TWI824511B - Pupil image detection system and method thereof - Google Patents

Abstract

A pupil image detection method includes: arranging subject’s eyes in a low-light or low-visibility environment; operating an IR capturing unit to capture at least one facial IR image which includes at least one eye portion; detecting at least one pupil IR image section in the facial IR image, with the pupil IR image section including a pupil outline; and executing a deep-learning module with the pupil IR image section to build a pupil IR-image-retrieving module which is capable of detecting or positioning an eye IR image position.

Description

Pupil image detection system and method thereof

The present invention relates to a pupil (infrared image) shooting and detection system and a method thereof; in particular, to a pupil and iris (iris) image detection system and a method thereof; more particularly to a pupil and iris status (for example, ：Construction status] image detection system and method thereof.

Pupil image capturing systems and methods are commonly used, for example: U.S. Patent Publication No. US-20190191995 "Systems and methods for capturing and analyzing pupil images to determine toxicology and neurophysiology", which discloses a pupil image capturing and processing system. and methods. The pupil image capturing and processing method is used to detect a pupillary light reflex (PLR) from a target pupil.

Following the above, the pupil image shooting and processing method in the aforementioned No. US-20190191995 uses an image data to define and classify the pupil light reflex, and outputs an analytical output accordingly. And for example, the image data can be captured from a smartphone.

Following the above, the analysis results of the aforementioned US-20190191995 are used to diagnose [diagnosis] or pre-diagnosis [prognosis] of brain conditions related to neurology or psychiatry.

Another commonly used pupil image capturing system and method thereof, for example For example: U.S. Patent Publication No. US-10108877, the invention patent "System for capturing pupil and method thereof", discloses a pupil image capturing system. The pupil image capturing system is used to capture a pupil image, and the pupil image capturing system includes an image capturing unit [image obtainer], an illumination unit [illuminator] and a pupil image capturing unit [pupil extractor].

Following the above, the image capture unit of the aforementioned US-10108877 is used to obtain (i.e., photograph) the front image data (front image data) of a user, and after obtaining (i.e., photographing) the front image data At that time, the lighting unit was used to provide lighting to the user.

Continuing with the above, the aforementioned US-10108877 is photographed by the lighting unit using a light strength state, and the pupil image capturing unit is used to receive the frontal image data, and the light intensity state has one or several A predetermined reference value is used, and the pupil image capturing unit is used to perform a binary process (thresholding process) in an eye region of interest (ROI) of the frontal image data.

Another commonly used pupil image capturing system and method thereof, for example: US Patent Publication No. US-9794468 "Image sensor, image capturing apparatus, focus detection apparatus, image processing apparatus, and control method of image capturing apparatus using pupil division in different "directions" invention patent discloses an image sensor, an image capture device, a focus detection device and an image processing device.

Following the above, the image sensor of the aforementioned US-9794468 has several pixels, and each pixel has a photoelectric conversion area separated to form several areas, and the pixels are arranged to form a two-dimensional shape, and the image The sensor includes a first pixel that separates each photoelectric conversion area from a first side direction to receive a light beam passing through a pupil region, and the image sensor includes a second pixel separating each photoelectric conversion region in a second direction to receive a light beam passing through the pupil district.

Following the above, the first pixel of the aforementioned US-9794468 outputs several signals from the photoelectric conversion area separated in the first direction, and several of the signals are parallel to the first direction, and the second pixel outputs from the photoelectric conversion area The photoelectric conversion area separated in the second direction outputs a plurality of signals, and a plurality of the signals are parallel to the second direction.

Another commonly used pupil image capturing system and method, such as the invention patent No. US-9412777 "Image sensor and image capturing apparatus for satisfactory pupil division by a focus detection pixel in which a light guide is formed", It discloses an image capturing device. The image capture device includes an image sensor, and the image sensors include a plurality of image forming pixels and a lens, and the lens is configured on several of the images. Shaped pixels.

Following the above, each of the several image forming pixels in the aforementioned US-9412777 includes a plurality of allocated photo-electric conversion units (divided photo-electric conversion units), and the photo-electric conversion units have a photo-electric conversion unit. function, and converts several images through different exit pupils of the imaging optical system, and outputs a focus detection signal.

Following the above, each of the image forming pixels in the aforementioned US-9412777 includes a first light guide and a second light guide, and the first light guide is disposed on one side of the lens, and The second light guide is disposed on one side of the photo-electric conversion unit, and the allocated number of the second light guide is greater than that of the first light guide.

