JP7415856B2 - Object recognition device, object recognition method and program - Google Patents

Description

The present invention relates to an object recognition device, an object recognition method, and a program for recognizing the position and orientation of a target object.

BACKGROUND ART Conventionally, as a technology related to object recognition that recognizes the position and orientation of a target object measured by a three-dimensional sensor, for example, a robot system disclosed in Patent Document 1 below is known. In this robot system, a three-dimensional sensor acquires scene point cloud data representing the surface shape of a target object, and a point pair feature value (PPF) that defines the geometric relationship of point pairs selected from this scene point cloud data is used. ) is calculated for each point pair, the pose of the target object in the 3D scene is recognized based on these point pair features and point pair features calculated from model point cloud data prepared in advance. Ru. This makes it possible to grasp the orientation of a specific object to be grasped from among the plurality of objects piled up randomly in the box, and to appropriately grasp the object.

Japanese Patent Application Publication No. 2018-189510

By the way, depending on the color, material, etc. of the object to be measured, the reflected light from the object becomes unstable, so that the measurement results may contain noise. For example, in the case of a member with a black surface, a metal member that is prone to specular reflection, a translucent resin member, etc., the reflected light from the target object is likely to be unstable, and the measurement results are likely to contain noise. If a recognition error occurs regarding the position and orientation of the target object due to the influence of the generated noise, there is a problem in that the robot is more likely to make a picking error of the target object.

For this reason, for example, if a high-performance sensor is introduced, not only will the cost of introducing the sensor increase, but the sensor itself will also be large, which may limit its use, such as making it impossible to mount it on a robot arm. Furthermore, although measurement accuracy can be improved by increasing the number of measurements, there is a problem that the response deteriorates because the measurement time becomes longer.

The present invention has been made to solve the above-mentioned problems, and its purpose is to provide a configuration that can suppress the influence of noise on recognition of the position and orientation of a target object.

In order to achieve the above object, the invention described in claim 1 of the claims is as follows:
an acquisition unit (11, 16) that acquires three-dimensional model data consisting of point cloud data representing the three-dimensional shape of the target object (W);
a first selection unit (11) that selects a plurality of pairs of point data from the three-dimensional model data acquired by the acquisition unit;
a first feature amount calculation unit (11) that calculates a feature amount for each of the plurality of sets of the pair of point data selected by the first selection unit;
a three-dimensional sensor (12) that measures three-dimensional measurement data consisting of point cloud data representing the shape of the surface of the target object in a predetermined measurement range;
a second selection unit (11) that selects a plurality of pairs of point data from the three-dimensional measurement data measured by the three-dimensional sensor;
a second feature amount calculation unit (11) that calculates feature amounts for each of the plurality of sets of the pair of point data selected by the second selection unit;
a recognition unit (11) that recognizes the position and orientation of the target object measured by the three-dimensional sensor based on the calculation result by the first feature calculation unit and the calculation result by the second feature calculation unit; ,
Equipped with
The point data measured by the three-dimensional sensor includes XYZ three-dimensional coordinate data whose Z axis is a direction from the three-dimensional sensor toward the target object,
The second selection unit includes one point data selected as a reference point from the three-dimensional measurement data and a coordinate system based on the reference point, in which the distance to the reference point is X with respect to the Z axis. The method is characterized in that point data having the smallest distance from the reference point in a predetermined coordinate system whose coordinates have been transformed to be larger than the axis and the Y-axis are selected as the pair of point data.
Note that the reference numerals in parentheses above indicate correspondence with specific means described in the embodiments described later.

In the invention of claim 1, the calculation result of the first feature amount calculation unit that calculates the feature amount for each of the plurality of pairs of point data selected by the first selection unit from the three-dimensional model data, and the measurement result by the three-dimensional sensor. 3D measurement data measured by the 3D sensor. The position and orientation of the target object are recognized by the recognition unit. The point data measured by the three-dimensional sensor includes XYZ three-dimensional coordinate data whose Z axis is the direction from the three-dimensional sensor toward the target object, and the second selection section selects a reference from the three-dimensional measurement data. One point data selected as a point and a predetermined coordinate system that is a coordinate system based on a reference point and whose coordinates are transformed so that the distance from the reference point is larger than the X and Y axes with respect to the Z axis. Point data having the smallest distance from the reference point are selected as a pair of point data.

