JPS60191373A - Recognizer for three-dimensional object - Google Patents

Abstract

PURPOSE:To recognize the form, etc. of a three-dimensional object in a short time by projecting the grid-shaped slit light to the surfase of an object to be recognized to form a slit image and calculating a two-dimensioal spase frequency spectrum from said slit image to calculate the similarity to a standard spectrum. CONSTITUTION:An object 60 to be recognized is put on a belt conveyor 10 and moved right or left. The grid-shaped slit light is projected to the object 60 from a projector 50, and the object 60 is picked up by an ITV camera 70. A picture data processing prt 80 digitizes the picked-up analog picture signals to store them to a built-in picture memory and also delivers the picture data D1 on the area of the object 60 to a recognizing part 100. The part 100 calculates a two- dimensional space frequency spectrum and then calculates the similarity between the picture data obtained by projecting the grid-shaped slit light to plural standard objects of standard forms designated previously and the storage value of a standard spectrum. Then it is decided that the object 60 has the form of a standard object having a standard spectrum which gives the highest similarity. Thus the form of a three-dimensional object can be recognized in a short time.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a three-dimensional object recognition device, and particularly to a three-dimensional object recognition device suitable for handling three-dimensional objects with a manipulator, recognizing the external world of a self-propelled robot, etc. .

[Background of the invention]

Conventionally, when handling an arbitrarily placed three-dimensional object using a manipulator, etc., a method of obtaining a distance image using a single slit light is generally used as a method of recognizing the target three-dimensional object. (Refer to Proceedings of the United Nations Conference, 1971, p. 3245). FIG. 1 shows an example of the configuration of a three-dimensional object recognition device using this method. A slit light 40 emitted from a slit light projector 20 forms an image 45 on the recognition target object 10, and a television camera 30 captures this image 45, takes it in as an electrical signal, and performs image processing by a device not shown. In order to recognize the three-dimensional shape of the recognition target object 10 in this manner, it is necessary to move the slit light 40 at appropriate intervals to scan the recognition target object 10. FIG. 2 shows an example of an image of scanning slit light obtained by the television camera 30 of FIG. 1 using the method described above. In the case of this method, if the scanning speed of the slit light 40 is slow, the entire image cannot be obtained as a single image unless the logical sum of the images captured by the television camera 30 is calculated at regular time intervals, so extra calculations are required. shall be. Furthermore, if the scanning speed is set sufficiently high, it is possible to obtain a distance image of the recognition target on the screen as shown in Figure 2 in 1/30 or 1/25 seconds, which is the scanning period of one screen of a television camera. However, a problem arises in that a special mechanism is required to scan the slit light at high speed. Furthermore, in order to obtain three-dimensional information of the recognition target from a distance image as shown in Fig. 2, a surface is detected from a discontinuous point of the gradient of a line segment on the distance image, and the direction of the surface is determined. [Objective of the Invention] An object of the present invention is to eliminate the need for a mechanism for scanning slit light as in the conventional art, and to scan the shape of a three-dimensional object in one plane using a simple algorithm. An object of the present invention is to provide a three-dimensional object recognition device that can recognize the direction, surface texture, etc. of objects in a short time.

[Summary of the invention]

In order to achieve the above object, the present invention uses lattice-like slit light and examines the pattern of the two-dimensional spatial frequency spectrum of a lattice-like pattern image obtained by projecting the slit light onto a recognition target. This system recognizes the directions, surface properties, etc. of nine planes of a three-dimensional object.

[Embodiments of the invention]

The present invention will be explained in detail below using examples.

FIG. 3 is an overall configuration diagram of an embodiment of the device of the present invention for recognizing a three-dimensional object flowing on a belt conveyor, which includes a grid-shaped slit light projector 50, an ITV camera 70, an image data processing unit 80, and a controller. section 90, recognition section 100 and display section 1
Consists of 10.

