WO2012039139A1 - 瞳孔検出装置及び瞳孔検出方法 - Google Patents
瞳孔検出装置及び瞳孔検出方法 Download PDFInfo
- Publication number
- WO2012039139A1 WO2012039139A1 PCT/JP2011/005336 JP2011005336W WO2012039139A1 WO 2012039139 A1 WO2012039139 A1 WO 2012039139A1 JP 2011005336 W JP2011005336 W JP 2011005336W WO 2012039139 A1 WO2012039139 A1 WO 2012039139A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- pupil
- image
- value
- resolution
- calculated
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/18—Eye characteristics, e.g. of the iris
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/18—Eye characteristics, e.g. of the iris
- G06V40/19—Sensors therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/59—Context or environment of the image inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
- G06V20/597—Recognising the driver's state or behaviour, e.g. attention or drowsiness
Definitions
- the present invention relates to a pupil detection device and a pupil detection method.
- erroneous detection may occur frequently because the resolution of the pupil part is too low.
- the angle of view of the imaging device increases to capture a wide range, and the resolution of the eye area cannot be secured sufficiently.
- the first feature that the luminance of the pupil portion is lower than the luminance around the pupil portion or the second feature that the pupil is circular or elliptical is used.
- the pupil contour on the image may be a polygon and may not be a circle or an ellipse. If pupil detection is performed using the above-described second feature in such a state, erroneous detection frequently occurs.
- the resolution is low, the shape is different from the pupil as described above, but a portion whose luminance is as low as the pupil portion is erroneously detected as the pupil.
- Patent Document 1 an eye area is detected from a face image, and zoom processing is performed on the detected eye area. Then, by photographing an eye region in which the edge of the pupil is sufficiently observed to be observed, the resolution of the pupil is ensured as much as necessary for pupil detection.
- Patent Document 2 there is a technique disclosed in Patent Document 2 as a technique for detecting an iris that is a part of an eye.
- an iris contour is detected in advance from the eye region, and the image of the eye region is converted so that the iris contour becomes a circle having a predetermined size.
- the zoom magnification when the zoom magnification is insufficient, if the pupil is small, the number of pixels at the edge of the pupil contour portion is reduced, and pupil detection becomes unstable. Therefore, it is necessary to set the zoom magnification so that a sufficient number of pixels with an edge can be secured even when the pupil is small.
- the zoom magnification is set so that a sufficient number of edge pixels can be secured even when the pupil is small, when the pupil is large, an excessive number of edge pixels is detected with respect to the originally required number of edge pixels. End up. At this time, since the number of pixels more than necessary for the detection performance is processed, there is a problem that the amount of calculation required for the pupil detection process increases.
- An object of the present invention is to provide a pupil detection device and a pupil detection method that can improve pupil detection accuracy even when the detection target image has a low resolution.
- a pupil detection device is a pupil detection device that detects an image of a pupil, and obtains an actual scale value of a peripheral region including the pupil, and calculates an actual scale predicted value of a pupil diameter.
- a normalization means for calculating an enlargement / reduction magnification based on the image, normalizing the image of the peripheral region based on the calculated enlargement / reduction magnification, and the pupil of the pupil from the normalized image of the peripheral region Detecting means for detecting an image.
- the pupil detection method of one aspect of the present invention is a pupil detection method for detecting an image of a pupil, acquires an actual scale value of a peripheral region including the pupil, calculates an actual scale predicted value of a pupil diameter, A target value of resolution is calculated based on the calculated actual scale predicted value, and an enlargement / reduction ratio is calculated based on the calculated target value of resolution and the actual scale value of the surrounding area, and the calculated The image of the peripheral area is normalized based on the enlargement / reduction magnification, and the pupil image is detected from the normalized image of the peripheral area.
- the present invention it is possible to provide a pupil detection device and a pupil detection method that can improve pupil detection accuracy even when the detection target image has a low resolution.
- Block diagram showing the configuration of the eye region detection unit Flow chart for explaining the operation of the pupil detection device
- the figure which shows the face image which is the target image Diagram showing an example of a resolution table The block diagram which shows the structure of the pupil detection apparatus based on Embodiment 2 of this invention.
- FIG. 1 is a block diagram showing a configuration of pupil detection apparatus 100 according to Embodiment 1 of the present invention.
- the pupil detection device 100 is provided, for example, in a passenger compartment of an automobile and is used by being connected to a line-of-sight detection device.
- This line-of-sight detection apparatus determines the pupil position based on the detection result of the pupil detection apparatus 100 and detects the line-of-sight direction of the driver.
- a case where the pupil detection device 100 is applied to a visual line detection device will be described.
- a pupil detection device 100 includes an eye region image acquisition unit 101, an eye region actual size calculation unit 102, a pupil state prediction unit 103, an actual size pupil diameter storage unit 104, a necessary resolution estimation unit 105, and a table storage.
- Unit 106 eye region image normalization unit 107, pupil detection unit 108, and pupil actual size calculation unit 109.
- the eye area image acquisition unit 101 acquires an eye area image and outputs it to the eye area actual size calculation unit 102.
- the eye area image acquisition unit 101 includes an image input unit 111 and an eye area detection unit 112.
- the image input unit 111 images an imaging target (that is, a person here).
- the image input unit 111 includes a stereo camera, and acquires a stereo image by the stereo camera.
- the target image data is output to the eye area detection unit 112.
- the image input unit 111 is installed in front of the driver's seat, for example, on a car handle or on a dashboard. As a result, the face of the driving driver is photographed by the image input unit 111.
- the eye area detection unit 112 detects an eye area image from the target image acquired from the image input unit 111.
