JPH1099278A - Visual axis detecting device and optical instrument - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce time required for one visual axis detection, by judging the reliability of calculated pupil circle by a calculation means and extracting a part of edges corresponding to the number of plural edges when the reliability is low. SOLUTION: A sequence controller checks whether a pupil circle correction is the first one or not (#101), and if it was the first one, the pupil circle estimation error CER is evacuated as CER0 (#103), and whether the total number of remaining edges is equal to or more than 10 or not is checked (#105). If it is equal to or larger than 10, the number is set as 'k=3' and k edges are extracted in the order (#110) to return to detection of the pupil center. If it is smaller than 10, whether it is equal to or larger than 5 is checked (#107), if it is equal to or larger than 5, the number is set as 'k=2' (#108) and k edges are extracted in order (#110) to return to detection of the pupil center.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a line-of-sight detecting device for detecting the line of sight of an observer observing a display and the like, and to an improvement in an optical device such as a camera having the line-of-sight detecting device.

[0002]

2. Description of the Related Art Hitherto, various so-called line-of-sight detection devices have been proposed which detect which position on a finder is observed by an observer. For example, JP-A-1-27473
In Japanese Patent Application Laid-Open No. 6-64, a parallel light beam from a light source is projected onto an eyeball of an observer, and a reflection image from a cornea, a so-called corneal reflection image (hereinafter also referred to as a P image (Purkinje image)) and an image forming position of a pupil are used. Looking for the visual axis. An example of this gaze detection device will be described with reference to FIGS.

FIG. 5 is a block diagram showing a main part of the electrical configuration of the eye-gaze detecting device. 1 is a sequence controller for controlling the sequence of the entire system, and 2 and 3 are observers for illuminating the observer's eyeballs. A pair of infrared light emitting diodes (hereinafter, referred to as IRED) juxtaposed horizontally with respect to
Reference numeral 4 denotes an IRED driver for driving the IREDs 2 and 3, 5 denotes an image sensor such as a CCD for reading an eyeball image illuminated by the IREDs 2 and 3, and 6 denotes an image signal read by the image sensor 5. An A / D converter for A / D conversion for easy processing, 7 is a push switch for setting start / stop of line-of-sight detection, and 8 is a battery for supplying power to the entire system.

FIG. 6 is a flowchart showing an example of a series of operations of the above-mentioned eye-gaze detecting device. FIG. 7 is an eyeball image projected on the image sensor 5 of FIG. 5 when an observer is in an eyepiece state close to the eye-gaze detecting device. FIG. 8 is a diagram showing an image output of a line where a P image exists on the image sensor 5 in FIG.

In FIG. 6, a sequence controller 1
First, the IREDs 2 and 3 are turned on toward the observer's eyeball (step # 1). Next, the eyeball image at that time is read by the image sensor 5, digitized via the A / D converter 6, and stored in the internal RAM as an image signal (step # 2). Then, the image signal is processed to extract a corneal reflection image (step # 3).

As a method of extracting the P image, the P image is considered to be a P image if the signal is at a certain level or more and has a large inclination, utilizing the fact that the P image is bright and its output is steep. For example, assume that the eyeball image is as shown in FIG. In this eyeball image, although P picture is a and b, and P picture condition to detect them, and the signal level is L ₁ (see FIG. 8) above, and is the slope is equal to or greater than a predetermined value, P image a, b in the horizontal line y _1, satisfies the condition as in FIG. 8, their position is IPa by gravity calculation or the like of the portion exceeding L _1, I
It is obtained as Pb.

Next, the sequence controller 1 processes the image signal to extract a pupil edge (step # 4).

[0008] As the way of extraction of the pupil edges, the pupil portion is dark, using the fact that the iris portion brighter than the signal level (and L ₂ as shown in FIG. 8 this) constant value level , and the addition of the intersection of the L ₂ when the predetermined period or more slope continues shall be deemed to be the pupil edge.

[0009] For example, an edge in the horizontal line y ₁ when the eyeball image was 7 as described above, e shown in FIG. 8
₁ and e ₆ are proper pupil edges. However, P
Image a, since base of b may thus satisfy the conditions of the pupil _{edge, e 2, e 3, e} 4, e 5 are processed as pupil edge despite not really a pupil edge. Therefore, in order to avoid this, as proposed in Japanese Patent Application Laid-Open No. 6-148509, after pupil edge extraction, edges regarded as pupils within a certain area around the P image are removed (step # 5).

