A dry eye detection method, system, device, and medium

Abstract

The application discloses a dry eye detection method, system, device and medium, frame-by-frame identification of eye lid point time sequence data, eyebrow eye point time sequence data and eye red probability in collected historical video information; calculation of eye lid distance and eyebrow eye distance to construct eye lid distance time sequence data set and eyebrow eye distance time sequence data set; calculation of blink recognition threshold and frown recognition threshold, identification of blink frames according to the eye lid distance time sequence data set and the blink recognition threshold, identification of frown frames according to the eyebrow eye point time sequence data set and the frown recognition threshold, and calculation of blink frame data of each blink frame and frown frame data of each frown frame; training of a dry eye detection model by using the blink frame data, the frown frame data and the eye red probability to obtain a target dry eye detection model for real-time detection of dry eye. The method provided by the application can provide a convenient, practical, accurate and large-scale application dry eye detection method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical treatment, in particular to a dry eye detection method, system, device and medium. BACKGROUND

[0002] Dry eye is a common eye disease, often accompanied by eye dryness, burning, stinging and foreign body sensation, etc. symptoms, and even can cause blurred vision and decline, seriously affecting the quality of life of patients. With the significant increase in the use rate of screen devices such as mobile phones and computers, the incidence of dry eye continues to be high, and in the white-collar high-risk population, the incidence is more than 50%. However, due to the large number of dry eye patients and the difficulty in completely curing them, the limited ophthalmic medical and health resources are heavily burdened, and a large number of dry eye patients have difficulty in obtaining professional diagnosis and treatment of ophthalmologists in time, leading to the aggravation of the disease, seriously affecting their quality of life, and even can induce corneal inflammation and other blinding eye diseases.

[0003] The existing dry eye detection method has the following problems: 1. It needs to be matched with hardware equipment or / and drug support, which is not convenient for daily use and wide promotion; 2. The data acquisition condition is high, and the data processing is insufficient, resulting in low accuracy of the detection result.

[0004] Therefore, how to improve the convenience, practicality and accuracy of dry eye detection has become a technical problem to be solved by the technical personnel in the field. SUMMARY

[0005] The present application provides a dry eye detection method and system to solve the technical problem of improving the convenience, practicality and accuracy of dry eye detection, and to achieve the effect of providing a convenient, practical, accurate and large-scale application of dry eye detection method.

[0006] In a first aspect, the present application provides a dry eye detection method, which comprises:

[0007] Collecting historical video information containing a face, and identifying eyelid point position time series data, eyebrow eye point position time series data and eye red probability in the historical video information frame by frame;

[0008] According to the eyelid point position time series data, the eyelid distance is calculated frame by frame to construct the eyelid distance time series data set, and according to the eyebrow eye point position time series data, the eyebrow eye distance is calculated frame by frame to construct the eyebrow eye distance time series data set;

[0009] Based on the fluctuation and blink amplitude abnormal value of the eyelid distance time series data set, the blink recognition threshold is calculated, and based on the fluctuation and eyebrow eye movement amplitude abnormal value of the eyebrow eye distance time series data set, the frown recognition threshold is calculated;

[0010] According to the size relationship between the eyelid distance time series data set and the blink recognition threshold, a blink frame is recognized, and according to the size relationship between the eyebrow-eye distance time series data set and the frown recognition threshold, a frown frame is recognized;

[0011] Based on the eyelid distance time series data set, blink frame data of each blink frame is calculated, and based on the eyebrow-eye distance time series data set, frown frame data of each frown frame is calculated;

[0012] The blink frame data, the frown frame data and the eye red probability are used to train the constructed dry eye detection model to obtain a target dry eye detection model;

[0013] In the real-time detection process, real-time blink frame data, real-time frown frame data and real-time eye red probability extracted from the collected real-time video information are input into the target dry eye detection model to obtain a dry eye detection result.

[0014] Preferably, the frame-by-frame recognition of the eyelid point position time series data, the eyebrow-eye point position time series data and the eye red probability in the historical video information comprises:

[0015] The 68 standard point positions of the human face in the video information are frame-by-frame recognized by using a dlib model:

[0016] The eyelid point position data and the eyebrow-eye point position data are selected from the 68 standard point positions;

[0017] The time series data of each frame of the video information is combined with the eyelid point position data to construct eyelid point position time series data;

[0018] The time series data is combined with the eyebrow-eye point position data to construct eyebrow-eye point position time series data;

[0019] Each frame of the video information is input into the trained eye red recognition model to obtain an eye red probability.

[0020] Preferably, the eyelid distance is defined as the difference between the vertical coordinate of the upper eyelid and the vertical coordinate of the lower eyelid;

[0021] The eyebrow-eye distance is defined as the difference between the vertical coordinate of the highest point position of the eyebrow and the vertical coordinate of the inner corner point position of the eye.

[0022] Preferably, the blink recognition threshold is calculated based on the fluctuation and blink amplitude outliers of the eyelid distance time series data set, comprising:

[0023] The global mean of the eyelid distance and the global standard deviation of the eyelid distance of the eyelid distance time series data set are calculated, and the global threshold of the eyelid distance is calculated according to the global mean of the eyelid distance and the global standard deviation of the eyelid distance.

[0024] The formula for calculating the eyelid distance global threshold value is:

[0025]

[0026] Wherein, represents the eyelid distance global threshold value, represents the eyelid distance global mean value, represents the eyelid distance global standard deviation, represents the blink recognition sensitive parameter;

[0027] The video frames of the first preset time period before and after the historical video information are selected to represent the blink amplitude abnormal frame, the eyelid distance local mean value and the eyelid distance local standard deviation of the eyelid distance time series data set corresponding to the blink amplitude abnormal frame are calculated, and the eyelid distance local threshold value is calculated according to the eyelid distance local mean value and the eyelid distance local standard deviation;

[0028] The formula for calculating the eyelid distance local threshold value is:

[0029]

[0030] Wherein, represents the eyelid distance local threshold value, represents the eyelid distance local mean value, represents the eyelid distance local standard deviation;

[0031] The eyelid distance global threshold value and the eyelid distance local threshold value are fused according to the first preset weight to obtain the blink recognition threshold value.

