Visual field detection equipment, system and method based on expanded reality

Abstract

The invention relates to the technical field of intelligent visual detection, in particular to visual field detection equipment, system and method based on extended reality (XR), and the method comprises the steps that an image feature extraction module collects image data in real time, extracts features and eliminates optical distortion; the fixation deviation correction module analyzes eye movement data in real time, quantifies fixation deviation and triggers visual prompt and correction; the adaptive stimulation generation module dynamically encrypts to generate a test point image and adaptively adjusts the test point image according to visual field analysis and test requirements; the dynamic adjustment image module predicts and optimizes a test point presentation sequence according to image features and fixation points, automatically encrypts a high-probability defect area and quickly jumps to a key screening area; and the visual field defect identification module extracts image features, calculates the defect probability of each region in real time by using a Bayesian network in combination with subject response data, and locates a high-risk region. The system provides an efficient and accurate visual field detection scheme through the synergistic effect of multiple modules.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent visual detection, in particular to a field of view detection device, system and method based on extended reality. BACKGROUND

[0002] Vision field (VF) refers to the spatial range that can be seen by the eyes when the eyes are fixed and not moving, which is often expressed in angles. Vision field can reflect the visual function characteristics of the entire visual pathway from the retina to the visual cortex. When any part of the visual pathway is diseased, corresponding defects will be shown in the visual field.

[0003] A perimeter is an essential device for detecting the visual field, and is a basic method for diagnosing and monitoring glaucoma, macular diseases of the fundus, optic nerve damage and other ophthalmic and other visual and optic nerve diseases. At present, commonly used perimeters in clinical practice include Goldmann perimeter, Octopus perimeter, Humphrey perimeter, MP-1 or MP-3 microperimeter and MAIA microperimeter, etc. These visual field detection devices are large-scale equipment and can only be used in fixed medical sites such as hospitals. Most of these devices are concentrated in large-scale first-class and second-class hospitals, and it is difficult for most primary hospitals or community hospitals in China to equip them, resulting in many patients missing the best monitoring opportunity and treatment opportunity. Therefore, there is an urgent need for a visual field detection device that is highly popular and easy to carry.

[0004] A head-mounted display (HMD) is a head-mounted image display and observation device, which mainly consists of four parts: an image source, an optical system, a display device, and a circuit control. Its working principle is to combine the image generated by the image source with the external scene image through the optical system and image on the focal plane of the goggles system. After passing through the eyepiece system, it is converted into parallel light and projected into the eye through the combination glass in front of the user's eye. According to the availability, HMD can be classified into virtual reality (VR) devices, augmented reality (AR) devices and extended reality (XR) devices. In recent years, with the rapid development of HMD technology, VR, AR or XR has more and more advanced and intelligent technical advantages in visualizing and imaging. Based on advanced XR technology (such as computer image and target recognition), the environmental information around the user becomes interactive and digitally operable, and the virtualized information related to the environment and objects can be superimposed on the real world, which is expected to meet the requirements of human eye visual field threshold detection. Therefore, the present application displays a visual field detection device based on HMD including but not limited to XR and a digital image processing method.

[0005] In the field of digital image processing, in order to ensure the accuracy of visual information, various advanced feature extraction algorithms are used to deeply perceive the current environment. Image segmentation, edge detection, texture feature analysis, pixel brightness histogram analysis and salient point distribution. Through these diversified features, the system can accurately perceive the current environmental lighting conditions and comprehensively evaluate the visual complexity of the test point image, providing key inputs for adaptive stimulus generation and subsequent analysis.

[0006] Although the prior art has made significant progress in image processing, in the specific application field of visual detection based on image presentation, it still faces serious challenges. In many cases, the key visual stimuli that need to be accurately identified or presented may only show extremely weak, low-contrast or blurred features in the image. More difficultly, these weak image signals are easily masked by complex and variable environmental lighting, inherent noise of display devices, and artifacts generated during image acquisition and processing. The existing image feature extraction and analysis algorithms often fail to effectively identify the potential key features when faced with such low signal-to-noise ratio, blurred details or high visual complexity. This constitutes the core problem that needs to be solved in current image-based visual detection technology, limiting the system's ability to perceive and judge subtle changes in complex scenarios.

[0007] Therefore, the present application proposes a visual field detection device, system and method based on extended reality. SUMMARY

[0008] The present application provides a visual field detection device, system and method based on extended reality, which improves the level of visual field health assessment through intelligent diagnosis. The system first extracts multi-dimensional features from the visual field and eye images, and eliminates optical distortion to ensure high-precision geometric position. At the same time, all visual field data are subjected to fine image preprocessing. Next, the system analyzes the eye movement data of the subject in real time, and uses image processing technology to accurately quantify the fixation deviation; when the deviation exceeds the preset threshold, a rendering correction visual prompt is generated immediately, with its brightness, contrast and other attributes dynamically adjusted in real time, effectively guiding the subject to maintain fixation.

[0009] Based on these preprocessed visual field data, the system can extract image features related to defects and combine them with the subject's response data to use Bayesian networks to predict the probability of potential defect areas in real time, clearly identifying high-probability defect areas. In addition, the system continuously assesses the environmental lighting and visual complexity of the test points, intelligently adjusts the brightness, contrast and presentation interval of the test points, and performs noise reduction enhancement. More importantly, the system dynamically encrypts the test points and adjusts the presentation sequence based on the predicted defect areas, which significantly optimizes the detection efficiency and greatly improves the diagnostic accuracy.

[0010] To achieve the above purpose, the present application provides the following technical solutions: An extended reality-based visual field detection system, comprising: An image feature extraction module: for real-time acquisition of image data of the eye and the environment by the built-in camera of the XR glasses, extraction of eye features, line-of-sight features and environmental features from the image data, while eliminating optical distortion; A fixation deviation correction module: for real-time analysis of eye movement data, quantification of fixation deviation, triggering of visual cues and correction when exceeding the threshold; An adaptive stimulus generation module: for dynamically generating new test point images according to visual field analysis and test requirements, and adaptively adjusting the brightness and contrast of the test point images in combination with environmental lighting and visual sensitivity of the subject, and performing noise reduction and enhancement processing on the test point images; A dynamic adjustment image module: for optimizing the test point presentation sequence according to image features and gaze point prediction, reducing saccade distance, automatically encrypting test points for high-probability defect areas, and quickly jumping to key areas for screening when defects are found; A visual field defect identification module: for extracting texture, shape, edge, and pixel motion-related image features, combining subject response data, and using Bayesian networks to calculate the probability of each area defect in real time to accurately locate high-risk visual field defect areas.

