A display screen environment adaptive brightness adjustment method and device

Abstract

The application discloses a display screen environment adaptive brightness adjusting method and device, which fuses environmental illumination data and facial brightness data to form a brightness perception data frame, constructs an illumination distribution matrix through blink cycle shielding and scene type detection; performs illumination jump characteristic analysis on the illumination distribution matrix to identify a high-illumination jump area, determines an adjusting reference position through continuous jump attenuation priority enhancement analysis, positions a backlight adjusting node after comparing the adjusting reference position with the brightness perception data frame; determines backlight driving parameters according to the backlight adjusting node and generates a backlight optimization sequence, generates an adjusting configuration scheme through advanced multi-scene brightness fusion; determines an adjusting period based on visual masking window alignment of the adjusting configuration scheme, and executes PWM closed-loop modulation to output a brightness adjusting instruction. The application integrates multi-source perception fusion, jump identification and visual masking execution into a unified adjusting process, and improves the matching degree of display screen brightness adjustment and user visual comfort.

Description

Technical Field

[0001] This invention relates to the field of display device control technology, and in particular to a method and apparatus for adaptive brightness adjustment of a display screen based on environmental conditions. Background Technology

[0002] Laptops face particularly significant variations in lighting conditions during mobile use. Moving from an indoor office environment to the outdoors, the periodic changes in sunlight due to cloud cover, and the shifting of artificial lighting between meeting rooms and corridors all cause rapid or gradual changes in screen illuminance. Laptops' built-in ambient light sensors are typically deployed at a single point, directly triggering brightness adjustments based on real-time illuminance readings. This limited perception capability makes it insufficient to distinguish between direct light interference and gradual changes in background diffused light, resulting in adjustments that fail to accurately reflect the actual light exposure to the user's eyes.

[0003] The user's physiological state also affects the accuracy of the adjustment. Blinking and visual fatigue introduce invalid samples into the facial data collected by the built-in camera, interfering with scene recognition. Existing solutions lack a mechanism to actively eliminate this kind of physiological noise. At the adjustment execution level, the feature recognition of illumination jumps remains at the threshold judgment level, unable to distinguish between transient noise and continuous jumps. The adjustment action lags behind scene switching and does not utilize the visual masking characteristics of the human eye to select low-perceived interference moments. Users have a clear perception of the screen brightness adjustment process, affecting the mobile office and daily use experience. In addition, insufficient perception accuracy leads to the laptop screen brightness being consistently too high, and the backlight drive power consumption cannot be effectively reduced according to changes in ambient light, exacerbating unnecessary battery consumption. Summary of the Invention

[0004] This invention discloses a display screen environment adaptive brightness adjustment method and device, which aims to solve problems such as single ambient light perception dimension, insufficient recognition of illuminance jump features and lack of optimization of brightness adjustment execution timing. By establishing a complete adjustment closed loop from multi-source light feedback fusion, scene type detection, jump feature analysis, prediction fusion to visual masking window alignment, the display screen brightness can be automatically adjusted to adapt to the user's visual state.

[0005] The first aspect of this invention provides a method for adaptive brightness adjustment of a display screen based on environmental conditions, comprising the following steps: Acquire ambient illuminance data and user facial brightness data, and perform multi-source light feedback fusion based on the ambient illuminance data and the user facial brightness data to form a brightness perception data frame; A scene response sequence is constructed using the brightness perception data frame. A brightness sampling point set is extracted by blink cycle masking scene type detection of the scene response sequence. An illuminance distribution matrix is ​​constructed based on the brightness sampling point set. Illuminance abrupt change feature analysis is performed on the illuminance distribution matrix to identify high illuminance abrupt change regions. In the high illuminance abrupt change regions, continuous abrupt change attenuation priority enhancement analysis is performed to determine the adjustment reference position. The adjustment reference position is compared with the brightness perception data frame to determine the backlight adjustment node. Based on the backlight adjustment node, the PWM driver parameters are adjusted to determine the backlight driving parameters. The backlight driving parameters are used to control the generation of a backlight optimization sequence through the driving circuit. The backlight optimization sequence is then used to generate an adjustment configuration scheme by performing advanced multi-scene brightness fusion. Illumination modulation analysis is performed on the intensity change components in the ambient illuminance data to obtain modulation amplitude data. The adjustment period is determined by aligning the visual masking window of the modulation amplitude data with the adjustment configuration scheme. The brightness adjustment command is then executed according to the adjustment period using PWM closed-loop modulation.

[0006] A second aspect of the present invention provides a display screen environment adaptive brightness adjustment device, comprising: The data fusion module is used to acquire ambient illuminance data and user facial brightness data, and to perform multi-source light feedback fusion based on the ambient illuminance data and the user facial brightness data to form a brightness perception data frame; The scene detection module is used to construct a scene response sequence using the brightness perception data frame, perform blink cycle masking scene type detection on the scene response sequence to extract a brightness sampling point set, and construct an illuminance distribution matrix based on the brightness sampling point set. The reference positioning module is used to perform illuminance jump feature analysis on the illuminance distribution matrix to identify high illuminance jump regions, perform continuous jump attenuation priority enhancement analysis in the high illuminance jump regions to determine the adjustment reference position, and compare the adjustment reference position with the brightness perception data frame to determine the backlight adjustment node. The backlight control module is used to determine the backlight driving parameters by adjusting the PWM driver parameters according to the backlight adjustment node, and to generate a backlight optimization sequence by controlling the driving circuit through the backlight driving parameters. The backlight optimization sequence is then used to generate an adjustment configuration scheme by performing advanced multi-scene brightness fusion. The modulation output module is used to perform illuminance modulation analysis based on the intensity change components in the ambient illuminance data to obtain modulation amplitude data, align the adjustment period with the adjustment configuration scheme based on the visual masking window of the modulation amplitude data, and execute PWM closed-loop modulation output brightness adjustment commands according to the adjustment period.

[0007] The beneficial effects of this invention are reflected in the following points: First, by combining the multi-directional illuminance acquisition of the ambient light sensor array with the facial brightness perception of the front-facing camera to establish a light feedback fusion channel, the changes in direct light and the trend of scattered light are separated and differentially weighted onto the facial brightness data, making the brightness perception results closer to the actual light received by the user's eyes. Combined with blink timing markers, fatigue sampling periods are eliminated, and brightness sampling points are extracted differentially according to scene type. While ensuring the quality of the perception data, an illuminance distribution matrix covering various lighting environments is constructed, improving the effectiveness and scene adaptability of the brightness perception data. Second, by performing self-emission suppression gradient analysis on the temporal change distribution of the illuminance distribution matrix, the self-emission interference of the device and the real ambient light jump are distinguished. After verifying the cross-frame continuity, high illuminance jump regions are identified. Within the high illuminance jump region, the adjustment reference position of the jump energy concentration is determined through interval distribution analysis and attenuation weight reconstruction. After comparing the reference position with the brightness perception data frame, the backlight adjustment node is accurately located, ensuring that the timing of adjustment intervention is consistent with the spatiotemporal distribution of the actual illuminance jump characteristics. Finally, the PWM drive parameters are dynamically determined based on the light sensing status of the backlight adjustment node. Advanced brightness parameters are generated by predicting scene switching trends and weighted and fused with the backlight optimization sequence, ensuring that brightness adjustment is established before scene switching occurs. The masking window boundary is reconstructed using the visual masking effect caused by rapid eye movement, aligning the execution timing of the adjustment configuration scheme to a low-perceived interference window to determine the adjustment cycle. This reduces the user's subjective perception of the brightness adjustment process and improves the smoothness and timing of adjustment execution. Simultaneously, the PWM duty cycle is precisely compressed according to the ambient light state, reducing the average power consumption of the backlight drive circuit. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a display screen environmental adaptive brightness adjustment method according to the present invention.

[0009] Figure 2 This is a structural block diagram of a display screen environmental adaptive brightness adjustment device according to the present invention. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0012] The technical solutions of the embodiments of this application will be described below.

[0013] like Figure 1 As shown, this embodiment of the invention provides a display screen environmental adaptive brightness adjustment method, including the following steps S11-S15: Step S11: Obtain ambient illuminance data and user facial brightness data, and perform multi-source light feedback fusion based on the ambient illuminance data and user facial brightness data to form a brightness perception data frame.

[0014] Specifically, ambient illuminance data and user facial brightness data are acquired. Ambient illuminance data is collected by a light sensor array deployed around the display device. The sensor array covers three directions: the front, left, and right sides of the device. The results from these three directions are combined and weighted to obtain a representative value of the current ambient light intensity. The sensor in front has the highest weight because it most directly reflects the light intensity in the user's line of sight. User facial brightness data is continuously collected at a frame rate of 30 frames per second using the laptop's built-in front-facing camera. After locating the face region in each frame, two values ​​are extracted: the overall facial brightness average and the local brightness average around the pupils. The overall facial brightness reflects the overall light received by the user's face under the current lighting conditions, while the local brightness around the pupils is more sensitive to changes in ambient light than the overall facial brightness. Both ambient illuminance and user facial brightness data are timestamped. The two data streams are fused after being timestamped. Resampling is triggered when the alignment error exceeds 33ms, with 33ms corresponding to the time interval between two adjacent frames. If face detection fails for more than 5 consecutive frames during user facial brightness data acquisition, the corresponding time period is marked as a missing face. The user facial brightness data for the missing face period is filled with the value of the previous valid frame, with a filling time not exceeding 500ms. After that, the ambient illumination data drives the fusion process separately. The ambient illumination data and user facial brightness data have different dimensions. Before entering the fusion process, they are normalized to the range of 0 to 1. The normalization reference is determined by the upper limit of the range of their respective sensors.

[0015] In some embodiments, the step of performing multi-source light feedback fusion to form a brightness-sensing data frame based on the ambient illuminance data and the user's facial brightness data includes: performing illuminance spectrum decomposition on the ambient illuminance data to generate direct light components and scattered light components; using the direct light components to perform pupil region-priority illuminance weighting on the user's facial brightness data to form a composite facial brightness; performing temporal correlation fusion of the composite facial brightness and the scattered light components to generate fused light-sensing parameters; and performing frame structure grouping to form a brightness-sensing data frame based on the fused light-sensing parameters.