[0032] Preferably, the calculation of the frown recognition threshold value based on the fluctuation of the eyebrow-eye distance time series data set and the eyebrow-eye movement amplitude abnormal value comprises:

[0033] The global mean value and the global standard deviation of the eyebrow-eye distance time series data set are calculated, and the global threshold value of the eyebrow-eye distance is calculated according to the global mean value and the global standard deviation of the eyebrow-eye distance;

[0034] The formula for calculating the global threshold value of the eyebrow-eye distance is:

[0035]

[0036] Wherein, represents the global threshold value of the eyebrow-eye distance, represents the global mean value of the eyebrow-eye distance, represents the global standard deviation of the eyebrow-eye distance, represents the frown recognition sensitive parameter;

[0037] selecting video frames in a first preset time period of the first preset time period to represent eyebrow-eye movement amplitude abnormal frames, calculating a local mean of eyebrow-eye distance and a local standard deviation of eyebrow-eye distance of the eyebrow-eye distance time series data set corresponding to the eyebrow-eye movement amplitude abnormal frames, and calculating a local threshold of eyebrow-eye distance according to the local mean of eyebrow-eye distance and the local standard deviation of eyebrow-eye distance;

[0038] The formula for calculating the local threshold of eyebrow-eye distance is:

[0039]

[0040] wherein, the local threshold of eyebrow-eye distance, the local mean of eyebrow-eye distance, the local standard deviation of eyebrow-eye distance;

[0041] The global threshold of eyebrow-eye distance and the local threshold of eyebrow-eye distance are fused according to a second preset weight to obtain the frown recognition threshold.

[0042] Preferably, based on the eyelid distance time series data set, the blink frame data of each blink frame is calculated, including:

[0043] Selecting video frames in a second preset time period before and after each blink frame to represent blink-related frames, and selecting the maximum eyelid distance and the minimum eyelid distance of the blink-related frames from the eyelid distance time series data set;

[0044] The first time series interval in which the eyelid distance of each blink-related frame is less than the blink recognition threshold is calculated, and the first time series interval is taken as the blink time;

[0045] The mean value of the blink time is calculated to obtain the average blink time;

[0046] According to the size of the blink time and the preset blink threshold, it is judged whether the blink frame is complete blink or incomplete blink, and the number of complete blinks and the number of incomplete blinks in the blink frame are counted;

[0047] The complete blink time standard deviation and the complete blink time average value of the blink time corresponding to the complete blink are calculated, and the ratio of the complete blink time standard deviation and the complete blink time average value is taken as the blink interval time variability;

[0048] According to each blink time, the minimum blink time is determined;

[0049] According to the second time series interval between the minimum eyelid distance of each blink frame, the blink interval is calculated, and the maximum blink interval is selected from the blink interval.

[0050] calculating a third time series interval of the eyelid distance of each of the blink frames from less than the blink recognition threshold to the eyelid distance minimum value, taking the third time series interval as a closed-eye time;

[0051] calculating a fourth time series interval of the eyelid distance of each of the blink frames from the eyelid distance minimum value to greater than the blink recognition threshold, taking the fourth time series interval as an open-eye time;

[0052] selecting a minimum closed-eye time from the closed-eye times, and calculating an average closed-eye time according to each of the closed-eye times;

[0053] selecting a minimum open-eye time from the open-eye times, and calculating an average open-eye time according to each of the open-eye times;

[0054] taking the eyelid distance maximum value, the eyelid distance minimum value, the complete blink times, the incomplete blink times, the blink time minimum value, the average blink time, the maximum blink interval, the blink interval time variability, the minimum closed-eye time, the average closed-eye time, the minimum open-eye time, and the average open-eye time as blink frame data.

[0055] Preferably, based on the eyebrow-eye distance time series data set, calculating a frown frame data of each of the frown frames comprises:

[0056] selecting a video frame within a third preset time period before and after each of the frown frames as a frown-related frame, and selecting a maximum eyebrow-eye distance and a minimum eyebrow-eye distance corresponding to the frown-related frame from the eyebrow-eye distance time series data set;

[0057] calculating a fifth time series interval of the eyebrow-eye distance of each of the frown-related frames being less than the frown recognition threshold, and determining a frown time and a frown times according to the fifth time series interval;

[0058] calculating a frown time standard deviation and a frown time average value of the frown time, and taking a ratio of the frown time standard deviation and the frown time average value as a frown interval time variability;

[0059] determining a maximum frown time according to each of the frown times;

[0060] calculating a frown interval according to a sixth time series interval between the minimum eyebrow-eye distance of each of the frown frames, and selecting a minimum frown interval from the frown interval;

[0061] The maximum eyebrow-eye distance, the minimum eyebrow-eye distance, the frown time, the frown frequency, the frown interval time variation, the maximum frown time, and the minimum frown interval are taken as frown frame data.

[0062] In a second aspect, the present application also provides a dry eye detection system for implementing the dry eye detection method described above, which comprises a data acquisition and processing unit, a first calculation unit, an identification threshold calculation unit, an identification unit, a second calculation unit, a model training unit, and a real-time detection unit.

[0063] The data acquisition and processing unit is configured to acquire historical video information containing a face, and identify eyelid point position time sequence data, eyebrow-eye point position time sequence data, and eye red probability in the historical video information frame by frame.

[0064] The first calculation unit is configured to calculate eyelid distance frame by frame based on the eyelid point position time sequence data to construct an eyelid distance time sequence data set, and calculate eyebrow-eye distance frame by frame based on the eyebrow-eye point position time sequence data to construct an eyebrow-eye distance time sequence data set.

[0065] The identification threshold calculation unit is configured to calculate a blink identification threshold based on fluctuations and blink amplitude outliers of the eyelid distance time sequence data set, and calculate a frown identification threshold based on fluctuations and eyebrow-eye movement amplitude outliers of the eyebrow-eye distance time sequence data set.

[0066] The identification unit is configured to identify a blink frame according to the size relationship between the eyelid distance time sequence data set and the blink identification threshold, and identify a frown frame according to the size relationship between the eyebrow-eye distance time sequence data set and the frown identification threshold.

