WO2021171396A1

WO2021171396A1 - Pupil position detection device, visual line detection appratus, and pupil position detection method

Info

Publication number: WO2021171396A1
Application number: PCT/JP2020/007602
Authority: WO
Inventors: 遼平村地; 亮介虎間
Original assignee: 三菱電機株式会社
Priority date: 2020-02-26
Filing date: 2020-02-26
Publication date: 2021-09-02

Abstract

An eye region extraction unit (3) extracts an eye region from an image acquired by an image acquisition unit (2). A pupil position detection unit (4) detects a plurality of pupil positions from the eye region extracted by the eye region extraction unit (3). A second reliability calculation unit (6) calculates a spread degree of the plurality of pupil positions on the basis of positional relationships between the plurality of pupil positions detected by the pupil position detection unit (4), and calculates a second reliability that serves as an index of certainty of the plurality of pupil positions, on the basis of the spread degree.

Description

Pupil position detection device, line-of-sight detection device, and pupil position detection method

The present disclosure relates to a pupil position detection device, a line-of-sight detection device, and a pupil position detection method.

The conventional pupil position detection device executes template matching to detect pupil position candidates from the image of the eye region, calculates the likelihood of each detected pupil position candidate, and selects the pupil position candidate having the maximum likelihood. It was determined to be the position (see, for example, Patent Document 1).

Japanese Unexamined Patent Publication No. 2013-215549

In template matching, a huge number of templates are required to support all imaging environments, but it is difficult to prepare such a huge number in advance. Therefore, the number of templates actually prepared is less than the number required to support all imaging environments. Therefore, in the conventional pupil position detecting device, the calculated likelihood may not be an appropriate value depending on the imaging environment, and there is a problem that the reliability of the pupil position cannot be evaluated correctly.

The present disclosure has been made to solve the above problems, and an object of the present disclosure is to evaluate the reliability of the pupil position without being affected by the imaging environment.

The pupil position detection device according to the present disclosure is composed of an image acquisition unit that acquires an image captured by a subject, an eye area extraction unit that extracts an eye area from an image acquired by the image acquisition unit, and an eye area extraction unit. The degree of spread of the plurality of pupil positions is calculated based on the positional relationship between the pupil position detection unit that detects a plurality of pupil positions from the extracted eye region and the plurality of pupil positions detected by the pupil position detection unit. It is provided with a second reliability calculation unit that calculates a second reliability that is an index of the certainty of the plurality of pupil positions based on the degree of spread.

According to the present disclosure, since the degree of spread of a plurality of pupil positions is not affected by the imaging environment, the reliability of the pupil positions can be evaluated without being affected by the imaging environment.

It is a block diagram which shows the structural example of the line-of-sight detection apparatus which concerns on Embodiment 1. FIG. 2A and 2B are diagrams showing a hardware configuration example of the line-of-sight detection device according to the first embodiment. It is a figure which shows the example of the eye area image extracted by the eye area extraction part. It is explanatory drawing which shows the example of a filter. It is explanatory drawing which shows the example of the sweep by a filter. It is explanatory drawing which shows the example of the vector corresponding to the 1st luminance gradient value, the example of the vector corresponding to the 2nd luminance gradient value, and the example of the luminance gradient vector. It is explanatory drawing which shows the example of the luminance value in the attention image unit, and the example of the luminance value in each image unit arranged around the attention image unit. It is explanatory drawing which shows the example of the luminance gradient vector corresponding to the attention image unit. It is explanatory drawing which shows the example of the evaluation area. It is explanatory drawing which shows the example of the luminance gradient vector corresponding to each evaluation image unit. FIG. 11A is a diagram showing an example of three pupil positions having a small degree of spread, and FIG. 11B is an explanatory diagram showing an example of a method of calculating the second reliability in the example of FIG. 11A. FIG. 12A is a diagram showing an example of three pupil positions having a large degree of spread, and FIG. 12B is an explanatory diagram showing an example of a method of calculating the second reliability in the example of FIG. 12A. It is explanatory drawing which shows the example of the calculation method of the line-of-sight angle with respect to the yaw direction. It is explanatory drawing which shows the example of the calculation method of the line-of-sight angle with respect to a pitch direction. It is a flowchart which shows the operation example of the line-of-sight detection apparatus which concerns on Embodiment 1. FIG. It is a block diagram which shows the structural example of the line-of-sight detection apparatus which concerns on Embodiment 2. FIG. It is explanatory drawing which shows the example of the pupil position in the 1st reliability calculation region and the brightness value distribution around it. It is explanatory drawing which shows the example of the concentric image unit group set in the 1st reliability calculation area. It is explanatory drawing which shows the example of the calculation method of the 3rd reliability. It is a flowchart which shows the operation example of the line-of-sight detection apparatus which concerns on Embodiment 2.

Hereinafter, in order to explain the present disclosure in more detail, a mode for carrying out the present disclosure will be described with reference to the accompanying drawings.
Embodiment 1.
FIG. 1 is a block diagram showing a configuration example of the line-of-sight detection device 10 according to the first embodiment. The line-of-sight detection device 10 includes a pupil position detection device 1 and a line-of-sight angle calculation unit 11. The pupil position detection device 1 includes an image acquisition unit 2, an eye region extraction unit 3, a pupil position detection unit 4, and a second reliability calculation unit 6. Further, the image pickup device 12 is connected to the pupil position detection device 1.

2A and 2B are diagrams showing a hardware configuration example of the line-of-sight detection device 10 according to the first embodiment. The functions of the image acquisition unit 2, the eye area extraction unit 3, the pupil position detection unit 4, the second reliability calculation unit 6, and the line-of-sight angle calculation unit 11 in the line-of-sight detection device 10 are realized by a processing circuit. That is, the line-of-sight detection device 10 includes a processing circuit for realizing the above functions. The processing circuit may be a processing circuit 100 as dedicated hardware, or may be a processor 101 that executes a program stored in the memory 102.

As shown in FIG. 2A, when the processing circuit is dedicated hardware, the processing circuit 100 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, or an ASIC (Application Specific Integrated Circuit). ), FPGA (Field Processor Gate Array), or a combination thereof. The functions of the image acquisition unit 2, the eye area extraction unit 3, the pupil position detection unit 4, the second reliability calculation unit 6, and the line-of-sight angle calculation unit 11 may be realized by a plurality of processing circuits 100, or the functions of each unit may be realized. May be collectively realized by one processing circuit 100.

