WO2020186480A1 - 近视预测系统及方法 - Google Patents

近视预测系统及方法 Download PDF

Info

Publication number
WO2020186480A1
WO2020186480A1 PCT/CN2019/078885 CN2019078885W WO2020186480A1 WO 2020186480 A1 WO2020186480 A1 WO 2020186480A1 CN 2019078885 W CN2019078885 W CN 2019078885W WO 2020186480 A1 WO2020186480 A1 WO 2020186480A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
myopia
eye
user
change trend
Prior art date
Application number
PCT/CN2019/078885
Other languages
English (en)
French (fr)
Inventor
杨智宽
蓝卫忠
李响
吴砚
朱均伟
Original Assignee
杭州镜之镜科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 杭州镜之镜科技有限公司 filed Critical 杭州镜之镜科技有限公司
Publication of WO2020186480A1 publication Critical patent/WO2020186480A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment

Definitions

  • the invention relates to the technical field of myopia prevention and control, in particular to a myopia prediction system and a myopia prediction method.
  • the technical problem to be solved by the present invention is to provide a myopia prediction system and method capable of personalized prediction of the user's myopia trend in view of one or more of the above-mentioned defects in the prior art.
  • the first aspect of the present invention provides a myopia prediction system, including:
  • a terminal control device that communicates with the multiple data collection devices, and uploads eye behavior data collected by the multiple data collection devices
  • Mobile terminal equipment used to collect user's myopia change trend data
  • the cloud server is used to receive the eye behavior data uploaded by the terminal control device, and use the eye behavior data as the model input data and the myopia change trend data as the verification result to train the deep learning model; and use the trained The deep learning model processes the eye behavior data to be predicted, and obtains the prediction results of myopia change trend matching the user's eye habits.
  • the eye behavior data includes: date, user ID, average eye distance, average eye time, average ambient light, average outdoor exposure time, average head tilt Angle and current myopia prevention and control means sequence value
  • the myopia change trend data includes: date, user ID, user age, left eye refractive power change trend label value, and right eye refractive power change trend label value.
  • the mobile terminal device is also used to collect personal characteristic data of the user, and the cloud server merges eye behavior data and personal characteristic data into user eye data, As model input data; the personal characteristic data includes at least: user ID, user age, student gender, student area and genetic information.
  • the left eye refractive power change trend label value and the right eye refractive power change trend label value both represent the corresponding refractive power change trend by the following values:
  • the refractive power change trend label value is set to the third label value.
  • the deep learning model adopts an attention encoder to process the eye data of N consecutive users at time t and before time t, and perform classification by a softmax encoder, Output the probabilities corresponding to the label values of the three refractive power change trends.
  • the data collection device includes:
  • a data collection module for collecting eye behavior data, where the eye behavior data includes at least eye distance data
  • the main control module is used to process the current eye distance data using the training model, detect whether the current sitting posture is a standard sitting posture or a non-standard sitting posture, and generate a reminder instruction when the current sitting posture is a non-standard sitting posture;
  • the reminder module is used to generate a reminder signal to remind the user when receiving a reminder instruction
  • the feedback module is used for receiving the feedback signal input by the user, and sending the feedback signal to the main control module to update the training model.
  • the main control module includes:
  • the real-time detection unit is used to input the current eye distance data as a test sample into the training model, and output the detection result data including the label value, where the label value is used to identify the standard sitting posture or the non-standard sitting posture;
  • the state detection unit is used to generate a reminder instruction when the tag value of the detection result data corresponds to a non-standard sitting posture
  • the model training unit is used to add the current eye distance data and its label value as a training sample to the training sample set when the feedback signal is not received, and to correct the label value corresponding to the current eye distance data when the feedback signal is received.
  • the training sample set is added; the model training unit uses the training sample set to update the training model.
  • the data collection module includes:
  • Matrix type ranging unit used to sample the distance in the preset rectangular measurement area
  • the main control module generates a sample sequence according to multiple distance values sampled in the rectangular measurement area, and inputs the sample sequence as a test sample into the training model.
  • the data collection device further includes:
  • Input module for receiving the height value input by the user
  • Lifting mechanism for adjusting the height position of the distance measuring sensor of the matrix type distance measuring unit
  • the main control module further includes: a sensor height calculation unit for determining the preset height position of the distance measuring sensor according to the height value input by the user, and sending a signal to the lifting mechanism to control the distance measuring sensor to reach the preset height position; wherein The preset height position h satisfies the following formula:
  • w is the desktop depth
  • FOV is the field of view angle of the ranging sensor
  • H is the height value input by the user.
  • the present invention also provides a myopia prediction method, including the following steps:
  • the present invention collects the user's eye behavior data by installing multiple data collection devices, and uploads them to the server uniformly, combines the myopia change trend data collected by mobile terminal equipment, and trains the deep learning model to accurately predict the user The change trend of myopia after a preset period of time.
  • the present invention can perform sitting posture detection based on the current eye distance data, generate a reminder instruction when the current sitting posture is a non-standard sitting posture, and continuously correct training samples by receiving feedback signals from the user when judging abnormalities Therefore, the training model is continuously updated during use, which can better match the user's characteristics, such as height, and make the judgment result more accurate.
  • the present invention adopts a matrix ranging method, uses multiple eye data in an area to determine sitting posture, and through a reasonable height setting, it can effectively cover the head and chest area, making the measurement result more accurate.
  • Figure 1 is a schematic diagram of the composition of a myopia prediction system according to a preferred embodiment of the present invention
  • Figure 2 is a schematic diagram of an encoder model according to a preferred embodiment of the present invention.
  • Fig. 3 is a schematic diagram of softmax according to a preferred embodiment of the present invention.
  • FIG. 4 is a graph showing an increase in myopia of an individual in the later stage generated according to a preferred embodiment of the present invention and a prevention and control effect graph of different myopia prevention and control methods predicted by the preferred embodiment of the present invention;
  • Figure 5 is a block diagram of a data acquisition device according to a preferred embodiment of the present invention.
  • FIG. 6 is a block diagram of the principle of the main control module in the data acquisition device of the preferred embodiment of the present invention.
  • FIG. 7 is a schematic diagram of a first appearance of a data acquisition device according to a preferred embodiment of the present invention.
  • FIG. 8 is a diagram of the use state of the data acquisition device according to the preferred embodiment of the present invention.
  • Figure 9 is a schematic diagram of sitting posture physiological analysis
  • FIG. 10 is a schematic diagram of adjusting the height position of the distance measuring sensor according to the present invention.
  • FIG. 11 is a schematic diagram of sampling of a matrix measurement unit of a data acquisition device according to a preferred embodiment of the present invention.
  • Fig. 12 is a distance measurement spatial relationship diagram of a data acquisition device according to a preferred embodiment of the present invention.
  • Fig. 13 is a diagram of a preset rectangular measurement area of a data acquisition device according to a preferred embodiment of the present invention.
  • Figure 14 is a schematic diagram of the principle of the SVM trainer.
  • FIG. 1 is a schematic diagram of the composition of a myopia prediction system according to a preferred embodiment of the present invention.
  • the myopia prediction system provided by this embodiment includes a plurality of data collection devices 100, a terminal control device 200, a cloud server 300, and a mobile terminal device 400.
  • multiple data collection devices 100 are used to collect user's eye behavior data.
  • the main purpose of the data collection device 100 is to monitor students' eye behavior data, which may include but not limited to: average eye distance, average eye time, average ambient light, average outdoor exposure time, average head angle, and current myopia prevention and control Means sequence value. And transmit the data to the terminal control device 200.
  • the data collection device 100 may be installed in a specific place, such as a campus.
  • the data collection device 100 includes but is not limited to:
  • the terminal control device 200 communicates with the multiple data collection devices 100 and uploads the eye behavior data collected by the multiple data collection devices 100.
  • the terminal control device 200 is a terminal product installed in a specific place to manage all the data collection devices 100.
  • the terminal control device 200 communicates with all devices at regular intervals, obtains all collected eye behavior data, and uploads the data to the cloud server 300.
  • the functions of the terminal control device 200 include but are not limited to:
  • Data transmission receiving data collected by the data collecting device 100 and transmitting the data to the cloud server 300.
  • Equipment management responsible for managing all data collection devices 100 in the site, including adding, deleting, updating, and viewing devices.
  • the mobile terminal device 400 is used to collect user's myopia change trend data.
  • the mobile terminal device 400 may be an electronic device such as a mobile phone/tablet, and the user can input myopia change trend data through the mobile terminal device 400.
  • the cloud server 300 communicates with the terminal control device 200 and the mobile terminal device 400, and is used to receive the eye behavior data uploaded by the terminal control device 200, and use the eye behavior data as model input data, and use the myopia change trend data as verification As a result, the deep learning model is trained.
  • the cloud server 300 also uses the trained deep learning model to process the eye behavior data to be predicted to obtain the prediction result of the myopia change trend.
  • the cloud server 300 has a big data learning function. By collecting hundreds of millions of eye data and all users’ follow-up myopia development diagnosis information, it forms a platform with big data as samples and deep learning methods.