Another commonly used pupil image capturing system and method thereof, for example For example: US Patent Publication No. US-7999217, "Image-capturing element including photoelectric conversion cells that pupil-divide object light and output a ranging signal", discloses an image-capturing element. The image pickup element includes a pair of photoelectric conversion cells. The pair of photoelectric conversion cells is used to separate a pupil-divide object light into a first direction and a second direction in order to output a distribution signal. ranging signal].

Following the above, the photoelectric conversion cell of the aforementioned US-7999217 includes a photo-receiving element to receive object light and generate the distribution signal, and the photoelectric conversion cell includes a first light protective cover layer. light-shielding layer] and a second light-shielding layer, and the first light-shielding layer has a first light-transmitting area, and the second light-shielding layer has a second light-transmitting area district.

Following the above, the pair of photoelectric conversion cells in the aforementioned US-7999217 includes a first photoelectric conversion cell, wherein the first light transmission area is biased to the first direction and the second photoelectric conversion cell in the first photoprotective cover layer. The light transmission area is biased to the second direction in the second light protective cover layer, and the pair of photoelectric conversion cells includes a second photoelectric conversion cell, wherein the first light transmission area is biased in the first light protection cover layer. Set to the second direction and the second light transmission area on the second light protective cover layer is biased to the first direction.

Obviously, the aforementioned U.S. Patent Publication No. US-20190191995, U.S. Patent Publication No. US-10108877, No. US-9794468, No. US-9412777 and No. US-7999217 have disclosed various related pupil-related invention patents. Image capturing system and method thereof, but it still has the technical shortcomings of being unable to properly solve the problem of interfering with capturing pupil images under various environmental conditions (such as low-light environments or tracking faces or eye orientations that are deflected to various angles), and it is also unable to properly Improve the ability to photograph pupils through spectacle lenses (for example: reflections from the surface of spectacle lenses or contact lenses) Technical shortcomings of imaging.

Therefore, conventionally used pupil image capturing systems and methods must have a need to further improve the interference in capturing pupil images caused by environmental site conditions (for example, low light environments or tracking faces or eyes that are tilted to various angles), or There is also a need for how to appropriately improve the factors that affect the pupil image captured through spectacle lenses (for example, the reflection on the surface of spectacle lenses or contact lenses).

In short, the aforementioned invention patent application No. US-20190191995, US Patent Publication No. US-10108877, US-9794468, US-9412777 and US-7999217 are only It is only for reference and explanation of the current technological development status of the technical background of this invention, and is not intended to limit the scope of the invention.

In view of this, in order to meet the above needs, the present invention provides a pupil image (infrared image) detection system and method, which places a subject in a low-light environment or a low-resolution environment, and uses an infrared photography unit At least one facial infrared image is captured from the subject, and at least one pupil infrared image area is searched for in the facial infrared image, and a pupil area acquisition model is established using a deep learning model using the pupil infrared image area to improve Technical shortcomings of conventional pupil imaging systems and methods (for example: conventional use requires taking off glasses or contact lenses).

The main purpose of the preferred embodiment of the present invention is to provide a pupil image (infrared image) detection system and method, which places a subject in a low-light environment or a low-resolution environment, and uses an infrared photography unit to capture the image from the pupil image (infrared image). The subject captures at least one facial infrared image, and searches for at least one pupil infrared image area in the facial infrared image, and uses the pupil infrared image area to build a pupil area acquisition model with a deep learning model, thereby achieving the goal of achieving the pupil area. Image (infrared image) detection and the purpose of improving its detection accuracy (avoiding taking off glasses or contact lenses).

In order to achieve the above objectives, the pupil image detection system in a preferred embodiment of the present invention includes:

An infrared photography unit used to capture at least one facial infrared image from a person to be tested, and the person to be tested is appropriately in a low-light environment or a low-resolution environment;

a computing unit configured to receive the facial infrared image and search for at least one pupil infrared image area in the facial infrared image, and the pupil infrared image area includes a pupil outline;

A deep learning model, the connection of which is provided to the computing unit, and the deep learning model is used to perform representation learning calculations on the pupil infrared image area; and

A pupil area capture model is established from the pupil infrared image area using the deep learning model, and the pupil area capture model can be used to detect or locate an eye position in an infrared image.

The infrared photography unit of the preferred embodiment of the present invention is selected from a network camera, a network surveillance camera, a wireless network surveillance camera or other devices with similar infrared photography functions.

