JP2762463B2 - Focus detection device - Google Patents

Description

Description: FIELD OF THE INVENTION The present invention relates to an improvement in a focus detection device used for a camera or the like.

BACKGROUND OF THE INVENTION Conventionally, as one type of a camera focus detection device,
It is known that a shift of two subject images formed by a light beam that has passed through a divided pupil region of a photographing lens is observed to detect a focus state of the photographing lens. For example, U.S. Pat. No. 4,185,191 discloses an apparatus in which a fly-eye lens group is arranged on a predetermined imaging plane of a photographic lens and two images which are shifted according to the defocus amount of the photographic lens are generated.

In addition, an aerial image formed on a predetermined image forming plane by two side-by-side secondary image forming optical systems is guided to two photoelectric conversion element arrays, respectively, and a so-called “2” for detecting a relative displacement between the two images is performed. The next imaging method is disclosed in JP-A-55-118019, JP-A-55-155331.
No. 6,009,036. Although this method has a slightly longer overall length, it has the advantage of not requiring a special optical system.

FIG. 8 shows an optical system of the focus detection device of the secondary imaging system. A field lens 3 is disposed with the same optical axis 2 as the taking lens 1 whose focus is to be detected. Two secondary imaging lenses 4a and 4b are arranged at positions symmetrical with respect to the optical axis 2 on the rear side. Further behind it, a photoelectric conversion element array 5
a and 5b are arranged. Stops 6a and 6b are provided near the secondary imaging lenses 4a and 4b. The field lens 3 forms an image of the exit pupil of the photographing lens 1 substantially on the pupil planes of the two secondary imaging lenses 4a and 4b. As a result, the light beams incident on the secondary imaging lenses 4a and 4b are emitted from the non-overlapping equal-area areas corresponding to the secondary imaging lenses 4a and 4b on the exit pupil plane of the photographing lens 1. It will be. The aerial image formed near the field lens 3 is a secondary imaging lens 4
When the image is re-imaged on the surfaces of the photoelectric conversion element arrays 5a and 5b by a and 4b (the focus may be out of focus), based on the displacement of the aerial image position in the optical axis direction, the photoelectric conversion element arrays 5a and 5b 2 on the surface of 5b
The image will change its position. FIG. 9 shows how the relative positions of the two images are displaced. As shown in FIGS. 9 (b) and 9 (c), focusing on the in-focus state in FIG. The two images formed on the surfaces of the photoelectric conversion element arrays 5a and 5b move in directions opposite to each other. If the image intensity distribution is photoelectrically converted by the photoelectric conversion element arrays 5a and 5b, and the relative displacement between the two images is detected using the signal processing circuit, the focus state of the photographing lens 1 can be known.

As a method for processing a photoelectrically converted signal, for example, a method disclosed in US Pat. No. 4,250,376 is known. This is because when the image signals output from the two photoelectric conversion element arrays are A (i) and B (i) (where i = 1 to N), in the above example, for an appropriate constant k, Is calculated by an analog arithmetic circuit or digitally, and the moving direction of the photographing lens is determined by the sign of the value of the correlation amount V.

In addition, the applicant of the present application Or, Is calculated, and the moving direction of the photographing lens is determined based on whether the correlation amount V is positive or negative. (JP-A-59-107313
Further, in a focus detection device that determines the in-focus state based on the shift between two images, one image is relative to the other image using a relationship in which the shift amount of the two images is substantially proportional to the defocus amount. There is known a method of calculating the amount of movement of a photographing lens by displacing the imaging lens. This method has been known for a long time from a baseline distance meter type focus detection device. TT
An L-type focus detection device is also known from Japanese Patent Application Laid-Open Nos. 56-75607 and 57-45510.

For example, the image signal represented by B (i) is relatively moved in circuit processing with respect to the image signal represented by A (i), and V (m) = Σ | A (i) −B ( i + 1−m) | −Σ | A (i + 1) −B (im) | (4) The correlation amount V (m) is set in the range m ₁ ≦ m ≦ m ₂ of the set relative displacement amount m. The operation is repeated for each integer value. FIG. 10 is a diagram in which the value of the correlation amount V (m) is plotted against the relative displacement amount m. When the two images coincide with each other, the correlation amount V (m) should be 0. Therefore, in FIG. 10, an image shift amount corresponding to 1.5 bits exists.

