JP2877399B2 - Gaze detection method of gaze detection device - Google Patents

Description

The present invention relates to a step of obtaining a photoelectric conversion signal distribution curve including a peak component corresponding to a first Purkinje image and a signal component corresponding to a light beam reflected from the fundus. Detecting the position of the first Purkinje image by performing signal processing on the photoelectric conversion signal distribution curve, and detecting the green position of the iris of the eye by excluding the signal components including the first Purkinje image from signal processing targets The present invention relates to an improvement of a gaze detection method of a gaze detection device, comprising the steps of: determining a pupil center based on a position of a pupil edge; and detecting a gaze direction from a positional relationship between the pupil center and the first Purkinje image.

(Prior Art) Recently, a camera, for example, a single-lens reflex camera, has a plurality of focusing zones within a field of view of a viewfinder, and has a position optically substantially conjugate to the plurality of focusing zones. An in-focus field of the auto-focus optical system corresponding to each focusing zone of the viewfinder is provided, a focusing zone in a direction in which the photographer's line of sight is directed is selected, and an auto-focusing zone corresponding to the selected focusing zone is selected. A device that focuses on a subject by using a focusing optical system based on focusing information on the subject that appears to overlap with the selected focusing zone has been proposed (see Japanese Patent Application No. 63-143259).

In order to detect the direction of the line of sight, a parallel light beam is emitted to the photographer's eye using a light transmission system (not shown), and the reflected light beam from the eye is detected using a one-dimensional line sensor (CCD) of a light receiving system. 10, the photoelectric conversion signal distribution curve 3 including the peak component 1 corresponding to the first Purkinje image having the highest reflection intensity and the signal component 2 corresponding to the light flux reflected from the fundus, as shown in FIG. The position of the first Purkinje image is detected by performing signal processing on the photoelectric conversion signal distribution curve 3, and the position of the edge of the pupil of the eye is detected by excluding the signal component 1 including the first Purkinje image from signal processing targets. A line-of-sight detection device is being developed which determines the pupil center based on the position of the edge of the pupil and detects the positional relationship between the pupil center and the first Purkinje image. In addition, 4 is a signal component based on the light flux reflected from the iris,
The skirt of the photoelectric conversion signal distribution curve 3 is formed.

By the way, conventionally, the direction of the line of sight is detected by performing signal processing described below on the photoelectric conversion signal distribution curve 3.

That is, as shown in FIG. 11, two slice levels SL ₁ and SL ₂ are prepared, and coordinates i ₁ and i ₂ for obtaining the edge of the pupil by slicing the signal component 2 by the slice level SL _1.
, And slice the peak component 1 to obtain coordinates PI ₁ and PI ₂ . Then, the pupil center is obtained by averaging the coordinates i ₁ and i _2, and the position of the first Purkinje image is obtained by averaging the coordinates PI ₁ and PI ₂ . The direction of the line of sight is detected from the position of the center of the pupil and the position of the first Purkinje image (Japanese Patent Application No. 62-14 / 1987).
No. 6067).

(Problems to be Solved by the Invention) However, the signal processing method disclosed in Japanese Patent Application No. 62-146067 discloses a method of processing two bit elements B ₁ , B _{1 at} positions sliced by slice levels SL ₂ , SL ₁ . _Since only the coordinates PI ₁ , PI ₂ , i ₁ and i ₂ are obtained using only B ₂ ′ and B ₁ ′ and B ₂ ′,
When noise is mixed in the photoelectric conversion signal distribution curve 3, there is a problem that the noise is easily affected by the noise.

Therefore, as shown in FIG. 12, the frequency of the bit element exhibiting the same output value is obtained, the output value of the most frequent bit element is set to the division level BL, and the division peak component 1 is calculated using the division level BL. '(14 as shown) and the division signal component 2' is divided into (13 as shown) and inverts the divided peak component 1 'and the divided signal component 2' and the right and left bordering the central O _2, respectively Prepare the inverted component. Then, the divided peak component 1 'and the inverted component are shifted from each other in the direction in which the bit elements are arranged, thereby obtaining the absolute value of the difference between the output values.
The absolute value of the difference between the output values is obtained by shifting the divided signal component 2 'and the inverted component 2' in the direction in which the bit elements are arranged. Next, a method has been proposed in which each bit element having the minimum absolute value is determined for the divided peak component 1 'and the divided signal component 2', and the position of the first Purkinje image and the center of the pupil are determined (details are given below). See Japanese Patent Application No. 63-143259). According to this method, the position of the first Purkinje image and the center of the pupil can be obtained by removing the influence of noise as much as possible.

