Method and system for measuring optical center distance of myopia prevention and control lenses for children

Abstract

The present application relates to the technical field of lens assembly, and discloses a kind of optical center distance measurement method and system for children myopia prevention and control lens, the method includes obtaining face video stream under multiple variable viewing distances, intercepts eye image sequence with timestamp, extracts pupil coordinate set and pupil offset vector sequence;According to the pupil offset direction and amplitude, determine the nasal side offset state, converge the line of sight of both eyes and map the convergence coordinates to viewing distance condition data;Match eye scene label, weight clustering is carried out to scene pupil coordinates, and target pupil position is obtained;Quantify the defocus misregistration between target pupil position and lens center, complete optical center calibration ranging in combination with pupil coordinates, obtain target optical center distance, and the optical center of children myopia prevention and control lens is positioned and assembled according to the target optical center distance.The method can improve the defocus control effect of children myopia prevention and control lens.

Description

Technical Field

[0001] This invention relates to the field of lens assembly technology, and in particular to a method and system for measuring the optical center distance of lenses for myopia control in children. Background Technology

[0002] Currently, in the process of fitting myopia control lenses for children, it is necessary to understand the spatial correspondence between the eye position and the center of the lens under different eye use conditions. Therefore, a non-contact visual measurement method that can continuously track the spatial coordinates of the face and eyes is needed, such as using facial recognition technology to provide a reliable basis for eye position.

[0003] In one existing technology, the fitting personnel use a pupillary distance ruler or electronic pupillary distance meter to measure the distance. They switch between the left and right eye measurement channels to make the measuring cursor coincide with the corneal reflection point or the center of the pupil, read the pupillary distance of one eye or both eyes, determine the optical center position of the lens based on the pupillary distance value, and enter the parameters into the processing equipment to complete the lens centering, recentering, edge grinding and assembly. Existing technologies use a fixed head position and a sight target to read the single pupillary distance and determine the lens processing center using a pupillary distance ruler or electronic pupillometer. Although the industry has established standardized fitting procedures such as pupillary height measurement, frame fit adjustment, and regular check-ups to compensate for the shortcomings of single static measurement, these measures are still based on static point measurement data under limited discrete viewing distances (far / near). However, when children read, write, and use electronic devices, their viewing distances change continuously. The horizontal inward displacement of the pupil center in the frame coordinate system caused by binocular convergence exceeds the conventional alignment tolerance threshold of defocused lenses. Therefore, traditional contact or short-term static measurement methods cannot continuously and non-contactly obtain the continuous drift trajectory of the pupil center under various viewing distances during children's natural eye use. As a result, the optical center after processing can only be set as a compromise, and the actual line of sight is very likely to deviate from the lens defocus design area.

[0004] In summary, existing technologies are insufficient to improve the defocus control effect of myopia prevention lenses for children. Summary of the Invention

[0005] This invention provides a method and system for measuring the optical center distance of lenses for myopia control in children, so as to improve the defocus control effect of lenses for myopia control in children.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for measuring the optical center distance of lenses for myopia prevention and control in children, comprising: Acquire a facial video stream under varying viewing distances, and extract temporal images of the eyes from the facial video stream to obtain an eye image sequence and corresponding timestamps; Based on the eye image sequence and the corresponding timestamp, the pupil temporal trajectory is extracted to obtain a set of pupil coordinates and a sequence of pupil offset vectors. The nasal displacement state is determined based on the pupil offset vector sequence to obtain the nasal offset state identifier. The visual field spatial convergence is performed based on the nasal offset state identifier to obtain the convergence coordinate point. The focal length relationship depth mapping is performed based on the convergence coordinate point to obtain the visual distance condition data. Based on the viewing distance condition data, viewing distance scene label matching is performed to obtain eye use scene labels. Then, based on the eye use scene labels, the pupil coordinate set is subjected to scene position weighted clustering to obtain the target pupil position under different eye use scenarios. Based on the target pupil position under different eye use scenarios and the preset lens center, the defocus misalignment is quantified to obtain the defocus misalignment amount; The optical center distance of the target is obtained by optical center calibration and distance measurement based on the defocus misalignment, the pupil coordinate set, and the lens center.

[0007] In a second aspect, the present invention provides an optical center distance measurement system for myopia control lenses for children, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention extracts the pupil temporal trajectory from a sequence of eye images with timestamps to obtain a set of pupil coordinates and a sequence of pupil offset vectors. It filters the left and right nasal-side movements of the eyes based on the lateral displacement amplitude and its direction relative to the vertical line of the face, and synchronously pairs them according to the timestamps to generate a nasal-side offset status identifier. When this status identifier is triggered, it combines the spatial pose of the binocular image acquisition device to converge the gaze of both eyes, and uses the focal length mapping relationship to convert the convergence coordinates into visual distance condition data. Since the displacement amplitude and movement direction can exclude non-nasal-side eye movements, the synchronous pairing of the left and right eyes can identify the convergence state of both eyes, solving the problem that a single static pupillary distance measurement cannot correlate the nasal-side movement of the pupil with the actual visual distance, and achieving reliable acquisition of the pupil convergence state and corresponding visual distance under varying visual distances.

[0010] (2) This invention obtains eye-use scene labels by matching viewing distance condition data with viewing distance intervals in a scene mapping table, filters pupil coordinates according to eye-use scene labels, obtains cluster centers by spatial clustering, and then performs scene weight inverse proportional allocation and coordinate weighted summation based on the variance of coordinate offset within the cluster. Since eye-use scene labels can distinguish pupil position data according to actual viewing distance, and coordinate offset variance can reflect the positional stability of each cluster, clusters with concentrated distributions can obtain higher weights in coordinate fusion, solving the problem that a single pupil position is difficult to reflect the true gaze position of children in different eye-use scenarios, and improving the scene adaptability and stability of the target pupil position.

[0011] (3) This invention quantifies the defocus misalignment between the target pupil position and the lens center under different eye usage scenarios, combines the defocus misalignment amount and pupil coordinate set to calculate the visual axis eccentricity distance, and calibrates the lens center based on the corneal vertex coordinates and wearing posture constraints. It then calculates and corrects the spatial Euclidean distance between the lens center and the corneal vertex coordinates to obtain the target optical center distance and performs optical center positioning and assembly accordingly. Since the defocus misalignment amount can provide a calibration basis for visual deviation, and the corneal vertex coordinates and wearing posture constraints can correct the positional deviation of the lens center under actual wearing conditions, it solves the problem that fixed pupillary distance data cannot take into account both visual distance changes and wearing posture, improves the accuracy of optical center positioning and assembly of myopia control lenses for children, and enhances the defocus control effect. Attached Figure Description

[0012] Figure 1 This is a schematic flowchart of the optical center distance measurement method for myopia control lenses for children provided in the first embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for measuring the optical center distance of lenses for myopia control in children, comprising the following steps: S11, acquire the facial video stream under varying viewing distances, and perform temporal image extraction of the eyes from the facial video stream to obtain an eye image sequence and corresponding timestamps; S12, based on the eye image sequence and the corresponding timestamp, extract the pupil temporal trajectory to obtain a pupil coordinate set and a pupil offset vector sequence; S13, determine the nasal displacement state based on the pupil offset vector sequence to obtain the nasal offset state identifier, and perform binocular visual space convergence based on the nasal offset state identifier to obtain the convergence coordinate point, and perform focal length relationship depth mapping based on the convergence coordinate point to obtain visual distance condition data. S14, perform visual distance scene label matching based on the visual distance condition data to obtain eye use scene labels, and perform scene position weighted clustering on the pupil coordinate set based on the eye use scene labels to obtain the target pupil position under different eye use scenarios. S15, quantify the defocus misalignment of the line of sight based on the target pupil position and the preset lens center under different eye use scenarios, and obtain the defocus misalignment amount; S16, perform optical center calibration distance measurement based on the defocus misalignment amount, the pupil coordinate set, and the lens center to obtain the target optical center distance.