Here, in addition to the false pupil edge around the P image (hereinafter simply referred to as a false edge), a false edge is generated by eyelashes and various other noises. The black dots in FIG. 7 represent all edges including false edges that can be detected.

In order to eliminate these false edges, the system controller 1 uses the P image position information based on the P image position information as proposed in Japanese Patent Laid-Open No. 4-347132.
The coordinate range of a probable edge is limited, and edges outside this range are excluded (step # 6). As a result, if a limited range is set as in the range F in FIG. 7, other edges, for example, e ₁₀ and e ₁₁ will be excluded. After the elimination work, the center of the pupil is detected (step # 7). Here, the center of the pupil is obtained by using the least square method or the like, utilizing the fact that the pupil portion is circular (including an elliptical shape). After this calculation, a pupil circle estimation error CER is calculated based on the obtained pupil center position and each pupil edge position in order to check the reliability of the calculated value (step #).
8).

The method of calculating the pupil circle estimation error CER is performed using a square error, as proposed in Japanese Patent Application Laid-Open No. 4-347131.

[0013] Next, the system controller 1 checks whether the following threshold C _T pupil circle estimating error CER obtained in the step # 8 is the probable (step # 9), if C _T below If so, the eyeball rotation angle is calculated based on the obtained pupil center position and P image position (step # 11), and the line-of-sight detection ends.

On the other hand, if the pupil circle estimation error CER is larger than C _T (NO in step # 9), it is determined that the pupil edge still contains many false edges, and the false edge is further eliminated, that is, Perform "pupil circle correction" (step # 10) and calculate the pupil center again (step # 7).
Return to

FIG. 9 is a flow chart for explaining "pupil circle correction" in step # 10, and the description will be made in accordance with this.

In FIG. 9, the sequence controller 1
First checks whether this pupil circle correction is the first time (first correction) (step # 501). If it is the first time, the pupil circle estimation error CER is retracted as the pupil circle estimation error CER ₀ (step # 503), and one edge is removed (step # 505). The order of extraction is the order stored in the sequence controller 1, and is usually the order of extraction.

If it is determined in step # 501 that the correction of the pupil circle is not the first time, that is, after the correction of the pupil circle for the first time or the second or subsequent time, step # 9 of FIG.
If the pupil circle estimation error CER is not equal to or less than the threshold value C _T and “pupil circle correction” is performed again in step # 10, the obtained latest pupil circle estimation error CER is the pupil circle estimation error CER before the edge is extracted. That is, it is checked whether the error is smaller than the pupil circle estimation error CER ₀ (this means that the extracted edge is a false edge and is approaching a perfect circle) (step # 502). Here, "CER
If it is not <CER ₀ ”, the extracted edge is not a false edge and is returned to the original (step # 504), and the next edge is extracted in the arrangement order (step # 505).

In the above step # 502,
If “CER <CER ₀ ”, the extracted edge is kept extracted, the pupil circle estimation error CER is saved as CER ₀ (step # 503), and the next edge is extracted in the order of arrangement (step # 505). ).

[0019]

However, in the above-mentioned conventional eye-gaze detecting apparatus, only one edge is removed per pupil circle correction operation regardless of the total number of edges, so that all false edges are eliminated. By the time the pupil-circle estimation error CER becomes equal to or less than C _T (that is, the pupil-circle correction is completed), the number of false edges is small, and the edge to be conveniently extracted at the beginning is a false edge. Except for the number of edges, the larger the number of edges, the more the step # 7 shown in FIG.
The number of times "detection of pupil center" in step # 8 and "calculation of pupil circle estimation error CER" in step # 8 increases, and specifically, steps # 7 → # 8 → # 9 → # 10 →
The processing of # 7... Is repeated, which leads to a longer detection time per gaze detection.

(Object of the Invention) It is an object of the present invention to provide a visual axis detecting device and an optical apparatus which can reduce the time required for one visual axis detection and can perform more efficient visual axis detection.

[0021]

In order to achieve the above object, the present invention according to the first to third and sixth aspects of the present invention comprises a plurality of illumination means for illuminating an observer's eyeball, A light-receiving means for receiving light reflected from the eyeball, and an extraction for extracting, based on an image signal obtained by the light-receiving means, an edge which can be regarded as a pupil edge in the eyeball image and positions of a plurality of corneal reflection images by the lighting means Means for calculating a pupil circle by treating the pupil portion as a circular shape from the plurality of extracted corneal reflection images and the plurality of edge information, obtaining a pupil center from the pupil circle, and determining the pupil center and the plurality of corneal reflections. A gaze detecting apparatus comprising: a calculating unit configured to detect a line of sight of an observer from a position of an image; wherein the calculating unit determines the reliability of the calculated pupil circle, and when the reliability is low, the plurality of edges are determined. Some edges depending on the number Than to pull out, and the structure provided with the pupil circle correcting means for correcting the said pupil circle.

