JPH06148509A - Line-of-sight detector - Google Patents

Abstract

PURPOSE:To accurately extract a pupil edge at all time. CONSTITUTION:A pupil edge extracting means is provided with a first memory means for storing the lowest brightness value in every area which is obtained in a light-intercepting means 4, a second memory means for storing the lowest brightness value in the whole area which is obtained in the light-intercepting means and a pupil edge discriminating means for comparing the lowest brightness value in every area with the loft brightness value in the whole area, respectively, and when the loft brightness value in every area is not the same as the lowest brightness value in the whole area, eliminating the edge in the area by reason that it is inappropriate for a pupil edge, and the edge within the area having the loft brightness value which is not equal to the loft brightness value (higher than the lowest brightness value) in the whole area corresponding to the lowest brightness value of a pupil edge is eliminated by reason that it is inappropriate for a pupil edge.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement of a visual axis detecting device arranged in a video camera or the like.

[0002]

2. Description of the Related Art A conventional method for calculating the position of a Purkinje image (corneal reflection image) is disclosed in JP-A-61-272552. This JP-A-61-272552 describes as follows.

FIG. 16 shows a state in which an observer's pupil 50 and an iris 51 are imaged on a sensor array portion 31a of one semiconductor one-dimensional image pickup device (here, 31), and FIG. 17 shows a semiconductor one-dimensional image at that time. Sensor array unit 31 of the image pickup device 31
The distribution of the brightness of the image on a is shown.

Since the iris 51 has a higher reflectance than the pupil 50, it becomes bright as an image. In FIG. 17, points 53 and 54 at which the brightness greatly changes are the pupils 50 in FIG.
And a boundary 52 of the iris 51 corresponds to two points 53a and 54a intersecting the sensor array portion 31a. Reference numeral 55 shown in FIG. 17 is an output corresponding to the Purkinje image PI.

For example, data digitized by an A / D converter for each pixel is used as the semiconductor one-dimensional image pickup device 31.
If the points at which the light level 56 first changes to the dark level 57 are found respectively, the boundary 52 between the pupil 50 and the iris 51 is determined to be the sensor array unit 3
Two points 53a and 54a intersecting 1a can be obtained. Then, the center of the photographer's pupil is obtained by taking the average of these two intersecting points 53a and 54a.

[0006]

However, the actual signal waveforms obtained from the semiconductor one-dimensional image pickup device are shown in FIG.
As shown in FIG. 21, it is difficult to accurately extract the pupil edge in the conventional method that simply finds the first point where the light level changes to the dark level.

Further, it is often the case that the Purkinje image is outside the pupil, and that the pupil is eclipsed by the eyelid in the detection in the vertical direction. Even in such a case, the initial change point from the bright level to the dark level does not become the pupil edge, and the center of the pupil cannot be calculated by the conventional method.

(Object of the Invention) It is an object of the present invention to provide a visual axis detecting device capable of always accurately extracting a pupil edge.

[0009]

According to the present invention, the pupil edge extraction means has a first storage means for storing the lowest luminance value for each area obtained by the light receiving means and all the light receiving means. The second storage means for storing the lowest luminance value in the area is compared with the lowest luminance value for each area and the lowest luminance value in all areas, and the lowest luminance value in each area is the lowest in all areas. When the brightness value is not equal to the edge, an edge in the area is provided as a pupil edge discriminating means that is excluded as an inappropriate pupil edge,
Not equal to the minimum brightness value in all areas corresponding to the minimum brightness value of the pupil edge (higher than this minimum brightness value)
The edge in the area having the lowest brightness value is excluded as an inappropriate pupil edge.

Further, according to the present invention, it is determined whether or not there is an edge in the periphery of the Purkinje image extracted by the Purkinje image extracting unit in the pupil edge extracting unit, and if there is, this edge is the pupil edge. A pupil edge discriminating means is provided to eliminate the pupil edge as an inappropriate one, and it is clear that the pupil edge is not located around the Purkinje image. Therefore, the edges existing around the Purkinje image are inappropriate as pupil edges. I'm trying to exclude it.

Further, according to the present invention, it is determined in the pupil edge extraction means whether or not the positions of the respective extracted edges and the Purkinje image extracted by the Purkinje image position extraction means have a predetermined positional relationship, and the Purkinje image is extracted. An edge that does not have a predetermined positional relationship with the image is excluded as an inappropriate pupil edge, and a pupil edge discriminating unit is provided. Since the pupil edge and the Purkinje image have a predetermined positional relationship, an edge that does not have this predetermined relationship is The pupil edge is excluded as an inappropriate one.

In the pupil edge extraction means, the extracted edges are statistically processed, that is, the average value and standard deviation of the extracted edges are obtained, and those outside the range defined by these values are extracted as pupil edges. A pupil edge discriminating means for eliminating inappropriate edges is provided as the pupil edge, and a pupil edge within a range defined by the average value and standard deviation of a large number of edges including the pupil edge is regarded as a pupil edge, and a pupil edge is outside the range. As an inappropriate edge, it is excluded.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below based on the illustrated embodiments.

FIG. 1 is a block diagram showing the schematic arrangement of a camera equipped with a line-of-sight detection apparatus according to the first embodiment of the present invention.

In FIG. 1, 1 is an MPU (micro processing unit), 2 is a memory, and 3 is a CCD described later.
And a driver circuit for driving an IRED (infrared light emitting diode), 4 is a CC which is a two-dimensional image sensor
D and 5 are IRED groups composed of a plurality of IREDs that project infrared light that is insensitive to the observer's eyes, and are arranged in pairs in a horizontal / vertical direction at predetermined intervals, that is, in pairs. Has been done. Reference numeral 6 is a lens drive unit for performing AF (autofocus), 7 is an aperture drive unit, and 8 is a shutter unit.

FIG. 2 is a flow chart showing the main operation of the camera provided with the visual axis detection device having the above-mentioned structure.

When a line-of-sight detection request is made by turning on the switch SW1 interlocking with the first stroke operation of the release button (not shown) of the camera, the MPU 1 makes a step 102
The routine for detecting the line of sight from is entered.

In step 102, initialization processing of variables used for calculation is performed, and in the next step 103, accumulation time is set (step 103). This is done in consideration of the presence / absence of glasses and the intensity of external light, and at the same time, IR
The IRED pair is also selected to be turned on from within the ED group 5. Then, the process proceeds to the storage control operation after step 104.