Noise caused by unstable reflected light from a target object tends to increase in the direction toward the target object. That is, in the XYZ three-dimensional coordinate data of the point data in the real coordinate system before the above-described conversion, the Z coordinate value is more susceptible to noise than the X coordinate value and the Y coordinate value. Then, when selecting multiple pairs of point data with the smallest distance and calculating the feature values for each, point data affected by noise regarding the Z coordinate value was incorrectly selected, resulting in accurate Feature values cannot be calculated.

Therefore, in the second selection section, the distance to the reference point is the smallest in the predetermined coordinate system in which the coordinates are transformed so that the distance to the reference point is larger with respect to the Z axis than the X and Y axes as described above. Point data is selected. This is because the point data selected in this way is selected so that the X-coordinate value and the Y-coordinate value are given more importance than the Z-coordinate value, which is more susceptible to noise. This makes it difficult to select point data that is likely to be noise among a pair of point data for calculating a feature amount, so it is possible to suppress the influence of noise on recognition of the position and orientation of the target object.

According to the second aspect of the invention, it is possible to realize an object recognition method that has the same effects as the first aspect.
According to the third aspect of the invention, it is possible to realize a program that has the same effects as the first aspect.

FIG. 1 is a block diagram schematically illustrating an object recognition device according to a first embodiment. FIG. 3 is an explanatory diagram illustrating feature amounts calculated by PPF from a pair of point data. 3 is a flowchart illustrating the flow of object recognition processing performed by the control unit of the object recognition device. FIG. 3 is an explanatory diagram illustrating the relationship between a virtual coordinate system and a real coordinate system in the first embodiment. FIG. 3 is an explanatory diagram illustrating an example of calculating a virtual coordinate distance. FIG. 3 is an explanatory diagram illustrating a measurement accuracy ratio of Z accuracy, X accuracy, and Y accuracy; FIG. 7 is an explanatory diagram illustrating the relationship between a virtual coordinate system and a real coordinate system in a first modified example of the first embodiment. FIG. 7 is an explanatory diagram illustrating the relationship between the virtual coordinate system and the real coordinate system when θ≧θ′ in the second modification of the first embodiment. FIG. 7 is an explanatory diagram illustrating the relationship between the virtual coordinate system and the real coordinate system when θ<θ' in the second modification of the first embodiment.

[First embodiment]
The object recognition device, object recognition method, and program according to the first embodiment will be described below with reference to the drawings.
The object recognition device 10 shown in FIG. 1 is a device that recognizes the position and orientation of a target object measured by a three-dimensional sensor. More specifically, the object recognition device 10 is mounted on a robot arm (not shown) that grasps and moves a workpiece W, which is a target object, and uses three-dimensional data of the workpiece W acquired in advance and stored in a storage unit. The robot arm is configured to use the model data to output information regarding the position, orientation, etc. of the workpiece W to be gripped to the control unit of the robot arm. Note that the three-dimensional model data of the workpiece W is composed of point cloud data representing the three-dimensional shape of the workpiece W, and each point cloud data includes three-dimensional coordinate data and a normal vector. .

The object recognition device 10 includes a control section 11 made up of a CPU, etc., a three-dimensional sensor 12, a storage section 13 made up of a semiconductor memory, etc., as well as an operation section 14, a display section 15, and a device for communicating with external devices such as a robot arm. It includes a communication section 16 and the like.

The three-dimensional sensor 12 measures three-dimensional measurement data consisting of point cloud data representing the three-dimensional shape of the surface of the target object by receiving reflected light from a target object such as the workpiece W in a predetermined measurement range. It works like this. A large number of point data constituting the point group data each include XYZ three-dimensional coordinate data and a normal vector whose Z axis is the direction from the three-dimensional sensor 12 toward the target object.

In the object recognition device 10 configured as described above, object recognition is realized by a predetermined program (hereinafter also referred to as an object recognition program) executed by the control unit 11 that functions as a computer that controls the object recognition device 10. Through the process, the position and orientation of the workpiece W are recognized. In this object recognition process, the three-dimensional The position and orientation of the workpiece W measured by the original sensor 12 are recognized, and the recognition results are output to the control section of the robot arm.