More specifically, if the calculated pupil circle has low reliability, the number of edges to be extracted by one pupil circle correction is changed in accordance with the number of the plurality of edges. As the number of pupil circles increases, the number of edges to be extracted by one pupil circle correction is increased to correct the pupil circle.

According to another aspect of the present invention, a plurality of illumination means for illuminating an eyeball of an observer, and receiving light reflected by the illumination means from the eyeball of the observer. Based on the image signal obtained by the light receiving unit, an extracting unit that extracts an edge that can be regarded as a pupil edge in the eyeball image and a position of a plurality of corneal reflection images by the lighting unit, and the extracted The pupil portion is calculated as a circular shape by treating the pupil portion as a circular shape from the plurality of corneal reflection images and the plurality of pieces of edge information, and the pupil center is obtained from the pupil circle. A gaze detecting apparatus comprising: a gaze detecting device, wherein, when the calculated pupil circle has low reliability, the calculated pupil circle is calculated from a temporary pupil center position estimated from the positions of the plurality of corneal reflection images. For each edge of Based on the distance information dividing the plurality of edges into a plurality of groups, than to pull out in groups in one pupil circle modifications, and the configuration provided with the pupil circle correcting means for correcting the said pupil circle.

More specifically, if the calculated pupil circle is unreliable, the pupil circle is corrected in one pupil circle correction by removing the pupil circle from a group farther away.

[0025]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail based on illustrated embodiments.

FIG. 1 is a flowchart showing the operation of the main part of the eye-gaze detecting apparatus according to the first embodiment of the present invention. The electric configuration of the eye-gaze detecting device is the same as that of FIG. 5 described above, and a series of operations of the eye-gaze detecting device is the same as that of the process for “pupil circle correction” in step # 10 of FIG. Since it is the same as FIG. 10, the description is omitted here.

The pupil circle correcting operation will be described below with reference to the flowchart of FIG.

In FIG. 1, a sequence controller 1
First, it is checked whether the pupil circle correction is the first time (first correction) (step # 101). For the first time,
The pupil circle estimation error CER is saved as CER ₀ (step # 103), and it is checked whether the total number of remaining edges is 10 or more (step # 105). If the number is 10 or more, “k = 3” is set (step # 106), and k edges are extracted in the order of arrangement (step # 110).
To step # 7.

If it is determined in step # 105 that the total number of remaining edges is not 10 or more, it is checked whether it is 5 or more (step # 107). Where 5
If the number is equal to or more than "k = 2" (step # 10)
8), remove k edges in the arrangement order (step # 11)
0), and return to step # 7 in FIG.

If it is determined in step # 107 that the total number of remaining edges is not five or more, "k = 1" is set (step # 109), and k edges are removed in the arrangement order (step # 110). To step # 7.

If it is determined in step # 101 that the correction of the pupil circle has not been performed for the first time, that is, after the first or second or subsequent circle correction, the process returns to step # 7 in FIG. 6, and then returns to step # 9 again. in pupil circle estimating error CER is not less than the threshold C _T, when performing "pupil circle correction" at step # 10 again, the obtained latest pupil circle estimating error CE
R is smaller than the pupil circle estimation error CER before the edge is extracted, that is, the pupil circle estimation error CER ₀ (this is one or more of the extracted edges are false edges or include false edges). Is checked) (step # 102). Here, unless “CER <CER ₀ ” (step # 1)
02 (NO at step # 10), the extracted edge or edge group is returned to the original as it does not include a false edge (step # 10).
4), proceed to the next step # 105.

Steps after step # 105 are as described above. According to the total number of remaining edges (after extraction), if the total number of edges is 10 or more, three, the total number of edges is five or more, and 1
If the number is less than 0, two edges are extracted, and if the total number of edges is less than five, one edge is extracted, and the process returns to step # 7 in FIG.

