JP3447356B2 - Object position detection method and object position detection device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object position detecting method and an object position detecting method for detecting the position of an object.

[0002]

2. Description of the Related Art There is a strong need for an image recognition device such as a visual device for an industrial robot or a product inspection device for an automation line. Conventionally, there has been a strong demand for a recognition device using a correlator using an optical matched filter. Development has been done vigorously. Further, recently, development of a recognition device using a neural network has been actively conducted. However, all of these recognition devices have a low ability to recognize an image that has been deformed by shifting, rotating, scaling, or the like, that is, the generalization ability is low, and improvement of this generalization ability is not practical. Had been the biggest challenge of.

One answer to this problem is David Casasent of Carnegie Mellon University (David C
asasent et.al. "Real-time deformation invariant opt
icalpattern recognition using coordinate tranforma
tions ", Appl. Opt., Vol.26, pp.938-942 (1987)). The method of using their correlators is as follows. First, as shown in FIG. 101 is irradiated with coherent light 102, and then a computer-generated hologram (CGH) 10
When the Fourier transform lens L ₁ 104 performs the Fourier transform after superimposing the phase information for the coordinate transform by 3, the information obtained by performing the desired coordinate transform on the input image is obtained on the coordinate transform surface 105. In the case of the above-mentioned document, for example, the desired coordinate transformation is a logarithmic polar coordinate transformation in which even if transformations of scaling and rotation occur in the input image, those transformations are transformed into shift amounts.

Next, as shown in FIG. 15, the information after the logarithmic polar coordinate conversion is applied to a liquid crystal television (LCTV).
a double diffraction system consisting of two Fourier transform lenses L ₂ 107 and L ₃ 108,
Matched spatial filter (MSF) created by applying the same logarithmic polar coordinate transformation to the reference image
The coherent light 110 is incident on the shift invariant correlation optical system in which the filtering by 109 is combined, and is incident. The camera 111 detects the correlation peaks of the reference image and the test image on the correlation plane P, and recognizes them. Is to do. In fact, Casasent et al. In the above-referenced document show an example in which good matching results are obtained for rotation and scaling deformations using the above method.

Further, the US Air Force Institute of Tech
Kenneth H Fielding et al. of nology, although there is no detailed explanation, develop this and make the input image a Fourier spectrum (the product of the Fourier transform of the input image and its conjugate information) and then perform the same processing. In addition to rotation and enlargement / reduction, a good matching result was obtained with respect to shift deformation, and an actual example is shown (Kenneth H. Fie
lding et.al. "Position, scale, and rotation invari
ant holographic associative memory ", Opt.Eng., Vol.
28, pp.849-853 (1989)).

[0006]

Fielding et al.'S method, which is an extension of the method of Casasent et al., And has generalization ability for shift deformation, is based on the coordinates of the Fourier spectrum of the input image. It is used for recognition after conversion, but the information around the 0th-order light, which is strong in its spectral intensity, is relatively similar in any input image, and even if other parts are different, it is greatly affected by that part, The recognition error becomes large. In addition, due to the operation of taking the product of the Fourier transform of the input image and its conjugate information that was added when obtaining the Fourier spectrum, the information about the phase and the information about the negative amplitude are missing from the Fourier transform information of the input image. However, since recognition is not performed with complete information, a large recognition error occurs depending on the input image.

The present invention has been made in view of the above circumstances, and an object thereof is to provide an object position detecting method and apparatus capable of accurately recognizing an image which has been deformed by shifting, rotating, enlarging or reducing. It is to provide it.

[0008]

The object position detecting method of the present invention which achieves the above object reads a Fourier transform hologram of an input object image by using Fourier transform information of the input object image,
Further, the method is characterized in that the position of the input object image is detected by the convolution image of the input object image obtained by performing the inverse Fourier transform on the input object image.

Further, the object position detecting apparatus of the present invention comprises an image input means for taking an input object image into the system, and a Fourier transform for performing Fourier transform on the taken input object image to obtain Fourier transform information. Means, a reference light beam means for generating a reference light beam for producing a hologram with a desired carrier, and a Fourier transform hologram for recording a wavefront in which the Fourier transform information of the input object image and the reference light flux are recorded. Recording means for producing, reading means for reading the recorded Fourier transform hologram with Fourier transform information of an input object image, and inverse Fourier transform means for further inverse Fourier transforming the read information. , And a detection means for detecting convolution information of the input object image obtained by the inverse Fourier transform. It is an butterfly.

Another object position detecting device of the present invention is
An image input means for capturing an input object image in the system, a Fourier transform means for applying a Fourier transform to the captured input object image, and a squared Fourier transform information of the input object image by the Fourier transform means. , An inverse Fourier transform means for further inverse Fourier transforming the squared Fourier transform information into convolution information of the input object image, and a detection means for detecting this convolution information. It is characterized by that.

Another object position detecting method of the present invention is to read a Fourier transform hologram of an input object image with Fourier transform information of the input object image, and further perform inverse Fourier transform on the Fourier transform information to obtain an Fourier transform of the input object image. This is a method characterized by detecting the position of this input object image from an image on which the DC component of the hologram is applied.

Another object position detecting apparatus of the present invention is an image input means for taking an input object image into the system, and a Fourier transform for subjecting the taken input object image to Fourier transform to obtain Fourier transform information. Means, a reference light beam means for generating a reference light beam for producing a hologram with a desired carrier, and a Fourier transform hologram for recording a wavefront in which the Fourier transform information of the input object image and the reference light flux are recorded. Recording means for producing, reading means for reading the recorded Fourier transform hologram with Fourier transform information of an input object image, and inverse Fourier transform means for further inverse Fourier transforming the read information. , In order to detect the information in which the DC component of the Fourier transform hologram is applied to the input object image among the information obtained by the inverse Fourier transform It is characterized in that it has a detection means.

[0013]

[0014]

As shown in the block diagram of FIG. 1, the image processing apparatus according to the present invention described above has an image pickup means 1 for inputting input object information as an image and a position of an object inputted by the image pickup means 1. Object position detecting means 2 for detecting
And moving means 3 for moving the image pickup means 1 from the position of the object obtained by the object position detecting means 2 to move the object on the optical axis, and object information moved by the moving means 3 on the optical axis. It is intended to solve the above-mentioned problems and achieve the above-mentioned object by configuring with the recognition means 4 for inputting and recognizing.

According to the above configuration, the position (shift amount and direction from the optical axis) of the inputted object to be recognized is detected, and the image pickup means 3 is moved from this position information to recognize the above. The object information to be moved can be sent to the recognition means 4 after being moved on the optical axis. That is, since the same information can be sent to the recognition means 4 regardless of the position of the object on the input screen, it is possible to improve the generalization ability regarding shift deformation without missing information. . further,
By using the method of Casasent et al. Described above in this recognition means, it is possible to obtain the generalization ability with respect to the transformation of rotation and enlargement / reduction. It is possible to provide a new image processing device that can perform collation / recognition with high accuracy.