[0015] In step S11, a facial video stream under varying viewing distances is acquired, and eye time-series images are extracted from the facial video stream to obtain an eye image sequence and corresponding timestamps.

[0016] Specifically, the facial video stream is acquired by a binocular image acquisition device that performs binocular targeting. During acquisition, the child's head is positioned within the effective depth of field of the binocular image acquisition device, and the child sequentially gazes at multiple distance targets corresponding to the lens product design document. The image resolution is selected based on the condition that the pupil diameter within the effective eye area is no less than 12 pixels. If this condition is not met, the resolution is increased or the distance between the binocular image acquisition device and the child is shortened. The sampling period is the reciprocal of the frame rate. After acquisition begins, the frame rate, exposure time, and gain are not changed, thus maintaining consistency in the temporal interval, brightness scale, and spatial resolution of the facial video stream.

[0017] It should be noted that the variable viewing distance mentioned in this invention refers to the continuously switching gaze distance of children in different eye use scenarios, including but not limited to reading and writing distance, electronic device usage distance, and distance for looking into the distance; wherein, the reading and writing distance is 0.25 meters to 0.40 meters, corresponding to near-distance eye use scenarios; the electronic device usage distance is greater than 0.40 meters and does not exceed 0.80 meters, corresponding to medium-distance eye use scenarios; the distance for looking into the distance is greater than 0.80 meters, corresponding to far-distance eye use scenarios, such as looking at a blackboard or looking into the distance outdoors; during data collection, children are asked to look at the corresponding targets in sequence according to the above three distance intervals to cover the entire range of viewing distances in children's daily eye use.

[0018] Before performing face recognition on each image frame, the pixel matrix is ​​normalized according to the upper limit of pixel values. The calibration target is a planar target covered with a checkerboard pattern. The target size and the side length of the checkerboard are calibrated by measurement. The spatial coordinates of the checkerboard corner points are known. During calibration, the target is placed at the minimum measurement distance, the maximum measurement distance, and at least three intermediate distances between them. At least 10 images are collected at each distance. The corner point detection algorithm is used to extract the pixel coordinates of the checkerboard corner points, and a correspondence is established with the known spatial coordinates to obtain the intrinsic parameter matrix, distortion coefficients, and dual-target calibration extrinsic parameters. The aspect ratio of the original image is maintained and input into the face recognition function attached to the binocular image acquisition device. When multiple facial regions appear in a single image frame, the facial region that simultaneously contains the center of both eyes and has the largest area is selected. If the areas are the same, the facial region with higher confidence is selected. The sampling period is the reciprocal of the rated frame rate of the binocular image acquisition device. The first frame uses a unified hardware clock to write the acquisition start timestamp. The timestamp of each subsequent frame is determined by adding the cumulative value of the sampling periods completed before that frame to the acquisition start timestamp.

[0019] Based on the left and right corner points output by the facial recognition function, the minimum bounding rectangles for the left and right corner points are calculated respectively. These rectangles are then expanded outwards by 1 / 4 of their width to obtain a complete cropping range covering the eyelids and pupils. The corresponding timestamp is preferentially read from the acquisition time written by the unified hardware clock. If no acquisition time is written, the timestamp is calculated by adding the first frame acquisition time to the product of the frame number and the sampling period. The eye image sequence is arranged in ascending order according to the corresponding timestamps. Each record includes the corresponding timestamp, eye type, the origin coordinates of the cropping range in the facial video stream, and the cropped pixel matrix, thus establishing a direct correspondence between the cropping operation and the eye image sequence and the corresponding timestamp. Image frames are marked as invalid if the facial region is missing, either eye corner point is outside the image boundary, the pixel matrix overexposure ratio exceeds 1%, or the time interval between adjacent image frames is greater than two sampling periods. The number of effective eye images should cover the fixation period of targets at each distance. The number of effective eye images for targets at each distance should not be less than the number of frames generated within 1 second at the rated frame rate of the binocular image acquisition device, and should not be less than 30 frames. If the rated frame rate is less than 30 frames per second, the fixation period of the target at that distance should be extended until 30 effective eye images are obtained. Otherwise, pupil temporal trajectory extraction should not be performed.

[0020] The facial video stream, eye image sequence, and corresponding timestamps generated through the above checks have a unified data format, stable sampling interval, and clear effective boundaries, making them repeatable inputs for subsequent timing operations. Before formal acquisition, flash signals simultaneously entering the field of view of two acquisition channels are used to check channel synchronization, and the absolute difference in the timestamps of the first brightness peak of the flash signals in the two acquisition channels is determined as the synchronization error. The upper limit of the synchronization error tolerance is taken as the smaller value between one sampling period and the channel synchronization accuracy specified in the technical parameters of the binocular image acquisition device. Acquisition begins when the synchronization error is not greater than the upper limit of tolerance; if it exceeds the upper limit of tolerance, acquisition stops and channel resynchronization is performed. If the error still exceeds the limit after resynchronization, the measurement is terminated. The synchronization error, the upper limit of the synchronization error tolerance, the effective frame ratio, and the processing branch are all written into the acquisition record, ensuring that the rationality of the facial video stream data can be quantitatively checked before entering S12.

[0021] In step S12, pupil temporal trajectory is extracted based on the eye image sequence and the corresponding timestamp to obtain a set of pupil coordinates and a sequence of pupil offset vectors.

[0022] Specifically, the output color space in the technical parameters of the binocular image acquisition device is first read, and then each color eye image is converted to grayscale pixel by pixel according to the weighting coefficients of the red, green, and blue channels specified in the color space standard. When the output color space is BT.601, the channel weights specified in BT.601 are read; when the output color space is BT.709, the channel weights specified in BT.709 are read; when the output color space is sRGB, the grayscale conversion channel weights of sRGB are the same as those of BT.709, and the channel weights specified in BT.709 are read; when other color spaces are used, the grayscale conversion coefficients specified in the technical parameters of the device are used, and the original pixel values ​​of a single-channel image are directly read to obtain grayscale data with the same size as the original image.