As shown in FIG. 2B, when the processing circuit is the processor 101, the functions of the image acquisition unit 2, the eye area extraction unit 3, the pupil position detection unit 4, the second reliability calculation unit 6, and the line-of-sight angle calculation unit 11 Is realized by software, firmware, or a combination of software and firmware. The software or firmware is described as a program and stored in the memory 102. The processor 101 realizes the functions of each part by reading and executing the program stored in the memory 102. That is, the line-of-sight detection device 10 includes a memory 102 for storing a program in which the step shown in the flowchart of FIG. 15 described later will be executed as a result when executed by the processor 101. In addition, this program causes a computer to execute the procedures or methods of the image acquisition unit 2, the eye area extraction unit 3, the pupil position detection unit 4, the second reliability calculation unit 6, and the line-of-sight angle calculation unit 11. I can say.

Here, the processor 101 is a CPU (Central Processing Unit), a processing device, an arithmetic unit, a microprocessor, or the like.
The memory 102 may be a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), an EPROM (Erasable Program ROM), or a flash memory, and may be a non-volatile or volatile semiconductor memory such as a hard disk or a flexible disk. It may be an optical disc such as a CD (Compact Disc) or a DVD (Digital Versaille Disc).

Some of the functions of the image acquisition unit 2, the eye area extraction unit 3, the pupil position detection unit 4, the second reliability calculation unit 6, and the line-of-sight angle calculation unit 11 are realized by dedicated hardware, and some of them are realized. May be realized by software or firmware. As described above, the processing circuit in the line-of-sight detection device 10 can realize the above-mentioned functions by hardware, software, firmware, or a combination thereof.

The imaging device 12 images the periphery of the face of the subject whose pupil position is detected and the line of sight is detected in chronological order. The image pickup device 12 outputs the captured image I1 to the image acquisition unit 2.

The image acquisition unit 2 acquires the image I1 of the subject imaged by the image pickup device 12 from the image pickup device 12 and outputs it to the eye region extraction unit 3.

The eye region extraction unit 3 extracts the target person's eye region from the image I1 acquired by the image acquisition unit 2, and transfers the extracted image of the eye region (hereinafter referred to as “eye region image I2”) to the pupil position detection unit 4. Output.

FIG. 3 is a diagram showing an example of the eye region image I2 extracted by the eye region extraction unit 3. The eye area extraction unit 3 detects the face area of the subject from the image I1 acquired by the image acquisition unit 2, and from the detected face area, the outer corner FP1, the inner corner of the eye FP2, the upper eyelid FP3, and the lower eyelid, which are the feature points of the right eye. FP4 is detected, and the right eye region image I2 is extracted from the image I1 based on the positions of the detected eye feature points. Similarly, the eye region extraction unit 3 detects the outer corners of the eyes, the inner corners of the eyes, the upper eyelid, and the lower eyelid, which are the feature points of the left eye, from the face region, and based on the positions of the detected eye feature points, the left eye from the image I1. Extract the eye area image.

The eye region image I2 is composed of a plurality of units (hereinafter referred to as "image units") U arranged in two directions (that is, the X direction and the Y direction) orthogonal to each other. Here, each image unit U is composed of one pixel. Alternatively, each image unit U is composed of a plurality of pixels adjacent to each other.

The eye region extraction unit 3 extracts either the right eye region or the left eye region from the image I1. When the eye region extraction unit 3 extracts only one of the right eye and the left eye, that is, the pupil region of one eye, the pupil position detection unit 4 detects the pupil position of the one eye, and the second reliability calculation unit 6 , The second reliability of the pupil position of the one eye detected by the pupil position detection unit 4 is calculated.
Alternatively, the eye region extraction unit 3 may extract the right eye region and the left eye region. In this case, the pupil position detection unit 4 detects the pupil position of the right eye and the pupil position of the left eye, and the second reliability calculation unit 6 detects the second reliability of the pupil position of the right eye and the second reliability of the pupil position of the left eye. Calculate the reliability.

The pupil position detection unit 4 detects one or more positions (hereinafter, referred to as “pupil position”) indicating that it is within the pupil region from the eye region image I2 extracted by the eye region extraction unit 3. The pupil position detection unit 4 outputs information including the position of the eye feature point and one or more pupil positions to the second reliability calculation unit 6. The pupil position detection unit 4 may detect the pupil position from the eye region image I2 by a well-known method, and the method is not limited. The template matching as described above may be used, or the method based on the luminance gradient vector as described below may be used. The likelihood in template matching corresponds to the evaluation value E in the method based on the luminance gradient vector described later.

A method of calculating the luminance gradient vector V2 will be described with reference to FIGS. 4 to 8.
FIG. 4 is an explanatory diagram showing an example of the filter F. FIG. 5 is an explanatory diagram showing an example of sweeping by the filter F. The filter F shown in FIG. 4 is used for calculating the luminance gradient vector V2. The filter F is applied so as to sweep the eye region image I2 as shown in FIG.

The filter F has a luminance value B_L and a attention image in the image unit U_L arranged to the left of the attention image unit U_I for each image unit (hereinafter, may be referred to as "attention image unit") U_I in the eye area image I2. The difference value (hereinafter referred to as “first luminance gradient value”) ΔB_X from the luminance value B_R in the image unit U_R arranged to the right of the unit U_I is calculated. Further, the filter F is a difference value between the luminance value B_U in the image unit U_U arranged above the attention image unit U_I and the luminance value B_D in the image unit U_D arranged below the attention image unit U_I (hereinafter, "second"). It is called "luminance gradient value".) ΔB_Y is calculated.

FIG. 6 is an explanatory diagram showing an example of the vector V2_X corresponding to the first luminance gradient value ΔB_X, an example of the vector V2_Y corresponding to the second luminance gradient value ΔB_Y, and an example of the luminance gradient vector V2. The luminance gradient vector V2 is represented by the sum of the vector V2_X corresponding to the first luminance gradient value ΔB_X and the vector V2_Y corresponding to the second luminance gradient value ΔB_Y. Therefore, the pupil position detection unit 4 calculates the first luminance gradient value ΔB_X and the second luminance gradient value ΔB_Y, and uses the calculated first luminance gradient value ΔB_X and the second luminance gradient value ΔB_Y to obtain the luminance gradient vector V2. The angle corresponding to the direction (hereinafter referred to as “luminance gradient angle”) θ is calculated.

