognized, which is input from a detector such as a range sensor, is noise-erased to form shape data, and a shape model is constructed from the segments, parameters, and feature points of the shape data. It is characterized by using a method of comparing the shape data and the past shape model, detecting the current parameters and feature points, updating the shape model, and outputting the feature points. Further, it is characterized in that a recording medium recording a program for controlling execution of the method in the operation of the apparatus is used.

More specifically, in order to solve the problem, the present invention achieves the above-mentioned object by adopting a novel characteristic construction means or method ranging from a superordinate concept to a subordinate concept enumerated below. Done.

That is, the first feature of the device of the present invention is a device for recognizing a predetermined feature point of a recognition target object that may be deformed and moved, and is an input / output path for data. Shape, a shape measurement unit that forms measurement data input through the input / output path into shape data that is a multidimensional coordinate sequence, a profile accumulation unit that accumulates the shape data from the shape measurement unit, and an initial time The shape data accumulated in the profile accumulating section is segmented by a line segment, a quadrangle, a polygonal line, or a figure having one or more line segments or curves as constituent elements, and a parameter of a segment and an angle between each segment and A shape model having a feature point that is a contact point or a closest point between each segment is constructed automatically or by manual input, and The shape model construction unit for outputting, the shape model storage unit for storing the shape model constructed by the shape model construction unit, the current shape data accumulated in the profile accumulation unit and the shape model storage unit Each segment matching unit that collates the shape model and extracts the position information of each segment, and creates the current shape model by detecting the feature points and the parameters from the position information in each segment matching unit, According to another aspect of the present invention, there is employed a configuration of a shape recognition device including: a feature point detection unit that updates the shape model in the shape model storage unit and outputs the feature point through the input / output path.

A second feature of the device of the present invention is that the shape measuring unit and the profile accumulating unit in the first feature of the device of the present invention mutually accumulate the shape data from the shape measuring unit. A data storage unit and a pre-processing unit that erases noise of the shape data stored in the shape data storage unit are provided, and the profile storage unit stores the shape data from the pre-processing unit. This is in adopting the configuration of the shape recognition device.

A third feature of the device of the present invention is that the detection of the parameter in the first or second feature of the device of the present invention is performed by using the evaluation value of the evaluation function described with the position of the segment as a variable. This is due to the adoption of the configuration of the shape recognition device.

The first feature of the method of the present invention is a method for recognizing a predetermined feature point of a recognition target object which may be deformed and moved, and the shape is obtained from the input measurement data of the recognition target object. Forming shape data having a concatenation or set of undeformed segments, accumulating the shape data, and if the shape data is at the initial time, the accumulated shape data is a line segment, a rectangle, a polygonal line, or 1
Performs segmentation by a figure with one or more line segments or curves as constituent elements, and automatically creates a shape model with the parameters of the segment and the angle between each segment, and the feature points that are the contact points or nearest points between each segment. Position information is obtained by collating the accumulated current shape data with past shape models if the shape data is not at the initial time. Then, a current shape model is created by detecting the current parameter and the current feature point from the position information, the past shape model is updated to the current shape model, and the feature point is output. It is in adopting the configuration of the recognition method.

A second feature of the method of the present invention is the adoption of a configuration of a shape recognition method in which the shape data in the first feature of the method of the present invention is data of a multidimensional coordinate sequence.

A third feature of the method of the present invention is that the shape recognition method according to the first or second feature of the method of the present invention is configured so that noise of the measurement data is also removed. It is in.

A fourth feature of the method of the present invention is that the detection of the parameter in the first, second or third feature of the method of the present invention is performed by using the position of the segment as a variable and the evaluation value of the evaluation function. The configuration of the shape recognition method is performed by estimating the parameters by using.