First, in step 104, the CCD
The driver circuit 3 gives an instruction for performing the charge clearing operation of No. 4
Do against. Upon receiving this instruction, the driver circuit 3 performs a clear operation to erase the charges remaining in the memory zone of the CCD 4, the charge transfer line, and the like. Then step 1
05, the I selected in step 103 above
At the same time as transmitting the IRED selection signal to the driver circuit 3 to turn on the RED pair, the accumulation signal is set to the high level to start accumulation by the CCD 4, and when the above set accumulation time has elapsed, the accumulation signal is set to the low level and accumulation is performed. To finish. The IRED pair selected in synchronization with this accumulation is turned on.

Next, the MPU 1 proceeds to the operations after step 106 for performing the processing of extracting the optical image block (Purkinje image candidate) and the pupil edge candidate.

First, in step 106, the image signal for one line of the CCD 4 is sequentially read through the driver circuit 3, A / D converted, and the value is stored in the memory 2. Then, in the next step 107, the optical image block (Purkinje image candidate) and the pupil edge candidate are extracted using this data. This processing is performed for the number of lines of the CCD 4.

The "extraction of the pupil edge" performed in step 107 will be described with reference to the flowcharts of FIGS. 3 and 4.

First, in step 151, the MPU 1 reads the data for one line, and in step 15
In step 2, the minimum value MINL on this line and its coordinates are obtained from this data. Then, this "minimum value MINL" is compared with the "minimum value MIN + constant C" in the lines that have been read so far, and whether this "minimum value MINL" is smaller than the "minimum value MIN + constant C". If the value is substantially equal to this value, the processes in and after step 154 are continued, but if not, the extraction of the horizontal pupil edge in this line is not performed, and the process proceeds to the operation in step 169 and later described later in FIG.

The extraction of the pupil edge is performed by first extracting the pupil edge in the horizontal direction.

The minimum value MIN is used to extract the pupil edge in the horizontal direction.
The process starts from the L coordinate (step 154).

First, in step 155, the counter is decremented to detect the left pupil edge. Then, it is checked whether or not the output signal monotonically decreases over n pixels. That is, it is checked whether the following equations d [j] <d [j-1] <d [j-2] <... <d [j- (n-1)] <d [j-n] are satisfied. . At this time, if the monotonous decrease continues, the count is continued during that time and the slope length D is obtained.

Next, at step 156, whether the minimum value d [j] of the slope is substantially equal to the minimum value MIN of the line read so far, that is, "d [j] ≤MIN + constant C". ”Is established. Furthermore, it is checked whether or not the difference between the pixels in all the slopes and the adjacent pixels is a certain value Cde or more. For example, d [j-n] -d [j], d [j-n] -d [j-
(N-1)] ... By doing so, it is possible to detect a slope (a slope of the pupil edge) which rises from the pupil portion where the signal intensity is substantially equal to the minimum value, with an inclination of a certain value or more.

Next, in steps 157 to 160, processing for removing the influence of the luminance change due to the eyelashes is performed.

The change in luminance due to the eyelashes is as shown in FIG. Therefore, by checking whether or not there is a down edge B paired with an up edge A immediately after that, it can be determined whether or not the detected edge is eyelashes.

If the above three conditions are satisfied, the vicinity of the pixel (j-2D) (for example, j-2D ± several pixels) is examined,
If there is a value that is approximately equal to the minimum value (MIN), it is determined that this luminance change is due to the eyelashes.

Those which satisfy the above four conditions are regarded as pupil edges, and in step 161, the line number 1 and the pixel number (j
+ J−D) / 2 is stored in the memory 2.

After that, the MPU 1 proceeds to the routine after step 163 for detecting the pupil edge on the opposite side (right side).

On the contrary, if the above four conditions are not satisfied, the process returns from step 160 to step 158 to move to the adjacent pixel and perform the same processing.

Similar processing is performed in the routine (steps 163 to 167) for detecting the pupil edge on the opposite side (right side).

That is, (1) The value increases monotonically over n pixels (the length D of the slope is obtained).

(2) d [j] which is the minimum value of the slope
Is approximately equal to the minimum value (MIN).

(3) The difference between the pixels in all the slopes and the adjacent pixels is not less than the constant value Cde.

(4) There is no down edge immediately after the up edge (not a change in brightness due to eyelashes). When the four conditions are satisfied, the position is regarded as the pupil edge, the line number L and the pixel number (j + j + D) are stored in the memory 2 as the right pupil edge information, and the horizontal pupil edge extraction processing is completed.

On the contrary, when the above four conditions are not satisfied, the processing is moved to the adjacent pixel.

Next, the vertical pupil edge shown in FIG. 4 is extracted.

Since this processing cannot use the information of all pixels due to the capacity of the memory 2, the image information for m lines is stored in the memory 2 and extracted.

First, the minimum value MINL of the line and the minimum value MIN (Lm) of the line (Lm) before the mth line are compared, and "MINL <MIN (Lm)", and If the minimum value MINL is smaller than or substantially equal to the minimum value MIN among the lines read so far, upper pupil edge extraction processing is performed (step 169 → 170 →
171), or “MIN ≧ MIN (L-m)”, and a value in which the minimum value MIN (L-m) is less than or substantially equal to the minimum value MIN in the lines read so far. If so, lower pupil edge extraction processing is performed (step 1
69 → 174 → 175). Otherwise, the vertical pupil edge extraction processing is not performed (steps 169 → 170 → 17).
8, steps 169 → 174 → 178).

The upper pupil edge extraction processing is performed by checking whether or not the same three conditions as those in the horizontal direction are satisfied (steps 171 and 172). That is, (1) it decreases monotonically over n pixels (however, due to the memory capacity, if n> m, it monotonically decreases over m pixels).

(2) Line L, which is the minimum slope value
Is approximately equal to the minimum value (MIN).

(3) The difference between the pixels in all the slopes and the adjacent pixels is not less than the constant value Cde. If the three conditions are satisfied, the position is regarded as the pupil edge, and the line number (L
+ 1-D) [or (L + 1-m)] and the pixel number j are stored in the memory 2. In the case of n <m, the slope length D is obtained similarly to the extraction of the pupil edge in the horizontal direction.

Similarly, the lower pupil edge extraction processing is performed by checking whether or not the three conditions are satisfied (steps 175 and 176). That is, (1) it monotonically decreases over n pixels (however, in the case of n> m, it monotonically increases over m pixels due to the memory capacity).

(2) Line L, which is the minimum slope value
The value of -D (or line L-m) is approximately equal to the minimum value (MIN).