FIG. 7 shows an example of the luminance value B_I in the attention image unit U_I and an example of the luminance values B_D, B_L, B_R, B_U in the individual image units U_D, U_L, U_R, U_U arranged around the attention image unit U_I. It is explanatory drawing which shows. For example, as shown in FIG. 7, it is assumed that the luminance value B_L is 44, the luminance value B_R is 16, the luminance value B_U is 47, and the luminance value B_D is 18. In this case, by applying the filter F to the attention image unit U_I, the first luminance gradient value ΔB_X is calculated to −28 and the second luminance gradient value ΔB_Y is calculated to −29. Further, the luminance gradient angle θ is calculated to be 46 °. FIG. 8 is an explanatory diagram showing an example of the luminance gradient vector V2 corresponding to the attention image unit U_I.

The pupil position detection unit 4 uses the luminance gradient vector V2 to calculate the evaluation value E corresponding to each image unit U in the eye region image I2. The evaluation value E is based on the number n of the luminance gradient vectors V2 toward the individual image unit U (that is, the attention image unit U_I). A method of calculating the evaluation value E will be described with reference to FIGS. 9 and 10.

FIG. 9 is an explanatory diagram showing an example of the evaluation area EA. First, the pupil position detection unit 4 sets an evaluation region (hereinafter referred to as “evaluation region”) EA including the attention image unit U_I. The evaluation area EA includes the attention image unit U_I and includes N image units (hereinafter, may be referred to as “evaluation image unit”) U_E different from the attention image unit U_I. N is any integer greater than or equal to 2.

FIG. 9 shows an example of the evaluation area EA. In the example shown in FIG. 9, the shape of the evaluation region EA is square, and the image unit U_I of interest is arranged at the center of the evaluation region EA. Further, the evaluation area EA includes 48 evaluation image units U_E. That is, N = 48.

The evaluation area EA has a size smaller than a predetermined size (hereinafter referred to as "reference iris size"). The reference iris size corresponds to the size of a standard human iris.

Next, the pupil position detection unit 4 calculates the angle θ'corresponding to the inclination of the straight line connecting the attention image unit U_I and the individual evaluation image units U_E. Next, the pupil position detection unit 4 calculates the difference value Δθ between the corresponding luminance gradient angle θ and the corresponding angle θ'for each evaluation image unit U_E. When the calculated difference value Δθ is less than the predetermined threshold value Δθth, the pupil position detection unit 4 determines that the corresponding luminance gradient vector V2 has a direction toward the attention image unit U_I. On the other hand, when the calculated difference value Δθ is equal to or greater than the threshold value Δθth, the pupil position detection unit 4 determines that the corresponding luminance gradient vector V2 does not have a direction toward the attention image unit U_I.

Next, the pupil position detection unit 4 calculates the number n of the luminance gradient vectors V2 toward the attention image unit U_I based on the result of such determination. At this time, the number n is calculated as a value of 0 or more and N or less. The pupil position detection unit 4 calculates an evaluation value E according to the calculated number n. That is, the larger the number n, the higher the evaluation value E is calculated. In other words, the smaller the number n, the lower the evaluation value E is calculated.

FIG. 10 is an explanatory diagram showing an example of the luminance gradient vector V2 corresponding to each evaluation image unit U_E. For example, as shown in FIG. 10, it is assumed that the evaluation region EA includes 48 evaluation image units U_E (N = 48). Of the 48 luminance gradient vectors V2 corresponding to the 48 evaluation image units U_E, it is assumed that 19 luminance gradient vectors V2 have a direction toward the attention image unit U_I (n = 19). In the figure, the solid line arrow indicates the luminance gradient vector V2 having a direction toward the attention image unit U_I. On the other hand, the arrow by the broken line indicates the luminance gradient vector V2 having no direction toward the attention image unit U_I.

In this case, the pupil position detection unit 4 calculates the number n to 19. Further, the pupil position detection unit 4 calculates an evaluation value E according to the calculated number n. For example, the evaluation value E based on n / N is calculated. The pupil position detection unit 4 detects the pupil position PP in the eye region image I2 based on the calculated evaluation value E. More specifically, the pupil position detection unit 4 extracts an evaluation value E that is equal to or higher than a predetermined threshold value from a plurality of evaluation values E corresponding to the plurality of image units U in the eye region image I2. , The position corresponding to the extracted evaluation value E is detected as the pupil position PP. At this time, the pupil position detection unit 4 may detect a plurality of pupil position PPs when there are a plurality of evaluation values E that are equal to or higher than the threshold value. Further, the pupil position detection unit 4 detects coordinate values C_X and C_Y indicating the position of the image unit U corresponding to the pupil position PP in the eye region image I2 for each of the detected one or more pupil position PPs.

The pupil position detection unit 4 detects the position corresponding to the maximum value (that is, the maximum value) of the plurality of evaluation values E corresponding to the plurality of image units U in the eye region image I2 as the pupil position PP. You may.

Here, the principle that the pupil position PP is detected based on the evaluation value E will be described.

First, the eye region image I2 is usually composed of a region corresponding to the inside of the ocular fissure (hereinafter referred to as "intraocular fissure region") and a region corresponding to the outside of the ocular fissure (hereinafter referred to as "extraocular fissure region"). It is configured. The extraocular region is located around the intraocular region. The intraocular fissure region includes a region corresponding to the pupil (hereinafter referred to as “pupil region”), a region corresponding to the iris (hereinafter referred to as “iris region”), and a region corresponding to the sclera (hereinafter referred to as “sclera region”). ".) Is included. The scleral region is located around the iris region. The iris region is located around the pupil region. The shape of the iris region is circular. The shape of the pupil region is circular.

Second, the brightness in the pupil region is usually lower than the brightness in the iris region. Therefore, in the intraocular fissure region, an edge based on the brightness discontinuity is generated at the boundary between the pupil region and the iris region. Moreover, the brightness in the iris region is lower than the brightness in the sclera region. Therefore, in the intraocular fissure region, an edge based on the brightness discontinuity is generated at the boundary between the iris region and the scleral region.

According to these principles, it is highly probable that the number n corresponding to the image unit U in the pupil region is larger than the number n corresponding to the image unit U outside the pupil region. Therefore, the evaluation value E corresponding to the image unit U in the pupil region is likely to be higher than the evaluation value E corresponding to the image unit U outside the pupil region. Therefore, the pupil position PP can be detected by detecting the evaluation value E which is equal to or higher than the threshold value among the evaluation values E.