[0011] Preferably, the image feature extraction module includes: extracting feature information in the image, the feature information including eye features, line-of-sight features, and environmental features; eliminating optical distortion of the eye feature information to ensure geometric position accuracy of the image.

[0012] Preferably, the fixation deviation correction module includes: real-time analysis of current eye movement data, combination with a pre-set fixation area model, quantification of fixation deviation degree and direction through image processing technology, alarm and correction when exceeding the set threshold; dynamically generating and rendering a corrective visual cue image on the visual field image through image superposition and fusion technology according to the fixation deviation degree and direction; dynamically adjusting the brightness, contrast, size, color and animation properties of the corrective visual cue image according to the duration and amplitude of the fixation deviation.

[0013] Preferably, the adaptive stimulus generation module includes: dynamically generating new test point images according to visual field analysis and test requirements; dynamically adjusting the brightness and contrast of the test point images through histogram equalization according to the current environmental lighting conditions, the visual sensitivity of the subject, and the fixation performance of the previous test point image, and performing real-time image noise reduction and enhancement processing on the test point images.

[0014] Preferably, the dynamic image adjustment module includes: identifying the region of interest after segmenting the field of view image in combination with the current field of view image content, automatically generating and encrypting new test point images using the ORB feature point detection algorithm; based on the visual similarity, spatial distance and semantic association between the test point images, using the simulated annealing algorithm to optimize the test point presentation sequence; based on the sequence adjustment of the fixation point prediction, predicting the future fixation point through the CNN-LSTM neural network model and adjusting the presentation sequence of the test point image accordingly.

[0015] Preferably, the field defect identification module includes: image preprocessing of the acquired field of view data to form standardized field of view data; extracting image abnormal features related to field defects from the standardized field of view data, the image abnormal features including texture abnormalities, uneven brightness and color, geometric distortion and shape defects, edge and gradient abnormalities, and pixel-level motion abnormalities; Based on the image features and subject response data, the Bayesian network is used to predict the potential defect image region in real time and quantitatively evaluate the defect probability and confidence of the potential defect image region; based on the defect probability and confidence, the high probability region of the field defect is identified and located.

[0016] Preferably, the dynamic image adjustment module further includes: for the region predicted as high probability defect and suspected to have a boundary, dynamically inserting additional test point images to accurately locate the defect boundary; for the region continuously predicted as healthy and with extremely low defect probability, after confirming that the key points are normal, reducing the number of non-key test points in the region and skipping the test.

[0017] Preferably, the dynamic image adjustment module further includes: adjusting the standard test sequence to preferentially test the predicted high probability defect region and its adjacent edge region; if high probability defects are continuously detected in a certain region, the system suspends the comprehensive test of the current region, preferentially performs rapid screening on the remaining key regions, and then returns to the detailed detection of the defect region after completion.

[0018] An extended reality-based visual field detection device, comprising: the visual field detection device includes a portable glasses main body structure, the main body structure is built-in with a memory and a processor, and a dioptric adjustment knob is arranged to support dioptric adjustment and identification from-5D to 0D; the memory is used to store image data, model parameters and adaptive strategies; the processor is used to run image processing algorithms and Bayesian prediction models; the dioptric adjustment imaging principle adopts the Birdbath scheme, that is, the Micro-LED display source is refracted and reflected through the lens group, the picture is enlarged and projected into the human eye, the micro distance between the display source and the lens group is controlled through the adjustment knob, and the near and far positions of the virtual image in the retina are fine-tuned.

[0019] A kind of based on the field of view detection method of extended reality, specific steps include: S10. Real-time acquisition of image data of eye and environment, extract eye features, line-of-sight features and environmental features from the image data, while eliminating optical distortion; S20. Real-time analysis of eye movement data, quantify the degree of fixation deviation, trigger visual cues and correct when exceeding threshold; S30. According to the field of view analysis and test requirements, dynamically encrypted new test point image is generated, and the brightness and contrast of the test point image are adaptively adjusted in combination with the ambient light and the visual sensitivity of the tested person, and the test point image is enhanced; S40. According to image features and gaze point prediction, optimize test point presentation sequence, reduce saccade distance, automatically encrypt test points for high probability defect area, and quickly jump to key area screening when defect is found; S50. By extracting texture, shape, edge, pixel motion related image features, combining with the response data of the subject, using Bayesian network to calculate the probability of each area defect in real time, accurately positioning the high-risk visual field defect area.

[0020] Compared with the prior art, the beneficial effects of the present application are: 1. The system determines the device parameters by image analysis to eliminate optical distortion, ensures the geometric position accuracy of the image, and on this basis, uses intelligent vision technology to extract and process multi-dimensional image features such as eye features, line-of-sight features and environmental features, to accurately quantify the fixation quality and correct the fixation deviation. The system can predict potential defect image areas in real time based on the acquired visual field data and response data, and identify high probability areas of visual field defects. It greatly improves the accuracy of visual field detection, the detection rate of early and local defects, and provides intelligent diagnostic assistance.

[0021] 2. The system makes full use of intelligent vision technology and image analysis algorithm to perform image segmentation, region of interest identification and gaze point prediction on the visual field image. Based on this, the system can dynamically adjust the visual stimulus image presentation strategy, including dynamically generating new test point images and adjusting the test point image presentation sequence. This adaptive and intelligent test strategy based on image feature analysis, combined with real-time image feature analysis to accurately lock high-risk areas, improves detection efficiency while ensuring detection accuracy.

[0022] 3. The system employs sophisticated image analysis and processing techniques to extract and evaluate real-time features of the current environmental lighting conditions and the visual complexity of the test point images through intelligent vision. Based on these evaluation results, the brightness, contrast, color, and other parameters of the test point images are dynamically adjusted, and the presentation rhythm is optimized. Combined with real-time noise reduction and image enhancement techniques, the system ensures clear and distinguishable stimuli. When a fixation deviation is detected, the system automatically generates a correction prompt, improves the subject's cooperation, reduces fatigue interference, and thus enhances the reliability and repeatability of the detection data. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A structural schematic diagram of a visual field detection system based on extended reality is provided for the present application; Figure 2 A flow step diagram of a visual field detection method based on extended reality is provided for the present application; Figure 3 A flow chart of closed-loop control of a visual field detection system based on extended reality is provided for an embodiment of the present application; Figure 4 A hardware architecture diagram of an extended reality XR device is provided for the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. 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.