However, the signal processing method disclosed in Japanese Patent Application No. 63-143259 discloses a distribution of a divided peak component 1 'and a divided signal component 2'.
It is assumed that the distribution is symmetrical about the center, and if ghosts are mixed in the optical system of the eye gaze detection device, the symmetry is impaired and the center of the pupil is accurately detected. The problem remains that it is difficult to determine the position of the first Purkinje statue.

Therefore, an object of the present invention is to provide a gaze detection method of a gaze detection device that can detect the direction of a gaze with high accuracy by removing the influence of noise, ghost, and the like as much as possible.

(Means for Achieving the Object) A gaze detection method of a gaze detection apparatus according to the present invention includes a step of performing signal processing on a photoelectric conversion signal distribution curve to detect a position of a first Purkinje image in order to achieve the above object. Performing a process of searching for a bit element corresponding to the pee value of the photoelectric exchange signal distribution curve, and a position of a first Purkinje image from a bit element outputting a peak value and a plurality of bit elements sandwiching the bit element And the step of interpolating.

Further, the step of excluding the peak component including the first Purkinje image from the target of the signal processing is to exclude a plurality of bit elements from the target of the signal processing with the bit element corresponding to the peak value interposed therebetween. I do.

Further, the step of determining the edge of the pupil of the eye includes obtaining the inflection point on the left side and the inflection point on the right side of the photoelectric conversion distribution curve based on the first straight line by Newton's method in order to determine the edge of the pupil. And performing a process of obtaining an intersection between the obtained second straight line and the third straight line by linear approximation.

(Operation) Since the gaze detection method of the gaze detection device according to the present invention obtains the center of the pupil and the position of the first Purkinje image by using a large number of bit elements, the influence of noise, ghost, and the like is removed as much as possible. The gaze direction can be detected with high accuracy.

(Embodiment) An embodiment in which the visual line detection device according to the present invention is applied to an autofocus optical system of a single-lens reflex camera will be described below with reference to the drawings.

FIG. 3 shows the viewfinder field of view of the single-lens reflex camera. The viewfinder field 10 is provided with three focusing zones 11, 12, and 13. Focus zone 1
The separation distance between 1, 12, and 13 is about 9 mm here. Single-lens reflex cameras have their respective focusing zones 11, 12, 13
An autofocus optical system (not shown) is provided in correspondence with. The autofocus optical system has a focus field (not shown) corresponding to each of the focusing zones 11, 12, and 13 at a position substantially conjugate with the three focusing zones 11, 12, and 13. Each autofocus optical system has CCDs 14, 15, and 16, as shown in FIG. The images of the subjects 17, 18, and 19 are re-imaged on the CCDs 14, 15, and 16 by a pair of separator lenses (not shown) constituting a part of the autofocus optical system.

CCDs 14, 15, and 16 are processing circuits 20 of a gaze detection device described later.
In the following, a schematic configuration of an optical system of the visual line detection device will be described with reference to FIG.

In FIG. 2, reference numeral 21 denotes a pentaprism, 22 denotes an eyepiece, 23 denotes a frame of a camera, 24 denotes a light transmission system of a visual line detection device, 25 denotes a light receiving system of the visual line detection device, and 26 denotes a beam splitter. The light transmission system 24 is generally constituted by a light source 27 'and a compensator prism 28'. The light source 27 'generates infrared light, and the infrared light is transmitted through the compensator prism 28' and the pentaprism 21 to the eyepiece 22.
And is converted into a parallel light beam by the eyepiece lens 22.
The infrared light beam is guided to the finder window 28 via the beam splitter 26. When the user puts his / her eyes 29 on the finder window 28, the photographer can see the subject 17, for example, as shown in FIG.
18, 19 can be seen. At the same time, an infrared light beam is projected on the eye 29 of the photographer.

By the projection of the infrared light beam, as shown in FIG. 4, a first Purkinje image PI is formed on the cornea 30 of the eye 29 of the photographer. Part of the infrared light beam passes through the cornea 30 and reaches the fundus. If the positional relationship between the first Purkinje image PI and the center 32 of the pupil 31 is determined, the rotation angle of the eye can be determined, and this point is disclosed in detail in Japanese Patent Application No. 63-143259. .