(3) The difference between the pixels in all the slopes and the adjacent pixels is not less than the constant value Cde. If the three conditions are satisfied, the position is regarded as the pupil edge, and the line number (L
+ 1-D) [or (L + 1-m)] and the pixel number j are stored in the memory 2. In the case of n <m, the slope length D is obtained similarly to the extraction of the pupil edge in the horizontal direction.

The above processing is performed for all pixels within the effective range of one line.

Finally, the lowest value in this area is stored.

In order to save the memory capacity, the entire screen is divided in the vertical direction and the lowest value is stored in that area. After determining which area the line L corresponds to, the minimum value MINn (n: area number) in that area is compared with the minimum value MINL of the line, and "MINn>
If it is "MINL", "MINn = MINL" is updated.

Returning to FIG. 2 again, when it is determined in step 108 that this process has been completed for all lines, the process proceeds to step 109 and the Purkinje image selection process is started.

In the first processing here, the block widths in the horizontal and vertical directions of each optical image block (Purkinje image candidate) are examined, and only those widths which are equal to or less than the constant CH are adopted. Those having a larger block width than this are excluded because they are optical images other than the Purkinje image, such as ghosts of glasses and reflection of external light from the eyelids.

In the second processing, the vertical coordinate of the optical image block adopted in the first processing is examined. If a certain light image block and a certain light image block overlap or are in contact with each other in the vertical direction, it is determined that the two light image blocks have substantially the same vertical coordinates. That is, the upper, lower, left, and right boundary positions of the n-th optical image block are respectively Bny1, B
Assuming ny2, Bnx1, and Bnx2, the upper, lower, left, and right boundary positions of the m-th optical image block are respectively Bmy1, Bmy.
2, Bmx1 and Bmx2, if the conditional expressions of Bny1 ≦ Bmy2 and Bny2 ≧ Bmy1 are satisfied, it is determined that the vertical coordinates of the two optical image blocks are substantially equal. Then, the light image blocks without the substantially equal vertical coordinates are eliminated. After that, the horizontal block intervals Δmn of the optical image blocks having substantially the same values are examined, and only those having a value within a predetermined range (Δ1 ≦ Δmn ≦ Δ2) are adopted (where Δ1 and Δ2 are constants).

For example, when the three vertical coordinates of the optical image blocks B1, B2, B3 are substantially equal, the interval Δ12 between B1 and B2, the interval Δ23 between B2 and B3, and the interval Δ between B3 and B1.
31 is determined, and it is checked whether it is within a predetermined value. As a result, if Δ12 and Δ23 are within a predetermined value, the combination of [B1, B2] [B2, B3] is adopted as the Purkinje image candidate pair. To do. At this point, if there is one Purkinje image candidate pair (if the number of optical image blocks is two), this is adopted as the Purkinje image. If the Purkinje image cannot be determined at this point, the following processing is performed.

In the third processing, the average values of the maximum luminance values of the Purkinje image pairs are compared, and the one having the maximum value is selected as the Purkinje image.

For example, if the optical image blocks B1 and B2 and B3 and B4 are selected by the second processing, (Ima
x1 + Imax2), (Imax3 + Imax4) are calculated, and if there is a relation of “the former ≧ the latter”, B1, B2
Is a Purkinje image, and if there is a relationship of “the former <latter”, the optical image blocks B3 and B4 are Purkinje images (however, Imax1, Imax2, Imax3, Imax4
Is the maximum brightness of B1, B2, B3, B4).

Alternatively, the average of the maximum luminance values of the Purkinje image pairs may be compared and the one having the maximum value may be selected.

The coordinates of the Purkinje image pairs P1 and P2 and the Purkinje image Pc determined as described above are P1x = (Bnx1 + Bnx2) / 2 P1y = (Bny1 + Bny2) / 2 P2x = (Bmx1 + Bmx2) / 2 P2y = (Bmy1 + Bmy1 + Bmy1) / 2 Pcx = (P1x + P2x) / 2 Pcy = (P1y + Pyx) / 2 However, P1x, P1y, P2x,
P2y, Pcx and Pcy are Purkinje image pairs P1 and P2
And the horizontal and vertical coordinates of the Purkinje image Pc. Also, n
It is assumed that the 1st and mth optical image blocks are selected as the Purkinje image.

Next, the routine proceeds to step 110, where a "pupil circle calculation" operation is performed.

This "pupil circle calculation" operation is performed as follows by steps 201 to 225 shown in FIGS.

In the pupil edge extraction process described above, the MPU
1 stores the extracted coordinates of the pupil edge, the minimum brightness MINn of the area n and the minimum brightness MIN0 of the entire image in the memory 2.

Therefore, first, the MPU 1 compares the minimum luminance (MINn) of the area where the pupil edge is extracted with the minimum luminance (MIN0 + constant C) of the entire image, and when the condition of MINn ≦ MIN0 + constant C is not satisfied, , The pupil edge existing in the area is excluded as an inappropriate one.

Then, the pupil edge around the Purkinje image obtained in the previous step 109 is eliminated. This is in an area centered on the horizontal and vertical coordinates (Pnx, Pny) of the Purkinje image, for example, (Pnx-constant,
Pny-constant), (Pnx-constant, Pny + constant),
(Pnx + constant, Pny−constant), (Pnx + constant, P
It is carried out by excluding those in the quadrangle surrounded by four points (ny + constant). Since there are two Purkinje images, this processing is performed for each Purkinje image.

Next, edges that are considered to be inappropriate due to the positional relationship with the Purkinje image are eliminated.

The relationship between the Purkinje image and the pupil is as shown in FIG. However, in this figure, the position of the Purkinje image is fixed for ease of explanation, but in the actual eye movement, both the Purkinje image and the pupil move. Those that fall within the range of the table below from FIG. 8 can be regarded as proper edges as pupil edges.

[0067] The values of α1, α2, α3, α4 may be variable in consideration of the pupil diameter. In this case, the pupil diameter is estimated from the photometric information used for AE. If the pupil diameter is small (bright), the values of α1, α2, α3, α4 are large, and if the pupil diameter is large (dark), α1, α2. , Α3, α4 are reduced.

Further, the average value m and the standard deviation σ of the coordinates of the pupil edge candidates selected so far are obtained, and [average value m-
a * standard deviation σ-average value m + a * standard deviation σ (however, a
Is a constant)]. This calculation may be performed only in the horizontal direction or may be performed in both the horizontal and vertical directions. Also, this process may be performed twice. That is, the average value m ′ and the standard deviation σ ′ of those within the range of [average value m−a * standard deviation σ−average value m + a * standard deviation σ] are obtained again, and the range defined by these two quantities [average Only values within the range of m′−a * standard deviation σ ′ to average value m ′ + a * standard deviation σ ′] are adopted.