A first feature of the recording medium of the present invention is a recording medium in which a program for controlling a device for recognizing a predetermined feature point of a recognition target object that may be deformed and moved by a computer is recorded, The measurement data obtained by measuring the recognition target object is input to a shape measurement unit through an input / output path, and the shape measurement unit expresses the measurement data of the recognition target object as shape data of a multidimensional coordinate sequence. The first step of accumulating the shape data in the shape data accumulating unit, and the pre-processing unit performs noise removal processing in the multidimensional coordinates on the shape data accumulating in the shape data accumulating unit, The second step of accumulating the shape data subjected to the removal processing in the profile accumulating unit and the shape data accumulated in the profile accumulating unit are In the third step of determining whether or not it is at the initial time, and in the third step, if the shape data at the initial time, the shape model construction unit stores in the profile storage unit at the initial time. In constructing a shape model from the formed shape data, a fourth step of segmenting a figure using the shape data as a line segment, a quadrangle, a polygonal line, or one or more line segments or curves, From the shape data stored in the profile storage unit, parameters of angle between each segment and each segment and feature points that are contact points or nearest points between each segment can be automatically or manually extracted. The fifth step is to construct a shape model described by the segment, the parameter and the feature point. And the step of 6,
In the seventh step of storing the shape model in the shape model storage unit, the eighth step of outputting the feature points of the shape model to the computer and terminating, and the third step, If the shape data is not the shape data, each segment matching unit compares the past shape model stored in the shape model storage unit with the current shape data stored in the profile storage unit, Ninth to extract the position information of
And a tenth step of detecting the current parameters and feature points by the feature point detection unit from the position information, and the feature point detection unit creates a shape model having the current parameters and feature points. Then, a series of procedural processing including an eleventh step of updating and storing in the shape model storage unit and a twelfth step of outputting the current feature point to the computer through the input / output path and ending the processing. This is in adopting the configuration of the recording medium in which the shape recognition program to be executed is recorded.

A second characteristic of the recording medium of the present invention is that the detection of the parameter in the first characteristic of the recording medium of the present invention is performed by using the evaluation value of the evaluation function described with the position of the segment as a variable. The configuration of the recording medium that stores the shape recognition program is as follows.

[0034]

BEST MODE FOR CARRYING OUT THE INVENTION The principle of the present invention is to recognize a predetermined feature point of a recognition target object that may be deformed or moved, and to connect or aggregate segments that are parts that do not deform the shape of the object. Since the feature points are the contact points or the nearest points between the segments, the state between the segments in the shape data is described by a parameter. Therefore, even when the recognition target object is deformed, the shape model is calculated from the shape data. Can be constructed, and the current parameters and feature points can be detected from the position information obtained by comparing the shape model according to the past recognition result with the current shape data.

Hereinafter, with reference to the accompanying drawings, a device example 1 and a device example 2 according to an embodiment of the present invention, and a method example 1,
Method example 2, method example 3 and method example 4 will be described in order.

(Device Example 1) This device example will be described with reference to FIG. FIG. 1 is a block diagram of an example of this apparatus. An example of this device is an input / output path which is a path for inputting / outputting data, a shape measuring section 1 which forms measurement data from the input / output path as shape data which is a two-dimensional coordinate sequence, and the shape measuring section. The shape data storage unit 2 for storing the shape data 1 and the shape data storage unit 2
Preprocessing unit 3 for eliminating the noise of the shape data accumulated in
A profile accumulating section 4 for accumulating noise-eliminated shape data in the pre-processing section 3; and automatically constructing a shape model from the shape data accumulated in the profile accumulating section 4 to input / output the characteristic points. The shape model construction unit 5 for outputting to the road, the shape model storage unit 6 for storing the shape model constructed by the shape model construction unit 5, the current shape data and the shape model storage unit 6 in the profile storage unit 4.
Each segment matching unit 7 that collates the shape model in to extract the position information of each segment, and detects the parameters and feature points from the position information in each segment matching unit 7 to create the current shape model, A feature point detection unit 8 that updates the shape model in the shape model storage unit 6 and outputs the feature point to the input / output path.

(Device Example 2) This device example is a modification of the device example 1 and will be described with reference to FIG. FIG. 2 is a block diagram of the example of the apparatus. Compared with FIG. 1, instead of the shape model construction unit 5, a shape model input unit 5a for constructing a shape model from the shape data accumulated in the profile accumulation unit 4 by manual input through an input / output path is provided. The shape model storage unit 6 is a device for storing the shape model constructed by the shape model input unit 5, and the other parts are the same and the description thereof will be omitted.

(Method Example 1) This method example is applied corresponding to the above-mentioned device example 1,
Similarly, FIG. 1 and other FIGS. 3 and 4 will be described.
FIG. 3 is an explanatory diagram showing the representation of the shape model of this method example, and FIG. 4 is an explanatory diagram showing the comparison between the shape model and the shape data.

First, a description will be given with reference to FIG. Measurement data obtained by measuring a recognition target object to be measured by a detector such as a range sensor is sent to the shape measuring unit 1 through an input / output path (not shown), and the shape measuring unit 1 sends the measurement data of the shaped object to the two-dimensional coordinates. Expressed by the shape data of the column.