[0023] The median filter window should not exceed 1 / 4 of the minor axis length of the fitted ellipse to avoid filtering out the true pupil edge. At least 30 clear eye calibration images should be acquired during equipment delivery calibration. The grayscale difference amplitude should be calculated on both sides of the manually confirmed pupil boundary, and the 5th percentile of all valid grayscale difference amplitudes should be written into the binocular image acquisition equipment calibration file as the minimum resolvable grayscale difference. A curve formed by candidate edge pixels with a first-to-last distance of no more than 2 pixels is considered a closed curve. Closed curves that touch the eye image boundary or enclose an area less than 1% of the eye image area are deleted. If there is more than one candidate closed curve, the closed curve with the lowest internal average grayscale and the largest number of consecutive pixels in its contour is selected, thus ensuring that the grayscale difference calculation and contour selection stably correspond to the pupil contour. For candidate edge pixels, the square root of the sum of the square of the horizontal and vertical grayscale differences of the same pixel is taken as the grayscale difference amplitude and compared with the minimum resolvable grayscale difference.

[0024] The pixel coordinates in the pupil contour are fitted using a least-squares ellipse. The fitting object is a quadratic curve containing horizontal quadratic terms, horizontal and vertical intersection terms, vertical quadratic terms, horizontal linear terms, vertical linear terms, and a constant term, with the combination of quadratic term coefficients constrained to satisfy the ellipse discrimination condition. The fitting output includes the ellipse center, major axis, minor axis, rotation angle, and fitting residual. The fitting residual is the root mean square distance from the pupil contour pixels to the fitted ellipse. During device delivery calibration, the 95th percentile of the distance between the manually labeled pupil center and the fitted ellipse center is determined as the maximum pupil positioning error. The effective pupil position coordinates are written into the pupil coordinate set according to the corresponding timestamp. Each record retains the corresponding timestamp, eye type, horizontal coordinate, vertical coordinate, fitting residual, major axis, and minor axis fields. During equipment delivery calibration, at least 30 static gaze images of targets at fixed distances were used. After removing images with blinks, occlusions, and truncated edges, the two-dimensional distance between the pupil center of the intermediate frame and the average pupil center of the adjacent frames was calculated frame by frame. All valid distances were arranged in ascending order, and the 99.7 percentile was used to determine the outlier threshold for pupil position. If the two-dimensional distance between the intermediate record and the average coordinates of the two preceding and following records exceeds the outlier threshold, the intermediate record was marked as an outlier. The pupil coordinate set was arranged in ascending order according to the corresponding timestamp and in left-eye and right-eye order, with consistent origin, unit, eye type, and quality fields to meet the data requirements for subsequent orientation determination and scene selection. The fitting residual was the square root of the average squared distance from all pupil contour pixels to the fitted ellipse.

[0025] Within the same eye, two adjacent valid pupil position coordinates are extracted. The horizontal coordinate of the latter is subtracted from the horizontal coordinate of the former, and the vertical coordinate of the latter is subtracted from the vertical coordinate of the former. These are then combined into a motion vector in the order of horizontal and vertical components. The horizontal and vertical components retain their signs and are expressed in pixels. When comparing speeds, they are divided by the difference between the corresponding timestamps, without replacing the original motion vector with speed. To verify the repeatability of the pupil position coordinates, the mean horizontal and vertical coordinates are calculated for odd and even frames during the static fixation phase at a fixed distance target. The two-dimensional Euclidean distance between the two mean coordinates is then calculated. All effective distances for all eyes and at the target are sorted in ascending order. The static fixation fluctuation is determined by the 95th percentile, and the sample size, screening criteria, and static fixation fluctuation are written into the binocular image acquisition device calibration file. The 95th percentile corresponds to approximately twice the standard deviation. During re-extraction, only the median filter window can be switched between 3×3 and 5×5 without changing the minimum resolvable grayscale difference or the maximum pupil positioning error. Each record in the pupil offset vector sequence simultaneously stores the corresponding timestamps and eye type of the two preceding and following pupil position coordinates, and is arranged in ascending order according to the corresponding timestamps.

[0026] In step S13, the nasal displacement state is determined based on the pupil offset vector sequence to obtain a nasal offset state identifier. Then, binocular visual space convergence is performed based on the nasal offset state identifier to obtain convergence coordinate points. Finally, focal length relationship depth mapping is performed based on the convergence coordinate points to obtain visual distance condition data, including: The motion vectors in the pupil offset vector sequence are orthogonally decomposed to obtain two-dimensional Cartesian coordinate components; The horizontal axis displacement component is extracted from the two-dimensional Cartesian coordinate components to obtain the lateral displacement value; The facial midline is located based on the facial video stream to obtain the position of the facial midline. The direction of horizontal displacement is determined based on the value of the lateral displacement and the position of the vertical line of the face. The nasal side movement state is filtered based on the lateral displacement value and the horizontal displacement direction to obtain the nasal side offset state identifier.

[0027] When the nasal offset status indicator indicates that nasal movement has occurred, external parameters are obtained from the binocular image acquisition device to obtain the rotation vector and translation vector; Construct a three-dimensional spatial pose matrix based on the rotation vector and the translation vector; Based on the three-dimensional spatial pose matrix, the pupil coordinates of both eyes in the pupil coordinate set are spatially transformed to obtain the gaze direction of both eyes; The convergence coordinates are obtained by triangulating the direction of the binocular lines of sight based on the epipolar geometric constraints.

[0028] The convergence coordinates are input into a pre-established focal length mapping relationship for coordinate distance conversion to obtain the spatial depth value; The spatial depth value is associated with the nasal side offset status identifier to obtain the nasal side movement distance record; The nasal-side movement distance record is structured and encapsulated to obtain the distance condition data.

[0029] The step of filtering the nasal side movement state based on the lateral displacement value and the horizontal displacement direction to obtain the nasal side offset state identifier includes: The lateral displacement value is compared with a preset movement determination threshold, and lateral displacement values ​​that are greater than the movement determination threshold are filtered out to obtain the lateral displacement value exceeding the threshold. From the above threshold lateral displacement values, the lateral displacement values ​​of the corresponding horizontal displacement directions toward the vertical line of the face are selected according to the left eye and the right eye respectively, to obtain the left eye nasal side movement displacement value and the right eye nasal side movement displacement value. Based on the corresponding timestamp, the left eye nasal side movement displacement value and the right eye nasal side movement displacement value are synchronously paired to obtain a pair of bilateral nasal side movement data; For each pair of bilateral nasal side movement data, the corresponding timestamp, the displacement value of the left eye nasal side movement, and the displacement value of the right eye nasal side movement are written into a status data item, and the status field of the status data item is set to nasal side convergence status to obtain the nasal side offset status identifier.

[0030] Specifically, before orthogonal decomposition, the motion vector is rotated and corrected in a plane based on the angle between the line connecting the inner corners of the left and right eyes in the same image frame and the horizontal axis of the facial video stream, so that the corrected horizontal axis is parallel to the line connecting the inner corners of the left and right eyes. Rotation correction is not performed on the image frame if the length of the line connecting the inner corners of the left and right eyes is less than 1 / 5 of the width of the facial region or if the corner confidence is lower than the lower confidence limit of the facial recognition function. The corrected horizontal component is read from the two-dimensional Cartesian coordinate components, and its absolute value is determined as the horizontal displacement value. The sign of the horizontal component is used as the direction information for subsequent horizontal displacement direction calculations. If the absolute horizontal distance from the current pupil position to the vertical line of the face is less than the corresponding distance from the previous pupil position, the horizontal displacement direction is determined to be towards the vertical line of the face; if it is greater, it is determined to be away from the vertical line of the face. If the difference between the two does not exceed the maximum pupil positioning error, it is marked as having no determined direction.