First, the object position detecting method used in the object position detecting means 2 for detecting the position (shift amount and direction from the optical axis) of the inputted object, which is the core of the image processing apparatus according to the present invention, explain in detail.

First, the origin is taken on the optical axis, the input image of the object to be recognized is g (x, y), and the image when the object to be recognized is on the optical axis is f (x, y). ), And considering that the object g (x, y) to be recognized is f (x, y) shifted by (a, b) from the origin, this g
(X, y) is written as g (x, y) = f (x-a, y-b) = f (x, y) * δ (x-a, y-b) (1) be able to. Here, “*” indicates convolution integration. Next, the Fourier transform information of this g (x, y) is G (u, v), and the reference light flux is R (u,
v), the Fourier transform hologram recording the interference of these two wavefronts has the information | G | ² + | R | ² + G ^* · R + G · R ^* ... (2) (however, “ ^* ”Is the complex conjugate and“ ・ ”
Indicates multiplication, respectively. Further, for simplicity, variables irrelevant to the description will be omitted hereinafter. ). Further, when the Fourier transform hologram represented by the equation (2) is read by the Fourier transform information G (u, v) of the input object, the read information is (| G | ² + | R | ² )・ G + G ^*・ G ・ R + G ・ G ・ R ^*・ ・ ・ (3) Further, when this information is subjected to an inverse Fourier transform, finally, information of F ⁻¹ {(| G | ² + | R | ² ) · G + G ^* · G · R + G · G · R ^* } (4) Will be obtained. In addition, the inverse Fourier transform was represented as "F ^-1 ".

Next, the expression (4) will be examined item by item. First, the first term is: F ⁻¹ {(| G | ² + | R | ² ) · G} = F ⁻¹ {(| G | ² + | R | ² ) * g = F ⁻¹ {(| G | ² + | R | ² ) * f * δ (x-a, y-b) · (5), which is the image f (x, y) when the object to be recognized is on the optical axis. A position where an image on which a direct current component (so-called zero-order component) of a Fourier transform hologram is applied appears in the direction in which the Fourier transform information G (u, v) at the time of recording the hologram, and which has object information on the optical axis. In contrast, it is shown that the positions are exactly apart from each other by the shift amount (a, b).

Next, the second term is a ^{F -1 {G * · G ·} R} = g * * g * r = f * * f * r ··· (6). The Fourier transform information of the reference light flux R (u, v) is represented as r (x, y). This is because the correlation image of the image f (x, y) when the object to be recognized is on the optical axis is the reference light flux R when recording the Fourier transform hologram.
It appears in the direction of (u, v), and indicates that it is at exactly the same position as when there is object information on the optical axis.

Next, the third term is: F ⁻¹ {G · G · R ^* } = g * g * r ^* = f * f * δ (x-2a, y-2b) * r ^* ... (7) This is because the convolution image of the image f (x, y) in the case where the object to be recognized is on the optical axis is a direction conjugate with the traveling direction of the reference light flux R (u, v) at the time of recording the Fourier transform hologram. , Which indicates that the position is apart from the position when there is object information on the optical axis by exactly twice the shift amount (a, b).

As shown by the equations (5) to (7), the above-mentioned Fourier transform hologram of the input object image is read out with the Fourier information of the input object image, and further, the object of the present invention is subjected to inverse Fourier transform. According to the position detection method,
If the information represented by the equation (5) is detected in the traveling direction of the Fourier transform information G (u, v) at the time of recording the hologram, the information is relative to the position when there is object information on the optical axis. Since the positions are apart by exactly the shift amount (a, b), if the position is detected, the position of the object to be recognized, that is,
The shift amount and direction from the case where the object to be recognized is at the origin (on the optical axis) can be detected.

Further, if the convolution information expressed by the equation (7) is detected in a direction conjugate with the traveling direction of the reference light flux R (u, v) at the time of recording the Fourier transform hologram, the information is obtained as the optical axis. Since the position is just twice the shift amount (a, b) with respect to the position when there is object information above, if the position is detected and halved, the position of the object to be recognized, that is, The object to be recognized is the origin (on the optical axis)
The shift amount and direction from the case of can be detected.

In the above description of the object position detecting method,
A series of procedures for finding the position of an object (3)
Although the expression (7) is used for the expression, the object position detecting device for realizing the object position detecting method will be described next. This is, as shown in FIG. 2, at least an image input means 5 for taking in an input object image in the apparatus, and a Fourier transform means for performing Fourier transform on the taken input object image to obtain Fourier transform information. 6, a reference light beam means 7 for generating a reference light beam for producing a hologram with a desired carrier, and a Fourier transform hologram by recording a wavefront in which the Fourier transform information of the input object image and the reference light beam are recorded. Recording means 8 for producing
Reading means 9 for reading the recorded Fourier transform hologram with the Fourier transform information of the input object image, inverse Fourier transform means 10 for further inverse Fourier transforming the read information, and this inverse Fourier transform. It is composed of a detection means 11 for detecting information.

The object position detecting apparatus records a Fourier transform hologram of an object to be recognized, which is read out as Fourier transform information of the input object image, and further, the read out information is subjected to inverse Fourier transform. It is clear from the above description of the method that the position of the object to be recognized is obtained by the above.

In the above description of the object position detecting method,
For a series of procedures for finding the position of an object,
As described above using equations (3) to (7), in the present method,
Further examining Eq. (7), in order to obtain this convolution information, the Fourier transform information of the object obtained by Fourier transforming the object to be recognized is squared (rather than multiplying by taking the phase conjugate, (Multiply)
Then, it is understood that the inverse Fourier transform may be further performed. Next, an object position detecting device for realizing this object position detecting method by obtaining convolution information will be described. This is, as shown in the block diagram of FIG. 3, at least an image input means 5 for taking an input object image into the apparatus, and a Fourier transform means 6 for applying a Fourier transform to the taken input object image. , An integrating means 12 for squaring the Fourier transform information of the input object image by the Fourier transform means 6, and an inverse means for further inverse Fourier transforming the squared Fourier exchange information into convolution information of the input object image. The Fourier transform means 10 and the detection means 11 for detecting the convolution information are included. With this object position detection device, the convolution information of the object to be recognized can be obtained, and the position of the object to be recognized can be further obtained.
It will be clear from the above description.

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Specific embodiments for implementing the present invention will be described below. First, in the present image processing apparatus, the aforementioned object for realizing the object position detecting method used in the aforementioned object position detecting means 2 is realized. Regarding the position detecting device, the first embodiment
A detailed description will be given in the fourth embodiment.

