JPS6361108A - Apparatus for confirming three-dimensional object - Google Patents

Abstract

PURPOSE:To obtain the titled apparatus capable of confirming and inspecting a three-dimensional object in a real time in a production line, by confirming the entire shape of the three-dimensional object from the unevenness relation of the object obtained by projecting a pattern to the object and the surface relation of the object preliminarily given without directly extracting all of the surfaces of the object. CONSTITUTION:The indication of a shape, for example, the indication of a rectangular parallelopiped is sent to a surface relation processing apparatus 8 from a console 4. The surface relation of each shape is stored in a block 42 and the information of the property of the rectangular parallelopiped of the block 43 is referred to by indication. Light is projected to an object 9 to be measured by a TV camera 1 and two projectors 2 obliquely mounted to said camera 1 and the part impinged against the surface of the object 9 of the pattern thereof is extracted by an unevenness position detector 7. From the extracted part, the coordinates of the apex calculated in an image processor 3 are inputted to the apparatus 8. Then, the number of apices are checked and a deficient apex is calculated. By this method, high speed three-dimensional confirmation can be confirmed without requiring the confirmation of all surfaces.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a three-dimensional object recognition device, and in particular, a three-dimensional object recognition device suitable for placement in a production line etc. where three-dimensional shapes need to be recognized at high speed. Regarding a ratio device.

[Conventional technology]

Conventionally, many studies have been conducted on methods for recognizing three-dimensional objects, such as binocular stereoscopic vision and slit light full-surface operation methods. "Measurement", Norio Aoki and 3 others, Institute of Electronics and Communication Engineers Technical Research Report, VoQ83°Nα309. (1984)
In this method, the three-dimensional position of the object is determined by applying a plurality of slit patterns to the object and viewing it stereoscopically with two eyes using two cameras.

[Problem that the invention seeks to solve]

However, according to this method, it is necessary to analyze multiple pattern shapes captured by two cameras, which increases the amount of processing per image.
Recognition at several hundred m5ec) is difficult. Furthermore, since the properties of the shape of the target three-dimensional object are not used as knowledge, there is a high possibility that the object recognition will be incorrect.

An object of the present invention is to provide a three-dimensional object recognition device that can recognize and inspect three-dimensional objects in real time on a production line or the like.

[Means for solving problems]

To recognize a three-dimensional object, we only need to know all the equations for each surface that makes up the object. However, for example, in the case of a rectangular parallelepiped or an object made of a combination of rectangular parallelepipeds, it is known in advance that adjacent surfaces form right angles, so it is not necessarily necessary to find equations for all the surfaces. For example, in the case of a rectangular parallelepiped, if the directions of two adjacent planes and the coordinates of the vertices of those planes are known, the coordinates of the remaining vertices and the directions of the planes can be estimated and found.

Therefore, in the present invention, a pattern is projected onto an object, and the overall shape of a three-dimensional object is calculated from the unevenness relationship of the pattern and the relationship between object surfaces given in advance, without directly extracting all the surfaces of the object. It performs high-speed recognition and displays the results as a developed diagram.

[Effect]

For example, when recognizing a rectangular parallelepiped, first the direction of one reference plane and the spatial position of the vertex of that plane are determined, and then the planes that are considered to be adjacent to that plane are found. The remaining surfaces are calculated from the surface relationships of the rectangular parallelepiped, and the resulting three-dimensional object is displayed in the form of a developed view on three surfaces 1, for example, a plan view, a front view, and a side view.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a configuration diagram of this embodiment, including a TV camera 1.

It consists of two slit light projectors 2, an image processing device 3, a console 4, a display 5, a monitor 6, an unevenness detection device 7, a one-surface relationship processing device 8, and an object 9. The object 9 will be described below using a rectangular parallelepiped as an example.

The two projectors 2 are attached obliquely to the TV camera 1, and in combination with the TV camera 1 constitute a spatial position measuring device that detects a spatial position. The principle of the spatial position measuring device used in this embodiment will be explained using FIG. As shown in FIG. 2, when the pattern emitted from the projector 2 hits the surface A, the image of the hit point P on the pattern is formed at a position yI on the imaging surface of the TV camera 1. Furthermore, when the surface A' is hit, the image of the point P2 on the pattern is formed at the position y2. in this way.

