JP3232873B2 - Gaze direction detection device for vehicles - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a gaze direction detecting device for a vehicle which measures a gaze direction of a vehicle driver in a non-contact manner and uses the measured gaze direction for an interface such as a switch operation.

[0002]

2. Description of the Related Art Conventionally, as an input device for controlling various devices on a vehicle, a manual mechanical switch, an electric switch, an electronic switch, and the like are known. In such an input device using a conventional switch, since a series of operations of searching for a switch to be operated and visually operating the switch by hand is required, the line of sight must be shifted to the switch. Attention to the front is neglected. Further, even if switches are provided on the steering column located on the front side to reduce the movement of the line of sight, as the number of switches increases, the time required for the line of sight to deviate from the front of the vehicle increases in order to identify the required switch. In addition, switches must be installed within easy reach, which limits the design of the device.

As a countermeasure against such a problem, for example, Japanese Patent Laying-Open No. 2-134130 discloses a vehicular gaze direction detecting device that measures the gaze direction of a vehicle driver in a non-contact manner and uses the measured gaze direction for an interface such as a switch operation. It has been disclosed.
In this method, a human corneal sphere is imaged by an illumination light source and a camera, the coordinates of the reflected light from the light source and the coordinates of the camera are connected, and an equation of a straight line passing through the center of the corneal sphere is obtained. The two operations are performed by a set of two illumination light sources and a camera, and the line of sight of the driver is detected by a triangulation method as a linear direction connecting the central coordinates of the corneal sphere and the central coordinates of the pupil. .

[0004]

However, in the above-mentioned line-of-sight direction detecting device, a means for detecting the degree of focus of the reflected image of the eyeball is not provided, so that the head of the vehicle driver or the like is not used. When the position fluctuates, the degree of focus of the eyeball reflection image required for gaze direction detection is reduced, and it is difficult to accurately specify the eyeball reflection image, so that the gaze direction cannot be obtained with high accuracy. There is. Therefore, in view of the above-mentioned conventional problems, the present invention provides a vehicle gaze direction detecting device that measures the gaze direction of a vehicle driver in a non-contact manner. It is an object of the present invention to provide a gaze direction detecting device for a vehicle that can accurately determine a gaze direction using an image most focused on a vehicle.

[0005]

In order to achieve the above object, according to the present invention, an invisible light is applied to a driver's eyeball and a reflected image from the eyeball is captured by an image input unit. In a gaze detection device that detects a gaze direction of a driver,
Coaxial system in which the optical axis of the lens of the image input unit and the irradiation direction match
An illumination, and a non-coaxial illumination provided at a position that is not coaxial with the lens.
Irradiating the invisible light in the bright, the image input unit each irradiation for different focus positions have the focus position adjustment function
It is configured to input each image in which lights are alternately turned on ,
A focusing degree calculating means for calculating a focusing degree of the image for each focus position; a maximum focused image selecting means for selecting an image having a maximum focusing degree from a plurality of input images; Eyeball reflection image extraction means for extracting a retinal reflection image and a corneal reflection image from data of an image having a degree, and gaze direction calculation means for calculating a gaze direction of a driver based on the retinal reflection image and the corneal reflection image. The above-mentioned focusing degree calculating means includes a retinal reflection image candidate extracting means for extracting a retinal reflection image candidate from the input image from the image input unit, and the small area including the retinal reflection image candidate as the focusing degree of the image. Calculate the degree of focus of the eyeball reflection image
The extracting means includes coaxial illumination of a small area having a maximum focus degree and
Retinal anti-retina based on difference of image data by non-coaxial illumination
The projection was extracted .

The maximum focus image selecting means outputs a focus position adjustment command to the image input unit based on a comparison of the degree of focus between images at different focus positions, and the image input unit adjusts the focus position. The focus position can be changed based on the command.