  • the prediction result of myopia change trend matches the user's eye habits.
  • the theoretical basis of the present invention believes that the user's eye data can be used to predict the myopia risk trend, that is, the user's eye data is used as input to predict the myopia risk trend (output).
  • the myopia risk trend as three categories, as follows:
  • a and b can be adjusted and optimized according to actual conditions.
  • the value of a is 0.8D-1.2D.
  • the eye behavior data collected by the data collection device 100 may include, but is not limited to: average eye distance, average eye time, average ambient light, average outdoor exposure time, average head angle, and sequence value of current myopia prevention and control measures.
  • the eye behavior data may also include basic information such as date and user ID, and the eye behavior data may constitute data set 1:
  • Average eye distance The average eye distance of the student in the past month; among them, using eyes at close range is an important factor leading to myopia.
  • Average eye use time The average daily eye use time of the student in the past month; the time of continuous eye use, eye load and myopia also have a strong positive correlation.
  • Average ambient light The student's ambient light intensity when using eyes at close range in the past month; if the ambient light intensity is too dark, it will increase the load on the eyes, which is related to myopia.
  • Average outdoor exposure time The average daily outdoor exposure time of the student in the past month; the study found that sunlight has a mitigation effect on myopia, so outdoor exposure time is also related to myopia.
  • Average head tilt angle The average head tilt angle of the student when using eyes at close range in the past month; whether a person tilts the head when using the eyes, bad head tilt habits can also lead to different development of myopia in the left and right eyes.
  • a sample case is as follows:
  • the myopia change trend data collected by the mobile terminal device 400 includes but is not limited to date, user ID, user age, left eye refractive power change trend label value, and right eye refractive power change trend label value.
  • Make up data set 2 is not limited to date, user ID, user age, left eye refractive power change trend label value, and right eye refractive power change trend label value.
  • a sample case is as follows:
  • the myopia change trend data can be calculated by the myopia refractive power data received by the mobile terminal device 400, for example, the myopia refractive power data is stored as the following data set:
  • myopia is represented by +
  • hyperopia is represented by -.
  • the present invention is a classification problem.
  • the user uses eye data to predict the risk of diopter deepening, that is, the risk of myopia in recent eye habits. Based on clinical experience, the present invention assumes the increase in myopia diopter and eye behavior Habits are related to personal characteristics.
  • the eye behavior data is used as the user's eye data input model, that is, the aforementioned data set 1 is used as the model input data.
  • eye behavior data and personal characteristic data are combined into user eye data as model input data.
  • the mobile terminal device 400 is also used to collect personal characteristic data of the user.
  • the personal characteristic data includes but is not limited to: user ID, user age, student gender, student area, and genetic information.
  • Data set 3 represents all personal characteristic data, such as genetic information, gender characteristics, etc.
  • the gender of the student is 1 for boys and 2 for girls; among them, myopia is represented by +, and hyperopia is represented by -.
  • the area is indicated by the telephone area code.
  • a sample case is as follows:
  • the cloud server of the present invention can use a neural attention encoder for training, as shown in Figure 2.
  • the deep learning model uses an attention encoder to process the continuous N user eye data at time t and before time t, and classifies it through the softmax encoder, and outputs the probabilities corresponding to the label values of the three refractive power change trends.
  • N is the preset value. specifically,
  • I t wherein t is the time user data eye, i tN, i tN + 1 , i tN + 2 whil i t-1 for the N successive user data eye before time t, the data are
  • the data of set 1 is encoded by the encoder
  • O t is the vector encoded at time t
  • the attention mechanism is input to adjust the recent weights
  • C t is the predicted data at time t.
  • S j is the probability of the j-th category, and the j-th value in the vector represented by a j .
  • X is the vector that has undergone attention processing
  • the gradient descent algorithm is used to update the gradient, and the cross entropy is used as our loss function.
  • the form of the cross entropy function is as follows:
  • the value in log is the Softmax value of the correct classification of this group of data. The larger the proportion of it, the smaller the Loss of this sample.
  • the cloud server of the present invention can further use the analysis result to generate an individual's late-sighted increase trend graph, as shown in FIG. 4.
  • -1.00 means myopia is 100°, and other things can be known in the same way; this system can predict the increasing trend of myopia in the later period of an individual every (3 months/year).
  • the present invention can also predict the prevention and control effects of different myopia prevention and control means, as shown in FIG. 4.
  • the above methods can assess the increasing trend of individual myopia. For some group scenes, it is necessary to assess the group's myopia risk from a more systematic perspective.
  • the cloud server of the present invention can select groups with one type (or multiple types) of tags (for example, the same school, the same area, etc.), and comprehensively evaluate the myopia risk trend of the group to judge the myopia risk of the entire group. To formulate a comprehensive myopia prevention and control program.
  • the present invention starts with the above model, and assessing the development trend of myopia can start with data on eye behavior habits, and preventing and controlling myopia can start from controlling eye behavior.
  • Starting from theory from multiple perspectives such as data collection, eye evaluation, and behavior intervention, combined with specific scenarios, such as campuses.
  • This program provides a friendly product solution for myopia prediction or further assessment of myopia status from the entire system network. Therefore, the present invention specifically addresses the needs of myopia risk assessment, and formulates a low-cost, good-effect, easy-to-expand and maintain eye habit assessment and prediction plan.
  • the present invention can also use deep learning to evaluate the myopia risk for groups with different characteristics such as the whole group.
  • a data collection device 100 that can be used for sitting posture detection is also provided to improve the function of the entire myopia prediction system.
  • FIG. 5 is a block diagram of a data acquisition device according to a preferred embodiment of the present invention.
  • the data collection device 100 provided in this embodiment includes: a data collection module 110, a main control module 120, a reminder module 130 and a feedback module 140.
  • the data collection module 110 is used to collect eye behavior data, and the eye behavior data includes at least eye distance data.
  • the main control module 120 is connected to the data collection module 110, and is used to process the current eye distance data using the training model, detect whether the current sitting posture is a standard sitting posture or a non-standard sitting posture, and generate a reminder instruction when the current sitting posture is a non-standard sitting posture .
  • the training model is preferably an SVM (Support Vector Machine) model.
  • the reminder module 130 is connected to the main control module 120 and is used to generate a reminder signal to remind the user when receiving a reminder instruction.
  • the reminding module 130 may remind the user through, for example, sound and light reminders.
  • the feedback module 140 is connected to the main control module 120, and is configured to receive a feedback signal input by the user, and send the feedback signal to the main control module 120 to update the training model.
  • the present invention can perform sitting posture detection based on the current eye distance data, and generate a reminder instruction when the current sitting posture is a non-standard sitting posture.
  • the method of the present invention discards the traditional logic of determining whether it is a bad sitting posture by directly performing threshold judgment based on distance measurement, but uses machine learning to determine the sitting posture, which can better match the characteristics of the user, such as height.
  • FIG. 6 is a functional block diagram of the main control module 120 in the myopia prediction system of the preferred embodiment of the present invention.
  • the main control module 120 includes a real-time detection unit 121, a state detection unit 122 and a model training unit 123.
  • the real-time detection unit 121 is configured to input the current eye distance data x as a test sample into a training model such as an SVM model, and output detection result data including a label value y, where the label value is used to identify a standard sitting posture or a non-standard sitting posture.
  • a training model such as an SVM model
  • detection result data including a label value y, where the label value is used to identify a standard sitting posture or a non-standard sitting posture.
  • the model training unit 123 is used for adding the current eye distance data and its label value as a training sample to the training sample set when the feedback signal is not received.
  • the model training unit 123 updates the training model using the updated training sample set.
  • the model training unit 123 may use the latest training sample set to update the training model every time it receives a feedback signal. In a more preferred embodiment of the present invention, the model training unit 123 may use the latest training sample set to update the training model when judging that the number of received feedback signals exceeds a preset threshold. For example, if the feedback signal is received more than 5 times, the training model is updated with the latest training sample set.
  • the housing 1 of the data acquisition device includes a base 2 and a cylindrical housing 3 vertically installed on the base 2.
  • a feedback button 5 is installed on the base 2, for example, a touch sensing method is adopted.
  • a distance measuring sensor 4 is installed at the upper end of the cylindrical housing 3, and the data collection device can collect eye behavior data through the distance measuring sensor 4.
  • the distance measuring sensor 4 preferably adopts an infrared distance measuring sensor, which includes a pair of infrared signal emitting diodes and receiving diodes. The propagation of infrared light takes time.
  • the time it takes to be received by the receiving diode after hitting the reflector and being received by the receiving diode multiplied by the propagation speed of the infrared light can calculate the distance measuring sensor 4 and the measured object
  • the distance between the eyes is the distance between the eyes.
  • the data collection device can be installed on the edge of a desk in front of a user, such as a student, so that the user is measuring.
  • the distance measuring sensor 4 can also be a distance measuring light sensor, which detects the distance between the eyes and the light intensity at the same time.