[0067] The second calculation unit is configured to calculate blink frame data of each blink frame based on the eyelid distance time sequence data set, and calculate frown frame data of each frown frame based on the eyebrow-eye distance time sequence data set.

[0068] The model training unit is configured to train a completed dry eye detection model by using the blink frame data, the frown frame data, and the eye red probability, to obtain a target dry eye detection model.

[0069] The real-time detection unit is configured to input real-time blink frame data, real-time frown frame data, and real-time eye red probability extracted from acquired real-time video information into the target dry eye detection model to obtain a dry eye detection result in a real-time detection process.

[0070] In a third aspect, the present application also provides a computer device, which comprises a memory, a processor and a transceiver connected through a bus; the memory is used for storing a set of computer program instructions and data, and transmitting the stored data to the processor; the processor executes the program instructions stored in the memory to perform the dry eye detection method described above.

[0071] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, when the computer program is executed, the dry eye detection method described above is realized.

[0072] The present application provides a dry eye detection method, system, device and medium, compared with the prior art, the beneficial effects of the embodiment of the present application are at least one of the following:

[0073] (1) A dry eye detection method is provided, which is convenient, practical, accurate and can be widely applied, so that patients can independently complete dry eye screening and obtain corresponding treatment or relief suggestions, provide timely and effective self-management scheme for patients, and improve their quality of life.

[0074] (2) The dry eye patients are screened in time, self-help diagnosis and treatment suggestions are provided for patients with mild symptoms, and referral treatment is helped for patients with moderate and severe symptoms, so as to promote the efficient allocation of medical resources. BRIEF DESCRIPTION OF DRAWINGS

[0075] Figure 1 It is a dry eye detection method step schematic diagram provided by a preferred embodiment of the present application;

[0076] Figure 2 It is a 68 key point schematic diagram of a face provided by a preferred embodiment of the present application;

[0077] Figure 3 It is a blink judgment fold line diagram provided by a preferred embodiment of the present application;

[0078] Figure 4 It is a structure schematic diagram of a dry eye detection system provided by a preferred embodiment of the present application;

[0079] Figure 5 It is a structure schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0080] The embodiments of the present application are described below with reference to the drawings, and the examples are given only for the purpose of illustration and are not to be construed as limiting the present application, and the accompanying drawings are used for reference and illustration only, and do not constitute a limitation on the scope of protection of the present application. Based on the examples in the present application, all other examples obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application. In the description of the present application, the terms "first", "second", "third" and the like are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second", "third" and the like can be explicitly or implicitly included one or more features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0081] In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are only for the purpose of illustration, and are not intended to indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be construed as limiting the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood in specific cases.

[0082] In the description of the present application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present application have the same meaning as understood by those skilled in the art. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments, and are not intended to limit the present application. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood in specific cases.

[0083] Please refer to Figure 1 In the embodiments of the present application, a dry eye detection method is provided, which comprises:

[0084] S1, collecting historical video information containing a face, and identifying eyelid point position time series data, eyebrow eye point position time series data and eye red probability in the historical video information frame by frame.

[0085] S2, calculate eyelid distance frame by frame according to the eyelid point position time series data to construct an eyelid distance time series data set, calculate eyebrow-eye distance frame by frame according to the eyebrow-eye point position time series data to construct an eyebrow-eye distance time series data set.

[0086] S3, calculate a blink recognition threshold based on the fluctuation and blink amplitude outliers of the eyelid distance time series data set, and calculate a frown recognition threshold based on the fluctuation and frown amplitude outliers of the eyebrow-eye distance time series data set.

[0087] S4, identify a blink frame according to the size relationship between the eyelid distance time series data set and the blink recognition threshold, and identify a frown frame according to the size relationship between the eyebrow-eye distance time series data set and the frown recognition threshold.

[0088] S5, calculate blink frame data of each blink frame based on the eyelid distance time series data set, and calculate frown frame data of each frown frame based on the eyebrow-eye distance time series data set.

[0089] S6, train the completed dry eye detection model using the blink frame data, the frown frame data and the eye red probability to obtain a target dry eye detection model.

[0090] S7, in the real-time detection process, input real-time blink frame data, real-time frown frame data and real-time eye red probability extracted from the collected real-time video information into the target dry eye detection model to obtain a dry eye detection result.

[0091] In the preferred embodiment of the present application, for the collection of video information, a dedicated video collection device can be used, or a smart phone can be used, which is convenient for users to detect dry eye syndrome on a daily basis. In the present application, the collection of video information by a smart phone is taken as an example for illustration. The historical video information containing a face is collected by a mobile phone. The collection of historical video information can be in the form of live recording, or the already recorded video can be uploaded in various ways. The diverse video information collection methods improve the convenience of information collection. The video collection time can be set according to the requirements, and is generally 1 minute.

[0092] After the collection of historical video information is completed, the information related to the face in the historical video information is identified frame by frame. In the embodiment of the present application, the dlib model is used to identify 68 standard points of facial features in the video information frame by frame. The dlib model is a C++ library containing machine learning algorithms, which is particularly good at image processing and deep learning, and is widely used in the fields of computer vision, machine learning and data analysis. In the field of face recognition, dlib can detect 68 key points of a face (such as Figure 2The key points are crucial for subsequent eyelid feature extraction and eyebrow-eye feature extraction. The dlib face key point detector has the characteristics of fast speed and high accuracy. The eyelid point data is selected from the 68 standard points, and the eyebrow-eye point data is selected from the 18-27 standard points and the 37-48 standard points of the eyelid point data. The time sequence data of each frame of video information is combined with the eyelid point data to construct the eyelid point time sequence data, and the time sequence data is combined with the eyebrow-eye point data to construct the eyebrow-eye point time sequence data.

[0093] For obtaining the eye red probability, each frame of video information is input into the trained eye red recognition model to obtain the eye red probability. The eye red recognition model is generally constructed in combination of image processing and computer vision recognition, and is used for detecting and identifying the probability of the eye being in a red state. Common model construction algorithms include support vector machine, random forest and deep learning classifier.