Here, the expression of the two-dimensional coordinate sequence is an expression method using a two-dimensional image plane as shown in FIG. The shape data formed by the shape measuring unit 1 is
It is stored in the shape data storage unit 2.

The preprocessing unit 3 performs noise removal processing on the shape data stored in the shape data storage unit 2 in the two-dimensional coordinate sequence, and stores the shape data subjected to the noise removal processing in the profile storage unit 4. accumulate.

The shape model construction unit 5 constructs a shape model from the shape data stored in the profile storage unit 4 at the initial time. The construction of the shape model is performed by extracting constituent elements from the shape data and segmenting the constituent elements, thereby defining parameters of each segment and an angle between the respective segments, and a feature point that is a contact point or a nearest point between the segments. The shape model described in 1) is automatically created, the shape model is stored in the shape model storage unit 6, and the feature points are output to the outside through the input / output path.

Here, as the shape model, as shown in FIG. 3B, segments seg1, seg2, parameters α (seg1, seg2), characteristic points Q (x, y).
The number of line segments of each segment that intersects the characteristic point, the length of the line segment, the angle between the line segments, and the end point information of the line segment are described as the constituent elements of each segment, and each of the above-mentioned The segment information and the angle between the respective segments are described, and the x and y coordinates of the characteristic point are described as the characteristic point. In the following symbols, the subscript s means the start point and e means the end point, and so on.

The description of each segment is as follows, for example. Segment description: Number of line segments, line segment length, angle between line segments, line segment end point information example: 2 seg1 ... 1, L1, null, ((x _1s , y _1s ),
(X _1e , y _1e )) seg2 ... 1, L2, null, ((x _2s , y _2s ),
(X _2e , y _2e )) Parameter: angle (intersecting segment 1, intersecting segment 2) α (seg1, seg2) Feature point: (x coordinate, y coordinate) Q (x, y)

Next, when new shape data is newly stored in the profile storage unit 4, the segment matching unit 7 stores the past shape model stored in the shape model storage unit 6 and the profile. The feature point detection unit 8 compares the current shape data stored in the storage unit 4 to extract the position information of each segment, and detects the parameters and feature points of the current shape data from the position information. Then, a shape model is created from the current parameters and feature points.

Here, the extraction of the position information and the detection of the parameters and the characteristic points are performed, for example, by first using the shape model at time t-1 shown in FIG. 4A and the shape at time t shown in FIG. 4B. By comparing with the data, the positional information of each constituent element for each segment is obtained and extracted, the parameter α _t is determined similarly as in (c), and then the shape model at time t-1 is described. From the characteristic points Q (x _{t − 1} , y _{t − 1} )
The current feature point Q (x _t , y _t ) is calculated and detected.

Finally, the feature point detection unit 8 updates and stores the shape model having the current parameters in the shape model storage unit 6, and outputs the current feature points to the outside through the input / output path. To do.

(Method Example 2) This method example corresponds to the apparatus example 2 and will be described with reference to FIG. Compared with the method example 1,
In the construction of the shape model, the profile storage unit 4
It is not a method of automatically constructing a shape model from the shape data from, but it is a procedure to construct a shape model by manually inputting segmentation of shape data and extraction of parameters and feature points through an input / output path. . The other parts are the same as in Method Example 1, and the description thereof is omitted.

(Method Example 3) This method example corresponds to the device example 1 and the device example 2,
The method example 1 and the method example 2 are different in the procedure of modeling the shape data, and will be described with reference to FIG. FIG. 5A is an explanatory diagram of a shape model in which a segment is a long quadrangle (quadrangle), and FIG. 5B is an explanatory diagram of a shape model in which a segment is a polygonal line. The segment represented by the rectangular shape models the plate to be welded as it is, and the segment represented by the polygonal line models the surface of the plate to be welded that may be measured. Since the method example 3 is the same as the method example 1 and the method example 2 except for the modeling of the shape data, the description thereof will be omitted.

In the modeling of the shape data in this example of the method, first, when the segment is a long quadrangle, referring to FIG. 5A, the shape model is a long quadrangle segment seg.
1, seg2, parameter α (seg1, seg2),
The number of line segments of each side of each segment of a rectangular rectangle that intersects with the feature point Q, the length of the line segment, and the angle between the line segments, which are configured from the feature points Q (x, y) , The end point information of the line segment is described, the segment information and the angle between the segments are described as parameters, and the x and y coordinates of the feature point are described as the feature point Q.