- the eye region detection unit 112 includes a face detection unit 121, a face part detection unit 122, and an eye region determination unit 123, as shown in FIG.
- the face detection unit 121 detects a face image from the target image acquired from the image input unit 111, and outputs the detected face image data to the face part detection unit 122.
- the face part detection unit 122 detects a face part group (that is, the mouth corner, the corner of the eye, the eyes, etc.) from the face image data acquired from the face detection part 121, and determines the position coordinates of each face part together with the face image data to determine the eye area. Output to the unit 123.
- a face part group that is, the mouth corner, the corner of the eye, the eyes, etc.
- the eye area determination unit 123 acquires the position coordinates of each face part from the face part detection unit 122, and based on the acquired position coordinates of each face part, the position and size (width / height) of the eye area in the face image. Determine). Then, the eye area determination unit 123 cuts out the eye area image from the face image and sets it as the eye area image. The position and size of the eye area in the face image are output to the eye area actual size calculator 102 together with the eye area image as an eye area detection result. The position and size of the eye area are calculated for each of the right eye and the left eye.
- the eye area actual size calculation unit 102 determines the eye area based on the eye area image data (stereo image data in the first embodiment) acquired from the eye area image acquisition unit 101 and the eye area detection result. Calculate the actual scale value.
- the actual scale value of the eye area is a value representing how much the eye area on the target image actually has.
- the actual scale value of the eye area is, for example, as the width and height of the eye area, the actual size (for example, 30 mm in width and 20 mm in height) of the object (face area around the eyes) or 1 on the eye area image. It is represented by a distance per pixel (for example, 1 pixel 0.75 mm).
- the pupil state prediction unit 103 calculates an actual scale prediction value of the pupil diameter. Specifically, the pupil state prediction unit 103 determines the actual size of the pupil in the eye region image acquired by the eye region image acquisition unit 101 based on the past actual size pupil diameter held in the actual size pupil diameter storage unit 104. To calculate an actual scale prediction value.
- the actual size pupil diameter storage unit 104 stores the actual size of the pupil acquired from the pupil actual size calculation unit 109 together with the photographing time. That is, the actual size pupil diameter storage unit 104 holds the pupil diameter (history) derived from the pupil image acquired in the past by the eye region image acquisition unit 101.
- the required resolution estimation unit 105 calculates a target value of resolution based on the actual scale prediction value calculated by the pupil state prediction unit 103.
- a resolution table stored in the table storage unit 106 is used.
- an actual scale value candidate group of pupil diameters is associated with an image resolution necessary for pupil detection in each actual scale value candidate. Therefore, the necessary resolution estimation unit 105 uses the resolution table to determine the resolution target value based on the actual scale prediction value calculated by the pupil state prediction unit 103 and the necessary resolution corresponding to the actual scale prediction value. calculate.
- the table storage unit 106 holds the resolution table described above.
- the resolution table is a table in which a plurality of pupil diameters are associated with resolutions that are calculated in advance through experiments or simulations and that are necessary for stably detecting pupils with each pupil diameter.
- images of each pupil diameter are taken at a plurality of resolutions, the resolution with the best pupil detection result is selected from each resolution image, and that resolution is used as the pupil resolution for each pupil diameter. It is done by attaching.
- the eye area image normalization unit 107 calculates the enlargement / reduction magnification based on the target value of the resolution calculated by the necessary resolution estimation unit 105 and the actual scale value of the eye area calculated by the eye area actual size calculation unit 102. calculate. Specifically, the eye region image normalization unit 107 is used for the actual scale value of the eye region calculated by the eye region actual size calculation unit 102 and the eye region image acquired from the eye region image acquisition unit 101. Based on the number of pixels, the number of pixels per unit length is calculated, and the ratio between the calculated number of pixels per unit length and the target value of the resolution calculated by the necessary resolution estimation unit 105 is obtained. Calculate the reduction ratio.
- the eye area image normalization unit 107 normalizes the eye area image acquired from the eye area image acquisition unit 101 based on the calculated enlargement / reduction ratio.
- the normalized eye region image (that is, the normalized eye region image) is output to the pupil detection unit 108.
- the eye region image normalization unit 107 calculates the actual size value of the eye region in the normalized eye region image and outputs it to the pupil actual size calculation unit 109.
- the pupil detection unit 108 detects a pupil image from the normalized eye region image acquired from the eye region image normalization unit 107.
- the coordinates of the pupil center and the pupil diameter in the detected pupil image are output to the line-of-sight detection unit (not shown) and the pupil actual size calculation unit 109, respectively.
- the actual pupil size calculation unit 109 calculates the actual size of the pupil from the actual size value of the eye region in the normalized eye region image acquired from the eye region image normalization unit 107 and the pupil diameter acquired from the pupil detection unit 108. . This actual size pupil diameter is held in the actual size pupil diameter storage unit 104.
- FIG. 3 is a flowchart for explaining the operation of the pupil detection device 100.
- the flow diagram of FIG. 3 also includes a processing flow in the above-described line-of-sight detection apparatus.
- the processing flow shown in FIG. 3 starts with the start of the shooting operation.
- the shooting operation may be started by a user operation, or may be started with some external signal as a trigger.
- step S201 the image input unit 111 images an imaging target (that is, a person here). Thereby, a target image is acquired.
- the image input unit 111 is, for example, a digital camera provided with a CMOS image sensor and a lens. Accordingly, the PPM (Portable Pix Map File format) image captured by the image input unit 111 is temporarily stored in an image storage unit (not shown) (for example, a memory space of a PC) included in the image input unit 111. Then, the data is output to the eye area detection unit 112 in the PPM format.