The recognition target object 60 is placed on the belt conveyor 10 and moves to the right or left. In order to recognize the direction of the two planes of the object to be recognized, a lattice-shaped slit light projector 5, which will be described later, is used.
0, a lattice-shaped slit light (hereinafter referred to as slit light for simplicity) is projected onto the recognition target object 60, and the recognition target object is imaged by the ITV camera 70. The ITV camera 70 differs depending on the type of slit light source. For example, when using a visible light source such as a halogen lamp, it is usually
Using a TV camera (hereinafter referred to as ITV camera)
If an infrared light emitting element is used, an infrared television camera is used. The ITV camera 70 is provided with an optical system control unit 75 for adjusting the zoom ratio, focal length, etc. in order to be able to respond to changes in the geometric size and focal length of the recognition target object 60. There is. Image data processing section 8
0 is an analog image signal V captured by the ITV camera 70
(t) is digitized and stored in the built-in image memory, while outputting image data D1 of the designated recognition target area among the stored image data to the recognition unit 100. Also,
If necessary, the command data C1 may be output to the control section 90 to control the optical system control section 75 or the lattice-shaped slit light projector 50.

The control unit 90 controls the image data processing unit 80 or the recognition unit 1.
The optical system controller 75 or the lattice-shaped slit light projector 50 is controlled by inputting the command data C2 from 00. The recognition unit 100 sends command data C3 to the image data processing unit 80, receives the image data, calculates a two-dimensional spatial frequency spectrum, and uses the spectrum pattern to determine nine three-dimensional shapes of the object to be recognized as described later. Recognizes the direction of the surface and the properties of the surface. Further, the 0 display unit 110 may output command data C2 to the control unit 90 as necessary.
The area to be subjected to the two-dimensional spatial frequency calculation performed by the recognition unit 100 is displayed, the recognition results are displayed, and the like.

The above is an outline of the embodiment shown in FIG. 3, and the details of each part will be described next. FIG. 4 shows a grid-like slit light projector 5.
0 shows an example of the grid-like slit plate 510.
A light source 52 is placed in front, and the light emitted through a grid-like slit plate 51 is projected by a lens 53. The lattice-like slit plate 51 can be switched to one with a different lattice spacing by a motor 54 depending on the situation.

Further, the focus of the projected slit image can be automatically adjusted by the servo motor 55. Note that in this method, a transmissive liquid crystal can be used as the lattice-shaped slits, and in this case, the lattice spacing is electrically cut and the light from the 52A and 52B is projected as a parallel slit beam by the cylindrical lenses 56A and 56B. The parallel slit beams of these two parallel slit light projectors 50A and 50B are installed on a support stand 57 so that the slit directions differ by 90° from each other, and both generate lattice-shaped slit lights. Whichever of these is used, there is no need for a mechanism for scanning the slit light as in the conventional example, and the ITV camera 70 can capture the entire image by scanning once.

FIG. 6 shows a specific embodiment of the image data processing section 80 shown in FIG. 3, and is composed of an image signal digitizer 81, an image data memory 82, and a control section 83. The image signal digitizer 81 digitizes the image signal V (t) (analog signal) from the ITV 70 and stores it in the image data memory 82 . The image data memory 82 stores the image data output from the image signal digitizer 81 in the memory, and outputs a specified number of image data DI from a specified address to the recognition unit according to a data output command from the control device 83. Output to 100. The control device 83 decodes the command data C3 sent from the recognition unit 100, and controls the digitizing start operation of the image signal digitizer 81, the data read start address from the image data memory 82, etc. In addition, the signal level of the image signal V(t) is checked from the digitized data of the image signal V(t), and in order to obtain appropriate image data, Projector 5
It is also possible to issue command data C1 for controlling 0. 7th
The figure shows the relationship between the command data C3 output from the recognition unit 100 and the image memory space when specifying a data area in the image data memory for which the two-dimensional spatial frequency is to be calculated. It is. In the same figure,
X, , Y indicate readout start addresses in the X and Y directions on the image, and N, , M indicate the size of the two-dimensional spatial frequency calculation area A in the X and Y directions in the X and Y directions.