[0031] For planar rotation correction, the product of the horizontal displacement component before correction and the cosine of the rotation angle is added to the product of the vertical displacement component and the sine of the rotation angle to obtain the corrected horizontal displacement component; the product of the vertical displacement component and the cosine of the rotation angle is subtracted from the product of the horizontal displacement component and the sine of the rotation angle to obtain the corrected vertical displacement component. Under normal symmetry, the position of the facial perpendicular line is taken as the midpoint of the horizontal coordinates of the inner canthus of the left and right eyes. During the initial measurement, the device compares the widths of the left and right palpebral fissures formed by the inner and outer canthi of the left and right eyes. When the difference between the widths of the two palpebral fissures exceeds 10% of the width of the larger palpebral fissure, it is determined that the inner canthi are significantly asymmetrical, and the midpoint of the line connecting the outer canthi of the left and right eyes is used to determine the position of the facial perpendicular line; when the confidence of either outer canthus is lower than the lower confidence limit of the facial recognition function, the center point of the bridge of the nose marked and saved by the fitting personnel in the facial image during the initial measurement is read as the reference for the facial perpendicular line.

[0032] The preset movement determination threshold is determined during the delivery and calibration phase of the binocular image acquisition equipment and written into the binocular image acquisition equipment calibration file. The upper limit of displacement is determined by the static gaze calibration sequence of each fixed distance target during the equipment delivery and calibration phase. The calibration sequence retains only image frames that are binocularly synchronized, without blink occlusion, and whose pupil fitting residual does not exceed the maximum pupil positioning error. The absolute value of the corrected lateral displacement of adjacent effective frames of the left and right eyes is calculated respectively, summarized by eye and distance target and arranged in ascending order. The 95th percentile of all effective values ​​is determined as the upper limit of displacement.

[0033] The motion determination threshold is the larger of the upper limit of displacement and twice the maximum pupil positioning error. The twice maximum pupil positioning error is used to cover the cumulative effect of the positioning errors of two adjacent frames and is uniformly converted to pixel units. From the lateral displacement values ​​exceeding the threshold, records of horizontal displacement directions towards the vertical line of the face for both the left and right eyes are selected to obtain the nasal side movement displacement values ​​for the left and right eyes. Records for each eye are scanned in ascending order according to their corresponding timestamps. When the difference between the corresponding timestamps is no greater than one sampling period, a pair of nasal side movement data is formed, with each record participating in only one pairing. If any lateral displacement value in a pair of nasal side movement data exceeds half the width of the eye image, the pair is marked as invalid to exclude mislocalization of the eye region. The calibration file of the binocular image acquisition device simultaneously saves the sample number of the upper limit of displacement, the filtering conditions, the corresponding percentiles, the motion determination threshold, and the measurement number. This is directly read during runtime without using an external training database, preset weights, or adjustable coefficients.

[0034] It should be noted that the status field uses a two-bit binary code: 00 indicates no nasal movement, 01 indicates both eyes converge towards the nose simultaneously, and 10 indicates invalid data or the status cannot be determined. When the status field in the nasal offset status identifier is 01, the rotation vector and translation vector are read from the calibration file of the binocular image acquisition device. The rotation vector represents the rotation axis and rotation angle of the right eye acquisition coordinate system relative to the left eye acquisition coordinate system, and the translation vector represents the baseline direction and length between the origins of the two acquisition coordinate systems. After calibration, the reprojection error is verified using the calibration image that was not used in the solution. If the error exceeds the spatial measurement error specified in the device's technical parameters, recalibration is performed. The binocular gaze direction is calculated using the pupil gaze calibration function established before the child's first measurement. This function maps the normalized pupil coordinates to the corresponding unit direction vector for each eye.

[0035] The fitting samples for the pupil gaze calibration function were obtained from the child's effective fixation images of the calibration target before the first measurement. For each eye, at least nine calibration target directions were used: center, top, bottom, left, right, and four diagonal directions. For each direction, at least 10 frames of normalized pupil coordinates and unit direction vectors determined by the spatial position of the calibration target were retained to meet the pupil positioning error requirements. Each unit gaze direction component was fitted using a quadratic polynomial containing a constant term, a normalized horizontal coordinate linear term, a normalized vertical coordinate linear term, a normalized horizontal coordinate square term, a normalized horizontal and vertical coordinate product term, and a normalized vertical coordinate square term.

[0036] The above six coefficients require at least six sets of linearly independent calibration target directions to form six equations. In this embodiment, at least nine directions are used for overdetermined solution using the least squares method, covering the effective field of view specified by the technical parameters of the binocular image acquisition device product in the central and surrounding directions. If the rank of the sample matrix is ​​less than 6 or there are fewer than 10 effective frames in any direction, fitting is stopped and re-acquisition is performed. After obtaining the three components of the unit gaze direction, they are normalized according to the three-dimensional vector magnitude. If the input normalized pupil coordinates exceed the coverage range of the calibration samples, the corresponding record is marked as invalid and extrapolation is not performed. All coefficients are written into the calibration file of the binocular image acquisition device according to the eye and the current measurement number, and are not updated during runtime.

[0037] Based on the intrinsic parameter matrices of the two acquisition channels, the pupil coordinates of both eyes in the pupil coordinate set that are closest to the timestamp corresponding to the state data item and whose time difference is no greater than one sampling period are converted into normalized pupil coordinates, and then substituted into the pupil gaze calibration function. The solution function is a bivariate linear equation with respect to two gaze parameters. The equation coefficients are directly composed of the dot product of two unit direction vectors and the difference between the two origins, without setting the number of iterations. The pre-established focal length mapping relationship is formed by the binocular image acquisition device during the delivery calibration phase. The physical distance of the calibration target covers the minimum to maximum measurement distance specified in the device product technical parameters. For each focal length mapping sample pair, the number of samples and the interquartile range of the depth components are recorded simultaneously. When the interquartile range exceeds 5% of the absolute value of the median of the depth components of the same node, the node is re-acquired. The left eye gaze and the right eye gaze are respectively composed of the gaze origin of the corresponding eye, the unit gaze direction vector, and the distance parameter along that direction. The closest point between the two lines of sight is obtained by the aforementioned binary linear equation. The three coordinate components of the two closest points are added together and then divided by 2. The resulting three-dimensional coordinates are used as the convergence coordinate point. If the distance between the closest points of the two lines of sight exceeds the spatial measurement error specified in the technical parameters of the binocular image acquisition device, the convergence coordinate point is marked as invalid.