(First Embodiment) An explanation will be given with reference to FIG. 4 showing the optical system. First, the substantially parallel light flux 20 is generated by the coherent light source 21 and the beam expander 22. The coherent light source 21 may be any light source having a coherence that can generate hologram interference fringes with good contrast, but an argon laser having a wavelength of 514.5 nm is used as an example. Although the beam expander 22 is a conventional means, it is configured by using a focusing lens 22a, a pinhole 22b, and a collimating lens 22c. This substantially parallel light flux 20 is
Mirror 23 installed to make the optical system compact
After the optical path is bent at, at the beam splitter 24,
It is divided into a writing / reading light beam 25 and a reference light beam 26. Into the writing / reading light flux 25, the object information to be recognized which is picked up by the image pickup means 1 is inputted into the apparatus by the image input means 5. The image pickup means 1 may be a tube type or a solid-state image pickup element as long as it can pick up a two-dimensional image, but in this embodiment, a CCD camera with a zoom lens is used. An image of an object to be recognized, which is picked up by the CCD, is converted into a TV signal such as NTSC by a circuit in the camera. In order to capture this signal and give image information to the writing / reading light flux 25 as a change in intensity. As the image input means 5, an electrically addressed spatial light modulator, in this embodiment, a transmissive liquid crystal display element which has recently been used in a CRT of a TV or a computer is used.

The writing / reading light beam 25 carrying the object information to be recognized passes through a Fourier transform lens which is a Fourier transforming means 6, and is further divided into a writing light beam 28 and a reading light beam 29 by a beam splitter 27. One of the writing light beams 28 is irradiated onto the reflection type spatial light modulator which is the recording means 8 via the mirrors 30 and 31 and the beam splitter 32. On the reflection type spatial light modulator which is the recording means 8, the object information to be recognized is the Fourier transform information by the Fourier transform means 6. As the reflection type spatial light modulator which is the recording means 8, a BSO is used.
A reflective spatial light modulator using a photorefractive crystal such as (Bi ₁₂ SiO ₂₀ ) or a TN (Twisted Nemati)
c) or a reflective spatial light modulator using liquid crystal of ferroelectricity, LN (LiNbO ₃ ), DKDP (K
D ₂ PO ₄ ) and other reflective spatial light modulators, YIG (Y ₃ Fe ₅ O ₁₂ ) and other reflective magnetic modulators, DMD (Deformable Mirror Devic)
Although it can be used in various other ways such as a reflection type spatial modulator using a deforming mirror such as e), this embodiment has a relatively high resolution and high-speed responsiveness, and is a reflection type for easy handling. Type liquid crystal spatial modulator was used.

On the other hand, the reference light beam means 7 is composed of the reference light beam 26 and the mirror 33 described above.
6 arranges the mirror 33 at an appropriate position and at an appropriate angle, so that the Fourier transform information of the object to be recognized on the reflective liquid crystal spatial light modulator 8 is approximately parallel to the desired angle. Try to ride it as a light flux. At this time, the computer serving as the control means 34 gives a signal for setting the reflective liquid crystal spatial light modulator 8 to the writing mode to the liquid crystal spatial light modulator 8 via the driver 35 of the liquid crystal spatial light modulator 8, and , A signal to open the shutter 36 and close the shutter 37 respectively.
The read light beam 29 is cut off by giving it to each shutter via 8 and 39, and the Fourier transform information of the object to be recognized by the above-described write light beam 28 and the reference light beam 2
By irradiating a substantially parallel light beam with a desired angle by 6,
The same Fourier transform hologram as shown in the equation (2) is recorded on the liquid crystal spatial modulator 8 which is a recording means. The shutters 36 and 37 are made of the same type of liquid crystal as that used for the liquid crystal spatial modulator 8 which is the recording means in consideration of the ease of control, but if the operating speed is sacrificed to some extent, Others such as mechanical ones may be used.
In the figure, the polarizer 13 is arranged between the shutter 36 and the reflective liquid crystal spatial light modulator 8 in the optical path through which the writing light beam 28 and the reference light beam 26 pass. This means that the writing light beam 28 reflected by the mirror surface of the beam splitter 32 is S
Since the reference light beam 26 that has a large number of polarization components and the reference light beam 26 that passes through the beam splitter 32 has a large amount of P-polarization components and there are few components that interfere with each other, the two polarizers 13 cause two polarization components in the pass axis direction of the polarizer 13. This is for enabling interference, and the arrangement of the polarizer 13 improves the contrast of interference fringes.

Next, the Fourier transform hologram formed on the reflection type liquid crystal spatial light modulator 8 is read out by the read-out means 9, which reads out the read light beam 29 as a pentaprism 40 as an erecting prism. The light beam is incident on the reflective liquid crystal spatial light modulator 8 through the beam splitter 41 in a direction directly opposite to the writing light beam 28. Here, the beam splitter 41 is configured by replacing it with a polarization beam splitter, or a polarizer is provided on the incident side of the normal beam splitter 41 (exit side of the pentaprism 40) and the exit side of the beam splitter 41 (inverse Fourier transform). It can be replaced with an analyzer arranged on the incident side of the conversion lens 10. With such a configuration, unnecessary light can be cut and the output contrast can be increased. The pentaprism 40 is used because the Fourier transform information of the read light beam 29 is the reverse side of the Fourier transform information of the write light beam 28 (opposite to the direction in which the image is inverted by a mirror or the like). Is to correct the direction of the Fourier transform information of the write light beam 28 and the Fourier transform information of the read light beam 29.
Match above. Further, this erecting prism may be replaced with a mirror, and an erecting prism may be placed at the position of the mirror 30 or the mirror 31. In practice, the control means 34 further gives a signal for setting the reflective liquid crystal spatial light modulator 8 to the read mode via the driver 35, and further the shutter 3
The signals from the write light beam 28 and the reference light beam 26 are cut by applying a signal for closing the shutter 6 and opening the shutter 37 to the respective drivers 38, 39, and the Fourier transform of the object to be recognized by the read light beam 29 is performed. Only the information is read by applying it to the reflective liquid crystal spatial light modulator 8. By this operation, the information shown in the equation (3) is obtained.

Next, using the inverse Fourier transform lens 10 as the inverse Fourier transform means 10 for further inverse Fourier transforming the read information, the inverse Fourier transform information of the information read on the screen 42, that is, ( 4)
Get the information shown in the formula. This information is further provided to the detection means 1
Two-dimensional detector 43 which is 1, C in the case of this embodiment
It is transmitted to the CD camera through the imaging lens 44 and detected. In the figure, for simplification, the detecting means is shown only in one direction among the information of three directions to be detected. In the figure,
The direction of observing the information represented by formula (7) is the direction of (A), the information represented by formula (6) is the direction of (C), and the information represented by formula (5) is Observed in the direction of (B). Further, in this drawing, the angle of the reference light beam 26 is exaggerated from the actual one in order to avoid complication of the drawing, so that the obtained information in three directions is obtained by separate inverse Fourier transform means and detection means. Although the angle of the actual reference light beam is relatively small, there are cases in which it can be detected by a set of a large-diameter inverse Fourier transform lens capable of performing an inverse Fourier transform at one time and a detection means. Many.

As an example, FIG. 5 shows the processing result of this device when a circle with a diameter of 2.5 mm is input as an input object on the optical axis.
(Also in this example, detection is performed by a pair of inverse Fourier transform lenses and detection means). FIG. 5 is a drawing of a negative in which information on three directions obtained on the screen 42 is photographed. (A) in the figure is the information of the equation (7) corresponding to the direction of (A) of FIG. 4, (B) is the information of the equation (5) corresponding to the direction of (B) of FIG. ,
(C) is the information of the expression (6) corresponding to the direction of (C) of FIG.

Next, a change in the information represented by the equations (5) to (7) when the object was displaced in one direction perpendicular to the optical axis was detected. The results are shown in Table 1 and FIG.

[0034]

The values in Table 1 are the CCD as the detecting means 43.
After AD conversion of the information detected in, input into the computer,
The peak position of each information obtained as the coordinate of the pixel of the CCD is converted into the original distance coordinate by the calculation in the computer in consideration of the (imaging) magnification in the process in the middle. In FIG. 6, circles show changes in the information represented by the equation (7). When this result is fitted by the least squares method, a straight line like a solid line is obtained. The slope of this straight line is found to be 2.04 by this fitting, which is just in agreement with the theory. Further, the square marks in FIG. 6 indicate changes in the information represented by the equation (5), and when this result is fitted by the least squares method, a straight line like a solid line is obtained. The slope of this straight line is found to be 1.03 by this fitting, which also exactly matches the theory. Further, in FIG. 6, the triangular marks show the changes in the information expressed by the equation (6). When this result is fitted by the least squares method, a straight line like a solid line is obtained. The slope of this straight line is found to be 0 by this fitting, which is exactly the same as the theory. In this example, the input object was moved in only one direction for simplicity.
Needless to say, the same detection can be performed in the other direction or in the case where both directions are deviated.

From the above, according to the apparatus shown in this embodiment, the position of the object to be recognized is just half the peak position of the information represented by the equation (7), and is further represented by the equation (5). It is obtained as the same position as the peak position of the information
It has been found that the method of the invention can detect the position of the object in theory.

In this embodiment, the same Fourier transform lens 6 and inverse Fourier transform lens 10 having the same focal length are used, and the optical path length from the image input means 5 to the Fourier transform lens 6 and the Fourier transform are used. The optical path length from the lens 6 to the spatial light modulator 8, the optical path length from the spatial light modulator 8 to the inverse Fourier transform lens 10, and the optical path length from the inverse Fourier transform lens 10 to the screen 42 are
Although it is adjusted to this focal length, the Fourier transform lens 6
And the inverse Fourier transform lens 10 may have different focal lengths, and in this case, it is obvious that the detected object position may be multiplied by a coefficient according to the difference in the focal lengths.

In this embodiment, the optical path length between the image input means 5 and the Fourier transform lens 6 is set so that the phase information of the object can be handled. If not handled, there is no problem even if the optical path length from the image input means 5 to the Fourier transform lens 6 is shortened to make the system compact.

(Second Embodiment) FIG. 7 shows the optical system of the essential part of this embodiment. This embodiment is a modification of the first embodiment, and the image input means 5 is of the optical writing type. This is a modification of the reflective spatial light modulator, and the configuration and operation of the other parts are the same as in the first embodiment.

As the optical writing type reflection type spatial light modulator, the same element as that which can be used for the recording means 8 in the first embodiment can be used. Therefore, also in the case of this embodiment, the reflective liquid crystal spatial light modulator is used as the image input means 5. The object information S to be recognized is imaged on the direct reflection type liquid crystal spatial light modulator 5 by the imaging lens which is the imaging means 1. This information is recorded by the computer, which is the control means 5b, by giving a signal for setting the liquid crystal spatial light modulator 5 to the writing mode to the liquid crystal spatial light modulator 5 via the driver 5a of the liquid crystal spatial light modulator 5. It Further, a signal for setting the liquid crystal spatial light modulator 5 to the read mode is given to the liquid crystal spatial light modulator 5 via the driver 5a of the liquid crystal spatial light modulator 5, and the same coherent light source 21 and beam beam as in the first embodiment are provided. A substantially parallel light beam 2 generated by the panda 22
0 is bent by the mirror 23 and the beam splitter 24 and is incident on the liquid crystal spatial light modulator 5, and is read out as its reflected light. This read information is
The read light beam 25 is further Fourier-transformed by the Fourier transform means 6. Further, of the light flux 20, the light flux that has passed through the beam splitter 24 is sent to the reference light flux means 7 as a reference light flux 26. From this point onward, the configuration and operation are similar to those of the first embodiment, but it is clear that the position of the object can be detected as in the first embodiment.

Also in this embodiment, as in the first embodiment,
The beam splitter 24 is configured by replacing it with a polarization beam splitter, or a polarizer is provided on the incident side of the normal beam splitter 24 (exit side of the mirror 23) and the exit side of the beam splitter 24 (incident on the Fourier transform lens 6). It can be replaced by the one with an analyzer placed on the side). With such a configuration, unnecessary light can be cut and the output contrast can be increased.

Although the optical writing type reflective spatial light modulator has been described here, it is apparent that the optical writing type transmissive spatial light modulator can be configured in substantially the same manner. Is.

(Third Embodiment) This embodiment, like the first embodiment, also shows an object position detecting device which realizes the above-described object position detecting method, and the spatial light modulator is a transmissive type. Although the optical system is changed to achieve this, the basic configuration and operation are the same.

Explaining with reference to the diagram showing the optical system of FIG. 8, first, a substantially parallel light beam 45 is generated by a coherent light source and a beam expander similar to those of the first embodiment (not shown), and the beam splitter 46 It is divided into a writing / reading light beam 47 and a reference light beam 48. In the writing / reading light flux 47, the object information to be recognized, which is imaged by the image pickup means 1 similar to the first embodiment not shown, is an electric address type spatial light modulator such as a liquid crystal which is the image input means 5. Input by displaying on. The writing / reading light flux 47 carrying the object information to be recognized should be recognized on the transmissive spatial light modulator which is the recording means 8 via the mirror 49 and the Fourier transform lens which is the Fourier transform means 6. Put the Fourier transform information of the object. The transmissive spatial light modulator which is the recording means 8 includes BSO (Bi ₁₂ SiO ₂₀ )
Transmissive spatial light modulator using photorefractive crystal, etc., transmissive spatial light modulator using liquid crystal such as TN (Twisted Nematic) or ferroelectric, or Seed
(Self Electro-Optic Device) and other MQW (Multiple
Quantum Wells) or etalon type transmission type spatial modulator using a bulk structure semiconductor such as ZnS can be used in various other ways, but in this embodiment, it has relatively high resolution and high-speed response. A transmissive liquid crystal spatial modulator was used for easy handling.

On the other hand, the reference light beam means 7 is composed of the reference light beam 48 and the mirror 50 described above.
Is arranged at an appropriate position and at an appropriate angle, the Fourier transform information of the object to be recognized on the transmissive liquid crystal spatial light modulator 8 described above is added to the reference light flux 48 as a substantially parallel light flux of a desired angle. Will make you ride more. At this time, the computer as the control means 51 gives a signal to the liquid crystal spatial light modulator 8 via the driver 52 of the liquid crystal spatial light modulator 8 to set the transmissive liquid crystal spatial light modulator 8 to the write mode. Further, by giving signals for opening the shutter 53 and the shutter 54 via the drivers 55 and 56, respectively, the transmission type liquid crystal spatial light modulator 8
The Fourier transform hologram of the object to be recognized by the writing / reading light beam 47 and the substantially parallel light beam of the desired angle by the reference light beam 48 are the same as those shown in the equation (2). Will be formed. Similar to the first embodiment, the shutters 55 and 56 are made of the same kind of liquid crystal as that used for the liquid crystal spatial modulator 8 which is the recording means in consideration of the ease of control. Others, such as mechanical ones, may be used if they are sacrificed a little.

Next, the Fourier transform hologram formed on the transmissive liquid crystal spatial light modulator is read by the reading means 9, which serves as the writing / reading light beam 47.
Is used as it is, and further, the control means 51 gives a signal for setting the transmissive liquid crystal spatial light modulator 8 to the read mode through the driver 52 of the liquid crystal spatial light modulator 8, and
Signals for closing the shutter 54 and opening the shutter 53 are given through the respective drivers 56, 55 to cut the information from the reference light beam 48, and the writing / reading light beam 47 Fourier transform of the object to be recognized. Only the conversion information is given to the transmissive liquid crystal spatial light modulator 8 to read it. By this operation, the information shown in the equation (3) can be obtained.

Next, as the inverse Fourier transform means 10 for further inverse Fourier transforming the read information, the inverse Fourier transform lens 10 is used, and the inverse Fourier transform information of the information read on the screen 57, that is, ( 4)
The information shown in the formula can be obtained. This information is further detected by the detection means 11 (not shown) similar to that in the first embodiment. In the figure, the direction of observing the information represented by formula (7) is the direction of (A), the information represented by formula (6) is represented by the direction of (C), and further represented by formula (5). Information is observed in the direction of (B).

Also in this drawing, in order to avoid complication of the drawing, the angle of the reference light beam 48 is exaggerated from the actual one, as in the first embodiment, so that the obtained information in three directions is obtained. Are used to use separate inverse Fourier transform means and detection means, but the angle of the actual reference light beam 48 is relatively small, and in fact, a large aperture inverse Fourier transform capable of performing inverse Fourier transform in three directions at a time. In many cases, it can be detected by one set of lens and detection means. Further, instead of the mirror 50 used as the reference light flux means 7, the beam splitter 4
In some cases, it is better to incline the reference beam 48 as a substantially parallel beam having a desired angle by tilting the beam 6.

Although a concrete processing example is omitted in the present embodiment, it is obvious that the method of the present invention can detect the position of the object theoretically in consideration of the above contents and the result of the first embodiment.

Also in this embodiment, as in the first embodiment,
Fourier transform lens 6 and inverse Fourier transform lens 1 used
The same lens having the same focal length is used for 0, and the optical path length from the image input unit 5 to the Fourier transform lens 6, the optical path length from the Fourier transform lens 6 to the spatial light modulator 8, and the spatial light modulator 8 are used. To the inverse Fourier transform lens 10 and the optical path length from the inverse Fourier transform lens 10 to the screen 57 are adjusted to this focal length.
It is obvious that the Fourier transform lens 6 and the inverse Fourier transform lens 10 may have different focal lengths, and in this case, the detected object position may be multiplied by a coefficient depending on the difference in the focal length.

Further, in this embodiment, the optical path length between the image input means 5 and the Fourier transform lens 6 is set in order to handle the phase information of the object, but if the phase information of the object is not handled, the image input is performed. There is no problem even if the system is made compact by shortening the optical path length from the means 5 to the Fourier transform lens 6.

(Fourth Embodiment) In this embodiment, of the above-mentioned object position detecting method, an apparatus for obtaining a convolution image represented by the equation (7) is realized by a computer or the like.

Explaining with reference to FIG. 3, the object information to be recognized is input to the frame memory which is the image input means 5 by the image pickup means 1 such as a CCD not shown and AD-converted. The AD-converted object information is sent to the FFT processor in the computer which is the Fourier transform means 6, subjected to the Fourier transform, and stored in the memory in the computer. Further, the Fourier-transformed and stored information is called from the memory by the accumulating means 12, the information is squared for each pixel on the computer, and stored again in the memory. The information squared for each pixel is sent to the FFT processor in the computer that is the inverse Fourier transform means 10, is subjected to the inverse Fourier transform, and is stored in the memory in the computer as convolution information. Finally, this convolution information is detected by the detecting means 1.
The coordinate of the peak position is detected, and the coordinate of half the position is output as the coordinate of the object position.

Next, a concrete processing example of the convolution image obtained by this object position detecting device is shown in FIGS. First, in FIG. 9, as shown in FIG. 9A, the character S is fetched into the device, and 100 × 1
It shows the result of processing after AD conversion into information of 00 pixels. The letter S shown in FIG. 5A has coordinates (50,
The information is at coordinates (35, 35) shifted by (-15, -15) from the origin (optical axis) at 50). A bird's-eye view of the convolution image resulting from this processing is shown in FIG. 7B, and a contour map is shown in FIG. The height c in FIG. 7B shows a normalized convolution value. The peak position of this convolution image is shown in FIG.
As shown in (c), the coordinates are (20, 20), and the shift amount is (-30, -30) from the origin, so that exactly half of them is in the vector indicating the position of the input image. I am doing it. Similarly, 100 × 10 shown in FIG.
The character L AD-converted into information of 0 pixels has coordinates (50,
The information is at coordinates (55, 63) shifted by (5, 13) from the origin (optical axis) at 50). A bird's-eye view of the convolution image resulting from this processing is shown in FIG.
The contour map is shown in FIG. The peak position of this convolution image is at coordinates (60, 76) as shown in (b) and (c) of the figure, and the shift amount is (10, 26) from the origin, Exactly half of it coincides with the vector indicating the position of the input image.