[0007]

According to a first aspect of the present invention, in the image input unit having a focus position adjusting function, a common focus position is set for each different focus position.
Each image in which the axial illumination and the non-coaxial illumination are alternately turned on is input, and the degree of focus calculation unit calculates the image of each of the focus positions in a small area including the retinal reflection image candidate extracted from the input image. The focus degree is calculated, and the maximum focus image selection means selects an image having the maximum focus degree from the plurality of input images, and the eyeball reflection image extraction means selects the retina from the data of the maximum focus image. A reflection image and a corneal reflection image are extracted. At this time, the retinal reflection image is non-coaxial with the coaxial illumination of the small area.
It is extracted based on the difference of the image data by the system illumination.
First, since the degree of focus is calculated in a small area that includes a retinal reflection image candidate that can be easily visualized even if the image is blurred to some extent, position identification is easy. The gaze direction calculating means calculates and outputs the gaze direction of the driver based on the retinal reflection image and the corneal reflection image, so that the gaze direction can be measured even when the driver's head position changes.

[0008]

1 is a block diagram showing an embodiment of a gaze direction detecting device for a vehicle according to the present invention. An image input unit 4 using a CCD or the like is installed on an instrument panel so as to image the eyeball of the driver 16. On the front surface of the image input unit 4, a lens 3 whose focus position can be arbitrarily changed is provided. At the center of the lens 3, a coaxial illumination 1 composed of a near-infrared LED or the like is provided so that the optical axis of the image input unit 4 and the irradiation direction coincide with each other. Also, the illumination 1 is placed at a position where the relative relationship with the illumination 1 is known.
A non-coaxial illumination 2 having the same specifications as the above is provided. The imaging output from the image input unit 4 is A / D converted as image data by an A / D converter 5 and stored in an image memory 6. A pupil extraction unit 7 is connected to the image memory 6, and sequentially connected to the pupil position determination unit 8, a focus degree calculation unit 9, a maximum focus image selection unit 10, a lens driving actuator 11, an eyeball reflection image extraction unit 12 , A gaze direction calculation unit 13 is connected.

The pupil extraction unit 7 extracts a retinal reflection image candidate by the illumination 1 and the illumination 2 by calculating a difference between the respective image data stored in the image memory 6. The pupil position determination unit 8 calculates and extracts the center position of the retinal reflection image candidate and sets the small area A. The focus degree calculation unit 9 calculates the focus degree for the retinal reflection image candidate in the small area A. The lens driving actuator 11 drives the lens 3 to change the focus position. The maximum in-focus image selection unit 10 determines an image having the maximum in-focus degree.

An eyeball reflection image extraction unit 12 extracts a retinal reflection image and corneal reflection images by illumination 1 and illumination 2 from an image giving the maximum degree of focus. Then, the gaze direction calculation unit 13 calculates the gaze direction of the driver from the extracted reflected image position. The illuminations 1 and 2 are connected to an illumination light emission control unit 15 for controlling the light emission thereof. The illumination light emission control unit 15 and an overall control unit 14 including the A / D converter 5 for controlling the operation of the entire apparatus are provided. Have been.

FIG. 2 is a diagram showing the overall configuration of a line-of-sight switch system of a vehicle using the above-described line-of-sight direction detecting device.
Indicates the in-vehicle layout. In FIG. 2, the gaze direction detecting device shown in FIG.
The line-of-sight direction calculation unit 13 receives H through the line-of-sight stop determination unit 23.
A UD (head-up display) display control unit 24 is connected. The controller switching unit 27 is connected to the HUD display control unit 24. An air conditioner controller 28, a CD controller 29, a radio controller 30, a headlight controller 31, and a wiper controller 32 are connected to the controller switching unit 27. An air conditioner,
Each operation target item (area) such as a wiper is displayed. H
The content of the area to be operated is displayed on the UD display unit. The main switch 21 is connected to the overall control unit 21 and the steering switch 22 is connected to the HUD display control unit 24.

When the main switch 21 is pressed, the system is turned on, and the gaze direction calculating unit 13 of the gaze direction detecting device 20 calculates the gaze direction of the driver and outputs the calculated gaze direction to the gaze stop determining unit 23. The gaze stop determining unit 23 determines the area where the gaze is stopped and outputs the determined area to the HUD display control unit 24. The HUD display control unit 24 displays the contents of the determined area on the HUD display unit 25 and controls the air conditioner controller 28 and the CD by the controller switching unit 27.
Controller 29, radio controller 30, headlight controller 31, or wiper controller 32
Switch the item to select. Subsequently, a predetermined control target is controlled based on the user input determination from the steering switch 22.