  • the data acquisition device provided by the present invention is placed in front of the desk, with an opening right in front of the top, and a distance measuring sensor 4 inside, which analyzes the sitting posture by detecting the distance from the head and chest of the person.
  • the data collection device can adopt a matrix ranging method.
  • the data collection module 110 includes at least: a matrix ranging unit 111 for sampling the distance in a preset rectangular measurement area.
  • the matrix distance measuring unit 111 can detect the distance to the head and chest of a person through the distance measuring sensor 4 having a certain angle to the horizontal.
  • the ranging chip of the matrix ranging unit 111 is a matrix ranging method. Each measurement returns a matrix array (n*n) within a certain angle:
  • the height position of the distance measuring sensor 4 of the data collection device is fixed. In some other preferred embodiments of the present invention, the height position of the distance measuring sensor 4 of the data acquisition device is adjustable. Therefore, the data acquisition device also includes: an input module and a lifting mechanism.
  • the input module is used to receive the height value input by the user. For example, input the height value by voice or keystrokes.
  • the lifting mechanism is used to adjust the height position of the distance measuring sensor 4 of the matrix distance measuring unit, for example, by various methods known and applicable to those skilled in the art, such as installing a lifting mechanism control column housing 3 in the base 2 Lift up and down to adjust the height position of the distance measuring sensor 4 on it.
  • the main control module 120 further includes a sensor height calculation unit for determining the preset height position of the distance measuring sensor according to the height value input by the user, and sending a signal to the lifting mechanism to control the distance measuring sensor to reach the preset height position.
  • the preset height position h satisfies the following formula:
  • w is the depth of the desktop
  • FOV is the field of view of the ranging sensor
  • H is the height value entered by the user.
  • the present invention concludes that the height position of the distance measuring sensor 4 determined by the above method can well improve the measurement accuracy of the eye distance and improve the accuracy of posture prediction.
  • the effect of setting the height position of the distance measuring sensor 4 of the present invention will be analyzed with reference to FIG. 10.
  • the minimum value of the preset height position h of the distance measuring sensor 4 is w*tan(FOV/2), so that the measurement data is not blocked by the desktop.
  • the maximum value of the preset height position h of the distance measuring sensor 4 is (H/15)+w*tan(FOV/2). This is because, considering that the distance measuring sensor should be able to better measure the bending of the head and neck, the detection range of the measuring sensor must include the chest. According to the actual statistics, take 1/4 of the height on the desktop, that is The lower edge of the measurement area needs to be no lower than the chest position.
  • the proportion of human head height ranges from 5 years old to adults, roughly between 6 and 7.5.
  • a person’s body is at least 2/7.5 (2/6, that is, at least the head and upper chest, these two are about 2 heads high, and the total height of the person is 6 heads to 7.5 heads high) on the table , In order to have a more comfortable posture.
  • the following best relationship can be obtained:
  • the present invention also studies the FOV of the range-finder sensor in the above formula.
  • the range-finding sensor In order for the range-finding sensor to be able to detect most areas of the human body exposed on the desktop, it must be able to detect the height of the person (2/7.5 ⁇ 2/6) most of the area. After a lot of research and experience, the present invention has a better effect when the detection coverage rate reaches 30% and above. From this, the following formula can be obtained:
  • the desktop depth w (the size from the sitting side, the front and back direction of the desktop) is 40cm. Assuming that the device is installed directly in front of the student and far from the edge of the user's desk, it is approximately considered that the horizontal distance between the device and the user is 40 cm.
  • the FOV of the range-finder sensor of the present invention is preferably 27°.
  • the height of the ranging chip is h. If the FOV of the ranging sensor is 27° and the desktop depth w is 40cm, the result of each measurement is a 4*4 matrix, as shown in Figure 11. Show.
  • the exploreable area is a circular area with a radius of 9.6, as shown in Figure 12 and Figure 13.
  • the distance measurement of the present invention can collect the distance data of the chest and the head at the same time to judge the sitting posture. From the above information, the preset height position h of the distance measuring sensor of the device can be better designed so that the final matrix measurement area covers the head and chest well.
  • the main control module 120 generates a sample sequence according to multiple distance values sampled in the rectangular measurement area, and inputs the sample sequence as a test sample into the training model. Preferably, the main control module 120 uniformly samples 4*4 distance values in the rectangular measurement area to form a sample sequence.
  • Each current eye distance data includes the following 4*4 matrix eye distance values:
  • Dc is an eye distance data used to form a sample sequence.
  • d c11 , d c12 , d c13 , and d c14 are the distance values of the 4 sampling points of the first row of distribution, and so on, d c41 , d c42 , d c43 and d c44 are the 4 samples of the fourth row of distribution The distance value of the point.
  • the current eye distance data and the test time point are used to obtain a sequence of test samples and input into the training model.
  • the present invention can continuously learn independently during use. In the prior art, when determining the sitting posture, it does not take into account that each person's posture, habits, and desk distance are different, which may cause misjudgment.
  • the present invention introduces an autonomous learning mechanism to solve the above problems.
  • a feedback button is set on the device. When the user feels that the current machine judgment is abnormal, such as a bad sitting posture is not reminded, or a correct sitting posture is not reminded, press the feedback to the machine. When the user continues Feedback, internal algorithms will continue to self-learn and optimize.
  • the SVM algorithm is mainly used.
  • Support Vector Machine shows many unique advantages in solving small sample, nonlinear and high-dimensional pattern recognition, and can be extended to other machine learning problems such as function fitting.
  • the main idea of the SVM trainer is to find a linearly separable line (or hyperplane) that can separate the two types.
  • line or hyperplane
  • What we are looking for is the line (or hyperplane) with the largest distance when dividing the two types of targets among these straight lines (or hyperplanes).
  • We call such a straight line or hyperplane the best linear classifier, as shown in Figure 14.
  • x is the inputted eye condition data Dc.
  • the sigmoid function is:
  • W is the normal vector
  • T in w T represents transposition
  • b is the displacement
  • the learning method of the present invention is as follows:
  • the training samples with known sitting state are pre-recorded in the device to form a training sample set, including the eye distance data of the standard sitting posture and the eye distance data of the non-standard state, and the label values are respectively marked, and the training sample is used to compare the SVM model Conduct training.
  • the device starts to recognize the sitting posture.
  • the machine detects that the current sitting posture is not the standard sitting posture, it will remind and correct it.
  • the machine detects that the current sitting posture is the standard sitting posture, it will not remind.
  • the sitting posture data is processed for dimensionless standardization, and paired with the label value into the svm trainer for training.
  • the device also has detection functions such as light intensity.
  • the data collection module 110 further includes: a light intensity measuring unit 112 for measuring data of light intensity.
  • the main control module is also used to detect the light intensity, and generate a reminding instruction to the reminding module 140 when it is lower than a preset threshold.
  • the device also has the function of detecting the length of the eye.
  • the main control module 120 is also used to detect the eye duration, and generate a reminder instruction to the reminder module when it is higher than a preset threshold.
  • both the preset threshold of eye duration and the preset threshold of light intensity can be set according to needs.
  • the preset threshold of eye duration is 30min-60min, and the preset threshold of light intensity is 600-800lux.
  • the main control module 120 generates a reminding instruction when the light intensity is lower than a preset threshold.
  • the main control module 120 determines that the near vision state is satisfied when the average eye distance L ⁇ 60 cm, and starts timing.
  • the average eye distance L ⁇ 60cm is the average value of the distance within the preset rectangular measurement area.
  • the main control module 120 generates a reminder when the eye duration is higher than the preset threshold.
  • these two reminding signals can be used in separate reminding methods, such as vibration reminding. In this way, the user can only give feedback on the accuracy of the sitting posture detection.
  • the present invention also provides a myopia prediction method corresponding to the system, including:
  • the above method also includes:
  • the data collection step is used to collect eye behavior data, where the eye behavior data at least includes eye distance data.
  • the judgment step is used to process the current eye distance data using the training model, detect whether the current sitting posture is a standard sitting posture or a non-standard sitting posture, and generate a reminder instruction when the current sitting posture is a non-standard sitting posture.
  • Reminder step when receiving a reminder instruction, a reminder signal is generated to remind the user.
  • the feedback step is used to receive the feedback signal input by the user
  • the training model is updated in response to the feedback signal input by the user.
  • the judgment steps include:
  • the real-time detection sub-step is used to input the current eye distance data as the test sample into the training model, and output the detection result data containing the label value, where the label value is used to identify the standard sitting posture or the non-standard sitting posture;
  • the state detection sub-step is used to generate a reminder instruction when the tag value of the detection result data corresponds to a non-standard sitting posture.
  • the model training step adds the current eye distance data and its label value as a training sample to the training sample set when the feedback signal is not received, and corrects the label value corresponding to the current eye distance data before adding the training when the feedback signal is received.
  • Sample set ; and use the training sample set to update the training model.
  • the training sample set is used to update the training model when it is determined that the number of received feedback signals exceeds a preset threshold.
  • the myopia prediction method further includes the following steps:
  • w is the desktop depth
  • FOV is the field of view angle of the ranging sensor
  • H is the height value input by the user.