When the pupil position detection unit 4 detects a plurality of pupil positions, the second reliability calculation unit 6 calculates the degree of expansion of the plurality of pupil positions based on the positional relationship of the plurality of pupil positions. The second reliability calculation unit 6 calculates the second reliability, which is an index of the certainty of the plurality of pupil positions, based on the calculated spread degree. The second reliability calculation unit 6 obtains information including the position of the eye feature point, the plurality of pupil positions, the evaluation value E of each pupil position, and the second reliability common to the plurality of pupil positions. Output to the calculation unit 11.

In the pupil position evaluation method using the second reliability, if the distance spread of the plurality of pupil positions is large, in other words, if the positional variation is large, the pupil position detection unit 4 erroneously detects the pupil position. This is an evaluation method based on the idea that there is a high possibility. Since the pupil position detecting unit 4 tries to detect the pupil position in the pupil region, there is a tendency to detect a plurality of pupil positions having substantially the same evaluation value E in the pupil region. Therefore, when the distance spread of the plurality of pupil positions detected by the pupil position detection unit 4 is small, it is highly possible that the plurality of pupil positions exist in the pupil region, and thus the plurality of pupil positions are located. It can be said that it is highly reliable. On the other hand, when the distance spread of the plurality of pupil positions detected by the pupil position detection unit 4 is large, it is highly possible that at least one of the plurality of pupil positions exists outside the pupil region. It can be said that the positions of a plurality of pupils are unreliable. A method of calculating the second reliability will be described with reference to FIGS. 11A, 11B, 12A, and 12.

FIG. 11A is a diagram showing an example of three pupil positions PP1, PP2, and PP3 having a small degree of spread. FIG. 11B is an explanatory diagram showing an example of a method for calculating the second reliability in the example of FIG. 11A. As shown in FIG. 11A, it is assumed that the pupil position detection unit 4 detects three pupil positions PP1, PP2, PP3 from the eye region image I2. In FIG. 11B, the pupil position image unit U_21 corresponds to the pupil position PP1, the pupil position image unit U_22 corresponds to the pupil position PP2, and the pupil position image unit U_23 corresponds to the pupil position PP3. First, the second reliability calculation unit 6 sets the circumscribed rectangle 21 circumscribing the pupil position image units U_21, U_22, and U_23 with respect to the eye region image I2. Next, the second reliability calculation unit 6 calculates the diagonal length L1 based on the minimum coordinates (xmin, ymin) and the maximum coordinates (xmax, ymax) in the set circumscribing rectangle 21.

FIG. 12A is a diagram showing an example of three pupil positions PP4, PP5, and PP6 having a large degree of spread. FIG. 12B is an explanatory diagram showing an example of a method for calculating the second reliability in the example of FIG. 12A. As shown in FIG. 12A, it is assumed that the pupil position detection unit 4 detects three pupil positions PP4, PP5, and PP6 from the eye region image I2. In FIG. 12B, the pupil position image unit U_24 corresponds to the pupil position PP4, the pupil position image unit U_25 corresponds to the pupil position PP5, and the pupil position image unit U_26 corresponds to the pupil position PP6. First, the second reliability calculation unit 6 sets the circumscribed rectangle 22 circumscribing the pupil position image units U_24, U_25, and U_26 with respect to the eye region image I2. Next, the second reliability calculation unit 6 calculates the diagonal length L2 based on the minimum coordinates (xmin, ymin) and the maximum coordinates (xmax, ymax) in the set circumscribing rectangle 22.

The diagonal length L1 is a value representing the degree of spread of the pupil position image units U_21, U_22, and U_23. The smaller the value of the diagonal length L1, the higher the reliability of the pupil positions PP1, PP2, and PP3 corresponding to the pupil position image units U_21, U_22, and U_23. Similarly, the diagonal length L2 is a value representing the degree of spread of the pupil position image units U_24, U_25, and U_26. The smaller the value of the diagonal length L2, the higher the reliability of the pupil positions PP4, PP5, PP6 corresponding to the pupil position image units U_24, U_25, U_26. The second reliability calculation unit 6 may use the reciprocal of the diagonal lengths L1 and L2 as the second reliability, or may calculate the second reliability according to the diagonal lengths L1 and L2. When calculating the second reliability according to the diagonal lengths L1 and L2, the shorter the diagonal lengths L1 and L2, the larger the second reliability is calculated, and the longer the diagonal lengths L1 and L2, the higher the second reliability. Is calculated as a small value.

In the above description, the second reliability calculation unit 6 has set circumscribed rectangles circumscribing at a plurality of pupil positions, but a figure having a shape other than the rectangle may be set. Further, although the second reliability calculation unit 6 has calculated the diagonal length as the degree of spread, an index other than the diagonal length may be calculated as long as it is an index indicating the degree of spread (variation) of a plurality of pupil positions. ..

When the pupil position detection unit 4 is configured to detect only the pupil position corresponding to the maximum evaluation value E among the plurality of evaluation values E, the second reliability calculation unit 6 may perform a plurality of evaluations. A plurality of pupil positions corresponding to the evaluation value E larger than the predetermined value among the values E may be acquired from the pupil position detection unit 4. The second reliability calculation unit 6 calculates the degree of spread of the plurality of pupil positions acquired from the pupil position detection unit 4, and as a result, the pupil having the maximum evaluation value E detected by the pupil position detection unit 4. The reliability of the position can be evaluated.