[0094] Further, after identifying the historical video information, the eyelid distance is calculated frame by frame based on the eyelid point time sequence data. The eyelid distance is defined as the difference between the upper eyelid longitudinal coordinate and the lower eyelid longitudinal coordinate, as shown in Figure 2 For the right eye, the difference between the longitudinal coordinates of the two groups of upper and lower eyelid points 38 and 42, and 39 and 41, and for the left eye, the difference between the longitudinal coordinates of the two groups of upper and lower eyelid points 44 and 46, and 45 and 47. The average of the differences between the longitudinal coordinates of the four groups of upper and lower eyelid points of the right eye and the left eye of each frame is obtained, and the eyelid distance of the video frame is obtained. The eyelid distance of all frames is stored in a list data to construct the eyelid distance time sequence data set.

[0095] For the eyebrow-eye point time sequence data, the eyebrow-eye distance is calculated frame by frame. The eyebrow-eye distance is defined as the difference between the highest point of the eyebrow and the inner corner of the eye, as shown in Figure 2 For the right eye, the difference between the longitudinal coordinates of the two groups of upper and lower eyelid points 38 and 42, and 39 and 41, and for the left eye, the difference between the longitudinal coordinates of the two groups of upper and lower eyelid points 44 and 46, and 45 and 47. The average of the differences between the longitudinal coordinates of the four groups of upper and lower eyelid points of the right eye and the left eye of each frame is obtained, and the eyelid distance of the video frame is obtained. The eyelid distance of all frames is stored in a list data to construct the eyelid distance time sequence data set.

[0096] Further, based on the fluctuation of the eyelid distance time sequence data set and the blinking amplitude abnormal value, the blinking recognition threshold is calculated, as shown in Figure 3As shown, this is a line graph for blink detection. In this graph, the blue line represents the blink recognition threshold, which is also a dynamically changing line. In a preferred embodiment of the present invention, the calculation of the blink recognition threshold needs to consider special cases, such as some users whose eyelid distance increases or decreases when their eyes are open; some users who do not blink at all; some users whose eyes are not open at all, resulting in little change in the distance between their upper and lower eyelids; and some users whose eyelid distance changes little at all due to eye size. To eliminate noise caused by the above situations, this application uses a local eyelid distance threshold for correction. Specifically, the global mean and global standard deviation of the eyelid distance in the eyelid distance time series dataset are calculated, and the global eyelid distance threshold is calculated based on the global mean and global standard deviation of the eyelid distance. The eyelid distance time series dataset consists of the pixel distances of the upper and lower eyelids identified in each frame of a video. The specific format is: [62.0, 62.0, 63.5, 62.0, 62.5, 59.5, 56.0, 59.0, 58.5, 60.5, 61.5, 61.5, 63.0, 63.5, 65.0, 65.0, 66.0, 66.0, 67.5, 65.0, 66.5, 66.0, 65.5, 65.5, 66.0, 65.0, 65.0, 65.0, 65.0, 65.0, 64.5, 65.5, ...]. Since the video typically consists of 16 frames per second, this pixel distance is normalized using the global maximum value and 0. Calculate the global mean and global standard deviation of eyelid distance in the normalized eyelid distance time series dataset. The global mean of eyelid distance is used as the threshold baseline, and the global standard deviation of eyelid distance is used as an assessment of the overall fluctuation of eyelid distance. Then, calculate the global threshold of eyelid distance using the following formula:

[0097]

[0098] in, This indicates the global threshold distance from the eyelid. This represents the global average distance between the eyelids. This indicates the global standard deviation of the eyelid distance. This indicates the sensitive parameters for blink recognition.

[0099] formula The accuracy of the blink recognition threshold is controlled using the global standard deviation of eyelid distance. A larger global standard deviation indicates greater noise in the patient's blink data, meaning the distance between the eyelid distance and the global mean is smaller for each blink, requiring a threshold closer to the mean for detection. (Formula) Adjustments are made to address cases where the global standard deviation of eyelid distance is large, thereby improving the accuracy of the global threshold for eyelid distance.

[0100] For the calculation of the local threshold of the eyelid distance, the video frames of the first preset time period before and after the historical video information are selected to represent the blink amplitude abnormal frames, the local mean and the local standard deviation of the eyelid distance time series data set corresponding to the blink amplitude abnormal frames are calculated, and the local threshold of the eyelid distance is calculated according to the local mean and the local standard deviation of the eyelid distance, the calculation formula of the local threshold of the eyelid distance is:

[0101]

[0102] wherein, represents the local threshold of the eyelid distance, represents the local mean of the eyelid distance, represents the local standard deviation of the eyelid distance.

[0103] After obtaining the global threshold of the eyelid distance and the local threshold of the eyelid distance, the weight of the global threshold of the eyelid distance and the local threshold of the eyelid distance is set to be the first preset weight, the global threshold of the eyelid distance and the local threshold of the eyelid distance are fused according to the first preset weight, and the blink recognition threshold is obtained. In the preferred embodiment of the present application, the weight of the global threshold of the eyelid distance is 2 / 3, the weight of the local threshold of the eyelid distance is 1 / 3, and the final blink recognition threshold is as shown by the blue broken line. Figure 3

[0104] After obtaining the blink recognition threshold, the eyelid distance in the eyelid distance time series data set is compared with the blink recognition threshold, and the video frame corresponding to the eyelid distance less than the blink recognition threshold is the blink frame, as shown by the orange broken line. Figure 3 The green broken line is the identified blink frame.

[0105] Similarly, based on the fluctuation and the abnormal value of the eyebrow-eye distance, the blink recognition threshold is calculated, specifically, the global mean and the global standard deviation of the eyebrow-eye distance time series data set are calculated, and the global threshold is calculated according to the global mean and the global standard deviation of the eyebrow-eye distance, the calculation formula of the global threshold of the eyebrow-eye distance is:

[0106]

[0107] wherein, represents the global threshold of the eyebrow-eye distance, represents the global mean of the eyebrow-eye distance, represents the global standard deviation of the eyebrow-eye distance, represents the frown recognition sensitivity parameter.