More specifically, multiple pairs of point data are selected from the point cloud data, and as illustrated in FIG. 2, XYZ three-dimensional coordinate data (m1, m2) and Feature quantities (F1 to F4) are calculated from the normal vectors (n1, n2) using a calculation method called Point Pair Feature (PPF). The feature quantity calculated in this way corresponds to a four-dimensional quantity that expresses the geometric relationship between the displacement vector between two points and the normal vector at those two points, and the feature quantity is calculated for each pair of point data. The amount is calculated. Calculations of such feature quantities are performed for each of the three-dimensional measurement data of the workpiece W measured using the three-dimensional sensor 12 and the three-dimensional model data of the workpiece W acquired in advance, and the respective calculated feature quantities are matched. The position and orientation of the workpiece W are recognized according to the .

In particular, in this embodiment, regarding the calculation of feature amounts in three-dimensional measurement data, one point data selected as a reference point and a coordinate system based on this reference point, and the distance to the reference point is Z. In a predetermined coordinate system (hereinafter also referred to as a virtual coordinate system) whose coordinates have been transformed so that the axes are larger than the X-axis and Y-axis, the point data whose distance from the above reference point is the smallest is set as a pair of point data. select.

The reason why point data having the smallest distance from the reference point in the virtual coordinate system in which the coordinates have been transformed so that the distance from the reference point to the reference point becomes larger with respect to the Z axis is selected will be explained below.

Depending on the color, material, etc. of the object to be measured, the reflected light from the object may become unstable, so that the measurement results may contain noise. For example, when measuring a workpiece W made of a member with a black surface, a metal member that is prone to specular reflection, a semitransparent resin member, etc., the reflected light from the workpiece W is likely to be unstable, and the measurement results may be affected. Noise is likely to be included. If a recognition error occurs regarding the position and orientation of the workpiece W due to the influence of the generated noise, there is a problem in that the robot is more likely to make a picking error of the workpiece W.

As described above, the noise generated because the reflected light from the target object becomes unstable tends to increase in the direction toward the target object. In other words, when the direction toward the target object is taken as the Z axis, the XYZ three-dimensional coordinate data of the point data in the real coordinate system before the above-mentioned transformation is such that the Z coordinate value is noise compared to the X and Y coordinate values. susceptible to. Then, when selecting multiple pairs of point data with the smallest distance in the real coordinate system and calculating the feature values for each, point data affected by noise regarding the Z coordinate value was incorrectly selected. In some cases, it may become impossible to calculate accurate feature quantities.

Therefore, in the object recognition process performed by the control unit 11 in this embodiment, the reference point is set in a virtual coordinate system in which the coordinates are transformed so that the distance from the reference point is larger on the Z axis than on the X and Y axes, as described above. The point data whose distance to the point is the smallest and its reference point are selected as a pair of point data. This is because the point data selected in this way is selected so that the X-coordinate value and the Y-coordinate value are given more importance than the Z-coordinate value, which is more susceptible to noise. This makes it difficult to select point data that is likely to be noise among a pair of point data for calculating a feature amount, so it is possible to suppress the influence of noise on recognition of the position and orientation of the target object.

The object recognition process performed by the control unit 11 will be described in detail below with reference to the flowchart shown in FIG.
When the object recognition process is started by the control unit 11 in response to a predetermined operation on the operation unit 14, first, a three-dimensional model data acquisition process shown in step S101 is performed. In this process, three-dimensional model data of the workpiece W to be measured is acquired from an external device or the like via the communication unit 16 and stored in the storage unit 13. Note that the control unit 11 and the communication unit 16 may correspond to an example of an “acquisition unit”.