The line-of-sight angle calculation unit 11 outputs the position of the eye feature point, the plurality of pupil positions, the evaluation value E of each pupil position, and the second reliability common to the plurality of pupil positions, which is output by the pupil position detection device 1. Get information including and. The line-of-sight angle calculation unit 11 calculates the line-of-sight angle φ of the subject using the acquired information. For example, when the pupil position detection device 1 outputs the above information regarding only the right eye, the line-of-sight angle calculation unit 11 has a plurality of pupil positions as long as the second reliability included in this information is equal to or higher than a predetermined threshold value. The pupil position having the maximum evaluation value E is selected from the inside, and the line-of-sight angle φ is calculated based on the selected pupil position. Further, for example, when the pupil position detecting device 1 outputs the above information regarding the right eye and the above information regarding the left eye, the line-of-sight angle calculation unit 11 includes the second reliability included in the information of the right eye and the second reliability included in the information of the left eye. 2 If both of the reliabilitys are equal to or higher than the above threshold value, the pupil position having the maximum evaluation value E is selected from the plurality of pupil positions included in the information of the right eye and the plurality of pupil positions included in the information of the left eye. Then, the line-of-sight angle φ is calculated based on the selected pupil position. On the other hand, if only one of the second reliability included in the information of the right eye and the second reliability included in the information of the left eye is equal to or higher than the above threshold value, the line-of-sight angle calculation unit 11 has a second reliability equal to or higher than the above threshold value. The pupil position having the maximum evaluation value E is selected from the plurality of pupil positions corresponding to the degrees, and the line-of-sight angle φ is calculated based on the selected pupil position. If the second reliability is less than the above threshold value, the line-of-sight angle calculation unit 11 does not calculate the line-of-sight angle φ based on this information.

In the above description, the second reliability calculation unit 6 includes the position of the eye feature point, the plurality of pupil positions, the evaluation value E of each pupil position, and the second reliability common to the plurality of pupil positions. Although it is configured to output information, it may be configured to output information including the position of the eye feature point, the pupil position having the maximum evaluation value E, and the second reliability. In this case, if the second reliability included in the information acquired from the pupil position detection device 1 is equal to or higher than a predetermined threshold value, the line-of-sight angle calculation unit 11 gazes at the pupil position having the maximum evaluation value E. Calculate the angle φ.

The line-of-sight angle φ in this example includes the line-of-sight angle φ_X with respect to the yaw direction and the line-of-sight angle φ_Y with respect to the pitch direction. A method of calculating the line-of-sight angle φ_X will be described with reference to FIG. FIG. 13 is an explanatory diagram showing an example of a method of calculating the line-of-sight angle φ_X with respect to the yaw direction. Further, a method of calculating the line-of-sight angle φ_Y will be described with reference to FIG. FIG. 14 is an explanatory diagram showing an example of a method of calculating the line-of-sight angle φ_Y with respect to the pitch direction.

First, the line-of-sight angle calculation unit 11 calculates the positions P_FP1_X and P_FP2_X in the X direction using the information of the eye feature points. The position P_FP1_X corresponds to the outer corner of the eye FP1. The position P_FP2_X corresponds to the inner corner FP2.

Next, the line-of-sight angle calculation unit 11 calculates the position (hereinafter referred to as "first reference position") P_C_X in the X direction based on the calculated positions P_FP1_X and P_FP2_X. The first reference position P_C_X corresponds to an intermediate position between the outer corner FP1 and the inner corner FP2. That is, the first reference position P_C_X corresponds to the central portion of the intraocular fissure region with respect to the X direction.

Next, the line-of-sight angle calculation unit 11 calculates the position P_PP_X with respect to the X direction using the information on the pupil position. The position P_PP_X corresponds to the pupil position PP.

Next, the line-of-sight angle calculation unit 11 calculates the interval L_X_1 with respect to the first reference position P_C_X and the interval L_X_2 with respect to the position P_PP_X for the position P_FP1_X or the position P_FP2_X. FIG. 13 shows an example when the position P_FP2_X is used as a reference for the intervals L_X_1 and L_X_2.

The line-of-sight angle calculation unit 11 is preset with a value indicating the maximum value φmax_X of the line-of-sight angle φ_X. The line-of-sight angle calculation unit 11 calculates the line-of-sight angle φ_X by the following equation (1) using the set maximum value φmax_X and the calculated intervals L_X_1 and L_X_2.

Φ_X = φmax_X × (L_X_1-L_X_2) / L_X_1 (1)

The maximum value φmax_X is based on, for example, the following model M_X. That is, in the model M_X, if the position P_PP_X is the same as the first reference position P_C_X, the line-of-sight angle φ_X is 0 °. Further, in the model M_X, if the position P_PP_X is the same as the position P_FP1_X, the line-of-sight angle φ_X becomes a value (for example, −60 °) corresponding to the maximum value φmax_X. Further, in the model M_X, if the position P_PP_X is the same as the position P_FP2_X, the line-of-sight angle φ_X becomes a value (for example, + 60 °) corresponding to the maximum value φmax_X.

Further, the line-of-sight angle calculation unit 11 calculates the positions P_FP3_Y and P_FP4_Y in the Y direction by using the information of the eye feature points. The position P_FP3_Y corresponds to the upper eyelid FP3. The position P_FP4_Y corresponds to the lower eyelid FP4.

Next, the line-of-sight angle calculation unit 11 calculates the position P_C_Y in the Y direction (hereinafter referred to as "second reference position") based on the calculated positions P_FP3_Y and P_FP4_Y. The second reference position P_C_Y corresponds to an intermediate position between the upper eyelid FP3 and the lower eyelid FP4. That is, the second reference position P_C_Y corresponds to the central portion of the intraocular fissure region with respect to the Y direction.

Next, the line-of-sight angle calculation unit 11 calculates the position P_PP_Y with respect to the Y direction using the information on the pupil position. The position P_PP_Y corresponds to the pupil position PP.

Next, the line-of-sight angle calculation unit 11 calculates the interval L_Y_1 with respect to the second reference position P_C_Y and the interval L_Y_2 with respect to the position P_PP_Y for the position P_FP3_Y or the position P_FP4_Y. FIG. 14 shows an example when the position P_FP3_Y is used as a reference for the intervals L_Y_1 and L_Y_2.

The line-of-sight angle calculation unit 11 is preset with a value indicating the maximum value φmax_Y of the line-of-sight angle φ_Y. The line-of-sight angle calculation unit 11 calculates the line-of-sight angle φ_Y by the following equation (2) using the set maximum value φmax_Y and the calculated intervals L_Y_1 and L_Y_2.

Φ_Y = φmax_Y × (L_Y_1-L_Y_2) / L_Y_1 (2)

The maximum value φmax_Y is based on, for example, the following model M_Y. That is, in the model M_Y, if the position P_PP_Y is the same as the second reference position P_C_Y, the line-of-sight angle φ_Y is 0 °. Further, in the model M_Y, if the position P_PP_Y is the same as the position P_FP3_Y, the line-of-sight angle φ_Y becomes a value (for example, + 20 °) corresponding to the maximum value φmax_Y. Further, in the model M_Y, if the position P_PP_Y is the same as the position P_FP4_Y, the line-of-sight angle φ_Y becomes a value (for example, −20 °) corresponding to the maximum value φmax_Y.