[0108] ​The video frame of the first preset time period is selected to represent the abnormal frame of the eyebrow and eye movement amplitude, the local mean of the eyebrow and eye distance and the local standard deviation of the eyebrow and eye distance of the eyebrow and eye distance time series data set corresponding to the abnormal frame of the eyebrow and eye movement amplitude are calculated, the local threshold of the eyebrow and eye distance is calculated according to the local mean of the eyebrow and eye distance and the local standard deviation of the eyebrow and eye distance, and the calculation formula of the local threshold of the eyebrow and eye distance is:

[0109]

[0110] wherein, the local threshold of the eyebrow and eye distance is represented, the local mean of the eyebrow and eye distance is represented, the local standard deviation of the eyebrow and eye distance is represented.

[0111] After the global threshold of the eyebrow and eye distance and the local threshold of the eyebrow and eye distance are obtained, the weight of the global threshold of the eyebrow and eye distance and the local threshold of the eyebrow and eye distance is set to be a second preset weight, the global threshold of the eyebrow and eye distance and the local threshold of the eyebrow and eye distance are fused according to the second preset weight, and the frowning recognition threshold is obtained. In the preferred embodiment of the present application, the weight of the global threshold of the eyebrow and eye distance is 2 / 3, the weight of the local threshold of the eyebrow and eye distance is 1 / 3, and finally the frowning recognition threshold is obtained.

[0112] After the frowning recognition threshold is obtained, the eyebrow and eye distance in the eyebrow and eye distance time series data set is compared with the frowning recognition threshold, and the video frame corresponding to the eyebrow and eye distance less than the frowning recognition threshold is a frowning frame.

[0113] Further, in the preferred embodiment of the present application, based on the eyelid distance time series data set, the blink frame data of each blink frame is calculated, including the maximum eyelid distance, the minimum eyelid distance, the number of complete blinks, the number of incomplete blinks, the minimum blink time, the average blink time, the maximum blink interval, the blink interval time variability, the minimum closed eye time, the average closed eye time, the minimum open eye time and the average open eye time. Wherein, the maximum eyelid distance and the minimum eyelid distance are the maximum value and the minimum value of the eyelid distance of each blink in the eyelid distance broken line as shown in Figure 3 The video frame in the second preset time period before and after each blink frame is selected to represent the blink related frame, and the maximum eyelid distance and the minimum eyelid distance corresponding to the blink related frame are selected from the eyelid distance time series data set.

[0114] For the blink time, the duration of each blink, specifically, the duration of each blink related frame with the eyelid distance less than the blink recognition threshold is calculated as the first time series interval, and the first time series interval is taken as the blink time of each blink. As shown in Figure 3As shown, the blink time is the duration of the orange fold line less than the blue fold line, the average blink time is calculated by averaging each blink time, and the minimum blink time is determined according to each blink time.

[0115] Further, according to the size of the blink time and the preset blink threshold, it is judged whether the blink frame is a complete blink or an incomplete blink, and the number of complete blinks and the number of incomplete blinks in the blink frame are counted. In the preferred embodiment of the present application, the preset blink threshold is 0.2s, when the blink time is greater than or equal to 0.2s, it is a complete blink, and when the blink time is less than 0.2s, it is an incomplete blink.

[0116] Further, the complete blink time standard deviation and the complete blink time average value corresponding to the blink time of the complete blink are calculated, and the ratio of the complete blink time standard deviation and the complete blink time average value is taken as the blink interval time variability.

[0117] Further, the duration of the minimum eyelid distance of each blink frame is calculated, which is the second time sequence interval, and the second time sequence interval is taken as the blink interval, and the maximum blink interval is selected from the blink interval.

[0118] Further, the duration of the minimum eyelid distance of each blink frame is calculated, which is the second time sequence interval, and the second time sequence interval is taken as the blink interval, and the maximum blink interval is selected from the blink interval. Figure 3 The duration of the orange fold line less than the blue fold line in the middle, the third time sequence interval is taken as the closed eye time. The minimum closed eye time is selected from the closed eye time, and the average closed eye time is calculated according to each closed eye time.

[0119] Further, the duration of the minimum eyelid distance of each blink frame is calculated, which is the second time sequence interval, and the second time sequence interval is taken as the blink interval, and the maximum blink interval is selected from the blink interval.

[0120] Similarly, for the frown frame, based on the eyebrow-eye distance time sequence data set, the frown frame data of each frown frame is calculated, including the maximum eyebrow-eye distance, the minimum eyebrow-eye distance, the frown time, the frown frequency, the frown interval time variability, the maximum frown time and the minimum frown interval.

[0121] Among them, the maximum eyebrow-eye distance and the minimum eyebrow-eye distance are the maximum and minimum of each frown eyebrow-eye distance. The frown associated frame is selected from the video frame in the third preset time period before and after each frown frame, and the maximum eyebrow-eye distance and the minimum eyebrow-eye distance corresponding to the frown associated frame are selected from the eyebrow-eye distance time sequence data set.

[0122] For the frown time, the time duration of each frown, specifically, by calculating the time duration of each frown associated frame with the eyebrow-eye distance being less than the frown recognition threshold, a fifth time sequence interval is obtained, the fifth time sequence interval is taken as the frown time of each frown, and the total number of frowns in the entire historical video information is determined, and the maximum frown time is determined according to each frown time.

[0123] Further, the frown time standard deviation and the frown time average value corresponding to the frown time are calculated, and the ratio of the frown time standard deviation and the frown time average value is taken as the frown interval time variability.

[0124] Further, the time duration between the minimum eyebrow-eye distance of each frown frame is calculated, a sixth time sequence interval is obtained, the sixth time sequence interval is taken as the frown interval, and the minimum frown interval is selected from the frown interval.

[0125] Further, the blink frame data, the frown frame data and the eye red probability are used to train the constructed dry eye detection model to obtain a target dry eye detection model. In the preferred embodiment of the present application, the dry eye detection model is constructed using the Catboost framework. Catboost has made a lot of improvements to the processing of category features, and embedded an innovative algorithm that automatically processes category features into numerical features. Specifically, some statistics are performed on the category features, the frequency of a certain category feature is calculated, and then a super parameter is added to generate a new numerical feature. The use of combined category features can take advantage of the relationship between features, greatly enriching the feature dimension and enabling more efficient training. CatBoost uses the concept of ordered boosting, which is a permutation-driven method that trains a model on a subset of data while calculating the residual on another subset, thereby preventing target leakage and overfitting. CatBoost can be easily integrated with deep learning frameworks, supporting various features of numbers, categories or text, saving time and effort on preprocessing, and greatly improving the convenience and efficiency of the dry eye detection on the smartphone.