[0025] Please refer to Figures 1 to 4 The present application provides a visual field detection device, system and method based on extended reality, and the technical solution is as follows: Figure 1 A visual field detection system based on extended reality is provided for the present application, which comprises: An image feature extraction module: for real-time acquisition of image data of the eye and the environment, extraction of eye features, line of sight features and environmental features from the image data, and elimination of optical distortion; A fixation deviation correction module: for real-time analysis of eye movement data, quantification of fixation deviation degree, triggering of visual prompt and correction when exceeding the threshold; An adaptive stimulus generation module: for dynamically generating new test point images according to visual field analysis and test requirements, and adaptively adjusting the brightness and contrast of the test point images based on environmental lighting and visual sensitivity of the subject, and performing noise reduction and enhancement processing on the test point images; Dynamic image adjustment module: used to optimize the test point presentation sequence according to image features and gaze point prediction, reduce saccade distance, automatically encrypt test points for high-probability defect areas, quickly jump to key area screening when defects are found; Visual field defect identification module: used to extract texture, shape, edge, and pixel motion-related image features, combine with subject response data, and use Bayesian network to calculate the defect probability of each area in real time to accurately locate high-risk visual field defect areas.

[0026] The specific implementation steps are referred to Figure 2 A flowchart of a visual field detection method based on extended reality is proposed for the present application.

[0027] Embodiment one This embodiment details the application of an intelligent visual field detection system based on extended reality to the visual field detection process of a suspected glaucoma patient. The system uses a lightweight and portable XR glasses as the core hardware platform, which integrates a high-resolution micro display screen, a high-frequency eye movement tracking camera, an ambient light sensor (ALS), and an inertial measurement unit (IMU) multiple sensors. Not only does it provide an immersive, controllable, and highly customizable virtual environment for visual stimulation, but more importantly, it can capture the eye movement data and environmental information of the subject in real time and dynamically adjust the detection strategy based on this information, thereby overcoming the limitations of traditional methods.

[0028] The workflow of the entire system is highly integrated and cyclically feedback, referring to Figure 3 A flowchart of the closed-loop control of a visual field detection system based on extended reality is proposed for the present application. This closed-loop control is the key to realizing intelligent and adaptive detection of the system. Refer to Figure 4 A hardware architecture diagram of the extended reality XR device of the present application.

[0029] Make full use of the unique advantages of XR technology, especially eye movement tracking, sensor fusion, and real-time computing power, to achieve detection accuracy, efficiency, user experience, or data dimensions beyond traditional perimeters, rather than just reproducing standard procedures.

[0030] As the perception front end of the entire system, the image feature extraction module is the cornerstone of accurate eye movement tracking and environmental perception. It is responsible for identifying and quantifying eye features, gaze features, and environmental feature information from the raw image data captured by the internal (towards the eyes) and external (towards the environment) cameras of the extended reality headset.

[0031] Specifically, the eye features mainly include the corrected pupil center position, diameter and shape, which are the basis for achieving high-precision eye tracking and fixation analysis; the gaze features include the real-time gaze point, gaze direction vector and eye movement patterns, such as fixation, saccade and microsaccade, which are used to monitor fixation stability and predict the visual focus of the subject; the light conditions obtained by the ambient light sensor ALS and the background complexity obtained by analyzing the external camera image are used to adaptively adjust the presentation parameters of the stimulus image.

[0032] Specifically, new test point images are dynamically encrypted and generated according to the visual field analysis and test requirements, and the stimulus point diameter and cursor size are adaptively adjusted according to the preset standard (such as Goldmann type III), and the brightness, contrast of the test point image are adaptively adjusted in real time according to the current background brightness, ambient light and visual sensitivity of the subject, and the test point image is processed by noise reduction and enhancement.

[0033] According to the commonly used 24-2 standard detection strategy for glaucoma, the initial test points are intelligently generated in the visual field of the patient, the default light point brightness is 1000 asb, or is adjusted according to the highest brightness of the XR glasses, the size is Goldmann type III, and the spatial distribution follows the 50-100 or so grids commonly used in the 24-2 detection strategy. For example, the 76 grids in the embodiment. The system will automatically set the initial background brightness according to the patient's age, refractive error and other information, the background light brightness is 3 asb or above, the typical background light brightness is 31.5 asb, the typical examination distance is 300 mm, the stimulus point duration is 100-200 ms, and the interval time range is 1.5-4 s, ensuring compatibility with international standards.

[0034] Specifically, when the subject fails to respond at the previous test point and the eye tracking shows that the fixation is stable, the system will determine that the stimulus may not be perceived due to insufficient intensity. At this time, the system will increase the brightness of the next and repeated test points by 4 dB, and may also fine-tune the cursor size from Goldmann type III to type IV, and perform real-time image noise reduction on the test point image to reduce background visual noise interference, and perform edge enhancement processing to make the test point outline clearer, ensuring that it can be accurately captured even at the subject's lower visual sensitivity.

[0035] The system has multiple standard test strategies, including 24-2, 10-2 and 30-2, and can intelligently select and dynamically adjust according to the patient's disease type and detection purpose.

[0036] Further, in the extended reality head-mounted display, the barrel / pillow distortion of the optical lens can cause eye feature positioning deviation, affecting the eye movement tracking accuracy. To this end, the system adopts Zhang Zhengyou calibration method in advance: through multi-angle shooting of the chessboard calibration board, the camera intrinsic parameters are calculated, including focal length, principal point, distortion coefficient and extrinsic parameters, and a distortion model is established. In subsequent real-time processing, the reverse mapping is performed on each frame of eye image, the target pixel coordinates are converted to floating point coordinates in the original distorted image using the distortion coefficient, and the undistorted image is reconstructed through bilinear interpolation, ensuring the geometric position accuracy of features such as pupil and corneal reflection points.

[0037] Specifically, the subject wears an extended reality head-mounted display, and the system collects data using the built-in eye movement tracking camera and external ambient light sensor ALS. First, the optical distortion is corrected by Zhang Zhengyou calibration method to ensure the accuracy of eye feature extraction. Subsequently, the system analyzes the eye movement data of the subject and establishes a 1.5 degree visual angle fixation stability baseline.