Landscapes

  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Ophthalmology & Optometry (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Epidemiology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
  • Eye Examination Apparatus (AREA)

Abstract

本发明涉及一种近视预测系统及方法，其中系统包括：多个数据采集装置，用于采集用户的用眼行为数据；终端控制设备，与所述多个数据采集装置通讯，并将通过所述多个数据采集装置采集的数据上传；移动终端设备，用于上传用户的近视变化趋势数据；云端服务器，用于接收终端控制设备上传的用眼行为数据，并以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验证结果，对深度学习模型进行训练；并利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得与用户用眼习惯匹配的近视变化趋势预测结果。本发明可以准确地预测用户在预设时间段后的近视变化趋势，并可进一步在个体近视变化趋势的基础上评估整个群体中不同特征群体的近视风险。

Description

近视预测系统及方法 技术领域
本发明涉及近视防控技术领域，尤其涉及一种近视预测系统及近视预测方法。
背景技术
随着时代的发展，近视变得越来越普遍，其中青少年近视情况尤为令人担忧，青少年近视患病率呈逐年上升的趋势。因此中小学生近视防控迫在眉睫。
当前已有的用眼习惯检测的产品，都更多的应用在消费者领域，单一用户场景，而对于近视加深最严重的一些特殊场景(例如学校)，缺乏有效、性价比高的控制手段。
因此，亟待开发一种能够采集用眼数据对近视进行预测的系统及方法。
发明内容
本发明要解决的技术问题在于，针对上述现有技术中的一个或多个缺陷，提供一种能够个性化的对使用者的近视趋势进行预测的近视预测系统及方法。
为了解决上述技术问题，本发明第一方面，提供了一种近视预测系统，包括：
多个数据采集装置，用于采集用户的用眼行为数据；
终端控制设备，与所述多个数据采集装置通讯，并将通过所述多个数据采集装置采集的用眼行为数据上传；
移动终端设备，用于采集用户的近视变化趋势数据；
云端服务器，用于接收所述终端控制设备上传的用眼行为数据，并以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验证 结果，对深度学习模型进行训练；并利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得与用户用眼习惯匹配的近视变化趋势预测结果。
在根据本发明所述的近视预测系统中，优选地：所述用眼行为数据包括：日期、用户ID、平均用眼距离、平均用眼时长、平均环境光照、平均户外暴露时长、平均偏头角度和当前近视防控手段序列值；所述近视变化趋势数据包括：日期、用户ID、用户年龄、左眼屈光度数变化趋势标签值和右眼屈光度数变化趋势标签值。
在根据本发明所述的近视预测系统中，优选地：所述移动终端设备还用于采集用户的个人特征数据，所述云端服务器将用眼行为数据和个人特征数据合并为用户用眼数据，作为模型输入数据；所述个人特征数据至少包括：用户ID、用户年龄、学生性别、学生地区和遗传信息。
在根据本发明所述的近视预测系统中，优选地：所述左眼屈光度数变化趋势标签值和右眼屈光度数变化趋势标签值均通过以下数值表示相应的屈光度数变化趋势：
在近视屈光度数一年增加数大于a时，或者三个月增加数大于0.25a时，将屈光度数变化趋势标签值设为第一标签值；
在近视屈光度数一年增加数为b到a之间时，或者三个月增加数为0.25b到0.25a之间时，将屈光度数变化趋势标签值设为第二标签值；
在近视屈光度数一年增加数小于b时，或者三个月增加数小于0.25b时，将屈光度数变化趋势标签值设为第三标签值。
在根据本发明所述的近视预测系统中，优选地：所述深度学习模型采用attention编码器对t时刻及t时刻之前连续的N个用户用眼数据进行处理，并通过softmax编码器进行分类，输出三种屈光度数变化趋势标签值对应的概率。
在根据本发明所述的近视预测系统中，优选地，所述数据采集装置包括：
数据采集模块，用于采集用眼行为数据，所述用眼行为数据至少包括用眼距离数据；
主控模块，用于利用训练模型对当前的用眼距离数据进行处理，检测当前坐姿为标准坐姿或者非标准坐姿，在当前坐姿为非标准坐姿时生成提醒指令；
提醒模块，用于在接收提醒指令时产生提醒信号以提醒用户；
反馈模块，用于接收用户输入的反馈信号，并将所述反馈信号发送给所述主控模块对训练模型进行更新。
在根据本发明所述的近视预测系统中，优选地，所述主控模块包括：
实时检测单元，用于将当前的用眼距离数据作为测试样本输入训练模型，输出包含标签值的检测结果数据，其中标签值用于标识标准坐姿或非标准坐姿；
状态检测单元，用于检测结果数据的标签值对应非标准坐姿时生成提醒指令；
模型训练单元，用于在未接收到反馈信号时将当前的用眼距离数据及其标签值作为训练样本加入训练样本集，在接收到反馈信号时更正当前的用眼距离数据对应的标签值再加入训练样本集；所述模型训练单元利用所述训练样本集更新训练模型。
在根据本发明所述的近视预测系统中，优选地，所述数据采集模块包括：
矩阵式测距单元，用于对预设的矩形测量区域内的距离进行采样；
所述主控模块根据矩形测量区域内采样的多个距离数值生成样本序列，作为测试样本输入训练模型。
在根据本发明所述的近视预测系统中，优选地，所述数据采集装置还包括：
输入模块，用于接收用户输入的身高值；
升降机构，用于调整矩阵式测距单元的测距传感器的高度位置；
所述主控模块还包括：传感器高度计算单元，用于根据用户输入的身高值确定测距传感器的预设高度位置，并发送信号给所述升降机构控制测距传感器达到预设高度位置；其中预设高度位置h满足以下公式：
w*tan(FOV/2)≤h≤(H/15)+w*tan(FOV/2)；
其中，w为桌面深度，FOV为测距传感器的视场角，H为用户输入的身高值。
本发明还提供了一种近视预测方法，包括以下步骤：
采集用户的用眼行为数据；
采集用户的近视变化趋势数据；
以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验证结果，对深度学习模型进行训练；并利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得与用户用眼习惯匹配的近视变化趋势预测结果。
在完成个人近视趋势预测的基础上，可以通过统计学分析方法，对一整个群体进行近视趋势变化评估，从而更好的了解群体近视风险现状，制定群体的近视防控方案。
实施本发明的近视预测系统及方法，具有以下有益效果：
1、本发明通过安装多个数据采集装置来采集用户的用眼行为数据，并统一上传至服务器，结合移动终端设备采集的近视变化趋势数据，通过对深度学习模型进行训练，从而准确的预测用户在预设时间段后的近视变化趋势。
2、本发明通过机器学习的方法，可以基于当前的用眼距离数据进行坐姿检测，在当前坐姿为非标准坐姿时生成提醒指令，并通过接收用户在判断异常时的反馈信号，不断更正训练样本集，从而在使用过程中不断的更新训练模型，更能够匹配使用者的特征，例如身高等，使得判断结果更加准确。
3、本发明通过采用矩阵式测距方式，利用一个区域内多个用眼数 据来判断坐姿，并通过合理的高度设置使其可以有效覆盖头部和胸部区域，使得测量结果更为准确。
附图说明
图1为根据本发明优选实施例的近视预测系统的组成示意图；
图2为根据本发明优选实施例的编码器模型示意图；
图3为根据本发明优选实施例的softmax示意图；
图4为根据本发明优选实施例生成的个体后期的近视增加趋势图与本发明优选实施例预测不同近视防控手段的防控效果图；
图5为根据本发明优选实施例的数据采集装置的模块框图；
图6为本发明优选实施例的数据采集装置中主控模块的原理框图；
图7为本发明优选实施例的数据采集装置的第一种外形示意图；
图8为本发明优选实施例的数据采集装置的使用状态图；
图9为坐姿生理学分析示意图；
图10为根据本发明的测距传感器的高度位置调节示意图；
图11为根据本发明的优选实施例的数据采集装置的矩阵式测量单元采样示意图；
图12为根据本发明的优选实施例的数据采集装置的测距空间关系图；
图13为根据本发明的优选实施例的数据采集装置的预设矩形测量区域图；
图14为SVM训练器原理示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明的一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本发明 保护的范围。
请参阅图1，为根据本发明优选实施例的近视预测系统的组成示意图。如图1所示，该实施例提供的近视预测系统包括：多个数据采集装置100、终端控制设备200、云端服务器300和移动终端设备400。
其中多个数据采集装置100，用于采集用户的用眼行为数据。数据采集装置100的主要目的是监测学生的用眼行为数据，可以包括但不限于：平均用眼距离、平均用眼时长、平均环境光照、平均户外暴露时长、平均偏头角度和当前近视防控手段序列值。并将数据传输给终端控制设备200。数据采集装置100可以安装在特定场所，例如校园等。该数据采集装置100包括但不限于：
(1)用眼习惯监测眼镜，搭配有传感器的眼镜，可以实施采集用户的用眼距离、用眼时长、环境光照等用眼相关数据；
(2)依托于眼镜上的智能监测设备，比如吸附在镜腿上的智能模块；
(3)其他可穿戴设备，比如夹在衣领上的智能监测设备。
终端控制设备200，与所述多个数据采集装置100通讯，并将通过所述多个数据采集装置100采集的用眼行为数据上传。
终端控制设备200是安装在特定场所用于管理所有数据采集装置100的终端产品。当特定场所例如教室内所有数据采集装置100接入终端控制设备200，该设备每隔一定时间和所有设备通信，获取所有采集的用眼行为数据，并将数据上传至云端服务器300。该终端控制设备200的功能包括但不限于：
(1)数据传输：接收数据采集装置100采集的数据，并将数据传输给云端服务器300。
(2)设备管理：负责管理场所内所有数据采集装置100，包括装置的增加、删除、更新、查看。
(3)状态查询：随时查询所有已注册的数据采集装置100的工作状态。
(4)异常上报：所有已注册的数据采集装置100发生异常，自动上报异常。
移动终端设备400，用于采集用户的近视变化趋势数据。该移动终端设备400可以为手机/平板等电子设备，用户可通过该移动终端设备400输入近视变化趋势数据等。
云端服务器300，与终端控制设备200和移动终端设备400通讯，用于接收所述终端控制设备200上传的用眼行为数据，并以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验证结果，对深度学习模型进行训练。该云端服务器300还利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得近视变化趋势预测结果。该云端服务器300具有大数据学习功能，通过收集亿万条用眼数据，以及所有用户的后续近视发展诊断信息，形成了一套以大数据为样本、以深度学习为方法的平台，所获得的近视变化趋势预测结果是与用户用眼习惯相匹配的。
下面对本发明的近视预测系统的运行原理进行详细说明。