[0126] The blink frame data, the frown frame data and the eye red probability are input into the constructed dry eye detection model for training to obtain a target dry eye detection model. In the real-time detection process, the user to be identified uploads real-time video information through the smartphone or records real-time video information using the front camera of the smartphone, and obtains the real-time blink frame data, the real-time frown frame data and the real-time eye red probability after information extraction. The real-time blink frame data, the real-time frown frame data and the real-time eye red probability are input into the target dry eye detection model to obtain a dry eye detection result.

[0127] In the preferred embodiment of the present application, historical video information containing a human face is collected, and eyelid point position time sequence data, eyebrow eye point position time sequence data and eye red probability in the historical video information are identified frame by frame; eyelid distance is calculated frame by frame according to the eyelid point position time sequence data to construct an eyelid distance time sequence data set, and eyebrow eye distance is calculated frame by frame according to the eyebrow eye point position time sequence data to construct an eyebrow eye distance time sequence data set; a blink recognition threshold is calculated based on the fluctuation of the eyelid distance time sequence data set and the abnormal value of the blink amplitude, and a frown recognition threshold is calculated based on the fluctuation of the eyebrow eye distance time sequence data set and the abnormal value of the eyebrow eye movement amplitude; a blink frame is identified according to the size relationship between the eyelid distance time sequence data set and the blink recognition threshold, and a frown frame is identified according to the size relationship between the eyebrow eye distance time sequence data set and the frown recognition threshold; blink frame data of each blink frame is calculated based on the eyelid distance time sequence data set, and frown frame data of each frown frame is calculated based on the eyebrow eye distance time sequence data set; the constructed dry eye detection model is trained using the blink frame data, the frown frame data and the eye red probability, and a target dry eye detection model is obtained; in the real-time detection process, real-time blink frame data, real-time frown frame data and real-time eye red probability extracted from the collected real-time video information are input into the target dry eye detection model, and a dry eye detection result is obtained. The dry eye detection method provided by the present application can realize the effect of providing a convenient, practical, accurate and large-scale application dry eye detection method, which helps to improve the quality of life of patients and promote the efficient allocation of limited medical resources.

[0128] Correspondingly, as Figure 4 shown, based on a dry eye detection method, the present application also provides a dry eye detection system, which realizes the dry eye detection method disclosed in the present application, comprising: a data acquisition and processing unit 1, a first calculation unit 2, an identification threshold calculation unit 3, an identification unit 4, a second calculation unit 5, a model training unit 6 and a real-time detection unit 7.

[0129] The data acquisition and processing unit 1 is used to collect historical video information containing a human face, and identify eyelid point position time sequence data, eyebrow eye point position time sequence data and eye red probability in the historical video information frame by frame.

[0130] The first calculation unit 2 is used to calculate eyelid distance frame by frame according to the eyelid point position time sequence data to construct an eyelid distance time sequence data set, and calculate eyebrow eye distance frame by frame according to the eyebrow eye point position time sequence data to construct an eyebrow eye distance time sequence data set.

[0131] The identification threshold calculation unit 3 is configured to calculate a blink identification threshold based on fluctuations in the eyelid distance time series data set and blink amplitude outliers, and calculate a frown identification threshold based on fluctuations in the eyebrow-eye distance time series data set and eyebrow-eye movement amplitude outliers.

[0132] The identification unit 4 is configured to identify a blink frame according to the size relationship between the eyelid distance time series data set and the blink identification threshold, and identify a frown frame according to the size relationship between the eyebrow-eye distance time series data set and the frown identification threshold.

[0133] The second calculation unit 5 is configured to calculate blink frame data of each of the blink frames based on the eyelid distance time series data set, and calculate frown frame data of each of the frown frames based on the eyebrow-eye distance time series data set.

[0134] The model training unit 6 is configured to train the constructed dry eye detection model using the blink frame data, the frown frame data and the eye red probability, to obtain a target dry eye detection model.

[0135] The real-time detection unit 7 is configured to input real-time blink frame data, real-time frown frame data and real-time eye red probability extracted from real-time video information collected in a real-time detection process into the target dry eye detection model, to obtain a dry eye detection result.

[0136] The specific limitations of the dry eye detection system can refer to the above limitations of the dry eye detection method, which will not be repeated here. Those skilled in the art can realize that the modules and steps described in combination with the embodiments disclosed in the present application can be realized in hardware, software or both. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0137] As shown in Figure 5 The computer device provided by the embodiment of the present application includes a processor, a memory and a computer program stored in the memory and configured to be executed by the processor. The processor implements the steps in the above dry eye detection method embodiments when executing the computer program, such as steps S1-S7 described in Figure 1 .

[0138] Those skilled in the art can understand that the schematic Figure 5The computer device is only an example and does not constitute a limitation on the computer device, which can include more or fewer components than shown, or combine some components, or have different components, for example, the computer device can also include an input / output device, a network access device, a bus, etc.

[0139] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is a control center of the computer device, and connects various parts of the computer device through various interfaces and lines.

[0140] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the computer device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required for a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0141] The modules integrated in the computer device can be stored in a computer readable storage medium in the form of software function units when they are realized in the form of software function units and sold or used as independent products. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0142] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. The program can include the processes of each method embodiment when executed.

[0143] Correspondingly, the embodiment of the present application provides a computer readable storage medium including a stored computer program, wherein when the computer program runs, the device where the computer readable storage medium is located executes the steps in the dry eye detection method of the above-mentioned embodiment, for example Figure 1 the steps S1-S7 in the above-mentioned embodiment.