[0038] When the convergence point is within the center area of ​​the binocular image acquisition device's screen, its depth component along the device's optical axis is read as the focal length mapping independent variable. When the horizontal distance from the projection position of the convergence point to the center of the screen exceeds 1 / 3 of the screen width, the spatial Euclidean distance from the convergence point to the origin of the binocular acquisition coordinate system is calculated, and the obtained value is used as the focal length mapping independent variable. Focal length mapping sample pairs are arranged in ascending order according to the focal length mapping independent variable. When the current focal length mapping independent variable is between two adjacent sample pairs, the interval ratio of the current independent variable relative to the previous node is first calculated, and then this ratio is multiplied by the difference in physical distance between the two adjacent nodes. The resulting increment is added to the physical distance of the previous node to obtain the spatial depth value. When the convergence point coincides with the interpolation node, the physical distance of that node is directly read. When it is between two nodes, linear interpolation is performed, and extrapolation of values ​​outside the node interval is not performed.

[0039] The focal length mapping sample nodes must cover all viewing distance interval boundaries in the scene mapping table. When the current focal length mapping independent variable is less than the minimum node or greater than the maximum node, a nasal movement viewing distance record is still established, with the physical distance of the minimum or maximum node written in, and a near-end boundary marker or a far-end boundary marker added, so that S14 assigns the record to the first or last eye-use scene. When the status field is 00, a nasal movement viewing distance record is still established, with the spatial depth value written in the physical distance of the maximum focal length mapping node and a far-end boundary marker added to preserve the far-distance pupil position in a near-parallel state of both eyes. When the status field is 10 or any required field is missing, an invalid record is established and the reason for failure is saved. The valid spatial depth value and the nasal offset status marker are associated according to the corresponding timestamp in the status data item, and a consistency check of nasal movement and viewing distance change is performed on continuous records. When there is only a single record, no trend check is performed, but two closest point distances, focal length mapping interval numbers, and boundary markers are retained.

[0040] In step S14, visual distance scene label matching is performed based on the visual distance condition data to obtain visual scene labels. Then, based on the visual scene labels, scene location weighted clustering is performed on the pupil coordinate set to obtain the target pupil position under different visual scenes, including: The spatial depth value is extracted from the line-of-sight condition data to obtain the line-of-sight value; The viewing distance value is matched with the viewing distance range in the preset scene mapping table to obtain the matching scene range; Read the scene labels corresponding to the matching scene interval to obtain the eye use scene labels.

[0041] Based on the eye-use scene label, the pupil position coordinates are filtered from the pupil coordinate set to obtain the scene pupil coordinate set; Spatial clustering is performed on the pupil coordinate set of the scene to obtain the cluster centers; Calculate the variance of the pupil position coordinates relative to the center of the cluster within each cluster; The scene association weights are obtained by inversely allocating scene weights based on the coordinate offset variance. The target pupil position under different eye use scenarios is obtained by performing a weighted summation of coordinates based on the cluster center and the scene association weight.

[0042] Specifically, the preset scene mapping table is established from the lens product design documents of children's myopia control lenses. Each record includes a lower limit of the viewing distance interval, an upper limit of the viewing distance interval, and a scene label. The lower and upper limits of the interval use the same length unit as the spatial depth value, and the lens product model and document version are recorded simultaneously when writing. Valid nasal movement viewing distance records are read one by one from the viewing distance condition data, and the spatial depth value is extracted as the viewing distance value, while inheriting the corresponding timestamp and status fields. When the viewing distance value matches the upper limit of an internal interval, it is assigned to the next matching scene interval; when it matches the upper limit of the last interval, it is assigned to the last matching scene interval. When the viewing distance value is greater than or equal to the lower limit of the viewing distance interval for a certain eye use scenario and less than the upper limit of that interval, the eye use scenario label is written; records with near-end boundary markers or far-end boundary markers are directly written to the first or last eye use scenario label, respectively.

[0043] Based on the timestamps corresponding to the eye use scene labels, pupil position coordinates for the same eye type with a timestamp difference of no more than one sampling period are searched in the pupil coordinate set. The scene pupil coordinate set uses the top left corner of the facial video stream as the origin, with units in pixels, and is not mixed with the frame coordinate system before clustering. If a scene use label has no valid pupil position coordinates for any eye type, the scene pupil coordinate set corresponding to that eye type is marked as empty, and data from the other eye type is not used to fill it in. Spatial clustering adopts a deterministic pointwise clustering function based on Euclidean distance. The clustering distance threshold is the sum of the maximum pupil positioning error and the static gaze fluctuation, both of which are read from the calibration file of the binocular image acquisition device corresponding to the same measurement number and unified to pixel units. The scene pupil coordinate set is input into the clustering function in ascending order of the corresponding timestamps. The first cluster center is established with the first pupil position coordinate, and then the Euclidean distance from the subsequent coordinates to all existing cluster centers is calculated. A new cluster is created when the minimum Euclidean distance is greater than the clustering distance threshold. After each coordinate is written, the cluster center is updated with the average of the horizontal and vertical coordinates within the cluster. The distance from the coordinate to the cluster center is the square root of the sum of the squares of the differences in their horizontal and vertical coordinates.

[0044] To avoid short-term mislocalization leading to independent clusters, a valid sample count check is performed on each cluster after clustering. The minimum valid sample count is the number of frames generated within 0.1 seconds at the rated frame rate of the binocular image acquisition device, rounded up, and must not be less than 3 frames. This value is written into the calibration file of the binocular image acquisition device after the rated frame rate is determined. When the number of valid clusters is 0, the application eye scene label and eye-specific label will be marked as having no valid clustering results, and the target pupil position will not be output. For each valid cluster, the average horizontal and vertical coordinates of all pupil position coordinates within the cluster are calculated to obtain the cluster center. The horizontal and vertical coordinates of the cluster center are stored as double-precision values. The calculation is not rounded up beforehand; the decimal places are retained according to the device coordinate precision after the target pupil positions under different eye usage scenarios are formed. The cluster center is composed of the average horizontal and vertical coordinates of all pupil position coordinates within the cluster. The coordinate offset variance is obtained by averaging the squared two-dimensional Euclidean distances from each coordinate within the cluster to the cluster center. Both operations use all valid samples from the cluster and do not use additional letter variables.

[0045] Before calculating the scene association weights, the scene pupil coordinate set is divided into stationary gaze records and saccade records using the movement determination threshold of S13. Records with monocular displacement exceeding the movement determination threshold are saved as saccade records but do not participate in the representative gaze position weighting. Therefore, the subsequent coordinate offset variance reflects the stability of stationary gaze rather than the reading saccade range. The scene association weight of each effective cluster is allocated according to the reciprocal of its coordinate offset variance and normalized by the sum of the reciprocals of the variances of all effective clusters. When the variance is less than the lower variance limit, it participates in the reciprocal calculation according to the lower variance limit, which is the square of the maximum pupil positioning error in the binocular image acquisition device calibration file. Stationary gaze clusters with smaller variances are given higher weights to reduce the impact of blink edge frames and positioning jitter on the representative gaze position. When there is only one effective cluster, its scene association weight is set to 1. The target pupil position under different eye use scenarios is obtained by multiplying the center of each effective cluster by its scene association weight and then summing the coordinate components. Each record is arranged in ascending order according to the lower limit of the viewing distance range in the scene mapping table. Under the same viewing scene label, the left eye records are arranged first, followed by the right eye records.