Next, point data selection processing shown in step S103 is performed, and a plurality of pairs of point data are selected from the acquired three-dimensional model data as described above. Subsequently, a three-dimensional model feature calculation process shown in step S105 is performed, and feature quantities (hereinafter also referred to as three-dimensional model feature quantities) using the above-mentioned PPF are calculated for each pair of selected point data. be done. Note that the point data selection process in step S103 may correspond to an example of a "first selection step," and the control unit 11 that executes this point data selection process may correspond to an example of a "first selection unit." Further, the three-dimensional model feature calculation process in step S105 corresponds to an example of a "first feature calculation step", and the control unit 11 that executes this three-dimensional model feature calculation process performs a "first feature calculation step". This may correspond to an example of "part".

When the three-dimensional model feature quantity is calculated in this way, the three-dimensional measurement data measurement process shown in step S107 is performed, and the three-dimensional sensor 12 collects point cloud data representing the shape of the surface of the workpiece W in a predetermined measurement range. Three-dimensional measurement data consisting of is measured.

Next, point data selection processing shown in step S109 is performed, and a plurality of pairs of point data are selected from the measured three-dimensional measurement data using the coordinate transformation described above. Note that the point data selection process in step S109 may correspond to an example of a "second selection step," and the control unit 11 that executes this point data selection process may correspond to an example of a "second selection unit."

Specifically, in this embodiment, as shown in FIG. 4, in a virtual coordinate system in which a perfect circle C1 in a real coordinate system is transformed into an ellipse C2, points (hereinafter referred to as The distance between the pair candidate point Pc) and the reference point Po (hereinafter also referred to as virtual coordinate distance) is calculated, and the pair candidate point Pc with the shortest virtual coordinate distance is selected. A and b in FIG. 4 are the aspect ratio parameters of the ellipse C2, and the smaller the short axis b is with respect to the long axis a, the more the Z coordinate value affects the X and Y coordinate values with respect to the virtual coordinate distance. It becomes smaller.

Therefore, the actual coordinate distance between the pair candidate point Pc located on the perfect circle C1 in FIG. 4 and the reference point Po is D1, and the distance between the straight line connecting this pair candidate point Pc and the reference point Po and the ellipse C2 in FIG. When the real coordinate distance to the intersection is D2, the conversion coefficient r2 for converting from the real coordinate system to the virtual coordinate system is based on D2/D1 and using the angle θ of the pair candidate point Pc in FIG. Equation (1) in FIG. 4 holds true. Note that the angle θ is expressed by equation (2) in FIG. 4 when the coordinates of the reference point Po are (0, 0, 0) and the coordinates of the pair candidate point Pc are (x, y, z).

Therefore, based on the conversion coefficient r2 obtained from the aspect ratio parameters a and b of the ellipse C2 and the angle θ, the virtual coordinate distance is obtained as D1/r2, and the pair candidate point Pc at which this virtual coordinate distance is the shortest is found. , and select it as a pair with that reference point. For example, when the conversion coefficient r2 is calculated as 0.7 and the actual coordinate distance D1 is measured as 1 mm, the virtual coordinate distance is calculated as 1.43 mm.

In this way, by selecting the pair candidate point Pc for which the virtual coordinate distance in the virtual coordinate system is the shortest, it is possible to reduce the influence of the Z coordinate value, which tends to cause noise, on the X and Y coordinate values. can.

For example, as illustrated in FIG. 5, in a virtual coordinate system in which a perfect circle C1 in the real coordinate system is transformed into an ellipse C2 with a ratio of major axis a to minor axis b of 1:0.5, the reference point Po(0, 0, 0), a case is assumed in which a virtual coordinate distance is calculated between a pair candidate point Pc1 (1, 1, 0) and a pair candidate point Pc2 (1, 0, 1). Since the pair candidate point Pc1 (1, 1, 0) has θ=0° and r2=1, the virtual coordinate distance is calculated as 1.41. On the other hand, since θ=45° and r2=0.63 for the pair candidate point Pc2 (1,0,1), the virtual coordinate distance is calculated as 2.24. Therefore, the pair candidate point Pc1 (1, 1, 0) whose virtual coordinate distance is short is selected.

If the virtual coordinate system is not considered as described above, the actual coordinate distance from the reference point Po to the pair candidate point Pc1 is the same as the actual coordinate distance from the reference point Po to the pair candidate point Pc2, so the above-mentioned It can be seen that by considering such a virtual coordinate system, the influence of the Z coordinate value can be reduced on the X and Y coordinate values.

