Eyeball tremor identification method and system
By performing multi-level filtering and interval segmentation on pupil data, combined with rhythmic analysis, nystagmus can be identified, solving the problems of subjectivity and misjudgment in existing nystagmus identification technologies, and achieving high-precision and automated nystagmus diagnosis.
Patent Information
- Application Number
- CN202511373009.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-24
- Publication Date
- 2025-11-21
AI Technical Summary
Existing methods for identifying nystagmus rely on manual observation, which is subject to strong subjectivity, low accuracy, low efficiency, and difficulty in quantitative analysis. Furthermore, automated methods are weak in noise resistance and cannot effectively distinguish between spontaneous eye movements and pathological nystagmus.
By acquiring pupil data, performing multi-level filtering preprocessing, segmenting into interval data segments, detecting fast and slow phase modes, and combining rhythmic feature analysis, the velocity variation coefficient is used to determine nystagmus events and extract nystagmus parameters.
It achieves high-precision and highly interference-resistant automated nystagmus recognition, reduces the false positive rate, provides objective quantitative evidence for clinical diagnosis, and improves diagnostic consistency and efficiency.
Smart Images
Figure CN120982986A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of medical auxiliary diagnosis, in particular to a nystagmus recognition method and system. BACKGROUND
[0002] Nystagmus, also known as eye tremor, is a common ophthalmic disease, which is characterized by involuntary and rhythmic back-and-forth movement of the eyeball. According to the direction of the eyeball movement, it can be divided into horizontal type, vertical type, rotational type and other types, among which the horizontal type of nystagmus is the most common in clinical practice. In clinical diagnosis, the characteristic parameters of nystagmus, such as slow phase velocity and duration, play a crucial role in accurate diagnosis of diseases and scientific evaluation of treatment effect.
[0003] The existing nystagmus recognition mainly relies on manual observation by doctors, which has the following defects: first, the subjectivity is too strong. Due to the differences in experience, professional knowledge and judgment standards of different doctors, the consistency of the diagnosis results is low. Second, the accuracy is limited. The manual observation method often fails to accurately capture the tiny or short-term nystagmus phenomenon, which may lead to missed diagnosis or misdiagnosis. Third, the parameters are missing. Manual observation cannot quantitatively analyze key indicators such as slow phase velocity, which makes doctors lack accurate data support when diagnosing and evaluating the disease. Fourth, the efficiency is low. Doctors need to observe the eyeball movement of patients in real time, which not only consumes a lot of time and energy, but also easily leads to fatigue and negligence when facing a large number of patients, further reducing the quality and efficiency of diagnosis. Fifth, the accuracy verification of nystagmus recognition based on spectral angle mapping (SAM) and other segmentation techniques is limited by the integrity of the eye data.
[0004] In order to overcome the drawbacks of manual observation, some studies have begun to use automatic methods based on eye tracking to recognize nystagmus. However, these automatic nystagmus recognition methods have weak noise resistance and are easily affected by pupil recognition errors; at the same time, they cannot effectively distinguish between voluntary eye movement and pathological nystagmus, increasing the risk of misjudgment. In addition, the existing nystagmus recognition methods lack analysis of the rhythmic characteristics of nystagmus, which restricts their value in clinical application. SUMMARY
[0005] In order to solve the above technical problems, the present application provides a nystagmus recognition method, and on the other hand, provides a nystagmus recognition system.
[0006] The technical problems solved by the present application can be realized by the following technical solutions:
[0007] A nystagmus recognition method, comprising:
[0008] acquire pupil data, the pupil data comprising time series information and corresponding pupil position coordinate information;
[0009] pre-process the pupil data to obtain pre-processed pupil data;
[0010] segment the pre-processed pupil data to form a plurality of interval data segments;
[0011] detect fast phase and slow phase patterns according to speed characteristics of each interval data segment;
[0012] when it is detected that fast phase and slow phase alternating patterns appear continuously for at least a preset number of times and rhythm characteristics meet rhythm requirements, determine that an ocular tremor event occurs; wherein the rhythm characteristics are a speed variation coefficient.