The description of each segment is as follows, for example. Segment description: Number of line segments, line segment length, angle between line segments, line segment end point information

Example: 2 seg1 ... 4, line1 ... L1, (line2 (90), line3
(90)), ((x _11s , y _11s ), (x _11e , y _11e )) line2 ... L2, (line1 (90), line4
(90)), ((x _12s , y _12s ), (x _12e , y _12e )) line3 ... L3, (line1 (90), line4
(90)), ((x _13s , y _13s ), (x _13e , y _13e )) line4 ... L4, (line2 (90), line3
(90)), ((x _14s , y _14s ), (x _14e , y _14e ))

Seg2 ... 4, line1 ... L1, (line2 (90), line3
(90)), ((x _21s , y _21s ), (x _21e , y _21e )) line2 ... L2, (line1 (90), line4
(90)), ((x _22s , y _22s ), (x _22e , y _22e )) line3 ... L3, (line1 (90), line4
(90)), ((x 23s, y 23s), (x 23e, y 23e)) line4 ... L4, (line2 (90), line3
(90)), ((x _24s , y _24s ), (x _24e , y _24e ))

Parameter: angle (intersecting segment;
Line, intersecting segment; line) α (seg1; line1, seg2; line1) Feature point: (x coordinate, y coordinate) Q (x, y)

Next, when the segment is a polygonal line, FIG.
Referring to (b), the shape model is a rectangular segment seg1, seg2, a parameter α (seg1, se
g2), composed of feature points Q (x, y), and as the constituent elements of each segment, the number of line segments of each segment of the polygonal line intersecting the feature points, the length of the line segment, the angle between the line segments, and the line The end point information of the minutes is described, the segment information and the angle between the segments are described as parameters, and the x and y coordinates of the feature points are described as the feature points.

The description of each segment is as follows, for example. Segment description: number of line segments, line segment length, angle between line segments, line segment end point information seg1 ... 2, line1 ... L1, (line2 (90)), ((x
_11s , y _11s ), (x _11e , y _11e )) line2 ... L2, (line1 (90)), ((x
_12s , y _12s ), (x _12e , y _12e ))

Seg2 ... 2, line1 ... L1, (line2 (90)), ((x
_21s , y _21s ), (x _21e , y _21e )) line2 ... L2, (line1 (90)), ((x
_22s , y _22s ), (x _22e , y _22e ))

Parameter: angle (intersecting segment;
Line, intersecting segment; line) α (seg1; line1, seg2; line1) Feature point: (x coordinate, y coordinate) Q (x, y) Note that a segment is composed of one or more line segments or curves. Even in the case of a figure, it is possible to describe it by adding parameters required for expression.

(Method Example 4) This method example is different from the method example 1 in the modeling procedure, and will be described with reference to FIG. FIG. 6 is an explanatory diagram showing the relationship between the shape data and the shape model in this method example.

[0060]

[Outer 1]

[0061]

[Outside 2]

[0062]

[Outside 3]

[0063]

[Outside 4]

[0064]

[Outside 5]

Next, when the shape recognition method of Method Example 1 is expressed in a program and recorded on a recording medium and shape recognition is performed, the program is read by a computer (not shown) and the apparatus is executed by the computer. A procedure of the program in the case of performing the control in accordance with the execution of the program will be described with reference to FIGS. Figure 7
3 is a flowchart of a shape recognition program according to this embodiment.

The flow of each step will be described. Measurement data obtained by measuring a recognition target object to be measured by a detector such as a range sensor is sent to the shape measuring unit 1 through an input / output path (not shown), and the shape measuring unit 1 measures the measurement data of the recognition target object in two dimensions. It is represented by the shape data of the coordinate sequence, and the shape data of the shape measuring unit 1 is stored in the shape data storage unit 2 (step ST1).

In the preprocessing unit 3, the shape data stored in the shape data storage unit 2 is subjected to noise removal processing in the two-dimensional coordinate sequence, and the shape data subjected to the noise removal processing is stored in the profile storage unit 4. It is accumulated (step ST2).

Next, it is determined whether or not the shape data stored in the profile storage unit 4 is at the initial time (step ST3).

If it is the shape data at the initial time in step ST3, the shape model construction process is performed. When constructing a shape model from the shape data stored in the profile storage unit 4 at the initial time, the shape model construction unit 5 segments the shape data into line segments (step ST4).