FIG. 8 shows one embodiment of the recognition unit 100. Therefore, the spatial frequency calculation device 101 and the pattern recognition device 10
These operations are the most characteristic part of the present invention. So, to explain this recognition part in detail,
First, a general Fourier transform of a time-varying signal is used to analyze the frequency components of the signal. Similarly, the Fourier transform with spatial frequency as a parameter is
It handles signals related to spatial one-dimensional or two-dimensional positions and is used to analyze the periodicity of spatial signals. The video signal captured by a television camera is a signal that changes over time, and at the same time, it is also a signal that accurately correlates time and position (position on the screen).
The signal visualized on a CRT (CRT, etc.) is a spatial signal. Therefore, when a video signal is viewed as a one-dimensional signal related to time, the frequency that appears in its Fourier transform represents the number of repetitions of the signal per unit time (cycle 7 seconds 2, etc.), but it changes spatially. When viewed as a signal, the spatial frequency is the number of signal repetitions per unit length (cycles/c
m, etc.). This space (10) In this case, since it is two-dimensional, let μ and ν be the frequencies,
Fourier transform G(μ, ν) with these parameters as ′! 1′,! l) The spatial frequency spectrum is given by the following discrete Fourier transform.

x exp (-j 2πmν/M) - (1) However, g (n + m) is imaged with an ITV camera at 7°,
Discrete image data digitized by the image data processing unit 80, which includes images at a total of NXM positions, N in the X direction and M in the Y direction, as explained in FIG. This is shading data.

And they are n = O~N-1, m = 0-M
This is expressed as g(n+m) using the parameter −1 as a variable. Next, the meaning of projecting the lattice-shaped slit light onto a three-dimensional object and obtaining the two-dimensional spatial frequency spectrum of the resulting image is as follows. In other words, the two-dimensional spatial frequency spectrum (11) of the grayscale image data obtained by projecting grid-like slits onto the target object with constant slit spacing in the vertical and horizontal directions. Significant peaks are obtained at spatial frequency points corresponding to the period of the slit image in the vertical and horizontal directions. The gradation image data of the lattice pattern obtained by projection of the lattice slit light is basically a black and white binary pattern, and the repetition period of this black and white is extracted as a spatial frequency. For example, for the sake of simplicity, consider a one-dimensional spatial frequency spectrum, and as shown in Figure 9, the repetition of the white part with amplitude a and the black part with zero amplitude obtained by the slit light at every interval of length L becomes the slit image. Given, and assuming that the width of the white part caused by the slit light is sufficiently narrower than the width of the black part, this white part can be expressed as a Dirac δ function, and the lattice image data g (n ) is g (n) − Σ aδ (n0 pL) ・
・(21p=-flash is expressed, and the result of Fourier transformation of this is S(μ),
As shown in FIG. 10, a spectrum appears at a position that is an integral multiple of the spatial frequency μm17'r,=r. In reality, the white part at the top of the image (12) obtained by the slit light has a certain width, and the spatial frequency spectrum differs from that in Fig. 10, with the amplitude of the fundamental frequency r being the largest, and the amplitude of the second, The amplitude of the spectrum of the third and subsequent high frequency components exhibits a characteristic of rapidly decreasing. Furthermore, even if the interval between the projected slit lights is constant, if the angle of the projection surface of the target object with respect to the slit light changes, the interval between the white parts of the image obtained by the slit lights will also change accordingly. The inclination of each part of the target object can be determined by the value of the spatial frequency (fundamental frequency) μ that appears.

This also holds true for the two-dimensional spatial frequency spectrum as follows, and the present invention utilizes this property.

FIG. 11A shows a state in which the surface of the object to be recognized 60 is oriented perpendicular to the projection direction of the grid-like slit light 50. If the vertical direction of the figure is taken as the X-axis in the direction perpendicular to the drawing 71, then Assume that the lattice formed by the slit light on the surface of the object 60 has equal intervals of ε0 in both the x and X directions as shown in FIG. When the plane figure in Figure 11B is Fourier-transformed using equation (1) and the spatial frequency spectrum is obtained, the 9th
゜In the same manner as explained in Fig. 10, μ-μo-1/ε0. as shown in Fig. 11c. si=νo=1/ε.

