Eyelid detection system for assessing orbicularis oculi muscle motor function in patients with myasthenia gravis

Abstract

This invention relates to the field of image edge detection technology, specifically to an eyelid detection system for assessing the orbicularis oculi muscle movement function in patients with myasthenia gravis. The system includes a video frame acquisition module, an iris morphology reconstruction module, an interference level acquisition module, a gradient map correction module, and an eyelid edge detection module. By acquiring video frame data of the patient's eyes, the system corrects iris deformation in the iris region of the target image frame corresponding to eye movement. Based on the corrected iris region, the system analyzes the interference of iris morphology on eyelid edge detection, determines the degree of iris deformation interference, and then corrects the gradient values ​​of gradient points in the same area corresponding to the corrected iris region in the gradient map of the target image frame during eye movement, obtaining a corrected gradient map. Based on the corrected gradient map, the system determines the patient's eyelid edge detection result. This invention effectively improves the accuracy of eyelid edge detection for patients.

Description

Technical Field

[0001] This invention relates to the field of image edge detection technology, and more specifically to an eyelid detection system for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis. Background Technology

[0002] Myasthenia gravis (MG) is an acquired autoimmune disease characterized by impaired neuromuscular junction transmission. Ocular myasthenia gravis (OMG) patients often present with dysfunction of the orbicularis oculi muscle, leading to symptoms such as incomplete eyelid closure and ptosis, severely affecting eye health and visual function. With the development of computer vision technology, video image analysis has provided new insights into the quantification of motor function. By capturing dynamic eye behaviors (such as blinking frequency and eyelid opening and closing trajectories), objective assessment of orbicularis oculi muscle function can be achieved.

[0003] Because the movement of the orbicularis oculi muscle involves the coordination of multiple muscles, such as the orbicularis oculi muscle contracting to close the eyelid and the levator palpebrae superioris muscle controlling the opening, traditional image algorithms have difficulty accurately segmenting and tracking high-speed, subtle movements. In particular, the eyelid edge and iris texture will become pixel-level adhered under high-speed movement (>30 frames / second). Traditional edge detection is not accurate enough because the closure trajectory is broken due to interference from iris texture, resulting in jagged pseudo-edges. Summary of the Invention

[0004] To address the aforementioned technical problem of inaccurate eyelid edge detection in traditional methods for assessing orbicularis oculi muscle movement due to interference from iris texture, this invention aims to provide an eyelid detection system for assessing orbicularis oculi muscle movement function in patients with myasthenia gravis. The specific technical solution adopted is as follows:

[0005] In a first aspect, the present invention provides an eyelid detection system for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis, the system comprising:

[0006] The video frame acquisition module is used to acquire video frame data of the patient's eye, the video frame data including several blink cycles;

[0007] The iris morphology reconstruction module is used to correct the iris deformation in the target image frame corresponding to the eye movement in the video frame data to obtain the corrected iris region.

[0008] The interference level acquisition module is used to analyze the interference of iris morphology on eyelid edge detection based on the texture features of the corrected iris region, and to determine the degree of interference from iris deformation.

[0009] The gradient map correction module is used to acquire the gradient map of the target image frame corresponding to eye movement, and based on the degree of iris deformation interference, correct the gradient values ​​of gradient points in the same position area corresponding to the corrected iris region in the gradient map to obtain the corrected gradient map.

[0010] The eyelid edge detection module is used to determine the patient's eyelid edge detection results based on the corrected gradient map.

[0011] In conjunction with the first aspect above, in some possible implementations, the iris morphology reconstruction module includes:

[0012] The motion parameter acquisition unit is used to match the target image frame corresponding to eye movement in the video frame data with the scleral vascular feature area corresponding to the iris region in the previous adjacent image frame, and determine the motion parameters of eye movement in the target image frame corresponding to eye movement based on the matching result.

[0013] The grid division unit is used to divide the iris region in the target image frame corresponding to eye movement into several grids.

[0014] An initial mesh correction unit is used to apply elastic mechanical constraints to the plurality of meshes and correct the meshes based on the motion parameters of the eyeball rotation to obtain an initially corrected mesh.

[0015] The final mesh correction unit is used to construct an energy function with the goal of minimizing mesh deformation error, and to iteratively correct the initially corrected mesh based on the energy function to obtain the final corrected mesh;

[0016] An iris region correction unit is used to obtain the corrected iris region based on the final corrected grid.

[0017] In conjunction with the first aspect above, in some possible implementations, the step of matching the target image frame corresponding to eye movement in the video frame data with the scleral vascular feature area corresponding to the iris region in its previous adjacent image frame, and determining the motion parameters of eye rotation in the target image frame corresponding to eye movement based on the matching result, includes:

[0018] During eye movement, the bifurcation points on the vascular topology of the target image frame and its previous adjacent image frame are extracted as feature points from the scleral vascular feature area corresponding to the iris region.

[0019] Generate SIFT descriptors for the feature points of the iris region in the target image frame and its previous adjacent image frame corresponding to eye movement;

[0020] Based on the SIFT descriptor, the feature points of the iris region in the target image frame corresponding to eye movement and its previous adjacent image frame are matched to obtain several matching points;

[0021] Based on the matching points, an affine transformation equation system is established. The affine transformation equation system is solved to obtain the motion parameters of eye movement in the target image frame when the eye moves. The motion parameters include at least translation, rotation angle and scaling ratio.

[0022] In conjunction with the first aspect above, in some possible implementations, applying elastic mechanical constraints to the plurality of meshes includes:

[0023] Based on the distance from the centroid of the grid to the centroid of the pupil, the grids are divided into near-pupil grids and outer edge grids.

[0024] The ratio of the centroid movement distance of the grid to the radius of the iris region is determined as the deformation of the grid;

[0025] The deformation of the near-pupil mesh is set to be less than a first elastic force constraint threshold, and the deformation of the outer edge mesh is set to be less than a second elastic force constraint threshold, wherein the first elastic force constraint threshold is less than the second elastic force constraint threshold.

[0026] In conjunction with the first aspect above, in some possible implementations, the construction of the energy function with the objective of minimizing mesh deformation error includes:

[0027] Based on the difference between the centroid spacing of two adjacent grids after deformation and the centroid spacing of two adjacent grids before deformation, the spacing error between two adjacent grids is determined.