[0038] Through the multi-sensor space-time alignment subsystem, accurate spatial calibration and time synchronization of all eye movement, environmental and response data are ensured before fusion, completely eliminating data misplacement and laying a foundation for high-precision information fusion.

[0039] The hierarchical sensor fusion processor is responsible for efficiently utilizing these high-quality inputs for in-depth analysis and intelligent decision-making. The core lies in the hierarchical progressive fusion of data, where the bottom layer of sensors aligns and cleanses raw eye movement, light, response and other data, extracting primary features such as fixation points and blink frequency. The middle layer of sensors integrates and infers abstract features such as user fixation quality, visual sensitivity and real-time state. The top layer of sensors uses Bayesian networks and other methods for high-level visual field defect probability inference and dynamically optimizes subsequent test strategies.

[0040] This layered fusion mechanism enables the system to extract insights from multi-dimensional and multi-granularity data, achieving more accurate, intelligent and adaptive visual field detection, significantly improving screening and diagnostic capabilities.

[0041] Distortion correction is essentially a geometric transformation that rearranges the pixels of an image and displays them in straight lines, ensuring the geometric position accuracy of the image. In the corrected image, the pixel coordinates of eye features such as pupil center and corneal reflection points more accurately reflect their relative positions in the real space, providing accurate input for subsequent eye model calculations of gaze direction. Through this link, the system can extract high-precision, undistorted eye and environmental information from the original pixel data, providing a reliable foundation for subsequent gaze analysis and intelligent decision-making.

[0042] The accuracy of visual field testing highly depends on the ability of the subject to maintain stable fixation during the test. Even slight eye drift and irregular microsaccades lead to false positive and false negative results. Therefore, the system integrates an advanced fixation quality monitoring and real-time correction module to ensure the reliability of the detection data.

[0043] The system takes the raw gaze point data provided by the eye tracking module built-in the extended reality headset as input, 1000 points per second. These data include the coordinates of the gaze point on the extended reality display and the line of sight vector in the three-dimensional space for each timestamp.

[0044] The embodiment sets the fixation target as the center of a circular and elliptical area with a radius of 2 degrees of visual angle as the fixation area.

[0045] During the test, the subject occasionally loses fixation due to fatigue and involuntary eye movements. When testing the upper area of the left eye, the eye tracking and fixation analysis module detects that its gaze point deviates from the center of the fixation target by more than 2.5 degrees of visual angle and lasts for about 350 milliseconds. The system immediately triggers the feedback mechanism, a red warning box pops up in the extended reality headset, and the central fixation point starts to flash at a frequency of 2 Hz, accompanied by a soft voice prompt "Please keep your central gaze".

[0046] Due to the large deviation amplitude and long duration, the system dynamically adjusts the intensity of the corrective visual cues: a larger and brighter dynamic arrow pointing to the center point is superimposed, and histogram equalization processing is adopted to make it more eye-catching in the current background. As the subject's fixation returns to stability, the brightness, size, and flashing frequency of the arrow prompt will gradually weaken until it disappears.

[0047] This process ensures the reliability of data acquisition and avoids false positive and false negative results caused by unstable fixation. Through this precise fixation monitoring and correction mechanism, the detection error caused by unstable fixation can be minimized, and the reliability and clinical value of the data can be improved.

[0048] According to the 24-2 standard detection strategy commonly used for glaucoma, the initial test points are intelligently generated within the patient's visual field, with a default light point brightness of 1000 asb or adjusted according to the maximum brightness of the XR glasses, a size of Goldmann III type, and a spatial distribution following the 50-100 grid commonly used in the 24-2 detection strategy. This embodiment uses 76 grids. These test points are superimposed in the form of virtual light spots in the extended reality environment, and the patient can see the real environment through the semi-transparent screen while perceiving the virtual light spots.

[0049] The subject is tested in a clinic, the ambient light sensor of the extended reality head-mounted display detects that the indoor light is relatively dark in real time, and the virtual scene simulation effect of the test point to be presented in the current field of view image is insufficient, specifically manifested as that the contrast of the shadow of the blood vessels in the fundus is low, the system will immediately instruct the adaptive stimulation strategy engine to predict the possible area of visual field defect in real time, dynamically increase the test point density or adjust the test order of the area, and improve the detection efficiency and accuracy of early or local lesions, instead of fixedly testing all 76 points. The histogram equalization processing is applied to the test point image, the pixel value distribution is dynamically adjusted, the brightness histogram is more uniform, the contrast of the image is significantly enhanced, and the image is more easily distinguished by the subject in the dim background.

[0050] When the subject fails to respond at the previous test point, and the eye tracking shows that the fixation is stable, the system will judge that the stimulation may not be perceived due to insufficient intensity. At this time, the system will increase the brightness of the next and repeated test point by 4dB, and perform real-time image noise reduction processing on the test point image to reduce background visual noise interference, and perform edge enhancement processing to make the test point outline clearer, so that the test point can be accurately captured even under the possible low visual sensitivity of the subject.

[0051] No matter how bright or dark the light in the clinic is, and how complex the background is, the system can dynamically adjust the stimulation parameters to ensure that the test point is presented with constant visual intensity in any environment, eliminating the influence of environmental light changes on the detection results in traditional methods. At the same time, according to the real-time fixation performance of the subject and the preliminary judgment of the visual sensitivity, the system can individually adjust the stimulation intensity to ensure that the test is neither too difficult to cause fatigue nor too simple to fail to detect real defects.

[0052] Further, in combination with the content of the current field of view image, a new test point image is automatically identified and generated in the area rich in visual information and needing to be focused on by using an ORB feature point detection algorithm. The feature points detected by the ORB algorithm are usually located in areas with significant gray scale changes and textures in the image, including corner points, edges and high-contrast areas. In the area where the feature points detected by the ORB are densely distributed, and in the area that needs to be focused on which is manually preset, additional test points are automatically generated. These points are not simply added to the fixed grid, but are generated by increasing the density at the center and periphery of the feature point cluster, and are adaptively added based on the distribution density of the feature points.

[0053] The field of view image is segmented by image segmentation, the image is divided into different semantic regions, and the interesting regions are automatically identified based on the algorithm, and then the test point image is generated by increasing the density in the interesting regions.

[0054] Specifically, the simulated mapping image of the subject's fundus and the semantic understanding image of the augmented reality scene are processed by a deep learning-based semantic segmentation network DeeplabV3+. The image is divided into multiple semantic regions, including the macular area, the para-macular area, the nasal visual field, the temporal visual field, the superior and inferior arcuate areas, and the physiological blind spot area.