- an image storage unit for example, a memory space of a PC
- step S ⁇ b> 202 the face detection unit 121 detects a face image from the target image acquired from the image input unit 111.
- FIG. 4 is a diagram illustrating a face image that is a target image.
- the horizontal direction of the image is the X axis and the vertical direction of the image is the Y axis, and one pixel is one coordinate point.
- a candidate image that is, a feature image candidate
- the extracted feature image candidate is compared with a feature image representing a face area prepared in advance.
- the similarity is obtained, for example, as a reciprocal of the absolute value of the difference between the average face Gabor feature amount acquired in advance and the Gabor feature amount extracted by scanning the target image.
- the face detection unit 121 extracts a face area candidate group from the image 400 in FIG. 4, compares the extracted face area candidate group with a template prepared in advance, and determines the face area candidate with the highest correlation. It is detected as a face image 401.
- the face area detection processing may be performed by detecting a skin color area from the image (that is, skin color area detection) or by detecting an ellipse part (that is, ellipse detection). However, it may be performed by using a statistical pattern identification method. In addition, any method may be employed as long as the technique can perform the face detection.
- step S ⁇ b> 203 the face part detection unit 122 detects a face part group (that is, mouth corner, corner of the eye, an eye, etc.) from the face image acquired from the face detection part 121, and sets the position coordinates of each face part to the eye region determination part 123. Output to.
- the search area for the face part group is the face image 401 specified in step S202.
- FIG. 4 shows a face component group 402.
- a two-dimensional coordinate such as an end point of a face part such as a mouth corner, an eye corner, or an eye or a nostril is detected using a separability filter.
- the learning unit learns the correspondence between a plurality of face images and the position of the face part corresponding to the face image in advance, and the face part detection unit 122 receives the correspondence relation when the face image 401 is input.
- the part with the highest likelihood of the above may be detected as a face part.
- the face part detection unit 122 may search for a face part from the face image 401 using a standard face part template.
- step S ⁇ b> 204 the eye region determination unit 123 determines an eye region from the face image acquired from the face detection unit 121 and the face component group acquired from the face component detection unit 122.
- a rectangular area 403 including the corners of the eyes and the eyes is determined as the eye area, and the upper left corner coordinates and the lower right corner coordinates of the rectangle are acquired as the eye area detection results.
- the parameters indicating the position and size of the eye region rectangular upper left corner coordinates and lower right corner coordinates are used.
- the eye area actual size calculation unit 102 calculates the actual scale value of the eye area from the eye area image data (stereo image data in the first embodiment) acquired from the eye area image acquisition unit 101. Specifically, the eye area actual size calculator 102 calculates a feature distance (number of pixels) p between feature points on the image from the eye area image data. The feature distance between feature points on the image is, for example, the distance between the corners of the eyes and the eyes on the image. Then, the eye area actual size calculation unit 102 calculates x / p by multiplying the reference distance x by the feature distance p.
- the feature distance p is, for example, an average distance (for example, 28 mm) between the corners of the eyes and the eyes in the real space. Therefore, x / p represents an actual size value corresponding to one pixel.
- step S206 the pupil state prediction unit 103 calculates an actual scale predicted value of the pupil diameter.
- the pupil state prediction unit 103 is based on the past actual size pupil diameter held in the actual size pupil diameter storage unit 104, and the pupil state included in the eye region image acquired from the eye region image acquisition unit 101 is determined. Predict the actual size. For example, when the actual size pupil diameter one frame before is D t ⁇ 1 and the actual size pupil diameter two frames before is D t ⁇ 2 , the pupil state prediction unit 103 calculates the actual scale predicted value D t of the pupil diameter by the formula ( Calculated according to 1).
- pupil state predicting unit 103 the actual scale prediction value D t of the pupil diameter may be calculated by equation (2).
- V m is the average human miosis (here, representing the smaller pupil).
- the pupil state prediction unit 103 may perform state prediction using a Kalman filter or the like.
- step S ⁇ b> 207 the necessary resolution estimation unit 105 calculates a resolution target value based on the actual scale prediction value calculated by the pupil state prediction unit 103.
- a resolution table stored in the table storage unit 106 is used for the calculation of the resolution target value.
- the actual scale value candidate group of the pupil diameter is associated with the image resolution necessary for pupil detection in each actual scale value candidate.
- the resolution table stored in the table storage unit 106 may be held in a graph format as shown in FIG. 5A, or may be held in a table format as shown in FIG.
- the tendency characteristics of the resolution table are as follows. (1) The larger the actual size pupil diameter (that is, the actual scale value candidate), the larger the corresponding necessary resolution value. (2) The necessary resolution monotonously decreases with respect to the actual scale value candidate and converges to a constant value.
- the necessary resolution estimation unit 105 uses the resolution table based on the actual scale prediction value a calculated by the pupil state prediction unit 103 and the necessary resolution b corresponding to the actual scale prediction value.
- the resolution target value b / a is calculated.
- the resolution target value b / a is the number of pixels per unit length of an actual size necessary for stably detecting a pupil having a predicted pupil diameter.
- step S ⁇ b> 208 the eye area image normalization unit 107 enlarges based on the target value of the resolution calculated by the necessary resolution estimation unit 105 and the actual scale value of the eye area calculated by the eye area actual size calculation unit 102. / Calculate the reduction ratio.
- the eye region image normalization unit 107 is used for the actual scale value of the eye region calculated by the eye region actual size calculation unit 102 and the eye region image acquired from the eye region image acquisition unit 101.
- the number of pixels per unit length is calculated, and the ratio between the calculated number of pixels per unit length and the target value of the resolution calculated by the necessary resolution estimation unit 105 is obtained. Calculate the reduction ratio.