The pupil infrared image area of the preferred embodiment of the present invention includes a pupil shape, a pupil peripheral stripe or other identifiable features.

In a preferred embodiment of the present invention, the pupil area capture model is selected from a feature vector capture model or an edge detection capture model.

The deep learning model in the preferred embodiment of the present invention is selected from a pre-trained deep learning model or an unsupervised deep learning model.

In order to achieve the above objectives, the pupil image detection method in a preferred embodiment of the present invention includes:

Appropriately place a subject under test in a low-light environment or a low-visibility environment;

Using an infrared photography unit to capture at least one facial infrared image from the subject, and the facial infrared image includes at least one eye part;

Searching for at least one pupil infrared image area in the facial infrared image, and the pupil infrared image area includes a pupil outline; and

A pupil area capture model is established using a deep learning model using the pupil infrared image area, and the pupil area capture model can be used to detect or locate an eye position in an infrared image.

In a preferred embodiment of the present invention, the infrared photography unit is further combined with at least one infrared irradiation unit, and the infrared irradiation unit irradiates the person to be measured with an infrared ray.

The infrared irradiation unit of the preferred embodiment of the present invention has a predetermined wavelength of infrared ray, and the predetermined wavelength of infrared ray is between 0.76 μm and 1000 μm.

In the preferred embodiment of the present invention, the pupil area extraction model is optimized and accelerated data processing using a transfer learning model, so as to further obtain an optimized pupil area extraction model.

The eye position of the infrared image in the preferred embodiment of the present invention can be used to locate or adjust the facial features position of the infrared image, the T-shaped part position of the infrared image or other facial positions of the infrared image.

1: Test taker

10: Infrared photography unit

11: Infrared irradiation unit

20:Arithmetic unit

30: Deep learning model

31: Pupil area capture model

100: Pupil infrared image

S1: Steps

S2: Step

S3: Steps

S4: Steps

Figure 1: Block diagram of the pupil image detection system according to the first preferred embodiment of the present invention.

Figure 2: A schematic flow chart of the pupil image detection method according to the first preferred embodiment of the present invention.

Figure 3: A schematic diagram of a series of images in which the pupil image detection system and its method according to the preferred embodiment of the present invention track and operate at various facial orientations and angles and mark the position of the eye image.

Figure 4: A schematic diagram of a series of images of the pupil image detection system and its method according to the preferred embodiment of the present invention, which track and operate on the eye image position while wearing glasses.

Figure 5: A schematic diagram of a series of images of the pupil image detection system and its method according to the preferred embodiment of the present invention, which track and operate on the eye image position without wearing glasses.

Figure 6: Block diagram of the pupil image detection system according to the second preferred embodiment of the present invention.

In order to fully understand the present invention, preferred embodiments will be exemplified and described in detail below with the accompanying drawings, which are not intended to limit the present invention.

The pupil image (infrared image) detection system and method of the preferred embodiment of the present invention are suitable for use in various security systems or various medical systems and are combined with peripheral equipment products in a wired or wireless manner, such as desktop computers. , laptop computer, workstation computer, smart phone, smart watch, bracelet or other wearable device, but it is not To limit the scope of application of the present invention.

Based on the above, the pupil image (infrared image) detection system and method thereof in the preferred embodiment of the present invention are suitable for selective combination and application in various workplaces, such as: farms, aquaculture fields, and construction sites. site〕, dockyard site, machine factory, chemical factory, electricity facility site, high voltage tower maintenance site, tunnel construction site , tunnel maintenance project site, sewer maintenance project site, road maintenance project site or coal mine, but they are not used to limit the scope of application of the present invention.

Following the above, the pupil image (infrared image) detection system and method thereof according to the preferred embodiment of the present invention are also suitable and can be selectively combined and implemented in various applications. A military or dangerous disaster site, such as: bomb disposal scene, mine land, fire scene, fire investigation site, disaster site or mountain disaster Search and rescue site [mountain rescue site], but it is not used to limit the scope of application of the present invention.

Following the above, the pupil image (infrared image) detection system and method thereof in the preferred embodiment of the present invention can be selected as pre-processing for face image recognition operations, which is suitable for various face image authenticity recognition devices and related application equipment, such as : Various types of computer systems, home appliance product control systems (such as the Internet of Things), automation control systems, medical care systems, access control systems or security systems, but they are not used to limit the scope of application of the present invention.