The light beam reflected from the fundus and the light beam reflected from the cornea 30 are guided again to the beam splitter 26 via the finder window 28. The beam splitter 26 provides a light receiving system.
Reflected towards 25. The light receiving system 25 includes a reduction lens 33, a mirror 34, a re-imaging lens 35, and a one-dimensional line sensor (CCD) 36.
It is roughly composed of The one-dimensional line sensor 36 has a number of bit elements having a predetermined width. On the one-dimensional line sensor 36, the images of the pupil edges 37, 37 that emerge as silhouettes based on the reflected light beam from the fundus and the first Purkinje image PI based on the reflected light beam from the cornea 30 are re-imaged.
The one-dimensional line sensor 36 outputs a photoelectric conversion signal corresponding to the re-imaging. The photoelectric conversion signal distribution curve 3 includes a peak component 1 corresponding to the first Purkinje image PI, a signal component 2 corresponding to a light beam reflected from the fundus, and a signal component 4 corresponding to a reflected light beam from the iris. The photoelectric conversion signal forming the photoelectric conversion signal distribution curve 3 is input to the processing circuit 20.

The processing circuit 20 performs processing described below based on a predetermined calculation program.

(1) Step of detecting the position of the first Purkinje image PI Step of searching for the bit element corresponding to the peak of the photoelectric conversion signal distribution curve 3 This processing is to find the maximum value by a commonly known method. Initial value in value storage memory (not shown)
V _max = 0 gives (see S ₁ in FIG. 15), and temporarily maximum value V _max. Bit element of the one-dimensional line sensor 36 went reference in order (the number and N) from left to right, the output value of the bit element if any bit element that outputs a value greater than the provisional maximum value V _max update the as a new provisional maximum value V _max, number I of the temporary maximum value V _max had corresponding bit element
_max is stored in a bit number storage memory (not shown).

That is, compared with the provisional maximum value V _max and the output V of the I-th bit element (I) (see _{S 2), V (I)} V max = V (I) when the ≧ V _max, I _max = The processing of I is performed (see S3). V (I) <(see S4) and increments the count number when the V _max, makes the determination of S2.

The processing from S1 to S4 is repeated from I = 1 to N,
By executing this process up to the final bit elements (N th bit element), the maximum value storage memory V _max true maximum value (peak value) V _max is stored, the bit number storage memory I
The _max number I _max of the bit element and outputs the true maximum value V _max is stored.

Next, a step of _obtaining the position of the first Purkinje image PI as the interpolation coordinate X _peak by using the bit element that outputs the maximum value.

First, the number I of the bit element that outputs the true maximum value V _max
max, calling each true maximum value V _max bit number storage memory, the maximum value storage memory.

Also, call the number I _max-1 bit element to the left of the bit element number I _max from the memory, the output value V of the bit element of that number I _max-1 a (I _max-1) in the memory V _-1 Remember
Then, call the number I _{max + 1} bit elements right next to the bit element number I _max from the memory, and stores the output value V of the number I _{max + 1-bit} elements (CImax + 1) in the memory V ₊₁ (S5 reference).

Next, it is determined whether or not V ₋₁ ≦ V ₊₁ (see S6). Next, the process proceeds to S7 or S8 to perform a process of _{obtaining the} interpolated coordinates _Xpeak .

When the output value V _-1 of the bit element of the number I _max-1 and the output value V ₊₁ of the bit element of the number I _{max + 1} are equal to each other, the number I _max
Although the center position of the bit element of _max can be considered as the peak value, generally, the output value V _-1 of the bit element of number I _max-1 and the output value V ₊₁ of the bit element of number I _{max + 1} are Are not always equal, and the bit element has a width, so X _peak is determined by interpolation.

That is, when the inequality V ₋₁ <V ₊₁ (or V ₋₁ ≦ V ₊₁ ) is satisfied, a straight line A and a straight line B are used as shown in FIG. As the intersection of (See S8 in FIG. 15). If the inequality does not hold, (See S7 in FIG. 15).

In this way, the position of the first Purkinje image PI is obtained as the interpolation coordinate X _peak using the bit element outputting the peak value and the plurality of bit elements sandwiching the bit element.