As shown in FIG. 3A, the main switch 21 and the steering switch 22 are mounted on a steering 34. In addition, illumination 1, image input unit 4,
The lighting 2 and the like are installed on an instrument panel board panel in front of the driver's seat, and a HUD display unit 25 and a line-of-sight switch area 26 are set on a windshield 33. The line-of-sight switch area 26 further displays the names of various controllers in HUD, as shown in detail in FIG. 3B. FIG. 3C shows details of the steering switch 22.

Next, the operation of the embodiment constructed as described above will be described. First, the principle of operation of extracting a retinal reflection image and a corneal reflection image required for gaze direction calculation will be described first. Of the retinal reflection image and the corneal reflection image, the retinal reflection image can specify the approximate position even when blurred to some extent, whereas the corneal reflection image makes it difficult to specify the position even when the blur is slight. Become. This is because the spatial frequency components included in both images are different. That is, since the corneal reflection image contains more high frequency components than the retinal reflection image, the out-of-focus blur acts as a high-frequency cut filter, so that the out-of-focus corneal reflection image has a large amount of power reduction and is visualized. It becomes difficult.

For this reason, taking advantage of the fact that the position of the retinal reflection image can be specified even if the focus is out of focus as described above, a small area A containing the blurred retinal reflection image candidate extracted from the image is set. A plurality of images are input while the focus position of the lens of the image input unit is shifted, and the degree of focus is calculated for the small area A of each image by frequency analysis or the like. An image to be given a degree of focus is determined.
Since the edge (pupil plane) of the retinal reflection image substantially coincides with the image formation position of the corneal reflection image, if the retinal reflection image is focused, the corneal reflection image is also focused. By extracting each reflection image from the maximum in-focus image, the gaze direction can be calculated.

Next, the operation of the gaze direction detecting device 20 will be described. FIGS. 4 and 5 are flowcharts showing the flow of the gaze direction calculation process in the gaze direction detection device 20.

In step 101, the steering 3
By pressing the main switch 21 provided in 4, the control is started. In step 102, the number n of extracted images is set to zero. In step 103, n is incremented. In step 104, the illumination 1 is turned on and the illumination 2 is turned off by the control signal of the illumination light emission control unit 15. In step 105, the image signal input from the image input unit 4 is A / D converted by the A / D converter 5 and stored in the image memory 6. This image is designated as In1 (x, y). In step 106, the illumination 1 is turned off and the illumination 2 is turned on by the control signal of the illumination light emission control unit 15. In step 107, the image signal input from the image input unit 4 is A / D converted by the A / D converter 5 and stored in the image memory 6. This image is defined as In2 (x, y).

In step 108, it is checked whether n is 1. If n is 1, the flow advances to step 109, and the flow advances to a flow for determining a retinal reflection image candidate thereafter.
When n is other than 1, the routine proceeds to step 115. In step 109, the pupil extraction unit 7 sets the image In1 (x,
y) is subtracted from the image In2 (x, y) to obtain an image In3
Generate (x, y). In step 110, the image In3
(X, y) is binarized by a fixed threshold value Th1 to generate a binary image In4 (x, y). In step 111, a labeling process is performed on the image In4 (x, y), and the regions are numbered.

In step 112, the area Ri of each region obtained as a result of the labeling is compared with predetermined thresholds S1 and S2 (S1 <S2), and S1 <Ri
<Only the region that satisfies S2 is extracted. That is, the image I
In n4 (x, y), in addition to the retinal reflection image candidate, various noises such as a spectacle lens reflection image, a spectacle frame reflection image, and a part of a face caused by a change in external illumination are included. Could be. These noises are generally irregular,
In addition, since the area is indefinite, it can be clearly distinguished from a retinal reflection image candidate that is observed as a circle or an ellipse having an expected area. Therefore, here, sorting is performed first based on the area of the region.

S1 and S2 are the expected pupil diameters (diameter of 2 to 8 mm) estimated from the photographing magnification of the image input unit 4.
Are set to the upper limit and the lower limit of the number of corresponding pixels. The spectacle lens reflection image due to light from the illumination light source is also observed as a circular area. For example, by reducing the aperture of the lens of the image input unit 4, the area of the spectacle lens reflection image is always larger than the area of the retinal reflection image candidate. Is small, the two can be identified by the area.