1、本发明的理论基础认为，通过用户用眼数据，可以用于预测近视风险趋势，即以用户用眼数据作为输入，来预测近视风险趋势(输出)。近视风险趋势我们定义为3个类，如下：
高：在近视屈光度数一年增加数大于a时，或者三个月增加数大于0.25a时，将屈光度数变化趋势标签值设为第一标签值，例如+3；
中：在近视屈光度数一年增加数为b到a之间时，或者三个月增加数为0.25b到0.25a时，将屈光度数变化趋势标签值设为第二标签值，例如+2；
低：在近视屈光度数一年增加数小于b时，或者三个月增加数小于0.25b时，将屈光度数变化趋势标签值设为第三标签值，例如+1。并且近视屈光度数增加数为0时，也处于该低风险。
上述a和b的阈值可以根据实际情况调节，调优。优选地，a的取值为0.8D-1.2D。b的取值为0.4D-0.6D。更优选地，a＝1.0D，b＝0.5D。 即：
高：在近视屈光度数一年增加数大于1.0D时，或者三个月增加数大于0.25D时，将屈光度数变化趋势标签值设为第一标签值，例如+3；
中：在近视屈光度数一年增加数为0.5D-1.0D时，或者三个月增加数为0.125D-0.25D时，将屈光度数变化趋势标签值设为第二标签值，例如+2；
低：在近视屈光度数一年增加数小于0.5D时，或者三个月增加数小于0.125D时，将屈光度数变化趋势标签值设为第三标签值，例如+1。并且近视屈光度数增加数为0时，也处于该低风险。
2、建立样本，建立模型前期，采集较多的用眼行为数据，并定期(3个月)通过眼科验光操作，得到当前近视屈光度数。有如下两个数据样本集。
数据集1：
数据采集装置100采集的用眼行为数据可以包括但不限于：平均用眼距离、平均用眼时长、平均环境光照、平均户外暴露时长、平均偏头角度和当前近视防控手段序列值。该用眼行为数据还可以包括日期和用户ID等基本信息，该用眼行为数据可以构成数据集1：
[日期，用户ID，平均用眼距离，平均用眼时长，平均环境光照，平均户外暴露时长，平均偏头角度，当前近视防控手段序列值]
其中：
平均用眼距离：过去1个月该学生的平均用眼距离；其中近距离用眼是导致近视的一个重要因素。
平均用眼时长：过去1个月该学生的平均每天用眼总时长；持续用眼的时间，用眼负荷和近视也有很强的正相关。
平均环境光照：过去1个月该学生的近距离用眼时的环境光照强度；如果环境光照强度太暗，会导致用眼负荷增加，与近视有关。
平均户外暴露时长：过去1个月该学生的平均每天户外暴露时长； 研究发现，太阳光对近视有缓解作用，因此在户外的暴露时间也和近视有相关关系。
平均偏头角度：过去1个月该学生的近距离用眼时的平均偏头角度；人用眼时是否偏头，不好的偏头习惯也会导致左右眼近视发展不同。
当前近视防控手段序列的示例为：
*0：无
*1：单焦框架眼镜
*2：渐进多焦框架眼镜
*3：角膜塑形镜
*4：直分界整体双光镜
*5：多焦点软性角膜接触镜
若随着技术发展，新的技术防控手段出现，可以增加新的类别，算法依旧包容。
一个样本案例如下：
[20181007,11358,32,270,158,22,2,0]
[20181107,11358,31,255,162,20,2,0]
[20181207,11358,30,243,160,22,2,1]
数据集2：
移动终端设备400采集的近视变化趋势数据包括但不限于日期、用户ID、用户年龄、左眼屈光度数变化趋势标签值和右眼屈光度数变化趋势标签值。构成数据集2：
[日期，用户ID，用户年龄，左眼屈光度数变化趋势标签值，右眼屈光度数变化趋势标签值]
一个样本案例如下：
[20181228,11358,15,+2,+2]
[20180928,11358,15,+3,+3]
该近视变化趋势数据可以通过移动终端设备400接收的近视屈光度数数据来计算，例如近视屈光度数数据作为以下数据集存储：
[20180928,11358,15,-125,-225]
[20181228,11358,15,-150,-250]
其中近视用+表示，远视用-表示。
3.选择模型，本发明是一个分类问题，通过用户用眼数据去预测屈光度数加深的风险，即近期用眼习惯的近视风险，根据临床经验，本发明假定近视屈光度数的增加和用眼行为习惯和个人特征相关。
在本发明的一些实施例中，将用眼行为数据作为用户用眼数据输入模型，即上述数据集1作为模型输入数据。
在本发明的另一些更优选的实施例中，将用眼行为数据和个人特征数据合并为用户用眼数据，作为模型输入数据。在该实施例中，移动终端设备400还用于采集用户的个人特征数据。该个人特征数据包括但不限于：用户ID、用户年龄、学生性别、学生地区和遗传信息。
数据集3：
[用户ID，用户年龄，学生性别，学生地区，父亲右眼屈光度数，父亲左眼屈光度数，母亲右眼屈光度数，母亲左眼屈光度数，爷爷右眼屈光度数，爷爷左眼屈光度数，奶奶右眼屈光度数，奶奶左眼屈光度数，外公右眼屈光度数，外公左眼屈光度数，外婆右眼屈光度数，外婆左眼屈光度数]
数据集3表征所有个人特征数据，如遗传信息，性别特征等。
其中，学生性别，男生为1，女生为2；其中近视用+表示，远视用-表示。地区用电话区号表示。
一个样本案例如下：
[11358,15,1,020,-125,-225,-100,-200,-100,-175,-125,-225,-100,-225,-225,-100]
考虑学生当前的近视屈光度数样本值是一个长时间的时间序列问 题，并且是受更多的近期影响，因此本发明的云端服务器可以采用neural attention编码器来进行训练，如图2所示。该深度学习模型采用attention编码器对t时刻及t时刻之前连续的N个用户用眼数据进行处理，并通过softmax编码器进行分类，输出三种屈光度数变化趋势标签值对应的概率。N为预设值。具体地，
其中i t为t时刻的用户用眼数据,i t-N、i t-N+1、i t-N+2……i t-1为t时刻之前连续的N个用户用眼数据，均为数据集1的数据，通过encoder进行编码，O t为t时刻编码后的向量，输入attention注意力机制来调整近期的权重,C t为t时刻的预测数据。
同时我们在C t叠加一个softmax编码器，来解决多分类问题，使其输出各个不同值的概率，softmax公式为：
Figure PCTCN2019078885-appb-000001
S j为第j个类别的概率，a j表示的向量中的第j个值。
X为已经经过attention处理的向量，y为当前近视风险级别，其中定义为y＝1为低风险，y＝2为中风险，y＝3为高风险。Softmax(Ct)会输出一个三维向量，来分别描述y＝1、y＝2、y＝3的发生概率，其中softmax示意图3所示。
求解的过程中使用梯度下降算法来更新梯度，用交互熵作为我们的损失函数，交叉熵函数形式如下：
Figure PCTCN2019078885-appb-000002
取log里面的值就是这组数据正确分类的Softmax值，它占的比重越大，这个样本的Loss也就越小。
4、通过对上述模型输入大量数据集，进行训练。然后当需要对用 户用眼数据进行预测时，可以将该用户用眼数据输入模型，则模型可以输出接下来的近视风险趋势。
本发明的云端服务器还可进一步利用分析的结果生成个体后期的近视增加趋势图，如图4所示。其中，-1.00表示近视100°，其他同理可知；本系统可以以每(3个月/年)预测个体后期的近视增加趋势。同时，由于加入了近视防控手段数据，本发明同样可以预测不同近视防控手段的防控效果，如图4所示。
5、以上方法可以评估个体的近视增加趋势，对于一些群体场景，需要从更加系统的角度去评估群体的近视风险。本发明的云端服务器可以选取具有一类(或多类)标签的群体(例如同一个学校、同一个地区等)，将该群体的用眼近视风险趋势进行综合评估，判断整个群体的近视风险，从而制定全面的近视防控方案。
综上所述本发明从上述模型入手，评估近视发展趋势可以用眼行为习惯的数据入手，防控近视则可以从控制用眼行为入手。从理论出发，从数据采集、用眼评估、行为干预等多角度，并结合特定场景，例如校园等。建立一整套产品解决方案，本方案从整个系统组网上来提供一种友好的近视预测或进一步进行近视现状评估的产品方案。因此，本发明专门针对近视风险评估的需求，制定成本低、效果好、便于拓展、维护的用眼习惯评估及预测方案。本发明还可以使用深度学习评估针对整个群体等不同特征群体的近视风险。
在本发明更优选的实施例中，还提供了一种可用于对坐姿检测的数据采集装置100，以完善整个近视预测系统的功能。
请参阅图5，为根据本发明优选实施例的数据采集装置的模块框图。如图5所示，该实施例提供的数据采集装置100包括：数据采集模块110、主控模块120、提醒模块130和反馈模块140。
其中，数据采集模块110用于采集用眼行为数据，该用眼行为数据至少包括用眼距离数据。
主控模块120与所述数据采集模块110连接，用于利用训练模型 对当前的用眼距离数据进行处理，检测当前坐姿为标准坐姿或者非标准坐姿，在当前坐姿为非标准坐姿时生成提醒指令。该训练模型优选为SVM(支持向量机)模型。
提醒模块130与所述主控模块120连接，用于在接收提醒指令时产生提醒信号以提醒用户。优选地，该提醒模块130可以通过例如声光的提醒方式来提醒用户。
反馈模块140与所述主控模块120连接，用于接收用户输入的反馈信号，并将所述反馈信号发送给所述主控模块120对训练模型进行更新。
由此可见，本发明通过机器学习的方法，可以基于当前的用眼距离数据进行坐姿检测，在当前坐姿为非标准坐姿时生成提醒指令。并且本发明的方法抛弃了传统的根据测距直接进行阈值判断来确定是否为不良坐姿的逻辑，而是采用机器学习来判定坐姿，更能够匹配使用者的特征，例如身高等。
请参阅图6，为本发明优选实施例的近视预测系统中主控模块120的原理框图。如图6所示，主控模块120包括：实时检测单元121、状态检测单元122和模型训练单元123。
实时检测单元121用于将当前的用眼距离数据x作为测试样本输入训练模型例如SVM模型，输出包含标签值y的检测结果数据，其中标签值用于标识标准坐姿或非标准坐姿。例如，生成的标签值y＝-1表示非标准坐姿，即不良坐姿；y＝1表示标准坐姿。
状态检测单元122用于检测结果数据的标签值对应非标准坐姿时生成提醒指令，即检测标签值y＝-1时生成提醒指令。
模型训练单元123用于在未接收到反馈信号时将当前的用眼距离数据及其标签值作为训练样本加入训练样本集。当实时检测单元121对当前的用眼距离数据Dc进行判断后，如果收到反馈模块140的反馈信号，则表示坐姿判断的结果不准确，需要将当前的用眼距离数据Dc对应的标签值y更正为相反姿态的标签值。例如，将当前的用眼距离 数据Dc作为测试样本输入训练模型时，检测出标签值y＝1时，如果没有收到反馈信号，则将该当前的用眼距离数据Dc及标签值y作为一个训练样本加入训练样本集，如果收到反馈信号，则将标签值y更正为y＝-1，并将该当前的用眼距离数据Dc及标签值y作为一个训练样本加入训练样本集。相应地，当检测出标签值y＝-1时，如果没有收到反馈信号，则将该当前的用眼距离数据Dc及标签值y作为一个训练样本加入训练样本集，如果收到反馈信号，则将标签值y更正为y＝1，并将该当前的用眼距离数据Dc及标签值y作为一个训练样本加入训练样本集。模型训练单元123利用更新的训练样本集更新训练模型。在本发明的一种实施方式中，模型训练单元123可以在每次接收到反馈信号后就利用最新的训练样本集更新训练模型。在本发明更优选的实施方式中，模型训练单元123可以在判断接收的反馈信号次数超过预设阈值时利用最新的训练样本集更新训练模型。例如，如果接收的反馈信号超过5次，则重新用最新的训练样本集更新训练模型。