On the other hand, in step # 102,
If “CER <CER ₀ ”, the extracted edge or edge group is left as it is and the pupil circle estimation error CE
R is saved as CER ₀ (step # 103), and the process also proceeds to step # 105. Similarly, if the total number of edges is 10 or more, the number is 3 and the total number of edges is 5 or more and 10 in accordance with the total number of remaining edges. If it is less than 2, two edges are extracted, and if the total number of edges is less than 5, one edge is extracted, and the process returns to step # 7 in FIG.

According to the first embodiment, when pupil circle correction is performed, processing is performed such that the greater the total number of edges, the greater the number of edges to be removed for each pupil circle correction. Therefore, the time required for one gaze detection can be reduced.

For example, when the total number of initial edges is 12 and the false edges are the last 3 in the arrangement order, the conventional example requires pupil circle correction to be performed 12 times. In the embodiment, only four times are required.

In this embodiment, every time the pupil circle correction is performed once, the least square method calculation is performed once in "detection of pupil center" in step # 7 of FIG. 6 after returning to the main routine. Therefore, reducing the number of times leads directly to shortening of the operation time as it is, and the effect is enormous.

(Second Embodiment) FIG. 2 is a flowchart showing the operation of the main part of the visual line detection device according to a second embodiment of the present invention. The electric configuration of the eye-gaze detecting device is the same as that of FIG. 5 described above, and a series of operations of the eye-gaze detecting device is the same as that of the process for “pupil circle correction” in step # 10 of FIG. Since it is the same as FIG. 10, the description is omitted here.

In the first embodiment, it is determined that if the “total number of remaining edges” is 10 or more, 3 is removed, if 5 or more and less than 10, 2 is removed, and if it is less than 5, 1 is removed. However, this may be expressed by a functional expression of “the total number of remaining edges” and “the number of edges to be removed”. By doing so, more appropriate processing according to the situation at each time can be performed. FIG. 2 is a flowchart showing this embodiment, and the operation of correcting the pupil circle will be described below.

In FIG. 2, the sequence controller 1
First checks whether this pupil circle correction is the first time (first correction) (step # 201). If it is the first time, the pupil circle estimation error CER is saved as CER ₀ (step # 203), and the number k of edges to be extracted is determined according to the following equation (1) (step # 205).

K ← (total number of remaining edges) / 4 + 1 (1) Then, k edges are extracted in the arrangement order (step # 20)
6), and return to step # 7 shown in FIG.

In step # 201, pupil circle correction is 1
If it is determined that it is not the first time, that is, after correcting the pupil circle for the first or second time, the process returns to step # 7 of FIG.
If the ER is not less than the threshold value C _T and the “pupil circle correction” of step # 10 is performed again, the obtained latest pupil circle estimation error CER is the pupil circle estimation error CER before the edge is extracted, ie, the pupil circle estimation. It is checked whether the error is smaller than the error CER (this means that one or more extracted edges are false edges or include false edges) (step # 202). .
Here, if not “CER <CER ₀ ” (step # 2)
02 (NO in step # 2), the extracted edge or edge group is returned to the original state as not including the false edge (step # 2).
04), the number k of edges to be extracted is determined according to the above equation (1) (step # 205), k edges are extracted in the order of arrangement (step # 206), and the process returns to step # 7 shown in FIG.

On the other hand, in step # 202,
If “CER <CER ₀ ”, the extracted edge or edge group is left as it is and the pupil circle estimation error CE
R is saved as CER ₀ (step # 203), the process also proceeds to step # 205, and the same processing is performed thereafter.

(Third Embodiment) FIG. 3 is a flow chart showing the operation of the main part of a visual axis detection device according to a third embodiment of the present invention. The electric configuration of the eye-gaze detecting device is the same as that of FIG. 5 described above, and a series of operations of the eye-gaze detecting device is the same as that of the process for “pupil circle correction” in step # 10 of FIG. Since it is the same as FIG. 10, the description is omitted here.

In the first and second embodiments, the "number of edges to be removed" is uniquely determined with respect to the "total number of remaining edges" in the correction of the pupil circle. However, the number of remaining edges is large. Only in the case, based on the distance information for each edge from the temporary pupil center position estimated from the P image position, the remaining pupil edges are divided into a group farther from the temporary pupil center position and a group closer thereto. , It is highly likely that the farthest group is a group of false edges, so that if one pupil circle correction is performed to sneak out for each group,
More efficient pupil circle correction becomes possible. FIG. 3 is a flowchart showing this embodiment, and the operation of correcting the pupil circle will be described below.