In step 113, the area remaining as a result of the selection processing in step 112 is selected based on the area shape. Here, the ratio F of the area of the area to the circumscribed rectangle of the area is calculated. Since the retinal reflection image is observed in a circular or elliptical shape, the ratio F is equal to or more than a certain value, whereas, for example, the spectacle frame reflection is an elongated region along the frame, and is temporarily equivalent to a retinal reflection image candidate. Are identified because F becomes smaller.
The area remaining as a result of the selection processing in step 113 is determined as a retinal reflection image candidate. The pupil extraction unit 7 determines a retinal reflection image candidate in this manner.

Next, in step 114, the pupil position determining section 8 finds the center of gravity (xg, yg) of the extracted retinal reflection image candidate, and sets the small area A centered on the centroid position and containing the retinal reflection image candidate. Set. The small area A has two points (xg-α, yg-
.beta.) and (xg + .alpha., yg + .beta.) are set at the upper left and lower right, respectively.

Subsequently, at step 115, the focusing degree calculating means 9 focuses on the retinal reflection image candidate in the area on the image Isn1 (x, y) corresponding to the small area A set by the pupil position determining section 8. Calculate the degree of focus. As an evaluation parameter indicating the degree of focus, the sum of power spectra when the small area A is Fourier transformed, the sum of power spectra at a certain spatial frequency or higher when the small area A is Fourier transformed, or the image Isn1 (x, The sum of outputs when a differential filter is applied to y) can be considered.

Here, a method of calculating the degree of focus Gn will be described by taking as an example the sum of outputs when a differential filter is applied to the image Isn1 (x, y). FIG. 6 shows an example of the differential filter.
Shown in The filter E1 of (a) is added to the image Isn1 (x, y).
Is applied, the vertical edge strength Et (x, y) of the image Isn1 (x, y) is calculated. When the filter E2 of (b) is applied to the image Isn1 (x, y), the image Isn1 is obtained.
The horizontal edge strength Ey (x, y) of (x, y) is calculated. This filtering operation is called convolution, and is executed by a product-sum operation between corresponding pixels when a 3 × 3 filter is arranged on an image. At this time, Et (x,
y) + Ey (x, y) is defined as the edge strength at point (x, y). Et (x, y) + Ey (x, y) is calculated at all points of the image Isn1 (x, y), and the sum thereof is obtained as the edge strength of the image Isn1 (x, y). In general, the higher the degree of focus of an image, the higher the number of high spatial frequency components included in the image, and thus the higher the edge intensity. Therefore, the calculated edge strength is defined as the focus degree Gn. Regardless of which evaluation parameter is used, the higher the degree of focusing, the larger the value.

In step 116, it is checked whether n is 2 or more. If n is not 2 or more, in step 117, the lens drive signal is sent from the maximum focus image selecting unit 10 to the lens drive actuator 11, and the focus position of the lens 3 is changed from the initial position by Δd. Then, the process returns to step 103, and the processes from step 104 to step 108 and step 115 are repeated, and a new focus degree Gn is obtained. If n is equal to or greater than 2 in the check in step 116, the process proceeds to step 118.
In step 118, the maximum in-focus image selection unit 10 determines whether or not the degree of focus Gn-1 has become maximum.
When Gn-1 <Gn, the process proceeds to step 119, returns to step 103, repeats steps 104 to 108 and step 115, and sets the focus degree Gn-1 when Gn-1> Gn for the first time to the maximum focus degree. Gmax, and an image that gives the maximum focus degree Gmax is a maximum focus image Imax1 (x,
y), Imax2 (x, y), and the routine proceeds to step 120.

In step 119, if G1 <G2, the focus position of the lens is changed by $ d as described above, and the routine returns to step 103. Thereafter, in step 119, the focus position is similarly changed by Δd. on the other hand,
If G1> G2, the focus position is changed by -d from the position of G2, and the process returns to step 103 again. Thereafter, in step 119, the focus position is similarly changed by-△ d.

In step 120, the eyeball reflection image extraction unit 12 generates an image Imax1 (x,
y), Imax2 (x, y), a retinal reflection image, illumination 1,
The corneal reflection image by the illumination 2 is extracted. Here, images Ismax1 (x, y) and Ismax2 corresponding to small area A
The following processing is performed only for (x, y). Ismax1
(X, y) -Ismax2 (x, y) is calculated, and the calculation result is binarized using a fixed threshold value Th1. The extracted region is a retinal reflection image. Find the position of the center of gravity of the extracted area. The obtained position of the center of gravity is defined as (xgmax1, ygmax1).

