JP5585826B2 - Object recognition method and apparatus - Google Patents

Description

  The present invention relates to a target object recognition method and apparatus for generating a target object area occupied by a target object in a virtual coordinate space based on the shape data of the target object obtained by measuring or imaging the target object.

  Conventionally, in order to recognize a target object, by measuring the distance from the measurement point to each measured point on the target object or by imaging the target object, the shape data of the target object (shape data) ) To get. Based on the shape data, an object region occupied by the object is generated in a two-dimensional or three-dimensional virtual coordinate space. In this way, the shape of the object is expressed in the virtual coordinate space.

  In addition, there exists the following patent document 1 as a prior art document of this application. In Patent Document 1, the feature amount of each divided region of the recognition target is obtained based on the three-dimensional shape data of the recognition target, and the degree of coupling with another divided region adjacent to the divided region is calculated based on the feature amount. When the degree of coupling is equal to or less than the integration threshold, the divided areas are integrated with other divided areas.

JP 2006-185132 A

  However, in the three-dimensional virtual coordinate space, an unclear shape portion may occur in the outer peripheral portion of the target object region generated as described above. Examples of unclear shapes are shown in FIGS. 1A and 1B.

  As shown in FIG. 1A, two target object regions may be generated for a single target object. For example, an object that is not originally related to the target object may exist in the vicinity of the target object, and another target object region is generated for such an object. In addition, unevenness may exist in the background of the object, and another object region is generated with respect to such uneven portions. Furthermore, when acquiring data on the shape of an object using a laser distance meter, the reflection intensity of the laser is too strong, so that another object that reflects the laser with such intensity A body region is generated.

  As shown in FIG. 1B, object regions are generated for two partially overlapping objects, but these object regions are connected to each other. For this reason, there are cases where the boundaries between these object areas cannot be recognized, or these object areas cannot be recognized as separate areas.

  In addition, as shown in FIG. 1B, a small region that should not originally exist in a narrow space between two adjacent objects may be generated as another object region. For example, in the narrow space, the side surface of the target object may be generated as another target object area, or may be generated as another target object area due to the presence of a discontinuous part (discontinuous part) in the background. is there.

  Therefore, an object of the present invention is to generate an unclear shape portion on the outer periphery of the object area when generating the object area occupied by the object in the virtual coordinate space based on the object shape data. It is an object of the present invention to provide an object recognition method and apparatus capable of reducing the performance.

To achieve the above object, according to the present invention, there is provided an object recognition method for expressing the shape of an object in a virtual coordinate space,
By measuring the distance from the measurement point to each measured point on the target object or by capturing the target object, the shape data acquisition step for acquiring the target object shape data;
Based on the shape data, an area generation step for generating an object area occupied by the object in the virtual coordinate space;
An area correction step for performing correction processing on the generated object area,
The correction process includes at least one of a reduction process and an enlargement process,
The outer peripheral part or inner peripheral part of the target body area is the processing target area,
(A) The reduction process is a process of deleting a processing target area that is a part of the target object area from the target object area,
(B) The object recognition method is characterized in that the enlargement process is a process that is an area other than the object area and integrates the object area adjacent to the object area into the object area. The

  According to a preferred embodiment of the present invention, the processing target area is an outer peripheral portion of the target object area, and as a result of performing the reduction process in the area correction step, the target object area is divided into a plurality of target object areas. In this case, the plurality of target object areas are set as different target object areas in the virtual coordinate space.

  According to a preferred embodiment of the present invention, when the object region disappears as a result of performing the reduction process in the region correction step, the object region is deleted in the virtual coordinate space.

  Preferably, in the region correction step, the enlargement process is performed after the reduction process.

  According to another embodiment of the present invention, a region other than the target region is included inside the target region, and the target region is included in the other region. And the correction process is an enlargement process for integrating the inner periphery part into the object region.