Next, the pupil center and the pupil radius are obtained by using the pupil edges selected as described above. As this method, the least square method may be used.

Then, the process proceeds to step 111, and the rotation angle of the eyeball and the viewpoint position on the focusing plate of the camera are calculated using the Purkinje image and the position of the center of the pupil. And
In step 112, the AF point and the like are determined based on this viewpoint position, and the camera is controlled.

(Second Embodiment) FIGS. 9 and 10 are flow charts showing the operation of the main part of the camera equipped with the visual axis detection device according to the second embodiment of the present invention. The circuit configuration is the same as that of the first embodiment shown in FIG. 1, and the main routine of the camera is the first embodiment shown in FIG. The same circuit number and step number shall be given.

Similar to the first embodiment, the switch SW which is interlocked with the first stroke operation of the release button (not shown) of the camera.
When a line-of-sight detection request is made by turning on 1 etc., MP
U1 enters the visual axis detection routine from step 102.

In step 102, initialization processing of variables used for calculation is performed, and in the next step 103, accumulation time is set (step 103). This is performed in consideration of the presence / absence of glasses, the intensity of external light, and the like, and at the same time, the IRED pair is selected so that the IRED group 5 is turned on. Then, the process proceeds to the storage control operation after step 104.

First, in step 104, the CCD
The driver circuit 3 gives an instruction for performing the charge clearing operation of No. 4
Do against. Upon receiving this instruction, the driver circuit 3 performs a clear operation to erase the charges remaining in the memory zone of the CCD 4, the charge transfer line, and the like. Then step 1
05, the I selected in step 103 above
At the same time as transmitting the IRED selection signal to the driver circuit 3 to turn on the RED pair, the accumulation signal is set to the high level to start accumulation by the CCD 4, and when the above set accumulation time has elapsed, the accumulation signal is set to the low level and accumulation is performed. To finish. The IRED pair selected in synchronization with this accumulation is turned on.

Next, the MPU 1 proceeds to the operation after step 106 for performing the process of extracting the optical image block (Purkinje image candidate) and the pupil edge candidate.

First, in step 106, the image signal for one line of the CCD 4 is sequentially read through the driver circuit 3, A / D converted, and the value is stored in the memory 2. Then, in the next step 107, the optical image block (Purkinje image candidate) and the pupil edge candidate are extracted using this data. This processing is performed for the number of lines of the CCD 4.

Here, the "pupil edge extraction" performed in step 107 will be described with reference to the flowcharts of FIGS. 9 and 10.

First, in step 301, the MPU 1 reads the data for one line, and in step 30
In step 2, the minimum value MINL on this line and its coordinates are obtained from this data. Then, this "minimum value MINL" is compared with the "minimum value MIN + constant C" in the lines that have been read so far, and whether this "minimum value MINL" is smaller than the "minimum value MIN + constant C". If the value is substantially equal to this, the processing from step 304 onward is continued, but if not, the extraction of the horizontal pupil edge in this line is not performed, and the operation proceeds to step 319 or later, which will be described later in FIG.

The extraction of the pupil edge is performed by first extracting the pupil edge in the horizontal direction.

Minimum pupil edge extraction in the horizontal direction is MIN
The process starts from the L coordinate (step 304).

First, in step 305, the counter is decremented to detect the left pupil edge. Then, it is checked whether or not the output signal monotonically decreases over n pixels. That is, it is checked whether the following equations d [j] <d [j-1] <d [j-2] <... <d [j- (n-1)] <d [j-n] are satisfied. . At this time, if the monotonous decrease continues, the count is continued during that time and the slope length D is obtained.

Next, at step 306, whether the minimum value d [j] of the slope is approximately equal to the minimum value MIN of the line read so far, that is, "d [j] ≤MIN + constant C ”Is established. Furthermore, it is checked whether or not the difference between the pixels in all the slopes and the adjacent pixels is a certain value Cde or more. For example, d [j-n] -d [j], d [j-n] -d [j-
(N-1)] ... By doing so, it is possible to detect a slope (a slope of the pupil edge) which rises from the pupil portion where the signal intensity is substantially equal to the minimum value, with an inclination of a certain value or more.

Next, in steps 307 to 310, processing for removing the influence of the luminance change due to the eyelashes is performed.

The change in luminance due to the eyelashes is as shown in FIG. 5 as described above. Therefore, by checking whether or not there is a down edge B paired with an up edge A immediately after that, it can be determined whether or not the detected edge is eyelashes.

If the above three conditions are satisfied, the vicinity of the pixel (j-2D) (for example, j-2D ± several pixels) is examined,
If there is a value that is approximately equal to the minimum value (MIN), it is determined that this luminance change is due to the eyelashes.

Those which satisfy the above four conditions are regarded as pupil edges, and in step 311, the line number 1 and the pixel number (j
+ J−D) / 2 and the lowest brightness at that slope (d [j] in this case) are stored in memory 2.

After that, the MPU 1 proceeds to the routine after step 313 for detecting the pupil edge on the opposite side (right side).

On the contrary, if the above four conditions are not satisfied, the process returns from step 310 to step 308 to move to the adjacent pixel and perform the same processing.

Similar processing is performed in the routine (steps 313 to 317) for detecting the opposite (right) pupil edge.

That is, (1) It is monotonically increasing over n pixels (the length D of the slope is obtained).

(2) d [j] which is the minimum value of the slope
Is approximately equal to the minimum value (MIN).

(3) The difference between the pixels in all slopes and the adjacent pixels is equal to or more than a constant value Cde.

(4) There is no down edge immediately after the up edge (not a change in luminance due to eyelashes). When the four conditions are satisfied, the position is regarded as the pupil edge, the line number 1 and the pixel number (j + j + D) are stored in the memory 2 as the right pupil edge information, and the horizontal pupil edge extraction processing is completed.

On the contrary, when the above four conditions are not satisfied, the processing is moved to the adjacent pixel.

Then, the pupil edge in the vertical direction shown in FIG. 10 is extracted.

Since this processing cannot use the information of all the pixels due to the capacity of the memory 2, the image information for m lines is stored in the memory 2 and extracted.

First, the value of the minimum value MINL of the line and the value of the minimum value MIN (Lm) of the line (Lm) before the mth line are compared, and "MINL <MIN (Lm)", and If the minimum value MINL is less than or substantially equal to the minimum value MIN among the lines read so far, upper pupil edge extraction processing is performed (step 319 → 320 →
321), or “MIN ≧ MIN (L-m)”, and a value in which this minimum value MIN (L-m) is less than or substantially equal to the minimum value MIN in the lines read so far. If so, lower pupil edge extraction processing is performed (step 3).
19 → 324 → 325). Otherwise, the vertical pupil edge extraction process is not performed.