[0046] The viewing distance ranges in the scene mapping table are determined based on children's daily eye-use behavior data. Ranges less than or equal to 0.4 meters correspond to reading and writing scenarios; ranges greater than 0.4 meters but less than or equal to 0.8 meters correspond to electronic device scenarios; and ranges greater than 0.8 meters correspond to blackboard viewing scenarios. The viewing distance range division is based on the classification standards for children's fixation distances in different eye-use scenarios in the reference "Guidelines for Appropriate Technologies for Myopia Prevention and Control in Children and Adolescents (Updated Version)," and refers to the generally accepted formula for calculating accommodation demand in the field of optometry, i.e., the reciprocal of the fixation distance equals the accommodation demand. 0.4 meters corresponds to 2.5D accommodation demand, which is the upper limit for continuous near-field eye use; 0.8 meters corresponds to 1.25D accommodation demand, which is for intermediate-distance eye use. The upper limit of this classification has been adopted by major domestic optometry clinical guidelines. The interval boundary can be adjusted according to the average arm length and height of children of different ages. The adjustment value is set by the fitting personnel according to the actual situation of the child. For example, for children under 6 years old, whose arm length is relatively short and whose reading and writing distance is usually 0.25 meters to 0.35 meters, the upper limit of the reading and writing scenario can be adjusted to 0.35 meters. For adolescents over 12 years old, whose arm length is relatively long and whose reading and writing distance is usually 0.30 meters to 0.45 meters, the upper limit of the reading and writing scenario can be adjusted to 0.45 meters. The adjusted interval boundary is written into the calibration file of the binocular image acquisition device and locked by eye to ensure that the boundary is consistent within the same child's measurement cycle.

[0047] It should be noted that the scene interval matching, eye-specific spatial clustering, coordinate offset variance inverse ratio allocation, and coordinate weighted summation used in S14 are all general numerical processing methods that can be directly implemented in this field. When performing half-stability verification on the target pupil position, the scene pupil coordinate set under the same eye usage scene label and eye type is divided into a first half and a second half according to the corresponding timestamp. When the number of samples is odd, the middle record is assigned to the second half, and the same clustering and weighting process is repeated. The verification passes when the two-dimensional Euclidean distance between the two half target pupil positions is not greater than the clustering distance threshold; if it exceeds the clustering distance threshold, the eye usage scene label and eye type are marked as unstable and re-collected once. If it still fails after re-collection, the target pupil position is not output and the reason for failure is saved.

[0048] In step S15, the defocus misalignment is quantified based on the target pupil position and the preset lens center under different eye usage scenarios to obtain the defocus misalignment amount.

[0049] Specifically, the preset lens center is read from the lens product design document of the children's myopia control lens and represented according to the frame coordinate system specified in the lens product design document. When the coordinate unit in the lens product design document is not millimeters, it is first converted to millimeters according to the unit conversion relationship given in the document, and is not directly subtracted from the pixel coordinates. The target pupil position under different eye use scenarios is converted from the pixel coordinates of the facial video stream to the frame coordinate system and then compared with the lens center. The planar homography calibration points are taken from the four frame frame reference marks on the upper nasal side, lower nasal side, upper temporal side and lower temporal side that have defined frame coordinates in the frame product design document; when the frame does not have inherent reference marks, removable high-contrast circular calibration stickers are attached at the corresponding four points, and the frame coordinates of the center of the circle are written into the calibration file of the binocular image acquisition device for this measurement.

[0050] During runtime, the horizontal and vertical coordinates of the target pupil position and the homogeneous component 1 are input into the planar homography transformation, and then the first two components are normalized using the third output component. Each set of calibration points simultaneously records the pixel center of the same reference marker in the facial video stream and the known center in the frame coordinate system. Four sets of non-collinear calibration points form eight independent scalar equations, which are solved using the proportional normalization constraint of the transformation matrix. When there are more than four sets of calibration points, least squares is used to minimize the sum of squared coordinate residuals before and after the transformation of all calibration points. If any three calibration points are approximately collinear, the matrix rank is insufficient, or the target pupil position is outside the convex hull of the calibration points, the coordinate transformation record is marked as invalid and recalibrated; planar extrapolation is not performed.

[0051] It should be noted that the requirement that the four sets of calibration points are not collinear and that any three points are not approximately collinear is to ensure the numerical stability of the homography matrix solution; the calibration points should be distributed in four regions: the upper nasal side, the lower nasal side, the upper temporal side, and the lower temporal side, covering the four quadrants of the frame; the pixel coordinates of the target pupil position in the facial video stream, after homography transformation, must fall within the convex hull enclosed by the four calibration points. If the transformation result is outside the convex hull range, it is unreliable and the facial video stream should be re-acquired.

[0052] For each visual scene label and eye type, the horizontal misalignment component is obtained by subtracting the horizontal coordinate of the lens center from the horizontal coordinate of the target pupil position in the frame coordinate system; the vertical misalignment component is obtained by subtracting the vertical coordinate of the lens center from the vertical coordinate of the target pupil position. The defocus misalignment amount is formed by relating the visual scene label, eye type, target pupil position, lens center, horizontal misalignment component, vertical misalignment component, and radial misalignment value, with all length fields using millimeters.

[0053] Before calculating the defocus misalignment, the target pupil position and the lens center are compared with the effective optical area specified in the lens product design document. The effective optical area is a circular region on the lens surface centered on the lens center, with its diameter specified in the lens product design document. The diameter of the effective optical area varies for different lens models, typically ranging from 8 mm to 12 mm. This area corresponds to the central photopic zone in the defocused lens used for clear imaging, and the pupil center should be located within this area to ensure defocus control. When the radial misalignment value is greater than the radius of the effective optical area, it is considered an abnormal coordinate transformation or eye-specific matching, and the lens product model, eye-specific matching, and planar homography transformation are rechecked. The radial misalignment value is the square root of the sum of the squares of the horizontal and vertical misalignment components. The defocus misalignment value simultaneously stores two misalignment components, the radial misalignment value, and their corresponding visual scene labels and eye-specific matching.

[0054] In step S16, optical center calibration distance is performed based on the defocus misalignment amount, the pupil coordinate set, and the lens center to obtain the target optical center distance, including: The visual axis offset distance is obtained by calculating the defocus misalignment amount and the pupil coordinate set. Extract the corneal vertex coordinates from the eye image sequence corresponding to the pupil coordinate set; The center of the lens is calibrated by wearing posture constraint based on the visual axis eccentricity distance and the corneal vertex coordinates to obtain the corrected lens center. The spatial Euclidean distance between the center of the corrective lens and the coordinates of the corneal vertex is calculated to obtain the target optical center distance.