In order to achieve the above object, according to the present invention, there is provided a target object recognition device for expressing the shape of a target object in a virtual coordinate space,
By measuring the distance from the measurement point to each measured point on the object, or by capturing the object, shape data acquisition means for acquiring the object shape data;
Based on the shape data, an area generating means for generating an object area occupied by the object in the virtual coordinate space;
Area correction means for performing correction processing on the generated object area,
The correction process includes at least one of a reduction process and an enlargement process,
The outer peripheral part or inner peripheral part of the target body area is the processing target area,
(A) The reduction process is a process of deleting a processing target area that is a part of the target object area from the target object area,
(B) The object recognition apparatus is characterized in that the enlargement process is a process that is an area other than the object area and integrates the object area adjacent to the object area into the object area. The

  According to the present invention described above, (A) a reduction process for deleting an outer peripheral portion of an area included in the target body area from the target body area, and (B) outside the target body area for the target body area generated in the virtual coordinate space. And at least one of the enlargement processing for integrating the outer peripheral portion of the region adjacent to the target body region into the target body region. A simple shape portion is deleted from the object region or integrated into the object region. Thereby, it can suppress that an indefinite shape part generate | occur | produces in the outer peripheral part of a target object area | region.

It is a figure explaining the problem of a prior art. It is a functional block diagram of the target object recognition apparatus by embodiment of this invention. 3 is a flowchart of an object recognition method according to an embodiment of the present invention. (A) is an IVA-IVA arrow line view of FIG. 2, (B) shows the some object area | region produced | generated with respect to the 2 object body of (A). It is a flowchart which shows a reduction process. It is a figure explaining the measurement of the number of triangular meshes. (A) shows the result of performing the reduction process on each object region shown in FIG. 4B, and (B) shows the result of performing the enlargement process on each object region shown in (A). Show. It is a flowchart which shows an area number assigning process. It is a flowchart which shows an expansion process. The result of performing the enlargement process only twice for each object region in FIG. It is a figure for demonstrating other embodiment of this invention.

  A preferred embodiment of the present invention will be described with reference to the drawings. In addition, the same code | symbol is attached | subjected to the common part in each figure, and the overlapping description is abbreviate | omitted.

  FIG. 2 is a functional block unit of the object recognition device 10 according to the embodiment of the present invention. The object recognition apparatus 10 is an apparatus that expresses the shape of the object 1 in a virtual coordinate space. The object recognition apparatus 10 performs shape data acquisition means 3 that performs step S1 described later, area generation means 5 that performs step S2 described below, and An area correction means 7 for performing step S3 is provided. The shape data acquisition means 3 acquires the shape data of the target object 1 by measuring the distance from the measurement point to each measured point on the target object 1 or by imaging the target object 1. The area generation unit 5 generates an object area occupied by the object 1 in the virtual coordinate space based on the shape data. The area correction means 7 performs a correction process on the generated object area. The area correction means 7 includes a reduction processing means 7a that performs a reduction process described later, an area number assignment means 7b that performs an area number assignment process described later, and an expansion processing means 7c that performs an enlargement process described later.

  The area generation means 5 and the area correction means 7 may be constituted by a CPU (central processing unit) of a computer and a numerical calculation program. Each numerical calculation program may be stored in an appropriate storage unit so that it can be executed by the region generation unit 5 and the region correction unit 7.

  FIG. 3 is a flowchart of the object recognition method. As shown in FIG. 3, the object recognition method includes a shape data acquisition step S1, an area generation step S2, and an area correction step S3.

(Shape data acquisition step)
In the shape data acquisition step S1, the shape data of the object 1 is acquired. This shape data may be acquired by shape data acquisition means 3 such as a laser distance meter or a camera.
When the shape data acquisition means 3 is a laser distance meter, the distance from the measurement point to each measurement point on the object 1 is measured by emitting laser light from the measurement point and receiving the reflected laser. measure. These distance data are acquired as the shape data. When the shape data acquisition means 3 is a laser distance meter, the above-mentioned virtual coordinate space is preferably a two-dimensional or three-dimensional coordinate space.
When the shape data acquisition unit 3 is a camera, the target 1 included in the image data is acquired as the above-described shape data by imaging the target 1. The camera may be, for example, a CCD (Charge-Coupled Device) camera. When the shape data acquisition means 3 is a camera, the virtual coordinate space described above is preferably a two-dimensional coordinate space.

  FIG. 4A is an IVA-IVA arrow view of FIG. 2 and shows an example of the object 1. In the example of FIG. 4A, the target 1 is an object (here, a bag) that is gripped and transported by a robot hand. Moreover, in the example of FIG. 4A, the two objects 1 partially overlap.