The upper pupil edge extraction processing is performed by checking whether or not three conditions similar to those in the horizontal direction are satisfied (steps 321 and 322). That is, (1) it decreases monotonically over n pixels (however, due to the memory capacity, if n> m, it monotonically decreases over m pixels).

(2) Line L, which is the minimum slope value
Is approximately equal to the minimum value (MIN).

(3) The difference between the pixels in all the slopes and the adjacent pixels is not less than the constant value Cde. If the three conditions are satisfied, the position is regarded as the pupil edge, and the line number (L
+ 1-D) [or (L + 1-m)], the pixel number j, and the minimum luminance in the slope are stored in the memory 2.
In the case of n <m, the slope length D is obtained similarly to the extraction of the pupil edge in the horizontal direction.

Similarly, the lower pupil edge extraction processing is performed by checking whether or not the three conditions are satisfied (steps 325 and 326). That is, (1) it monotonically decreases over n pixels (however, in the case of n> m, it monotonically increases over m pixels due to the memory capacity).

(2) Line L, which is the minimum slope value
The value of -D (or line L-m) is approximately equal to the minimum value (MIN).

(3) The difference between the pixels in all the slopes and the adjacent pixels is not less than the constant value Cde. If the three conditions are satisfied, the position is regarded as the pupil edge, and the line number (L
+ 1-D) [or (L + 1-m)], the pixel number j, and the minimum luminance in the slope are stored in the memory 2.
In the case of n <m, the slope length D is obtained similarly to the extraction of the pupil edge in the horizontal direction.

The above processing is performed for all pixels within the effective range of one line.

Returning to FIG. 2 again, when it is determined in step 108 that this process has been completed for all lines, the process proceeds to step 109, and the Purkinje image selection process is started.

In the first processing here, the block widths in the horizontal and vertical directions of each optical image block are examined, and only those widths of which are equal to or less than the constant C are adopted. Those having a larger block width than this are excluded because they are optical images other than the Purkinje image, such as ghosts of glasses and reflection of external light from the eyelids.

In the second process, the vertical coordinate of the optical image block adopted in the first process is examined. If a certain light image block and a certain light image block overlap or are in contact with each other in the vertical direction, it is determined that the two light image blocks have substantially the same vertical coordinates. That is, the upper, lower, left, and right boundary positions of the n-th optical image block are respectively Bny1, B
Assuming ny2, Bnx1, and Bnx2, the upper, lower, left, and right boundary positions of the n-th optical image block are Bmy1, Bmy, respectively.
2, Bmx1 and Bmx2, if the conditional expressions of Bny1 ≦ Bmy2 and Bny2 ≧ Bmy1 are satisfied, it is determined that the vertical coordinates of the two optical image blocks are substantially equal. Then, the light image blocks without the substantially equal vertical coordinates are eliminated. After that, the horizontal block intervals Δmn of the optical image blocks having substantially the same values are examined, and only those having a value within a predetermined range (Δ1 ≦ Δmn ≦ Δ2) are adopted (where Δ1 and Δ2 are constants). At this point, if there is one Purkinje image candidate pair (if the number of optical image blocks is two), this is adopted as the Purkinje image. If the Purkinje image cannot be determined at this point, the following processing is performed.

In the third processing, the average value of the maximum luminance values of the pair of Purkinje images is first obtained, and then the values are compared, and the largest one is selected as the Purkinje image.

For example, blocks B1 and B2, B3 and B4
Is selected in the second processing, the average value Iave1, of the brightness values of the optical image blocks B1, B2, B3, B4 is
Iave2, Iave3, and Iave4 are calculated by the following equation.

Iave1 = SI [1] / [(B1x2-B1x1 + 1) * (B1y2-B1y2 + 1)] Iave2 = SI [2] / [(B2x2-B2x1 + 1) * (B2y2-B2y2 + 1)] Iave3 = SI [3] / [(B3x2-B3x1 + 1) * (B3y2-B3y2 + 1)] Iave4 = SI [4] / [(B4x2-B4x1 + 1) * (B4y2-B4y2 + 1)] Then, "Iave1 + Iave2" and "Iave3 +".
Iave4 ”, and if there is a relation of“ former ≧ latter ”, B1 and B2 are Purkinje images, and“ former <latter ”
B3 and B4 are Purkinje images.

Further, the average values of the maximum luminance values of the Purkinje image pairs may be compared and the one having the maximum value may be selected.

The coordinates of the Purkinje image pairs P1 and P2 and the Purkinje image Pc determined as described above are P1x = SIx [n] / SI [n] P1y = SIy [n] / SI [n] P2x = SIx [ m] / SI [m] P2y = SIy [m] / SI [m] Pcx = (P1x + P2x) / 2 Pcy = (P1y + Pyx) / 2 However, P1x, P1y, P2x,
P2y, Pcx and Pcy are Purkinje image pairs P1 and P2
And the horizontal and vertical coordinates of the Purkinje image Pc. Also, n
It is assumed that the 1st and mth optical image blocks are selected as the Purkinje image. However, SI [n], SIx [n],
SIy [n] is the sum of the luminance of the n-th block, the sum of the products of luminance and horizontal coordinates, and the sum of the products of luminance and vertical coordinates.

Next, the routine proceeds to step 110, where a "pupil circle calculation" operation is performed.

This "pupil circle calculation" operation is performed as follows by steps 401 to 423 shown in FIG.

In the above-mentioned pupil edge extraction processing, the MPU
Reference numeral 1 stores the coordinates of the extracted pupil edge, the lowest luminance MIN (e) at the edge (slope) and the lowest luminance MIN0 of the entire image in the memory 2 (e is an edge number).

Therefore, first, the MPU 1 compares the minimum luminance (MIN (e)) of the area where the pupil edge is extracted with the minimum luminance (MIN0 + constant C) of the entire image, and MIN (e) ≤MIN0 + constant C If the condition is not satisfied, the pupil edge existing in the area is excluded as an inappropriate one.

Then, the pupil edge around the Purkinje image obtained in the previous step 109 is eliminated. This is in an area centered on the horizontal and vertical coordinates (Pnx, Pny) of the Purkinje image, for example, (Pnx-constant,
Pny-constant), (Pnx-constant, Pny + constant),
(Pnx + constant, Pny−constant), (Pnx + constant, P
It is carried out by excluding those in the quadrangle surrounded by four points (ny + constant). Since there are two Purkinje images, this processing is performed for each Purkinje image.