[0028] The spacing error between all two adjacent grids is summed to obtain the spacing error term;

[0029] The difference between the actual deformation of all meshes and their corresponding elastic force constraint thresholds is accumulated to obtain the biological constraint error term;

[0030] The energy function is obtained by weighted summation of the spacing error term and the biological constraint error term.

[0031] In conjunction with the first aspect above, in some possible implementations, the iterative correction of the initially corrected mesh based on the energy function to obtain the final corrected mesh includes:

[0032] In each iteration of the correction process, the partial derivative of the energy function with respect to the coordinate position of each grid is calculated to obtain the gradient of each grid, and the position of each grid is adjusted along the negative direction of its gradient according to a set step size to obtain the corrected grid.

[0033] The difference between the energy function value corresponding to the previous iteration correction process and the energy function value corresponding to the current iteration correction process is determined as the energy difference. When the energy difference is less than a set energy threshold, the next iteration correction process is stopped, and each corrected grid obtained in the current iteration correction process is taken as the final corrected grid.

[0034] In conjunction with the first aspect above, in some possible implementations, the interference level acquisition module includes:

[0035] The texture feature recognition unit is used to extract the texture features of the iris region in the corrected iris region and the iris region in the reference image frame corresponding to the patient's maximum eye opening in the video frame data;

[0036] The feature difference acquisition unit is used to determine the feature difference value between the texture features of the corrected iris region and the texture features of the iris region in the reference image frame;

[0037] The velocity suppression coefficient determination unit is used to determine the velocity suppression coefficient based on the eye movement velocity during eye movement;

[0038] The deformation interference acquisition unit is used to determine the degree of iris deformation interference based on the feature difference value and the velocity suppression coefficient.

[0039] In conjunction with the first aspect above, in some possible implementations, extracting the texture features of the iris region in the reference image frame corresponding to the corrected iris region when the patient's eyes are maximally open in the video frame data includes:

[0040] The corrected iris region and the iris region in the reference image frame are each uniformly divided into several fan-shaped regions.

[0041] In the fan-shaped region, the radial gray-level gradient distribution is statistically analyzed to generate a gradient histogram, and the gradient histograms of all fan-shaped regions are used as the texture features of the corresponding iris regions.

[0042] In conjunction with the first aspect above, in some possible implementations, the gradient map correction module includes:

[0043] The gradient map acquisition unit is used to acquire the gradient map of the target image frame corresponding to eye movement in the video frame data;

[0044] The gradient adjustment parameter acquisition unit is used to determine the gradient adjustment parameter of each gradient point in the same location area based on the gradient density and edge degree of each gradient point in the same location area corresponding to the corrected iris region in the gradient map.

[0045] The gradient adjustment unit is used to correct the gradient value of each gradient point in the same location area of ​​the gradient map based on the gradient adjustment parameters and the degree of iris deformation interference, so as to obtain the corrected gradient map.

[0046] In conjunction with the first aspect above, in some possible implementations, the eyelid edge detection module includes:

[0047] An eyelid edge recognition unit is used to obtain the initial eyelid edge based on the corrected gradient map;

[0048] An abnormal edge point recognition unit is used to identify abnormal edge points in the initial eyelid edge;

[0049] The edge point correction unit corrects the position of the abnormal edge point based on the position distribution of the effective edge points in the neighborhood of the abnormal edge point if the number of effective edge points in the neighborhood of the abnormal edge point is greater than a set threshold. Otherwise, it corrects the position of the abnormal edge point based on the distribution of the corresponding position points of the abnormal edge point in the preceding image frames in the video frame data, thereby obtaining the corrected eyelid edge.

[0050] Secondly, the present invention also provides an eyelid detection method for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis, the method comprising the following steps:

[0051] Acquire video frame data of the patient's eyes, the video frame data including several blink cycles;

[0052] Iris deformation correction is performed on the iris region in the target image frame corresponding to eye movement in the video frame data to obtain the corrected iris region;

[0053] Based on the texture features of the corrected iris region, the interference of iris morphology on eyelid edge detection is analyzed to determine the degree of interference from iris deformation.

[0054] The gradient map of the target image frame corresponding to eye movement is obtained, and the gradient values ​​of the gradient points in the same position area corresponding to the corrected iris region in the gradient map are corrected based on the degree of iris deformation interference, so as to obtain the corrected gradient map.

[0055] Based on the corrected gradient map, the patient's eyelid edge detection results are determined.

[0056] Thirdly, the present invention also provides an eyelid detection device for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis, comprising a memory and a processor. The memory stores executable computer program code, and the processor retrieves and runs the executable computer program code from the memory, causing the device to perform the steps implemented by the modules in the first aspect or any possible implementation thereof.

[0057] Fourthly, the present invention also provides a computer program product comprising: computer program code, which, when run on a computer, causes the computer to perform the method described in the first aspect or any possible implementation thereof.

[0058] Fifthly, the present invention also provides a computer-readable storage medium storing computer program code that, when executed on a computer, causes the computer to perform the steps implemented by the modules in the first aspect or any possible implementation thereof.

[0059] This invention offers the following advantages: It corrects iris deformation in the target image frame corresponding to eye movements in patient video frame data, obtaining a corrected iris region to eliminate interference from iris deformation caused by eye movements. Then, based on the texture features of the corrected iris region, it analyzes the interference of iris morphology on eyelid edge detection, determines the degree of iris deformation interference, and corrects the gradient values ​​of gradient points in the same area corresponding to the corrected iris region in the gradient map of the target image frame during eye movements, obtaining a corrected gradient map. This adaptively suppresses the gradient interference of iris texture on eyelid edges during blinking. Finally, based on this corrected gradient map, it determines the accurate eyelid edge detection result for the patient. This invention overcomes the edge breakage and artifact problems of traditional algorithms in patient eye movement scenarios, improves the accuracy of patient eyelid edge detection, provides an accurate data foundation for the quantitative assessment of orbicularis oculi muscle function, and thus improves the accuracy of the quantitative assessment of orbicularis oculi muscle function. Attached Figure Description

[0060] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1This is a schematic diagram of the structure of an eyelid detection system for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis, according to an embodiment of the present invention.