[0055] Based on the preset clinical knowledge base and machine learning model, the "region of interest" of the current detection task is automatically identified; within these identified ROIs, the system dynamically increases the density of test points, which helps to more accurately depict and quantify the visual field defects in specific areas. According to the test requirements, the resolution of the test point image is dynamically adjusted, and the super-resolution image processing technology is used to generate higher detail test point images in areas that require high precision measurement.

[0056] Further, for those areas that are expected to have small defects and require fine boundary delineation, the system can generate test point images with higher resolution. In the early stages of glaucoma, the lesions often start from small, low-contrast dark spots or sensitivity decreases, and these subtle visual stimuli are easily overlooked by traditional methods or cannot be effectively presented due to display limitations. By providing lossless and sharp stimuli through ESRGAN, the system can maximize the detection sensitivity of these weak visual stimuli, thereby significantly reducing the false negative rate and avoiding missed diagnosis of those already occurring but not yet significant lesions. This ability fundamentally enhances the system's ability to capture "marginal differences", providing clinicians with more early and accurate diagnostic evidence, which is crucial for early intervention and preservation of visual function in neurodegenerative eye diseases such as glaucoma.

[0057] Further, optimizing the test point presentation sequence can significantly improve detection efficiency and subject experience, reduce fatigue, and detect key lesions earlier. If adjacent test points have similar visual properties, including brightness, color, and background, they can be grouped and presented. The nearest untested point to the current test point is presented first to minimize the saccade distance. Genetic algorithms are used to plan the optimal access sequence of test points. Test points belonging to the same anatomical region and the same type of visual function are grouped and presented. A "cost" matrix is established between test points, including distance, similarity, and semantic weight, and an optimization algorithm is run to find the optimal path.

[0058] Based on gaze point prediction sequence adjustment, combined with eye movement data, the future gaze point is predicted by image processing algorithms, and the presentation sequence of the test point image is adjusted accordingly to guide the subject's vision: Using real-time eye movement data, convolutional neural networks (CNN) are used to extract local features in time series, and then these features are input into the LSTM layer for sequence modeling to predict the most gazed location within the next 500 milliseconds.

[0059] The system initiates a standard detection sequence. The prediction module, in combination with the patient's medical history, identifies a high-risk area under the left eye temporal side. The adaptive stimulation engine responds by increasing the test point density from 2 degrees to 1 degree in that area. When the system predicts that the patient's gaze is about to slightly deviate to the lower nasal side, it presents the next test point in that area and its vicinity in advance, reducing the time for the patient to find the stimulus, and shortening the average response time by about 50 milliseconds. The system also uses a genetic algorithm to group test points within the same arcuate sector and optimize their presentation order to minimize saccadic distance and improve detection efficiency. When the system predicts that the patient is about to fixate on a certain area, it can present the next test point in that area and its vicinity in advance.

[0060] According to the patient's overall reaction speed and stability, the stimulation interval time is adjusted in real time, or the initial stimulus brightness is adjusted in a specific area according to the preliminary detection results to approach the threshold faster. This can effectively reduce the time for the patient to find the stimulus point and improve the reaction speed. When the system predicts that the patient's gaze is about to deviate from the test area, it presents a guiding test point in advance to pull the patient's gaze back. This predictive adjustment is essentially a visual guide. It makes the appearance of the stimulus point more consistent with the patient's natural eye movement pattern, rather than passively waiting for the patient to find the stimulus point.

[0061] Further, before making predictions, the original visual field data needs to be standardized and optimized: remove abnormal data points caused by fixation instability, blinking, device errors, etc. Standardize different stimulus intensity and response time data to a unified scale. Visual field detection is usually done at discrete points. In order to form a continuous "visual field image", bilinear interpolation is used to convert discrete threshold and response data into a continuous probability density map of visual field defects. At the same time, use smoothing algorithms such as Gaussian filtering to reduce image noise.

[0062] Normalize the sensitivity value and response probability to the range of 0-1 for image processing and model input. Based on the patient's response data, a density map representing the sensitivity and extent of visual field defects is generated.

[0063] Further, macro features that can reflect pathological changes of visual field defects are extracted from the pre-processed probability density map of visual field defects. A normal visual field sensitivity map usually presents smooth and uniform texture. The defect area shows rough, chaotic and sudden interruption of texture.

[0064] Further, the various image features extracted include the degree of texture abnormality, brightness deviation, edge sharpness, and the original response data including whether to see, reaction time, fixation stability.

[0065] Latent variable nodes represent intermediate pathological states, while target nodes represent the probability of visual field defects in each visual field region. Conditional probability tables (CPTs) define the conditional probability of each node given its parent node; these CPTs can be learned from a professional knowledge training library containing features and diagnostic results from a large number of known cases of visual field defects.

[0066] Once new visual field data and response data are acquired and features are extracted, these features are input as evidence into the Bayesian network. The posterior probability of each target node is updated in real time using a probabilistic inference algorithm, thus predicting the potential defective image region and its probability of defect in real time.

[0067] Furthermore, based on the prediction results of the image analysis model, the defect probability and confidence level of each potential defective image region are quantitatively evaluated. The output of the Bayesian network is the defect probability value of each pixel block in the visual field, ranging from 0 to 1. Based on the defect probability value, high-probability regions of visual field defects are identified and located. A value greater than 0.7 indicates a high probability of defect. Pixels are labeled as "defective" if their value is greater than 0.7 and "normal" if their value is less than or equal to 0.7, based on this threshold. Then, connected component analysis is used to identify consecutive pixel regions with a probability higher than the threshold as independent "high-probability defect regions." The system highlights these identified high-probability defect regions on the visual field image, presenting them intuitively to operators and doctors.

[0068] Bayesian networks effectively handle incomplete and fuzzy input data, providing robust decision support through probabilistic inference. Their network structure intuitively represents the causal and correlational relationships between variables, greatly enhancing model interpretability and enabling users to clearly understand the logic behind decisions. Furthermore, Bayesian networks can seamlessly integrate multi-source information for comprehensive judgment. With the continuous input of new data, the model can dynamically update its predictions, adaptively adjusting and optimizing in real time.

[0069] Furthermore, for areas predicted to have high probability of defects and suspected boundaries, additional test point images are dynamically inserted to improve the localization accuracy of defect boundaries and the detection rate of focal defects: when the prediction module identifies a region with a defect probability higher than 70% and whose edges are unclear, the system will immediately provide feedback to the stimulus generation module.