A spectrum S+-8 (μ0.shi0) with a strong fundamental frequency appears at the spatial frequency (μ0.shi0). Next, in the state shown in Fig. 11A, the surface of the object 6o is slightly rotated around a line in the X direction on that surface, as shown in Fig. 12A.
At this time, the slit interval in the X direction is ε, the same as in Fig. 11B.
It is 0. However, due to the above rotation, as shown in Figure 12B,
The axial distance becomes small as ε1 < ε0, so the first
As shown in Figure 2C, the fundamental frequency spectrum S1 is (μl
, νo) + μm=1/ε! 〉Appears at the point μ0. Furthermore, when rotating from the state of FIG. 11A to FIG. 13A around one line in the X direction on the object 600 plane, similarly, ε2<ε0 becomes as shown in FIG. As shown in the figure (μ0.shi1) + ν1=
Appears in 1/ε2〉ν0. In this way, in the case of a flat surface, a strong spectrum appears only at one point in the spatial frequency (11A), and the inclination of the surface appears as a change in its position.

Above, we have explained the relationship between the obtained grid image and the spatial frequency spectrum when the object to be recognized is a plane. However, it is clear that this spatial frequency spectrum also shows the spectral distribution specific to a three-dimensional object. The case is shown below. Figure 14A shows a cylinder 6 as a recognition target object.
11A shows a situation in which lattice-shaped slit light similar to that shown in FIG. 11A is projected onto 0. However, the cylinder 60 is a side view.

In the case of this cylinder, as shown in FIG. 14B, the lattice spacing in the X direction is constant, but the parallel slit optical lattice spacing in the X direction gradually narrows on the left side of the cylinder and widens on the right side. However, in this case, the left half of the cylindrical surface captured by the camera 70 is cut out and processed.

Therefore, when the two-dimensional spatial frequency spectrum of this entire image is calculated, a spectrum as shown in FIG. 14C is produced. In the case of a block or cylinder, the spatial frequency ν in the X direction is constant, and many spectra appear in the X direction. Next, the 15th A
The figure shows a case where the object to be recognized is a (15) sphere 60, and the image produced by the slit light at this time is as shown in Fig. 15B, where the parallel slit light grating spacing in the X direction is a constant interval as in the case of a cylinder. However, since the slit light in the X direction is distorted by the curvature of the spherical surface, it forms an arc convex to the left as shown in the figure, and the interval between the arcs is narrow on the left and wide on the right. The two-dimensional spatial frequency spectrum of the entire image is as shown in FIG. 15C. In other words, the characteristics of the two-dimensional spatial frequency spectrum in the case of this sphere are
Spectrum 81.82 of three points on the low frequency side of the spatial frequency ν in the X direction that occurs due to distortion of the slit light in the direction.
It is S3. As mentioned above, due to the difference in the pattern of the two-dimensional spatial frequency spectrum of images obtained by projecting lattice slit light onto a three-dimensional object, the shapes of three-dimensional objects such as nine planes, cylinders, and spheres and planes are Three-dimensional direction recognition is possible.

The present invention utilizes the properties of the two-dimensional spatial frequency spectrum as described above, and returning to the embodiment shown in FIG. 3, the pattern shown in FIG. The recognition device 102 is an IT device in FIG.
First, the entire screen of image data captured by the V camera 70 and captured by the image data processing section 80 is displayed on the display device 110. Assume that this display screen is as shown in FIG. 16A, for example. Next, a two-dimensional spatial frequency calculation area A is determined from this screen as shown in FIG. 16B, and the image data processing section 8
Command data C3 indicating this area A is sent to 0. Then, as explained in FIGS. 6 and 7, the data processing unit 80 outputs the grayscale data g(n+m)+n=0 to N− of the calculation target area A.
1. m=o to M-1 are sent as data DI to the spatial frequency calculation device 101 in FIG. v) is sent to the pattern recognition device 102. The device 102 uses this input spectrum G(μ
, ν) to determine the shape of the object in area A on the screen.
Z2+... (For example, Zt = sphere, Z2 = cylinder 9, etc.)
The spatial frequency spectrum GI (μ, ν) of (1
7) Store it in the local memory. Then, the similarity Q1 between the input two-dimensional spatial frequency G (μ, ν) of the target object and the above-mentioned stored two-dimensional spatial frequency spectrum G1 is calculated using equation (3).
The shape of the object Z+ that gives the highest degree of similarity is determined to be the shape of the target object.