As described above, when a plurality of pairs of point data are selected from the three-dimensional measurement data in the point data selection process in step S109, the three-dimensional measurement feature calculation process shown in step S111 is performed to select the selected points. A feature amount (hereinafter also referred to as a three-dimensional measurement feature amount) using the above-mentioned PPF is calculated for each pair of point data. Note that the three-dimensional measurement feature calculation process in step S111 corresponds to an example of a "second feature calculation step", and the control unit 11 that executes this three-dimensional measurement feature calculation process performs a "second feature calculation step". This may correspond to an example of "part".

Next, the recognition process shown in step S113 is performed, and the position and orientation of the workpiece W measured by the three-dimensional sensor 12 are determined based on the matching between each calculated three-dimensional model feature and each three-dimensional measurement feature. Recognized. Then, the output process shown in step S115 is performed, and information regarding the recognition result etc. of the workpiece W as described above is outputted to the control unit of the robot arm via the communication unit 16. Note that the control unit 11 that executes the recognition process in step S113 may correspond to an example of a “recognition unit”.

In this way, the control unit of the robot arm can acquire the position and orientation of the workpiece W to be gripped from the object recognition device 10, so that the workpiece W can be gripped accurately.

As explained above, in the object recognition device 10 according to the present embodiment, the feature amount is calculated in step S105 for each of the plurality of pairs of point data selected from the three-dimensional model data in the point data selection process in step S103. From the calculation results of the three-dimensional model feature amount calculation process and the three-dimensional measurement data measured by the three-dimensional sensor 12, feature amounts are calculated for each of the plurality of pairs of point data selected in the point data selection process of step S109. Based on the calculation result of the three-dimensional measurement feature amount calculation process of step S111, the position and orientation of the workpiece W measured by the three-dimensional sensor 12 are recognized by the recognition process of step S113. The point data measured by the three-dimensional sensor 12 includes XYZ three-dimensional coordinate data whose Z axis is the direction from the three-dimensional sensor 12 toward the workpiece W. In the point data selection process in step S109, three-dimensional From the measurement data, coordinate transformation is performed using one point data selected as a reference point and a coordinate system based on the reference point such that the distance to the reference point is larger on the Z axis than on the X and Y axes. Point data having the smallest distance from the reference point in the predetermined coordinate system are selected as a pair of point data.

In this way, in the point data selection process in step S109, the distance to the reference point is the smallest in the predetermined coordinate system in which the coordinates are transformed so that the distance to the reference point is larger on the Z axis than on the X and Y axes. Point data is selected. This is because the point data selected in this way is selected so that the X-coordinate value and the Y-coordinate value are given more importance than the Z-coordinate value, which is more susceptible to noise. This makes it difficult to select point data that is likely to be noise among a pair of point data for calculating a feature amount, so it is possible to suppress the influence of noise on recognition of the position and orientation of the target object.

Note that in the coordinate transformation used in the point data selection process shown in step S109 above, a perfect circle C1 in the real coordinate system is transformed into an ellipse C2 with a ratio of major axis a to minor axis b of 1:0.5. is not limited to this, but a process of converting at another ellipse ratio may be employed depending on the target object or the like. For example, when performing object recognition processing on a target object with a measurement accuracy ratio as shown in Fig. 6, since the Z accuracy is approximately 100 times the X accuracy and Y accuracy, : Processing may be adopted in which a perfect circle in the real coordinate system is transformed into an ellipse with the short axis b of 1:0.01.

Furthermore, the coordinate transformation used in the point data selection process shown in step S109 is not limited to the virtual coordinate system in which a perfect circle in the real coordinate system is transformed into an ellipse as described above; Any predetermined coordinate system may be used as long as the coordinates are transformed so that the distance is larger on the Z axis than on the X and Y axes.