[0016] Illuminance spectrum decomposition is performed on ambient illuminance data to generate direct light and diffuse light components. The illuminance spectrum decomposition takes the time-domain sequence of ambient illuminance data as input and performs a short-time Fourier transform on the sequence. The transform window length is set to 1 second corresponding to 30 frames, and the step size between windows is 10 frames. The energy distribution of each frequency component within the window reveals the stability characteristics of the current illumination. The low-frequency part of the ambient illuminance data spectrum corresponds to the gradual trend of light intensity change, which mainly originates from changes in ambient background diffuse light and is extracted as the diffuse light component. The high-frequency part corresponds to the rapid fluctuation of light intensity, which originates from the instantaneous intensity change of the direct beam caused by the movement of obstructions or changes in the direction of the light source and is extracted as the direct light component. The boundary frequency between low and high frequencies is set to 0.5 Hz. Spectral energy below 0.5 Hz is classified as the diffuse light component, and energy above 0.5 Hz is classified as the direct light component. The light intensity fluctuation frequency caused by indoor lights being briefly blocked by passersby is usually higher than 1 Hz and is classified as the direct light component rather than the diffuse light component. The instantaneous amplitude of the direct light component reflects the drastic change in the intensity of the current direct light. A larger amplitude indicates stronger direct light interference, corresponding to a larger weight adjustment amplitude when weighting the pupil area. The rate of change of the amplitude of the scattered light component reflects the gradual trend of the ambient background light. A low rate of change indicates that the ambient light is in a stable state, while a persistently high rate of change suggests that the user's lighting environment is experiencing a slow and continuous intensity drift. From dusk to night, the amplitude of the scattered light component continuously declines, exhibiting a typical unidirectional gradual change pattern.

[0017] The direct light component is used to weight the user's facial brightness data based on pupil-area illuminance, forming a composite facial brightness. Visual discomfort caused by direct light illuminating the area around the pupil is far more intense than when it illuminates other areas of the face; therefore, the weighting of the direct light component prioritizes the local brightness term around the pupil in the user's facial brightness data. The instantaneous amplitude of the direct light component in the current frame is normalized and multiplied by 0.8 as the increment of the weighting coefficient for the local pupil brightness term. This increment is added to the base weighting coefficient of 1.0, resulting in a final weighting coefficient range of 1.0 to 1.8. The upper limit of 1.8 prevents excessive weighting of the pupil area when direct light is extremely strong, which could distort the weighting result. The weighting coefficient for the overall facial brightness term is complementary to that of the local pupil brightness term, and the sum of the two coefficients remains at 2.0. Increasing the weighting coefficient for the local pupil brightness term leads to a corresponding decrease in the overall facial brightness term coefficient. In a certain frame, when the normalized amplitude of the direct light component is 0.6, the local pupil coefficient is 1.48 and the overall facial brightness coefficient is 0.52. The composite facial brightness is obtained by weighted averaging of two weighted brightness values. The weighted average is calculated by dividing the sum of the two weighted values ​​by the sum of the two coefficients. The weighted average reflects the effective brightness level that has the greatest impact on user visual comfort in the current frame. When the amplitude of the direct light component is zero, both coefficients degenerate to 1.0, and the composite facial brightness degenerates to the average facial brightness. For frames with missing facial filler in the user's facial brightness data, composite facial brightness is marked with filler labels. These labels indicate that the light perception information at that moment comes from interpolation rather than actual measurement. When the user briefly turns to the side, causing the camera to temporarily lose sight of the face, the corresponding segment's composite facial brightness enters the filler label state.

[0018] The synthesized facial brightness and the scattered light component are temporally correlated and fused to generate fused light perception parameters. Temporal correlation fusion uses the Pearson correlation coefficient between the two signals within the most recent 1-second window as the basis for weight adjustment. The correlation coefficient reflects the synchronicity between the fluctuation rhythm of background scattered light and the measured facial brightness change. High synchronicity indicates that the current facial brightness is mainly dominated by ambient background light, and the scattered light component receives a higher weight in the fusion. When the two signals diverge, it indicates that the facial brightness is independently driven by direct light or local light sources, and the weight of the synthesized facial brightness is correspondingly increased. When users use the device in overcast diffuse light environments, the rhythms of the two signals are similar, and the correlation is usually high. Upon entering an area with direct sunlight, the sunlight first impacts the pupil area, causing a rapid increase in facial brightness, while the scattered light component has not yet responded synchronously. A significant phase difference appears between the two signals, and the correlation coefficient decreases accordingly. The fusion weight shifts towards the measured facial brightness side, making the fused light perception parameters closer to the user's current actual visual lighting state rather than the average level of ambient background light. The fused light sensing parameters are obtained by linearly superimposing the weighted values ​​of the scattered light components and the weighted values ​​of the facial composite brightness. The sum of the two weights is 1. The weight allocation is driven by the correlation coefficient to update each frame in real time. Frames with padding annotations in the facial composite brightness are excluded when calculating the correlation coefficient of the corresponding time domain window. When the number of effective frames is less than 15, the values ​​of the previous effective window are used. During this period, low confidence annotations are added to the fused light sensing parameters.

[0019] The luminance sensing data frame is formed by grouping the fused light sensing parameters into a frame structure. The frame structure grouping uses the temporal sequence of the fused light sensing parameters as the organizing object, dividing the fused light sensing parameter values ​​within a continuous time window into fixed-length frame units according to the frame rate. The duration of each frame unit is aligned with the frame rate of the front-facing camera, set to 33ms. This uniformity in frame unit duration ensures that the temporal resolution of the luminance sensing data frame is strictly consistent with the camera's acquisition rhythm. Within each frame unit, the mean describes the steady-state light sensing level, the maximum value records the instantaneous peak light sensing intensity, and the variation amplitude characterizes the dynamic fluctuation range within the frame. These three values ​​are combined to form the frame description vector for that frame unit. Each frame unit in the luminance sensing data frame is accompanied by a low-confidence annotation of the fused light sensing parameters at the corresponding time. The low-confidence frame unit identifier originates from uncertain light sensing segments where facial filling is missing, and the light sensing estimation accuracy is lower than that of normal acquisition frames. Frames whose changes in ambient light parameters exceed 50% of the intra-frame mean are marked as high dynamic range (HMR) frames. HMR frames are annotated with dynamic markers in the brightness perception data frames, indicating a rapid change in light intensity at the corresponding moment. When a user suddenly moves from a stable indoor environment to direct sunlight, multiple consecutive frames in the brightness perception data frame will show HMR markers. Brightness perception data frames are generated sequentially in chronological order, with no gaps between adjacent frame units. The frame sequence completely covers the entire light perception history from the start of data acquisition to the current moment, ensuring that all types of light evolution are fully recorded. When the mean component in the brightness perception data frame remains consistently low, it indicates that the current ambient light is in an energy-saving and compressible state. A typical scenario is when a user spends a long time in a dimly lit room or uses the device at night; the consistently low mean component reflects a margin for brightness reduction in such environments.

[0020] Step S12: Construct a scene response sequence using brightness perception data frames, perform blink cycle masking scene type detection on the scene response sequence to extract brightness sampling point set, and construct an illuminance distribution matrix based on the brightness sampling point set.

[0021] Specifically, a scene response sequence is constructed using brightness-sensing data frames. The frame description vectors of each frame unit in the brightness-sensing data frames are read sequentially in chronological order. A continuous temporal signal is constructed using the mean component of the frame description vector as the main axis. The difference in the mean of adjacent frame units reflects the inter-frame rate of change in light perception level; a consistently positive rate of change indicates an upward trend in light perception level, while a consistently negative rate indicates a downward trend. The scene response sequence is based on the frame sequence of the brightness-sensing data frames. Two attributes, inter-frame rate of change and trend direction, are added to the frame description vector of each frame unit. The inter-frame rate of change is obtained by normalizing the difference in the mean of adjacent frames by the frame duration. The trend direction is determined by the sign mode of the rate of change in the most recent 5 frames. The weight of the inter-frame rate of change at the corresponding position in the scene response sequence for low-confidence labeled frame units in the brightness-sensing data frames is reduced to 0.5. The trend direction attribute of low-confidence frame units uses the result of the previous non-low-confidence frame to avoid data padding interfering with trend judgment. In scene response sequences, the absolute value of the rate of change of highly dynamic labeled frame units is usually large. A concentrated occurrence of highly dynamic frames indicates a continuous and drastic change in the current lighting environment. When a user moves from indoors to outdoors, highly dynamic labels appear consecutively in the brightness perception data frames, corresponding to a sustained high rate of change in the corresponding scene response sequence segment. The length of the scene response sequence is the same as the length of the brightness perception data frame sequence, and it is appended frame by frame in real time. When a new frame unit is added, the trend direction attribute of the most recent 5 frames is updated synchronously.

[0022] In some embodiments, the step of performing blink cycle masking scene type detection and extracting a brightness sampling point set on the scene response sequence includes: performing facial action recognition on the scene response sequence to obtain blink timing markers; performing high-frequency blink fatigue state removal on the scene response sequence based on the blink timing markers to generate an effective sampling sequence; performing scene type classification on the effective sampling sequence to obtain scene classification results; and performing sampling point extraction based on the scene classification results to form a brightness sampling point set.

[0023] Facial motion recognition is performed on the scene response sequence to obtain blink timing markers. Facial motion recognition uses the front-facing camera image at the corresponding moment of each frame unit in the scene response sequence as the analysis object. The eye region is extracted from the face detection results corresponding to each frame of the scene response sequence. The cropping range is based on the line connecting the centers of the two eyes, extending upwards and downwards by 20% of the face height. The degree of eye opening and closing is measured by the ratio of the vertical pixel span to the horizontal pixel span of the eye. When the ratio is less than 0.15, it is judged as a closed eye state. A complete blink event is judged when the closed eye state lasts for more than 2 frames and there is one open eye state before and after it. A single frame of closed eye does not constitute a blink event to exclude frame rate jitter misjudgment. In the blink timing marker, the start and end frame positions define the elimination boundaries, while the duration of frames serves as the basis for judging micro-blinks and tightening the fatigue threshold. The duration of frames reflects the duration of this blink; the normal blink duration is usually between 100 and 400 ms, corresponding to 3 to 12 frames. Eye-closing events exceeding this range are not included in the blink timing marker but are recorded separately as abnormal eye-closing. In the scene response sequence, frames where face detection fails cannot extract the eye region, resulting in missing blink timing markers at the corresponding locations. These missing periods are handled conservatively during the fatigue state elimination phase, being marked as unverified segments rather than being directly eliminated. The distribution of the interval between two adjacent blink events in the blink timing marker reflects the user's current blink rhythm. Shorter and more concentrated intervals indicate that the user's eyes are in a state of continuous stress. When users stare at a high-brightness screen for a long time, the blink interval is significantly compressed, and the median interval of the corresponding segment in the blink timing marker is significantly lower than the normal level.