In FIG. 3, the sequence controller 1
First, it is checked whether this pupil circle correction is the first correction (first correction) (step # 301). If it is the first time, it is first checked whether the total number of remaining edges is 5 or more (step # 306). If not, the pupil circle estimation error CER is saved as CER ₀ (step # 303). One edge is removed in the arrangement order (step # 305), and the process returns to step # 7 shown in FIG.

[0046] If remaining edges total number of five or more in the step # 306, first determining the pupil center position C ₀ tentative based on the position of the P images. Of the two P image coordinates, the left side is (IP ₁ , JP ₁ ) and the right side is (IP ₂ , JP ₂ ), and the temporary pupil center position C ₀ (IC, JC) is given by IC ← (IP ₁ + IP _{2). ) / 2 JC ← (JP 1} + JP 2) / 2-10 to (step # 307).

Next, the distance from the pupil center position C ₀ is determined for each edge (step # 308). For example, if the coordinates of a certain edge are (IE ₁ , JE ₁ ), the distance EC _{1 to be obtained} is EC ₁ = ＩＣ [(IC-IE ₁ ) ² + (JC-JE ₁ ) ² ] (2) Becomes Next, an average value m of the obtained distance information is obtained (step # 309). Then, the remaining edges are divided into two groups as follows (step # 31)
0).

Edge of group A ← (distance from C ₀ ≧ m) Edge of group B ← (distance to C ₀ <m) Here, group A is located farther from the provisional pupil center position C ₀ . The group and the B group are closer groups.

FIG. 4 shows an example in which the remaining edges are divided into an A group and a B group.

Empirically, since there is a high possibility that the false edge exists farther from the center of the pupil, the sequence controller 1 uses this to silently remove the A group as a false edge (step # 311). Then, the pupil circle estimation error CER (this is the value before still being extracted) is calculated as CER ₀
(Step # 312), and returns to Step # 7 shown in FIG.

If it is determined that step # 301 has failed for the first time, the latest estimated pupil circle estimation error C
It is checked whether or not the ER is smaller than the pupil circle estimation error CER before the edge is removed, that is, the pupil circle estimation error CER ₀ (step # 302). Here "C
If ER <CER _{0 is} not satisfied, the extracted edge or edge group is returned as if it did not include a false edge (step # 304), and one edge is removed in the arrangement order (step # 305), and FIG. And returns to step # 7.

In the above step # 302,
If “CER <CER ₀ ”, the extracted pupil circle error C
ER is saved as CER ₀ (step # 303),
One next edge is removed in the arrangement order (step # 305),
It returns to step # 7 shown in FIG.

[0053] The direction in this embodiment, are divided into a group and close groups farther from the pupil center position C ₀ provisional for all edges, the edge is viewed from the center of the pupil by the extraction conditions When the original pupil is divided into a plurality of groups such as an upper edge, a lower edge, a right edge, a left edge, etc., the temporary pupil center position coordinates C
A group farther from ₀ and a group closer to ₀ may be divided.

According to each of the above embodiments, when performing pupil circle correction in gaze detection, the greater the total number of edges, the greater the number of edges to be extracted each time the pupil circle correction is performed (first embodiment).
And the second mode) or only when the total number of remaining edges is large, the remaining edges are far from the pupil center position based on the distance information for each edge from the temporary pupil center position estimated from the P image position. The third group is divided into a near group and a near group, and it is determined that the far group is likely to be a group of false edges, and one group is removed by one pupil circle correction (third embodiment). Since such a process is performed (the algorithm is configured), the time required for one gaze detection can be significantly reduced as compared with the related art.

(Correspondence between the Invention and the Embodiment) In each of the above embodiments, the IREDs 2 and 3 and the IRED driver 4
Are the illumination means of the present invention, the imaging device 5 is the light receiving means of the present invention, and the portion of the sequence controller 1 for executing the operations of steps # 3 and # 4 shown in FIG. 1 is a section for executing the operations of steps # 7, # 8, # 9, # 10, and # 11 (the section for performing the processing of FIG. 1, FIG. 2 or FIG. 3) shown in FIG. Each corresponds to a pupil circle correcting unit. Further, the circular shape according to claim 1 or 4 includes not only a circular shape but also an elliptical shape.

The correspondence between the components of the embodiment and the components of the present invention has been described above. However, the present invention is not limited to the configuration of the embodiment, and the functions and features described in the claims can be obtained.
Needless to say, any configuration may be used as long as the functions of the embodiment can be achieved.