[0055] Specifically, for each effective defocus misalignment, the visual scene label, eye type, and lens center are read as associated fields. Then, the pupil position coordinates that match the timestamp corresponding to the visual scene label and have the same eye type are extracted from the pupil coordinate set. The extracted pupil position coordinates are transformed to the frame coordinate system according to the homography transformation in S15. If the number of samples is less than the minimum number of effective samples specified in S14, the visual axis eccentricity distance under that visual scene and eye type is not calculated. To ensure that each effective visual scene and eye type combination participates equally and to avoid the difference in the number of sampling frames changing its contribution, the visual axis eccentricity distance is calculated using the following formula defined in this scheme.

[0056] ; In the formula, Indicates the distance of the line of sight eccentricity. and Let the horizontal and vertical coordinates of the i-th pupil position be represented, respectively, in the s-th effective eye use scenario and eye type combination. and These represent the horizontal and vertical coordinates of the center of the preset lens for the same eye type, respectively. This represents the number of valid samples for the combination, where M represents the number of valid eye-use scene and eye-specific combinations. All coordinates are in millimeters. The calculation order is as follows: calculate the square of the distance relative to the lens center for each sample, calculate the average of the squares within the combination and take the square root, and then perform an equal-weighted average over all valid combinations. The equal-weighted average ensures that each eye-use scene and eye-specific combination contributes consistently to the visual axis eccentricity distance, preventing a particular scene from occupying a higher proportion due to a larger number of sampling frames. When M is 0, visual axis eccentricity distance is not generated, and the current optical center calibration distance measurement is stopped.

[0057] Image frames with a timestamp difference of no more than one sampling period corresponding to the effective defocus misalignment amount are read from the eye image sequence corresponding to the pupil coordinate set. Near-infrared illumination units near the optical axis of the binocular image acquisition device are used to form corneal reflective points. When there is more than one candidate corneal reflective region, the connected region whose gray-scale weighted center is closest to the geometric center of the pupil contour is selected, and the gray-scale weighted center is calculated using the pixel gray-scale values ​​within the region as weights. During device calibration, the 95th percentile of the distance between the manually labeled corneal reflective point and the gray-scale weighted center is used to determine the corneal reflective point positioning error and written into the binocular image acquisition device calibration file. Based on the epipolar geometric constraints of the two acquisition channels, the gray-scale weighted centers of corneal reflective points of the same eye and corresponding timestamp are matched. Matches are retained when the distance from the matching point to the corresponding epipolar line is no greater than the corneal reflective point positioning error. The median is calculated for each of the three coordinate components of all effective three-dimensional corneal vertices, and three-dimensional corneal vertices whose coordinate components deviate from the median by more than three times the absolute median deviation are deleted. The horizontal coordinate of the gray-weighted center of the corneal reflective area is obtained by summing the products of the gray-value of each pixel in the area and its horizontal coordinate, and then dividing by the sum of the gray-values ​​of all pixels. The vertical coordinate is calculated using the same steps. The summation range is all pixels within the selected corneal reflective area.

[0058] The corneal reflective point positioning error is used to determine whether the corneal reflective points matched by the left and right acquisition channels correspond to the same physical point on the cornea. During matching, the distance from the corneal reflective point in the left eye image to the epipolar equation of the right eye image is calculated. If the distance is greater than the corneal reflective point positioning error, the matching is considered incorrect, the corneal reflective point is discarded, and it is not included in the three-dimensional corneal vertex calculation.

[0059] The frame outline is extracted from an eye image sequence where the difference between the measured time of the corneal vertex coordinates and the actual time is no more than one sampling period. At least four non-overlapping frame frame plane fitting feature points are selected along the perimeter of the frame outline at equal arc length intervals, covering the upper nasal side, lower nasal side, upper temporal side, and lower temporal side regions. If the area of ​​the triangle formed by any three frame frame plane fitting feature points is less than the square of the device spatial measurement error, the feature points are considered approximately collinear, and calibration is stopped. The frame wearing plane is calculated by minimizing the sum of the squared signed distances from all valid feature points to the candidate plane, and the plane normal vector is normalized to a unit vector. The corneal vertex coordinates are projected onto the frame wearing plane along the normal direction, and the planar distance between the lens center and the projected corneal vertex coordinates is calculated. The direction from the lens center to the projected corneal vertex coordinates is normalized to a unit direction.

[0060] The visual axis eccentricity distance is not a trigger threshold for binary classification of frame-by-frame events, but rather a target residual eccentricity formed by weighting each effective eye-use scenario with the eye-specific combination. When the current plane distance is greater than the visual axis eccentricity distance, only the difference between the two is shifted to make the calibrated plane distance equal to the visual axis eccentricity distance. When the current plane distance is not greater than the visual axis eccentricity distance, the lens center remains unchanged to avoid amplifying existing small eccentricities. The plane correction amount is the larger of the difference between the current plane distance and the visual axis eccentricity distance and 0. The lens center is then shifted along the aforementioned unit direction by this plane correction amount to obtain the corrected lens center. A dual check is performed on the corrected lens center, checking both the effective optical area of ​​the lens and the mountable area of ​​the frame. If it is located at the boundary of either area, it is processed as valid. If it is located within both areas, it is transformed to the binocular acquisition coordinate system using the two unit basis vectors and the unit normal vector of the frame wearing plane. If it exceeds either area, calibration is stopped and the cause of failure is saved.

[0061] The target optical center distance is the spatial Euclidean distance between the center of the corrective lens and the corneal vertex coordinates of the same eye in the frame-wearing coordinate system. First, the differences between the three coordinate components are calculated separately. Then, the square root of the sum of the squares of the three differences is taken, and the resulting length unit is millimeters. The target optical center distance is calculated separately for the left and right eyes. Each record includes the eye type, target optical center distance, center of the corrective lens, corneal vertex coordinates, number of valid samples, and measurement time. Based on the effective target optical center distance, the lens positioning device is used for optical center positioning and assembly. The coordinate origin, horizontal axis direction, and length unit of the positioning device are aligned with the frame coordinate system before assembly. The planar Euclidean distance between the actual lens center fed back by the lens positioning device and the center of the corrective lens is determined as the assembly difference.

[0062] For assembly tolerance, the optical center eccentricity tolerance corresponding to the current lens model and frame model combination in the lens product design document is read first. If the lens product design document does not specify this, the horizontal and vertical allowable deviations of the optical center specified in the applicable technical standard for the product are read, and the smaller value is taken and converted to millimeters. Assembly is stopped if neither the design document nor the applicable technical standard provides this information. Lens fixing is completed when the assembly difference is not greater than the assembly tolerance. If it exceeds the assembly tolerance, repositioning is performed once. If the deviation still exceeds the tolerance after repositioning, assembly is stopped. Before completing the positioning and assembly, optical center calibration distance is performed twice consecutively on the same child. The frame model, lens product model, and binocular image acquisition device calibration document remain unchanged between the two acquisitions. If the difference between the two target optical center distances is greater than the assembly tolerance, the measurement is marked as unstable and re-acquired. If the difference is not greater than the assembly tolerance, the average of the two results is taken as the final target optical center distance. The final target optical center distance, eye type, two measurements, assembly difference, assembly tolerance, and the name and version of the source document are all written into the assembly record.