请结合参阅图7和图8分别为本发明优选实施例的数据采集装置的外形示意图以及使用状态图。如图所示，该数据采集装置的壳体1包括底座2以及竖直安装在所述底座2上的柱状壳体3。其中底座2上安装有反馈键5，例如采用触摸感应方式。柱状壳体3的上端部安装有测距传感器4，数据采集装置可以通过该测距传感器4采集用眼行为数据。测距传感器4优选采用红外测距传感器，包括一对红外信号的发射二极管与接收二极管。红外线的传播是需要时间的，根据红外线从发二极射管发出，碰到反射物回来后被接收二极管接收到所消耗的时间乘以红外线的传播速度可以计算出测距传感器4与被测物之间的距离，即用眼距离。该数据采集装置可以安装在使用者例如学生正前方的书桌边缘，从而正对使用者进行测量。该测距传感器4处也可以为测距测光传感器，同时对用眼距离和光强进行检测。
下面对本发明的坐姿检测原理进行详细说明。
从生理学分析坐姿原理，书写姿势可以大致分为正常、趴写、左 偏、右偏、前伸。其中不良坐姿内在原因是颈部和躯干弯曲，外在表现是头部和胸部弯曲，如图9所示。理想坐姿是头部不弯曲，躯干也不弯曲，即近似竖直坐姿，而实际上用户的习惯性坐姿会有一定程度的颈部和躯干弯曲。针对上述坐姿分析，本发明提供的数据采集装置放在书桌前，顶部正前方开口，内有测距传感器4，通过检测与人的头部、胸部距离来分析坐姿。
优选地，该数据采集装置可以采用矩阵式测距方式。数据采集模块110至少包括：矩阵式测距单元111，用于对预设的矩形测量区域内的距离进行采样。例如，矩阵式测距单元111可以通过与水平方面有一定角度的测距传感器4检测与人的头部、胸部距离。该矩阵式测距单元111的测距芯片是一种矩阵式测距方式，每次测量，返回一定角度内的一个矩阵队列(n*n)：
d 11 … d 1n
… … …
d n1 … d nn
在本发明的一些实施例中，数据采集装置的测距传感器4的高度位置固定。在本发明另一些优选的实施例中，数据采集装置的测距传感器4的高度位置可调。因此，该数据采集装置还包括：输入模块和升降机构。其中输入模块用于接收用户输入的身高值。例如通过语音或者按键等方式输入该身高值。升降机构则用于调整矩阵式测距单元的测距传感器4的高度位置，例如通过本领域基础技术人员熟知并能应用的各种方式将，例如在底座2中安装升降机构控制柱状壳体3升降，从而调整其上测距传感器4的高度位置。相应地，主控模块120还包括：传感器高度计算单元，用于根据用户输入的身高值确定测距传感器的预设高度位置，并发送信号给升降机构控制测距传感器达到预设高度位置。其中预设高度位置h满足以下公式：
w*tan(FOV/2)≤h≤(H/15)+w*tan(FOV/2)；
其中，w为桌面深度，FOV为测距传感器的视场角，H为用户输 入的身高值。
本发明通过大量研究及经验总结，得出通过上述方法确定的测距传感器4的高度位置可以很好地提高用眼距离的测量精度，提高姿势预测的准确性。下面结合图10对本发明测距传感器4的高度位置设定效果进行分析。
一方面，测距传感器4的预设高度位置h的最小取值为w*tan(FOV/2)，可以使得测量数据不被桌面遮挡。
另一方面，测距传感器4的预设高度位置h的最大取值为(H/15)+w*tan(FOV/2)。这是因为，考虑到测距传感器要更好的能测量到头部和颈部弯曲，所以测量传感器的探测范围一定要包含胸部，根据实际统计数据，取桌面上身高的1/4处，即测量区域的下边沿需要不低于胸部位置。
而人的头高比例从5岁到成人，大致在6-7.5之间。在正常学习时，人的身体至少有2/7.5(2/6，即至少头部、上胸部，这两个约占2头高，人总高度为6头高到7.5头高)在桌面上，才能有比较舒适的姿势。综上，可以得到以下最佳的关系式：
h-w*tan(FOV/2)≤H*(2/7.5)*(1/4))；
即h≤(H/15)+w*tan(FOV/2)。
此外，本发明还对上列公式中测距传感器的视场角FOV进行了研究，为了测距传感器要能探测人体在桌面上露出的大部分区域，即要能探测人身高的(2/7.5～2/6)的大部分区域。本发明经过大量研究及经验总结，在探测覆盖率达到30％及以上效果更佳，由此可得以下公式：
(2w*tan(FOV/2))/(H*(2/6))>30％
根据国家标准GB/T 3976-2014学校课桌椅功能尺寸及技术要求，桌面深度w(从坐人侧，桌面前后方向的尺寸)为40cm。假设装置安装与学生正前方远离用户的书桌边缘，则近似认为，装置与用户的水 平距离为40cm。对于6-18岁的群体，身高范围为100-180cm，代入极值H为180cm，w＝40cm，可以得到tan(FOV/2)>(9/40)。由此可知FOV角度需要大于25.36°才能得到较佳的探测覆盖率，根据现有芯片选型，本发明的测距传感器的视场角FOV优选为27°。
在本发明的一个实施例中，测距芯片高度为h，若测距传感器的视场角FOV为27°，桌面深度w为40cm，每次测量结果是4*4的矩阵，如图11所示。可探索面积为半径为9.6的一个圆形区域，如图12和图13所示。考虑到矩阵测量，如图13所示，最终确定的矩形测量区域为一个正方形区域，边长优选为12～15cm，更优选为9.6*sin45°*2＝13.57cm。
由前可知，不良坐姿有躯干弯曲和颈部弯曲，因此本发明的测距可以同时采集到胸部和头部距离数据，来判断坐姿。从以上信息可以更优设计装置的测距传感器的预设高度位置h，使最终矩阵测量的区域很好地覆盖头部和胸部。
主控模块120根据矩形测量区域内采样的多个距离数值生成样本序列，作为测试样本输入训练模型。优选地，主控模块120在矩形测量区域内均匀采样4*4个距离数值形成样本序列。每个当前的用眼距离数据包括以下4*4矩阵的用眼距离数值：
Figure PCTCN2019078885-appb-000003
其中Dc为一个用眼距离数据，用于构成样本序列。d c11、d c12、d c13、和d c14为第一行分布的4个采样点的距离值，以此类推，d c41、d c42、d c43和d c44为第四行分布的4个采样点的距离值。
将当前的用眼距离数据连同测试时间点得到测试样本的序列，输入训练模型。
本发明可以在使用过程中不断的自主学习。现有技术中在进行坐姿判别时，没有考虑每个人身姿、习惯、课桌距离不同，会造成误判。 本发明引入自主学习机制来解决以上问题，在装置上设置一个反馈键，当用户感觉目前机器判断异常时，例如不良坐姿未提醒，或者正确坐姿乱提醒时，按下反馈给机器，当用户持续反馈，内部算法会持续自学习优化。
在本发明的优选实施例中，主要使用SVM算法。支持向量机(Support Vector Machine，SVM))在解决小样本、非线性及高维模式识别中表现出许多特有的优势，并能够推广应用到函数拟合等其他机器学习问题中。SVM训练器主要思想是：找到一个能够将两类分开的线性可分的直线(或者超平面)。实际上有许多条直线(或超平面)可以将两类目标分开来，我们要找的其实是这些直线(或超平面)中分割两类目标时，有最大距离的直线(或超平面)。我们称这样的直线或超平面为最佳线性分类器，如图14。其中y＝-1我们认为是不良坐姿，y＝1我们认为是标准坐姿，x为输入的用眼状态数据Dc。
图中超平面的方程可以表示为：
w Tx+b＝0
其中，sigmoid函数为：
Figure PCTCN2019078885-appb-000004
W为法向量，w T中的T代表转置，b是位移量。
我们只需要判断样本数据是属于哪一个类，若大于0.5就是y＝1的类，反之属于y＝0类。
即w Tx i+b＞0，y i＝1；w Tx i+b＜0，y i＝-1。
本发明的学习方法如下：
1.装置中预先录入已知坐姿状态的训练样本构成训练样本集，包括标准坐姿的用眼距离数据和非标准状态的用眼距离数据，并分别标出标签值，利用该训练样本对SVM模型进行训练。
2.当用户坐在装置前，装置即开始识别坐姿，当机器检测到当前坐 姿非标准坐姿，则进行提醒纠正，当机器检测到当前坐姿为标准坐姿，则不进行提醒。
3.用户收到提醒/未收到提醒，若感觉到当前设备的反馈不准确，按下反馈键，则进行纠正；当用户在一定时间内未反馈，则认为设备的判定正确，将正确的标准坐姿持续更新数据集，并打上“标准坐姿”的标签。将不良坐姿持续更新数据集，并打上“非标准坐姿”的标签。
4.将坐姿数据做无量纲标准化处理，与标签值成对放入svm训练器，进行训练。
5.再次检测用户坐姿，再次将坐姿数据输入svm训练器，判定为不良坐姿/标准坐姿。
在本发明的更优选实施例中，该装置还具有光照强度等检测功能。例如，数据采集模块110还包括：光强测量单元112，用于测量光照强度的数据。主控模块还用于对光照强度进行检测，在低于预设阈值时产生提醒指令给提醒模块140。此外，该装置还具有用眼时长的检测功能。主控模块120还用于对用眼时长进行检测，在高于预设阈值时产生提醒指令给提醒模块。本发明中用眼时长的预设阈值和光照强度的预设阈值均可根据需要设置，优选地，用眼时长的预设阈值为30min～60min，光照强度的预设阈值600～800lux。例如，主控模块120在光照强度低于预设阈值时产生提醒指令。主控模块120在平均用眼距离L＜60cm时判断满足视近状态，并开始计时。平均用眼距离L＜60cm为预设的矩形测量区域内的距离数值平均值。相应地，主控模块120在用眼时长高于预设阈值时产生提醒。为了与坐姿检测的反馈信号区分开，这两种提醒信号可以通过单独的提醒方式，例如震动提醒方式。这样使用者可以只对坐姿检测的准确性进行反馈。
本发明还相应提供了一种与系统相对应的近视预测方法，包括：
采集用户的用眼行为数据；
采集用户的近视变化趋势数据；
以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验 证结果，对深度学习模型进行训练；并利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得与用户用眼习惯匹配的近视变化趋势预测结果。
上述方法还包括：
数据采集步骤，用于采集用眼行为数据，所述用眼行为数据至少包括用眼距离数据。
判断步骤，用于利用训练模型对当前的用眼距离数据进行处理，检测当前坐姿为标准坐姿或者非标准坐姿，在当前坐姿为非标准坐姿时生成提醒指令。
提醒步骤，在接收提醒指令时产生提醒信号以提醒用户。
反馈步骤，用于接收用户输入的反馈信号；
模型训练步骤，响应于用户输入的反馈信号，对训练模型进行更新。
其中，判断步骤包括：
实时检测子步骤，用于将当前的用眼距离数据作为测试样本输入训练模型，输出包含标签值的检测结果数据，其中标签值用于标识标准坐姿或非标准坐姿；
状态检测子步骤，用于检测结果数据的标签值对应非标准坐姿时生成提醒指令。
模型训练步骤则在未接收到反馈信号时将当前的用眼距离数据及其标签值作为训练样本加入训练样本集，在接收到反馈信号时更正当前的用眼距离数据对应的标签值再加入训练样本集；并且利用所述训练样本集更新训练模型。优选地，该步骤在判断接收的反馈信号次数超过预设阈值时利用所述训练样本集更新训练模型。
在本发明更优选实施例中，该所述近视预测方法还包括以下步骤：
(1)接收用户输入的身高值；
(2)根据用户输入的身高值确定测距传感器的预设高度位置，其 中预设高度位置h满足以下公式：
w*tan(FOV/2)≤h≤(H/15)+w*tan(FOV/2)；
其中，w为桌面深度，FOV为测距传感器的视场角，H为用户输入的身高值。
(3)调整矩阵式测距单元的测距传感器达到预设高度位置，例如通过升降方式。
应该理解地是，本发明的近视预测系统及数据采集装置和方法的原理相同，因此对数据采集装置和近视预测系统的实施例的详细阐述也适用于近视预测方法。
最后应说明的是：以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (11)