Next, the image Ismax1 (x, y) -Ismax2
A pixel having the maximum luminance gradation in (x, y) is obtained. The obtained pixels (xpsmax1, ypsmax1) are corneal reflection images due to illumination 1. Further, Ismax2 (x, y) -Ismax1
By calculating (x, y), a pixel having the maximum luminance gradation in the small area A is obtained. The obtained pixels (xpsmax2, ypsmax2)
It is a corneal reflection image by illumination 2.

Thereafter, in step 121, the line-of-sight direction calculation unit 13 extracts the three pixels extracted by the eyeball reflection image extraction unit 12.
The gaze direction of the driver is calculated from the two reflected image positions. The line-of-sight direction is calculated by the positional relationship between the three reflected images, the relative relationship between the installation positions of the two illuminations, the optical parameters (focal length, light receiving element size) of the image input unit, and the average size of the eyeball (corneal sphere radius 7. 8 mm, a distance of 4.2 mm between the center of the corneal sphere and the center of the pupil).

More specifically, as shown in the explanatory diagram of the generation of a corneal reflection image in FIG. 7, a point R 'at which a virtual corneal reflection image is generated by the illumination 1, the focal position T of the lens, and the center O of the corneal sphere are 1 It exists on the straight line L. A line segment connecting the illumination 2 and a regular reflection point Q on the corneal surface that generates a virtual image of the corneal reflection image, and a line segment connecting the point R generating a virtual image of the corneal reflection image by the illumination 2 and the focal position T of the lens. , The bisector of the angle intersects the straight line L at the point O, and the distance between Q and O is 7.8 mm. O is the center of the corneal sphere. Here, FIG. 8 is a diagram showing a modeled general eyeball configuration, in which a spherical surface having a radius of 4.2 mm around O, a straight line connecting the center of gravity of the retinal reflection image on the image and the focal position of the lens. Of the intersections on the side closer to the image input unit 4 is the pupil center P. The half line OP satisfying the above conditions gives the driver's line of sight.
As described above, the gaze direction of the driver can be calculated when the corneal reflection image and the pupil center of the coaxial illumination 1 and the corneal reflection image of the non-coaxial illumination 2 are determined.

Here, the operation after the line-of-sight direction calculation will be described by taking as an example a case where the set temperature of the air conditioner is controlled by the present apparatus. After the gaze direction is calculated in steps 101 to 121, the gaze stop determining unit 23 determines where the gaze is stopped in the gaze switch area 26. This is determined based on whether or not the same area is viewed a certain number of times in succession. If it takes 400 msec for one gaze direction calculation, for example, if the gaze direction stops for two consecutive times, that is, for about 0.8 sec, it is determined that the same area is viewed.

When the gaze stop determining unit 23 determines that the gaze direction is stopped in the air conditioner unit of the gaze switch area 26, the HUD display unit 25 displays the current air conditioner set temperature as shown in FIG. Is displayed. Where H
When the air conditioner set temperature is displayed on the UD display unit 25, the up / down switch of the steering switch 22 functions to perform up / down adjustment of the set temperature. The driver operates the up / down button to display the HUD display. Part 25
Adjust to a predetermined temperature while watching. Fixed time, eg 5
If the up-down button is not operated for a second, it is determined that the operation has been completed, and all the switch functions are stopped. When some operation is required again, the main switch 2
Pressing 1 starts the operation.

Since this embodiment is configured as described above, by immediately following the driver's head position and adopting the maximum in-focus image, even if the driver's head position fluctuates.
The gaze direction can be calculated with high accuracy. In the above embodiment, the amount of lens focus position adjustment is Δd
The lens was moved by a plurality of steps, for example, two steps, and a local maximum point was searched for. Then, only around the local maximum point, the lens was moved one step at a time, and the true local maximum point was determined. By calculating, the maximum in-focus position can be detected at high speed. Further, the resolution of the focus position detection can be increased by further reducing the focus position adjustment amount Δd of the lens in multiple steps near the maximum focus position.