Next, edges that are considered inappropriate due to the positional relationship with the Purkinje image are eliminated.

The relationship between the Purkinje image and the pupil is as shown in FIG. However, in this figure, the position of the Purkinje image is fixed for ease of explanation, but in the actual eye movement, both the Purkinje image and the pupil move. From FIG. 12, it is possible to consider that those falling within the range shown by the diagonal lines in FIG. 13 are appropriate edges as pupil edges. O is the center position (average position) of the two Purkinje images. Also, β1, β
The values of 2, β3, β4, β5, β6, β7, β8 may be variable in consideration of the pupil diameter. In this case, the pupil diameter is estimated from the photometric information used for AE, and if the pupil diameter is small (when it is bright), β1, β2, β3, β4, β5, β6, β
If the values of 7 and β8 are small and the pupil diameter is large (in the dark), β1, β2, β3, β4, β5, β6, β7, β
Increase the value of 8.

Further, the method of the first embodiment and the method of the second embodiment may be used in combination, and in this case, the pupil edge considered appropriate by both methods is subjected to the following processing. Can be used in.

Further, the average value m and the standard deviation σ of the coordinates of the pupil edge candidates selected so far are obtained, and [average value m−
a * standard deviation σ-average value m + a * standard deviation σ (however, a
Is a constant)]. This calculation may be performed only in the horizontal direction or may be performed in both the horizontal and vertical directions. Also, this process may be performed twice. That is, the average value m ′ and the standard deviation σ ′ of those within the range of [average value m−a * standard deviation σ−average value m + a * standard deviation σ] are obtained again, and the range defined by these two quantities [average Only values within the range of m′−a * standard deviation σ ′ to average value m ′ + a * standard deviation σ ′] are adopted.

Next, the pupil center and the pupil radius are obtained by using the pupil edges selected as described above. As this method, the least square method may be used.

After that, the process proceeds to step 111, and the rotation angle of the eyeball and the viewpoint position on the focus plate of the camera are calculated using the Purkinje image and the position of the center of the pupil. And
In step 112, the AF point and the like are determined based on this viewpoint position, and the camera is controlled.

According to the first and second embodiments described above, in the process of obtaining the positions of a large number of pupil edges using the two-dimensional image sensor, the lowest value in a certain area (or the lowest brightness corresponding to the pupil edge). ) Is stored with the coordinates, and if this value is not approximately equal to the minimum value of the entire image, it is excluded as an inappropriate edge, and there is an edge around the Purkinje image position to be extracted (determined). If there is not a predetermined relationship, it is determined whether there is a predetermined relationship between the Purkinje image and the extracted edge. Exclude this as an inappropriate edge, or statistically process the extracted edges, that is, obtain the average value and standard deviation,
It is determined whether or not it falls within the range defined by these values, and if it is outside this range, it is excluded as an inappropriate edge. Can be extracted and the pupil (pupil center, pupil diameter) can be calculated.

(Third Embodiment) FIGS. 14 and 15 are flowcharts showing the operation of the main part of the camera equipped with the visual axis detection device according to the third embodiment of the present invention, which is different from those of the other embodiments. The point is that a line sensor is used as the light receiving means and a single IERD is turned on for processing. Therefore, the circuit configuration is different from that of the first embodiment shown in FIG. 1 only in that the CCD serves as a line sensor and the IRED group serves as one IRED, and the others are the same. Further, since the main routine of the camera is the first embodiment shown in FIG. 2, the same step numbers will be given in the description of the parts related to this operation.

Similar to the first embodiment, the switch SW interlocked with the first stroke operation of the release button (not shown) of the camera.
When a line-of-sight detection request is made by turning on 1 etc., MP
U1 enters the visual axis detection routine from step 102.

In step 102, the initialization of the variables used in the calculation is performed, and in the next step 103, the accumulation time is set. Then, the process proceeds to the storage control operation after step 104.

First, in step 104, the driver circuit 3 is instructed to perform the charge clearing operation of the line sensor. Upon receiving this instruction, the driver circuit 3 performs a clear operation to erase the charges remaining in the memory zone of the line sensor, the charge transfer line, and the like. Then, in step 105, I
At the same time as transmitting the RED drive signal to the driver circuit 3,
The accumulation signal is set to the high level to start accumulation by the line sensor, and when the accumulation time set above has elapsed, the accumulation signal is set to the low level and the accumulation is completed.

Next, the MPU 1 proceeds to the operations after step 106 for performing the processing of extracting the optical image block (Purkinje image candidate) and the pupil edge candidate.

First, in step 106, the image signal on the line sensor is sequentially read through the driver circuit 3, A / D conversion is performed, and the value is stored in the memory 2. Then, in the next step 107, the optical image block (Purkinje image candidate) and the pupil edge candidate are extracted using this data.

Here, the "extraction of the pupil edge" performed in the above step 107 will be described with reference to the flowchart of FIG.

First, in step 501, the MPU 1 reads the data on the line sensor, and in the next step 502, the minimum value MINL and its position (coordinates) are obtained from this data. Then, in step 503, the sensor is divided in the horizontal direction, and the minimum value in each area is obtained.

After step 504, the extraction of the pupil edge in the horizontal direction is started from the coordinates of the minimum value MINL.

First, in step 505, the counter is decremented to detect the left pupil edge. Then, it is checked whether or not the output signal monotonically decreases over n pixels. That is, it is checked whether the following equations d [j] <d [j-1] <d [j-2] <... <d [j- (n-1)] <d [j-n] are satisfied. . At this time, if the monotonous decrease continues, the count is continued during that time and the slope length D is obtained. Furthermore, it is checked whether or not the difference between the pixels in all the slopes and the adjacent pixels is a certain value Cde or more. For example, d [j-n] -d [j], d [j-n]
Check -d [j- (n-1)] .... By doing so, it is possible to detect a slope (a slope of the pupil edge) which rises from the pupil portion where the signal intensity is substantially equal to the minimum value, with an inclination of a certain value or more.

Next, in step 506, a process of removing the influence of the luminance change due to the eyelashes is performed.

The change in luminance due to the eyelashes is as shown in FIG. 6 as described above. Therefore, by checking whether or not there is a down edge B paired with an up edge A immediately after that, it can be determined whether or not the detected edge is eyelashes.

If the above three conditions are satisfied, the vicinity of the pixel (j-2D) (for example, j-2D ± several pixels) is examined,
If there is a value that is approximately equal to the minimum value (MIN), it is determined that this luminance change is due to the eyelashes.