(2) Step of removing the bit including the first Purkinje image PI from the processing target In this processing, as shown in FIG. 6, the bit element of the number that outputs the output value forming the peak component 1 is removed from the processing target. exclude, is achieved by not called the output value of a plurality of bit elements of the left and right bit element number I _max. In FIG. 6, the bit elements in the range of arrow C are excluded from processing.

The number of bit elements to be excluded from the processing is related to the optical performance of the line-of-sight detection device and the width of the bit elements of the one-dimensional line sensor 36. However, the shape of the photoelectric conversion distribution curve 3 is projected on a CRT and the shape is reflected. It is desirable to decide while watching.

Here, the fourth counting left with the number I _max of the bit element that outputs a true maximum value V _max (I _max -4) and true maximum value V
with the number I _max of the bit element that outputs _max counting right 4
Each bit element between the first and the second (I _max +4) is excluded from the processing target.

Next, at the time of detecting the edge of the iris of the eye, which will be described later, in order to determine the threshold level used for the left processing and the threshold level used for the right processing, the rightmost bit element I _{R of the} left curve 40 and the right Of the bit element I _L on the left side of the curve 41 of FIG.

That, and stores the bit element number I _max -5 at the right end of the bit element memory I _R. Next, the left bit element memory
The bit element number I _max +5 is stored in I _L (see FIG. 15).
S9).

(3) Step of detecting the edge of the iris of the eye (the pupil edge 37) This processing is performed using the output values of the bit elements excluding the bit element that outputs the peak component 1, and the left side of the peak component 1 Curve 40 present on the right and curve 41 present on the right
And is performed for both.

Since the processing on the right curve 41 and the processing on the left curve 40 are performed symmetrically with respect to the peak component 1, the processing on the left curve 40 will be described in detail here.

 The processing for the left curve 40 is performed in the following procedure.

The number I _R (I _R = I _max) located approximately to the right of the left curve 40
5), that is, the output value V of the bit element located on the other curve side of one curve 40.
A threshold level SL smaller than (I _R ) is set as shown in FIG. It is desirable to use half the output value V (I _R ) of the bit element of the rightmost number I _{R as} the threshold level SL. Therefore, first, SLL = V
Performing (I _R) / 2 processing (see S101 in FIG. 16).

Focusing on two adjacent bit elements among the output values forming the left curve 40, the output value of the left bit element (the output value of the smaller bit element (indicated by number I)) V (I) is the threshold level SL
_It is determined whether or not the output value of the bit element smaller than _L (the output value of the bit element with the larger bit number (indicated by the number I + 1)) V (I + 1) is larger than the threshold level SL _L. (See S102). This judgment is I = 1
To I _R (see S104). The two adjacent bit elements satisfying this condition are stored in the memory as numbers I ₁ and I ₂ (see FIG. 1) (see S103).

Next, using the Newton method, in order to obtain the inflection point 5 as the intersection of the linear approximation, executes processing for calculating a first straight K _1.

This first straight line K ₁ is the second bit element (I ₁ + the second bit element (this is indicated by the number I ₃ )) counting from the bit element of the number I ₁ to the right and the bit element of the number I ₁ In order to obtain the second bit element (I ₁ -the second bit element (this is indicated by the number I ₄ )) counted from the left to the memory I _UP , I ₁ +
2 is stored, in the memory I _LO stores the I ₁ -2, and stores the D = 4 in the memory D (see S104). Next, (see S105) and proceeds in a straight line calculation routine, the plurality of bit elements in the vicinity of the bit element the number I ₃ and number I ₁ based on the output value V _3, V ₄ of I _4, I ₂ obtaining a first straight line K ₁ compatible.
Here, and requests the slope A _L1 and the intercept B _L1 of the straight line K _1, as shown in FIG. 18, the slope A _{is, A = (V (I UP} ) -V (I LO)) / D sections B is obtained by B = (I _UP × V (I _LO ) −I _LO × V (I _UP )) / D.

Then, at the intersection with the first straight line K ₁ is zero level
Determine the X _1. The bit element centered immediately to the left of this position X ₁ (the bit element closest to position X ₁ ) is defined as the zero-cross bit element (number I ₅ ). This is the memory I _UP
Is obtained by storing the value of -B ₁ DIVA ₁ in

Here, BDIVA means that processing of BDIVA = SGN (B / A) * INT (ABS (B / A)) is performed, and ABS (B / A) is the absolute value of the value obtained by dividing B by A. , INT means that the fractional part of the B / A value is rounded down to be converted to an integer, and SGN means whether the sign is positive or negative. Thereby, a zero cross bit element is obtained.