[0144] The dry eye detection method, system, device and medium provided in the embodiment are used to solve the technical problem of improving the convenience, practicality and accuracy of dry eye detection. Historical video information containing a face is collected, and eyelid point position time sequence data, eyebrow-eye point position time sequence data and eye red probability in the historical video information are identified frame by frame. The eyelid distance is calculated frame by frame according to the eyelid point position time sequence data to construct an eyelid distance time sequence data set, and the eyebrow-eye distance is calculated frame by frame according to the eyebrow-eye point position time sequence data to construct an eyebrow-eye distance time sequence data set. The blink recognition threshold is calculated based on the fluctuation of the eyelid distance time sequence data set and the blink amplitude abnormal value, and the frown recognition threshold is calculated based on the fluctuation of the eyebrow-eye distance time sequence data set and the eyebrow-eye movement amplitude abnormal value. The blink frame is identified according to the size relationship between the eyelid distance time sequence data set and the blink recognition threshold, and the frown frame is identified according to the size relationship between the eyebrow-eye distance time sequence data set and the frown recognition threshold. The blink frame data of each blink frame is calculated based on the eyelid distance time sequence data set, and the frown frame data of each frown frame is calculated based on the eyebrow-eye distance time sequence data set. The dry eye detection model constructed is trained by using the blink frame data, the frown frame data and the eye red probability, and a target dry eye detection model is obtained. In the real-time detection process, the real-time blink frame data, the real-time frown frame data and the real-time eye red probability extracted from the collected real-time video information are input into the target dry eye detection model, and a dry eye detection result is obtained. The dry eye detection method provided in the application can realize the effect of providing a convenient, practical and accurate dry eye detection method that can be widely applied, and is helpful to improve the quality of life of patients and promote the efficient allocation of limited medical resources.

[0145] Each of the embodiments in the specification is described in a progressive manner, and the directly same or similar parts of each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments. Especially, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the description of the method embodiment. It should be noted that, each technical feature of the above-mentioned embodiments can be combined arbitrarily, in order to make the description simple, not all possible combinations of the technical features in the above-mentioned embodiments are described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the description.

[0146] The above-described embodiments only express several preferred embodiments of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the application. It should be noted that, for ordinary skilled in the art, without departing from the technical principles of the present application, a number of improvements and replacements can be made, and these improvements and replacements should be considered as the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A dry eye detection method characterized by, The method comprises: Collecting historical video information containing a human face, and identifying eyelid point position time series data, eyebrow-eye point position time series data and eye red probability in the historical video information frame by frame; Calculating eyelid distance frame by frame according to the eyelid point position time series data to construct an eyelid distance time series data set, and calculating eyebrow-eye distance frame by frame according to the eyebrow-eye point position time series data to construct an eyebrow-eye distance time series data set; Calculating a blink recognition threshold value based on fluctuations and blink amplitude outliers of the eyelid distance time series data set, and calculating a frown recognition threshold value based on fluctuations and eyebrow-eye movement amplitude outliers of the eyebrow-eye distance time series data set; Identifying blink frames according to the size relationship between the eyelid distance time series data set and the blink recognition threshold value, and identifying frown frames according to the size relationship between the eyebrow-eye distance time series data set and the frown recognition threshold value; Calculating blink frame data of each of the blink frames based on the eyelid distance time series data set, and calculating frown frame data of each of the frown frames based on the eyebrow-eye distance time series data set; Training a completed dry eye detection model by using the blink frame data, the frown frame data and the eye red probability to obtain a target dry eye detection model; In the real-time detection process, real-time blink frame data, real-time frown frame data and real-time eye red probability extracted from collected real-time video information are input into the target dry eye detection model to obtain a dry eye detection result.

2. The dry eye detection method according to claim 1, wherein The frame-by-frame identification of the eyelid point position time series data, the eyebrow-eye point position time series data and the eye red probability in the historical video information comprises: Identifying 68 standard points of human facial features frame by frame by using a dlib model: Selecting eyelid point position data and eyebrow-eye point position data from the 68 standard points; Combining time series data of each frame of the video information with the eyelid point position data to construct eyelid point position time series data; Combining the time series data with the eyebrow-eye point position data to construct eyebrow-eye point position time series data; Inputting each frame of the video information into a trained eye red recognition model to obtain an eye red probability.

3. The dry eye detection method according to claim 1, wherein The eyelid distance is defined as the difference between the vertical coordinates of the upper eyelid and the lower eyelid; The eyebrow-eye distance is defined as the difference between the vertical coordinates of the highest point of the eyebrow and the inner corner of the eye.

4. The dry eye detection method of claim 1, wherein The calculation of the blink recognition threshold value based on fluctuations and blink amplitude outliers of the eyelid distance time series data set comprises: Calculating the global mean and the global standard deviation of the eyelid distance time series data set, and calculating a global threshold value of the eyelid distance according to the global mean and the global standard deviation of the eyelid distance; The calculation formula of the global threshold value of the eyelid distance is: wherein, represents the global threshold of eyelid distance, represents the global mean of eyelid distance, represents the global standard deviation of eyelid distance, represents the blink recognition sensitivity parameter; Selecting video frames of a first preset time period before and after the historical video information to represent blink amplitude abnormal frames, calculating the local mean and the local standard deviation of the eyelid distance time series data set corresponding to the blink amplitude abnormal frames, and calculating a local threshold value of the eyelid distance according to the local mean and the local standard deviation of the eyelid distance. The formula for calculating the eyelid distance local threshold value is: wherein, represents the local threshold of the eyelid distance, represents the local mean of the eyelid distance, represents the local standard deviation of the eyelid distance; The eyelid distance global threshold value and the eyelid distance local threshold value are fused according to a first preset weight to obtain the blink recognition threshold value.

5. The dry eye detection method of claim 1, wherein The calculation of the frown recognition threshold value based on the fluctuation of the eyebrow-eye distance time series data set and the eyebrow-eye movement amplitude abnormal value includes: The global mean value and the global standard deviation of the eyebrow-eye distance time series data set are calculated, and the global threshold value of the eyebrow-eye distance is calculated according to the global mean value and the global standard deviation of the eyebrow-eye distance; The formula for calculating the global threshold value of the eyebrow-eye distance is: wherein, represents a global threshold of the eyebrow-eye distance, represents a global mean of the eyebrow-eye distance, represents a global standard deviation of the eyebrow-eye distance, represents a frown recognition sensitivity parameter; The video frames of the first preset time period before and after the historical video information are selected to represent the eyebrow-eye movement amplitude abnormal frame, the local mean value and the local standard deviation of the eyebrow-eye distance time series data set corresponding to the eyebrow-eye movement amplitude abnormal frame are calculated, and the local threshold value of the eyebrow-eye distance is calculated according to the local mean value and the local standard deviation of the eyebrow-eye distance; The formula for calculating the local threshold value of the eyebrow-eye distance is: wherein, represents the eyebrow-eye distance local threshold value, represents the eyebrow-eye distance local mean value, represents the eyebrow-eye distance local standard deviation; The global threshold value of the eyebrow-eye distance and the local threshold value of the eyebrow-eye distance are fused according to a second preset weight to obtain the frown recognition threshold value.