From the shape data accumulated in the profile accumulating section 4, the parameters of the respective segments and the angle between the respective segments and the characteristic points which are the contact points or the closest points between the segments are automatically extracted (step ST
5), model and build the shape model described by the segment, the parameter and the feature point, and finish the shape model construction process (step ST6).

The shape model is stored in the shape model storage unit 6 (step ST7), the characteristic points of the shape model are output to the computer through the input / output path, and the process is terminated (step ST8).

If the shape data stored in the profile data storage unit 4 in step ST3 is not the data at the initial time, it is stored in the shape model storage unit 6 by each segment matching unit 7. The past shape model is compared with the current shape data accumulated in the profile accumulating unit 4, position information of each segment is extracted (step ST9), and the current parameters and characteristics are extracted from the position information. The points are detected, and the feature point detection unit 8 creates the shape model of the current shape data (step ST10).

The feature point detection unit 8 updates the shape model having the current parameters and feature points in the shape model storage unit 6 and stores it (step ST11), and inputs and outputs the current feature point. The data is output to the computer through the path and the process ends (step ST12).

The above shape recognition program is the above method example 1.
If the construction of the shape model in steps ST4 to ST6 of the shape model construction processing is carried out manually, it corresponds to the method example 2. If the segment of step ST4 or ST9 is a rectangular segment or a segment of polygonal line, it corresponds to the method example 3.

If the procedure of estimating the parameters of the shape model using the evaluation value of the evaluation function in which the position of the segment is described as a variable is used for the parameter detection in the procedure of modeling the shape model in step ST10, the method example 4 described above is obtained. Will correspond to.

[0076]

EXAMPLES Next, specific examples of the embodiments of the present invention will be described. This embodiment relates to the above apparatus example and method example, and will be described with reference to FIG. 3 and FIGS. 8 to 10. FIG. 8 is a configuration diagram of a welding arc robot system to which the shape recognition device according to the present embodiment is applied.
FIG. 10 is a flowchart showing a procedure of shape recognition in welding, and FIG.
It is a schematic diagram of the intersection pipe which consists of two elbow pipes, (b) is an explanatory view showing shape data in a welding point in (a).

The present embodiment is intended to apply the above-mentioned apparatus example and method example to tracing point tracing in an arc welding robot, and the welding point as the tracing point is the same as the characteristic point or depends on the characteristic point, A shape recognition device in welding will be described with reference to FIG.

The arc welding robot system includes a robot arm 9 having a plurality of joints that can move freely in three dimensions, a welding torch 10 attached to the tip of the robot arm 9 for welding, and a range for taking in a two-dimensional coordinate sequence. A sensor 11, a computer 12 that integrally controls the robot arm 9, a welding torch 10, a range sensor 11, and a shape recognition device 13, a shape recognition device 13 corresponding to the device example 1 or the device example 2, and the computer 12. Is composed of a controller 14 that can be remotely operated and a welding target object 15 that is a target of welding.

The object to be welded 15 has welding points a, b,
Welding trajectories are drawn along each point of c, and each has a two-dimensional coordinate sequence a1, b1, c1, and as it moves from a to b to c, the characteristics of the feature points also change from concave to convex. However, the property of the feature point changes. In the present embodiment, it is possible to calculate the welding point when the characteristics of the feature point change.

Next, the welding process will be described with reference to FIG. First, the controller 12 is used to teach the computer 12 the initial posture of the robot arm 9 corresponding to the welding point a and the final posture corresponding to the welding point c (step ST1 ').

The robot arm 9 automatically moves to the welding point a which is the starting point based on the information on the taught initial posture, and the range sensor 11 causes the two-dimensional coordinate sequence shown in FIG. The computer 12 obtains the input data, displays the input data on the display of the controller 14, and sends the input data to the shape recognition device 13 (step ST2 ').

The shape recognition device 13 performs recognition processing at the initial time, models with segments and parameters between the segments, and automatically creates a shape model. The shape model is displayed on the display of the computer 12 and indicates a welding point (step ST3 ').
The welding torch 10 is controlled by the computer 12.
From then on, sparks are scattered at the welding point at the current time t and welding is performed (step ST4 ').

The computer 12 compares the posture of the robot arm 9 at the current time t with the final posture information to determine whether the welding route is on the way or at the end point. If the end point, the welding process is performed. Is completed, and if it is in the middle of the route, the process proceeds to the next step (step ST5 ').