Figure 1 shows a block diagram of a pupil image detection system according to a first preferred embodiment of the present invention. Please refer to Figure 1. For example, the pupil image detection system of the first preferred embodiment of the present invention includes an infrared photography unit 10, an arithmetic unit (or calculation unit) 20, and a deep learning model (deep learning, DL] 30 and a pupil area capture model 31.

Please refer to Figure 1 again. For example, the infrared camera unit 10 can be selected from a network camera [webcam], a network surveillance camera [IP cam], and a wireless network surveillance camera [for example: ArduCam module] Set] or other devices with similar infrared photography functions [for example: camera phones, cameras, video cameras or other devices with similar infrared photography or photography functions].

Please refer to Figure 1 again. For example, the infrared photography unit 10 is used to photograph a frontal face or an oblique angle face of at least one person to be tested 1 (or a subject) corresponding to infrared rays. Or the face with deviated eyes], and the infrared photography unit 10 can choose to automatically focus or automatically continuously focus on the person to be tested 1 (or the person being tested).

Please refer to Figure 1 again. For example, the computing unit 20 can be configured in a device, or the computing unit 20 can be selected from a workstation. A workstation computer, a desktop computer, a notebook computer, a tablet personal computer, a mobile communication device, a smartphone [smart phone] or other devices with computer functions, but this is not intended to limit the scope of the present invention.

Please refer to Figure 1 again. For example, the computing unit 20 can be connected to a communication module, and the communication module can be selected from a wired communication module or a wireless communication module. Wireless communication module, and the communication module can optionally connect to a predetermined device (for example: database device, medical equipment device, display device or other device).

Please refer to Figure 1 again. For example, the deep learning model 30 can optionally have at least one or several mathematical algorithms (mathematical algorithms), and the mathematical algorithms are selected from a linear regression (REG). model, an artificial neural network (ANN) model or other model with deep learning capabilities.

Please refer to Figure 1 again. For example, the deep learning model 30 can choose to have various deep learning architectures, and the deep learning architecture includes a deep neural network, a convolutional neural network, and a deep belief network. [deep relief network], a loop neural network or other deep learning architecture.

Please refer to Figure 1 again. For example, the pupil area capture model 31 is selected from a feature vector capture model or an edge detection capture model to obtain a pupil structure feature image, and the pupil area capture model The model 31 may further provide a pupil feature model, a pupil inspection model or other related models.

Generally speaking, the pupil is a circular hole structure, which can also be called The pupil is located in front of the vascular membrane of the eyeball and in the center of the iris. There is a pupillary sphincter (muscle inside the iris) on the annular edge of the pupil. It is a smooth muscle arranged along the annular shape. When it contracts, it shrinks the pupil to regulate and reduce the amount of light entering the eyeball. In addition, there is also a mydriasis muscle on the annular periphery of the pupil, which is a smooth muscle (internal iris muscle) arranged radially along the pupil. When relaxed (mydriasis muscle contraction), the pupil dilates to adjust and increase the amount of light entering the eyeball.

Generally speaking, the pupil outer frame usually appears as a black opening frame (black border outline), but it also has a brown, blue or other colored opening frame. In addition, the annular periphery of the pupil usually has a circular or slightly circular outline, and the pupil and its annular periphery are difficult to distinguish under dark or dim light.

Figure 2 shows a schematic flow chart of the pupil image detection method according to the first preferred embodiment of the present invention. Please refer to Figures 1 and 2, the pupil image detection method of the first preferred embodiment of the present invention includes step S1: First, for example, you can choose to use automatic [automatically], semi-automatic [semi-automatically] or manual [ Manually put the subject 1 (or the subject) in a low-light environment, a low-resolution environment (for example, dust interferes with the light environment, the face or eyes are deviated to various angles), or similar conditions environment.

Please refer to Figures 1 and 2 again. For example, the low light environment includes a low lighting environment, a shadow darkness environment or a no lighting environment, and the low visibility environment The environment includes dust interfering with the light environment, high light contrast environment, face or eyes tilted to various angles, blinking state or squinting state.

Please refer to Figures 1 and 2 again. The pupil image detection method of the first preferred embodiment of the present invention includes step S2: Next, for example, you can choose to use appropriate technical means in an automatic, semi-automatic or manual manner. The infrared photography unit 10 captures at least one or a series of (several) facial infrared images (for example: 300*300 pixel [pixel] infrared images) from the subject 1 at a predetermined distance, and the facial infrared image Include at least one eye part for subsequent eye area capture operations or pupil area capture operations.