As shown by the above results, it is clear that the apparatus according to the present embodiment can obtain the convolution information of the object to be recognized and the position of the object to be recognized. The explanation is not only theory
It turns out that it is actually required.

(Fifth Embodiment) Next, the first to fourth embodiments will be described.
An image processing apparatus for recognizing an object using the above-mentioned object position detecting apparatus for realizing the above-mentioned object position detecting method shown in the embodiment will be described below.

The apparatus according to the present embodiment has the configuration described above with reference to FIG. 1, and will be described more specifically with reference to FIG. 11. As the image pickup means 1, a CC having a zoom lens 14 is attached.
An image of a recognition target is picked up using the D camera 15 or the like, and this image is taken into the object position detecting means 2 by the image input means 5 (not shown) and processed by the devices shown in the first to fourth embodiments. As a result, the position information is obtained, and this position information is sent to the moving means 3.

The moving means 3 includes stepping motors 58 and 59 which are tilting mechanisms for moving the optical axis attached to the CCD camera 15 of the image pickup means 1, drivers 60 and 61 for moving the stepping motors, and an object. From the position information sent from the position detection means 2, an appropriate movement amount of the stepping motors 58 and 59, which are the tilting mechanism, is calculated,
It comprises a controller 62 that gives information for moving the calculated amount to the respective drivers 60, 61 (the position information here means the object position detecting means 2).
The information of formula (7) sent from is halved,
Alternatively, it is the information of the expression (5) or a combination of both of these information, and the origin may be calibrated in advance when there is an object on the optical axis. Although complicated, these pieces of information may be considered in terms of a positional relationship with the information of Expression (6). ).

With the above configuration, the position (shift amount and direction from the optical axis) of the input object to be recognized is detected, the image pickup means is moved from this position information, and the object information to be recognized is optically detected. It can be moved to the axis and then sent to the recognition means 4. That is, since the same information can be sent to the recognition means 4 regardless of the position of the object on the input screen, it is possible to improve the generalization ability regarding shift deformation without missing information. .

Next, the recognition means 4, which is shown in FIG.
Will be explained. The upper half of this optical system is a coordinate conversion optical system that performs logarithmic polar coordinate conversion of the input image with the same configuration as the prior art example shown in FIG. It is sent from the CCD camera of the image pickup means 1 whose position is on the optical axis by the above-mentioned device and mechanism, which is inputted to the image input means 63 which is a transmission type spatial light modulator such as liquid crystal placed on the input surface. The obtained image is irradiated with substantially parallel coherent light 64 generated by the same means as in the first embodiment, and further, phase information for coordinate conversion is superposed on the coherent light 64 by the CGH 65, and then the Fourier transform lens 66 is used by the Fourier transform lens 66. The Fourier transform is performed, and the coordinate transform information output on the coordinate transform surface 67 placed on the rear focal plane of the Fourier transform lens 66 is obtained by the detecting means 68 such as a CCD.

This information is further input to the lower half correlation optical system having the same configuration as that of the prior art example shown in FIG. 15 via the transmission type spatial light modulator 69 such as liquid crystal. The lower half of the correlation optical system includes a transmissive spatial light modulator 69 for writing the above-mentioned coordinate conversion information into the system, a substantially parallel coherent light 64, and coordinate conversion information input into the system by these. Fourier transform lens 70 for performing a Fourier transform of the image, and MSF 7 for superimposing reference image information for recognition placed on the rear focus of the Fourier transform lens 70.
1 and further output to another Fourier transform lens 72 for Fourier transforming this superimposed information, and a correlation surface 73 placed at the rear focal point of this other Fourier transform lens 72. CC for detecting correlation peak
It is composed of a detection means 74 such as D. This recognition means
As described in the explanation of the preceding example, the rotation of the input image and the deformation of enlargement / reduction are transformed into the shift amount in the coordinate transformation optical system of the upper half, and the correlation optical system of the lower half is the shift invariant. By performing the correlation processing, it can be seen that the recognition can be performed without the rotation and the deformation of the enlargement / reduction.

The MS of the image of the object that serves as a reference for recognition
F71 places the hologram dry plate at this position, inputs the object to be recognized on the optical axis by the image pickup means 1, and after the above-mentioned series of processing, performs coordinate conversion of the object information to be recognized formed on this surface. Further, the Fourier-transformed information and the reference light angled so as to have a desired spatial frequency are interfered with each other and recorded.

As described above, the image processing apparatus of this embodiment is
It can be seen that images that have undergone deformation such as shifting, rotation, and scaling can be recognized with high accuracy. Needless to say, a similar device can be constructed by using a JTC (joint transform correlator) or the like as the correlation system.

(Sixth Embodiment) The invention of this embodiment is a modification of the fifth embodiment, and by using the optical system shown in FIG. 13 as the recognition means 4, recognition of the input image (pattern classification) in real time. An image processing device that performs In the figure, the elements up to the upper half of the coordinate conversion optical system are composed of exactly the same elements as in the fifth embodiment. The lower half of the figure is subjected to Fourier transform of the information converted by this coordinate conversion optical system, and further, a Fourier transform optical system for sending to the neural network in the subsequent stage, and this Fourier transform. It is a neural network that recognizes information. In this Fourier transform optical system, a spatial light modulator 69 such as a liquid crystal for inputting coordinate transform information converted by the coordinate transform optical system into the system,
The substantially parallel coherent light 64, a Fourier transform lens 70 for Fourier transforming the coordinate transform information input into the system by these, and the Fourier transform lens 70.
And a detection means 76 such as a CCD for detecting the information output to the Fourier transform plane 75 placed at the rear focal point, sampling it at appropriate intervals, and sending it to the neural network 77 constructed on the computer. To be done. Since the Fourier transform optical system is a shift-invariant optical system, even if the input image is deformed, it is converted into the shift amount by the coordinate conversion optical system. Therefore, on the Fourier transform surface 75, the image is deformed. However, the same information is obtained,
It can be seen that the information after this processing is converted into information insensitive to deformation. For example, considering a back propagation model as the neural network 77, it is first necessary to determine the synaptic weight between the layers of the neural network 77 by learning. This is because the image to be recognized (pattern classification) is sequentially input on the optical axis by the image pickup means 1, and the neurons of the input layer of the neural network 77 are further Fourier-transformed by the coordinate transformation information of these images. Images are sequentially input. At this time, the neural network 77
The back-propagation learning rule may be used to determine the synapse weight between the layers so that the firing of the neurons in the output layer is selectively performed corresponding to the image to be recognized (pattern classification).
At the time of recognition (pattern classification), the synapse load between the layers is fixed, an image to be recognized (pattern classification) is presented on the input surface by the image input means 1, and firing of neurons in the output layer is detected. Good.