(Modification) The present invention can be applied to various cameras such as a silver halide camera, a video camera, and an electronic still camera. Further, the present invention can be applied to a device having a display, a device having an operation panel (because it can be used for detecting a line of sight of an operator who looks at the display and the operation panel), and the like. The present invention can be applied to other optical apparatuses and other devices, and further to a constituent unit.

Further, the present invention may have a configuration in which the above embodiments or their techniques are appropriately combined.

[0059]

As described above, according to the present invention,
When performing pupil circle correction in gaze detection, as the total number of pupil edges increases, the number of edges to be extracted each time pupil circle correction is increased, or a temporary pupil center position estimated from the position of the corneal reflection image From divide multiple edges into multiple groups based on the distance report for each edge, and assuming that the more distant group is likely to be a group of false edges,
Since the pupil circle is removed one by one by one pupil circle correction, the time required for one gaze detection can be shortened, and more efficient gaze detection can be performed.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an operation of correcting a pupil circle of a gaze detecting apparatus according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating an operation of correcting a pupil circle of a gaze detecting apparatus according to a second embodiment of the present invention.

FIG. 3 is a flowchart showing an operation of correcting a pupil circle by a gaze detecting apparatus according to a third embodiment of the present invention.

FIG. 4 is a diagram for helping to explain the operation of FIG. 3;

FIG. 5 is a block diagram showing a main part of an electrical configuration of the present invention and a conventional visual line detection device.

FIG. 6 is a flowchart showing a series of operations of the visual line detection device of FIG. 5;

FIG. 7 is a diagram showing an eyeball image projected on the image sensor of FIG. 5;

FIG. 8 is a diagram for explaining image output and extraction of a pupil edge of a line where a P image exists on the image sensor illustrated in FIG. 7;

FIG. 9 is a flowchart showing an operation of correcting a pupil circle of the conventional eye gaze detecting device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Sequence controller 2, 3 IRED 4 IRED driver 5 Image sensor

Claims (6)

[Claims] A plurality of illuminating means for illuminating an observer's eye, a light receiving means for receiving light reflected from the observer's eye by the illuminating means, and an image signal obtained by the light receiving means. Extracting means for extracting an edge which can be regarded as a pupil edge in the eyeball image and positions of a plurality of corneal reflection images by the illuminating means; and treating the pupil portion as a circular shape from the extracted plurality of corneal reflection images and a plurality of edge information. To calculate the pupil circle,
A pupil center obtained from the pupil circle, and a sight line detection device comprising: a pupil center; and a calculation unit configured to detect a line of sight of an observer from the positions of the plurality of corneal reflection images, wherein the calculation unit calculates the calculated pupil circle Judgment of the reliability, and when the reliability is low, a pupil circle correcting means for correcting the pupil circle by extracting some edges according to the number of the plurality of edges is provided. Eye-gaze detecting device. 2. The pupil circle correcting means, when the reliability of the calculated pupil circle is low, changes the number of edges to be removed by one pupil circle correction according to the number of the plurality of edges. The gaze detection device according to claim 1, wherein: 3. The pupil circle correcting means, when the reliability of the calculated pupil circle is low, increases the number of edges to be removed by one pupil circle correction as the number of the plurality of edges increases. The gaze detection device according to claim 2, wherein 4. A plurality of illuminating means for illuminating an observer's eye, a light receiving means for receiving light reflected from the observer's eye by the illuminating means, and an image signal obtained by the light receiving means. Extracting means for extracting an edge which can be regarded as a pupil edge in the eyeball image and positions of a plurality of corneal reflection images by the illuminating means; and treating the pupil portion as a circular shape from the extracted plurality of corneal reflection images and a plurality of edge information. To calculate the pupil circle,
A pupil center obtained from the pupil circle, and a sight line detection device comprising: a pupil center; and a calculation unit configured to detect a line of sight of an observer from the positions of the plurality of corneal reflection images. If the reliability is low, the plurality of edges are divided into a plurality of groups based on distance information for each edge from a temporary pupil center position estimated from the positions of the plurality of corneal reflection images, and one pupil An eye gaze detecting apparatus comprising a pupil circle correcting means for correcting the pupil circle by removing the group in the circle correction. 5. The pupil circle correcting unit according to claim 4, wherein when the calculated pupil circle has low reliability, the pupil circle correcting unit removes the pupil circle from a group farther away in one pupil circle correction. Eye gaze detection device. 6. An optical apparatus comprising the line-of-sight detection device according to claim 1.