For example, as a first modification of the present embodiment, as illustrated in FIG. 7, a virtual coordinate system is adopted in which a perfect circle C1 in the real coordinate system is transformed into a rhombus C3 (major axis a, minor axis b). Good too. In this virtual coordinate system, the actual coordinate distance between the pair candidate point Pc located on the perfect circle C1 in FIG. 7 and the reference point Po is D1, and the straight line connecting this pair candidate point Pc and the reference point Po is the diamond in FIG. When the real coordinate distance to the intersection with C3 is D3, the conversion coefficient r3 for converting from the real coordinate system to the virtual coordinate system uses the angle θ of the pair candidate point Pc in FIG. 7 based on D3/D1. Therefore, equation (3) in FIG. 7 is established. Note that the angle θ is expressed by equation (4) in FIG. 7 when the coordinates of the reference point Po are (0, 0, 0) and the coordinates of the pair candidate point Pc are (x, y, z).

By adopting a virtual coordinate system as shown in Figure 7, the selection of reference points and pair candidate points is performed more rigorously for the virtual coordinate system using an ellipse, so malfunctions due to low-accuracy recognition are tolerated. It can be applied to applications where it is required to guarantee highly accurate recognition results. Specifically, for example, a system that combines a high-precision three-dimensional sensor and a robot can be used for picking and assembling precision parts.

For example, as a second modification of the present embodiment, as illustrated in FIGS. 8 and 9, a virtual coordinate system in which a perfect circle C1 in the real coordinate system is transformed into a rectangle C4 (long axis a, short axis b) system may be adopted. In this virtual coordinate system, as shown in FIGS. 8 and 9, the actual coordinate distance between the pair candidate point Pc located on the perfect circle C1 and the reference point Po is D1, and the pair candidate point Pc and the reference point Po are When the real coordinate distance to the intersection of the connecting straight line and the rectangle C4 is D4, the conversion coefficient r4 for converting from the real coordinate system to the virtual coordinate system uses the angle θ of the pair candidate point Pc based on D4/D1. Therefore, equation (5) in FIG. 8 or equation (6) in FIG. 9 is established. Note that equation (5) in FIG. 8 holds true when θ≧θ', and equation (6) in FIG. 9 holds true when θ<θ'. In addition, the angle θ is expressed by equation (7) in FIGS. 8 and 9 when the coordinates of the reference point Po are (0, 0, 0) and the coordinates of the pair candidate point Pc are (x, y, z). The angle θ' is expressed by equation (8) in FIGS. 8 and 9.

By adopting a virtual coordinate system as shown in Figures 8 and 9, reference points and pair candidate points are selected more loosely, so it can be applied to applications where low-accuracy recognition is acceptable. can. Specifically, for example, it is a system that combines a low-precision three-dimensional sensor and a robot, and can be used to transport large pallets.

[Other embodiments]
Note that the present invention is not limited to the above-described embodiments and modifications, and may be embodied as follows, for example.
(1) The feature amount of the three-dimensional model data is not limited to being calculated from the three-dimensional model data acquired via the communication unit 16 as in steps S101, S103, and S105 in the object recognition process described above. Feature amounts of three-dimensional model data calculated in advance may be input to the object recognition device 10 via the communication unit 16 from an external device or the like. In this case, the object recognition device 10 may include an external device that calculates the feature amount of the three-dimensional model data.

(2) In object recognition processing, when recognizing the position and orientation of a target object that has the same shape as the previous time, the processing of steps S101, S103, and S105 is omitted by using the 3D model feature obtained last time. You may.

(3) The present invention is not limited to being applied to object recognition processing that calculates feature quantities using PPF, but is also applicable to multiple sets of point data from point cloud data, such as Fast Point Feature Histograms (FPFH). It can be employed in object recognition processing that selects and calculates feature points.

(4) The object recognition device 10 according to the present invention is not limited to being configured to be mounted on a robot arm that grasps and moves a workpiece W as a target object, but is also a device for recognizing other target objects, For example, it may be configured as an object recognition device for monitoring the outer periphery to determine the presence or absence of an intruder.