Those which satisfy the above three conditions are regarded as pupil edges, and in step 507, the pixel number (j + j-D) / 2 is stored in the memory 2 as left pupil edge information.

The MPU 1 performs this processing until the end of the effective range of the line sensor is reached (step 508). After that, the MPU 1 proceeds to the routine after step 509 for detecting the pupil edge on the opposite side (right side).

Similar processing is performed in the routine (steps 509 to 513) for detecting the pupil edge on the opposite side (right side).

That is: (1) It monotonically increases over n pixels (the slope length D is obtained).

(2) Regarding the pixels in all slopes, the difference from the adjacent pixel is equal to or more than the constant value Cde.

(3) There is no down edge immediately after the up edge (not a change in brightness due to eyelashes). If the three conditions are satisfied, the position is regarded as the pupil edge, the line number L and the pixel number (j + j + D) are stored in the memory 2 as the right pupil edge information, and the process is continued until the end of the effective range is reached. To do.

As described above, this processing is performed for all the pixels within the effective range on the line sensor.

Returning to FIG. 2 again, in step 108, the Purkinje image selection processing is performed in the same manner as in the first embodiment.

That is, in the first processing, the block width in the horizontal direction of each optical image block is checked, and only those widths of which are equal to or less than the constant CH are adopted. Objects having a larger block width than this are optical images other than the Purkinje image, such as ghosts of eyeglasses and reflection of external light from the eyelids, and are excluded.
If there is only one Purkinje image candidate at this point, this is adopted as the Purkinje image. If the Purkinje image cannot be determined at this point, the following processing is performed.

In the second processing, the average value of the luminance values of the Purkinje image is first obtained, and then the values are compared, and the largest one is selected as the Purkinje image.

For example, when the blocks B1, B2, and B3 are selected by the above processing, the optical image blocks B1 and B3 are selected.
2, average values Iave1, Iave2, I of the brightness values of B3
ave3 is calculated by the following equation.

Iave1 = SI [1] / (B1x2-B1x1 + 1) Iave2 = SI [2] / (B2x2-B2x1 + 1) Iave3 = SI [3] / (B3x2-B3x1 + 1) Then, Iave1, Iave2, and Iave3 are compared, The one with the highest value is the Purkinje image.

The coordinates of the Purkinje image Pc determined as described above are calculated as Pcx = SIx [n] / SI [n]. However, Pcx is the Purkinje image Pc
Is the horizontal coordinate of. Further, it is assumed that the nth optical image block is selected as the Purkinje image (step 10).
9).

Then, the process proceeds to step 110, and "pupil circle calculation" is performed as follows.

This "pupil circle calculation" is performed as follows in steps 601 to 617 of FIG.

In the above-mentioned pupil edge extraction processing, the MPU
Reference numeral 1 stores the coordinates of the extracted pupil edge, the lowest luminance MINn at the edge (slope) and the lowest luminance MIN0 of the entire image in the memory 2.

Therefore, first, the MPU 1 compares the minimum luminance (MIN (e)) of the area where the pupil edge is extracted with the minimum luminance (MIN0 + constant C) of the entire image, and MIN (e) ≤MIN0 + constant C If the condition is not satisfied, the pupil edge existing in the area is excluded as an inappropriate one.

Then, the pupil edge around the Purkinje image obtained in the previous step 109 is eliminated. This is performed by excluding those in the area centered on the horizontal position Pnx of the Purkinje image, for example, those in the range surrounded by (Pnx-constant) and (Pnx + constant).

Further, the average value m and the standard deviation σ of the coordinates of the pupil edge candidates selected so far are obtained, and [average value m−
a * standard deviation σ-average value m + a * standard deviation σ (however, a
Is a constant)].

Next, the pupil center is obtained by using the pupil edge selected as described above. This can be determined by the average value of the adopted pupil edges.

After that, the routine proceeds to step 111, where the rotation angle of the eyeball and the viewpoint position on the focusing plate of the camera are calculated using the Purkinje image and the position of the center of the pupil. And
In step 112, the AF point and the like are determined based on this viewpoint position, and the camera is controlled.

According to the third embodiment described above, the lowest value in a certain area (or the lowest luminance corresponding to the pupil edge) is stored together with the coordinates in the process of obtaining the positions of many pupil edges using the line sensor. However, if this value is not substantially equal to the minimum value of the entire image, it is excluded as an inappropriate edge, or it is determined whether or not there is an edge around the Purkinje image position to be extracted (determined). , If it exists, it is excluded as an inappropriate edge, or it is determined whether the positional relationship between the Purkinje image and the extracted edge has a predetermined relationship. It is excluded as an edge, or statistical processing is performed on the extracted edge, that is, the average value and the standard deviation are obtained, and it is determined whether or not it falls within the range defined by these values. Bako Because you have to eliminate as improper edge extraction and pupil (pupil center, pupil diameter) of the well pupil edge regardless of the state of the eyeball it becomes possible to calculate the.

[0160]

As described above, according to the present invention,
In the pupil edge extraction means, a first storage means for storing the lowest luminance value for each area obtained by the light receiving means, and a second storage for storing the lowest luminance value for all areas obtained by the light receiving means. Means, comparing the minimum brightness value of each area and the minimum brightness value in all areas, respectively, when the minimum brightness value in each area is not equal to the minimum brightness value in all areas, the edge in that area A pupil edge discriminating means for eliminating as an inappropriate pupil edge is provided, and an edge in an area having a lowest luminance value which is not equal to the lowest luminance value in all areas corresponding to the lowest luminance value of the pupil edge is a pupil edge. I try to eliminate it as inappropriate.

Further, according to the present invention, it is determined whether or not there is an edge in the periphery of the Purkinje image extracted by the Purkinje image extracting unit in the pupil edge extracting unit, and if there is, this edge is the pupil edge. A pupil edge discriminating means for eliminating as inappropriate is provided, and it is clear that the pupil edge is not located in the periphery of the Purkinje image. Therefore, the edge existing in the periphery of the Purkinje image is not suitable as the pupil edge. I'm trying to exclude it.

Further, according to the present invention, it is determined in the pupil edge extraction means whether or not the positions of the respective extracted edges and the Purkinje image extracted by the Purkinje image position extraction means have a predetermined positional relationship, and the Purkinje image is extracted. An edge that does not have a predetermined positional relationship with the image is excluded as an inappropriate pupil edge, and a pupil edge discriminating unit is provided. Since the pupil edge and the Purkinje image have a predetermined positional relationship, an edge that does not have this predetermined relationship is The pupil edge is excluded as an inappropriate one.