- the eye area image normalization unit 107 normalizes the eye area image acquired from the eye area image acquisition unit 101 based on the calculated enlargement / reduction ratio.
- a technique generally used in image processing such as bilinear or bicubic, is used.
- step S209 the pupil detection unit 108 detects a pupil image from the normalized eye region image acquired from the eye region image normalization unit 107.
- the coordinates of the pupil center and the pupil diameter in the pupil image are output to the line-of-sight detection unit (not shown) and the pupil actual size calculation unit 109, respectively.
- step S210 the line-of-sight detection unit (not shown) is calculated from, for example, a face direction vector representing the front direction of the face calculated from the coordinates of the face component group 402, and coordinates of the corners of the eyes, the eyes, and the pupil.
- the gaze direction is calculated from the gaze direction vector with respect to the front direction of the face.
- the face orientation vector is calculated by the following procedure, for example. First, the three-dimensional coordinates of the driver's facial parts group acquired in advance are converted by rotating and translating. Then, the converted three-dimensional coordinates are projected onto the target image used for pupil detection. Then, a rotation / translation parameter that most closely matches the face component group detected in step S203 is calculated. In this case, when the three-dimensional coordinates of the driver's facial parts group are acquired in advance, the set of the vector representing the direction in which the driver's face is facing and the vector rotated by the determined rotation parameter is the face orientation. Is a vector.
- the line-of-sight direction vector is calculated, for example, by the following procedure. First, when the face is facing in a predetermined direction, the driver's face parts group and the three-dimensional coordinates of the pupil center when looking in the same direction as the face direction are stored in advance. The detection of the center of the pupil is performed, for example, by taking the center of gravity of a pixel having a predetermined luminance or less in the eye region. Next, from the detected three-dimensional coordinates of the pupil, a position moved by a predetermined distance in the direction opposite to the line-of-sight direction is calculated as the eyeball center position.
- the predetermined distance is appropriately about 12 mm, which is a radius of a general adult eyeball, but is not limited to the above value, and an arbitrary value may be used.
- the three-dimensional coordinates of the eyeball center at the time of detection are calculated using the rotation / translation parameters of the face acquired in the face orientation vector calculation.
- a search is made as to which position on the sphere the detected pupil center is.
- a vector connecting the center of the eyeball and the searched point on the sphere is calculated as the line-of-sight direction.
- step S ⁇ b> 211 the actual pupil size calculation unit 109 calculates the pupil size based on the actual size value of the eye region in the normalized eye region image acquired from the eye region image normalization unit 107 and the pupil diameter acquired from the pupil detection unit 108. Calculate the actual size.
- the actual pupil size D rt is calculated by Equation (3), where n is the magnification of the normalized eye region image and n is the pupil diameter of the detected pupil on the normalized eye region image. This actual size pupil diameter is held in the actual size pupil diameter storage unit 104.
- step S212 end determination is performed.
- the end determination may be performed by manually inputting an end command, or may be performed by the pupil detection device 100 using an external signal as a trigger.
- step S212 If it is determined in step S212 that the process is to end, the process in FIG. 3 ends.
- the necessary resolution estimation unit 105 calculates the target value of the resolution based on the actual scale prediction value calculated by the pupil state prediction unit 103. Then, the eye area image normalization unit 107 enlarges / reduces based on the resolution target value calculated by the necessary resolution estimation unit 105 and the actual scale value of the eye area acquired by the eye area actual size calculation unit 102. The magnification is calculated, and the image of the eye area is normalized based on the calculated enlargement / reduction magnification. Then, the pupil detection unit 108 detects the pupil image from the eye region image normalized by the eye region image normalization unit 107.
- pupil detection is performed using a normalized image obtained by enlarging / reducing the eye area image with an enlargement / reduction magnification reflecting the substance, calculated based on the actual scale predicted value of the pupil diameter and the actual scale value of the eye area. Therefore, even if the detection target image has a low resolution, the pupil detection accuracy can be improved.
- the eye area image normalization unit 107 calculates the number of pixels per unit length based on the actual scale value of the eye area and the number of pixels used in the image of the eye area.
- the enlargement / reduction ratio is calculated by obtaining the ratio between the number of pixels and the resolution target value.
- the resolution target value is an ideal resolution
- the value corresponding to the current resolution is described as the actual scale value of the eye area.
- the present invention is not limited to this, and any actual scale value may be used as long as it is a peripheral region of the pupil and the resolution of the pupil image is not much different from that of the pupil image.
- the actual scale value of the eye area is obtained by a distance sensor.
- FIG. 6 is a block diagram showing a configuration of pupil detection apparatus 500 according to Embodiment 2 of the present invention.
- the pupil detection device 500 includes an eye area actual size calculation unit 501.
- the eye area actual size calculation unit 501 includes a distance measuring sensor, and directly detects the actual scale value of the eye area using the distance measuring sensor.
- the distance measuring sensor is, for example, a laser range sensor or a TOF (Time-Of-Flight) sensor.
- the actual scale value of the detected eye area is output to the eye area image normalization unit 107.
- the value corresponding to the current resolution is not limited to the actual scale value of the eye region, but is a peripheral region of the pupil and the resolution of the pupil image on the image is much different. Any real scale value in a non-existing region is sufficient.
- the eye area actual size calculation unit 501 includes the distance measuring sensor, and the eye area image normalization unit 107 calculates the target value and the calculated resolution.