Further, using the equation (2) (or the equation (3)), the correlation amount V (m) is calculated by the following equation for each integer value of the range m ₁ ≦ m ≦ m ₂ of the relative displacement amount m set similarly. By counting, the image shift amount can be obtained.

V (m) = {min ｛A (i), B (i + k−m)} − {min ｛A (i + k), B (im)} (5) However, these operations are performed on two identical images. Phase shift can be accurately detected when only the phase is shifted, but if there is a frequency component equal to or higher than the Nyquist frequency determined by the pitch of the photoelectric conversion element, or a low frequency component generated due to the optical system and shooting conditions. Causes an error in the correlation operation. U.S. Pat. No. 4,561,749 discloses a method of performing a filtering process to remove a frequency component that causes these error factors.

A method of determining a filter to be used when a plurality of filters are used in this conventional example will be described with reference to FIG. First, photoelectric conversion output accumulation (# 1) is performed to obtain an image signal. At first, a correlation operation (# 2) is performed using a signal that has not been subjected to the filtering process. Next, calculation of reliability (# 3)
Thus, the amount of noise or ghost included in the image signal is quantified based on the contrast of the image, the degree of coincidence between the two images, and the like. This reliability value is compared with a predetermined reliability threshold value (#
4) If sufficient reliability is obtained, focus determination is performed (# 5), lens driving is performed at an out-of-focus point, and release is permitted at the time of focusing.

If it is determined in step # 4 that the positive confidence value is lower than the reliability threshold value, the filtering process 1 (#
6), and the same correlation calculation, reliability calculation and reliability determination as in # 2 to # 4 are performed in # 7 to # 9. Similarly, if the reliability value is lower than the reliability threshold value, filtering processing 2 (# 10) by the second filter is performed, and the same correlation calculation, reliability calculation and reliability determination as in # 2 to # 4 are performed in # 11 to # 4. Do it again in # 13. Finally, if it is determined in step # 13 that the reliability value is lower than the reliability threshold, it is determined that distance measurement is impossible. Here, the filtering process 2 may be performed on the image signal after the filtering process 1 is performed, or may be performed on the original image signal on which the filtering process 1 is not performed.

As described above, besides the method of processing by sequentially switching the filter, there is a method of switching the filter depending on the type of the photographing lens or switching in the previous defocus state.

As described above, in the conventional focus detection device having a plurality of filters for an image signal, the filtering selection is to sequentially switch the filters until a good result is obtained. According to this method, the filtering process, the correlation operation, and the reliability judgment must be repeated many times. In addition, it may be necessary to start over from the accumulation of the photoelectric conversion output due to system restrictions, processing methods, and the like, and the responsiveness of the focus detection calculation will be extremely poor.

Further, in the method of switching the filter in advance depending on the type of the photographing lens, the previous defocus state, and the like, the effects of the subject state, noise, ghost, and the like are not considered, and the optimal filtering process is not always performed. .

(Object of the Invention) An object of the present invention is to solve the above-mentioned problems, improve the responsiveness when performing filtering processing of an image signal, and perform appropriate filtering selection in consideration of noise, ghost, and the like. The object of the present invention is to provide a focus detection device capable of performing the following.

(Features of the Invention) In order to achieve the above object, according to the first aspect of the present invention, the first and second light receiving sections each including a plurality of photoelectric elements receive light beams having optical parallax. A focus detection apparatus comprising: a shift calculating unit configured to calculate a shift amount based on an image signal from each photoelectric element of the first light receiving unit and an image signal from each photoelectric element of the second light receiving unit; First filter processing means for performing first filter processing to match characteristics between image signals input from the first light receiving unit and the image signal input from the second light receiving unit; A second filter different from the first filter processing in order to match the characteristics between the image signals input from the respective light receiving units.
A plurality of filter processing means including a second filter processing means for performing the filter processing described above, and an image signal from the first light receiving portion and an image signal from the second light receiving portion before the calculation of the shift amount. Determining means for determining which one of the plurality of filtering means is to be used in accordance with the determined relationship between image signals; and performing a filtering process in accordance with a result determined by the determining means. In this case, a shift operation is performed on the image signal.