With the above structure, in addition to the pattern classification ability of the ordinary neural network 77, even if the input image is deformed by shifting and rotation / enlarging / reducing, the generalization ability that can be recognized as the same object is large. It is clear that significant improvements can be achieved. Further, in the above embodiment, the back propagation model was considered as the neural network 77, but it is needless to say that the generalization ability can be greatly improved even when another model such as the Hopfield model is used. Absent.

In the above, the explanation has been made with emphasis on the addition of the improvement of the general-purpose ability concerning the transformation of rotation and scaling using the logarithmic polar coordinate transformation and the addition of the improvement of the generalization ability concerning the transformation of shift. However, the present invention is not limited to this, and it is clear that the same effect can be obtained by combining with other conversion, and a general image processing device is used instead of the recognition means. Needless to say, further effects can be expected as a line-of-sight switching device or the like, and further as a line-of-sight switching autofocus mechanism of a general imaging device.

The object position detecting method and apparatus of the present invention described above and the image processing apparatus using them are summarized as follows. (1) An object position detecting method characterized by detecting the position of this input object image using a convolution image of the input object image.

(2) Object position characterized by detecting the position of this input object image by reading the Fourier transform hologram of the input object image with the Fourier transform information of the input object image and further performing inverse Fourier transform Detection method.

(3) Image input means for taking an input object image into the system, Fourier transform means for subjecting the taken input object image to Fourier transform to obtain Fourier transform information, and a hologram with a desired carrier. A reference light beam means for generating a reference light beam for producing, and a recording means for producing a Fourier transform hologram by recording a wavefront that interferes with the Fourier transform information of the input object image and the reference light beam, Reading means for reading the recorded Fourier transform hologram with Fourier transform information of the input object image, inverse Fourier transform means for further inverse Fourier transforming the read information, and the inverse Fourier transformed information. An object position detecting device, comprising: a detecting unit for detecting

(4) Between the Fourier transforming means and the recording means, there is a dividing means for dividing the light flux to be written into the recording means and the light flux to be read out from the recording means, and the optical path of the write light flux, or The object position detecting apparatus according to (3) above, which has an erecting prism in any one of the optical paths of the read light flux.

(5) A polarizer is arranged in front of the writing light flux incident side of the recording means, and behind the semi-transmissive / semi-reflective film that reflects the writing light flux and transmits the reference light flux. The object position detection device according to (4) above.

(6) The object position detecting device according to the above (3), wherein the recording means comprises an optical writing type spatial modulator.

(7) Image input means for taking in the input object image in the system, Fourier transform means for subjecting the taken input object image to Fourier transform, and Fourier transform of the input object image by the Fourier transform means. Integral means for squaring the transform information, inverse Fourier transform means for further inverse Fourier transforming the squared Fourier transform information into convolution information of the input object image, and for detecting this convolution information And an object position detecting device.

(8) Image pickup means provided with an image forming optical system for forming an input object image as an optical image, and a photoelectric conversion means for converting the formed optical image into an electric signal to obtain an input object image signal. And an object position detecting means for detecting the position of the input object.

(9) Image input means for allowing the object position detecting means to take in the input object image signal into the system,
Fourier transform means for performing Fourier transform on the input object image by the captured input object image signal to obtain Fourier transform information, and reference light flux means for generating a reference light flux for producing a hologram with a desired carrier. Recording means for recording a wavefront in which the Fourier transform information of the input object image based on the input object image signal and the reference light flux are made to produce a Fourier transform hologram, and the recorded Fourier transform hologram as the input A reading means for reading the Fourier transform information of the object image, an inverse Fourier transform means for further inverse Fourier transforming the read information, and a detecting means for detecting the inverse Fourier transform information. The above (8) characterized by having
The image processing device described.

(10) The object position detecting means takes in the above-mentioned input object image signal into the system, and image input means, and Fourier transforms the input object image by this taken-in input object image signal to perform Fourier transform. Fourier transforming means for obtaining information, and integrating means for squaring the Fourier transform information of the input object image by the Fourier transforming means,
Inverse Fourier transform means for further inverse Fourier transforming the squared Fourier transform information into the convolution information of the input object image, and detection means for detecting the convolution information. The image processing device according to (8) above.

(11) Image pickup means for inputting input object information as an image, object position detection means for detecting the position of the input object, and image pickup based on the position of the object obtained by this object position detection means An image processing apparatus comprising: a moving unit that moves the unit.

(12) The object position detecting means includes an image inputting means for taking in the image into the system, a Fourier transforming means for subjecting the fetched image to Fourier transform to obtain Fourier transform information, and a desired A reference light beam means for generating a reference light beam for producing a hologram with a carrier, and a recording means for producing a Fourier transform hologram by recording a wavefront in which the Fourier transform information of the image and the reference light flux are interfered with each other; Reading means for reading the recorded Fourier transform hologram with Fourier transform information of the image, inverse Fourier transform means for further inverse Fourier transforming the read information, and the inverse Fourier transformed information The image processing apparatus according to (11) above, further comprising a detection unit for detecting

(13) The object position detecting means includes an image inputting means for taking in the image into the system, a Fourier transforming means for subjecting the fetched image to Fourier transform to obtain Fourier transform information, and Integration means for squaring the Fourier transform information of the image by the Fourier transform means, and inverse Fourier transform means for further inverse Fourier transforming the squared Fourier transform information to convolution information of the image; The image processing apparatus according to (11) above, further comprising a detection unit for detecting convolution information.

(14) Image pickup means for inputting input object information as an image, object position detection means for detecting the position of the input object, and image pickup based on the position of the object obtained by the object position detection means Image processing characterized by having a moving means for moving the means to move the input object on the optical axis and a recognizing means for inputting and recognizing the input object information moved on the optical axis by the moving means. apparatus.

(15) The object position detecting means includes an image inputting means for taking in the image into the system, a Fourier transforming means for subjecting the fetched image to Fourier transform to obtain Fourier transform information, and a desired A reference light beam means for generating a reference light beam for producing a hologram with a carrier, and a recording means for producing a Fourier transform hologram by recording a wavefront in which the Fourier transform information of the image and the reference light flux are interfered with each other; Reading means for reading the recorded Fourier transform hologram with Fourier transform information of the image, inverse Fourier transform means for further inverse Fourier transforming the read information, and the inverse Fourier transformed information The image processing apparatus according to (14) above, further comprising: a detection unit for detecting

(16) The object position detecting means includes an image inputting means for taking in the image into the system, a Fourier transforming means for performing a Fourier transform on the fetched image to obtain Fourier transform information, and Integration means for squaring the Fourier transform information of the image by the Fourier transform means, and inverse Fourier transform means for further inverse Fourier transforming the squared Fourier transform information to convolution information of the image; The image processing apparatus according to (14) above, further comprising a detection unit for detecting convolution information.