  1. 一种近视预测系统，其特征在于，包括：
    多个数据采集装置，用于采集用户的用眼行为数据；
    终端控制设备，与所述多个数据采集装置通讯，并将通过所述多个数据采集装置采集的用眼行为数据上传；
    移动终端设备，用于采集用户的近视变化趋势数据；
    云端服务器，用于接收所述终端控制设备上传的用眼行为数据，并以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验证结果，对深度学习模型进行训练；并利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得与用户用眼习惯匹配的近视变化趋势预测结果。
  2. 根据权利要求1所述的近视预测系统，其特征在于，
    所述用眼行为数据包括：日期、用户ID、平均用眼距离、平均用眼时长、平均环境光照、平均户外暴露时长、平均偏头角度和当前近视防控手段序列值；
    所述近视变化趋势数据包括：日期、用户ID、用户年龄、左眼屈光度数变化趋势标签值和右眼屈光度数变化趋势标签值。
  3. 根据权利要求1所述的近视预测系统，其特征在于，所述移动终端设备还用于采集用户的个人特征数据，所述云端服务器将用眼行为数据和个人特征数据合并为用户用眼数据，作为模型输入数据；
    所述个人特征数据至少包括：用户ID、用户年龄、学生性别、学生地区和遗传信息。
  4. 根据权利要求2所述的近视预测系统，其特征在于，所述左眼屈光度数变化趋势标签值和右眼屈光度数变化趋势标签值均通过以下数值表示相应的屈光度数变化趋势：
    在近视屈光度数一年增加数大于a时，或者三个月增加数大于 0.25a时，将屈光度数变化趋势标签值设为第一标签值；
    在近视屈光度数一年增加数为b到a之间时，或者三个月增加数为0.25b到0.25a之间时，将屈光度数变化趋势标签值设为第二标签值；
    在近视屈光度数一年增加数小于b时，或者三个月增加数小于0.25b时，将屈光度数变化趋势标签值设为第三标签值。
  5. 根据权利要求3所述的近视预测系统，其特征在于，所述深度学习模型采用attention编码器对t时刻及t时刻之前连续的N个用户用眼数据进行处理，并通过softmax编码器进行分类，输出三种屈光度数变化趋势标签值对应的概率。
  6. 根据权利要求1-5中任一项所述的近视预测系统，其特征在于，所述数据采集装置包括：
    数据采集模块，用于采集用眼行为数据，所述用眼行为数据至少包括用眼距离数据；
    主控模块，用于利用训练模型对当前的用眼距离数据进行处理，检测当前坐姿为标准坐姿或者非标准坐姿，在当前坐姿为非标准坐姿时生成提醒指令；
    提醒模块，用于在接收提醒指令时产生提醒信号以提醒用户；
    反馈模块，用于接收用户输入的反馈信号，并将所述反馈信号发送给所述主控模块对训练模型进行更新。
  7. 根据权利要求6所述的近视预测系统，其特征在于，所述主控模块包括：
    实时检测单元，用于将当前的用眼距离数据作为测试样本输入训练模型，输出包含标签值的检测结果数据，其中标签值用于标识标准坐姿或非标准坐姿；
    状态检测单元，用于检测结果数据的标签值对应非标准坐姿时生成提醒指令；
    模型训练单元，用于在未接收到反馈信号时将当前的用眼距离数 据及其标签值作为训练样本加入训练样本集，在接收到反馈信号时更正当前的用眼距离数据对应的标签值再加入训练样本集；所述模型训练单元利用所述训练样本集更新训练模型。
  8. 根据权利要求6所述的近视预测系统，其特征在于，所述数据采集模块包括：
    矩阵式测距单元，用于对预设的矩形测量区域内的距离进行采样；
    所述主控模块根据矩形测量区域内采样的多个距离数值生成样本序列，作为测试样本输入训练模型。
  9. 根据权利要求8所述的近视预测系统，其特征在于，所述数据采集装置还包括：
    输入模块，用于接收用户输入的身高值；
    升降机构，用于调整矩阵式测距单元的测距传感器的高度位置；
    所述主控模块还包括：传感器高度计算单元，用于根据用户输入的身高值确定测距传感器的预设高度位置，并发送信号给所述升降机构控制测距传感器达到预设高度位置；其中预设高度位置h满足以下公式：
    w*tan(FOV/2)≤h≤(H/15)+w*tan(FOV/2)；
    其中，w为桌面深度，FOV为测距传感器的视场角，H为用户输入的身高值。
  10. 一种近视预测方法，其特征在于，包括以下步骤：
    采集用户的用眼行为数据；
    采集用户的近视变化趋势数据；
    以用眼行为数据作为模型输入数据，以近视变化趋势数据作为验证结果，对深度学习模型进行训练；并利用训练后的深度学习模型对待预测的用眼行为数据进行处理，获得与用户用眼习惯匹配的近视变化趋势预测结果。
  11. 根据权利要求10所述的近视预测方法，其特征在于，所述方 法还包括：选取群体的近视变化趋势预测结果，分析整个群体中不同特征群体的近视风险。
PCT/CN2019/078885 2019-03-18 2019-03-20 近视预测系统及方法 WO2020186480A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910203610.4 2019-03-18
CN201910203610.4A CN109754885A (zh) 2019-03-18 2019-03-18 近视预测系统及方法