[0024] Based on blink timing markers, high-frequency blink fatigue states are eliminated from scene response sequences to generate valid sampling sequences. The determination of high-frequency blink fatigue uses the number of blinks per unit time as the core indicator, with a statistical window set at 60 seconds. A blink count exceeding 25 times within the window is considered a high-frequency blink fatigue state, corresponding to a blink frequency exceeding the upper limit of the normal range per minute. Blink events with abnormally short durations in the blink timing markers are marked as microblinks, defined as blinks with a duration of less than 3 frames. When microblinks occur frequently, they are counted separately as a fatigue determination indicator. When the frequency of microblinks exceeds twice the normal blink frequency, the fatigue determination threshold is lowered to 20 times per minute. This tightening of the threshold reflects the increased eye fatigue experienced by users under frequent microblinking conditions. When users continuously use the device in strong direct light environments, the ratio of microblinks to normal blinks remains consistently high, corresponding to earlier fatigue determination within the statistical window. In the scene response sequence, the frame units corresponding to the time periods identified as high-frequency blinking fatigue are removed from the valid sampling sequence. The removal range covers the period from the start of the fatigue state to 3 seconds after the state is resolved. The 3-second buffer ensures that unstable brightness perception data in the initial stage of fatigue recovery does not enter the scene classification. If the verification segment corresponding to the missing time period in the blink timing marker is in a normal state before and after it, it is retained and included in the valid sampling sequence. If either side is in a fatigue state, it is removed along with the missing time period. Each frame unit in the valid sampling sequence retains the original frame description vector and trend attributes of the scene response sequence. The removal operation does not change the content of the frame unit. If a user blinks frequently due to eye stimulation in a strong light environment, the valid sampling sequence after removing the corresponding time period only retains the frame units of the time period when the eye state is stable.

[0025] For example, the step of classifying the scene type of the effective sampled sequence to obtain the scene classification result includes: performing illuminance dynamic range statistics on the effective sampled sequence to generate dynamic range distribution features; performing cross-frame illuminance interpolation completion based on the dynamic range distribution features to identify strong light direct glare sections, low light sections at night, and rapid light change sections to generate three types of scene labels; performing confidence evaluation on the three types of scene labels to generate a scene confidence sequence; and performing optimal scene classification based on the scene confidence sequence to obtain the scene classification result.

[0026] Illuminance dynamic range statistics are performed on the effective sampling sequence to generate dynamic range distribution characteristics. The dynamic range of illumination is calculated as the difference between the maximum and mean components of the frame description vector for each frame unit in the effective sampling sequence. The formula is DR_f = V_max - V_mean, where DR_f is the dynamic range of a single frame, V_max is the maximum component of the frame description vector, and V_mean is the mean component. A larger range indicates the presence of localized strong light regions or rapid light pulses within that frame. The dynamic range distribution of all frame units within a 30-frame sliding window in the effective sampling sequence is statistically analyzed. Three statistical parameters are extracted: the mean, maximum, and cumulative number of frames exceeding the threshold. The cumulative number of frames exceeding the threshold is defined as the number of frame units within the window whose DR_f exceeds 0.3. A higher number of frames exceeding the threshold indicates a higher frequency of light fluctuations within that window. The dynamic range distribution feature consists of three statistics arranged in window time order, reflecting the evolution trend of lighting dynamics over time. Window segments with a consistently high mean dynamic range correspond to an overall unstable lighting environment, while window segments with a consistently low mean correspond to an overall stable lighting environment. Low-confidence labeled frame units in the effective sampling sequence do not participate in the calculation of the three statistics of the sliding window. When the number of effective frames in the window containing the low-confidence frame unit decreases, the accuracy of the statistics decreases accordingly. When the number of effective frames is less than 60% of the window length, the dynamic range distribution feature of that window is labeled with low samples. Low-sample windows are downweighted during the generation of the three types of scene identifiers. The rate of change of the number of frames exceeding the threshold in the dynamic range distribution feature from low to high reflects the accelerating trend of lighting dynamics. When the accelerating trend is consistently high, it corresponds to the lighting environment where the user is located transitioning from a stable state to a frequently fluctuating state. During the process of the user moving from an indoor corridor to an outdoor plaza, this accelerating trend value continuously increases within multiple consecutive windows.

[0027] Based on dynamic range distribution characteristics, cross-frame illuminance interpolation is used to identify three types of scene labels: strong direct light segments, low-light segments at night, and segments with rapid light changes. Cross-frame illuminance interpolation is performed on low-sample labeled windows in the dynamic range distribution characteristics. The statistics of the three most recent valid windows on both sides of the low-sample window are linearly interpolated, and the completed statistics replace the original values ​​of the low-sample window in the segment identification. Interpolation avoids data gaps that may cause scene boundary judgment errors. Window segments in the valid sampling sequence with a frame description vector mean component consistently higher than 0.6, a dynamic range mean consistently lower than 0.15, and a cumulative number of frames exceeding the threshold less than 5 are identified as strong direct light segments. A high mean component indicates an overall strong light level, while a low dynamic range mean and a small number of frames exceeding the threshold indicate stable and continuous strong light rather than pulse-like fluctuations. The combined judgment of these three conditions excludes the possibility of short-lived strong light pulses being misidentified as strong direct light segments. Window segments with a frame description vector mean component consistently below 0.25, a dynamic range mean consistently below 0.2, and a near-zero number of frames exceeding the threshold are identified as low-light nighttime segments. When users operate the device in dark environments, the dynamic range distribution characteristics of each window have a mean value below 0.15 for an extended period, and the entire segment is classified as low-light nighttime. Segments with a cumulative number of frames exceeding the threshold exceeding 15 within a 30-frame window and an absolute value of the dynamic range mean change rate exceeding 0.1 per window are identified as fast-changing light segments. The dual conditions of high-frequency exceeding the threshold and rapid change rate ensure that the identified fast-changing light segments correspond to real drastic changes in lighting rather than stable high-dynamic scenes. The three scene categories are labeled on a window-by-window basis. When the same window simultaneously meets multiple identification conditions, the highest priority label is selected based on the following order: fast-changing light, direct strong light, and low-light nighttime. Fast-changing light has the most urgent impact on backlight adjustment and therefore has the highest priority.

[0028] Confidence assessments are performed on three types of scene markers to generate scene confidence sequences. The confidence assessment is based on the marker result for each window in the three scene marker categories, using the degree of conformity between the statistics relied upon by the marker and the corresponding recognition conditions as the evaluation criterion; the higher the degree of conformity, the higher the confidence. The confidence of the strong light direct illumination segment marker is jointly determined by the magnitude of the frame description vector mean component exceeding 0.6 and the margin of the cumulative number of frames exceeding the threshold being less than 5 frames. The higher the mean component is above 0.6 and the less than 5 frames exceed the threshold, the higher the confidence of the strong light direct illumination segment marker. The confidence of the nighttime low light segment marker is jointly determined by the magnitude of the frame description vector mean component below 0.25, the magnitude of the dynamic range mean below 0.2, and the degree to which the number of frames exceeding the threshold is close to zero. The lower the mean component is below 0.25, the lower the dynamic range mean is below 0.2, and the closer the number of frames exceeding the threshold is to zero, the higher the confidence of the nighttime low light segment marker. The confidence level of the fast light change segment identifier is jointly determined by the amplitude of the number of frames exceeding the threshold by more than 15 and the absolute value of the rate of change of the mean dynamic range. The confidence level of the fast light change identifier is the highest when both values ​​are far above the threshold. The confidence level of each window in the scene confidence sequence ranges from 0 to 1. The value directly drives the voting weight allocation of the weighted classification and the downweighting of low-confidence windows. The confidence level of the low-sample labeled window is multiplied by a reduction factor of 0.7 based on the evaluation result. The reduction reflects the impact of the uncertainty of the interpolated data on the reliability of the identifier. Window identifiers with a confidence level below 0.4 in the scene confidence sequence are marked as low-confidence identifiers. Low-confidence identifiers reflect the insufficient matching degree between the lighting characteristics of the window and the recognition conditions. Windows during periods of alternating strong light and cloud shadows generally have low confidence levels because the statistics continue to oscillate around the threshold.

[0029] The optimal scene classification result is obtained by performing optimal scene classification based on the scene confidence sequence. Optimal scene classification uses the confidence value of each window in the scene confidence sequence as the sorting criterion. A weighted vote is performed on the label distribution within consecutive window segments, with each window's vote weight equal to its confidence value. The scene type with the highest weighted vote is taken as the optimal classification label for that segment. The division of consecutive window segments is based on the continuity and consistency of label types. Adjacent windows with the same label are grouped into the same segment; if the labels are different, they are split into new segments at the point of difference. If a segment is less than three windows long, it is merged with adjacent segments with higher confidence. Low-confidence label windows in the scene confidence sequence are replaced by the highest-confidence labels within each of the three adjacent windows in the voting process. If all three adjacent windows are low-confidence, the original label is retained and an "unconfirmed" label is added. Scene classification results are output in segments. Each segment includes three items: scene type identifier, segment start and end window number, and confidence-weighted average. The scene type identifier directly determines the subsequent sampling density strategy: one point is sampled every 3 frames for segments with rapid light changes, and one point is sampled every 10 frames for segments with strong direct light and low light at night. The segment start and end window number defines the distribution position of various lighting conditions on the time axis, which is used to predict the scene switching trend of the backlight optimization sequence and locate the historical scene type. The higher the confidence-weighted average, the more reliable the scene type determination of the segment. When it is lower than 0.5, the corresponding sampling density is halved to reduce unreliable sampling points from entering the illuminance distribution matrix. Segments to be confirmed and labeled are processed according to the type of adjacent high-confidence segments and do not trigger special sampling logic separately.