[0083]

As is apparent from the above description, the object position detecting method, apparatus and image processing apparatus using the same according to the present invention detect the shift amount of the input object to be recognized from the optical axis. Since this object is moved so as to be on the optical axis and then sent to the recognizing means, it is possible to improve the generalization ability with respect to shift deformation without missing information. Further, when a coordinate transformation optical system for performing logarithmic polar coordinate transformation is used as a recognition means in combination with a correlation optical system or a neural network, it is possible to recognize with high accuracy an image that has been deformed by shifting, rotation or scaling. Become.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an image processing apparatus according to the present invention.

FIG. 2 is a block diagram for explaining a configuration of an object position detection device of the present invention.

FIG. 3 is a block diagram for explaining the configuration of another object position detection device of the present invention.

FIG. 4 is a diagram showing an optical system of the object position detecting device according to the first embodiment.

FIG. 5 is a diagram showing a processing example of the object position detecting device according to the first embodiment.

FIG. 6 is a diagram showing a change in information with respect to a shift amount of an object in the device of the first embodiment.

FIG. 7 is a diagram showing a main part of an optical system of an object position detecting device according to a second embodiment.

FIG. 8 is a diagram showing an optical system of an object position detecting device according to a third embodiment.

FIG. 9 is a diagram for explaining a processing example by the object position detection device of the fourth embodiment.

FIG. 10 is a diagram for explaining another processing example by the object position detection device of the fourth embodiment.

FIG. 11 is a diagram illustrating a configuration of an image processing apparatus according to a fifth embodiment.

FIG. 12 is a diagram showing an optical system of a recognition means in a fifth embodiment.

FIG. 13 is a diagram showing an optical system of recognition means in a sixth embodiment.

FIG. 14 is a diagram showing a conventional logarithmic polar coordinate conversion optical system.

FIG. 15 is a diagram showing a conventional correlation optical system.

[Explanation of symbols]

S ... Object information 1 ... Imaging means 2 ... Object position detecting means 3 ... Means of transportation 4 ... Recognition means 5 ... Image input means 6 ... Fourier transform means (Fourier transform lens) 7 ... Reference luminous flux means 8 ... Recording means 9 ... reading means 10 ... Inverse Fourier transform means (inverse Fourier transform lens) 11 ... Detection means 12 ... Accumulation means 13 ... Polarizer 14 ... Zoom lens 15 ... CCD camera 20 ... Almost parallel light flux 21 ... Coherent light source 22 ... Beam expander 22a ... Lens for narrowing down 22b ... Pinhole 22c ... Collimating lens 23 ... Mirror 24 ... Beam splitter 25 ... Writing / reading light flux 26 ... Reference luminous flux 27 ... Beam splitter 28 ... Writing light flux 29 ... Read light flux 30, 31 ... Mirror 32 ... Beam splitter 33 ... Mirror 34 ... Control means 35 ... driver 36, 37 ... Shutter 38, 39 ... Driver 40 ... Upright prism (penta prism) 41 ... Beam splitter 42 ... Screen 43 ... Two-dimensional detector 44 ... Imaging lens 5a ... driver 5b ... Control means 45 ... Almost parallel light flux 46 ... Beam splitter 47 ... Writing / reading light flux 48 ... Reference luminous flux 49, 50 ... Mirror 51 ... Control means 52 ... driver 53, 54 ... Shutter 55, 56 ... driver 57 ... Screen 58, 59 ... Stepping motor 60, 61 ... driver 62 ... Controller 63 ... Image input means 64 ... Coherent light that is substantially parallel 65 ... CGH 66 ... Fourier transform lens 67 ... Coordinate conversion plane 68 ... Detection means 69 ... Spatial light modulator 70 ... Fourier transform lens 71 ... MSF 72 ... Fourier transform lens 73 ... correlation surface 74 ... Detecting means 75 ... Fourier transform plane 76 ... Detecting means 77 ... Neural network

Claims (8)

(57) [Claims] 1. A Fourier transform hologram of an input object image is read out by using Fourier transform information of the input object image, and further inverse Fourier transform is performed to obtain an input object image
An object position detection method characterized by detecting the position of this input object image from a convolution image . 2. An image input means for taking an input object image into the system, a Fourier transform means for subjecting the taken input object image to Fourier transform to obtain Fourier transform information, and a hologram with a desired carrier. A reference light beam means for generating a reference light beam for producing, a recording means for producing a Fourier transform hologram by recording a wavefront in which the Fourier transform information of the input object image and the reference light flux are interfered with each other, and Reading means for reading the recorded Fourier transform hologram with the Fourier transform information of the input object image, inverse Fourier transform means for further inverse Fourier transforming the read information, and the inverse Fourier transform .
An object position detecting device, comprising: detection means for detecting convolution information of the obtained input object image . 3. The optical recording type spatial modulator, wherein the recording means is an optical writing type spatial modulator.
The object position detection according to claim 2, characterized in that
Output device. 4. The Fourier transforming means and the recording means
Between the recording means and the light beam to be written in the recording means.
The above-mentioned writing
The optical path of the incident light flux or the optical path of the read light flux
The erecting prism is provided in any one of the above.
2. The object position detecting device according to item 2. 5. A front side of a writing light flux incident side of the recording means
And the reference light flux is reflected by the write light flux.
A polarizer is placed after the semi-transmissive semi-reflective film that transmits light.
The object position detecting device according to claim 4, characterized in that
Place 6. An image input means for taking an input object image into the system, a Fourier transform means for applying a Fourier transform to the taken input object image, and a Fourier transform of the input object image by the Fourier transform means. Integrating means for squaring the information, inverse Fourier transform means for further inverse Fourier transforming the squared Fourier transform information into convolution information of the input object image, and for detecting this convolution information An object position detecting device comprising: a detecting unit. 7. A Fourier transform hologram of an input object image
Read out the Fourier transform information of the input object image,
Input object image obtained by inverse Fourier transform of
In the image where the DC component of the Fourier transform hologram is applied to
Is characterized by detecting the position of this input object image.
Object position detection method. 8. An image for capturing an input object image in the system
The input means and the Fourier transform of the captured input object image
Fourier transform procedure for performing conversion and obtaining Fourier transform information
Steps and references for making holograms with desired carriers.
Reference luminous flux means for generating an illuminating luminous flux and said input
Waves that interfere the Fourier transform information of the object image and the reference light flux
A record for recording a surface and producing a Fourier transform hologram.
Enter the recording means and the recorded Fourier transform hologram.
Readout means for reading out by Fourier transform information of force object image
And further inverse Fourier transform this read information
Inverse Fourier transform means for
Of the information obtained, the Fourier transform hologram of the input object image
The detection means for detecting the information on which the DC component is applied
An object position detecting device having.