[0070] Within these high-probability defect areas and at their boundaries with healthy areas (i.e., defect boundaries), the system automatically increases the density of test points. The original 2-degree spacing between test points is changed to 1-degree or even 0.5-degree spacing, generating more test points.

[0071] Focal defect detection rate: For small, isolated focal defects, increased testing can significantly improve their detection probability. However, for large, continuous areas predicted as healthy or with a very low probability of defect, after confirming the absence of abnormalities at key points, the number of non-critical test points within the area should be appropriately reduced, and some tests should be skipped. When the prediction module consistently determines that a large area has a defect probability of less than 10% and is considered healthy, and the test results of the macula center and optic disc in that area are also normal, the system will provide feedback to the stimulation strategy engine. In these areas, the system can intelligently change the 2-degree interval to a 4-degree interval, or even skip some non-critical test points to shorten the detection time.

[0072] Intelligent strategies avoid unnecessary repeat testing in known healthy areas. They precisely concentrate computational and stimulus resources on high-risk or areas requiring further confirmation, significantly reducing overall testing time and alleviating the testing burden on patients. This efficient resource allocation not only improves testing efficiency but also optimizes the use of limited system performance.

[0073] "No abnormalities at key points" ensures that even sparse testing does not miss important information. This includes the center point, boundary points, and specific points of clinical importance within the sparse region.

[0074] Furthermore, the standard test sequence is adjusted to prioritize earlier and more concentrated testing of predicted high-probability defect areas and their adjacent edge areas. When the system starts detection and enters a new testing phase, it first queries the defect probability map of the prediction module. Test points predicted as high-probability defects and adjacent edges are placed at the beginning of the test sequence and tested continuously. This means the system will "skip" the currently ongoing routine sequence and prioritize these urgent and important areas.

[0075] If significant defects are detected rapidly and continuously in a certain area, the system can pause the full testing of the current area, prioritize jumping to other key areas for rapid screening, and then return to refine the previously detected defect area: When the system receives feedback from the subject that there is no response to stimulation at 3-5 adjacent test points, combined with the judgment of the real-time prediction module, the probability of a defect in that area rapidly increases. At this point, the system will determine that continuing to conduct detailed testing in this area is inefficient, as a large defect is highly suspected. It will immediately pause testing in the current area and, according to preset clinical protocols and prediction models, prioritize jumping to other diagnostically valuable "critical areas" for rapid screening. These critical areas include: the other eye, the central visual field, and other visual field areas related to this type of lesion.

[0076] After completing a rapid screening of other key areas, the system returns to the areas where significant defects were previously detected. At this point, a more refined strategy is employed, including increasing the density of test points, adjusting stimulus intensity, and enhancing fixation monitoring rigor to precisely delineate the boundaries, depth, and shape of the defects.

[0077] When the test subject detected the lower nasal region of the left eye, there was no response at five consecutive test points. Based on these "non-response" data and combined with previously collected eye movement stability data, the prediction and diagnostic analysis module used a constructed Bayesian network model to calculate in real time and predicted a high probability of visual field defects in this region, up to 90%. The system immediately fed this feedback to the adaptive stimulation strategy engine.

[0078] The system determined that there might be a large area of ​​damage in this region and paused detailed testing of that area. In order to achieve "early detection and early diagnosis", the system prioritized rapid screening of the subject's right eye to rule out the possibility of acute lesions in both eyes at the same time.

[0079] After confirming that the right eye was basically normal, the system returned to the lower nasal region of the left eye. At this point, the system used encrypted test points, adjusted stimulus intensity, and increased fixation monitoring strictness to accurately depict the defect boundary, depth, and shape of this region.

[0080] The goal is to achieve "early detection and early diagnosis." This involves rapidly identifying and ruling out lesions in high-risk areas, providing timely information for subsequent diagnostic decisions. This "breadth-first, depth-later" strategy effectively balances testing efficiency and diagnostic comprehensiveness, avoiding excessive time spent on large defect areas, thus obtaining the most comprehensive diagnostic information in the shortest possible time.

[0081] After the detection is completed, all raw visual field data are cleaned, standardized, and bilinearly interpolated to generate sensitivity maps. After filtering and noise reduction, the prediction and diagnostic analysis module begins in-depth processing.

[0082] The system extracts macroscopic features from the preprocessed sensitivity map to determine whether they conform to the typical arcuate defect characteristics of glaucoma. These features are input as evidence nodes into the Bayesian network model. Through probabilistic inference, the model outputs the defect probability and confidence level of each visual field region in the subject's left eye in real time.

[0083] Table 1: Quantitative Report on Visual Field Function Testing

[0084] Table 1 shows the quantitative report of visual field function testing. The report indicates that the subject has highly suspicious glaucomatous damage in the lower nasal visual field of the left eye, with a 95% probability of loss on the lower nasal side, covering an area of ​​15 square degrees, and light sensitivity decreased to -10 dB. The results suggest glaucomatous optic neuropathy in the left eye, and further diagnosis with fundus photography and OCT is recommended.

[0085] This embodiment, based on an augmented reality visual field detection system, constructs a highly intelligent, adaptive, and user-friendly visual field detection platform. From high-precision ocular feature extraction and real-time optical distortion elimination to advanced fixation quality monitoring and dynamic correction, to adaptive visual stimulation strategies based on predictive models, and finally to real-time visual field defect prediction and identification, this system comprehensively improves the efficiency, accuracy, and patient experience of visual field detection.

[0086] Through the synergistic effect of image feature extraction and fixation deviation correction, the system ensures high accuracy and reliability of the detection data. The adaptive stimulus generation and dynamic image adjustment module intelligently optimizes the test sequence, density, and stimulus parameters based on the subject's real-time status, ambient lighting, and detection needs, significantly shortening the detection time while improving subject comfort and cooperation. Furthermore, the visual field defect identification module utilizes a Bayesian network to fuse multi-source features and response data, calculating the probability of defects in each region in real time, providing highly confident and interpretable diagnostic evidence. The system overcomes many limitations of traditional visual field testing, offering a more advanced, efficient, and user-friendly solution for early screening, precise localization, and dynamic monitoring of ophthalmic diseases.

[0087] Example 2 The macula is the most visually acute area of ​​the retina, responsible for fine vision and color perception. Age-related macular degeneration (AMD), diabetic macular edema, and other diseases can severely impair central vision, leading to difficulties in daily activities such as reading and recognizing faces. The 10⁻² visual field test focuses on the central 10-degree range of the visual field and is the gold standard for assessing macular function. However, traditional methods face challenges in sensitivity and accuracy when detecting early, small, or irregularly shaped central scotomas due to unstable fixation, relatively coarse stimulus points, and fixed testing procedures. This embodiment aims to demonstrate how the system of the present invention significantly improves the accuracy, efficiency, and reliability of the 10⁻² central visual field test through its intelligent and adaptive characteristics.

[0088] Patient profile: A 65-year-old patient presents with central visual distortion and mild visual decline over the past six months. Preliminary fundus examination suggests early-stage wet age-related macular degeneration (wAMD). High-precision 10⁻² central visual field testing is required to accurately locate and quantify functional impairment.

[0089] Furthermore, after patient information and preliminary diagnosis are entered into the system, the system automatically recommends and selects the "10⁻² high-precision central visual field detection" strategy based on keywords such as "macular degeneration" and "central visual acuity loss." This strategy automatically loads a set of parameters optimized for macular function. Stimulus point parameters: The default stimulation point diameter is adjusted to a finer Goldmann type I or II to detect tiny dark spots. The cursor brightness adjustment step is also more precise, set to 1dB instead of the standard 2-4dB.

[0090] Background brightness: The background brightness setting should conform to international standards to ensure the comparability of the results.

[0091] Fixation accuracy requirements: Given the extremely high requirements for fixation in the center field of view, the threshold of the fixation deviation correction module is automatically set to a more stringent 1.0-degree viewing angle. Any response data that exceeds this range will be marked as low confidence.

[0092] Furthermore, high-precision central fixation monitoring and micro-saccade analysis are crucial in central visual field detection. Even minor fixation instability, such as micro-saccades and eye drift, can cause the test point to fall at different positions on the retina, leading to misjudgments. The system of this invention demonstrates its core advantage in this aspect: Specifically, the fixation deviation correction module utilizes the 1000Hz high-frequency eye-tracking camera built into the XR glasses to not only monitor macroscopic fixation deviations but also introduces real-time analysis of micro-saccade frequency, amplitude, and eye movement speed. The system establishes a multi-dimensional "fixation quality" model, continuously quantifying the patient's fixation stability at every moment.

[0093] When the system detects an abnormally high microsaccade frequency or an eye drift velocity exceeding a preset central visual field stability threshold, it indicates that the patient may be losing stable fixation. The system immediately triggers a gentle visual cue, with the central fixation point briefly flashing or changing color, guiding the patient to re-stabilize their fixation. During this period, any subject response data collected during low fixation quality is automatically and specially labeled by the system, and assigned a lower weight or discarded during the final analysis. This ensures that the final visual field map highly reflects true macular function, rather than fixation artifacts. This dynamic fixation quality assessment and real-time feedback mechanism based on eye tracking is crucial to ensuring the reliability of central visual field testing results.

[0094] Furthermore, the system employs a series of intelligent strategies for central field-of-view adaptive encryption and stimulus sequence optimization, targeting potentially small and irregular lesions in the macular region: In the initial stage of detection, the Bayesian network of the visual field defect identification module, based on the patient's initial response and combined with the built-in knowledge base of macular degeneration visual field defect patterns, predicts in real time the areas most likely to have defects, such as areas corresponding to suspicious neovascularization locations found in fundus examinations. The dynamic image adjustment module then responds accordingly, automatically increasing the test point density from the standard 2-degree spacing to 0.5-1.0 degrees in these high-risk areas, such as the fovea and parafovea, for ultra-high-density detection.

[0095] Super-resolution stimulus presentation: For predicted small or low-contrast defect areas, the adaptive stimulus generation module invokes the Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) technology to perform real-time super-resolution processing on the test point images of that area. This ensures that even extremely small and dark stimulus points are presented in the clearest and sharpest way, greatly improving the system's detection rate of early and subtle central dark points and solving the problem of insufficient display accuracy in traditional devices.

[0096] Sequence optimization and rapid screening: The dynamically adjusted image module utilizes simulated annealing to optimize the presentation sequence of test points within a 10-degree radius of the center, prioritizing the detection of high-probability defect areas and minimizing scanning distance to effectively reduce patient fatigue. If a clear central scotoma is rapidly detected in a certain area, the system will pause the refinement of that area and prioritize rapid screening of other key central areas, such as the parafovea on the other side, achieving an efficient "detect first, then refine" strategy.

[0097] After the detection is completed, the visual field defect identification module performs a comprehensive analysis of all valid data: This system constructs a multi-source information fusion framework based on Bayesian networks, integrating subject behavioral response data (stimulus awareness / reaction time), high-precision eye-tracking features (fixation stability score), and OCT image structural information (macular edema area). It quantifies lesion features from visual field sensitivity maps using intelligent feature extraction technology, such as the irregularity of central scotomas and the steepness of defect boundaries, achieving joint structure-function analysis. The system employs a probabilistic graphical model to establish conditional dependencies between parameters, supporting prior knowledge injection and online parameter learning. This significantly improves the early detection capability of macular lesions, especially the sensitivity to small lesions, and significantly enhances the accuracy of diagnostic results. Simultaneously, a multiple validation mechanism effectively controls the probability of false positives, providing a more reliable and objective basis for clinical decision-making.

[0098] The system uses a probabilistic inference algorithm to generate a high-resolution central visual field defect probability map. This image can accurately identify macular functional abnormalities, including absolute scotomas, relative scotomas, and functional areas corresponding to visual distortion, achieving millimeter-level precision. The quantitative report output by the system, as shown in Table 2, fully presents the central visual field (10⁻²) detection results. Standardized indicators are used to quantitatively assess the degree of visual field defect from multiple dimensions, providing objective evidence for clinical diagnosis.