[0030] Based on the scene classification results, sampling points are extracted to form a brightness sampling point set. Sampling point extraction is based on the scene type and confidence-weighted average of each segment in the scene classification results as the sampling strategy decision. Different scene types correspond to different sampling densities. The sampling density is highest in fast-light-change segments to capture key light-sensing nodes during rapid light changes, while the sampling density is moderate in strong direct light segments and low-light night segments. In fast-light-change segments, one sampling point is extracted every 3 frames. The sampling point is taken as the mean component of the corresponding frame unit's frame description vector. When the difference between adjacent sampling points exceeds 0.2, an intermediate sampling point is added between the two points, and the mean of the two ends is taken for the supplementary sampling point. In strong direct light segments and low-light night segments, one sampling point is extracted every 10 frames. Sampling points are forcibly extracted in scene segment boundary frames to mark scene transition locations. Boundary frame sampling points are marked with boundary labels to distinguish sampling points within the scene. Each sampling point in the brightness sampling point set carries a scene type identifier for its respective segment, a sampling frame number, and a corresponding mean component value. The scene type identifier ensures that sampling points of the same scene type are grouped into the same row index during matrix construction, guaranteeing temporal consistency within the row. In the scene classification results, the sampling density of segments with a confidence-weighted mean below 0.5 is reduced by 50% based on the corresponding strategy. Sampling in low-confidence segments is reduced to avoid introducing too many interfering sampling points due to unreliable scene judgments. Sampling points from low-confidence segments in the brightness sampling point set are marked with low-confidence labels. The total number of sampling points in the brightness sampling point set dynamically changes with scene complexity; the total number of sampling points is larger when there are many rapidly changing light segments, and smaller when the scene is mainly stable.

[0031] An illuminance distribution matrix is ​​constructed based on the luminance sampling point set. The sampling points in the luminance sampling point set are grouped according to scene type. Sampling points of the same scene type are grouped into the same group. The sampling points within each group are arranged in ascending order of sampling frame number. The order of frame number arrangement ensures that the temporal relationship within each group is preserved. The illuminance distribution matrix uses scene type as the row index and time window as the column index. Each matrix element is the weighted average of the mean components of all luminance sampling points within the corresponding time window for that scene type. The calculation formula is M(s,t)=Σ(w_i×p_i) / Σw_i, where M(s,t) is the matrix element for scene type s within time window t, p_i is the mean component of the i-th sampling point of that scene type within the window, and w_i is the weight of the corresponding sampling point. Low-confidence labeled sampling points use w_i=0.5, and normal sampling points use w_i=1.0. The time window length is consistent with the sliding window length in the scene response sequence, which is 30 frames. When a scene type has no sampling points within a specific time window, the corresponding matrix element is filled with a null value. When the number of low-confidence labeled sampling points in the luminance sampling point set exceeds 50% of all sampling points for the corresponding scene type within that window, a low-confidence label is added to the matrix element. The method for filling in missing elements in the illuminance distribution matrix is ​​to take the linear interpolation of the two nearest non-empty elements in the same row. If there are less than two non-empty elements in a row, the average value of elements from other scene types in the same column is used as the replacement value, with cross-scene filling annotations added. The numerical distribution of each row in the illuminance distribution matrix reflects the trend of illuminance level change over time under the corresponding scene type. Monotonically increasing values ​​within a row indicate that the light intensity of that scene type is continuously rising, while oscillating values ​​within a row indicate that the light intensity is fluctuating repeatedly. The rate of change between columns reveals the magnitude of the jump or drop in illuminance between adjacent time windows. The rate of change between columns in the row of scenes with rapid light changes is usually much higher than that in the rows of strong direct light and weak light at night. The difference in the rate of change between columns in the three types of scene rows constitutes the basic discriminant information for identifying illuminance jump regions.

[0032] Step S13: Analyze the illuminance jump characteristics of the illuminance distribution matrix to identify high illuminance jump regions. Perform continuous jump attenuation priority enhancement analysis in the high illuminance jump regions to determine the adjustment reference position. Compare the adjustment reference position with the brightness perception data frame to determine the backlight adjustment node.

[0033] In some embodiments, the step of analyzing the illuminance jump feature of the illuminance distribution matrix to identify high illuminance jump regions includes: establishing a temporal illuminance change distribution through the illuminance distribution matrix to generate temporal illuminance data; performing self-illumination suppression jump gradient analysis based on the temporal illuminance data to determine illuminance abrupt change regions; performing cross-frame persistence verification on the illuminance abrupt change regions to form persistent jump bands; and extracting and identifying high illuminance jump regions based on the persistent jump bands.

[0034] Illuminance distribution matrix is ​​used to establish the temporal distribution of illuminance changes, generating time-series illuminance data. The extraction of the difference between adjacent illuminance snapshots is performed independently, row by scene type. The change sequences for each scene type are independent and not merged. The sign of the difference reflects the direction of illuminance change, and the absolute value reflects the magnitude of change. The time-series illuminance data uses time windows as indices, recording the amount and direction of illuminance change for each scene type within each time window. Windows with larger absolute values ​​in the time-series illuminance data indicate more drastic illuminance changes within that time window. Matrix elements with cross-scene filled labels are added to the illuminance distribution matrix, corresponding to the change in the time-series illuminance data for that window, but their weight is reduced to 0.6. The introduction of these filled elements reduces the reliability of the illuminance change estimation for that window; the weight reduction reflects this uncertainty. In time-series illuminance data, the change sequence of rapid light change scenes typically exhibits high-frequency oscillations, contrasting sharply with the gradual changes in strong direct sunlight and low-light nighttime scenes. When users move from a stable indoor environment to a directly sunny area, the change sequence of rapid light change scenes in the time-series illuminance data shows continuous alternation of positive and negative values ​​with persistently large absolute values ​​during the transition period. The difference in change patterns among the three types of scene rows is the main basis for distinguishing the types of illuminance jumps. Low-confidence inherited labels are added to the time-series illuminance data changes corresponding to the low-confidence labeled matrix elements in the illuminance distribution matrix. The abrupt change judgment threshold for the windows corresponding to the low-confidence inherited labels is appropriately relaxed to avoid missing effective abrupt change areas due to insufficient data quality. The continuous judgment range is appropriately expanded for periods with a high concentration of low-confidence windows to ensure coverage integrity.

[0035] Illumination abrupt change gradient analysis was performed based on time-series illuminance data to identify illuminance abrupt change regions. In the time-series illuminance data, when self-illumination interference intervenes, the changes in the three scene rows respond highly synchronously. However, changes in real ambient light typically only affect some scene rows. When the Pearson correlation coefficient of the changes in the three scene rows exceeds 0.85, it is identified as self-illumination synchronous interference. The time-series illuminance data changes in the interference window, after deducting the mean component, are used in gradient calculation. The abrupt change gradient G is determined by the ratio of the absolute value of the change in the current window to the mean of the absolute values ​​of the changes in adjacent windows, i.e., G = |ΔL_t| / mean(|ΔL_{t-1}|,|ΔL_{t+1}|), where ΔL_t is the illuminance change in the current time window, and ΔL_{t-1} and ΔL_{t+1} are the illuminance changes in the previous and next windows, respectively. When G exceeds 2, the window is identified as a gradient anomaly. The scene type identifier and time window number of the gradient anomaly are recorded as abrupt change candidate points. Illuminance abrupt change regions consist of segments where gradient anomalies appear in two or more consecutive adjacent windows. Isolated gradient anomalies in a single window are not considered illuminance abrupt change regions. Isolated anomalies usually originate from transient optical noise and lack segmental characteristics. The brief flickering of streetlights and continuous changes in ambient light exhibit drastically different behaviors on the G-series; the former forms isolated peaks, while the latter forms continuous over-threshold segments. The difference in segment length between these two types of features is the primary basis for distinguishing between noise and genuine abrupt changes. The average gradient value of each window within the illuminance abrupt change region serves as the intensity representative value for that region. This intensity representative value reflects the average severity of illuminance changes within the region; a higher intensity representative value indicates more significant illuminance fluctuations and a more concentrated impact on visual adaptation.

[0036] Illuminance abrupt change regions are validated across frames to form persistent transition zones. The consistent direction of change distinguishes between real-world ambient light evolution and random noise disturbances. A sustained unidirectional change means that illuminance evolves in a definite direction within that region, having a far more significant impact on user visual adaptation than repeated fluctuations. In illuminance abrupt change regions, directional consistency requires that the sign of the change direction be the same for more than 70% of the time windows. Abrupt change regions with an intensity representative value exceeding 3 and passing directional consistency validation are upgraded to persistent transition zones. Regions with an intensity representative value below 3, even if passing directional consistency validation, are not upgraded but are marked as weak abrupt change regions for future use. Weak abrupt change regions are re-evaluated as supplementary candidates when the number of persistent transition zones is insufficient. When the illuminance abrupt change zone spans multiple scene type rows, each scene type row undergoes independent continuous verification. If a scene type row passes verification while other rows fail, the continuous change zone is marked as a single-scene dominant change. The backlight adjustment range corresponding to the single-scene dominant change only covers the brightness range of the dominant scene type. If the row with strong direct light passes verification while the row with weak nighttime light fails, the continuous change zone only drives the adjustment parameters corresponding to the strong light scene, avoiding the erroneous extension of the strong light change's impact to other scene types. When the low-confidence inherited annotation window in the continuous change zone exceeds 30% of the total number of windows in the abrupt change zone, an overall low-confidence annotation is added, and the recognition confidence of the low-confidence continuous change zone is correspondingly reduced. The larger the time span of the continuous change zone, the longer the duration of the illuminance change. Continuous change zones with larger time spans have higher intensity representative values ​​compared to the benchmark when extracting the change region, and the possibility of upgrading to a high illuminance change region is correspondingly increased.

[0037] High-illuminance transition regions are identified based on continuous transition zones. The descending order of intensity representative values ​​determines the order in which continuous transition zones enter the extraction process, prioritizing transitions with higher urgency for backlight adjustment. Sections within a continuous transition zone whose intensity representative value exceeds the average intensity of all continuous transition zones plus one standard deviation are identified as high-illuminance transition regions. Continuous transition zones below this threshold are classified as low-illuminance transition regions. Continuous transition zones in low-illuminance transition regions are re-evaluated as supplementary candidates when the number of high-illuminance transition regions is insufficient. The time range of high-illuminance transition regions is defined by the start and end time windows of the continuous transition zones, extending one time window at each end. This buffer window captures the transitional light changes at the edges of the high-illuminance transition regions. When a scene transitions from stable indoor lighting to strong outdoor light, the illuminance increase characteristics of the transition section fall precisely within the buffer window. Without a buffer, this transition information is truncated, causing the judgment of the adjustment reference position to deviate from the effective range of the true transition region. When a sustained transition zone marked with low confidence is upgraded to a high-intensity transition region, a low-confidence inherited label is added. The threshold for determining a low-confidence high-intensity transition region is relaxed to retain more transition signals. The degree of relaxation is positively correlated with the low confidence level to avoid introducing noise interference due to over-relaxation. When the boundary interval between two adjacent high-intensity transition regions is less than two time windows, they are merged into the same region. The intensity representative value after merging is the weighted average of the two regions' time spans. The merging rules prevent the same transition event from being split into two independent regions due to data quality fluctuations in individual windows. After merging, the boundary of the high-intensity transition region is recalculated based on the total time span after merging. The rule of extending each end by one buffer window also applies after merging.

[0038] In some embodiments, the step of performing a continuous jump attenuation priority enhancement analysis in the high illuminance jump region to determine the adjustment reference position includes: performing jump direction consistency statistics on the high illuminance jump region to obtain a continuous jump sequence; performing cumulative attenuation weight analysis based on the continuous jump sequence to generate a jump priority distribution; performing peak location on the jump priority distribution to obtain priority extreme value coordinates; and determining the adjustment reference position based on the priority extreme value coordinates.

[0039] Continuous jump sequences are obtained by statistically analyzing the consistency of jump direction in high-illuminance jump regions. The consistency of the sign of the change direction is the core criterion for distinguishing between sustained jumps and random fluctuations. When the sign of the change direction is the same for three or more consecutive time windows, the segment is identified as a jump direction consistency event. The start and end window numbers and direction signs of the event are recorded as a jump event description. The more consecutive windows, the higher the intensity of the jump in that direction. The continuous jump sequence consists of the descriptions of all jump direction consistency events within the high-illuminance jump region arranged chronologically. Positive direction events correspond to a continuous increase in illuminance, and negative direction events correspond to a continuous decrease in illuminance. Alternating positive and negative events indicate that the illuminance within the high-illuminance jump region is in an oscillating change state. The interference mechanism of oscillation mode and unidirectional jump mode on user visual adaptation is different. In oscillation mode, brightness adjustment needs to consider the response requirements of both directions. Isolated windows with inconsistent directions within a high-illuminance jump region are not included in any events with consistent jump directions. These isolated windows are marked with intervals in a continuous jump sequence; a greater number of intervals indicates a more unstable jump direction within the high-illuminance jump region. Regions with low stability typically generate lower priority peaks than stable unidirectional jump regions. Low-confidence inherited high-illuminance jump regions correspond to event descriptions in the continuous jump sequence with added low-confidence labels. The weights of these low-confidence labeled events are reduced, with the reduction degree positively correlated with the original low-confidence level. After reduction, the weights of relatively reliable events become more prominent in the continuous jump sequence, and the interference of isolated, extremely low-confidence events on the overall priority distribution is correspondingly suppressed.

[0040] For example, the step of generating a jump priority distribution based on the cumulative attenuation weight analysis of the continuous jump sequence includes: performing interval distribution analysis on the continuous jump sequence to generate short-interval dense components and long-interval sparse components; using the short-interval dense components to modulate the attenuation intensity of the long-interval sparse components to form a modulation attenuation sequence; reconstructing the attenuation weight distribution feature through the modulation attenuation sequence; and performing jump intensity hierarchical analysis on the attenuation weight distribution feature to generate a jump priority distribution.

[0041] Interval distribution analysis of continuous jump sequences yields short-interval dense components and long-interval sparse components. The time interval between two adjacent jump events with the same direction in the continuous jump sequence is divided into short and long intervals using the median as the cutoff value. This median cutoff adapts to the actual differences in the interval distribution range under different illumination scenarios, avoiding classification bias when the fixed threshold is too long or too short overall. Intervals below the median are classified as short intervals, and intervals above the median are classified as long intervals. The short-interval dense component consists of all short intervals in the continuous jump sequence and the jump events with the same direction on either side. The jump events within the dense component are closely time-dependent; the next jump begins before the effect of the previous jump has fully decayed. The decay effect creates energy superposition in the dense component region. When users walk in corridors with frequent changes in light and dark, the intervals of the continuous jump sequence are generally short, and all event pairs are classified into the short-interval dense component, exhibiting an overall dense superposition characteristic. The long-interval sparse component consists of all long intervals and their corresponding time segments in the continuous jump sequence. Sufficient calm periods exist between jump events within the sparse component, meaning the impact of the previous jump has largely dissipated before the next jump begins, and the decay processes of each jump event are independent and do not interfere with each other. The event density of the short-interval dense component is determined by dividing the number of events within the dense component segment by the total number of windows in the segment. Higher density indicates more frequent jump events in that segment. Segments with significantly higher density typically form prominent peak regions in the jump priority distribution, while segments with lower density correspond to dispersed jump events and gentler peaks.

[0042] A modulation attenuation sequence is formed by modulating the attenuation intensity of long-interval sparse components using short-interval dense components. The continuous energy replenishment of successive jumps in the dense segment slows down the attenuation rate of the preceding jump. This mechanism ensures that the attenuation rate of isolated jump events in the sparse segment is also modulated if they are within the influence extension range of the dense segment. The event density of the short-interval dense component is normalized and used as the modulation coefficient M_mod. M_mod is multiplied by the original attenuation coefficient of each jump event in the long-interval sparse component to obtain the modulated attenuation coefficient, i.e., α_adjusted = α_original × M_mod, where α_original is the basic attenuation coefficient, and the value of M_mod ranges from 1.0 to 1.1. The upper limit of the modulated coefficient is limited to 0.99 to prevent the coefficient from approaching 1.0 and causing the weight to accumulate infinitely. The modulation attenuation sequence records the attenuation coefficient sequence of each abrupt event in the long-interval sparse component after modulation. Events with larger modulation coefficients have slower weight attenuation and a wider influence coverage on the time axis. Isolated abrupt events in sparse segments acquire longer influence trails after modulation, effectively filling the weight gaps between adjacent events. When a user experiences occasional strong light exposure in a uniformly lit environment, the weight trail of the corresponding isolated abrupt event extends into the adjacent calm period after modulation, thus reasonably preserving the priority influence of the event. In the short-interval dense component, the attenuation coefficients of each abrupt event are directly retained as their original values ​​without modulation. The dense component itself exhibits a superposition effect through the accumulation of multiple abrupt weights. These two mechanisms work together in the modulation attenuation sequence to affect the final weight distribution, synergistically ensuring that the weight distribution can accurately respond to various interval patterns.

[0043] Attenuation weight distribution features are generated by reconstructing the weight distribution using a modulation attenuation sequence. The overlap of weights of adjacent events in densely packed segments on the time axis is the direct cause of superposition peaks. In the modulation attenuation sequence, the modulated attenuation coefficient replaces the original value to perform attenuation calculations within each event window. The increase in the tail weights increases the overlap area between the next and previous events on the time axis, and the superposition of weights in the overlapping area forms a local peak. The attenuation weight distribution features are constructed by arranging the weight values ​​of each time window after reconstruction from the modulation attenuation sequence according to window number. Windows with weight values ​​exceeding the average weight of all windows plus one standard deviation are marked as weight high points. The distribution density of weight high points reflects the concentration of abrupt energy on the time axis. The concentrated appearance of high points indicates that this period is the interval where the illuminance change has the strongest visual impact on the user. When a user suddenly enters a bright outdoor environment from a dimly lit room, a dense cluster of weight high points will appear in the corresponding time period in the attenuation weight distribution features. This segment also has a higher probability of being classified into the high-intensity layer when the abrupt intensity is stratified. In the continuous jump sequence, the reconstruction weights of the windows corresponding to low-confidence labeled events are distinguished by gray labels in the attenuation weight distribution feature. Gray-labeled windows participate in the calculation of mean and standard deviation in the weight distribution obtained from the modulation attenuation sequence but do not trigger the high-intensity layer determination. Zero-weight windows correspond to the interval label positions of the continuous jump sequence and are uniformly assigned to the lowest intensity layer during intensity stratification. The overall weight distribution pattern of the attenuation weight distribution feature reflects the spatial organization of jump energy in the current high-illuminance jump region. A distribution pattern biased towards a single peak indicates that the jump energy is highly concentrated, while a distribution pattern with multiple peaks indicates that there are multiple independent jump aggregation centers in the region. For peak location, for single peaks, the global maximum point is taken, while for multiple peaks, candidate extreme values ​​are extracted according to each aggregation center.

[0044] A stratified analysis of abrupt change intensity is performed on the attenuation weight distribution characteristics to generate a priority distribution of abrupt changes. The quartile segmentation method does not rely on a fixed threshold and can adapt to the differences in the distribution range of weight values ​​within different high-illuminance abrupt change regions. Windows above the third quartile are assigned to the high-intensity layer, those between the median and the third quartile are assigned to the mid-to-high-intensity layer, those between the first quartile and the median are assigned to the mid-to-low-intensity layer, and those below the first quartile are assigned to the low-intensity layer. Gray-marked windows in the attenuation weight distribution characteristics are not involved in the quartile calculation; intensity layer classification is based on the results of adjacent non-gray windows. When the intensity layers of adjacent windows are inconsistent, they are assigned to the lower intensity layer to conservatively estimate the abrupt change intensity in low-confidence regions. The jump priority distribution superimposes the influence of weighted high-point annotations on top of the intensity layer. Window with weighted high points has an additional priority boost within the same intensity layer. Sections with densely packed weighted high points form prominent priority peak regions in the jump priority distribution. In outdoor scenes where light repeatedly changes due to cloud cover, weighted high points often appear simultaneously in multiple consecutive windows in the attenuation weight distribution characteristics. These sections form peak plateaus rather than single spikes in the jump priority distribution. Plateau-type peaks correspond to a wider effective range for adjusting the reference position, allowing for greater flexibility in selecting the adjustment timing. Windows in high-intensity layers containing weighted high points have the highest priority, while windows in low-intensity layers without weighted high points have the lowest priority. The four intensity layers and weighted high-point annotations jointly determine the priority value of each window in the jump priority distribution. A higher priority value indicates a more concentrated impact of the illuminance jump on the visual state at that moment.

[0045] Peak location is performed on the jump priority distribution to obtain priority extreme value coordinates. When the interval between adjacent candidate extreme points in the jump priority distribution is less than 3 time windows, the window number with the higher priority value is selected as the representative for merging. The merging rule prevents excessive dense extreme points from local small oscillations in the priority distribution. After merging, the number of extreme points converges to the truly significant jump region. Peak location adopts a sliding window maximum detection method. The detection window length is set to 5 time windows. The window with the largest jump priority distribution value within the window and higher than the mean within the window is judged as a candidate extreme point. This multiple threshold suppresses false extremes in flat areas while retaining the detection sensitivity of true peak values. In the priority extreme value coordinates, the time window number points to the starting position of the reference frame. The priority value is synchronously transmitted to the adjustment reference position to determine the estimated range of the target offset for this adjustment. The two components jointly locate the spatiotemporal position of the extreme point in the jump priority distribution. The gray-marked window does not participate in the candidate extreme point judgment. Non-gray candidate extreme points in the neighborhood of the gray window are normally included in the priority extreme value coordinates if they meet the judgment conditions, ensuring that low confidence regions do not interfere with the extraction of effective extreme points. Priority extreme value coordinates are arranged in descending order of priority value. When multiple extreme points have similar priority values, those with earlier time window numbers are given priority. Extreme points with earlier times correspond to earlier illuminance jumps, and their impact on the user's current visual state is still within the active period. During the transition period when users move from indoors to outdoors, the earliest high-priority extreme points often correspond to the critical window with the steepest illuminance rise. Prioritizing the processing of these points can respond to the visual impact of the jump in the most timely manner. Extreme points with later times serve as auxiliary candidates and are used in place of the main extreme point when there are many high dynamic range frames or insufficient confidence.

[0046] The adjustment reference position is determined based on the priority extreme value coordinates. The starting frame number is obtained by multiplying the time window number corresponding to the highest priority extreme point in the priority extreme value coordinates by the window length of 30 frames. Adding a 15-frame center offset determines the main adjustment reference position. The center frame is the window's starting frame plus 15 frames, rather than the starting frame itself, because the center position provides the most balanced coverage of the leading and trailing edges of the transition area. Frames at the front of the window are usually in the transition and climbing phase, while frames at the end may have entered a stable range. Using the center frame as the reference ensures that the comparison window has coverage margins in both directions. The second highest priority extreme point is determined as the auxiliary adjustment reference position. The auxiliary position takes over the main position when the time window corresponding to the main adjustment reference position is a high dynamic range (HMR) frame density area. A HMR frame density area is defined as having more than 6 HMR annotation frames within 5 frames before and after the main position, and the main position falling within the period when the user's violent movements cause camera image shake. When HMR frames are too dense, the auxiliary position intervenes to ensure the stability of the adjustment reference. When the priority extreme value coordinates contain only one extreme point, the adjustment reference position is unique and no auxiliary position is set. When the adjustment reference position falls within the low-confidence inherited annotation range, a low-confidence annotation is added. The deviation judgment threshold for the low-confidence adjustment reference position is relaxed to 1.3 times the normal value to reduce the risk of misjudging data fluctuations as true deviations. The intensity axis values ​​of each extreme point in the priority extreme value coordinates are synchronously transmitted to the adjustment reference position record. The higher the intensity axis value, the greater the target offset of this brightness adjustment. When the intensity axis value is higher than twice the global mean of the jump priority distribution, a large adjustment range is taken. When it is lower than the global mean, the adjustment range estimation range is narrowed to avoid weak jumps triggering excessive brightness changes.

[0047] The backlight adjustment node is determined by comparing the adjustment reference position with the brightness sensing data frame. The frame number corresponding to the adjustment reference position is located in the brightness sensing data frame as the corresponding frame unit. Using the mean component of this frame unit as the reference mean, a comparison window of 11 frames is formed by extending 5 frames forward and backward. The light perception changes within this timeframe typically cover the peak region of a sudden change event. Frames whose mean component deviates from the reference mean in the comparison window exceeds a set threshold are marked as deviation frames. A concentrated distribution of deviation frames indicates rapid light perception changes near the adjustment reference position. The backlight adjustment node is determined by selecting the frame unit with the largest absolute deviation from the deviation frames. If there are no deviation frames in the comparison window, the frame unit corresponding to the adjustment reference position in the brightness sensing data frame is directly used as the backlight adjustment node. No deviation indicates that the light perception is stable at that moment, and the reference position itself is a suitable node for adjustment intervention. The comparison window for the low-confidence reference position is expanded to 17 frames (8 frames before and 8 frames after) in the brightness perception data frame. This expanded window compensates for the insufficient positioning accuracy in low-confidence locations. The threshold for judging deviation frames within the expanded window is also relaxed. When the overall confidence of the user's light-sensing data is low in a semi-obscured environment, the expanded window can capture effective deviation features outside the confidence range, improving the accuracy of backlight adjustment node positioning under low-confidence conditions. The auxiliary adjustment reference position undergoes the same comparison process in the primary position replacement scenario. The backlight adjustment node determined by the auxiliary position is output together with the primary position result. The primary position node is prioritized for PWM driver parameter adjustment, while the auxiliary node serves as a backup. The backlight adjustment node includes the mean component and amplitude component values ​​of the corresponding frame unit. The mean component determines the direction of brightness adjustment, and the amplitude component determines the starting point of amplitude estimation. The combination of these two values ​​directly corresponds to the selection of the adjustment strategy.

[0048] Step S14: Adjust the PWM driver parameters according to the backlight adjustment node to determine the backlight driving parameters, use the backlight driving parameters to control the backlight optimization sequence through the driving circuit to generate the backlight optimization sequence, and perform advanced multi-scene brightness fusion to generate an adjustment configuration scheme.

[0049] Specifically, the backlight driving parameters are determined by adjusting the PWM driver parameters based on the backlight adjustment node. The mean component carried by the backlight adjustment node determines the initial brightness estimate for this adjustment, while the amplitude component determines the adjustment direction and intensity. A high mean component and a continuously rising amplitude component correspond to a strong light entry scene, while a low mean component and a stable amplitude component correspond to a weak light steady-state scene. The combination of these two forms corresponds to different parameter adjustment strategies. The PWM duty cycle adjustment amount D is determined by a linear mapping between the deviation of the current node's mean component and the target brightness reference, i.e., D = D_base + k × (L_target - L_node), where D_base is the current duty cycle reference value, k is a mapping coefficient calibrated according to the brightness response characteristics of the display panel, with a value range of 0.5 to 1.5, L_target is the target brightness reference preset according to the scene type, and L_node is the mean component of the backlight adjustment node. When the amplitude component of the backlight adjustment node is high, it indicates that the light perception at the node is in a dynamic fluctuation state. In this case, the mapping coefficient k is appropriately narrowed to reduce the single adjustment amplitude, avoiding excessive adjustment that could cause brightness jitter while the light perception is still changing. When the user is in an indoor environment where the light fluctuates with the movement of curtains, the amplitude component is high, and the duty cycle adjustment in the backlight drive parameters is correspondingly conservative to smooth the brightness transition. When the brightness perception data frame is marked as being in an energy-saving compressible state, L_target is shifted towards lower brightness with the shift not exceeding 15% of the current baseline value, and D_base is adjusted accordingly. The duty cycle baseline value in the backlight drive parameters remains at a lower level after compression in energy-saving scenarios, reducing the actual operating power consumption of the backlight drive circuit. When the backlight adjustment node is accompanied by a low-confidence label, the parameter adjustment strategy switches to conservative mode. In conservative mode, the adjustment of D is limited to 60% of that in normal mode, limiting the amplitude to prevent the estimation error generated by the low-confidence node from being amplified to the drive parameter level. In the backlight driving parameters, the PWM duty cycle adjustment, the adjustment step rate, and the overshoot protection upper limit jointly define the target amplitude, pace, and safety boundary of this adjustment. The adjustment step rate is dynamically determined by the change amplitude component of the backlight adjustment node. When the change amplitude is high, the step rate is appropriately increased to speed up the response. The overshoot protection upper limit is taken as 95% of the rated brightness of the display panel. When the duty cycle corresponding to the rated brightness of the panel is 100%, the overshoot protection upper limit is locked at 95% to prevent the driving circuit from exceeding the safety range during the rapid rise phase.

[0050] A backlight optimization sequence is generated by controlling the drive circuit using backlight drive parameters. The drive circuit performs closed-loop adjustment of the backlight drive signal based on the PWM duty cycle adjustment and step rate in the backlight drive parameters. Within each step cycle, the drive circuit advances the current duty cycle one step towards the target value. At the end of the step cycle, the backlight brightness feedback signal is collected in real time through the backlight module feedback loop. When the error between the feedback signal and the target duty cycle exceeds 5%, the drive circuit automatically triggers compensation adjustment to eliminate steady-state error. The backlight optimization sequence records the backlight brightness level output by the drive circuit in each step cycle throughout the adjustment process. The brightness level is scaled as a percentage of the duty cycle. The difference in brightness level between adjacent step cycles reflects the adjustment advancement rate. A uniform rate indicates a smooth adjustment transition, while a sudden change in rate indicates that the drive circuit has been affected by external interference or internal limiting triggering at that moment. In the backlight driving parameters, the overshoot protection upper limit is forcibly cut off when the output of the driving circuit reaches the upper limit. At the cutoff moment, the output is locked at the upper limit value and no longer advances. The cutoff event is recorded at the corresponding moment in the backlight optimization sequence and a cutoff annotation is attached. The cutoff annotation serves as a reference for the brightness upper limit constraint in the advanced multi-scene brightness fusion stage. In the backlight optimization sequence, adjusting the step rate reflects the friendliness of brightness changes to human eye perception. When the step rate is too fast, the brightness jump is obvious, and when the step rate is too slow, the response to light changes is lagging. The balance point between the two is determined by the step rate parameter in the backlight driving parameters. This parameter is dynamically adjusted according to the change amplitude component of the backlight adjustment node. When the change amplitude component is higher than 0.3, a higher step rate is used to speed up the response, and when it is lower than 0.1, a lower step rate is used to ensure a smooth transition.

[0051] In some embodiments, the step of generating an adjustment configuration scheme by performing advanced multi-scene brightness fusion on the backlight optimization sequence includes: predicting scene switching trends in the backlight optimization sequence to obtain a predicted scene identifier; performing delay compensation advanced brightness estimation based on the predicted scene identifier to generate advanced brightness parameters; performing multi-scene weighted fusion of the advanced brightness parameters and the backlight optimization sequence to form a fused brightness sequence; and generating an adjustment configuration scheme based on the fused brightness sequence.

[0052] Scene switching trend prediction is performed on the backlight optimization sequence to obtain the predicted scene identifier. The trend prediction is based on the recent evolution slope of the brightness level in the backlight optimization sequence and the remaining duration of the current scene segment in the scene classification results. The two are combined to determine the probability of an upcoming scene switch. Switching trend prediction is triggered when the brightness slope continues to evolve towards the scene boundary and the remaining duration of the current scene segment is less than 5 time windows. When the linear fitting slope of the recent brightness level change in the backlight optimization sequence exceeds 1.5 times the mean, it is judged as a rapid evolution trend. Rapid evolution indicates that the current scene lighting characteristics are moving closer to the feature range of adjacent scene types. In low-light scenes at night, the brightness level continues to rise rapidly, indicating that the user is moving to a brighter lighting environment and predicting that they will soon enter a scene with direct strong light. The predicted scene identifier is jointly determined by the trend judgment result and the list of adjacent scene types. The adjacent scene types are the historical scene types that have appeared before and after the current scene segment in the scene classification results. Types that have appeared in the past are more likely to be predicted than types that have never appeared. The predicted scene identifier is accompanied by a prediction confidence level, which is determined by the product of the normalized value of the evolution slope and the normalized value of the frequency of occurrence of historical scenes. The two are normalized to the range of 0 to 1 respectively and then multiplied to ensure that the dimensions are consistent. The higher the product, the stronger the fit between the current evolution trend and the historical pattern, and the higher the reliability of the predicted scene identifier. When the confidence level of the predicted scene identifier is lower than 0.4, the current backlight optimization sequence maintains the original parameters and continues to execute without initiating pre-adjustment to avoid unnecessary brightness disturbances introduced by low-confidence predictions. When the user moves in a mixed light source environment with complex illuminance distribution, the historical scene types are dispersed, and the prediction confidence level is generally lower than the threshold. The backlight optimization sequence maintains stable output during this period.

[0053] Based on the predicted scene identifier, a delay compensation-based advanced brightness estimation is performed to generate advanced brightness parameters. The necessity of delay compensation stems from the fact that the PWM drive circuit requires a settling time from receiving the adjustment command to the actual stabilization of the backlight brightness. During this settling time, the displayed brightness may not be synchronized with the ambient light, resulting in a brief brightness lag perceived by the user. Advanced estimation eliminates this lag by initiating adjustment in advance before the actual scene switch. The difference between the target scene brightness benchmark L_pred corresponding to the predicted scene identifier and the brightness level L_cur at the end of the current backlight optimization sequence determines the advanced adjustment magnitude. The advanced adjustment amount A_advance = L_pred - L_cur, where A_advance is the advanced adjustment amount, and its unit is the same as L_pred and L_cur, which is a normalized brightness value. The advance frame count N_advance is determined by the advance adjustment amount and the delay compensation coefficient, i.e., N_advance = (A_advance / R_step) + N_delay, where R_step is the step rate, i.e., the maximum duty cycle change per frame, N_delay is the number of frames converted from the actual setup time of the PWM drive circuit, A_advance / R_step is the number of frames required to complete all adjustments at the step rate, and N_delay is the number of frames for setup time compensation. The sum of the two ensures that the brightness has been adjusted and the drive circuit has stabilized at the actual scene switching time. The longer the setup time, the larger N_delay is. The more advance frames, the earlier the pre-adjustment needs to be started. Before the user drives into the tunnel, the light changes from bright light to dim rapidly. If the setup time is too long, the advance adjustment needs to be started several frames before the scene switching to ensure that the brightness is basically in place after the switch. The advanced brightness parameter includes the amount of advance adjustment and the number of frames before adjustment trigger. Together, they describe the intervention range and trigger timing of the pre-adjustment. When the predicted scene is marked with a low confidence label, the number of frames before adjustment trigger in the advanced brightness parameter is reduced to 70% of the normal value. Low confidence prediction is conservatively handled in terms of adjustment timing to reduce brightness fluctuations caused by misprediction.

[0054] A multi-scene weighted fusion of the advanced brightness parameters and the backlight optimization sequence is performed to form a fused brightness sequence. Weighted fusion superimposes the brightness trajectory of the current scene in the backlight optimization sequence with the pre-adjustment trajectory planned by the advanced brightness parameters along the time axis. The pre-adjustment trajectory takes the advanced adjustment amount as its target endpoint and the time node specified by the advance frame number of adjustment triggers as its starting point. Before this node, the two trajectories are output independently; after this node, they are weighted and merged. The fusion weights change dynamically over time. At the fusion starting point, the current scene brightness weight is 0.8, and the pre-adjustment trajectory weight is 0.2. As time progresses, both weights linearly transition to 0.2 and 0.8 at the predicted scene switching time. This weight transition ensures a smooth and gradual transition of brightness from the current scene to the predicted scene, avoiding sudden brightness jumps caused by the pre-adjustment trajectory dominating. Fusion trajectories of different scene types are calculated independently and then superimposed according to scene weights. The scene type belonging to the current moment has the highest weight, and the weights of adjacent scene types decrease with confidence. The multi-scene superposition result is normalized and used as the brightness level of each moment in the final fused brightness sequence. The brightness level of the fused brightness sequence at each moment within the fusion segment falls between the brightness at the end of the current backlight optimization sequence and the target endpoint of the advance adjustment. When it exceeds this range, it is truncated to the boundary value, and the truncation event is marked and recorded in the fused brightness sequence. Fusion segments with low confidence in the predicted scene are marked with low-confidence fusion labels at the corresponding positions in the fused brightness sequence. The brightness level of the low-confidence fusion labeled segments corresponds to more conservative execution constraints. When the user's prediction confidence is insufficient under complex lighting conditions at dusk, the weight transition slope of the corresponding segment in the fused brightness sequence narrows, and the gradual increase in brightness is limited.

[0055] An adjustment configuration scheme is generated based on the fused brightness sequence. The brightness level at each moment in the fused brightness sequence is converted into the corresponding PWM target duty cycle. The conversion relationship is determined by the panel brightness characteristic curve, which typically exhibits a non-linear relationship. The influence of duty cycle changes on brightness is greater in the low brightness range than in the high brightness range. During the conversion, a piecewise linear approximation of the non-linear segment is required to ensure the accuracy of the target duty cycle calculation. The adjustment configuration scheme uses time frames as indices, recording the target PWM duty cycle, execution priority, and abnormal response actions for each frame. The execution priority is determined by the rate of change of the brightness level in the fused brightness sequence of the corresponding frame. Frames with higher change rates have higher execution priority, ensuring that adjustment commands during rapid scene switching periods do not miss the optimal execution time due to scheduling delays. In the adjustment configuration scheme, the execution priority of the low-confidence fused frame in the fused brightness sequence is lowered by one level, and the abnormal response action is set to roll back to the previous valid configuration. When the adjustment command of the low-confidence frame fails to execute, it automatically rolls back to avoid the brightness remaining in an intermediate state. After the adjustment configuration scheme is generated, it is transmitted to the PWM closed-loop modulation module in the form of a frame sequence. The execution timing of each frame configuration in the scheme is aligned with the timestamp of the backlight adjustment node to ensure that the adjustment action is triggered within the correct light-sensing state window. During the transition period when the user moves from indoors to outdoors, the corresponding transition frame in the adjustment configuration scheme has the highest execution priority and the target duty cycle decreases continuously. The adjustment command is sent to the drive circuit in the frame sequence.

[0056] Step S15: Based on the intensity change components in the ambient illuminance data, perform illuminance modulation analysis to obtain modulation amplitude data, determine the adjustment period based on the visual masking window alignment adjustment configuration scheme of the modulation amplitude data, and execute PWM closed-loop modulation to output brightness adjustment command according to the adjustment period.

[0057] Specifically, modulation amplitude data is obtained through illuminance modulation analysis based on the intensity variation components in ambient illuminance data. The extraction of intensity variation components targets the non-steady-state components in the time-domain sequence of ambient illuminance data. The steady-state component corresponds to the average light level, while the non-steady-state component corresponds to the periodic or non-periodic fluctuations of light intensity around the average. Separation between the two is achieved by performing a moving average filter on the time-domain sequence and taking the residual. The moving window length is set to 1 second to effectively separate gradual trends from rapid fluctuations. The modulation amplitude data consists of two indicators: the peak-to-peak value of the non-steady-state component and the modulation frequency. The peak-to-peak value reflects the magnitude of light intensity fluctuations, while the modulation frequency is determined by the zero-crossing rate of the non-steady-state component. A higher zero-crossing rate indicates more frequent crossings of the light intensity around the average, corresponding to a higher modulation frequency. Interference components introduced by the display screen's self-emission in the ambient illuminance data are removed before illuminance modulation analysis through self-emission suppression processing. The removal method is consistent with the self-emission suppression mechanism in the gradient transition analysis stage of S13, ensuring that the modulation amplitude data reflects the pure ambient light modulation characteristics. When the peak-to-peak value of the modulation amplitude data exceeds 20% of the average ambient illuminance data, it is considered an over-modulation environment. This occurs when a user uses the device for an extended period in an office with noticeable fluorescent light flicker. In this scenario, the modulation frequency of the modulation amplitude data corresponds to a multiple of the power supply frequency, and the peak-to-peak value is relatively large. The alignment strategy for the visual masking window in this scenario needs to differ from that in a low-modulation environment. When the modulation frequency of the modulation amplitude data is below 1Hz, it is considered slow modulation. Slow modulation typically originates from cloud cover or the slow changes in indoor lighting as people move in and out. In slow modulation scenarios, the fluctuation period of light intensity is much longer than the rapid eye movement interval, and effective capture of the masking window relies on a wider boundary search range.