Publications (1)

Publication Number Publication Date
WO2020186480A1 true WO2020186480A1 (zh) 2020-09-24

Family

ID=66408864

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/078885 WO2020186480A1 (zh) 2019-03-18 2019-03-20 近视预测系统及方法

Country Status (2)

Country Link
CN (1) CN109754885A (zh)
WO (1) WO2020186480A1 (zh)

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110349057A (zh) * 2019-06-27 2019-10-18 浙江天地人科技有限公司 一种学习状态数据采集装置和系统
CN110288266A (zh) * 2019-07-03 2019-09-27 爱尔眼科医院集团股份有限公司 一种近视风险评估方法及系统
CN110797118B (zh) * 2019-09-14 2023-10-27 天津市九米九视光技术有限公司 一种防控近视的监测系统以及计算方法
CN111243742B (zh) * 2020-01-14 2023-08-25 中科海微(北京)科技有限公司 一种能够分析儿童用眼习惯的智能眼镜
CN111755125B (zh) * 2020-07-07 2024-04-23 医渡云（北京）技术有限公司 分析患者测量指标的方法、装置、介质及电子设备
CN111820865B (zh) * 2020-07-24 2024-05-17 安徽猫头鹰科技有限公司 一种眼部视觉数据采集在线监控系统
CN112289446A (zh) * 2020-10-29 2021-01-29 美视(杭州)人工智能科技有限公司 一种预测青少年近视的计算机系统
CN112700858B (zh) * 2020-12-14 2022-05-03 济南瞳星智能科技有限公司 一种儿童青少年近视预警方法及设备
CN113017831A (zh) * 2021-02-26 2021-06-25 上海鹰瞳医疗科技有限公司 人工晶体植入术后拱高预测方法及设备
CN113057577A (zh) * 2021-03-23 2021-07-02 成都爱尔眼科医院有限公司 一种青少年近视预测试诊断系统
CN114947726B (zh) * 2022-05-10 2023-02-28 北京神光少年科技有限公司 一种分析用眼习惯和用眼强度的计算方法
CN114937307B (zh) * 2022-07-19 2023-04-18 北京鹰瞳科技发展股份有限公司 用于近视预测的方法及其相关产品
CN115547497B (zh) * 2022-10-09 2023-09-08 湖南火眼医疗科技有限公司 基于多源数据的近视防控系统及方法
CN117059269B (zh) * 2023-08-10 2024-04-26 成都艾视医院管理有限公司 一种基于深度学习的青少年近视预测方法及模型
CN117671908A (zh) * 2023-12-06 2024-03-08 广州视域光学科技股份有限公司 基于行为监测的近视防控系统及其防控方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013040047A1 (en) * 2011-09-12 2013-03-21 Cornell University Biopolymer films and methods of making and using same
CN104095609A (zh) * 2014-05-20 2014-10-15 大连戴安科技有限公司 一种集预防、治疗、测量于一体的新型穿戴式智能近视治疗仪
JP2017029282A (ja) * 2015-07-30 2017-02-09 キヤノン株式会社 検査装置及び検査装置の制御方法
CN107595239A (zh) * 2015-06-02 2018-01-19 杭州镜之镜科技有限公司 个人用眼监控系统
CN108281197A (zh) * 2018-01-25 2018-07-13 中南大学 一种分析环境因素与青少年近视眼之间关系的方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2621944C2 (ru) * 2015-11-16 2017-06-08 Государственное бюджетное образовательное учреждение высшего профессионального образования "Южно-Уральский государственный медицинский университет" Министерства здравоохранения Российской Федерации (ГБОУ ВПО ЮУГМУ Минздрава России) Способ прогнозирования прогрессирования миопии у детей
CN105468147B (zh) * 2015-11-19 2018-07-27 宁波力芯科信息科技有限公司 一种预防近视的智能设备、系统及方法
CN108364687A (zh) * 2018-01-10 2018-08-03 北京郁金香伙伴科技有限公司 眼球状态预测方法及预测模型构建方法和设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013040047A1 (en) * 2011-09-12 2013-03-21 Cornell University Biopolymer films and methods of making and using same
CN104095609A (zh) * 2014-05-20 2014-10-15 大连戴安科技有限公司 一种集预防、治疗、测量于一体的新型穿戴式智能近视治疗仪
CN107595239A (zh) * 2015-06-02 2018-01-19 杭州镜之镜科技有限公司 个人用眼监控系统
JP2017029282A (ja) * 2015-07-30 2017-02-09 キヤノン株式会社 検査装置及び検査装置の制御方法
CN108281197A (zh) * 2018-01-25 2018-07-13 中南大学 一种分析环境因素与青少年近视眼之间关系的方法

Also Published As

Publication number Publication date
CN109754885A (zh) 2019-05-14

Similar Documents

Publication Publication Date Title
WO2020186480A1 (zh) 近视预测系统及方法
JP6010719B1 (ja) 健康管理サーバおよび健康管理サーバ制御方法並びに健康管理プログラム
CN111178313B (zh) 一种监督用户坐姿的方法及设备
CN109819402B (zh) 用于监督改善学习习惯的方法及其系统
CN105373767A (zh) 一种智能手机使用的眼睛疲劳检测方法
CN112365107B (zh) 一种基于人工智能的近视风险评估方法、装置及系统
CN112932402A (zh) 一种基于人工智能及智能感知的自助视力筛查系统
US20210082056A1 (en) Information processing apparatus and information processing method
CN112464793A (zh) 一种在线考试作弊行为检测方法、系统和存储介质
US20230237699A1 (en) Method and system for itelligently controlling children's usage of screen terminal
CN111734974B (zh) 一种具有坐姿提醒功能的智能台灯
CN111402094B (zh) 一种具有防近视功能的学生学习用辅助系统
CN105147237A (zh) 视力测试方法
CN110020628A (zh) 基于面部检测的坐姿检测方法、系统、设备及存储介质
CN110623631A (zh) 一种基于近视眼镜的学习监护系统
CN109783999B (zh) 校园近视防控装置及方法
KR20200000680U (ko) 학습 집중도 향상 장치
CN111700584A (zh) 一种视力智能检测系统
CN115240849A (zh) 电子产品使用特征及视疲劳的记录和提醒方法及系统
CN206964368U (zh) 智能学生椅
CN112633411A (zh) 坐姿检测装置及坐姿检测方法
CN112418022A (zh) 一种人体数据的检测方法及装置
CN112530141A (zh) 一种基于tof传感器的不良坐姿监测方法
JP6069652B1 (ja) 健康管理サーバ及びその制御方法並びに健康管理メッセージアプリケーションプログラム
CN109143616A (zh) 一种智能监测眼镜及配套眼镜的智能镜盒

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19920265

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19920265

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19920265

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 09.03.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19920265

Country of ref document: EP

Kind code of ref document: A1