[0062] Figure 2 This is a flowchart of the steps of an eyelid detection method for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis, according to an embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram illustrating the division of the iris region into a honeycomb hexagonal grid according to an embodiment of the present invention;

[0064] Figure 4 This is a schematic diagram of the structure of an eyelid detection device for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis, according to an embodiment of the present invention. Detailed Implementation

[0065] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings.

[0066] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0067] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0068] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0069] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0070] Although operations or steps are described in a specific order in the accompanying drawings in the embodiments of the present invention, this should not be construed as requiring these operations or steps to be performed in the specific order or serial order shown, or requiring all of the shown operations or steps to be performed to obtain the desired result. In the embodiments of the present invention, these operations or steps may be performed serially; they may be performed in parallel; or a portion of these operations or steps may be performed.

[0071] Furthermore, it is understood that the data involved in the technical solutions of this invention (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains, and all parameters or indicators in the formulas involved in this invention are normalized values ​​that have eliminated the influence of dimensions.

[0072] To address the problem of inaccurate eyelid edge detection in existing technologies during orbicularis oculi muscle movement assessment due to iris texture interference, this invention provides an eyelid detection system for assessing orbicularis oculi muscle movement function in patients with myasthenia gravis. This system is essentially a software system composed of modules that implement corresponding functions, as illustrated in the structural diagram below. Figure 1 As shown. The core of this system is to implement an eyelid detection method for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis. Each module in the system corresponds to a step in the method, and the corresponding flowchart is shown below. Figure 2 As shown in the diagram. The following section provides a detailed description of each module of the system, following the specific steps of this method.

[0073] The video frame acquisition module 100 is used to acquire video frame data of the patient's eyes, the video frame data including several blink cycles.

[0074] In this embodiment of the invention, to acquire video frame data of the eyes of a patient with myasthenia gravis, the patient's head is fixed to a support, and a uniform LED light source (color temperature 5500K) is used as ambient lighting to avoid shadows and reflections. The video resolution is set to 1920×1080 pixels, covering the iris and eyelid areas, and a high frame rate camera is used to continuously record for 5 minutes, including multiple blink cycles, capturing the dynamic behavior of the patient's eyes, including natural blinks and involuntary eyelid opening and closing movements (such as rapid eye closing and opening), thereby obtaining the raw video frame data of the patient's eyes.

[0075] After preprocessing the acquired raw video frame data of the patient's eyes, including continuous frame sampling, grayscale conversion, histogram equalization, and region cropping, the final video frame data of the patient's eyes is obtained. It should be noted that the assessment of the patient's orbicularis oculi muscle movement in this embodiment of the invention does not require real-time processing, but rather involves batch data collection followed by batch processing.

[0076] During high-speed video capture (>30 frames / second), the physical distance between the eyelid edge and the iris is extremely small during rapid eyelid closure, resulting in pixel overlap in the image. The iris contains complex radial textures and pigment variations, and its local gradient features are highly similar to the gradient features of the eyelid edge. When the eyelid moves at high speed, traditional algorithms (such as Canny edge detection) easily misidentify iris textures as eyelid edge fragments, leading to the generation of jagged pseudo-edges or broken trajectories.

[0077] The iris morphology reconstruction module 200 is used to perform iris deformation correction on the iris region in the target image frame corresponding to eye movement in the video frame data, so as to obtain the corrected iris region.

[0078] In the video frame data of the patient acquired by the video frame acquisition module 100, the image frame corresponding to eye movement is selected as the target image frame. Then, for the target image frame corresponding to eye movement, deformation correction is performed on the iris region in the target image frame to eliminate iris deformation interference caused by eye movement, thereby obtaining the corrected iris region.

[0079] In this embodiment of the invention, the corneal reflective point is located in the video frame data under a near-infrared light source. Specifically, based on the grayscale of the pupil region in any image frame, the bright spot region at the center of the pupil is segmented using the Otsu thresholding method. Simultaneously, the coordinate position of the centroid of the corneal reflective point is calculated based on the outline shape of the bright spot region. Using the previous image frame as a reference, the displacement of the centroid of the corneal reflective point within that image frame is measured. When the ratio of the displacement of the centroid of the corneal reflective point to the radius of the iris region exceeds 5%, eye movement is determined to have occurred, and that image frame is taken as the target image frame corresponding to the eye movement. Thus, the target image frames corresponding to all eye movements in the patient's video frame data acquired by the video frame acquisition module 100 can be determined.

[0080] To correct iris deformation in the target image frame during eye movement and obtain the corrected iris region, the aforementioned iris morphology reconstruction module 200 further includes a motion parameter acquisition unit 201, a mesh generation unit 202, an initial mesh correction unit 203, a final mesh correction unit 204, and an iris region correction unit 205, specifically comprising:

[0081] The motion parameter acquisition unit 201 is used to match the target image frame corresponding to eye movement in the video frame data with the scleral vascular feature area corresponding to the iris region in the previous adjacent image frame, and determine the motion parameters of eye movement in the target image frame corresponding to eye movement based on the matching result.

[0082] In this embodiment of the invention, the target image frame corresponding to eye movement in the video frame data is matched with the scleral vascular feature area corresponding to the iris region in the previous adjacent image frame. Based on the matching result, the motion parameters of eye rotation in the target image frame corresponding to eye movement are determined, including:

[0083] First, during eye movement, the bifurcation points on the vascular topology of the target image frame and the scleral vascular feature area corresponding to the iris region in the previous adjacent image frame are extracted as feature points.

[0084] In this embodiment of the invention, for any target image frame corresponding to eye movement, the edge of the iris region is initially segmented using the Canny operator. Since the degree of eyelid opening and closing is not at its maximum, the edge segmentation of the iris region is not complete and accurate, but it can be used to coarsely locate the iris region.

[0085] Within a 1.2mm (approximately 60 pixels) range extending outward from the outer edge of the iris region, the scleral vascular feature area, namely the white area between the upper and lower eyelids, is delineated. After morphological processing and edge smoothing, the bifurcation points on the vascular topology can be extracted as feature points.

[0086] Secondly, SIFT descriptors are generated for the feature points of the iris region in the target image frame and its previous adjacent image frame, corresponding to eye movement.

[0087] In this embodiment of the invention, for feature points of the iris region in the target image frame and the previous adjacent image frame during eye movement, a 128-dimensional SIFT descriptor is generated for these feature points. The SIFT descriptor contains position, main direction, and scale information, and records the connection relationship of the bifurcation points (such as ternary / quadrilateral structures) as topological constraints.