(Region generation step)
In region generation step S <b> 2, the region generation means 5 generates a target object region occupied by the target object 1 in the virtual coordinate space based on the shape data of the target object 1. This object region may be expressed in a virtual coordinate space using a mesh (for example, a triangular mesh) or a lattice. FIG. 4B shows a plurality of object regions generated for the two objects 1 in FIG. 4A by this region generation step S2. That is, FIG. 4B shows the object region generated based on the shape data obtained by laser measurement as described above with respect to the two objects 1 from the angle seen as in FIG. . FIG. 4B shows the case of the object region generated in the two-dimensional virtual coordinate space.

  Note that the object region may be expressed in a two-dimensional virtual coordinate space, or may be expressed in a three-dimensional virtual coordinate space. In the case of expressing in a three-dimensional virtual coordinate space, shape data capable of recognizing the three-dimensional shape of the object 1 is acquired in the shape data acquisition step S1. For example, shape data that can recognize a three-dimensional shape may be acquired by performing laser measurement on the target 1 as described above (more preferably, from a plurality of directions).

  In the area generation step S2, when a plurality of object areas are generated, as shown in FIG. 4B, each object area has a different area number in order to identify the object areas from each other. Is granted. In FIG. 4B (and FIGS. 6, 7, and 10), the white circle is the apex of the triangular mesh.

Each object region may be specified (represented) by any of the following specifying methods 1 to 3.
・ Specific method 1
Each object region is represented by one vertex or a collection of two or more vertices. In this case, the object can be recognized based on a set of these vertices (vertex coordinates).
・ Specific method 2
Among the vertices included in the target object area, each vertex located on the outermost side of the target object area is specified, and a curve that makes a round to pass through these vertices is defined as the boundary of the target object area. . The area within this boundary is set as the object area.
・ Identification method 3
A new boundary is specified by expanding (moving) the boundary used in the identification method 2 by a set distance to the outside of the object region within the boundary. The area within the new boundary is set as the object area. In FIG. 6 and the like, the object region (the hatched region) is expressed by this specifying method 3.

  In the example of FIG. 4B, each vertex of the triangular mesh corresponds to a point to be measured on the object 1 by laser measurement. Further, in this example, in the region generation step S2, for each target region, the region number of the target region is given to each vertex of the triangular mesh in the target region.

In area correction step S3, the area correction means 7 performs correction processing on the generated object area. The correction process includes at least one of the reduction process and the enlargement process (both in the present embodiment). In this embodiment, the enlargement process is performed after the reduction process. In the present embodiment, the outer peripheral portion of the target body region is a processing target region (hereinafter referred to as a region outer peripheral portion).
(A) The reduction process is a process of deleting an outer peripheral portion of the area included in the target object area to be subjected to this process from the target object area.
(B) The enlargement process is a process that integrates an outer peripheral portion of an area that is located outside the object area to be processed and is adjacent to the object area into the object area.
When the object region is specified using the specifying method 1 described above, the region outer peripheral portion is each vertex that exists over the entire outer periphery of the target object region. When the object region is specified by using the specifying method 2 or 3 described above, the region outer peripheral portion is a region near the boundary used for the specification. In this case, the region near the boundary may extend over the entire outer periphery of the object region.

As a result of performing the reduction process, when the target body area is divided into a plurality of target body areas, the plurality of target body areas are set as different target body areas in the virtual coordinate space.
On the other hand, when the object region disappears as a result of the reduction process, the object region is deleted in the virtual coordinate space.

  In the present embodiment, as shown in FIG. 3, the area correction step S3 includes step S31 for performing the above-described reduction process, step S32 for newly assigning an area number, and step S33 for performing the above-described enlargement process. . Hereinafter, the reduction process (step S31), the area number assignment process (step S32), and the enlargement process (step S33) will be described in detail.

(Reduction processing)
In step S31 for performing the reduction process, the object region may be reduced by deleting all the vertices in the outermost part (corresponding to the outer periphery of the region) in the object region.

  FIG. 5 is a flowchart showing the reduction process. Step S31 for performing the reduction process includes steps S311 to S318 as shown in FIG. Steps S311 to S318 are performed for each object region.