However, when 11GI (K, t> +1=V) Wanzi Shoulder Stone C7he t is used, and when spectral patterns are compared using the similarity QI described above, the imaging conditions of the recognition target object and the imaging conditions of the standard object are It is necessary to consider the difference in G.
and the spatial frequency at which the largest peak of GI appears (
Either one, for example G, so that the values of μ, ν) match
μ of (μ, ν).

After linearly transforming both ν, the calculation of equation (3) is performed.

〔Effect of the invention〕

(18) As described above, according to the present invention, the two-dimensional spatial frequency spectrum of the image data obtained by projecting the grid-like slit light onto the recognition target object and the representative three-dimensional object (plane 9 cylinder 2
By comparing the standard two-dimensional spatial frequency spectrum of a sphere, etc., it is possible to recognize the shape of a three-dimensional object with simpler processing than conventional methods. There is an effect that properties such as unevenness in nine directions can also be recognized.

[Brief explanation of the drawing]

1 and 2 are diagrams showing an embodiment of a conventional object light projector using slit light, FIG. 6 is a diagram showing an embodiment of an image data processing section, and FIG. 7 is a diagram showing a recognition section in FIG. 6. Figure 8 is a diagram showing an example of the recognition unit, and Figures 9 and 10 are diagrams explaining the relationship between spatial frequency calculation area designation data and image areas. Figures 11A to 13A are diagrams illustrating the relationship between pickpockets (19); Figures 11B to 13B are diagrams showing the states when the plane of the target object is arranged at various angles to the light beam; Figures 11B to 13B are gt
Explanatory diagram of images corresponding to Figures IA to 13A, 11C
Figure ~ Figure 13C is part 2 of the image of Figure 11B ~ Figure 13B.
Diagrams showing dimensional frequency spectra, Figures 14A to 14C
The figure is a diagram showing a layout, an image, and a spectrum when a cylinder is the target object, and Figures 15A to 15C are diagrams showing a layout, an image, and a spectrum when a sphere is a target object.
FIGS. 16A and 16B are diagrams showing the process of displaying a recognition target object in the recognition unit and specifying a two-dimensional spatial frequency calculation area. 50... Grid-shaped slit light projector, 51... Grid-shaped slit plate, 52... Light source, 53... Lens, 54
... Slit plate rotation motor, 55... Lens drive motor, 56... Cylindrical lens, 57... Support stand, 60... Recognition target object, 70... ITV camera, 80... - Image data processing section, 81... Image signal digitizer, 82... Image data memory, 83...
Control device, 90... Control unit, 100... Recognition unit, 1
01... Spatial frequency (20) arithmetic device, 102... Pattern recognition device, 110...
・Display section. Agent Patent Attorney Masami Oni Akimoto 114-Figure 6 also '1 Figure □X X,. Ki 1 diagram also 15B diagram 15c diagram

Claims (1)

[Claims] 1. Projection means for projecting lattice-shaped slit light onto the surface of an object to be recognized, and a slit image formed on the surface of the object to be recognized by the projection by the means, and the image is stored in a data memory as image data. input means for inputting image data within a specified region of the slit image to calculate a two-dimensional spatial frequency spectrum; standard spectrum storage means for storing a two-dimensional spatial frequency spectrum of image data obtained by projecting the grid-like slit 2 as a standard spectrum;
pattern recognition means for calculating the degree of similarity between the dimensional spatial frequency spectrum and each of the standard spectra and determining that the object to be recognized has the shape of a standard object having a standard spectrum that gives the highest degree of similarity; Characteristic 3D object recognition device. 2. In the three-dimensional object recognition device according to claim 1, the lattice-shaped slit light projection means is a projector, each of which outputs parallel slit lights, and the parallel slit lights are orthogonal to each other. A three-dimensional object recognition device characterized by being configured by combining two devices. 3. In the three-dimensional object recognition device according to claim 1 or 2, a transmissive liquid crystal is used for the slits for forming the lattice-shaped slit light, and the slit interval can be changed electrically. 3 characterized by having a structure
Dimensional object recognition device.