From the information of the initial posture of the robot arm 9, the posture at the current time t and the final posture, and the position information of the welding point at the past time and the current time t and the characteristic point information,
The welding point at the next time t + 1 is mathematically predicted to determine the welding path, and the robot arm 9 is controlled and moved along the welding path (step ST6 ').

The shape data of the cross section of the object to be welded from the range sensor 11 after the movement is transferred to the computer 1.
2 is sent to the shape recognition device 13, and the shape recognition device 1
3 performs the recognition process to detect the parameters and characteristic points at time t + 1 after the movement to obtain the welding point, and the step ST
It returns to 4 '(step ST7').

As described above, the welding process is the flow of the above steps. Next, a more specific flow of the welding process will be described with reference to FIG. 3, FIG. 4, FIG. 8 and FIG. 10, taking the case where the welding process is also applied to the method example 1 as an example.

Instead of the object 15 to be welded, an intersecting elbow pipe 16 consisting of two intersecting pipes in FIG. 10A is used as the object to be welded. The cross elbow pipe 16 has welding points A, B,
10C, the welding point A,
Two-dimensional coordinate sequence A1, B1, C corresponding to B and C, respectively
1 and each characteristic point.

In the concrete welding process, the welding points A, B and C are welded in the order of the welding point A which is the starting point, the welding point B which is a point in the path, and the welding point C which is the ending point. This is the case. First, the tip of the robot arm 9 is connected to the welding point A.
And teach the initial posture to the computer 12,
Next, the tip of the robot arm 9 is guided to the welding point C, and the final posture is taught to the computer 12 using the controller 14 (step ST1 ').

Next, based on the taught information, the robot arm 9 is moved to the welding point A which is the starting point, and the shape recognition device 13 eliminates noise in the measurement data input from the range sensor 11. It is removed to obtain the shape data as shown in FIG. 3 (a) (step ST2 ').

Next, using the line segment extraction technique and the like, as shown in FIG.
A shape model as shown in (b) is constructed. The shape model is composed of two segments, the constituent elements of each segment, the parameter α, and the characteristic point Q (x, y) are automatically described, and the computer 12 specifies the welding point from the characteristic points. (Step ST3 ').

Under the control of the computer 12, the welding torch 10 disperses sparks at the welding point at the present time t to perform welding (step ST4 '). The computer 12 compares the posture of the robot arm 9 at the current time t with the final posture to determine whether it is the welding point C in the middle of the welding path or at the end point, and the welding point C at the end point. If so, the welding process is terminated, and if it is in the middle of the welding path, the process proceeds to the next step (step ST5 ').

The computer 12 uses the information on the initial posture, the posture at the current time t, and the final posture of the robot arm 9, the position information of the welding point at the past time and the current time t, and the characteristic point information, The welding point at the next time t + 1 is mathematically predicted to determine the welding path, and the robot arm 9 is controlled and moved along the welding path (step ST6 ').

The computer 12 sends the measurement data of the cross section of the cross elbow pipe 16 from the range sensor 11 after the movement to the shape recognition device 13, and the shape recognition device 13 performs the recognition process at time t + 1 to perform the welding point. Calculate the position of.

For example, in FIG. 4, t-1 → t, t → t
By replacing +1 with the shape model at the time t shown in FIG. 4A and the shape data at the time t + 1 after the movement shown in FIG. 4B, the shape model of each component between the segments of the shape model is compared. The position information is obtained, the parameter α _{t + 1} is detected as shown in FIG. 4 (c), and the feature point Q (x _{t + 1} ,
y _{t + 1)} to detect. After obtaining the welding point from the characteristic point after the movement, the process returns to the step ST4 '(step ST
7 ').

Next, the flow of the welding process will be described with reference to FIG. The program for performing the welding process has the flow of the following step (ST). The robot arm 9 is guided to the starting point and the ending point of the welding point, and the initial posture and the final posture are taught to the computer 12 using the controller 14 (step ST1 ').

The robot arm 9 is moved to the starting point,
In the shape recognition device 13, noise of the measurement data input from the range sensor 11 is removed to obtain shape data (step ST2 '), and a shape model is constructed to automatically determine line segment segments, parameters, and feature points. And a welding point is specified from the characteristic points (step ST
3 ').

Sparks are scattered from the welding torch 10 to the current welding point to perform welding (step ST4 '), and the current posture and the final posture of the robot arm 9 are compared with each other to determine whether the robot arm 9 is in the middle or at the end of the welding path. It is determined whether there is any, and if it is the end point, the welding is ended (step ST5 ').