- the enlargement / reduction magnification is calculated based on the actual scale value of the eye area measured by the distance measuring sensor, the image of the eye area is normalized based on the calculated enlargement / reduction magnification, and the pupil detection unit 108 A pupil image is detected from the converted image of the eye region.
- the actual scale value of the eye area can be detected without using an image, so that even if the detection target image has a low resolution, a more accurate actual scale value of the eye area can be acquired. .
- the pupil detection accuracy can be further improved.
- the calculation method of the actual scale predicted value of the pupil diameter is switched based on the balanced or non-balanced state of illuminance.
- FIG. 7 is a block diagram showing a configuration of pupil detection apparatus 600 according to Embodiment 3 of the present invention.
- the pupil detection device 600 includes an illuminance sensor 601 and a pupil state prediction unit 602.
- the illuminance sensor 601 measures the illuminance around the pupil detection device 600 and the imaging target at a predetermined cycle, and sequentially outputs the measured illuminance to the pupil state prediction unit 602.
- the pupil state prediction unit 602 determines whether the illuminance is in an equilibrium state or a non-equilibrium state based on the illuminance history measured by the illuminance sensor 601, and based on the determination result, the actual scale of the pupil diameter Switch the calculation method of the predicted value.
- the actual scale prediction value of the pupil diameter is set to at least two pupil diameters detected in the past by the pupil detection unit 108. Calculate based on That is, when it is determined that the illuminance is in an equilibrium state, the pupil state prediction unit 602 calculates the actual scale prediction value of the pupil diameter using the above-described equation (1).
- the pupil scale prediction unit 602 determines the actual scale prediction value of the pupil diameter based on the pupil diameter and the miosis speed detected in the past by the pupil detection unit 108. To calculate. That is, when the pupil state prediction unit 602 determines that the illuminance is in a non-equilibrium state, the pupil state prediction unit 602 calculates the actual scale prediction value of the pupil diameter using the above equation (2).
- the pupil state prediction unit 602 determines whether the illuminance is in an equilibrium state or an unbalanced state based on the measurement value history of illuminance, and based on the determination result. Then, the calculation method of the actual scale predicted value of the pupil diameter is switched.
- each functional block used in the description of each of the above embodiments is typically realized as an LSI which is an integrated circuit. These may be individually made into one chip, or may be made into one chip so as to include a part or all of them. Although referred to as LSI here, it may be referred to as IC, system LSI, super LSI, or ultra LSI depending on the degree of integration.
- the method of circuit integration is not limited to LSI, and implementation with a dedicated circuit or a general-purpose processor is also possible.
- An FPGA Field Programmable Gate Array
- a reconfigurable processor that can reconfigure the connection and setting of circuit cells inside the LSI may be used.
- the pupil detection device described in each of the above embodiments is applied to an information providing device mounted on an information terminal such as a personal computer, an OA device, a mobile phone, or a moving means such as an automobile, an airplane, a ship, or a train. Useful.
- the present invention can also be applied to a monitoring device, an alarm device, a robot, a video / audio reproduction device, and the like.