Then, the fourth bit element I _UP -4 (number I ₆ ) counted to the left of the zero-cross bit element (number I ₅ ) is stored in the memory I _LO (see S106). Then, the process proceeds (see S107) a straight line calculation routine of Figure 18, on the basis of the output value V ₆ of bit element number I ₆ and the output value V ₅ bit element of the zero-cross bit numbers I _5, a plurality A second straight line that conforms to the output value of the bit element and has a smaller slope than the first straight line
Determine the K _2. Here, finding the slope A ₂ and the intercept B ₂ straight lines K _2.

Then, a first straight line K ₁ the intersection X ₂ with a second straight line K ₂ obtained by mathematical methods. That is, the intersection point X ₂ is calculated as _{X 2 = (B 1 -B 2} ) / (A 2 -A 1) ( see S108).

Next, determine the bit element IX ₂ centered on the left at the position closest to the intersection X _2. This is obtained by performing an operation of IX ₂ = X ₂ DIV1.

The fourth bit element (number and output value V ₇ from the bit element of that number I ₇ counted to the right the corresponding counted from the intersection IX ₂ to the right to the second bit element (No. I ₇₎ I ₈ , in this case, coincides with I ₂ ) based on the output value V ₈ , a process of obtaining a third straight line K ₃ that matches the output values of the plurality of bit elements near the intersection X ₂ I do.

That is, X ₂ +2 is stored in the memory I _L0 and the memory I _UP
Is stored as I _L0 +4, and a bit element to be used is specified (see S109). Next, the process proceeds to the straight line calculation routine of FIG. 18, and the slope A ₃ and the intercept B ₃ are obtained (see S110).

Then, based on a second straight line K ₂ and the third straight line K _3, the intersection point _{X edge, X edge = (B} 3 -B 2) / obtained by calculation of (A ₃ -A ₂₎ (S111 see ).

Then, treat the intersection X _edge as inflection point 5,
The inflection point 5 is treated as the pupil edge, and the positions of the pupil edges 37, 37 are determined for the left curve 40 and the right curve 41, respectively. Then, the positions of the edges 37 of the pupil are averaged to obtain the pupil center.

That is, it is determined whether or not the limit point _TL is larger than the X _edge (see S113). When the limit point _TL is larger than X _edge , the processing of X _edge = _TL is performed, and the right side processing is performed (see S11).
When the limit point _TL is smaller than or equal to X _{edge, the} process directly proceeds to the right side process (see S11 in FIG. 17).
Then, in the right side processing, processing substantially similar to the left side processing is performed.

After the right side processing is completed, an average X _mean of the intersection X _edge (left) of the right side processing and the intersection X _edge (right) of the right side processing is obtained to obtain the center of the pupil (see S12).

Next, using the average X _mean and the interpolated coordinates X _peak , the position X
Obtain _obj (see S13).

X _obj = 2.467 * (X _mean −X _peak ) The coefficients of this equation are obtained based on the principle of gaze detection in Japanese Patent Application No. 63-143259.

7 to 9 show actual measurement examples of the distribution curve of the photoelectric conversion signal output from the one-dimensional line sensor 36. FIG. 7 shows the photoelectric conversion when the line of sight is directed to the central focusing zone 12. Signal distribution curve 3, FIG. 8 shows a photoelectric conversion signal distribution curve 3 when the line of sight is directed to the left focusing zone 11, and FIG. 9 shows a photoelectric conversion signal distribution when the line of sight is directed to the right focusing zone 13. Curve 3 is shown, and the direction of the line of sight is the central focusing zone 12.
Is set to X = 0 mm, the position when the line of sight is directed to the left focusing zone 11 is obtained as X = −9 mm, and when the line of sight is directed to the right focusing zone 13, X = 9 mm. Good detection results were also obtained in the processing.