[0099] Table 2: Quantitative Report on Central Visual Field (10-2) Functional Testing

[0100] This embodiment fully demonstrates that the system of the present invention has significant advantages over existing technologies when performing the highly demanding 10⁻² central visual field detection: through dynamic fixation quality assessment, microsaccade analysis, and data filtering, it eliminates fixation instability, the biggest source of interference, ensuring the authenticity and reliability of central visual field data. Combined with Bayesian network-guided adaptive encryption and ESRGAN super-resolution stimulation, the system can effectively detect minute, early-stage macular degeneration that is difficult to detect with traditional equipment, gaining valuable time for early intervention. Intelligent sequence optimization and adaptive strategies significantly shorten detection time while maintaining accuracy, reducing fatigue and difficulty in cooperation for patients, especially elderly patients. The system not only "detects" defects but also provides precise quantitative data on the area, depth, shape, and location of defects, providing richer and more reliable evidence for clinical diagnosis, efficacy evaluation, and disease monitoring.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A field-of-view detection system based on augmented reality, characterized in that, include: Image feature extraction module: used to collect image data of the eyes and environment in real time through the built-in camera of XR glasses, extract eye features, gaze features and environmental features from the image data, and eliminate optical distortion; Fixation deviation correction module: used to analyze eye movement data in real time, quantify the degree of fixation deviation, and trigger visual prompts and correction when it exceeds the threshold; Adaptive stimulus generation module: used to dynamically generate new test point images based on visual field analysis and testing requirements, and adaptively adjust the brightness and contrast of the test point images in combination with ambient lighting and the visual sensitivity of the test subject, and perform noise reduction and enhancement processing on the test point images. Dynamic image adjustment module: It is used to optimize the test point presentation sequence based on image features and gaze point prediction, reduce the scanning distance, automatically encrypt test points for high-probability defect areas, and quickly jump to the key area for screening when a defect is detected. Visual field defect recognition module: It is used to extract image features related to texture, shape, edge, and pixel motion, and combine them with subject response data to calculate the probability of defects in each region in real time using a Bayesian network, so as to accurately locate high-risk visual field defect areas.

2. The field-of-view detection system based on augmented reality according to claim 1, characterized in that, The image feature extraction module includes: extracting feature information from the image, including eye features, gaze features, and environmental features; and eliminating optical distortion of the eye feature information to ensure the geometric position accuracy of the image.

3. The field-of-view detection system based on augmented reality according to claim 1, characterized in that, The fixation deviation correction module includes: real-time analysis of current eye movement data, combined with a preset fixation region model, quantifying the degree and direction of fixation deviation using image processing technology, and triggering an alarm and correction when the deviation exceeds a set threshold; dynamically generating and rendering a corrective visual cue image on the visual field image based on the degree and direction of fixation deviation using image overlay and fusion technology; and dynamically adjusting the brightness, contrast, size, color, and animation attributes of the corrective visual cue image based on the duration and magnitude of fixation deviation.

4. The field-of-view detection system based on extended reality according to claim 1, characterized in that, The adaptive stimulus generation module includes: Based on visual field analysis and testing requirements, new test point images are dynamically generated and encrypted. Based on the current ambient lighting conditions, the subject's visual sensitivity, and fixation performance of the previous test point image, the brightness and contrast of the test point images are dynamically adjusted through histogram equalization, and real-time image noise reduction and enhancement processing is performed on the test point images.

5. The field-of-view detection system based on extended reality according to claim 1, characterized in that, The dynamically adjusted image module includes: segmenting the current field-of-view image based on its content, identifying regions of interest, and automatically generating and encrypting new test point images using the ORB feature point detection algorithm; optimizing the test point presentation sequence using the simulated annealing algorithm based on the visual similarity, spatial distance, and semantic association between test point images; and adjusting the sequence based on gaze point prediction by using a CNN-LSTM neural network model to predict future gaze points and adjust the presentation sequence of test point images accordingly.

6. The field-of-view detection system based on extended reality according to claim 1, characterized in that, The visual field defect identification module includes: performing image preprocessing on the acquired visual field data to form standardized visual field data; extracting image abnormality features related to visual field defects from the standardized visual field data, wherein the image abnormality features include texture abnormalities, brightness and color unevenness, geometric distortion and shape defects, edge and gradient abnormalities, and pixel-level motion abnormalities; Based on the image features and subject response data, a Bayesian network is used to predict potential defect image regions in real time, and the defect probability and confidence of the potential defect image regions are quantitatively evaluated; based on the defect probability and confidence, high-probability regions of visual field defects are identified and located.

7. The field-of-view detection system based on augmented reality according to claim 5, characterized in that, The dynamic image adjustment module further includes: for areas predicted to have a high probability of defects and suspected boundaries, dynamically inserting additional test point images to accurately locate the defect boundaries; for areas continuously predicted to be healthy and with a very low probability of defects, after confirming that there are no abnormalities in the key points, reducing the number of non-key test points in the area and skipping the test.

8. The field-of-view detection system based on augmented reality according to claim 5, characterized in that, The dynamic image adjustment module further includes: adjusting the standard test sequence, prioritizing testing of predicted high-probability defect areas and their adjacent edge areas; if high-probability defects are continuously detected in a certain area, the system pauses the full testing of the current area, prioritizes rapid screening of other key areas, and then returns to the detailed detection of the defect area after completion.

9. A field-of-view detection device based on augmented reality, characterized in that, include: The visual field detection device includes a portable glasses main body structure, which has a built-in memory and processor, and is equipped with a diopter adjustment knob to support diopter adjustment and labeling from -5D to 0D; the memory is used to store image data, model parameters and adaptive strategies; the processor is used to run image processing algorithms and Bayesian prediction models; the diopter adjustment imaging principle adopts the Birdbath scheme, that is, the Micro-LED display source is refracted and reflected by the lens group, so that the image is magnified and projected into the human eye. By adjusting the knob, the tiny distance between the display source and the lens group is controlled, and the distance of the virtual image projected into the human eye on the retina is finely adjusted.

10. A field-of-view detection method based on extended reality, characterized in that, The specific steps include: S10. Real-time acquisition of image data of the eye and environment, extraction of eye features, gaze features and environmental features from the image data, and elimination of optical distortion; S20. Analyze eye movement data in real time, quantify the degree of fixation deviation, and trigger visual cues and correction when the threshold is exceeded; S30. Based on the visual field analysis and testing requirements, dynamically encrypt and generate new test point images, and adaptively adjust the brightness and contrast of the test point images in combination with ambient lighting and the visual sensitivity of the test subject, and perform noise reduction and enhancement on the test point images. S40. Based on image features and gaze point prediction, optimize the test point presentation sequence, reduce scanning distance, automatically encrypt test points for high-probability defect areas, and quickly jump to key areas for screening when defects are detected; S50. By extracting image features related to texture, shape, edge, and pixel motion, and combining them with subject response data, a Bayesian network is used to calculate the probability of defects in each region in real time, and to accurately locate high-risk visual field defects.

Images