[0013] Preferably, the pre-processing comprises multi-stage filtering processing, and the multi-stage filtering processing comprises:
[0014] perform median filtering processing on the pupil data to obtain first pupil data;
[0015] perform Gaussian smoothing filtering processing on the first pupil data to obtain second pupil data;
[0016] perform low-pass filtering processing on the second pupil data to obtain the pre-processed pupil data.
[0017] Preferably, the segmenting the pre-processed pupil data comprises:
[0018] calculate a speed of pupil position in a horizontal direction;
[0019] according to a direction of the speed, take a time corresponding to a change in speed direction as a segmentation point;
[0020] segment a curve drawn based on the pupil data according to the segmentation point to form a plurality of interval data segments.
[0021] Preferably, the segmenting the pre-processed pupil data further comprises:
[0022] screen the plurality of interval data segments, filter out interval data segments with a length less than a preset length, and retain interval data segments with a length not less than the preset length.
[0023] Preferably, the detecting fast phase and slow phase patterns according to speed characteristics of each interval data segment comprises:
[0024] calculate a moving speed of each interval data segment;
[0025] The interval data segment is marked according to the moving speed of the interval data segment, and the fast-slow phase mode of each interval data segment is determined; wherein, when the moving speed is greater than a preset speed threshold and the number of pixels moved by the interval data segment in a unit time exceeds a preset threshold, the corresponding interval data segment is marked as a fast phase mode; otherwise, it is marked as a slow phase mode.
[0026] Preferably, the moving speed is a ratio of a displacement difference value and a time span of the corresponding interval data segment, the displacement difference value is a Euclidean distance between two points at the beginning and the end of the corresponding interval data segment, and the time span is a difference between the end time and the start time of the corresponding interval data segment.
[0027] Preferably, the preset speed threshold is 3 times of the average speed of the moving speeds of all interval data segments.
[0028] The preset threshold is 5.
[0029] Preferably, the method further comprises:
[0030] When the fast-slow phase alternating mode appears continuously for at least a preset number of times but the rhythmicity feature does not meet the rhythmicity requirement, it is determined that an autonomous eye movement event occurs.
[0031] Preferably, the method further comprises extracting an nystagmus parameter, the nystagmus parameter comprising a nystagmus start time, a nystagmus end time, a nystagmus number, starting point information of each nystagmus slow phase, and a slow phase speed; wherein, the slow phase speed is a ratio of a displacement difference value and a time span between two points at the beginning and the end of the slow phase.
[0032] In another aspect, an eye tremor recognition system is provided for implementing the eye tremor recognition method as described above, comprising:
[0033] A data acquisition module is configured to acquire pupil data, the pupil data comprising time sequence information and corresponding pupil position coordinate information;
[0034] A preprocessing module is connected to the data acquisition module and configured to preprocess the pupil data to obtain preprocessed pupil data;
[0035] An interval segmentation module is connected to the preprocessing module and configured to segment the preprocessed pupil data to form a plurality of interval data segments;
[0036] A phase classification module is connected to the interval segmentation module and configured to detect a fast phase and a slow phase mode according to the speed feature of each interval data segment;
[0037] An nystagmus detection module is connected to the phase classification module and configured to determine that an eye tremor event occurs when the fast-slow phase alternating mode appears continuously for at least a preset number of times and the rhythmicity feature meets the rhythmicity requirement; wherein, the rhythmicity feature is a speed variation coefficient.
[0038] The advantages or beneficial effects of the technical scheme of the present application are as follows:
[0039] The present application can realize quantitative analysis of rhythmic characteristics, accurately distinguish between voluntary eye movement and pathological nystagmus by combining phase characteristics and rhythmic analysis, reduce the misjudgment rate, has the characteristics of high recognition accuracy, strong anti-interference ability and high automation, provides objective quantitative basis for clinical diagnosis of nystagmus, and can be widely applied to ophthalmic clinical diagnosis. BRIEF DESCRIPTION OF DRAWINGS
[0040] Figure 1 In the preferred embodiment of the present application, the flowchart of the nystagmus recognition method is shown in the figure.