(Effect) According to the gaze detection method of the gaze detection device according to the present invention, there is an effect that the direction of the gaze can be detected with high accuracy by removing the influence of noise, ghost, and the like as much as possible.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram for finding an inflection point by the Newton's method using the processing circuit of the eye-gaze detecting device according to the present invention, and FIG. FIG. 3 is a view showing a finder visual field of a single-lens reflex camera incorporating the visual line detection device shown in FIG. 2, and FIG. 4 is a view explaining a first Purkinje image formed on the eye using the visual line detection device. FIG. 5 is an explanatory diagram for explaining a process for obtaining a peak position using the processing circuit shown in FIG. 2, and FIG. 6 is an explanatory diagram for explaining a process for removing a peak component. 7 to 9 are explanatory diagrams for explaining an actual measurement example of a photoelectric conversion signal distribution curve, FIG. 10 is an explanatory diagram of a photoelectric conversion signal distribution curve, and FIG. 11 is a gaze detection of a conventional gaze detection device. FIGS. 12 to 14 show an example of a method. Diagram showing another example of the sight line detection method of the detection device, FIG. 15-FIG. 18 is a flow chart, of the sight line detection method of the visual line detection device according to the present invention. DESCRIPTION OF SYMBOLS 1 ... Peak component, 2 ... Signal component 3 ... Photoelectric conversion signal distribution curve, 5 ... Inflection point 20 ... Processing circuit, 31 ... Pupil 32 ... Pupil center, 36 ... One-dimensional line sensor (CCD) 37 ... Pupil edge, PI: First Purkinje image K ₁ … First straight line, K ₂ … Second straight line K ₃ … Third straight line, X _edge … Interpolated coordinates

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G02B 7/11 A61B 3/00

Claims (5)

(57) [Claims] A step of obtaining a photoelectric conversion signal distribution curve including a peak component corresponding to a first Purkinje image and a signal component corresponding to a light flux reflected from the fundus; Detecting a position of one Purkinje image; removing a peak component including the first Purkinje image from a signal processing target; detecting a position of an edge of a pupil of the eye; Determining the pupil center based on the pupil center, and detecting the direction of the line of sight from the positional relationship between the pupil center and the first Purkinje image, the signal processing the photoelectric conversion signal distribution curve, Detecting the position of the first Purkinje image includes performing a process of searching for a bit element corresponding to a peak value of the photoelectric conversion signal distribution curve, and outputting a peak value and a bit element that outputs the peak value. And interpolating steps the position of the first Purkinje image and a plurality of bit elements sandwiching the Tsu preparative element, the line-of-sight detection method of the visual axis detecting apparatus characterized by comprising a. 2. The method according to claim 1, wherein the step of excluding the peak component including the first Purkinje image from signal processing is performed by excluding a plurality of bit elements from signal processing with a bit element corresponding to the peak value interposed therebetween. 2. The method according to claim 1, wherein
7. A gaze detection method for the gaze detection device according to item 5. 3. The step of determining the edge of the pupil of the eye includes the step of determining the inflection point on the left side and the inflection point on the right side of the photoelectric conversion distribution curve to determine the edge of the pupil by Newton's method. The eye-gaze detecting device according to claim 2, wherein a process is performed to obtain a point of intersection by a linear approximation between a second straight line and a third straight line obtained based on the straight line. 4. The step of excluding a peak component including the first Purkinje image from a target of signal processing is performed by excluding a plurality of bit elements from a target of signal processing with a bit element corresponding to the peak value interposed therebetween. The step of obtaining the edge of the pupil includes, for the left curve and the right curve of the peak component excluded from the signal processing target, the output value of the bit element located on the other curve side of one curve. Setting a threshold level smaller than the output value of the bit element on the other curve side based on the threshold level, and two bit elements adjacent to the threshold level and one of the bit elements Obtaining two bit elements whose output value is smaller than the threshold level and whose output value of the other bit element is larger than the threshold level; Determining a first straight line that fits the output values of the plurality of bit elements in the vicinity of the bit element; and determining a zero-cross bit element located closest to the position where the first straight line intersects the zero level; A step of obtaining a second straight line having a smaller inclination than the first straight line, which is adapted to the plurality of bit elements based on output values of the plurality of bit elements near the zero-cross bit element, Obtaining a third straight line that is suitable for the output values of the plurality of bit elements near the intersection of the straight line and the second straight line; and determining the intersection of the third straight line and the second straight line. Determining the inflection point of the photoelectric conversion signal distribution curve and obtaining an edge of the pupil based on the inflection point of the left curve and the inflection point of the right curve. A gaze detection method for the gaze detection device according to claim 1 . 5. The line of sight according to claim 4, wherein the step of obtaining the center of the pupil is the step of obtaining the center of the pupil as a median between the left inflection point and the right inflection point. A gaze detection method for the detection device.