The method of calculating the line-of-sight angles φ_X and φ_Y is not limited to these specific examples. Various known techniques can be used to calculate the line-of-sight angles φ_X and φ_Y. Detailed description of these techniques will be omitted.

In the above description, the line-of-sight detection device 10 includes a line-of-sight angle calculation unit 11, and the line-of-sight angle calculation unit 11 calculates the line-of-sight angle φ of the target person, but the present invention is not limited to this configuration. For example, the line-of-sight detection device 10 may include a line-of-sight direction calculation unit (not shown), and the line-of-sight direction calculation unit may calculate the line-of-sight direction of the target person. Various known techniques can be used for the calculation method of the line-of-sight direction. Detailed description of these techniques will be omitted.

At least one of the line-of-sight angle and the line-of-sight direction of the target person calculated by the line-of-sight detection device 10 is used, for example, for determining the state of the target person. Here, consider an example in which the line-of-sight detection device 10 and the image pickup device 12 are mounted on the vehicle, and the image pickup device 12 takes an image of the driver as a target person. In this example, the line-of-sight detection device 10 detects at least one of the line-of-sight angle and the line-of-sight direction of the driver who is the target person. Then, the driver monitoring device (not shown) mounted on the vehicle uses at least one of the driver's line-of-sight angle and line-of-sight direction, and the driver's state is a predetermined state (hereinafter referred to as "warning target state"). .) Is determined. When the driver monitoring device determines that the driver is in the warning target state, the driver monitoring device outputs a voice or the like to warn the driver. The warning target states are the state where the driver is driving aside, the state where the driver is not checking backward at the timing when the driver should check backward, the state where the driver's attention is low, and the driver is dozing. It is a state where the driver is not able to drive, and a state where the driver cannot drive. Various known techniques can be used to determine the state subject to warning. Detailed description of these techniques will be omitted.

FIG. 15 is a flowchart showing an operation example of the line-of-sight detection device 10 according to the first embodiment. The process shown in FIG. 15 is repeatedly executed when a predetermined condition is satisfied (for example, when the ignition power of the vehicle equipped with the line-of-sight detection device 10 is turned on).

First, the image acquisition unit 2 acquires the image I1 captured by the subject from the image pickup device 12 (step ST1). Next, the eye region extraction unit 3 extracts the eye region image I2 from the image I1 acquired by the image acquisition unit 2 (step ST2). Next, the pupil position detection unit 4 detects a plurality of pupil position PPs from the eye region image I2 extracted by the eye region extraction unit 3 (step ST3). Next, the second reliability calculation unit 6 calculates the degree of spread of the plurality of pupil position PPs based on the positional relationship of the plurality of pupil position PPs detected by the pupil position detection unit 4, and based on the degree of spread. The second reliability, which is an index of the certainty of the plurality of pupil position PPs, is calculated (step ST4). Next, if the line-of-sight angle calculation unit 11 determines that the value of the second reliability calculated by the second reliability calculation unit 6 is equal to or greater than a predetermined threshold value, the evaluation value E is based on the highest pupil position PP. The line-of-sight angle φ of the subject is calculated (step ST5).

As described above, the pupil position detection device 1 according to the first embodiment includes an image acquisition unit 2, an eye region extraction unit 3, a pupil position detection unit 4, and a second reliability calculation unit 6. The image acquisition unit 2 acquires an image captured by the subject. The eye region extraction unit 3 extracts the eye region from the image acquired by the image acquisition unit 2. The pupil position detection unit 4 detects a plurality of pupil positions from the eye region extracted by the eye region extraction unit 3. The second reliability calculation unit 6 calculates the degree of expansion of the plurality of pupil positions based on the positional relationship of the plurality of pupil positions detected by the pupil position detection unit 4, and the plurality of pupils based on the degree of expansion. The second reliability, which is an index of the certainty of the position, is calculated. Since the degree of spread of the plurality of pupil positions is not affected by the imaging environment, the pupil position detecting device 1 can evaluate the reliability of the pupil positions without being affected by the imaging environment.

Further, the line-of-sight detection device 10 according to the first embodiment includes a pupil position detection device 1 and a line-of-sight angle calculation unit 11. The line-of-sight angle calculation unit 11 calculates the line-of-sight angle based on the plurality of pupil positions output by the pupil position detection device 1 and the second reliability common to the plurality of pupil positions. Since the line-of-sight detection device 10 calculates the line-of-sight angle based on the highly reliable pupil position evaluated by a method that is not affected by the imaging environment, the line-of-sight angle can be calculated accurately.

Embodiment 2.
FIG. 16 is a block diagram showing a configuration example of the line-of-sight detection device 10 according to the second embodiment. The line-of-sight detection device 10 according to the second embodiment has a configuration in which a first reliability calculation unit 5 and a third reliability calculation unit 7 are added to the line-of-sight detection device 10 of the first embodiment shown in FIG. Is. In FIG. 16, the same or corresponding parts as those in FIG. 1 are designated by the same reference numerals, and the description thereof will be omitted.

The first reliability calculation unit 5 and the third reliability calculation unit 7 shown in FIG. 16 are realized by the processing circuit 100 shown in FIG. 2A. Alternatively, the functions of the first reliability calculation unit 5 and the third reliability calculation unit 7 are realized by the processor 101 shown in FIG. 2B executing a program stored in the memory 102.

The pupil position detection unit 4 of the second embodiment outputs information including the positions of the eye feature points and the plurality of pupil positions to the first reliability calculation unit 5 in addition to the second reliability calculation unit 6. The second reliability calculation unit 6 of the second embodiment provides information including the positions of the eye feature points, the plurality of pupil positions, and the second reliability common to the plurality of pupil positions, in the third reliability calculation unit 7. Output to.

The first reliability calculation unit 5 calculates the brightness difference between the pixels at the pupil position and the pixels around the pupil position for each pupil position detected by the pupil position detection unit 4. The first reliability calculation unit 5 calculates the first reliability, which is an index of the certainty of the pupil position, based on the calculated luminance difference. The first reliability calculation unit 5 outputs information including the positions of the eye feature points, the plurality of pupil positions, and the first reliability of each pupil position to the third reliability calculation unit 7. The pupil position evaluation method using the first reliability is an evaluation method based on the idea that the pupil is darker than the periphery (iris, etc.). An example of the first reliability calculation method will be described with reference to FIGS. 17 and 18.