If in the middle of the welding path, from the information on the initial posture, the current posture, and the final posture of the robot arm 9, the past time and the position information of the welding point at the present time, and the characteristic point information, The welding point at the time after the next movement is mathematically predicted to determine the welding path, and the robot arm 9 is controlled and moved along the welding path (step ST6 ').

After the movement, the measurement data of the cross section of the object to be welded from the range sensor 11 is used as the shape recognition device 1.
3, the shape recognition device 13 performs shape recognition processing to detect the parameters of the time after movement and the positions of the feature points, obtain the welding points from the feature points after the movement, and proceed to step ST4 ′. Return (step ST7 ').

Although the flow of the welding process is based on the method example 1 described above, if the shape model is constructed by manual input in the step ST3 ', it corresponds to the method example 2. Similarly, step ST3 '
In Example 3, if the method of using a rectangular segment or a segment of a polygonal line is adopted as the modeling procedure, the method example 3
In step ST7 ′, if the procedure of estimating the parameters of the shape model using the evaluation value of the evaluation function described with the position of the segment as a variable is adopted in the step ST7 ′, it corresponds to the method example 4. Will be things.

In the embodiments and examples of the present invention described above, the shape data is a two-dimensional coordinate sequence, but it is not limited to two-dimensional, and shape recognition is performed as shape data of a multidimensional coordinate sequence. It may be a device.

Although the embodiments and examples of the present invention have been described above, the present invention is not necessarily limited to the above-described means and techniques, and the object referred to in the present invention can be achieved and the present invention can be realized. It is possible to make appropriate changes and modifications within a range that has the above effect.

[0103]

As described above, according to the present invention,
Shape recognition is possible even when the characteristics of the feature points of the target object of shape recognition change, and the number of feature points is reduced, and the simplification of the shape model construction simplifies the shape recognition procedure. .

[Brief description of drawings]

FIG. 1 is a block configuration diagram showing a first device example according to an embodiment of the present invention.

FIG. 2 is a block configuration diagram showing a second device example according to an embodiment of the present invention.

FIG. 3 is an explanatory diagram showing a shape model of a first method example according to an embodiment of the present invention.

FIG. 4 is an explanatory diagram showing a comparison between a shape model and shape data in the above.

FIG. 5A is an explanatory diagram of a shape model in which a segment is a rectangular shape, and FIG. 5B is a shape model in which a segment is a polygonal line in a third method example according to an embodiment of the present invention. FIG.

FIG. 6 is an explanatory diagram showing a relationship between shape data and a shape model in a fourth method example according to the embodiment of the present invention.

7 is a flowchart of the shape recognition program in FIG.

FIG. 8 is a configuration diagram of a welding arc robot system to which the shape recognition device according to the embodiment of the present invention is applied.

FIG. 9 is a flowchart showing a welding processing procedure of the above.

10A is a schematic view of a crossed elbow pipe which is an object to be welded, and FIG. 10B is an explanatory view showing shape data at a welding point in FIG. 10A.

FIG. 11 is a block flow chart showing steps of processing of a first conventional measuring method.

FIG. 12 is an explanatory diagram of a measurement processing procedure for tracking a moving ship in the above.

FIG. 13 is a block flow diagram showing the processing steps of a second conventional measuring method.

FIG. 14 is an explanatory diagram of a measurement procedure for tracking a welding tracing point in the above.

[Explanation of symbols]

A, B, C, a, b, c ... Welding points A1, B1, C1, a1, b1, c1 ... Two-dimensional coordinate sequence of welding points 1 ... Shape measuring unit 2 ... Shape data storage unit 3 ... Preprocessing unit 4 Profile storage unit 5 Geometric model building unit 5a Geometric model input unit 6 Geometric model storage unit 7 Segment matching unit 8 Feature point detection unit 9 Robot arm 10 Welding torch 11 Range sensor 12 Computer 13 ... Shape recognition device 14 ... Controller 15 ... Welding object 16 ... Cross pipe 17, 17a ... Input data part 18, 18a ... Shape model data part 19 ... Ship

Front page continuation (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06T 1/00

Claims (9)