- the pupil detection device and pupil detection method of the present invention can improve pupil detection accuracy even when the detection target image has a low resolution.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Ophthalmology & Optometry (AREA)
- Human Computer Interaction (AREA)
- Eye Examination Apparatus (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
Description
[瞳孔検出装置の構成]
図1は、本発明の実施の形態1に係る瞳孔検出装置100の構成を示すブロック図である。瞳孔検出装置100は、例えば、自動車の車室内に設けられ、視線検出装置と接続されて使用される。この視線検出装置は、瞳孔検出装置100の検出結果に基づいて瞳孔位置を決定し、ドライバーの視線方向を検出する。以下では、特に、瞳孔検出装置100が視線検出装置に適用された場合について説明する。
以上の構成を有する瞳孔検出装置100の動作について説明する。図3は、瞳孔検出装置100の動作説明に供するフロー図である。図3のフロー図には、上記した視線検出装置における処理フローも含まれている。
ステップS201において、画像入力部111は、撮像ターゲット(つまり、ここでは、人物)を撮像する。これにより、ターゲット画像が取得される。
ステップS202で顔検出部121は、画像入力部111から取得されるターゲット画像から顔画像を検出する。図4は、ターゲット画像である顔画像を示す図である。なお、撮像した顔画像では、例えば、画像横方向をX軸、画像縦方向をY軸とし、1画素が1座標点である。
ステップS203で顔部品検出部122は、顔検出部121から取得される顔画像から顔部品群(つまり、口角、目尻、目頭など)を検出し、各顔部品の位置座標を目領域決定部123へ出力する。顔部品群の探索領域は、ステップS202において特定された顔画像401である。図4には、それぞれ顔部品群402が示されている。
ステップS204で、目領域決定部123は、顔検出部121から取得される顔画像と、顔部品検出部122から取得される顔部品群から、目の領域を決定する。
ステップS205において、目領域実寸算出部102は、目領域画像取得部101から取得される目領域画像データ(実施の形態1では、ステレオ画像データ)から、目領域の実スケール値を算出する。具体的には、目領域実寸算出部102は、目領域画像データから、画像上の特徴点間の特徴距離(画素数)pを算出する。この画像上の特徴点間の特徴距離とは、例えば、画像上の目尻と目頭との間の距離である。そして、目領域実寸算出部102は、基準距離xを特徴距離pによって乗算することにより、x/pを算出する。特徴距離pは、例えば、実空間における目尻と目頭との間の平均的な距離(例えば、28mm)である。従って、x/pは、1画素に相当する実寸値を表す。
ステップS206において、瞳孔状態予測部103は、瞳孔径の実スケール予測値を算出する。
ステップS207で、必要解像度推定部105は、瞳孔状態予測部103において算出された実スケール予測値に基づいて、解像度のターゲット値を算出する。解像度のターゲット値の算出には、テーブル記憶部106に記憶されている解像度テーブルが用いられる。この解像度テーブルでは、瞳孔径の実スケール値候補群と、各実スケール値候補において瞳孔検出に必要な画像解像度とが対応付けられている。
ステップS208において、目領域画像正規化部107は、必要解像度推定部105において算出された解像度のターゲット値と、目領域実寸算出部102において算出された目領域の実スケール値とに基づいて、拡大/縮小倍率を算出する。具体的には、目領域画像正規化部107は、目領域実寸算出部102で算出された目領域の実スケール値と、目領域画像取得部101から取得される目領域画像に用いられている画素数とに基づいて、単位長当たりの画素数を算出し、算出された単位長当たりの画素数と必要解像度推定部105で算出された解像度のターゲット値との比を求めることにより、拡大/縮小倍率を算出する。
ステップS209において、瞳孔検出部108は、目領域画像正規化部107から取得される正規化目領域画像から、瞳孔画像を検出する。瞳孔画像における、瞳孔中心の座標及び瞳孔径は、視線検出部(図示せず)及び瞳孔実寸算出部109へそれぞれ出力される。
ステップS211で、瞳孔実寸算出部109は、目領域画像正規化部107から取得される正規化目領域画像における目領域の実寸値と、瞳孔検出部108から取得される瞳孔径とから、瞳孔の実寸を算出する。
ステップS212においては、終了判定が行なわれる。終了判定は、人手による終了命令の入力によって行われても良いし、外的な何らかの信号をトリガに瞳孔検出装置100により行われても良い。
実施の形態2では、目領域の実スケール値が、距離センサによって求められる。
実施の形態3では、照度の平衡状態又は非平衡状態に基づいて、瞳孔径の実スケール予測値の算出方法が切り替えられる。
101 目領域画像取得部
102,501 目領域実寸算出部
103,602 瞳孔状態予測部
104 実寸瞳孔径記憶部
105 必要解像度推定部
106 テーブル記憶部
107 目領域画像正規化部
108 瞳孔検出部
109 瞳孔実寸算出部
111 画像入力部
112 目領域検出部
121 顔検出部
122 顔部品検出部
123 目領域決定部
601 照度センサ
Claims (7)
- 瞳孔の画像を検出する瞳孔検出装置であって、
前記瞳孔を含む周辺領域の実スケール値を取得する取得手段と、
瞳孔径の実スケール予測値を算出する第1の算出手段と、
前記算出された実スケール予測値に基づいて、解像度のターゲット値を算出する第2の算出手段と、
前記算出された解像度のターゲット値及び前記周辺領域の実スケール値に基づいて、拡大/縮小倍率を算出し、前記算出された拡大/縮小倍率に基づいて前記周辺領域の画像を正規化する正規化手段と、
前記正規化された周辺領域の画像から前記瞳孔の画像を検出する検出手段と、
を具備する瞳孔検出装置。 - 前記正規化手段は、前記周辺領域の実スケール値と前記周辺領域の画像に用いられている画素数とに基づいて、単位長当たりの画素数を算出し、前記算出された単位長当たりの画素数と前記解像度のターゲット値との比を求めることにより、前記拡大/縮小倍率を算出する、
請求項1に記載の瞳孔検出装置。 - 前記第1の算出手段は、前記検出手段によって過去に検出された瞳孔画像から求められる瞳孔径に基づいて、前記実スケール予測値を算出する、
請求項1に記載の瞳孔検出装置。 - 複数の瞳孔径と各瞳孔径に適した解像度とが対応付けられた解像度テーブルを記憶する記憶手段をさらに具備し、
前記第2の算出手段は、前記解像度テーブルを用いて、前記算出された実スケール予測値に対応する解像度のターゲット値を算出する、
請求項請求項1に記載の瞳孔検出装置。 - 前記取得手段は、測距センサを具備し、前記測距センサによって前記周辺領域の実スケールを計測する、
請求項1に記載の瞳孔検出装置。 - 前記第1の算出手段は、照度センサで計測された照度の履歴に基づいて、照度が平衡状態にあるのか又は非平衡状態にあるのかを判定し、判定結果に基づいて、瞳孔径の実スケール予測値の算出方法を切り替える、
請求項1に記載の瞳孔検出装置。 - 瞳孔の画像を検出する瞳孔検出方法であって、
前記瞳孔を含む周辺領域の実スケール値を取得し、
瞳孔径の実スケール予測値を算出し、
前記算出された実スケール予測値に基づいて、解像度のターゲット値を算出し、
前記算出された解像度のターゲット値及び前記周辺領域の実スケール値に基づいて、拡大/縮小倍率を算出し、
前記算出された拡大/縮小倍率に基づいて前記周辺領域の画像を正規化し、
前記正規化された周辺領域の画像から前記瞳孔の画像を検出する、
瞳孔検出方法。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP11826585.9A EP2620907B1 (en) | 2010-09-24 | 2011-09-22 | Pupil detection device and pupil detection method |
CN201180045125.0A CN103119623B (zh) | 2010-09-24 | 2011-09-22 | 瞳孔检测装置及瞳孔检测方法 |