[0034]

As described above, according to the present invention, a retinal reflection image candidate is extracted based on image data from an image input unit, the degree of focusing is calculated, the focus position of the lens is adjusted, and the maximum focusing is achieved. An image having a degree of focus is selected, and a retinal reflection image and a corneal reflection image are extracted from the selected maximum in-focus image and the gaze direction is calculated, so that even when the driver's head position fluctuates. And the gaze direction can be calculated with high accuracy. Since the area for calculating the degree of focus is limited to the small area including the retinal reflection image candidate, the gaze direction detection processing can be performed at high speed. Furthermore, since image data can be input using a set of coaxial illumination, non-coaxial illumination, and one imaging device, the size of the image input unit can be reduced.

[Brief description of the drawings]

FIG. 1 is a diagram showing a first embodiment of the present invention.

FIG. 2 is a diagram illustrating an overall configuration of a line-of-sight switch system of a vehicle using the line-of-sight direction detection device.

FIG. 3 is a diagram illustrating details of a vehicle-mounted layout and a main part of the vehicle-mounted layout of the eye-gaze switch system.

FIG. 4 is a diagram showing a flow of processing in a gaze direction detecting device.

FIG. 5 is a diagram showing a flow of processing in a gaze direction detecting device.

FIG. 6 is an explanatory diagram of a differential filter.

FIG. 7 is an explanatory diagram of corneal reflection image generation.

FIG. 8 is a diagram showing a modeled eyeball configuration.

FIG. 9 is an explanatory diagram of a display state on the HUD display unit.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 illumination 2 illumination 3 lens 4 image input unit 5 A / D conversion unit 6 image memory 7 pupil extraction unit 8 pupil position determination unit 9 focus degree calculation unit 10 maximum focus image selection unit 11 lens drive actuator 12 eyeball reflection image Extraction unit 13 Gaze direction calculation unit 14 Overall control unit 15 Illumination emission control unit 16 Driver 20 Gaze direction detection device 21 Main switch 22 Steering switch 23 Gaze stop determination unit 24 HUD display control unit 25 HUD display unit 26 Gaze switch area 27 Controller Switching unit 28 Air conditioner controller 29 CD controller 30 Radio controller 31 Headlight controller 32 Wiper controller 33 Windshield 34 Steering L Specular reflection point Q 'and point R' at which corneal reflection image is generated
O The center of the corneal sphere P The center of the pupil Q The specular reflection point by the non-coaxial illumination 2 Q 'The specular reflection point by the coaxial illumination 1 R The point at which a virtual corneal reflection image by the non-coaxial illumination 2 is generated R '' Point where the corneal reflection image of the virtual image by the coaxial illumination 1 occurs

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int. Cl. ⁷ , DB name) A61B 3/00-3/16 G02B 7/28

Claims (2)

(57) [Claims] An image is input by irradiating a driver's eyeball with invisible light.
The driver's line of sight by capturing the reflected image from the eyeball with the force part
In a gaze detection device that detects the direction,The optical axis of the lens of the image input unit and the irradiation direction coincide with each other.
Axial illumination and non-coaxial provided at a non-coaxial position with the lens
Irradiating the invisible light with the system lighting,  The image input unit has a focus position adjusting function and has different pin positions.
For each pointEach of the above lights was alternately turned on.the image
A focus degree calculator for calculating a focus degree of an image for each focus position
Output means and an image having the maximum focus degree from the plurality of input images.
Means for selecting the maximum focus image to be selected; and the maximum focus degreeHavingRetinal reflection image from image data
Cornea reflection imageToAn eyeball reflection image extracting means for extracting, a driver's gaze direction based on the retinal reflection image and the corneal reflection image.
Eye-gaze direction calculating means for calculating, wherein the degree of focus calculating means calculates the degree of focus from an input image from an image input unit.
A retinal reflection image candidate extracting means for extracting a retinal reflection image candidate is provided.
Including the retinal reflection image candidate as the degree of focus of the image
Calculate the focus of a small areaAnd The eyeball reflection image extracting means is configured to output the small
Difference in image data between coaxial and non-coaxial illumination of a region
Extract retinal reflection image based on minute A car characterized by
Dual-purpose gaze direction detection device. 2. The maximum focus image selection means outputs a focus position adjustment command to the image input unit based on a comparison of the degree of focus of images at different focus positions, and the image input unit adjusts the focus position. The gaze direction detecting device for a vehicle according to claim 1, wherein the focus position is changed based on a command.