In addition, according to the present invention, the first and second light receiving units each including a plurality of photoelectric elements receive light beams having optical parallax, and each of the photoelectric elements of the first light receiving unit. A focus detecting device provided with a shift calculating means for calculating a shift amount based on an image signal from the second light receiving unit and an image signal from each photoelectric element of the second light receiving unit, wherein the first light receiving unit and the second light receiving unit Filter processing means for performing filter processing in order to match characteristics between image signals respectively input from the image signal from the first light receiving unit and the second light receiving unit before calculating the shift amount. And deciding means for deciding whether or not to perform the filtering by the filtering means in accordance with the relationship between the image signals determined by the image signals. Image signal or file It is characterized in carrying out the shift operation on the image signal in a state in which data processing is not performed.

(Embodiment of the Invention) Fig. 1 is a block diagram showing a main part of a focus detection device according to an embodiment of the present invention.

Reference numeral 10 denotes a sensor device having a plurality of storage-type photoelectric conversion element arrays 10a and 10b each including a CCD sensor array or the like. The photoelectric conversion element arrays 10a and 10b receive light beams that have passed through a plurality of regions having different exit pupils of the photographing lens, similarly to the photoelectric conversion element arrays 5a and 5b shown in FIG. Reference numerals 11 and 12 denote filtering means for filtering two image signals output from the sensor device 10. 13 is a signal processing means for detecting the focus state of the image lens from the relative positional relationship between the two image signals subjected to the filtering processing.

Reference numeral 14 denotes a selection unit that determines which of the filtering units 11 and 12 to use or none of them according to the state of the image signal. An array of photoelectric conversion elements without using image signals
For accumulation control of 10a, 10b, a peak signal extracted from the photoelectric conversion element array 10a, 10b or output from a separately provided photoelectric sensor, or provided in the sensor device 10 exclusively for filtering control. It is also possible to use a photoelectric conversion signal output from the provided photoelectric sensor. In practice, the filtering means 11 and 12, the signal processing means 13 and the selecting means 14 are constituted by one microcomputer.

The operation of the embodiment shown in FIG. 1 will be described with reference to the flowchart of FIG.

In step # 20, the signal processing means 13 instructs the sensor device 10 to start accumulation, so that the sensor device 10 has a photoelectric conversion element array 10a,
The 10b photoelectric conversion output is stored. When accumulation is completed, # 2
In 1, the signal processing means 13 instructs the selection means 14 to perform a selection operation. The selecting means 14 takes in the image signal by the sensor device 10 and performs a filtering selection operation. In # 22, no filtering processing, one of the filtering processing 1 and the filtering processing 2 is selected. Details of the selection operation will be described later. If no filtering is selected, the image signal is directly input to the signal processing means 13, and the signal processing means 13 performs a correlation operation (# 23). When the filtering process is performed, the range of i becomes narrow, and the range of the detectable defocus amount becomes small. Furthermore, the focus detection accuracy may be reduced depending on the state of the image signal, and it may be desirable not to perform the filtering process.

When the filtering process 1 is selected, the image signal is input to the filtering means 11 and the filtering process 1 is performed.
(# 24) is performed. The filtering process 1 requires a relatively short calculation time because the number of pixels used in the filtering operation expression is small.

When the filtering process 2 is selected, the image signal is input to the filtering unit 12 and the filtering process 2 is performed.
(# 25) is performed. The filtering process 2 requires a relatively large number of pixels to be used in the filtering operation expression, so that the operation time is relatively long. However, the amount of noise and ghost that can be removed is superior to the filtering process 1. After each process, the signal processing means 13 performs a correlation operation in # 23, and
Perform the reliability calculation, determine the reliability in # 27, determine the focus in # 28, and perform the following processing.

Next, what filtering means is selected when the image signal is in a state will be described with reference to FIG.