US13/821,382 US8983235B2 (en) | 2010-09-24 | 2011-09-22 | Pupil detection device and pupil detection method |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2010-213780 | 2010-09-24 | ||
JP2010213780A JP5538160B2 (ja) | 2010-09-24 | 2010-09-24 | 瞳孔検出装置及び瞳孔検出方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2012039139A1 true WO2012039139A1 (ja) | 2012-03-29 |
Family
ID=45873637
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2011/005336 WO2012039139A1 (ja) | 2010-09-24 | 2011-09-22 | 瞳孔検出装置及び瞳孔検出方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US8983235B2 (ja) |
EP (1) | EP2620907B1 (ja) |
JP (1) | JP5538160B2 (ja) |
CN (1) | CN103119623B (ja) |
WO (1) | WO2012039139A1 (ja) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108182380A (zh) * | 2017-11-30 | 2018-06-19 | 天津大学 | 一种基于机器学习的鱼眼瞳孔智能测量方法 |
CN109711239A (zh) * | 2018-09-11 | 2019-05-03 | 重庆邮电大学 | 基于改进混合增量动态贝叶斯网络的视觉注意力检测方法 |
US10881806B2 (en) | 2015-12-14 | 2021-01-05 | Sanofi-Aventis Deutschland Gmbh | Drive mechanism for an injection device |
Families Citing this family (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101046677B1 (ko) * | 2011-03-15 | 2011-07-06 | 동국대학교 산학협력단 | 눈 위치 추적방법 및 이를 이용한 의료용 헤드램프 |
JP5746937B2 (ja) * | 2011-09-01 | 2015-07-08 | ルネサスエレクトロニクス株式会社 | オブジェクト追跡装置 |
JP6375607B2 (ja) | 2013-09-30 | 2018-08-22 | 株式会社ニコン | 電子機器、電子機器の制御方法、及び制御プログラム |
JPWO2015075894A1 (ja) * | 2013-11-19 | 2017-03-16 | 日本電気株式会社 | 撮像装置、瞳孔撮像装置、瞳孔径測定装置、瞳孔状態検出装置、及び瞳孔撮像方法 |
US9355315B2 (en) | 2014-07-24 | 2016-05-31 | Microsoft Technology Licensing, Llc | Pupil detection |
US10861180B2 (en) * | 2016-03-10 | 2020-12-08 | Ohio State Innovation Foundation | Measurements using a single image capture device |
JP6587254B2 (ja) * | 2016-09-16 | 2019-10-09 | 株式会社東海理化電機製作所 | 輝度制御装置、輝度制御システム及び輝度制御方法 |
WO2018061927A1 (ja) * | 2016-09-30 | 2018-04-05 | 日本電気株式会社 | 情報処理装置、情報処理方法およびプログラム記憶媒体 |
CN106419830A (zh) * | 2016-11-10 | 2017-02-22 | 任秋生 | 一种瞳孔直径测量方法 |
CN109725418B (zh) | 2017-10-30 | 2020-10-16 | 华为技术有限公司 | 显示设备、用于调整显示设备的图像呈现的方法及装置 |
TWI662946B (zh) * | 2018-07-02 | 2019-06-21 | 宏碁股份有限公司 | 瞳距運算裝置及方法 |
CN108986164B (zh) * | 2018-07-03 | 2021-01-26 | 百度在线网络技术(北京)有限公司 | 基于图像的位置检测方法、装置、设备及存储介质 |
WO2021024470A1 (ja) * | 2019-08-08 | 2021-02-11 | オムロン株式会社 | モニタリング装置 |
JP6717477B1 (ja) * | 2019-09-13 | 2020-07-01 | 株式会社スワローインキュベート | 画像処理方法、画像処理装置、及び画像処理プログラム |
KR20210073135A (ko) * | 2019-12-10 | 2021-06-18 | 삼성전자주식회사 | 눈 복원 기반의 눈 추적 방법 및 장치 |
CN113405505B (zh) * | 2020-03-16 | 2022-09-16 | 同方威视技术股份有限公司 | 基于多传感器的距离与身高的确定方法及其装置 |
JP6958828B1 (ja) * | 2020-12-22 | 2021-11-02 | 株式会社スワローインキュベート | 視距離推定方法、視距離推定装置、及び視距離推定プログラム |
CN116421135A (zh) * | 2023-04-27 | 2023-07-14 | 北京京东拓先科技有限公司 | 视力测试方法、装置、电子设备及存储介质 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07249197A (ja) * | 1994-03-10 | 1995-09-26 | Mitsubishi Electric Corp | 人物状態検出装置 |
JP2002282210A (ja) | 2001-03-27 | 2002-10-02 | Japan Science & Technology Corp | 視線検出方法及びその装置 |
JP2008090483A (ja) | 2006-09-29 | 2008-04-17 | Oki Electric Ind Co Ltd | 個人認証システム及び個人認証方法 |
JP2009080706A (ja) * | 2007-09-26 | 2009-04-16 | Honda Motor Co Ltd | 個人認証装置 |
JP2009282925A (ja) * | 2008-05-26 | 2009-12-03 | Sharp Corp | 虹彩認証支援装置及び虹彩認証支援方法 |
JP2010213780A (ja) | 2009-03-13 | 2010-09-30 | Kawasumi Lab Inc | 固定具、接続部材及び医療用具 |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6631208B1 (en) * | 1998-05-29 | 2003-10-07 | Fuji Photo Film Co., Ltd. | Image processing method |
AUPQ958400A0 (en) * | 2000-08-22 | 2000-09-14 | Cmx Technologies Pty Ltd | Validation of transactions |
JP4317465B2 (ja) | 2004-02-13 | 2009-08-19 | 本田技研工業株式会社 | 顔識別装置、顔識別方法及び顔識別プログラム |
EP1818869A1 (en) * | 2004-11-25 | 2007-08-15 | NEC Corporation | Face image synthesis method and face image synthesis device |
US8624962B2 (en) * | 2009-02-02 | 2014-01-07 | Ydreams—Informatica, S.A. Ydreams | Systems and methods for simulating three-dimensional virtual interactions from two-dimensional camera images |