[0088] Next, based on the SIFT descriptor, the feature points of the iris region in the target image frame corresponding to eye movement are matched with those in the previous adjacent image frame to obtain several matching points.

[0089] In this embodiment of the invention, vascular topology matching is performed on the target image frame corresponding to any eye movement and its preceding adjacent image frame. Specifically, by comparing the SIFT descriptors and connectivity relationships of feature points between the target image frame and its preceding adjacent image frame, pairwise matching is performed on the feature points of the target image frame and its preceding adjacent image frame. Matching is based on a cosine similarity greater than 0.7 for multidimensional features. If multiple sets of matching feature points exist in adjacent frames, the feature point with the highest cosine similarity is selected as the matching feature point. Since the specific implementation process of matching the feature points of the target image frame with its preceding adjacent image frame to obtain the matching feature point is prior art, it will not be elaborated here.

[0090] Finally, based on the matching points, an affine transformation equation system is established, and the established affine transformation equation system is solved to obtain the motion parameters of eye movement in the corresponding target image frame when the eye moves. The motion parameters include at least translation, rotation angle and scaling ratio.

[0091] In this embodiment of the invention, for any eye movement, at least three pairs of matching feature points are selected between the target image frame and its previous adjacent image frame to establish an affine transformation equation system. Because when the eye rotates, pixels in areas such as the iris and sclera in the image undergo overall displacement, rotation, or slight scaling (due to changes in viewing angle), the essence of affine transformation is to describe the mapping relationship from "original image coordinates" to "rotated image coordinates" using a set of linear equations. By solving the coefficients of the affine transformation equation system, the motion parameters of eye rotation can be calculated. These motion parameters include translation, rotation angle, scaling ratio, etc. Since the specific implementation process of establishing an affine transformation equation system through matching points and solving the equation system to obtain the motion parameters is prior art, it will not be elaborated here. It should be noted that the translation, rotation, and scaling of the eye are not solely due to eye movement (the eye does not rotate or scale in a static state), but are determined by factors such as head movement, eye movement, and changes in camera viewing angle.

[0092] The grid division unit 202 is used to divide the iris region in the target image frame corresponding to eye movement into several grids.

[0093] After obtaining motion parameters such as translation, rotation angle, and scaling ratio of eye movements, the iris deformation interference under dynamic eye movements is calculated, thereby achieving morphological correction of the iris region. Therefore, in this embodiment of the invention, the iris region in the target image frame corresponding to eye movements is divided into several equally sized honeycomb-shaped hexagonal grids. Figure 3 This diagram illustrates how the iris region in a target image frame is divided into a honeycomb-shaped hexagonal grid.

[0094] The initial mesh correction unit 203 is used to apply elastic mechanical constraints to the mesh and correct the mesh based on the motion parameters of the eyeball rotation to obtain the initially corrected mesh.

[0095] Elastic mechanical constraints are applied to each grid obtained by dividing the iris region in the target image frame during any eye movement. Under these elastic mechanical constraints, deformation correction is performed on each grid based on the motion parameters of eye rotation, thereby obtaining the deformation-corrected grid distribution.

[0096] Furthermore, applying elastic mechanical constraints to several grids includes: dividing the several grids into near-pupil grids and outer-edge grids based on the distance from the grid centroid to the pupil centroid; determining the ratio of the grid centroid movement distance to the iris region radius as the grid deformation; setting the deformation of the near-pupil grid to be less than a first elastic constraint threshold, and setting the deformation of the outer-edge grid to be less than a second elastic constraint threshold, wherein the first elastic constraint threshold is less than the second elastic constraint threshold.

[0097] In this embodiment of the invention, elastic mechanical constraints are applied to each grid based on the grid position information in the target image frame corresponding to any eye movement, that is: taking the pupil centroid in the target image frame as the origin and the radial distance to the pupil edge points in each direction, the grid centroid points ( Figure 3 Grids with the black dot in the center of the hexagonal mesh falling within 30% of the radial distance are called near-pupil grids, while grids with their centroids falling beyond 30% are called outer-edge grids. The ratio of the centroid movement distance of the grid to the radius of the iris region is used as the deformation. To constrain the rigidity of the near-pupil grid and avoid core texture distortion, the deformation of the near-pupil grid is set to be less than a first elastic force constraint threshold, which is set to 5%. Simultaneously, to allow the outer-edge iris grid to stretch and adapt to edge stretching deformation, the deformation of the outer-edge grid is set to be less than a second elastic force constraint threshold, which is set to 20%. Furthermore, each grid can deform independently, and overlap between grids after deformation is allowed.

[0098] Furthermore, under the aforementioned elastic mechanical constraints, each grid is calibrated based on the rotation parameters of eye rotation determined in the motion parameter acquisition unit 201. In this embodiment of the invention, deformation correction of each grid based on the motion parameters of eye rotation includes: first, moving the grid centroid position according to the eye translation amount, i.e., moving the grid centroid position according to the inverse vector of the translation amount; then, rotating the entire grid around the displaced centroid position according to the eye rotation angle, i.e., rotating the grid according to the inverse rotation angle; and then adjusting the grid spacing according to the scaling ratio, i.e., adjusting the grid spacing according to the inverse ratio of the scaling ratio. In other words, according to the motion parameters of eye rotation, the iris region in the target image frame corresponding to eye rotation is adjusted in the opposite direction to eliminate the deformation effect caused by eye rotation, thereby obtaining the initial grid correction result.

[0099] The final mesh correction unit 204 is used to construct an energy function with the goal of minimizing mesh deformation error, and to perform iterative correction on the initially corrected mesh based on the energy function to obtain the final corrected mesh.

[0100] It should be noted that during the mesh calibration process in the initial mesh correction unit 203, the mesh needs to be constrained by elastic forces, meaning that the deformation of each mesh cannot exceed a set deformation threshold. However, due to the existence of different elastic mechanical constraints on the meshes, deviations occur between the meshes and the actual iris texture and structure (such as excessive deformation near the pupil area and insufficient stretching at the outer edge). Therefore, it is also necessary to optimize the mesh node positions by setting an energy function, under the premise of satisfying the elastic mechanical constraints, so that the final shape conforms to both the overall motion trend and the local deformation characteristics.