10...Object recognition device 11...Control unit (acquisition unit, first selection unit, first feature calculation unit, second selection unit, second feature calculation unit, recognition unit)
12...Three-dimensional sensor 16...Communication section (acquisition section)
W...Work (target object)

Claims (3)

an acquisition unit that acquires three-dimensional model data consisting of point cloud data representing the three-dimensional shape of the target object;
a first selection unit that selects a plurality of pairs of point data from the three-dimensional model data acquired by the acquisition unit;
a first feature amount calculation unit that calculates a feature amount for each of the plurality of sets of the pair of point data selected by the first selection unit;
a three-dimensional sensor that measures three-dimensional measurement data consisting of point cloud data representing the shape of the surface of the target object in a predetermined measurement range;
a second selection unit that selects a plurality of pairs of point data from the three-dimensional measurement data measured by the three-dimensional sensor;
a second feature amount calculation unit that calculates feature amounts for each of the plurality of sets of the pair of point data selected by the second selection unit;
a recognition unit that recognizes the position and orientation of the target object measured by the three-dimensional sensor based on the calculation result by the first feature calculation unit and the calculation result by the second feature calculation unit;
Equipped with
The point data measured by the three-dimensional sensor includes XYZ three-dimensional coordinate data whose Z axis is a direction from the three-dimensional sensor toward the target object,
The second selection unit includes one point data selected as a reference point from the three-dimensional measurement data and a coordinate system based on the reference point, in which the distance to the reference point is X with respect to the Z axis. An object recognition device characterized in that point data having the smallest distance from the reference point in a predetermined coordinate system whose coordinates have been transformed so as to be larger than an axis and a Y axis are selected as the pair of point data. An object recognition method for recognizing the position and orientation of a target object measured by a three-dimensional sensor,
acquiring three-dimensional model data consisting of point cloud data representing the three-dimensional shape of the target object;
a first selection step of selecting a plurality of pairs of point data from the three-dimensional model data;
a first feature amount calculation step of calculating feature amounts for each of the plurality of sets of the pair of point data selected in the first selection step;
measuring three-dimensional measurement data consisting of point cloud data representing the shape of the surface of the target object in a predetermined measurement range using the three-dimensional sensor;
a second selection step of selecting a plurality of pairs of point data from the three-dimensional measurement data;
a second feature amount calculation step of calculating feature amounts for each of the plurality of sets of the pair of point data selected in the second selection step;
a step of recognizing the position and orientation of the target object measured by the three-dimensional sensor based on the calculation result of the first feature amount calculation step and the calculation result of the second feature amount calculation step;
Equipped with
The point data measured by the three-dimensional sensor includes XYZ three-dimensional coordinate data whose Z axis is a direction from the three-dimensional sensor toward the target object,
In the second selection step, one point data selected as a reference point from the three-dimensional measurement data and a coordinate system based on the reference point, the distance from which is X with respect to the Z axis. An object recognition method characterized in that point data having the smallest distance from the reference point in a predetermined coordinate system whose coordinates have been transformed so as to be larger than the axis and the Y axis are selected as the pair of point data. . A program executed by a computer that controls an object recognition device that recognizes the position and orientation of a target object measured by a three-dimensional sensor,
acquiring three-dimensional model data consisting of point cloud data representing the three-dimensional shape of the target object;
a first selection step of selecting a plurality of pairs of point data from the three-dimensional model data;
a first feature amount calculation step of calculating feature amounts for each of the plurality of sets of the pair of point data selected in the first selection step;
measuring three-dimensional measurement data consisting of point cloud data representing the shape of the surface of the target object in a predetermined measurement range using the three-dimensional sensor;
a second selection step of selecting a plurality of pairs of point data from the three-dimensional measurement data;
a second feature amount calculation step of calculating feature amounts for each of the plurality of sets of the pair of point data selected in the second selection step;
a step of recognizing the position and orientation of the target object measured by the three-dimensional sensor based on the calculation result of the first feature amount calculation step and the calculation result of the second feature amount calculation step;
cause the computer to execute
The point data measured by the three-dimensional sensor includes XYZ three-dimensional coordinate data whose Z axis is a direction from the three-dimensional sensor toward the target object,
In the second selection step, one point data selected as a reference point from the three-dimensional measurement data and a coordinate system based on the reference point, the distance from which is X with respect to the Z axis. A program characterized in that point data having the smallest distance from the reference point in a predetermined coordinate system whose coordinates have been transformed so as to be larger than an axis and a Y-axis are selected as the pair of point data.

Images