6. The dry eye detection method of claim 1, wherein The calculation of the blink frame data of each blink frame based on the eyelid distance time series data set includes: The video frames within the second preset time period before and after each blink frame are selected to represent the blink-related frames, and the maximum eyelid distance and the minimum eyelid distance corresponding to the blink-related frames are selected from the eyelid distance time series data set; The first time series interval in which the eyelid distance of each blink-related frame is less than the blink recognition threshold value is calculated, and the first time series interval is taken as the blink time; The mean value of the blink time is calculated to obtain the average blink time; According to the size of the blink time and the preset blink threshold value, it is judged whether the blink frame is complete blink or incomplete blink, and the number of complete blinks and the number of incomplete blinks in the blink frame are counted; The complete blink time standard deviation and the complete blink time average value of the blink time corresponding to the complete blink are calculated, and the ratio of the complete blink time standard deviation to the complete blink time average value is taken as the blink interval time variation degree; The minimum blink time is determined according to each blink time; The blink interval is calculated according to the second time series interval between the minimum eyelid distance of each blink frame, and the maximum blink interval is selected from the blink interval; The third time series interval in which the eyelid distance of each blink frame changes from less than the blink recognition threshold value to the minimum eyelid distance is calculated, and the third time series interval is taken as the closed-eye time; The fourth time series interval in which the eyelid distance of each blink frame changes from the minimum eyelid distance to greater than the blink recognition threshold value is calculated, and the fourth time series interval is taken as the open-eye time; The minimum closed-eye time is selected from the closed-eye time, and the average closed-eye time is calculated according to each closed-eye time; The minimum open-eye time is selected from the open-eye time, and the average open-eye time is calculated according to each open-eye time; The eyelid distance maximum value, the eyelid distance minimum value, the complete blink times, the incomplete blink times, the blink time minimum value, the average blink time, the maximum blink interval, the blink interval time variability, the minimum closed eye time, the average closed eye time, the minimum open eye time, and the average open eye time are taken as blink frame data.

7. The dry eye detection method of claim 1, wherein The system comprises a data acquisition and processing unit, a first calculation unit, an identification threshold calculation unit, an identification unit, a second calculation unit, a model training unit, and a real-time detection unit. The data acquisition and processing unit is configured to acquire historical video information containing a human face, and identify eyelid point position time series data, eyebrow-eye point position time series data, and eye red probability in the historical video information frame by frame. The first calculation unit is configured to calculate eyelid distance frame by frame based on the eyelid point position time series data to construct an eyelid distance time series data set, and calculate eyebrow-eye distance frame by frame based on the eyebrow-eye point position time series data to construct an eyebrow-eye distance time series data set. The identification threshold calculation unit is configured to calculate a blink identification threshold based on fluctuations and blink amplitude outliers of the eyelid distance time series data set, and calculate a frown identification threshold based on fluctuations and eyebrow-eye movement amplitude outliers of the eyebrow-eye distance time series data set. The identification unit is configured to identify blink frames according to the size relationship between the eyelid distance time series data set and the blink identification threshold, and identify frown frames according to the size relationship between the eyebrow-eye distance time series data set and the frown identification threshold. The second calculation unit is configured to calculate blink frame data of each blink frame based on the eyelid distance time series data set, and calculate frown frame data of each frown frame based on the eyebrow-eye distance time series data set. The system comprises a data acquisition and processing unit, a first calculation unit, an identification threshold calculation unit, an identification unit, a second calculation unit, a model training unit, and a real-time detection unit.

8. A dry eye detection system implementing the dry eye detection method of any one of claims 1 to 7, characterized by, The data acquisition and processing unit is configured to acquire historical video information containing a human face, and identify eyelid point position time series data, eyebrow-eye point position time series data, and eye red probability in the historical video information frame by frame. The first calculation unit is configured to calculate eyelid distance frame by frame based on the eyelid point position time series data to construct an eyelid distance time series data set, and calculate eyebrow-eye distance frame by frame based on the eyebrow-eye point position time series data to construct an eyebrow-eye distance time series data set. The identification threshold calculation unit is configured to calculate a blink identification threshold based on fluctuations and blink amplitude outliers of the eyelid distance time series data set, and calculate a frown identification threshold based on fluctuations and eyebrow-eye movement amplitude outliers of the eyebrow-eye distance time series data set. The identification unit is configured to identify blink frames according to the size relationship between the eyelid distance time series data set and the blink identification threshold, and identify frown frames according to the size relationship between the eyebrow-eye distance time series data set and the frown identification threshold. The second calculation unit is configured to calculate blink frame data of each blink frame based on the eyelid distance time series data set, and calculate frown frame data of each frown frame based on the eyebrow-eye distance time series data set. ​ The model training unit is configured to train the constructed dry eye detection model by using the blink frame data, the frown frame data, and the eye red probability, and obtain a target dry eye detection model. The real-time detection unit is configured to input real-time blink frame data, real-time frown frame data, and real-time eye red probability extracted from real-time video information collected in a real-time detection process into the target dry eye detection model, and obtain a dry eye detection result.

9. A computer device, characterized by: The computer device comprises a memory, a processor, and a transceiver connected through a bus; the memory is configured to store a set of computer program instructions and data, and transmit the stored data to the processor; the processor executes the program instructions stored in the memory to perform the dry eye detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer readable storage medium stores a computer program, and when the computer program is executed, the dry eye detection method according to any one of claims 1 to 7 is implemented.

Images