3 (a), 3 (b) and 3 (d) show image signals that have not been subjected to filtering processing. (A) is a good image signal without noise, ghost, and level difference between two images.
(B) is an image signal in which a level difference has occurred between two images, and (d) is an image signal in which a ghost having an opposite inclination has occurred between the two images. If correlation calculation is performed without filtering, accurate deviation detection can be performed by (a)
However, (b) and (d) cause errors in the calculation.
For example, filtering means 11 for image signal A (i)
A ′ (i) = − A (i−3) / 4−A (i−2) / 2−A
(I-1) + A (i) + A (i + 1) / 2 + A (i + 2) / 4 (6) The filtering means 12 is set to A "(i) =-0.0625A (i-5) -0.125A (i-4) -0.5625A (i-3) + 0.125A (i-1) + 1.25A (i) + 0.125A (i + 1) -0.5625A (i + 3) -0.125A (i + 4) -0.0625A (i + 5) ... (7) If the same is applied to the image signal (i), the impulse response of the filtering means 11 is as shown in FIG. 4A, and the impulse response of the filtering means 12 is as shown in FIG. The impulse response in Fig. 4 (a) is anti-symmetric with respect to i and (i-1). A feature of this filtering means is that A
The difference (differential) between (i) and A (i-1) is calculated, and the weighting is reduced for the outside of the adjacent pixels to further increase the certainty, and the difference between A (i + 1) and A (i-2) is obtained. And A (i
By obtaining the difference between (+2) and A (i-3), the DC component can be removed, that is, even if there is a level difference between the two images, it can be removed.

The impulse response in FIG. 4 (b) is symmetrical about i. Hereinafter, the filtering means having a symmetrical impulse response will be referred to as symmetrical filtering means. In particular, the symmetric filtering means is considered so as to exhibit characteristics equivalent to performing the antisymmetric filtering processing twice. For this reason, the symmetrical filtering means 12 works effectively even for an image signal containing a ghost image in which a level difference remains even through the antisymmetrical filtering means 11.

For example, if the processing by the filtering means 11 is performed on an image signal including a level difference as shown in FIG.
An image signal which does not include a level difference as shown in FIG. 3 (c) is obtained. However, even if an image signal including a ghost is processed by the filtering means 11 as shown in FIG. 3 (d), a level difference remains as shown in FIG. 3 (e). Focus detection cannot be performed. In the case of an image signal as shown in FIG. 3 (d), if the image signal is processed by the filtering means 12, it becomes as shown in FIG. 3 (f), and the level difference disappears. In the image signal that has been subjected to the symmetric filtering by the filtering unit 12, noise is somewhat noticeable,
The error is smaller than the focus detection using the image signals shown in FIGS. 3 (d) and (e).

That is, in the case of an image signal having neither a level difference nor a ghost,
If no filtering is selected, the filtering means 11 is selected for an image signal having only a level difference, and the filtering means 12 is selected for an image signal containing a ghost.
Is selected and the correlation calculation is performed, the detection error is the least.

 Next, the operation of the selection means 14 will be described.

(I) When the two images have the same shape as shown in FIG. 3A and only the phase difference is shifted. (II) The two images have the same shape as shown in FIG. When there is a level difference in addition to the phase difference (III) As shown in FIG.
The selecting means 14 determines the above three cases where the image level is high and the image A is high at the right edge. The degree of coincidence of the two images is used for these determinations. Regarding the two coincidences, first, a shift amount of two images is detected, one image signal is shifted by the shift amount, and the coincidence is calculated at an image position where the shift becomes zero.

The shift amount can usually be calculated by the above equation (4) or (5). However, if the amount of deviation is calculated by this method, it takes time to select the filtering, and if no filtering is performed, the same calculation is performed again.

Therefore, in order to detect the shift amount, the maximum value of the image signal,
Use the minimum value. This method will be described with reference to the flowchart of FIG.

In # 30, the maximum value A (i max) and the minimum value A (i min) of the A image
Ask for. The positions indicating them are defined as i max and i min, and the difference is defined as Δi. In step # 31, the same calculation is performed for the B image. When the extreme values correspond to each other, the difference Δi and the difference Δj become equal as shown in FIG. 6A (# 3).
2) The shift amount p between the two images is obtained by p = imax-jmax (# 33). However, as shown in FIG. 6 (b), when the extreme values do not correspond, the difference Δi is not equal to the difference Δj. In this case, in # 34, it is searched for a local maximum value B (j′max) near a position (jmin + Δi) separated from the minimum value position jmin of the B image by a difference Δi. Furthermore, the difference Δ
A maximum value A near a position (i min + Δi) separated by j
Search for (i'max). If there is a local maximum value, the process branches to # 36 at # 35. If not, the position (jmax-Δ) separated by −Δi from the maximum value position jmax of the B image in # 37
It is searched for a local minimum B (j'min) near i). Further, it is searched for a local minimum value A (i'min) near a position (imax-.DELTA.j) apart from the maximum value position imax of the A image by -.DELTA.j. If either of the local minima is found in # 38, the flow branches to # 39. If not, it is determined that the coincidence between the two images is very poor, and the filtering means 12 is selected in # 40. # 36
Then, using the newly obtained maximum value position j'max or i'max, the shift amount p is calculated as p = imax-j'max or p = i'max-j
It is calculated by a formula that can calculate max. Similarly in # 39,
The shift amount p is calculated by a formula that can be calculated as p = i min−j min or p = i ′ min−j min. Thus, the shift amount p
Can be easily obtained.