Please refer to Figures 1 and 2 again. For example, the predetermined distance [for example: 30 cm to 40 cm or other safe shooting distances] can prevent the person to be tested 1 (or the person being tested) from being too close when shooting. Pressure is felt, and the predetermined distance prevents the head of the subject 1 (or the subject) from having to be positioned or pressed against a fixed support.

Please refer to Figures 1 and 2 again, the pupil image detection method of the first preferred embodiment of the present invention includes step S3: Then, for example, using the computing unit in an automatic, semi-automatic or manual manner using appropriate technical means [ Or the computing unit] 20 searches for at least one pupil infrared image area in the facial infrared image, and the pupil infrared image area includes a pupil outline for subsequent machine learning (ML) operations or deep learning operations.

Please refer to Figures 1 and 2 again. The pupil image detection method of the first preferred embodiment of the present invention includes step S4: Then, for example, using appropriate technical means in an automatic, semi-automatic or manual manner to utilize the pupil infrared image. The deep learning model 30 is used to build the pupil area acquisition model 31 with a predetermined architecture (for example: SSD MobileNetV2 architecture, Nvidia Jetson Nano and its suite or other architectures), and the pupil area acquisition model 31 is used to obtain a pupil infrared ray The image 100 and its pupil position, and the pupil area capture model 31 and the pupil infrared image 100 can be used to detect or locate the eye position in an infrared image.

Please refer to Figures 1 and 2 again. For example, the pupil area acquisition model 31 can be optimized using a transfer learning model [transfer learning module] or other learning models with similar optimization and acceleration functions. ization and accelerated data processing to further obtain an optimized pupil area acquisition model.

Please refer to Figures 1 and 2 again. For example, the deep learning model 30 is selected from a pre-trained deep learning model or an unsupervised deep learning model. In addition, the pupil infrared image area includes a pupil shape, a pupil peripheral stripe or other identifiable features for performing other related subsequent operations (such as pupil feature extraction, pupil medical examination or other operations).

Please refer to Figures 1 and 2 again. For example, if the head of the subject 1 (or the subject) is too tilted when shooting, the pupil infrared image 100 [select any one or two pupil infrared images] ) and the infrared image eye position can be selected to perform positioning or adjustment of an infrared image facial features position, an infrared image T-shaped position or other infrared image face position or face authenticity recognition, or the infrared image eye position can be selected to perform Perform other related follow-up operations (for example: pupil feature extraction or pupil medical examination).

Figure 3 shows a schematic diagram of a series of images of the pupil image detection system and its method according to the preferred embodiment of the present invention, which track and operate at various facial orientations deflected to various angles and mark the position of the eye image. Please refer to the upper left side of Figure 3. For example, the pupil image detection system and method thereof according to the preferred embodiment of the present invention can selectively track and operate on various facial orientations and angles without wearing glasses. Mark the eye image location.

Please refer to the lower left side of Figure 3 again. For example, the pupil image detection system and method thereof according to the preferred embodiment of the present invention are used when wearing glasses [for example: spectacle lenses with optical functions or with absorption of specific wavelengths] [for example] : Yellow light, blue light or high-energy blue ultraviolet light [eyeglass lenses] can be selected to track and operate on various face tilts to various angles and mark the eye image position.

Please refer to Figures 1, 2 and 3 again, for example, The pupil image detection system and method of the preferred embodiment of the present invention can obtain the pupil area capture model 31 and the pupil infrared image 100, which can be used to detect or locate the eye position of the infrared image, as shown on the right side of Figure 3, or other It can be used to identify the authenticity of facial images or detect pupil structure (for example: pupil contour structure and its characteristics).

Figure 4 shows a schematic diagram of a series of images of the pupil image detection system and its method according to the preferred embodiment of the present invention, which track and operate on the eye image position while wearing glasses. Please refer to Figures 1 and 4. For example, the subject 1 [or subject] can successfully obtain the pupil infrared image 100 [pupil structure infrared image] by the pupil area capture model 31 while wearing various glasses. , such as: pupil contour structure and characteristics] and infrared image eye position, that is, if the subject 1 (or the subject) wears various contact lenses [or pupil dilation films], it can be regarded as the same.

Figure 5 shows a schematic diagram of a series of images of the pupil image detection system and its method according to the preferred embodiment of the present invention, which are used to track the position of the eye image without wearing glasses. Please refer to Figures 1 and 5. For example, another subject 1 (or subject) can successfully obtain the pupil infrared image 100 [pupil structure infrared ray] without wearing glasses and in a low-light (dim) environment. Images, such as: pupil outline structure and characteristics] and infrared image eye position.