In step S <b> 311, one vertex (triangular mesh or lattice vertex) is selected from one object region to be reduced. Let the vertex selected here be the vertex i.
In step S312, each vertex that is adjacent to the vertex i and belongs to the same object area as the vertex i is specified. Let each vertex identified here be a vertex j.
In step S313, for each vertex j, the number of triangular meshes to which the vertex j and vertex i belong is measured. The triangular mesh to be measured is limited to the triangular mesh belonging to the object region including the vertices i and j. For example, in the case of FIG. 6A, the number of triangular meshes (shaded portions in the figure) to which the vertices i, j belong is one, and in the case of FIG. 6B, the vertices i, j The number of triangular meshes (shaded portions in the figure) to which the symbol belongs is two.
In step S314, it is determined whether there is a combination of vertex j and vertex i for which the measured number of triangular meshes is one. If this determination is YES, the process proceeds to step S315, and if NO, the process proceeds to step S316.
In step S315, the vertex i is specified as a deletion candidate, and the process proceeds to step S317.
In step S316, the vertex i is maintained as it is as the vertex in the object area, and the process proceeds to step S317.
In step S317, it is determined in step S311 whether or not all the vertices in the target area currently targeted for the reduction process have been selected. If the determination in step S317 is yes, the process proceeds to step S318. If the determination in step S317 is no, the process proceeds to step S311, where one vertex that has not been selected in step S311 is selected, and each step is repeated as described above.
In step S318, all the vertices identified as deletion candidates in step S315 are not included in the target area currently targeted for reduction processing, and all the vertices are set as vertices outside the target area. Thereby, the object region is reduced. Note that each vertex of the deletion candidate is a vertex corresponding to the outer peripheral portion of the target body region.

By performing such a reduction process on each object region shown in FIG. 4B generated in step S2, each object region becomes as shown in FIG. 7A. Here, the object areas of area numbers 2 to 4 in FIG. 4B are considered to be extinguished because there are no vertices existing in the interior by the reduction process, and the virtual coordinates as shown in FIG. Deleted in space.
In the case of FIG. 4B, two objects 1 are generated as one object area having area number 1. However, two objects as shown in FIG. Divided into body regions.

  The above reduction process may be performed only once for each target body region, or may be performed twice or more. FIG. 7A shows the result of performing the above-described reduction process only once for each object region in FIG. 4B. The number of times of reduction processing may be determined according to the amount to be reduced by one reduction processing, the required recognition accuracy of the target object, and the like.

  By performing the reduction process, the object region can be divided as shown in FIG.

(Area number assignment process)
As shown in FIG. 7B, when the target area disappears or is divided by the reduction process, each target area is identified in order to identify each target area existing after the reduction process. An area number assigning process is performed for reassigning the area number. That is, in the region number assigning process, region numbers are reassigned to a plurality of target object regions after the reduction process. In FIG. 7A, new region numbers 1 and 2 are assigned to two divided object regions by the region number assigning process.
Note that the region number assigning process may be omitted when the object region has not disappeared or divided due to the reduction process. That is, step S33 is performed after step S31.

  FIG. 8 is a flowchart showing the area number assigning process. Step S32 for performing such an area number assigning process includes steps S321 to S326 as shown in FIG.

In step S321, after the above-described reduction process, one object area is selected from a plurality of object areas existing in the virtual coordinate space. In the second and subsequent steps S321, an object region that has not yet been selected is selected.
In step S322, a region number n to be assigned is determined. In step S321 for the first time, n = 1 is set, and thereafter, the value of n is incremented by one every time step S321 is passed.
In step S323, one vertex that has not yet been selected in the object region selected in step S321 is selected.
In step S324, an area number n is assigned to the vertex selected in step S323.
In step S325, it is determined whether or not the region number n is assigned in step S324 to all the vertices in the object region selected in step S321. If this determination is YES, the process proceeds to step S326, and if NO, the process returns to step S323.
In step S326, it is determined whether all object regions have been selected in step S321. If this determination is YES, the region number assignment process is terminated, and if NO, the process returns to step S321.

  In the area number assigning process, no area number is assigned to the vertices outside the object area.

(Enlargement processing)
In the enlargement process, by including the vertex that is adjacent to the outermost vertices in the target area and outside the target area (vertices corresponding to the outer peripheral part of the target area) in this target area, You may enlarge the body area.

  FIG. 9 is a flowchart showing the enlargement process. Step S33 for performing the enlargement process may be performed by steps S331 to S338 as shown in FIG. In the example of FIG. 9, only the vertices of the triangular mesh corresponding to the measurement points on the target 1 by the above-described laser measurement are the targets of the enlargement process.