FIG. 17 is an explanatory diagram showing an example of the pupil position image unit U_11 in the first reliability calculation region U_10 and the brightness value distribution around it. In FIG. 17, the pupil position image unit U_11 is an image unit U corresponding to the pupil position PP detected by the pupil position detection unit 4. The larger the value, the brighter the brightness value in the image unit U, and the smaller the value, the darker the brightness value. The first reliability calculation unit 5 sets an image unit (hereinafter, referred to as “first reliability calculation area U_10”) of a predetermined region centered on the pupil position image unit U_11 with respect to the eye region image I2. do. For example, the first reliability calculation area U_10 is a circular area having a radius of 4 image units. Since the pupil is circular and expands or contracts depending on the external light environment, it is desirable that the first reliability calculation region U_10 is also circular, and the size is at least equal to or larger than the size of a standard human pupil. Desirably the size.

FIG. 18 is an explanatory diagram showing an example of concentric image unit groups U_12, U_13, U_14, and U_15 set in the first reliability calculation area U_10. The first reliability calculation unit 5 sets m concentric image unit groups centered on the pupil position image unit U_11 with respect to the first reliability calculation region U_10. m is any integer greater than or equal to 1. In FIG. 18, m = 4, and the first reliability calculation unit 5 sets four concentric image unit groups U_12, U_13, U_14, and U_15 centered on the pupil position image unit U_11. If one image unit U is one pixel, the m image unit group becomes the "m pixel group".

The first reliability calculation unit 5 selects the darkest image unit U_12a from the concentric image unit group U_12. The brightness value of the image unit U_12a is 21. Further, the first reliability calculation unit 5 selects the darkest image unit U_13a from the concentric image unit group U_13. The brightness value of the image unit U_13a is 24. Further, the first reliability calculation unit 5 selects the darkest image unit U_14a from the concentric image unit group U_14. The brightness value of the image unit U_14a is 26. Further, the first reliability calculation unit 5 selects the darkest image unit U_15a from the concentric image unit group U_15. The brightness value of the image unit U_15a is 51. In this way, the first reliability calculation unit 5 selects m image units from the m image unit group.

Next, the first reliability calculation unit 5 selects the brightest image unit U_15a from the selected m = 4 darkest image units U_12a, U_13a, U_14a, U_15a. Next, the first reliability calculation unit 5 calculates the brightness difference 30 (= 51-21) between the brightness value 51 of the brightest image unit U_15a selected above and the brightness value 21 of the pupil position image unit U_11. Since the pupil is darker than the periphery (iris, etc.), the reliability of the pupil position PP corresponding to the pupil position image unit U_11 is high when the difference between the two luminance values is large. The first reliability calculation unit 5 may use the value of the luminance difference as it is as the first reliability, or may calculate the first reliability according to the value of the luminance difference. When calculating the first reliability according to the luminance difference value, the larger the luminance difference value is, the larger the first reliability is calculated, and the smaller the luminance difference value is, the smaller the first reliability is. Is calculated to.

When the pupil position detection unit 4 detects a plurality of pupil positions, the first reliability calculation unit 5 calculates the first reliability for each of the plurality of pupil positions.

The third reliability calculation unit 7 calculates the third reliability using the first reliability calculated by the first reliability calculation unit 5 and the second reliability calculated by the second reliability calculation unit 6. .. The third reliability calculation unit 7 outputs information including the position of the eye feature point, the plurality of pupil positions, and the third reliability of each pupil position to the line-of-sight angle calculation unit 11. A method of calculating the third reliability will be described with reference to FIG.

FIG. 19 is an explanatory diagram showing an example of a method for calculating the third reliability. Here, it is assumed that the pupil position detection unit 4 has detected the pupil positions PP4, PP5, and PP6 shown in FIG. 12A. The first reliability calculation unit 5 calculates that the first reliability of the pupil position PP4 is 60%, the first reliability of the pupil position PP5 is 90%, and the first reliability of the pupil position PP6 is 55%. The second reliability calculation unit 6 calculates that the second reliability common to the pupil positions PP4, PP5, and PP6 is 30%. In this case, the third reliability calculation unit 7 uses a predetermined first reliability weight (for example, 0.8) and a second reliability weight (for example, 0.2) to be used in FIG. As shown in the above, the third reliability of the pupil position PP4 is calculated as 54%, the third reliability of the pupil position PP5 is 78%, and the third reliability of the pupil position PP6 is calculated as 50%.

In the above description, the third reliability calculation unit 7 calculates the weighted average value obtained by weighted averaging the first reliability and the second reliability as the third reliability, but the third reliability is calculated by another method such as a simple average. The reliability may be calculated.

The line-of-sight angle calculation unit 11 acquires information output by the pupil position detection device 1 including the position of the eye feature point, the plurality of pupil positions, and the third reliability of each of the pupil positions. The line-of-sight angle calculation unit 11 calculates the line-of-sight angle φ of the subject using the acquired information. For example, when the pupil position detection device 1 outputs the above information regarding only the right eye, the line-of-sight angle calculation unit 11 selects the pupil position having the highest third reliability from the plurality of pupil positions included in this information. If the third reliability of the selected pupil position is equal to or higher than a predetermined threshold value, the line-of-sight angle φ is calculated based on the selected pupil position. Further, for example, when the pupil position detecting device 1 outputs the above information regarding the right eye and the above information regarding the left eye, the line-of-sight angle calculation unit 11 has a third reliability among the plurality of pupil positions included in the information. Selects the largest pupil position, and if the third reliability of the selected pupil position is equal to or greater than the above threshold value, the line-of-sight angle φ is calculated based on the selected pupil position. The line-of-sight angle calculation unit 11 does not calculate the line-of-sight angle φ when there is no pupil position having a third reliability that is equal to or higher than the above threshold value.

The pupil position detecting device 1 shown in FIG. 16 has a configuration in which a third reliability is calculated from the first reliability and the second reliability and the third reliability is output to the line-of-sight angle calculation unit 11. However, the first reliability and the second reliability may be output to the line-of-sight angle calculation unit 11. In the case of this configuration, the pupil position detecting device 1 does not include the third reliability calculation unit 7. Further, in the case of this configuration, for example, when the second reliability calculated for a certain eye area image I2 is less than a predetermined threshold value, the line-of-sight angle calculation unit 11 discards the eye area image I2 and the line of sight. The angle φ is not calculated. On the other hand, when the second reliability calculated for the eye area image I2 is equal to or higher than the above threshold value, the line-of-sight angle calculation unit 11 has the highest reliability among the pupil positions detected from the eye area image I2. The line-of-sight angle φ is calculated based on the large pupil position.

FIG. 20 is a flowchart showing an operation example of the line-of-sight detection device 10 according to the second embodiment. The process shown in FIG. 20 is repeatedly executed when a predetermined condition is satisfied (for example, when the ignition power of the vehicle equipped with the line-of-sight detection device 10 is turned on). Since the operations of steps ST1 to ST4 in FIG. 20 are the same as the operations of steps ST1 to ST4 in FIG. 15, the description thereof will be omitted.

In step ST11, the first reliability calculation unit 5 calculates the brightness difference between the pixels of the pupil position PP and the pixels around the pupil position PP for each pupil position PP detected by the pupil position detection unit 4. The first reliability, which is an index of the certainty of the pupil position PP, is calculated based on the brightness difference. The operation of step ST4 and the operation of step ST11 may be performed in the reverse order or in parallel. May be done in.

Next, the third reliability calculation unit 7 applies the first reliability for each pupil position PP calculated by the first reliability calculation unit 5 and the plurality of pupil position PPs calculated by the second reliability calculation unit 6. The weighted average with the common second reliability is calculated to obtain the third reliability (step ST12). Next, the line-of-sight angle calculation unit 11 selects the pupil position PP having the highest third reliability value calculated by the third reliability calculation unit 7 among the plurality of pupil position PPs detected by the pupil position detection unit 4. Based on this, the line-of-sight angle φ of the subject is calculated (step ST13).

As described above, the pupil position detecting device 1 according to the second embodiment includes the first reliability calculation unit 5. The first reliability calculation unit 5 calculates the brightness difference between the pixel at the pupil position and the pixel around the pupil position for each pupil position detected by the pupil position detection unit 4, and based on the brightness difference. The first reliability, which is an index of the certainty of the pupil position, is calculated. Even if the brightness value of each pixel in the eye region is affected by the imaging environment, the brightness difference between the pixel at the pupil position and the pixels around the pupil position is not affected by the imaging environment. No. 1 can evaluate the reliability of the pupil position without being affected by the imaging environment.

Further, the pupil position detection device 1 according to the second embodiment includes a third reliability calculation unit 7. The third reliability calculation unit 7 calculates the third reliability using the first reliability calculated by the first reliability calculation unit 5 and the second reliability calculated by the second reliability calculation unit 6. calculate. By combining the first reliability and the second reliability with different evaluation methods, the pupil position detecting device 1 can evaluate the reliability of the pupil position more accurately.

In the above description, the image acquisition unit 2, the eye area extraction unit 3, the pupil position detection unit 4, the first reliability calculation unit 5, the second reliability calculation unit 6, the third reliability calculation unit 7, and the line-of-sight angle calculation unit Although the functions of 11 are integrated in the in-vehicle device mounted on the vehicle, they may be distributed to the server device on the network, the mobile terminal such as a smartphone, the in-vehicle device, and the like.

It should be noted that, within the scope of the disclosure, it is possible to freely combine the embodiments, modify any component of each embodiment, or omit any component of each embodiment.

The line-of-sight detection device according to the present disclosure includes a driver monitoring device that monitors the state of a driver in a vehicle, an occupant monitoring device that monitors the state of each occupant including a driver in a vehicle, and the like. Suitable for use.

1 pupil position detection device, 2 image acquisition unit, 3 eye area extraction unit, 4 pupil position detection unit, 5 1st reliability calculation unit, 6 2nd reliability calculation unit, 7 3rd reliability calculation unit, 10 line-of-sight detection Device, 11 line-of-sight angle calculation unit, 12 image pickup device, 21 and 22, circumscribing rectangle, 100 processing circuit, 101 processor, 102 memory.

Claims

An image acquisition unit that acquires an image captured by the subject,
An eye area extraction unit that extracts an eye area from an image acquired by the image acquisition unit, and an eye area extraction unit.
A pupil position detection unit that detects a plurality of pupil positions from the eye region extracted by the eye region extraction unit, and a pupil position detection unit.
The degree of spread of the plurality of pupil positions is calculated based on the positional relationship of the plurality of pupil positions detected by the pupil position detection unit, and the degree of certainty of the plurality of pupil positions is an index based on the degree of spread. 2 A pupil position detecting device including a second reliability calculating unit for calculating reliability.
The pupil position detecting device according to claim 1, wherein the second reliability calculation unit sets circumscribed rectangles circumscribing the plurality of pupil positions, and sets the length of the diagonal line of the circumscribed rectangle as the degree of spread. ..
For each pupil position detected by the pupil position detection unit, the brightness difference between the pixel at the pupil position and the pixels around the pupil position is calculated, and the brightness difference is used as an index of the certainty of the pupil position. The pupil position detecting device according to claim 1, further comprising a first reliability calculating unit for calculating the first reliability.
The first reliability calculation unit sets m concentric pixel groups (m is an integer of 1 or more) centered on the pupil position detected by the pupil position detection unit, and sets the m pixel groups. The pupil position detecting device according to claim 3, wherein the darkest pixel is selected from each of the above, and the brightness difference between the brightest pixel among the selected m pixels and the pixel at the pupil position is calculated.
A third reliability calculation unit that calculates a third reliability using the first reliability calculated by the first reliability calculation unit and the second reliability calculated by the second reliability calculation unit. The pupil position detecting device according to claim 3, further comprising.
The pupil position detecting device according to claim 1 and
A line-of-sight detection device including a line-of-sight angle calculation unit that calculates a line-of-sight angle based on a plurality of pupil positions output by the pupil position detection device and a second reliability common to the plurality of pupil positions.
The image acquisition unit acquires the image captured by the subject,
The eye area extraction unit extracts the eye area from the image acquired by the image acquisition unit.
The pupil position detection unit detects a plurality of pupil positions from the eye region extracted by the eye region extraction unit.
The second reliability calculation unit calculates the degree of expansion of the plurality of pupil positions based on the positional relationship of the plurality of pupil positions detected by the pupil position detection unit, and the plurality of pupil positions based on the degree of expansion. A pupil position detection method for calculating a second reliability, which is an index of certainty.