The position on the TV camera 1 changes depending on where on the projection axis of the projector 2 the pattern hits the surface. As a result, T
The distance 0 from the V camera to the surface can be found. Since the projector 2 is attached obliquely to the TV camera 1, when the pattern hits an object, it forms an oblique band. When the diagonal line is viewed with the TV camera 1, the farther the diagonal plane is from the TV camera 1, the higher it rises on the TV screen. In addition, the y direction (TV scanning direction) is
The larger y is, the further to the right the image appears on the TV screen. From this relationship, the coordinate values (I, J
), the spatial position @(y,z) corresponding to that point is determined. Furthermore, since the projector 2 is fixed relative to the TV camera 1, the X coordinate can be determined from the Y coordinate by using the inclination of the pattern. Therefore, from the coordinates (I, J) in the TV screen of the pattern to the spatial position (xt y
Iz) can be obtained. Using the above principles,
The spatial position of any point on the pattern from the left and right projectors 2, 2 can be determined. The procedure for recognizing each face of a rectangular parallelepiped will be described in detail below.

FIG. 3 is a diagram showing a pattern on the object 9 when a pattern is projected onto the object 9 by the left and right projectors 2, 2, and the pattern is bent at the boundary between each surface. Figure 4 shows the pattern taken on the TV screen, R1*
Rat R3 is the bending point of the right projector pattern v Llt L2e L3 is the bending point of the left projector pattern.

In FIG. 5, a procedure for detecting these bending points using the unevenness detection device 7 is shown.

Block 11 is a pattern photographed with a TV camera, and block 12 is a screen photographing only the object when the projector is turned off. The difference between these two screens.

That is, when the brightness of each pixel is subtracted one by one, only the uneven pattern shown in block 13 is extracted. Therefore, as shown in block 14, a straight line is applied to the extracted pattern, and points that deviate from the same straight line are extracted as bending points. In this way, R
,,R2,R3,L,.

L2+L3 screen coordinates (I, J) and spatial coordinates (x+y+
z) is determined from the measurement principle described above.

Next, extract the reference plane SI and its vertices PI+P2+P3
+Calculate the spatial position of P4. This calculation procedure is shown in FIG. In FIG. 6, a block 21 detects a line segment RIR2 and L2 where the left and right patterns intersect. Line segments R, R2, and Lz are clearly on the same plane, so in order to extract that plane, 'I3 light image image processing is performed. Images taken with a TV camera are usually 25
Divided into 6 x 256 pixels.

Brightness is assigned to each pixel in steps from 0 to 255. In grayscale image processing, areas where the brightness changes significantly are extracted as boundaries of regions. As shown in block 21, the brightness of some pixels on the line segments R, R2, and LZ is sampled, and in block 22, the average value f and standard deviation σ are calculated. Since the brightness f of each pixel can be expected to be approximately constant on the same plane, If fol<α·σ (1) may be used as the plane extraction operator. Here, α is a parameter and is on the order of 1. Extract the area that satisfies equation (1).

If the contour is represented by a series of points, it will look like block 23.

Here, the number of point sequences and area of the surface extracted in block 24 are calculated and checked, and if the values are acceptable, the process moves to block 25. For example, if the number of point sequences is 3, or the area is 1 pixel or too many pixels, or other unacceptable values, the parameter α is slightly changed in block 26 so that α±Δα(Δα(α
) and repeat the extraction in block 23. block 25
Now, four straight lines are extracted from the extracted vertex sequence, and their intersection points P I t P 2 + P 3 + P4 are determined. Finally, the spatial positions of vertices Pl, PZ+P3+R4 are calculated. Since the screen location e3 (I, J) of PI + P2 + P3 + R4 is known, the screen coordinates (I, J
) and the spatial position (Xy ywZ). By the way, point R1 on the pattern
+ In Rat Ll t L2, since the screen coordinates and the spatial position are both known, the linear relationship x=axI+b,, J+cx (2) y=a
yI ten by J + cy (3)z = azI +
Assuming bzJ + Cz (4),
4 points RI, R21LI+ L2 (I.

From the relationship between J) and (x+y+z), the coefficient axl a
y+* aZ+ 1)Xy bY+ bZ+ cxl
C>'y cz can be found. Therefore, the vertex P
The spatial coordinates of I+ P2+ Pay R4 can be calculated using these coefficients.

Next, consider extraction of the adjacent surface S2 to the reference surface SI.

As shown in block 31 of FIG. 7, the brightness of the pixels on the adjacent line segment R2R3 of pattern line segment R on S, R2 is sampled, and using the same process as that in FIG. 6, block 32 Extract the contour of S2 and select the vertex Ps+ P6+
Calculate the screen coordinates of P7+R8.

Next, in block 33, the apex PI+PZtP of the reference plane S1
Check whether there is a vertex that is the same as 3+P4, and if there is a vertex that is the same as 3+P4, change the vertex numbers. Block 34 shows the result, R3 and P 8 tP2 and R7
are identified as the same vertex. Finally, apex P5+P6
We want to find the spatial position of 4 points R2+ R3+
Since the screen coordinates and spatial coordinates of Pay R3 are known, the coefficients can be determined using equations (2) to (4) regarding surface S2. Therefore, the spatial position can be calculated from the screen coordinates of R5°R6. Since the spatial positions of the vertices of the two faces of 5ltS2 are known, in the case of a rectangular parallelepiped, the coordinates of the vertices of the remaining faces can be found from the face relationships.