Please refer to Figures 1, 4, and 5 again. For example, in the face image authenticity identification, the pupil infrared image 100 [pupil structure infrared image, such as: pupil outline structure and its characteristics] and the infrared image eye position It can be used to confirm whether the subject 1 (or subject) is a real human head (including its face or eyes), so as to avoid using false images for face image identification.

Figure 6 shows a block diagram of the pupil image detection system of the second preferred embodiment of the present invention, which corresponds to the pupil image detection system of Figure 1. Please refer to Figure 6. Compared with the first embodiment, the present invention has The pupil image detection system of the second preferred embodiment includes an infrared photography unit 10, at least one infrared irradiation unit 11, an arithmetic unit (or calculation unit) 20, a deep learning model 30 and a pupil area capture model 31.

Please refer to Figure 6 again. For example, the infrared photography unit 10 is further combined or connected to the infrared irradiation unit 11, and the infrared irradiation unit 11 irradiates the person to be tested 1 [or the person being tested] with an infrared ray. ), and the infrared irradiation unit 11 has a predetermined wavelength infrared ray, and the predetermined wavelength infrared ray is between 0.76 μm and 1000 μm or other appropriate wavelength infrared ray range.

The foregoing preferred embodiments only illustrate the present invention and its technical features. The technology of this embodiment can still be appropriately implemented with various substantially equivalent modifications and/or substitutions; therefore, the scope of rights of the present invention shall depend on the appended patent application. The scope defined shall prevail. The copyright in this case is restricted to use for patent applications in the Republic of China.

1: Test taker

10: Infrared photography unit

20:Arithmetic unit

30: Deep learning model

31: Pupil area capture model

100: Pupil infrared image

Claims (10)

A pupil image detection system, which includes: an infrared photography unit, which is used to capture at least one face infrared image from a person to be tested, and the person to be tested is appropriately in a low-light environment or a low-resolution environment; a calculation A unit configured to receive the face infrared image and search for at least one pupil infrared image area in the face infrared image, and the pupil infrared image area includes a pupil outline; a deep learning model, the connection of which is provided in the computing unit , and the pupil infrared image area is used to perform representation learning calculations using the deep learning model; and a pupil area acquisition model is established from the pupil infrared image area using the deep learning model, and the pupil area acquisition model can be used Detect or locate an eye position in an infrared image, and the pupil area capture model is selected from a feature vector capture model or an edge detection capture model. According to the pupil image detection system described in item 1 of the patent application, the infrared photography unit is selected from a network camera, a network surveillance camera or a wireless network surveillance camera. According to the pupil image detection system described in item 1 of the patent application, the pupil infrared image area includes a pupil shape, a pupil peripheral stripe or a combination thereof. According to the pupil image detection system described in item 1 of the patent application, the pupil area capture model is selected from an optimized pupil area capture model. According to the pupil image detection system described in item 1 of the patent application, the deep learning model is selected from a pre-trained deep learning model or an unsupervised deep learning model. A pupil image detection method, which includes: appropriately placing a subject in a low-light environment or a low-resolution environment; using an infrared photography unit to capture at least one face infrared image from the subject, and the face infrared The image contains at least one eye part; Find at least one pupil infrared image area in the face infrared image, and the pupil infrared image area includes a pupil outline; and use the pupil infrared image area to build a pupil area capture model with a deep learning model, and the pupil area capture The acquisition model can be used to detect or locate the eye position in an infrared image, and the pupil area acquisition model is selected from a feature vector acquisition model or an edge detection acquisition model. According to the pupil image detection method described in item 6 of the patent application, the infrared photography unit is further combined with at least one infrared irradiation unit, and the infrared irradiation unit irradiates the person to be measured with an infrared ray. According to the pupil image detection method described in item 7 of the patent application, the infrared irradiation unit has a predetermined wavelength of infrared ray, and the predetermined wavelength of infrared ray is between 0.76 μm and 1000 μm. According to the pupil image detection method described in item 6 of the patent application, the pupil area capture model is optimized and accelerated data processing using a transfer learning model, so as to further obtain an optimized pupil area capture model. According to the pupil image detection method described in Item 6 of the patent application, the eye position of the infrared image can be used to locate or adjust the facial features position of the infrared image, the T-shaped part position of the infrared image or the face position of the infrared image.

Images