In step S331, one object region to be enlarged is selected from one or more object regions existing in the virtual coordinate space. In this example, the object area with the smallest area number is selected from object areas that have not yet been enlarged.
In step S332, one vertex (triangular mesh or lattice vertex) is selected from the object region selected in step S331. Let the vertex selected here be the vertex i.
In step S333, each vertex adjacent to the vertex i is specified. Let each vertex identified here be a vertex j.
In step S334, it is determined whether any vertex j is not assigned an area number. If this determination is YES, the process proceeds to step S335, and if NO, the process proceeds to step S336.
In step S335, the area number of the object area selected in step S331 is assigned to each vertex j to which no area number is assigned, and the process proceeds to step S336.
In step S336, it is determined whether all the vertices in the object region selected in step S331 have been selected in step S332. If this determination is YES, the process proceeds to step S337, and if NO, the process returns to step S332.
In step S337, each vertex j given the region number in step S335 is integrated with the target region selected in step S331, thereby expanding the target region. When step S337 is completed, the process proceeds to step S338.
In step S338, it is determined whether all object regions have been selected in step S331. If this determination is YES, the enlargement process is terminated, and if NO, the process returns to step S331.

  The enlargement process described above may be performed only once, or may be performed twice or more. FIG. 7B shows the result of performing the above-described enlargement process only once for each object region in FIG. FIG. 10 shows a result of performing the above-described enlargement process only twice for each object region in FIG. The number of enlargement processes may be determined according to the amount to be enlarged by one enlargement process, the required recognition accuracy of the target object, and the like.

  By performing the enlargement process, it is possible to fill a gap (between a plurality of object areas) generated by the reduction process, and it is possible to enlarge the object area that has been reduced by the reduction process.

The object 1 can be recognized based on the object area obtained by the object recognition method according to the embodiment of the present invention. This recognition may be recognition of the position, posture, shape, or area of the target 1.
When the target body 1 is a transported object (for example, a product part) that is gripped and transported by the robot hand, the control device transports based on the recognition of the transported object (recognition of position and orientation). By controlling the robot, the object 1 may be gripped and transported to a predetermined position by the transport robot.
When the object 1 is a person, the computer determines how many people are within the range of laser measurement or imaging in the shape data acquisition step based on the recognition of the human face (recognition of the shape). You can do it. In this case, the range (place) of laser measurement or imaging in the shape data acquisition step may be a ticket gate of a transportation facility (for example, a train) on which people are placed, a railroad crossing of a train track, or at the entrance of a building. There may be. By judging how many people are in the range of such laser measurement or imaging, you can know how crowded people are,
When the object 1 is an electroplating process target (for example, a nail, a nut, a bolt, an industrial machine or a car part), the surface area is calculated based on the recognition of the area of the process target. The time for electroplating can be determined based on the surface area.
When a product (for example, a press-processed product) is manufactured based on CAD data, the product may be the object 1. In this case, based on the recognition of the shape of the product, it can be determined whether the shape of the product matches the CAD data.

  The present invention is not limited to the above-described embodiment, and various changes can be made without departing from the scope of the present invention. For example, the following contents may be adopted alone or in any combination. In that case, the other points may be the same as described above.

  Each vertex of the triangular mesh described above may correspond to a point to be measured on the object 1 by laser measurement, or may not correspond to a point to be measured by laser measurement.

  In the above description, the reduction or enlargement of the object region by the reduction process or the enlargement process is performed based on the vertices of the mesh. However, the reduction or enlargement may be performed by a set amount (specific dimension). For example, in the case of using the boundaries of the identification methods 2 and 3 described above, the region outer peripheral portion has a width corresponding to the set amount in the direction from the outside to the inside of the target body region. Therefore, in the reduction process, the outer peripheral part of the width included in the target body area is deleted from the target body area, and in the enlargement process, the outer peripheral part of the width of the area located outside the target target area is changed to the target object area. Integrated.

  In the above description, the reduction process and the enlargement process are performed, but the enlargement process may not be performed.

  The reduction process described above may not be performed. In this case, instead of the reduction process, a process of deleting an object area whose size is smaller than the set dimension may be performed, and then the area number assignment process and the expansion process may be performed as described above. Here, the dimension may be the number of vertices included in the object region when using the above-described identification method 1, and when using the above-described identification method 2 or 3, It may be area or volume.

In the above description, the processing target area is the outer peripheral part of the target body area, but may be the inner peripheral part of the target body area. This inner peripheral portion Y (for example, the shaded portion in FIG. 11) is included in the unauthorized area X as shown in FIG. The unauthorized area X is an area other than the object area R located inside the object area R (shaded portion in FIG. 11). Further, the inner peripheral portion Y is a portion surrounded by the object region R.
When the object region R generated in the region generation step S2 includes the illegal region X, the region correction unit 7 performs the region correction step S3 (correction processing) before performing the region correction step S3 (correction processing). Alternatively, the following correction process is performed instead of the above-described area correction step S3.
The area correction means 7 performs an enlargement process for integrating the unauthorized area X into the target body area. That is, the region correction means 7 performs an enlargement process for integrating the inner peripheral portion Y (region having a preset dimension) into the target region R. The enlargement process is repeatedly performed until all of the unauthorized area X is integrated into the object area R. Thereby, the illegal area X can be eliminated.
In addition, the above-mentioned fraudulent area | region can generate | occur | produce by the recessed part or convex part of the target object 1 with an unevenness | corrugation on the surface. The concave portion or the convex portion may be generated as an illegal region in the region generation step S2.

1 object, 3 shape data acquisition means, 5 area generation means,
7 area correction means, 7a reduction processing means, 7b area number assignment means,
7c Enlargement processing means, 10 object recognition device

Claims (6)

An object recognition method for expressing a shape of an object in a virtual coordinate space,
By measuring the distance from the measurement point to each measured point on the target object or by capturing the target object, the shape data acquisition step for acquiring the target object shape data;
Based on the shape data, an area generation step for generating an object area occupied by the object in a virtual coordinate space using a triangular mesh ;
An area correction step for performing correction processing on the generated object area,
The correction process includes a reduction process ,
The outer peripheral part or inner peripheral part of the target body area is the processing target area,
(A) The reduction process is a process of deleting a processing target area that is a part of the target object area from the target object area,
In the reduction process,
(A-1) Select one vertex i of the triangular mesh in the object region,
(A-2) Identify each vertex j that is adjacent to the vertex i and belongs to the same object area as the vertex i,
(A-3) For each vertex j, measure the number of triangular meshes to which the vertex j and vertex i belong,
(A-1), (A-2), and (A-3) are performed for all vertices i in the object region, and all vertices whose number measured in (A-3) is one A target object recognition method , wherein i is not included in the target object area, and all the vertices i are vertices outside the target object area . The processing target area is the outer peripheral part of the target object area,
When the target region is divided into a plurality of target regions as a result of performing the reduction process in the region correction step, the target regions are separated from each other in the virtual coordinate space. The object recognition method according to claim 1, wherein the object recognition method is set as a region.   3. The object according to claim 2, wherein if the object region disappears as a result of performing the reduction process in the region correction step, the object region is deleted in the virtual coordinate space. 4. Recognition method. In the region correction step, after performing the reduction process, it has rows enlargement processing,
4. The object according to claim 2 , wherein the enlargement process is a process that integrates a processing target area that is an area other than the target object area and is adjacent to the target object area into the target object area . 5. Recognition method. Inside the target area, other areas other than the target area are included, and the processing target area is the inner peripheral portion that is included in the other area and surrounded by the target area,
The correction process includes an enlargement process for integrating the inner peripheral portion into a target body region ,
2. The object recognition method according to claim 1, wherein the enlargement process is a process of integrating a process target area that is an area other than the target object area and is adjacent to the target object area into the target object area. . An object recognition apparatus for expressing a shape of an object in a virtual coordinate space,
By measuring the distance from the measurement point to each measured point on the object, or by capturing the object, shape data acquisition means for acquiring the object shape data;
Based on the shape data, an area generating means for generating an object area occupied by the object in a virtual coordinate space using a triangular mesh ;
Area correction means for performing correction processing on the generated object area,
The correction process includes a reduction process ,
The outer peripheral part or inner peripheral part of the target body area is the processing target area,
(A) The reduction process is a process of deleting a processing target area that is a part of the target object area from the target object area,
In the reduction process,
(A-1) Select one vertex i of the triangular mesh in the object region,
(A-2) Identify each vertex j that is adjacent to the vertex i and belongs to the same object area as the vertex i,
(A-3) For each vertex j, measure the number of triangular meshes to which the vertex j and vertex i belong,
(A-1), (A-2), and (A-3) are performed for all vertices i in the object region, and all vertices whose number measured in (A-3) is one A target object recognition apparatus , wherein i is not included in a target object area, and all the vertices i are vertices outside the target object area .