When the shift amount p is obtained, the coincidence between the two is determined by the coincidence determination values S ₁ and S ₂ calculated by the following equations.

That is, if the two images match, both of the coincidence determination values S ₁ and S ₂ become 0, and if there is a level difference, the coincidence determination value S
₁ and S _{2 both} increase. If the ghost is generated, only the coincidence determination value S ₁ is increased, the coincidence determination value S ₂ is a small value.

Therefore, filtering selection is performed as follows using the matching degree determination values S ₁ and S ₂ .

S ₁ ≒ 0, S ₂ ≒ 0 No filtering processing S ₁ = large, S
₂ = large filtering means 11 (S ₁ 1S ₂ ) S ₁ = large, S ₂ = small filtering means 12 (S ₁ >
S ₂ ) Whether the coincidence determination values S ₁ and S ₂ are 0, large or small is determined by setting a certain threshold.

As described above, since the optimum filtering selection is performed first before performing the correlation operation, it is not necessary to repeat useless correlation operations before selecting the optimum filtering means.
Responsiveness can be improved. In addition, since the filtering selection is determined according to the state of the image signal (or according to the state of the photoelectric conversion signal of the light beam that has passed through the objective lens), appropriate filtering selection considering noise, ghost, and the like is performed. be able to. Therefore, the focus detection accuracy can be generally improved.

As another method of obtaining the shift amount p, it is possible to reduce the calculation time by thinning out the image signal and shorten the calculation time. It is publicly known (JP-A-59-107312) that the calculation time of the correlation operation itself for focus detection is shortened by this method.
This method is used for a part of the filtering selection operation. A method for calculating the amount of shift due to the thinned image signal will be described below.

The calculation of the shift amount using the procedure of calculating the correlation amount V (m) in the above equation (4) or (5) for different relative displacement amounts m requires a significantly long calculation time when the number of data N is large. Become. For example, in order to calculate the correlation amount V (m) of the equation (4) for one m, the difference in the absolute value code is 2N, the sum at the time of integration is 2N, and a total of 4N additions and subtractions are required. . On the other hand, since about ± N / 2 is obtained at the upper and lower limit values m ₁ and m _{2 of} m, the number of V (m) to be calculated is almost N. Thus, all the processing number is the subtraction of approximately 4N ² in approximate.
When the calculation is performed with the data number N reduced, for example, the seventh
An image signal A (i) of the original data as shown in FIG.
As shown in FIGS. 7 (a) and 7 (c), the number of data used is reduced. The image signal A '(i) in FIG. 7 (b) is obtained by selecting every other original data, and the image signal A "(i) in FIG. 7 (c) is the photoelectric conversion of two adjacent pixels. In terms of mathematical expressions, the image signal of the original data obtained by the photoelectric conversion element arrays 10a and 10b is represented by A.
(I) and B (i) (B (i) is not shown). In the case of FIG. 7 (b), the value obtained by the following equation is used as calculation data.

A '(i) = A (2i-1) (11) B' (i) = B (2i-1) (12) where i = 1 to N / 2 In the case of FIG. Equations (13) and (14) are used.

A ″ (i) = A (2i−1) + A (2i) (13) B ″ (i) = B (2i−1) + B (2i) (14) Equations (11) to (14) Conversion from original data with 1
A ′ (i) = A (2i), A ″ (i)
= A (2i) + A '(2i + 1) and so on. Since the number of data N is halved by these operations,
The amount of calculation for obtaining the shift amount is reduced to approximately 1/4.

By using the value of m when the correlation amount V (m) calculated as described above becomes closest to 0 as the deviation amount p, the above-described coincidence determination values S ₁ and S ₂ can be calculated. At this time,
If the calculation of the equations (8) and (9) is performed with the reduced data, the amount of calculation can be further reduced.

(Modification) In the illustrated embodiment, the filtering means 11 is an anti-symmetric type, and the filtering means 12 is a symmetric type. However, the present invention is not limited to this, and both may be anti-symmetric types. May be symmetrical. Further, the number of filtering means may be three or more. The filtering means may be an electric analog filter.

(Effect of the Invention) As described above, according to the first aspect of the present invention, before the calculation of the shift amount, the shift amount is determined by the image signal from the first light receiving unit and the image signal from the second light receiving unit. According to the relationship between the image signals, which of the plurality of filter processing units used for matching the characteristics between the image signals is determined, so that the image signal is filtered. Response can be improved,
Appropriate filter processing selection can be performed in consideration of noise, ghost, and the like.

Further, according to the second aspect of the present invention, before the calculation of the shift amount, according to the relationship between the image signal from the first light receiving unit and the image signal determined by the image signal from the second light receiving unit, Since it is determined whether or not to perform the processing by the filter processing means used to match the characteristics between the image signals, it is possible to improve the responsiveness when performing the image signal filter processing, and to improve the noise and the ghost. It is possible to select an appropriate filtering process in consideration of the above.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention, FIG. 2 is a flowchart showing the operation of one embodiment of the present invention, FIG. 3 is a waveform diagram showing an image signal, and FIG. The figure which shows the impulse response of the filtering means which concerns on an Example, 5th.
FIG. 6 is a flow chart showing the operation of the selecting means according to one embodiment of the present invention, FIG. 6 is a waveform diagram showing the operation of the selecting means according to the shift of the image signal, and FIG. 7 is used in the embodiment of the present invention. FIG. 8 is a schematic diagram showing an optical system of a conventional focus detection device, FIG. 9 is a diagram showing the principle of image shift in a conventional focus detection device, and FIG. FIG. 11 is a diagram showing a correlation calculation value by a conventional focus detection device. FIG. 11 is a flowchart showing a filtering selection operation of the conventional focus detection device. 1 ... photographic lens, 3 ... field lens, 4a, 4b ...
... Secondary imaging lenses, 5a, 5b ... Photoelectric conversion element arrays, 6a, 6b ...
... Aperture, 10 ... Sensor device, 10a, 10b ... Photoelectric conversion element array, 11,12 ... Filtering means, 13 ... Signal processing means, 14 ... Selection means.

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Akira Ishizaki 770 Shimonoge, Takatsu-ku, Kawasaki City, Kanagawa Prefecture Inside the Tamagawa Works of Canon Inc. (56) References JP-A-58-52607 (JP, A) JP-A-61- 105520 (JP, A) JP-A-2-15223 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G02B 7/00

Claims (2)

(57) [Claims] An optical signal having optical parallax is received by first and second light receiving portions each comprising a plurality of photoelectric elements, and an image signal from each of the photoelectric elements of the first light receiving section and the image signal from the first light receiving section. In a focus detection device provided with a shift calculating unit that calculates a shift amount based on an image signal from each photoelectric element of the second light receiving unit, the image input from the first light receiving unit and the image input from the second light receiving unit, respectively. First filter processing means for performing a first filter process for matching characteristics between signals, and matching characteristics between image signals respectively input from the first light receiving unit and the second light receiving unit. A plurality of filter processing units including a second filter processing unit that performs a second filter process different from the first filter process; and an image signal from the first light receiving unit before the calculation of the shift amount. According to the image signal from the second light receiving unit Determining means for determining which one of the plurality of filtering means is to be used in accordance with the determined relationship between image signals; and performing a filtering process in accordance with a result determined by the determining means. A focus detection device for calculating a shift with respect to a shifted image signal. 2. A method according to claim 1, wherein the first and second light receiving sections each comprising a plurality of photoelectric elements receive a light beam having an optical parallax, and an image signal from each of the photoelectric elements of the first light receiving section and the image signal. In a focus detection device provided with a shift calculating unit that calculates a shift amount based on an image signal from each photoelectric element of the second light receiving unit, the image input from the first light receiving unit and the image input from the second light receiving unit, respectively. Filter processing means for performing filter processing for matching characteristics between signals, and, before calculating the shift amount, determined by an image signal from the first light receiving unit and an image signal from the second light receiving unit. Determining means for determining whether or not to perform the filtering by the filtering means in accordance with the relationship between the image signals, wherein the image signal or the filter which has been subjected to the filtering in accordance with the determination result by the determining means is provided. No action is taken A focus detection device for calculating a shift with respect to an image signal in an inoperative state.