[0101] Furthermore, the energy function constructed above with the goal of minimizing mesh deformation error includes: determining the spacing error between two adjacent meshes based on the difference between the centroid spacing between two adjacent meshes after deformation and the centroid spacing between two adjacent meshes before deformation; accumulating all the spacing errors between two adjacent meshes to obtain a spacing error term; accumulating the difference between the actual deformation of all meshes and their corresponding elastic force constraint thresholds to obtain a biological constraint error term; and weighted summing the spacing error term and the biological constraint error term to obtain the energy function.

[0102] In this embodiment of the invention, with the goal of minimizing mesh deformation error, an energy function is established. Specifically, the difference between the centroid spacing of all pairs of adjacent meshes after deformation and the centroid spacing of adjacent meshes before deformation is divided by the centroid spacing of adjacent meshes before deformation and normalized. This normalized value is used as the adjacent mesh node spacing error. Then, the spacing errors of all adjacent mesh nodes are accumulated to obtain the spacing error term. The sum of the differences between the actual deformation of different grids and their corresponding elastic force constraint thresholds is used as the biological constraint error term. The spacing error term and biological constraint error term The weighted sum of the two error terms serves as the energy function. At this time there is In the formula, Indicates the spacing error term The weight, Represents biological constraint error term The weights are set in the embodiments of the present invention. .

[0103] Furthermore, based on the energy function constructed above, the initially corrected mesh corresponding to the iris region in the target image frame during eye movement is iteratively corrected to further eliminate iris deformation interference caused by eye movement, thereby obtaining the final corrected mesh.

[0104] Furthermore, based on the energy function, the initially corrected mesh is iteratively corrected to obtain the final corrected mesh. This includes: in each iterative correction process, calculating the partial derivative of the energy function with respect to the coordinate position of each mesh to obtain the gradient of each mesh, and adjusting the position of each mesh along the negative direction of its gradient according to a set step size to obtain each corrected mesh; determining the difference between the value of the energy function corresponding to the previous iterative correction process and the value of the energy function corresponding to the current iterative correction process as the energy difference; when the energy difference is less than a set energy threshold, stopping the next iterative correction process, and taking each corrected mesh obtained in the current iterative correction process as the final corrected mesh.

[0105] In this embodiment of the invention, the energy function values ​​of each grid after the initial transformation are obtained to obtain the initial energy value. Then calculate the energy function. For each grid coordinate, the partial derivative (gradient) is used to adjust the grid coordinates along the negative gradient direction, with a set step size of 1 pixel. This adjustment is based on the initial energy value. The energy difference between two adjacent iterations of the correction process is calculated (the difference between the energy function value of the previous iteration and the energy function value of the next iteration). If the energy difference between two adjacent iterations is less than a set energy threshold (0.01), the iteration correction is stopped, and the final coordinates of all grid nodes are output. This yields the final corrected iris grid shape.

[0106] The iris region correction unit 205 is used to obtain the corrected iris region based on the finally corrected grid.

[0107] Based on the final corrected iris grid shape obtained in the final grid correction unit 204, the image region within the grid changes synchronously with the grid shape to obtain the reconstructed and rotated iris shape, that is, the corrected iris region.

[0108] The interference level acquisition module 300 is used to analyze the interference of iris morphology on eyelid edge detection based on the texture features of the corrected iris region, and to determine the degree of interference of iris deformation.

[0109] The texture features in the reconstructed iris region determined in the iris morphology reconstruction module 200 are identified. Based on the texture features of the corrected iris region, the interference of iris morphology on eyelid edge detection is analyzed, the degree of iris deformation interference is determined, and the eyelid edge detection process in the corresponding target image frame during eye movement is corrected based on the degree of iris deformation interference, so as to obtain accurate eyelid edge detection results.

[0110] Furthermore, the aforementioned interference level acquisition module 300 includes a texture feature recognition unit 301, a feature difference acquisition unit 302, a velocity suppression coefficient determination unit 303, and a deformation interference acquisition unit 304, specifically comprising:

[0111] The texture feature recognition unit 301 is used to extract the texture features of the iris region in the corrected iris region and the iris region in the reference image frame when the patient's eyes are wide open to the maximum extent in the video frame data.

[0112] In the texture feature recognition unit 301, for the patient's video frame data acquired by the video frame acquisition module 100, when the patient's eyes are wide open to the maximum extent, the image frame with the clearest iris is captured and used as the reference image frame. The iris region in the reference image frame is then segmented. Furthermore, the texture features of the corrected iris region and the iris region in the reference image frame are extracted.

[0113] Furthermore, the texture features of the iris region in the corrected iris region and the iris region in the reference image frame when the patient's eyes are wide open to the maximum extent in the video frame data are extracted, including: dividing the corrected iris region and the iris region in the reference image frame into several fan-shaped regions; statistically analyzing the radial gray-level gradient distribution in the fan-shaped regions to generate gradient histograms, and using the gradient histograms of all fan-shaped regions as the texture features of the corresponding iris regions.

[0114] In this embodiment of the invention, taking the extraction of texture features of the iris region in a reference image frame as an example, the iris region in the reference image frame is converted from a Cartesian coordinate system to a polar coordinate system (similar to the radial expansion of radar scanning), and divided into fan-shaped regions at 10-degree angles, forming a total of 36 angular partitions. Within each fan-shaped region obtained from the division of the iris region in the reference image frame, the gray-level gradient distribution in the radial direction (from the pupil center outward) is statistically analyzed, generating 36 angular histograms. These histograms completely record the inherent radial texture features of the iris. In addition, the centroid position of the iris region in the reference image frame is extracted, and the maximum distance between the centroid and the edge point of the iris region is taken as the radius of the iris region.

[0115] Following the same method used to extract texture features from the iris region in the baseline image frame, texture features in the corrected iris region are extracted. Thus, the texture features of the corrected iris region and the corresponding iris region in the baseline image frame when the patient's eyes are fully open in the video frame data can be obtained.

[0116] The feature difference acquisition unit 302 is used to determine the feature difference value between the texture features of the corrected iris region and the texture features of the iris region in the reference image frame.

[0117] In this embodiment of the invention, since iris texture (such as radial stripes) is a stable feature, the gradient distribution of each angular partition of the corrected iris region in the target image frame corresponding to the current eye movement is compared with the static base feature histogram. Specifically, after normalizing the vertical axis units of the texture features of the corrected iris region in the target image frame and the texture features of the iris region in the reference image frame corresponding to the 36 angular histograms, the mean square error between the two normalized histograms is calculated. The mean square error This serves as the feature difference value between two texture features. It reflects the inherent texture interference of the iris region in the target image frame corresponding to the current eye movement. The smaller the difference, the greater the possibility that the iris region is affected by the inherent texture interference of the iris, and the more necessary it is to suppress the gradient expression of the inherent texture.

[0118] The velocity suppression coefficient determination unit 303 is used to determine the velocity suppression coefficient based on the eye movement velocity during eye movement.

[0119] In this embodiment of the invention, the eye movement speed during eye movement When the ratio of the number of pixels corresponding to the magnitude of eye translation in the motion parameters of eye movement to the image acquisition time interval exceeds 10 pixels per second, velocity-related inhibition is activated, and the velocity inhibition coefficient is obtained. In the formula: e represents an exponential function with the natural constant e as the base. It should be understood that this refers to the speed of eye movement during eye movements. Set the speed suppression coefficient when the speed is no more than 10 pixels per second. .

[0120] The deformation interference acquisition unit 304 is used to determine the degree of iris deformation interference based on the feature difference value and the velocity suppression coefficient.

[0121] In this embodiment of the invention, based on feature difference values and velocity suppression coefficient Calculate the degree of interference from iris deformation .

[0122] The gradient map correction module 400 is used to acquire the gradient map of the target image frame corresponding to eye movement, and based on the degree of iris deformation interference, correct the gradient values ​​of gradient points in the same position area corresponding to the corrected iris region in the gradient map to obtain the corrected gradient map.

[0123] Based on the degree of iris deformation interference determined in the interference degree acquisition module 300, the gradient of the gradient point in the same position area corresponding to the corrected iris region in the gradient map of the target image frame during eye movement is dynamically corrected to obtain the corrected gradient map. Then, based on the corrected gradient map, the eyelid edge can be accurately detected.

[0124] Furthermore, the gradient map correction module 400 includes a gradient map acquisition unit 401, a gradient adjustment parameter acquisition unit 402, and a gradient adjustment unit 403, specifically comprising:

[0125] The gradient map acquisition unit 401 is used to acquire the gradient map of the target image frame corresponding to eye movement in the video frame data.

[0126] In this embodiment of the invention, the conventional Sobel operator is used to perform edge detection on the target image frame corresponding to eye movement, thereby obtaining the gradient map of the target image frame. Since the process of using the conventional Sobel operator to perform edge detection on the image to obtain the gradient map is prior art, it will not be described in detail here.

[0127] The gradient adjustment parameter acquisition unit 402 is used to determine the gradient adjustment parameter of each gradient point in the same location area based on the gradient density and edge degree of each gradient point in the same location area corresponding to the corrected iris region in the gradient map.

[0128] In this embodiment of the invention, in the gradient map of the target image frame corresponding to eye movement, the same position region corresponding to the corrected iris region is determined. Within this same position region, the gradient density within a circular area with a radius of 50 is searched, centered on the position of each gradient point. This involves calculating the ratio of the total number of gradient points within the circular area to the area of ​​the circular area, and using this ratio as the gradient density of each gradient point. Simultaneously, the distance from the position of each gradient point to the centroid of the corrected iris region is calculated, and the ratio of this distance to the radius of the iris region is calculated, using this ratio as the edge degree of each gradient point. Furthermore, the product of the gradient density and edge degree of each gradient point is calculated, and this product is used as the gradient adjustment parameter for each gradient point, denoted as M. Thus, the gradient adjustment parameter for each gradient point within the same position region corresponding to the corrected iris region in the gradient map can be obtained. A larger value for the gradient adjustment parameter indicates that the gradient point is closer to the iris edge, more likely to intersect with the eyelid, and that the gradient around the gradient point is dense, potentially indicating complex iris textures.

[0129] The gradient adjustment unit 403 is used to correct the gradient value of each gradient point in the same location area of ​​the gradient map based on the gradient adjustment parameters and the degree of iris deformation interference, so as to obtain the corrected gradient map.

[0130] In this embodiment of the invention, within the same location region corresponding to the corrected iris region in the gradient map of the target image frame during eye movement, the gradient value of each gradient point within the same location region is corrected based on the degree of iris deformation interference and the gradient adjustment parameters of each gradient point. In the formula, This represents the corrected gradient value at the i-th gradient point within the same location region corresponding to the corrected iris region in the gradient map. This represents the gradient value of the i-th gradient point within the same location region corresponding to the corrected iris region in the gradient map. This represents the gradient adjustment parameter for the i-th gradient point within the same location region corresponding to the corrected iris region in the gradient map. This indicates the degree of interference from iris deformation. Specifically, the larger the gradient adjustment parameter M, the stronger the corresponding... The smaller the value, the greater the reduction in gradient at that gradient point location. Additionally, it is multiplied by a negative correlation mapping value indicating the degree of iris dynamic interference. The suppressed gradient value is obtained. .

[0131] Based on the corrected gradient values ​​of all gradient points within the same region corresponding to the corrected iris region in the gradient map, as well as the gradient values ​​of gradient points at other locations, a corrected gradient map is obtained. Following this method, the gradient amplitude of gradient points in the outer edge region of the iris can be reduced proportionally according to the gradient adjustment parameters and the degree of iris deformation interference. Especially during the rapid eyelid closure phase, this can suppress dense gradient expression at the eyelid-iris junction, effectively eliminating interference from false edge signals.

[0132] The eyelid edge detection module 500 is used to determine the patient's eyelid edge detection results based on the corrected gradient map.

[0133] Based on the corrected gradient map corresponding to eye movement obtained in the gradient map correction module 400, the eyelid edge detection result can be obtained.

[0134] Furthermore, the aforementioned eyelid edge detection module 500 includes an eyelid edge recognition unit 501, an abnormal edge point recognition unit 502, and an edge point correction unit 503, specifically comprising:

[0135] The eyelid edge recognition unit 501 is used to obtain the initial eyelid edge based on the corrected gradient map.

[0136] In this embodiment of the invention, based on the corrected gradient map corresponding to eye movement obtained in the gradient map correction module 400, the Canny algorithm is used to perform edge detection on the target image frame to obtain the eyelid Canny edge detection result, and the eyelid Canny edge detection result is used as the initial eyelid edge.

[0137] An abnormal edge point recognition unit 502 is used to identify abnormal edge points in the initial eyelid edge.

[0138] In this embodiment of the invention, abnormal edge points on the initial eyelid edge determined in the target image frame during eye movement are verified. Specifically, based on continuous tracking using optical flow, the displacement deviation of each edge point on the initial eyelid edge in adjacent image frames is calculated. When the displacement deviation exceeds 20% of the iris region radius, the edge point is determined to be an abnormal transition point, i.e., an abnormal edge point. Thus, each abnormal edge point in the initial eyelid edge can be identified. When an abnormal edge point is identified, a trajectory correction procedure needs to be initiated.

[0139] The edge point correction unit 503 corrects the position of the abnormal edge point based on the position distribution of the effective edge points in the neighborhood of the abnormal edge point if the number of effective edge points in the neighborhood of the abnormal edge point is greater than a set threshold. Otherwise, it corrects the position of the abnormal edge point based on the distribution of the corresponding position points of the abnormal edge point in the preceding image frames in the video frame data, thereby obtaining the corrected eyelid edge.

[0140] In this embodiment of the invention, a 3×3 pixel neighborhood is defined centered on each abnormal edge point identified in the initial eyelid edge. A preset threshold of 5 is set. If there are 5 or more valid edge points within the neighborhood, local edge direction interpolation repair is preferentially adopted. That is, the least squares method is used to fit a straight line to the valid edge points within the neighborhood of the abnormal edge point to obtain a fitted straight line, which is the dominant edge direction within the neighborhood. Furthermore, based on the edge direction, the projection point of the abnormal edge point on the fitted straight line is calculated, and the projection point is used as the corrected position coordinate of the abnormal edge point.

[0141] If the number of valid edge points in the neighborhood of an abnormal edge point is less than 5, a timing verification is initiated. The trajectory of the first 10 image frames corresponding to the target image frame during eye movement is backtracked and fitted for correction. That is, the position of the corresponding pixel in the first 10 image frames corresponding to the target image frame during eye movement is obtained using optical flow method, and the edge curves of these 10 positions in their respective image frames are obtained. Then, the average of the edge curves of these 10 image frames is calculated to obtain the backtracked edge. Similarly, the projection point of the abnormal edge point on the backtracked edge is calculated, and the projection point is used as the corrected position coordinate of the abnormal edge point.

[0142] The abnormal edge points in the initial eyelid edge identified in the target image frame during eye movement are corrected in the manner described above, thus completing the correction of the eyelid edge and finally obtaining the accurate eyelid edge in the target image frame during eye movement.

[0143] Furthermore, based on the accurate eyelid edges in the target image frame corresponding to the eye movement, the orbicularis oculi muscle motor function of myasthenia gravis patients is quantitatively assessed. Specifically, the vertical distance between the lowest point of the upper eyelid edge and the highest point of the lower eyelid edge in each frame is obtained as the eyelid opening / closing degree. The difference between the maximum (open eye) and minimum (closed or semi-closed eye) values ​​of the eyelid opening / closing degree in the video frame data is used as the patient's maximum eyelid opening / closing range. The average speed (unit: pixels / second) of the eyelid from maximum to minimum opening / closing degree (eye-closing process) is calculated to obtain the eye-closing speed. Based on these characteristics, doctors are assisted in assessing the orbicularis oculi muscle motor function of myasthenia gravis patients. For example, if the patient's maximum opening / closing range is less than 70% of the average for healthy individuals of the same age, or the average eye-closing speed is slower than the normal average by 50%, it suggests weakened orbicularis oculi muscle motor function. If the absolute value of the difference in the maximum opening / closing range between the two eyes exceeds 20%, or the difference in eye-closing speed exceeds 30%, it suggests a more significant unilateral orbicularis oculi muscle dysfunction.

[0144] Based on the same inventive concept, embodiments of the present invention also provide an eyelid detection method for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis, such as... Figure 2 As shown, the method includes the following steps:

[0145] Acquire video frame data of the patient's eyes, the video frame data including several blink cycles;

[0146] Iris deformation correction is performed on the iris region in the target image frame corresponding to eye movement in the video frame data to obtain the corrected iris region;

[0147] Based on the texture features of the corrected iris region, the interference of iris morphology on eyelid edge detection is analyzed to determine the degree of interference from iris deformation.

[0148] The gradient map of the target image frame corresponding to eye movement is obtained, and the gradient values ​​of the gradient points in the same position area corresponding to the corrected iris region in the gradient map are corrected based on the degree of iris deformation interference, so as to obtain the corrected gradient map.

[0149] Based on the corrected gradient map, the patient's eyelid edge detection results are determined.

[0150] Based on the same inventive concept, embodiments of the present invention also provide an eyelid detection device for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis, such as... Figure 4As shown, the device includes: a memory 41, a processor 42, and computer program code 43 stored in the memory 41 and running on the processor 42, wherein when the processor 42 executes the computer program code 43, the system can perform the steps implemented by each module in any of the aforementioned eyelid detection systems for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis.

[0151] In this embodiment of the invention, the system can be divided into functional modules according to the above method example. For example, each module can correspond to a separate functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0152] Based on the same inventive concept, embodiments of the present invention also provide a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute the steps implemented by each module in any of the aforementioned eyelid detection systems for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis.