[0041] Figure 2 In the preferred embodiment of the present application, the flowchart of the multi-stage filtering process is shown in the figure.
[0042] Figure 3 In the preferred embodiment of the present application, the flowchart of the interval segmentation is shown in the figure.
[0043] Figure 4 In the preferred embodiment of the present application, the flowchart of the phase classification is shown in the figure.
[0044] Figure 5 In the preferred embodiment of the present application, the structure block diagram of the nystagmus recognition system is shown in the figure. DETAILED DESCRIPTION
[0045] The technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0046] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.
[0047] The present application will be further described below with reference to the drawings and specific embodiments, but not as a limitation of the present application.
[0048] In the preferred embodiment of the present application, based on the above-mentioned problems existing in the prior art, a nystagmus recognition method is provided, as shown in the figure, comprising: Figure 1
[0049] S1, acquiring pupil data, the pupil data comprising time series information and corresponding pupil position coordinate information;
[0050] During the data acquisition phase, eye-tracking devices are used to acquire video image data of the eyes. The sampling frequency of the eye-tracking devices ranges from 25Hz to 400Hz, and the sampling frequency value can be selected according to specific needs and scenarios.
[0051] After processing the acquired eye video images, the position of the pupil is identified from the video images, and relevant information is extracted to output pupil data containing time-series information and corresponding pupil position coordinates. This data is presented in a specific format, namely {(t1,x1,y1),(t2,x2,y2),...,(t n ,x n ,yn)}. Wherein, each element (t,x,y) represents the pupil state information at a specific moment; t represents the timestamp; (x,y) represents the coordinate position of the pupil center in the image coordinate system at the corresponding moment t.
[0052] S2, preprocess the pupil data to obtain preprocessed pupil data;
[0053] Specifically, in response to the problem of high noise interference in existing nystagmus recognition methods, this embodiment adopts multi-level filtering preprocessing to remove noise caused by pupil shape fitting accuracy errors or misidentification, thereby reducing the impact of pupil recognition noise and improving data quality.
[0054] like Figure 2 As shown, the multi-stage filtering process includes:
[0055] S21, Perform median filtering on the pupil data to obtain the first pupil data;
[0056] S22, Gaussian smoothing filter is applied to the first pupil data to obtain the second pupil data;
[0057] S23, perform low-pass filtering on the second pupil data to obtain preprocessed pupil data.
[0058] In the preprocessing stage, the raw pupil data undergoes a first-stage filtering process. Impulse noise can occur due to external interference, sensor malfunctions, or data transmission errors. This noise appears as isolated outliers in the data, severely affecting the accuracy of the data and the reliability of the analysis results. In this embodiment, the first-stage filtering process employs nonlinear filtering methods such as median filtering to remove impulse noise.
[0059] Specifically, a 3×3 filter kernel is used to perform median filtering on the pupil center coordinates (x, y) to remove impulse noise caused by recognition errors. This filter kernel slides point-by-point across the two-dimensional plane of the pupil data. For each center point, all data values within that point and its neighborhood are sorted, and the median value of the sorted values is selected as the new value for that center point. Median filtering preserves edge information and detailed features of the data while removing isolated outliers and impulse noise, resulting in smoother and more stable data.
[0060] Then, the first pupil data undergoes a second-stage filtering process. Since some minor random fluctuations or interferences can cause high-frequency noise, this noise, unlike impulse noise, does not produce obvious outliers, but it can make the data coarse and unstable, affecting the judgment of the overall trend and characteristics of the data. In this embodiment, the second-stage filtering process uses linear filtering methods such as Gaussian smoothing to remove high-frequency noise from the data.
[0061] In the Gaussian smoothing filtering process, a Gaussian kernel is used, and its standard deviation σ is set. The σ value determines the width and shape of the Gaussian distribution. A larger σ value results in a smoother distribution of the Gaussian kernel and a more even filtering effect, but may lead to the loss of detailed data information; while a smaller σ value results in a more concentrated distribution of the Gaussian kernel, better preserving the detailed features of the data, but may be less effective at removing high-frequency noise. In this embodiment, the standard deviation σ of the Gaussian kernel is set to 1.5. Using a Gaussian kernel with σ = 1.5 to smooth the filtering result effectively removes high-frequency noise while better preserving the detailed information of the data, making the data smoother and more continuous, and further improving the data quality.
[0062] Next, the first pupil data undergoes a third-stage filtering process. To preserve the effective signal components in the data, the third-stage filtering process is a low-pass filtering process.
[0063] The order of a filter determines its attenuation characteristics; the higher the order, the faster the filter attenuates in the stopband. The cutoff frequency is the boundary between the passband and the stopband. Low-frequency effective signal components related to eye movement below the cutoff frequency can pass through the filter smoothly, while high-frequency noise and interference signals above the cutoff frequency will be significantly attenuated.
[0064] In this embodiment, a Butterworth low-pass filter is preferably used. The Butterworth low-pass filter has a flat passband and a monotonically decreasing stopband, effectively suppressing high-frequency signals while preserving low-frequency signals. The filter order is set to 2, and the cutoff frequency is set to 5Hz. Using a second-order Butterworth filter with a cutoff frequency of 5Hz for zero-phase filtering preserves the effective signal components.
[0065] S3, the preprocessed pupil data is segmented into multiple interval data segments;
[0066] During the data segmentation stage, a time-series curve is plotted based on the preprocessed pupil data, and the curve is divided into multiple interval data segments based on the increase or decrease trend of the location data.
[0067] Specifically, such as Figure 3 As shown, the segmentation of the preprocessed pupil data includes:
[0068] S31, Calculate the velocity of the pupil position in the horizontal direction;
[0069] Wherein, velocity is the ratio of the position difference between adjacent time points to the time interval. The velocity value is calculated using the following formula:
[0070] v k =(x k+1 -x k ) / (T k+1 -T k );
[0071] Where v represents velocity; x represents the x-coordinate of the pupil center; T represents time; and the subscript k represents the time index.
[0072] S32, based on the direction of velocity, the time corresponding to the change in velocity direction is used as the dividing point;
[0073] The data is segmented based on the changes in velocity. Specifically, the point where the velocity direction changes is determined as the segmentation point; that is, based on the comparison of the signs of velocity at adjacent time points, when the velocity v at time k... k The velocity v at time k+1 k+1 When the signs are opposite, it indicates that the direction of the velocity has changed between time k and time k+1. In this case, time k+1 is marked as the dividing point.
[0074] S33, based on the segmentation points, the curve drawn based on pupil data is segmented to form multiple interval data segments.
[0075] Specifically, a curve is pre-plotted based on pupil data, visually representing the changes in pupil data over time. After determining all segmentation points, the curve is segmented according to these points. By segmenting at these points, the curve is divided into multiple data segments. Each data segment represents the changes in pupil data within a relatively independent phase of eye movement, facilitating the analysis of eye characteristics at different stages of eye movement.
[0076] S33 also includes:
[0077] Filter multiple data intervals, removing data intervals shorter than a preset length and retaining data intervals longer than the preset length.
[0078] Specifically, during actual eye movements, there may be some brief and unrepresentative movement phases. The data segments corresponding to these phases are relatively short, which may be due to accidental interference or minor fluctuations.
[0079] To ensure the data quality and validity of subsequent analysis, this embodiment filters the multiple interval data segments obtained from curve segmentation. Therefore, a preset length is set as the filtering standard. In this embodiment, the preset length is 3 frames to filter out all interval data segments with a length less than 3 frames, retaining valid interval data segments with a length greater than or equal to 3 frames.
[0080] The retained valid interval data segments after filtering can comprehensively describe the eye movement stage represented by that interval. Each interval data segment contains the following key information:
[0081] The start index start_idx and end index end_idx: These respectively identify the start and end positions of the data segment in the original pupil data sequence, so as to facilitate the extraction of data in this segment from the original data;
[0082] Start time start_t and end time end_t: These define the time range corresponding to the interval, making it easier to associate the characteristics of eye movements with specific time points to better understand the temporal patterns of eye movements;
[0083] The start position (start_x, start_y) and end position (end_x, end_y) represent the coordinates of the pupil at the beginning and end of the interval, respectively, which can intuitively show the movement trajectory and displacement of the eyeball within the interval.
[0084] S4, based on the velocity characteristics of each data segment, detects fast phase and slow phase modes;
[0085] Specifically, in this embodiment, the phase classification method based on velocity characteristics achieves accurate differentiation between fast and slow phases.
[0086] Among them, such as Figure 4 As shown, based on the velocity characteristics of each data segment, the detection of fast and slow phase modes includes:
[0087] S41, calculate the movement speed of each data segment in the interval;
[0088] The moving speed is the ratio of the displacement difference of the corresponding data segment to the time span. The displacement difference is the Euclidean distance between the first and last points of the corresponding data segment, and the time span is the difference between the end time and the start time of the corresponding data segment.
[0089] Calculate the average speed based on the moving speed of all interval data segments;
[0090] S42, mark the interval data segments according to their moving speed, and determine the fast and slow phase modes of each interval data segment; wherein, when the moving speed is greater than a preset speed threshold and the number of pixels moved by the interval data segment per unit time exceeds the preset threshold, the corresponding interval data segment is marked as fast phase mode; otherwise, it is marked as slow phase mode.
[0091] In a preferred embodiment, the preset speed threshold is three times the average speed of all interval data segments moving.
[0092] The preset threshold is 5.
[0093] In the phase classification stage, the movement speed of each data segment is first calculated. Specifically, the Euclidean distance between the first and last points of the data segment is calculated as the displacement difference; the formula for calculating the displacement difference dd is:
[0094]
[0095] The time span of the data interval is calculated as the difference between the end time and the start time, i.e., the time span Δt = end_t - start_t;
[0096] Based on the ratio of displacement difference to time span, the moving speed v = d / Δt of the data segment in this interval is calculated.
[0097] Next, the interval data segments are marked as fast or slow phases based on their velocity characteristics. Specifically, the average velocity v_avg of all interval data segments is calculated. When the movement speed of an interval data segment is greater than 3 times the average velocity and exceeds 5 pixels / second, the interval data segment is marked as a fast phase; otherwise, it is marked as a slow phase.
[0098] S5, when a fast-phase and slow-phase alternation pattern is detected to occur at least a preset number of times and the rhythmicity feature meets the rhythmicity requirement, an nystagmus event is determined to have occurred; wherein, the rhythmicity feature is the velocity variation coefficient.
[0099] S5 also includes:
[0100] When a pattern of alternating fast and slow phases is detected to occur at least a preset number of times, but the rhythmic characteristics do not meet the rhythmic requirements, an autonomic eye movement event is determined to have occurred.
[0101] In this embodiment, the preset number of times is preferably 3.
[0102] Specifically, existing nystagmus identification methods struggle to distinguish between voluntary eye movements and pathological nystagmus, and lack analysis of nystagmus rhythmic characteristics.
[0103] In this embodiment, by introducing the velocity variation coefficient, the rhythmic characteristics of nystagmus can be quantitatively analyzed; at the same time, by combining the alternating fast and slow phase modes with rhythmic characteristic analysis, it is possible to effectively distinguish between spontaneous eye movements and pathological nystagmus, thereby reducing the misjudgment rate.
[0104] During the nystagmus detection phase, alternating patterns of fast and slow phases are detected. When at least three consecutive sets of alternating patterns of fast and slow phases are observed and simultaneously meet the rhythmic characteristics, nystagmus is determined to have occurred.
[0105] Rhythmic characteristics are determined using the velocity coefficient of variation. The velocity coefficient of variation is a statistical indicator that measures the degree of dispersion of data. It can objectively and accurately describe the variation pattern of nystagmus velocity.
[0106] When an alternating pattern of fast and slow phases is detected and the number of alternation cycles (1 cycle = 1 fast phase + 1 slow phase) is ≥ 3, calculate the velocity variation coefficients of these phases:
[0107] CV=σ(v1,v2,...,v m ) / μ(v1,v2,...,v m )
[0108] Where m represents the total number of phases involved in the calculation; σ represents the standard deviation; and μ represents the mean.
[0109] When the velocity variation coefficient is less than 0.6, the eye movement is considered to be rhythmic.
[0110] When at least three consecutive fast and slow phase alternation patterns occur, but the velocity variation coefficient (CV) is greater than or equal to 0.6, it is determined to be spontaneous eye movement and is not recorded as a nystagmus event, in order to distinguish spontaneous eye movement from nystagmus.
[0111] When at least three consecutive alternating patterns of fast and slow phases occur and the coefficient of variation (CV) is less than 0.6, it is considered a nystagmus event, and relevant information is recorded, including the start and end times of the nystagmus. The start time refers to the beginning time of the first slow phase (start_t), and the end time refers to the end time of the last fast phase (end_t).
[0112] In a preferred embodiment, the method further includes extracting nystagmus parameters, which include nystagmus start time, nystagmus end time, number of nystagmus counts, start point information of each slow phase of nystagmus, and slow phase velocity; wherein the slow phase velocity is the ratio of the displacement difference between the first and last points of the slow phase to the time span.
[0113] Specifically, this invention can automatically extract complete nystagmus parameters, providing objective quantitative evidence for clinical diagnosis.
[0114] In the parameter extraction step, the total number of nystagmus events is counted as the number of nystagmus events; the start time and end time of each nystagmus event are recorded; the start point information (start_t, start_x, start_y) of the slow phase in each nystagmus event is extracted; and the velocity of each slow phase is calculated: slow phase velocity = slow phase displacement difference / slow phase time span.
[0115] In a preferred embodiment of the present invention, a nystagmus recognition system is provided for implementing the nystagmus recognition method as described above. Figure 5 As shown, the system includes:
[0116] Data acquisition module 1 is used to acquire pupil data, which includes time series information and corresponding pupil position coordinate information;
[0117] Preprocessing module 2, connected to data acquisition module 1, is used to preprocess the pupil data to obtain preprocessed pupil data;
[0118] Interval segmentation module 3, connected to preprocessing module 2, is used to segment the preprocessed pupil data into multiple interval data segments;
[0119] Phase classification module 4, connected to interval segmentation module 3, is used to detect fast phase and slow phase modes based on the velocity characteristics of each interval data segment;
[0120] The nystagmus detection module 5 is connected to the phase classification module 4. It is used to determine that an nystagmus event has occurred when a pattern of alternating fast and slow phases is detected to occur continuously for at least a preset number of times and the rhythmic characteristics meet the rhythmic requirements. The rhythmic characteristics are the velocity variation coefficients.
[0121] The nystagmus recognition method and system of the present invention acquires pupil data containing time-series information and corresponding pupil position coordinate information; employs multi-level filtering preprocessing to remove noise caused by pupil shape fitting accuracy errors or misidentification; divides the time-series curve of pupil position coordinates into multiple interval data segments based on position change trends; calculates the movement speed of each interval and classifies it into fast phase / slow phase; identifies nystagmus by detecting three or more consecutive fast-slow phase alternation patterns and combining rhythmic analysis; and automatically extracts the nystagmus start / end time, number of occurrences, and slow phase characteristic parameters.
[0122] The method and system of this invention can be integrated into medical diagnostic equipment. By improving data quality through multi-level preprocessing and combining phase feature and rhythmic analysis, it can achieve automatic identification and parameter extraction of nystagmus, automating the entire process, reducing manual intervention, improving diagnostic consistency, and enhancing diagnostic accuracy and efficiency. At the same time, it can solve the problems of large noise interference and difficulty in distinguishing between spontaneous eye movements and pathological nystagmus in the prior art. It has the characteristics of high identification accuracy, strong anti-interference ability, and high degree of automation, providing objective quantitative evidence for the clinical diagnosis of nystagmus and can be widely used in ophthalmological clinical diagnosis.
[0123] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.
Claims
1. A method for identifying nystagmus, characterized in that, include: Acquire pupil data, which includes time series information and corresponding pupil position coordinates; The pupil data is preprocessed to obtain preprocessed pupil data; The preprocessed pupil data is segmented into multiple interval data segments; Based on the velocity characteristics of each data segment, detect fast and slow phase modes; When a pattern of alternating fast and slow phases is detected to occur at least a preset number of times and the rhythmic characteristics meet the rhythmic requirements, an nystagmus event is determined to have occurred; wherein, the rhythmic characteristics are the velocity variation coefficients.
2. The nystagmus recognition method according to claim 1, characterized in that, The preprocessing includes multi-stage filtering, which includes: The pupil data is subjected to median filtering to obtain the first pupil data; The first pupil data is processed by Gaussian smoothing filter to obtain the second pupil data; The second pupil data is subjected to low-pass filtering to obtain the preprocessed pupil data.
3. The nystagmus recognition method according to claim 1, characterized in that, The segmentation of the preprocessed pupil data includes: Calculate the velocity of the pupil position in the horizontal direction; Based on the direction of the velocity, the time corresponding to the change in the velocity direction is used as the dividing point; Based on the segmentation points, the curve drawn based on the pupil data is segmented to form multiple interval data segments.
4. The nystagmus recognition method according to claim 1, characterized in that, The segmentation of the preprocessed pupil data also includes: The multiple interval data segments are filtered out, and interval data segments with a length less than a preset length are retained.
5. The nystagmus recognition method according to claim 1, characterized in that, The detection of fast and slow phase modes based on the velocity characteristics of each data segment includes: Calculate the moving speed of each of the data segments in the interval; The fast and slow phase modes of each data segment are determined by marking it according to the moving speed of the data segment; wherein, when the moving speed is greater than a preset speed threshold and the number of pixels moved by the data segment per unit time exceeds the preset threshold, the corresponding data segment is marked as fast phase mode; otherwise, it is marked as slow phase mode.
6. The nystagmus recognition method according to claim 5, characterized in that, The moving speed is the ratio of the displacement difference of the corresponding data segment to the time span, the displacement difference is the Euclidean distance between the first and last points of the corresponding data segment, and the time span is the difference between the end time and the start time of the corresponding data segment.
7. The nystagmus recognition method according to claim 5, characterized in that, The preset speed threshold is three times the average speed of all interval data segments; The preset threshold is 5.
8. The nystagmus recognition method according to claim 1, characterized in that, Also includes: When a pattern of alternating fast and slow phases is detected to occur at least a preset number of times, but the rhythmic characteristics do not meet the rhythmic requirements, an autonomic eye movement event is determined to have occurred.
9. The nystagmus recognition method according to claim 1, characterized in that, It also includes extracting nystagmus parameters, which include nystagmus start time, nystagmus end time, number of nystagmus, start point information of each slow phase of nystagmus, and slow phase velocity; wherein, the slow phase velocity is the ratio of the displacement difference between the first and last points of the slow phase to the time span.
10. A nystagmus recognition system, characterized in that, A method for implementing nystagmus recognition as described in any one of claims 1-9 includes: The data acquisition module is used to acquire pupil data, which includes time series information and corresponding pupil position coordinate information; The preprocessing module, connected to the data acquisition module, is used to preprocess the pupil data to obtain preprocessed pupil data. An interval segmentation module, connected to the preprocessing module, is used to segment the preprocessed pupil data into multiple interval data segments; The phase classification module, connected to the interval segmentation module, is used to detect fast phase and slow phase modes based on the velocity characteristics of each interval data segment; The nystagmus detection module, connected to the phase classification module, is used to determine that a nystagmus event has occurred when a pattern of alternating fast and slow phases occurring at least a preset number of times is detected and the rhythmic characteristics meet the rhythmic requirements; wherein, the rhythmic characteristics are the velocity variation coefficients.