(57) [Claims] 1. An apparatus for recognizing a predetermined feature point of a recognition target object which may be deformed or moved, and which is an input / output path for inputting / outputting data and an input / output path through the input / output path. A shape measurement unit that forms the measured data into shape data that is a multidimensional coordinate sequence, a profile storage unit that stores the shape data from the shape measurement unit, and the shape that is stored in the profile storage unit at the initial time. The data is segmented by a line segment, a quadrangle, a polygonal line, or a figure with one or more line segments or curves as constituent elements, and the segment and the angle parameter between the segments and the contact point or the closest point between the segments are used. A shape model construction unit that automatically or manually constructs a shape model having a certain feature point, and outputs the feature point from the input / output path, The shape model storage unit that stores the shape model constructed by the shape model construction unit, and the current shape data accumulated in the profile accumulation unit and the shape model stored in the shape model storage unit are collated. Each segment matching unit that extracts the position information of each segment, and the current shape model is created by detecting the feature points and the parameters from the position information of each segment matching unit, and the shape of the shape model storage unit is created. And a feature point detection unit that outputs the feature point through the input / output path. 2. The shape measuring section and the profile accumulating section mutually include a shape data accumulating section for accumulating shape data from the shape measuring section and a shape data accumulating section for accumulating the shape data accumulated in the shape data accumulating section. The shape recognition device according to claim 1, wherein a pre-processing unit that eliminates noise is provided, and the profile storage unit stores the shape data from the pre-processing unit. 3. The shape recognition according to claim 1, wherein the detection of the parameter is performed by using an evaluation value of an evaluation function described with the position of the segment as a variable. apparatus. 4. A method for recognizing a predetermined feature point of a recognition target object which is likely to be deformed and moved, wherein a connection or a set of segments which are not shape-deformed from the input measurement data of the recognition target object. If the shape data has an initial time, the accumulated shape data forms a line segment, a square, a polygonal line, or one or more line segments or curves. Performs segmentation by the figure as an element, constructs a shape model with the parameters of the angle between the segments and the respective segments and the feature points that are the contact points or the nearest points between the segments automatically or by manual input, The feature points are output, and if the shape data is not at the initial time, the accumulated current shape data is compared with the past shape model. According to the above, the position information is obtained, the current shape model is created by detecting the current parameter and the current feature point from the position information, and the past shape model is updated to the current shape model. A shape recognition method characterized by outputting points. 5. The shape data is Data of multidimensional coordinate sequence, The shape recognition method according to claim 4, wherein: 6. The shape recognition method according to claim 4, wherein in forming the shape data, noise in the measurement data is also removed. 7. The detection of the parameter is performed by estimating the parameter using an evaluation value of an evaluation function described by using the position of the segment as a variable. The shape recognition method described in. 8. A recording medium for recording a program for controlling a device for recognizing a predetermined feature point of a recognition target object that may be deformed or moved by a computer, the measurement data being obtained by measuring the recognition target object. Is input to the shape measurement unit through an input / output path, the measurement data of the recognition target object is expressed by the shape measurement unit as shape data of a multidimensional coordinate sequence, and the shape data of the shape measurement unit is stored in the shape data storage unit. In the first step and in the preprocessing unit, the shape data stored in the shape data storage unit is subjected to noise removal processing in multidimensional coordinates, and the shape data subjected to the noise removal processing is stored. The second step of accumulating in the profile accumulating unit and whether the shape data accumulated in the profile accumulating unit is at the initial time In the third step of judging, and in the third step, if the shape data is at the initial time, the shape model construction unit calculates the shape model from the shape data stored in the profile storage unit at the initial time. In the construction of, the fourth step of segmenting a figure having the shape data as a line segment, a quadrangle, a polygonal line, or one or more line segments or curves as constituent elements, and stored in the profile storage unit. A fifth step of automatically or manually extracting from each of the shape data the parameter of each segment and the angle between the segments and the feature point which is the contact point or the closest point between the segments; A sixth step of constructing a shape model described by the parameters and the feature points, A seventh step of storing the model in the shape model storage unit, an eighth step of outputting the feature points of the shape model to the computer and terminating, and a third step of the shape data at an initial time. If not, each segment matching unit compares the past shape model stored in the shape model storage unit with the current shape data stored in the profile storage unit to determine the position of each segment. A ninth step of extracting information, a tenth step of detecting the current parameter and the feature point from the position information by a feature point detection unit, and the feature point detection unit of the current parameter and the feature point. An eleventh step of creating a geometric model having the following, updating and storing the geometric model in the geometric model storage unit; 12th step and, in a series of steps process is executed, a recording medium recording a shape recognition program, characterized in that it terminates with through the power line to the computer. 9. The record recording the shape recognition program according to claim 8, wherein the detection of the parameter is performed by using an evaluation value of an evaluation function described with the position of the segment as a variable. Medium.