[0063] For example, the assembly tolerance example values ​​are set with reference to GB 13511.1 "Eyeglasses - Part 1: Single Vision and Multifocal" and the spherical power-to-tolerance correspondence principle commonly used in the eyeglass manufacturing industry. When the absolute value of the spherical power is less than or equal to 4.00D, the horizontal allowable deviation of the optical center is ±2.0 mm and the vertical allowable deviation is ±1.0 mm; when the absolute value of the spherical power is greater than 4.00D, the horizontal allowable deviation is ±1.0 mm and the vertical allowable deviation is ±0.5 mm. When both the design document and the technical standard are recorded, the design document shall prevail. D is diopter, the international unit for measuring the optical power of a lens.

[0064] The second embodiment of the present invention provides an optical center distance measurement system for myopia control lenses for children, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0065] It should be noted that the optical center distance measurement system for children's myopia prevention lenses provided in this embodiment of the invention is used to execute all the process steps of the optical center distance measurement method for children's myopia prevention lenses described in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0066] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0067] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for measuring the optical center distance of lenses used for myopia control in children, characterized in that, include: Acquire a facial video stream under varying viewing distances, and extract temporal images of the eyes from the facial video stream to obtain an eye image sequence and corresponding timestamps; Based on the eye image sequence and the corresponding timestamp, the pupil temporal trajectory is extracted to obtain a set of pupil coordinates and a sequence of pupil offset vectors. The nasal displacement state is determined based on the pupil offset vector sequence to obtain the nasal offset state identifier. The visual field spatial convergence is performed based on the nasal offset state identifier to obtain the convergence coordinate point. The focal length relationship depth mapping is performed based on the convergence coordinate point to obtain the visual distance condition data. Based on the viewing distance condition data, viewing distance scene label matching is performed to obtain eye use scene labels. Then, based on the eye use scene labels, the pupil coordinate set is subjected to scene position weighted clustering to obtain the target pupil position under different eye use scenarios. Based on the target pupil position under different eye use scenarios and the preset lens center, the defocus misalignment is quantified to obtain the defocus misalignment amount; The optical center distance of the target is obtained by optical center calibration and distance measurement based on the defocus misalignment, the pupil coordinate set, and the lens center.

2. The method for measuring the optical center distance of lenses for myopia control in children according to claim 1, characterized in that, The step of determining the nasal displacement state based on the pupil offset vector sequence to obtain the nasal displacement state identifier includes: The motion vectors in the pupil offset vector sequence are orthogonally decomposed to obtain two-dimensional Cartesian coordinate components; The horizontal axis displacement component is extracted from the two-dimensional Cartesian coordinate components to obtain the lateral displacement value; The facial midline is located based on the facial video stream to obtain the position of the facial midline. The direction of horizontal displacement is determined based on the value of the lateral displacement and the position of the vertical line of the face. The nasal side movement state is filtered based on the lateral displacement value and the horizontal displacement direction to obtain the nasal side offset state identifier.

3. The method for measuring the optical center distance of lenses for myopia control in children according to claim 2, characterized in that, The step of filtering the nasal side movement state based on the lateral displacement value and the horizontal displacement direction to obtain the nasal side offset state identifier includes: The lateral displacement value is compared with a preset movement determination threshold, and lateral displacement values ​​that are greater than the movement determination threshold are filtered out to obtain the lateral displacement value exceeding the threshold. From the above threshold lateral displacement values, the lateral displacement values ​​of the corresponding horizontal displacement directions toward the vertical line of the face are selected according to the left eye and the right eye respectively, to obtain the left eye nasal side movement displacement value and the right eye nasal side movement displacement value. Based on the corresponding timestamp, the left eye nasal side movement displacement value and the right eye nasal side movement displacement value are synchronously paired to obtain a pair of bilateral nasal side movement data; For each pair of bilateral nasal side movement data, the corresponding timestamp, the displacement value of the left eye nasal side movement, and the displacement value of the right eye nasal side movement are written into a status data item, and the status field of the status data item is set to nasal side convergence status to obtain the nasal side offset status identifier.

4. The method for measuring the optical center distance of lenses for myopia control in children according to claim 1, characterized in that, The step of converging binocular gaze spaces based on the nasal offset status indicator to obtain the convergence coordinate point includes: When the nasal offset status indicator indicates that nasal movement has occurred, external parameters are obtained from the binocular image acquisition device to obtain the rotation vector and translation vector; Construct a three-dimensional spatial pose matrix based on the rotation vector and the translation vector; Based on the three-dimensional spatial pose matrix, the pupil coordinates of both eyes in the pupil coordinate set are spatially transformed to obtain the gaze direction of both eyes; The convergence coordinates are obtained by triangulating the direction of the binocular lines of sight based on the epipolar geometric constraints.

5. The method for measuring the optical center distance of lenses for myopia control in children according to claim 1, characterized in that, The step of performing focal length relationship depth mapping based on the convergence coordinate points to obtain view distance condition data includes: The convergence coordinates are input into a pre-established focal length mapping relationship for coordinate distance conversion to obtain the spatial depth value; The spatial depth value is associated with the nasal side offset status identifier to obtain the nasal side movement distance record; The nasal-side movement distance record is structured and encapsulated to obtain the distance condition data.

6. The method for measuring the optical center distance of lenses for myopia control in children according to claim 1, characterized in that, The step of performing visual distance scene label matching based on the visual distance condition data to obtain eye use scene labels includes: The spatial depth value is extracted from the line-of-sight condition data to obtain the line-of-sight value; The viewing distance value is matched with the viewing distance range in the preset scene mapping table to obtain the matching scene range; Read the scene labels corresponding to the matching scene interval to obtain the eye use scene labels.

7. The method for measuring the optical center distance of lenses for myopia control in children according to claim 1, characterized in that, The step of performing scene-position weighted clustering on the pupil coordinate set based on the eye usage scene label to obtain the target pupil position under different eye usage scenes includes: Based on the eye-use scene label, the pupil position coordinates are filtered from the pupil coordinate set to obtain the scene pupil coordinate set; Spatial clustering is performed on the pupil coordinate set of the scene to obtain the cluster centers; Calculate the variance of the pupil position coordinates relative to the center of the cluster within each cluster; The scene association weights are obtained by inversely allocating scene weights based on the coordinate offset variance. The target pupil position under different eye use scenarios is obtained by performing a weighted summation of coordinates based on the cluster center and the scene association weight.

8. The method for measuring the optical center distance of lenses for myopia control in children according to claim 1, characterized in that, The step of performing optical center calibration and ranging based on the defocus misalignment, the pupil coordinate set, and the lens center to obtain the target optical center distance includes: The visual axis offset distance is obtained by calculating the defocus misalignment amount and the pupil coordinate set. Extract the corneal vertex coordinates from the eye image sequence corresponding to the pupil coordinate set; The center of the lens is calibrated by wearing posture constraint based on the visual axis eccentricity distance and the corneal vertex coordinates to obtain the corrected lens center. The spatial Euclidean distance between the center of the corrective lens and the coordinates of the corneal vertex is calculated to obtain the target optical center distance.

9. An optical center distance measurement system for myopia control lenses for children, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1 to 8.

Images