The processing carried out in the one-plane relationship processing device 8 will be described in detail below.6 As shown in FIG. The surface relationship processing device of block 42 stores surface relationships for each shape, such as a cube, a rectangular parallelepiped, a cylinder, and a cone. When a rectangular parallelepiped is selected, the surface relationship of the rectangular parallelepiped stored in block 43 is referred to. The block 43 contains six pieces of information regarding the properties of the rectangular parallelepiped, such as the number of vertices being eight, the number of nine faces being six, and three orthogonal straight lines coming from each vertex. On the other hand, in block 44, the spatial positions of the vertices P1 to P6 found within the image processing device 3 as described above are input to the surface relationship processing device 8. Therefore, first, it is checked whether the number of vertices is 8.

Since there are only 6 vertices, the remaining 2 vertices are calculated, but the procedure is based on the 6 vertices.

Find a vertex with only two line segments, draw a straight line perpendicular to both of them, and find the intersection of these straight lines to find the remaining vertices P7+P8. The above processing is performed in block 45, but the flow is repeated in block 45.
As shown in the figure.

When the spatial coordinates of the eight vertices of the rectangular parallelepiped are determined in the above manner, the information is sent to the image processing device 3 (Fig. 1), and the rectangular parallelepiped is displayed on the display 5 as a plan view, front view, and side view. do. The display method will be explained in detail in FIG. In block 51, vertex PI+ P2+ P3+ P
Surface made from 4 5l-P2+P3 r P5 + P
Surface S2 made from 6, P I * P 2 t P
6 Each normal vector nl of surface S3 created from rP7
Find +n2+n3. In block 52, the spatial position of the vertex P I +P2+P3+P4 on Sl is transformed to the position seen from the n1 direction. In this way, since the surface S can be regarded as a planar image, it can be displayed on the display 5 as a plan view. Similarly, in block 53, the surface S2 is converted into a figure viewed from the n2 direction to form a front view, and in block 54, the surface S3 is converted to a figure viewed from the n3 direction to form a side view. Finally, in block 55, each surface may be displayed on the display 5. At this time, the scale of the spatial position and the scale of the screen coordinates are suitably adjusted to display the screen in a convenient size.

As described above, according to this embodiment, a rectangular parallelepiped can be recognized by extracting only two surfaces by cutting out the left and right patterns and utilizing the surface relationship.

〔Effect of the invention〕

According to the present invention, a three-dimensional shape can be recognized by using surface relationship information, resulting in the following effects.

(1) Since it is not necessary to recognize the shape of all surfaces of a solid, and three-dimensional recognition can be performed at high speed, real-time recognition on a production line or the like is possible.

(2) By adding new face-related information.

It is highly versatile as it can recognize various three-dimensional shapes.

[Brief explanation of drawings]

Figure 1 is an overall configuration diagram of a three-dimensional object recognition device according to an embodiment of the present invention, Figure 2 is a diagram of the principle of spatial position measurement, Figure 3 is a conceptual diagram of a pattern projected onto an object, and Figure 4 is a diagram of the principle of spatial position measurement. FIG. 3 is a TV screen diagram of the pattern shown in FIG. 5, FIG. 5 is a flowchart of pattern unevenness processing, FIG. FIG. 8 is a flowchart showing a surface relationship processing method, FIG. 9 is a detailed flowchart of block 45 shown in FIG. 8, and FIG. 1O is a flowchart showing a procedure for displaying object recognition results. 1...TV left camera 2...Slit light projector, 3...Image processing device, 4...Console, 5...
-Display, 6...Monitor, 7...Irregularity detection device, 8...Surface relationship processing device, 9...Object. Representative Patent Attorney Tadashi Akimoto Fig. 1 Fig. 2 Fig. 3 Fig. 4 Fig. 5 Fig. 6 Fig. 7 Fig. 8 Section 9 Fig. 10

Claims (1)

[Claims] 1. A spatial position measuring device that detects a spatial position using a planar image input device such as a TV camera, an image processing device that processes the planar image, and a projector that projects a linear pattern such as slit light. The object recognition device includes an unevenness detection device that extracts a portion that hits the object surface from the unevenness relationship of a pattern projected on the object, and a surface relationship processing device that provides the relationship between adjacent surfaces of the object. A three-dimensional object recognition device that recognizes the three-dimensional shape of objects without directly extracting their surfaces.