US7912252B2 (en) * | 2009-02-06 | 2011-03-22 | Robert Bosch Gmbh | Time-of-flight sensor-assisted iris capture system and method |
-
2010
- 2010-09-24 JP JP2010213780A patent/JP5538160B2/ja active Active
-
2011
- 2011-09-22 EP EP11826585.9A patent/EP2620907B1/en active Active
- 2011-09-22 WO PCT/JP2011/005336 patent/WO2012039139A1/ja active Application Filing
- 2011-09-22 US US13/821,382 patent/US8983235B2/en active Active
- 2011-09-22 CN CN201180045125.0A patent/CN103119623B/zh active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07249197A (ja) * | 1994-03-10 | 1995-09-26 | Mitsubishi Electric Corp | 人物状態検出装置 |
JP2002282210A (ja) | 2001-03-27 | 2002-10-02 | Japan Science & Technology Corp | 視線検出方法及びその装置 |
JP2008090483A (ja) | 2006-09-29 | 2008-04-17 | Oki Electric Ind Co Ltd | 個人認証システム及び個人認証方法 |
JP2009080706A (ja) * | 2007-09-26 | 2009-04-16 | Honda Motor Co Ltd | 個人認証装置 |
JP2009282925A (ja) * | 2008-05-26 | 2009-12-03 | Sharp Corp | 虹彩認証支援装置及び虹彩認証支援方法 |
JP2010213780A (ja) | 2009-03-13 | 2010-09-30 | Kawasumi Lab Inc | 固定具、接続部材及び医療用具 |
Non-Patent Citations (1)
Title |
---|
See also references of EP2620907A4 |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10881806B2 (en) | 2015-12-14 | 2021-01-05 | Sanofi-Aventis Deutschland Gmbh | Drive mechanism for an injection device |
CN108182380A (zh) * | 2017-11-30 | 2018-06-19 | 天津大学 | 一种基于机器学习的鱼眼瞳孔智能测量方法 |
CN108182380B (zh) * | 2017-11-30 | 2023-06-06 | 天津大学 | 一种基于机器学习的鱼眼瞳孔智能测量方法 |
CN109711239A (zh) * | 2018-09-11 | 2019-05-03 | 重庆邮电大学 | 基于改进混合增量动态贝叶斯网络的视觉注意力检测方法 |
CN109711239B (zh) * | 2018-09-11 | 2023-04-07 | 重庆邮电大学 | 基于改进混合增量动态贝叶斯网络的视觉注意力检测方法 |
Also Published As
Publication number | Publication date |
---|---|
CN103119623A (zh) | 2013-05-22 |
US8983235B2 (en) | 2015-03-17 |
EP2620907A4 (en) | 2014-03-05 |
EP2620907A1 (en) | 2013-07-31 |
US20130170754A1 (en) | 2013-07-04 |
EP2620907B1 (en) | 2017-11-08 |
CN103119623B (zh) | 2016-07-06 |
JP2012068937A (ja) | 2012-04-05 |
JP5538160B2 (ja) | 2014-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5538160B2 (ja) | 瞳孔検出装置及び瞳孔検出方法 | |
JP5529660B2 (ja) | 瞳孔検出装置及び瞳孔検出方法 | |
CN106598221B (zh) | 基于眼部关键点检测的3d视线方向估计方法 | |
EP2541493B1 (en) | Pupil detection device and pupil detection method | |
JP5661043B2 (ja) | 外光映り込み判定装置、視線検出装置及び外光映り込み判定方法 | |
EP2704056A2 (en) | Image processing apparatus, image processing method | |
JP5466610B2 (ja) | 視線推定装置 | |
CN111919222B (zh) | 识别图像中的对象的装置和方法 | |
JP2006260397A (ja) | 開眼度推定装置 | |
CN108369739B (zh) | 物体检测装置和物体检测方法 | |
CN113508420A (zh) | 物体追踪装置以及物体追踪方法 | |
JP2005318546A (ja) | 画像認識システム、画像認識方法、及び画像認識プログラム | |
CN110909561A (zh) | 眼睛状态检测系统及眼睛状态检测系统的操作方法 | |
CN108734066B (zh) | 一种手势图像获取设备及方法 | |
JP6452236B2 (ja) | 眼球識別装置及び眼球識別方法 | |
KR20160146567A (ko) | 가변적으로 빠르게 움직이는 객체를 검출하는 방법 및 장치 | |
JP2012226645A (ja) | 画像処理装置および方法、記録媒体並びにプログラム | |
JP6288770B2 (ja) | 顔検出方法、顔検出システム、および顔検出プログラム | |
US20190005675A1 (en) | Methods and Apparatus for Tracking A Light Source In An Environment Surrounding A Device | |
US20160066150A1 (en) | Dynamic Configuration of a Positioning System | |
KR101909326B1 (ko) | 얼굴 모션 변화에 따른 삼각 매쉬 모델을 활용하는 사용자 인터페이스 제어 방법 및 시스템 | |
JP2015106307A (ja) | 物体検出装置 | |
EP3496390B1 (en) | Information processing device, information processing method, and storage medium | |
KR101212316B1 (ko) | 공간 변환과 움직임 추정을 이용한 프레임 중앙점의 이동 경로 추출방법 | |
EP3796257A1 (en) | Estimation device, estimation method, and computer program product |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 201180045125.0 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 11826585 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13821382 Country of ref document: US |
|
REEP | Request for entry into the european phase |
Ref document number: 2011826585 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2011826585 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |