CN-120954076-B - Pupil detection and blink identification method and system based on horizontal Haar characteristics

CN120954076BCN 120954076 BCN120954076 BCN 120954076BCN-120954076-B

Abstract

The invention discloses a pupil detection and blink recognition method and system based on horizontal Haar features, wherein the method comprises the steps of obtaining a current eye image, and obtaining strongest pupil-iris contrast features and corresponding image areas thereof generated in the eye image by a plurality of scale horizontal Haar feature operators; obtaining scoring results of local strongest pupil-iris contrast characteristics under each scale according to a preset scoring mechanism, obtaining the highest score in the scoring results, judging a blink state according to a preset scoring threshold value if the highest score is lower than the scoring threshold value, expanding an image area based on the highest score to obtain a pupil image otherwise, obtaining fitting results of pupil outlines in the pupil image, and selecting the pupil image and the fitting results thereof as pupil detection results. According to the invention, through multi-stage image processing and feature enhancement, accurate detection of pupils in an eye region and efficient discrimination of blink behaviors are realized.

Inventors

LI XIANGDONG
SONG YIFAN
PAN ZHENGHUA
SHAN YIFEI

Assignees

浙江大学

Dates

Publication Date: 20260512
Application Date: 20250729

Claims (6)

1. A pupil detection and blink identification method based on horizontal Haar features, comprising: The method comprises the steps of obtaining a current eye image, and obtaining strongest pupil-iris contrast characteristics and corresponding image areas thereof generated in the eye image by a horizontal Haar characteristic operator with multiple scales, wherein the horizontal Haar characteristic operator consists of three adjacent rectangular areas, the middle rectangular area is used for extracting pupil gray characteristics, and the rectangular areas on two sides are used for extracting iris gray characteristics, and the three are used for jointly constructing a gray contrast structure between the pupil and the iris; obtaining scoring results of local strongest pupil-iris contrast characteristics under each scale according to a preset scoring mechanism, and obtaining the highest score in the scoring results; According to a preset scoring threshold, if the highest score is lower than the scoring threshold, determining that the eye blink state is achieved, otherwise, expanding an image area based on the highest score to obtain a pupil image; and obtaining a fitting result of the pupil outline in the pupil image, and selecting the pupil image and the fitting result thereof as a pupil detection result.
2. The method for horizontal Haar feature-based pupil detection and blink recognition as claimed in claim 1, wherein the pupil feature and the iris feature are respectively calculated in different manners, wherein the pupil feature is represented by an average gray value of the included dark portion pixels, and the iris feature is represented by an average gray value of the included pixels.
3. The method for pupil detection and blink recognition based on horizontal Haar features as claimed in claim 1, wherein the scoring mechanism employs a ratio relationship of average gray value of iris area to average gray value of pupil area to suppress interference of iris-sclera contrast features.
4. The method for pupil detection and blink recognition based on horizontal Haar features as claimed in claim 1, wherein the process of determining the blink status includes determining the blink status if the optimal Haar feature score is below a preset threshold.
5. The method for pupil detection and blink recognition based on horizontal Haar features as claimed in claim 1, wherein the fitting of the pupil profile adopts a full-scale iterative ellipse fitting method, including removing outliers and re-fitting until there are no outliers or insufficient number of sampling points.
6. A system of horizontal Haar feature based pupil detection and blink identification method according to any one of claims 1-5, comprising: The first acquisition module is used for acquiring a current eye image, and acquiring strongest pupil-iris contrast characteristics and corresponding image areas thereof generated in the eye image by a horizontal Haar characteristic operator with a plurality of scales; the second scoring module is used for obtaining scoring results which generate local strongest pupil-iris contrast characteristics under each scale according to a preset scoring mechanism and obtaining the highest score in the scoring results; The third judging module is used for judging a blink state according to a preset scoring threshold value, if the highest score is lower than the scoring threshold value, otherwise, expanding the corresponding image area based on the highest score to obtain a pupil image; And the fourth fitting module is used for acquiring a fitting result of the pupil outline in the pupil image and selecting the pupil image and the fitting result thereof as a pupil detection result.

Description

Pupil detection and blink identification method and system based on horizontal Haar characteristics Technical Field The invention belongs to the technical field of 3D vision estimation, and particularly relates to a pupil detection and blink recognition method and system based on horizontal Haar characteristics. Background Pupil detection and blink recognition are key pre-steps in eye movement analysis. Pupil detection is required to extract pupil position and contour information from an eye image including pupil, iris, sclera and eyelid area, and blink recognition is commonly used for evaluating attention level and judging availability of eye movement signals. However, the existing pupil detection method is easy to be interfered by factors such as illumination change, shooting angle, near infrared light source reflection, pupil shielding and the like under the condition of lacking priori knowledge, and has the problem of insufficient robustness. According to the pupil detection method based on the gray threshold, the characteristic that the gray value of the pupil area is lower than that of the iris and the sclera is utilized, and the pupil area and surrounding structures in an image are segmented by setting a fixed or self-adaptive gray threshold, so that preliminary positioning of the pupil is realized; A pupil detection method based on Haar characteristics comprises the steps of constructing a Haar characteristic operator by utilizing brightness contrast characteristics between pupils and irises, selecting a region with the largest characteristic value as a pupil region, automatically learning spatial distribution characteristics of pupils from eye images by constructing a Convolutional Neural Network (CNN) or a Transformer structure to predict the central position or outline of the pupils, using high-frame-rate visual change information captured by an event camera to realize pupil detection by identifying positive and negative event edge outlines caused by pupil displacement, using an eye map (EOG) based blink recognition method, using a potential difference between a cornea and a retina to sense voltage fluctuation caused by blink to judge blink behaviors, using a Magnetic Search Coil (MSC) based blink recognition method, using a micro coil attached on the eyelids and tracking inductance response change thereof in a magnetic field to estimate three-dimensional motion tracks of the eyelids in real time to judge behaviors, using an eye key point detection based blink recognition method, using an eye key point detection method, using an eye map (EOG) to obtain a geometric change by acquiring an eye key point position, performing dynamic analysis by calculating a key point position (EAR) of an eye map (EYEASPECT RATIO), a blink recognition method based on Pupil characteristic change, wherein the method monitors Pupil diameter (PS) change in real time, and the PS signal loss is regarded as eyelid shielding caused by blink behavior. The pupil detection method based on the gray threshold value has the defects that the gray threshold value needs to be manually adjusted and is easy to be interfered by cornea reflection of a near infrared light source, the pupil detection method based on the Haar characteristic is easy to be interfered by cornea reflection of the near infrared light source, the pupil detection method based on the deep learning is weak in generalization capability of eye images acquired under different environment conditions, meanwhile, the model training cost is high, the reasoning calculation amount is large, the pupil detection method based on the eye movement event is high, the cost is high, the blink recognition method based on the Electrooculogram (EOG) is low in signal-to-noise ratio and serious in drift and needs to be frequently calibrated, the blink recognition method based on the Magnetic Search Coil (MSC) is strong in invasiveness and poor in wearing experience, is only suitable for laboratory research environments, the blink recognition method based on eye key point detection is high in error detection rate and calculation cost due to the fact that the Eye Aspect Ratio (EAR) threshold value has individual difference and needs to be manually adjusted, and the blink recognition method based on pupil characteristic change is high in error detection rate and calculation cost. Therefore, it is desirable to provide a pupil detection and blink recognition method and system based on horizontal Haar features. Disclosure of Invention In order to solve the technical problems, the invention provides a pupil detection and blink identification method and system based on horizontal Haar characteristics, which realize the accurate detection of pupils in an eye area and the efficient discrimination of blink behaviors through multi-stage image processing and characteristic enhancement. In order to achieve the above object, the present invention provides a pupil detection and blink recognition method