[0153] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing computer program code, which, when executed on a computer, causes the computer to perform the steps implemented by each module in any of the aforementioned eyelid detection systems for assessing the orbicularis oculi muscle motor function in patients with myasthenia gravis.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An eyelid detection system for assessing the motor function of the orbicularis oculi muscle in patients with myasthenia gravis, characterized in that, The system comprises: a video frame acquisition module configured to acquire video frame data of a patient's eye, the video frame data comprising a plurality of blink cycles; an iris shape reconstruction module configured to perform iris deformation correction on an iris region in a target image frame corresponding to eye movement in the video frame data, to obtain a corrected iris region; an interference degree acquisition module configured to analyze interference of iris shape on eyelid edge detection based on texture features of the corrected iris region, and determine an iris deformation interference degree; a gradient map correction module configured to acquire a gradient map of the target image frame corresponding to eye movement, and correct gradient values of gradient points in the corrected iris region in the gradient map based on the iris deformation interference degree, to obtain a corrected gradient map; an eyelid edge detection module configured to determine an eyelid edge detection result of the patient based on the corrected gradient map; the interference degree acquisition module comprises: a texture feature identification unit configured to extract texture features of the corrected iris region and an iris region in a reference image frame corresponding to maximum eye opening of the patient in the video frame data; a feature difference acquisition unit configured to determine a feature difference value between the texture features of the corrected iris region and the texture features of the iris region in the reference image frame; a velocity suppression coefficient determination unit configured to determine a velocity suppression coefficient based on an eye movement velocity during eye movement; a deformation interference acquisition unit configured to determine the iris deformation interference degree based on the feature difference value and the velocity suppression coefficient.

2. The eyelid detection system for evaluating the movement function of the orbicularis oculi muscle of a myasthenic patient according to claim 1, characterized in that, The iris shape reconstruction module comprises: a movement parameter acquisition unit configured to match a sclera blood vessel feature area corresponding to the iris region in a target image frame corresponding to eye movement in the video frame data and a previous adjacent image frame, and determine movement parameters of eye rotation in the target image frame corresponding to eye movement based on a matching result; a grid division unit configured to divide the iris region in the target image frame corresponding to eye movement into a plurality of grids; an initial grid correction unit configured to apply an elastic mechanics constraint to the plurality of grids, and correct the grids based on the movement parameters of eye rotation, to obtain initial corrected grids; a final grid correction unit configured to construct an energy function with the objective of minimizing grid deformation error, and iteratively correct the initial corrected grids based on the energy function, to obtain final corrected grids; an iris region correction unit configured to obtain the corrected iris region based on the final corrected grids.

3. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 2, characterized in that The matching of the sclera blood vessel feature area corresponding to the iris region in the target image frame corresponding to eye movement in the video frame data and the previous adjacent image frame, and the determination of the movement parameters of eye rotation in the target image frame corresponding to eye movement based on the matching result, comprise: extracting bifurcation points on a blood vessel topological structure in the sclera blood vessel feature area corresponding to the iris region in the target image frame corresponding to eye movement and the previous adjacent image frame as feature points; generating SIFT descriptors of the feature points of the iris region in the target image frame corresponding to eye movement and the previous adjacent image frame; Based on the SIFT descriptor, the feature points in the iris region of the corresponding target image frame and its previous adjacent image frame during eye movement are matched to obtain a plurality of matching points; Based on the matching points, an affine transformation equation set is established, and the affine transformation equation set is solved to obtain the motion parameters of the eye rotation in the corresponding target image frame during eye movement, the motion parameters at least including a translation amount, a rotation angle and a scaling ratio.

4. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 2, characterized in that The elastic mechanics constraint on the plurality of grids includes: Based on the distance from the grid centroid to the pupil centroid, the plurality of grids are divided into near-pupil grids and outer-edge grids; The ratio of the centroid movement distance of the grid to the radius of the iris region is determined as the deformation variable of the grid; The deformation variable of the near-pupil grid is set to be less than a first elastic force constraint threshold, and the deformation variable of the outer-edge grid is set to be less than a second elastic force constraint threshold, the first elastic force constraint threshold being less than the second elastic force constraint threshold.

5. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 4, characterized in that The energy function is constructed with the minimum grid deformation error as the target, including: Based on the difference between the distance between the centroids of the two adjacent grids after deformation and the distance between the centroids of the same two adjacent grids before deformation, the distance error between the two adjacent grids is determined; The distance errors between all the two adjacent grids are accumulated to obtain a distance error term; The difference between the actual deformation variable of all the grids and the corresponding elastic force constraint threshold is accumulated to obtain a biological constraint error term; The distance error term and the biological constraint error term are weighted and added to obtain the energy function.

6. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 2, characterized in that Based on the energy function, the initial corrected grids are iteratively corrected to obtain the final corrected grids, including: In each iteration correction process, the partial derivative of the energy function with respect to the coordinate position of each grid is calculated to obtain the gradient of each grid, and the position of each grid is adjusted along the negative direction of its gradient by a set step to obtain the corrected grids; The difference between the value of the energy function corresponding to the last iteration correction process and the value of the energy function corresponding to the current iteration correction process is determined as the energy difference, and when the energy difference is less than a set energy threshold, the next iteration correction process is stopped, and the corrected grids obtained in the current iteration correction process are taken as the final corrected grids.

7. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 6, characterized in that The texture features of the corrected iris region and the iris region in the corresponding reference image frame in the video frame data when the patient opens his eyes to the maximum extent include: The corrected iris region and the iris region in the reference image frame are uniformly divided into a plurality of sector regions, respectively; In the sector region, the radial direction gray gradient distribution is counted to generate a gradient histogram, and the gradient histograms of all sector regions are taken as the texture features of the corresponding iris region.

8. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 1, characterized in that The gradient map correction module includes: A gradient map acquisition unit is configured to acquire a gradient map of a corresponding target image frame during eye movement in the video frame data; The gradient adjustment parameter acquisition unit is configured to determine a gradient adjustment parameter of each gradient point in the same position region based on the gradient density and the edge degree of each gradient point in the same position region corresponding to the corrected iris region in the gradient map; The gradient adjustment unit is configured to correct the gradient value of each gradient point in the same position region in the gradient map based on the gradient adjustment parameter and the iris deformation interference degree, and obtain a corrected gradient map.

9. The eyelid detection system for the assessment of the orbicularis oculi muscle motor function in patients with muscle weakness according to claim 1, characterized in that The eyelid edge detection module comprises: An eyelid edge identification unit configured to obtain an initial eyelid edge based on the corrected gradient map; An abnormal edge point identification unit configured to identify an abnormal edge point in the initial eyelid edge; An edge point correction unit configured to correct the position of the abnormal edge point based on the position distribution of the effective edge points in the neighborhood range of the abnormal edge point if the number of the effective edge points in the neighborhood range of the abnormal edge point is greater than a set number threshold, or correct the position of the abnormal edge point based on the distribution of the position points corresponding to the previous several image frames of the video frame data in which the abnormal edge point is located, so as to obtain a corrected eyelid edge.

Images