In the pupil edge extraction means, the extracted edges are statistically processed, that is, the average value and standard deviation of the extracted edges are obtained, and those outside the range defined by these values are extracted as pupil edges. A pupil edge discriminating means for eliminating inappropriate edges is provided as the pupil edge, and a pupil edge within a range defined by the average value and standard deviation of a large number of edges including the pupil edge is regarded as a pupil edge, and a pupil edge is outside the range. As an inappropriate edge, it is excluded.

Therefore, it becomes possible to always accurately extract the pupil edge.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a camera provided with a first embodiment device of the present invention.

2 is a flowchart showing a main operation of the camera of FIG.

3 is a flowchart showing an operation performed in step 107 of FIG.

FIG. 4 is a flowchart showing a continuation of the operation of FIG.

FIG. 5 is a diagram showing an example of changes in luminance due to eyelashes.

FIG. 6 is a flowchart showing an operation performed in step 110 of FIG.

FIG. 7 is a flowchart showing a continuation of the operation of FIG.

FIG. 8 is a diagram for explaining the positional relationship between the Purkinje image and the pupil during eye movement.

FIG. 9 is a flowchart showing an operation of a main part of a camera provided with a second embodiment device of the present invention.

FIG. 10 is a flowchart showing a continuation of the operation of FIG.

FIG. 11 is a flowchart showing an operation performed in step 110 of FIG. 2 in the second embodiment of the present invention.

FIG. 12 is a diagram for explaining the positional relationship between the Purkinje image and the pupil during eye movement.

FIG. 13 is a diagram showing an appropriate pupil edge range.

FIG. 14 is a flowchart showing an optical image block extracting operation of a camera provided with a third embodiment device of the present invention.

FIG. 15 is a flowchart showing the operation of calculating the pupil circle of a camera equipped with the device of the third embodiment of the present invention.

FIG. 16 is a diagram showing a state in which a pupil and an iris of an observer are imaged on a conventional semiconductor one-dimensional image sensor.

17 is a diagram showing a distribution of brightness of an image on the semiconductor one-dimensional image sensor of FIG.

FIG. 18 is a diagram showing an example of an image brightness distribution on an actual semiconductor one-dimensional image sensor.

FIG. 19 is a diagram showing another example of the actual distribution of the brightness of the image on the semiconductor one-dimensional image sensor.

FIG. 20 is a diagram showing another example of an actual image brightness distribution on a semiconductor one-dimensional image sensor.

FIG. 21 is a diagram showing still another example of the distribution of the brightness of the image on the actual semiconductor one-dimensional image sensor.

[Explanation of symbols]

 1 MPU 2 memory 4 CCD 5 IRED group

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Office reference number FI technical display location 7316-2K G03B 3/00 A

Claims (5)

[Claims] 1. A light projecting means for irradiating an eyeball of an observer looking through a finder system, a light receiving means for receiving a light flux reflected by the eyeball by the light projecting means, and a Purkinje image extracted from a signal from the light receiving means. Purkinje image extracting means for detecting the number of edges from the signal from the light receiving means as a pupil edge candidate, and a pupil edge extracting means for extracting a pupil edge from these, a Purkinje image extracting means and the pupil edge extracting means In a line-of-sight detection device comprising a calculation means for calculating the rotation angle of the optical axis of the eyeball based on information from each and detecting the line-of-sight direction of the observer, in the pupil edge extraction means, the light-receiving means First storage means for storing the lowest brightness value for each area obtained by the above, and second storage means for storing the lowest brightness value for all areas obtained by the light receiving means. , Comparing the minimum brightness value of each area and the minimum brightness value in all areas, if the minimum brightness value in each area is not equal to the minimum brightness value in all areas, the edge in that area is the pupil edge A line-of-sight detection device, which is provided with a pupil edge determination means that is excluded as inappropriate. 2. A light projecting means for irradiating an eyeball of an observer looking into the finder system, a light receiving means for receiving a light flux reflected by the eyeball by the light projecting means, and a Purkinje image extracted from a signal from the light receiving means. Purkinje image extracting means for detecting the number of edges from the signal from the light receiving means as a pupil edge candidate, and a pupil edge extracting means for extracting a pupil edge from these, a Purkinje image extracting means and the pupil edge extracting means In a line-of-sight detection device comprising a calculation means for calculating the rotation angle of the optical axis of the eyeball based on information from each and detecting the line-of-sight direction of the observer, in the pupil edge extraction means, the Purkinje image extraction It is determined whether or not there is an edge around the Purkinje image extracted by the means, and if there is, this edge is rejected as an unsuitable pupil edge. A line-of-sight detection device characterized in that a pupil edge discriminating means for removing the pupil line is provided. 3. A light projecting means for irradiating an eyeball of an observer looking through a finder system, a light receiving means for receiving a light flux reflected by the eyeball by the light projecting means, and a Purkinje image extracted from a signal from the light receiving means. Purkinje image extracting means for detecting the number of edges from the signal from the light receiving means as a pupil edge candidate, and a pupil edge extracting means for extracting a pupil edge from these, a Purkinje image extracting means and the pupil edge extracting means In the eye-gaze detecting device having a calculating means for calculating the rotation angle of the optical axis of the eyeball based on the information from each and detecting the eye-gaze direction of the observer, in the pupil edge extracting means, each edge extracted And whether the Purkinje image extracted by the Purkinje image extracting means has a predetermined positional relationship, and the edge not having the predetermined positional relationship with the Purkinje image is the pupil. A line-of-sight detection apparatus comprising a pupil edge discriminating means for eliminating an edge as an inappropriate one. 4. A light projecting means for irradiating an eyeball of an observer looking into the finder system, a light receiving means for receiving a light flux reflected by the eyeball by the light projecting means, and a Purkinje image extracted from a signal from the light receiving means. Purkinje image extracting means for detecting the number of edges from the signal from the light receiving means as a pupil edge candidate, and a pupil edge extracting means for extracting a pupil edge from these, a Purkinje image extracting means and the pupil edge extracting means In the line-of-sight detection device having a calculation means for calculating the rotation angle of the optical axis of the eyeball based on information from each and detecting the line-of-sight direction of the observer, in the pupil edge extraction means, to the extracted edge A visual axis detection device comprising a pupil edge discriminating means for performing statistical processing to eliminate an inappropriate edge as a pupil edge. 5. The pupil edge discriminating means is means for obtaining an average value and a standard deviation of the extracted edges, and rejecting an inappropriate edge as a pupil edge if it is out of a range defined by these values. The line-of-sight detection device according to claim 4.