[0058] In some embodiments, determining the adjustment period by aligning the visual masking window with the adjustment configuration scheme based on the modulation amplitude data includes: performing eye-tracking event detection on the modulation amplitude data to generate fast eye-tracking components and slow eye-tracking components; using the fast eye-tracking components to perform masking depth modulation on the slow eye-tracking components to form a masking depth time series; reconstructing window boundaries using the masking depth time series to generate a masking window boundary distribution; and determining the adjustment period by aligning the duration based on the masking window boundary distribution and the adjustment configuration scheme.

[0059] Eye-tracking event detection is performed on the modulation amplitude data to generate fast and slow eye-tracking components. Eye-tracking event detection is based on the temporal variation characteristics of the modulation frequency in the modulation amplitude data. A rapid jump in the modulation frequency within a short period corresponds to the visual system's stress response triggered by the user's eye saccades. A rapid jump is defined as the difference in modulation frequency between adjacent sampling points exceeding 50% of the mean. A rapid jump is defined as three or more consecutive rapid jumps occurring at the same sampling point. The time period of occurrence in the fast eye-tracking component is used to determine the candidate starting position of the masking window. The normalized modulation frequency jump amplitude directly serves as the modulation input for the masking depth coefficient. A larger jump amplitude indicates a stronger visual system stress triggered by the fast eye movement. Strongly stressed fast eye-tracking events generate larger modulation coefficients during the masking depth modulation stage, thus forming a wider effective masking window in the corresponding time period. The slow-motion eye-tracking component is extracted from the residual temporal sequence after excluding fast eye-tracking event segments in the modulation amplitude data. The portion of the residual sequence where the modulation frequency drifts slowly over time corresponds to the eye movement state when the user's gaze smoothly tracks the displayed content. When a user reads long text, their gaze moves smoothly from left to right along the line, corresponding to a low and regular drift rate in the modulation frequency of the slow-motion eye-tracking component. The amplitude envelope of the slow-motion eye-tracking component reflects the stability of the user's visual attention. A stable envelope indicates a stable visual state, while fluctuating envelopes indicate frequent switching of visual attention between different regions. A stable visual state provides a more reliable reference for reconstructing the masked window boundary.

[0060] Masking depth temporal sequence is formed by modulating the slow eye movement component using the fast eye movement component (SEM). The visual masking effect is significantly enhanced within a short time window after a fast eye movement (FEM). Within this window, brightness changes are partially masked by the visual system. Utilizing this physiological mechanism to perform brightness adjustment within the masking window can reduce the user's perceptual sensitivity to the adjustment process. The normalized amplitude of the jumps in each FEM event within the FEM component is used as the masking depth coefficient M_saccade. The masking depth modulation formula is M_depth = M_base × (1 + α_gain × M_saccade), where M_base is the base masking depth coefficient and α_gain is the modulation gain coefficient. A higher M_saccade value corresponds to a greater effective masking depth in the corresponding time segment, allowing for larger brightness adjustments within that period without triggering noticeable perception. The coefficient values ​​at each moment in the masking depth time series are determined frame-by-frame by slow eye-tracking components after fast eye-tracking masking depth modulation. Moments with higher coefficients correspond to low-perceptual-interference windows where brightness adjustment is executed. During the brief transition period after a user quickly scans the page and their gaze settles, the masking depth coefficient in the masking depth time series is significantly higher, making it an ideal time to insert brightness adjustment commands. When the fast eye-tracking component event density is too high, multiple consecutive moments in the masking depth time series are in a high masking depth state. These consecutive high masking windows are merged into a wide window during the boundary reconstruction phase. The masking depth coefficient within the wide window is the duration-weighted average of the sub-windows. The merged wide window can accommodate the concentrated execution of more frame configurations during the duration alignment phase.

[0061] The masking window boundary distribution is generated by reconstructing the window boundaries using masking depth temporal sequences. The window boundaries are determined by the rising and falling edges of the masking depth coefficient in the masking depth temporal sequence. The rising edge corresponds to the opening time of the masking window, and the falling edge corresponds to the closing time. Rising edge identification uses first-order differential detection; a difference value exceeding 0.3 times M_base is considered a rising edge trigger. This threshold preserves the main masking events while suppressing false detections caused by noise. The opening and closing times in the masking window boundary distribution define the usable execution interval of each window. The average masking depth coefficient serves as the priority allocation criterion for frame configuration. A higher average masking depth coefficient within a window indicates lower perceptual interference for brightness adjustment within that window, making it more suitable for larger brightness adjustment commands. When the opening time interval between two adjacent masking windows is less than two frames, they are merged into the same window. The average masking depth coefficient after merging is the duration-weighted average of the two windows. This merging operation prevents the masking window from being divided into too many short windows during rapid eye movements, thus avoiding fragmentation of the boundary distribution. In the masking window boundary distribution, short windows with a duration of less than 3 frames are marked as micro-windows. Micro-windows do not independently carry the execution segment of the adjustment configuration scheme during the duration alignment phase; they only serve as supplementary time slots when the capacity of adjacent windows is insufficient. In the modulation amplitude data, the masking window boundary distribution window intervals corresponding to slow modulation are usually relatively long. When windows are sparse, duration alignment requires delaying the execution of some frame configurations until the next available window. The upper limit of the delay is 3 frame configurations. If the upper limit is exceeded, the corresponding frame configuration is forced to be executed at the current moment without waiting.

[0062] The adjustment cycle is determined by duration alignment based on the masking window boundary distribution and adjustment configuration scheme. Duration alignment maps the execution timing of each frame configuration in the adjustment configuration scheme to the nearest available masking window in the masking window boundary distribution. The mapping principle is to prioritize the window with the highest average masking depth coefficient within the window. The higher the masking depth, the lower the probability of the user perceiving a brightness change during adjustment, and the less visual interference the adjustment transition causes to the user. During mapping, it is also necessary to verify whether the duration of the frame configuration matches the effective duration of the target masking window. If the duration of the frame configuration does not exceed the effective duration of the window, it is directly embedded. If it exceeds, the frame configuration is split into two adjacent masking windows for segmented execution. The split point is the moment with the lowest duty cycle change rate in the frame configuration to ensure a smooth brightness transition at the segment. In the adjustment configuration scheme, the frame configuration with higher execution priority occupies the window with the highest masking depth first, the frame configuration with lower execution priority uses the remaining windows, and when window resources are insufficient, the low-priority frame configuration is degraded to micro-window execution. The adjustment period is determined by the interval between two adjacent execution moments in the frame configuration execution timing sequence. When the interval is uniform, the adjustment period is fixed; when the interval is non-uniform, the median of each interval is used as the representative value. The median is more resistant to extremely long intervals than the mean. When users read long texts, their eye movement rhythm is stable, the intervals between windows in the masked window boundary distribution are uniform, and the adjustment period is close to a fixed value. However, when quickly browsing images, rapid eye movements are frequent, the window intervals are short and non-uniform, and the adjustment period is correspondingly shortened and dynamically changes with the eye movement rhythm. Once the adjustment period is determined, it is locked as the execution beat of this PWM closed-loop modulation. The stability of the adjustment period directly affects the user's overall perception of the smoothness of brightness adjustment.

[0063] The brightness adjustment command is output based on the PWM closed-loop modulation output according to the adjustment cycle. The PWM closed-loop modulation drives the sequential execution of each frame configuration in the adjustment configuration scheme, using the adjustment cycle as the beat. Within each adjustment cycle, the drive circuit reads the target PWM duty cycle and compares it with the measured backlight brightness feedback signal. If the error exceeds the set tolerance, compensation adjustment is performed within the adjustment cycle to eliminate steady-state error. The brightness adjustment command is generated by the PWM closed-loop modulation module at the beginning of each adjustment cycle. The command content specifies the adjustment target for this cycle with the target duty cycle, constrains the single change rate with the allowable adjustment step size, defines the execution window with the effective duration of this cycle, and limits the maximum duty cycle change rate within a single adjustment cycle to prevent excessively rapid brightness jumps when consecutive high-priority configurations are executed in a concentrated manner. In the adjustment configuration scheme, the brightness adjustment command step size corresponding to the low-confidence fusion-marked frame is compressed to 60% of the normal value. The brightness adjustment range in the low-confidence segment is limited. When the confidence of the predicted scene is insufficient, the advanced adjustment command generated will not produce obvious erroneous brightness states even if there is a deviation in the prediction under the compression constraint. If a sudden change in ambient light occurs during the execution of a brightness adjustment command, the execution of the command in the current adjustment cycle is interrupted, and the complete process from S13 to S15 is retried to ensure that the brightness adjustment always responds accurately to the current ambient light state. After the brightness adjustment command sequence is fully executed, the backlight brightness converges to the target brightness level. When the deviation between the measured brightness and the target brightness is less than the tolerance for three consecutive adjustment cycles, convergence is considered complete. After convergence, the PWM closed-loop modulation enters the hold mode, and the adjustment cycle is extended to reduce the frequency of ineffective adjustments. During the hold mode, if the average value component of the brightness sensing data frame is lower than 80% of the current target brightness reference for 10 consecutive adjustment cycles, a power-saving compression evaluation is triggered. If the evaluation is passed, the target duty cycle is further reduced to the energy-saving threshold and locked.

[0064] To implement the above method embodiment, a display screen environment adaptive brightness adjustment method is provided to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a display screen environment adaptive brightness adjustment device according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The display screen environment adaptive brightness adjustment device provided in this embodiment includes: Data fusion module 201 is used to acquire ambient illuminance data and user facial brightness data, and perform multi-source light feedback fusion based on the ambient illuminance data and the user facial brightness data to form a brightness perception data frame; Scene detection module 202 is used to construct a scene response sequence using the brightness perception data frame, perform blink cycle masking scene type detection on the scene response sequence to extract a brightness sampling point set, and construct an illuminance distribution matrix based on the brightness sampling point set; The reference positioning module 203 is used to perform illuminance jump feature analysis on the illuminance distribution matrix to identify high illuminance jump regions, perform continuous jump attenuation priority enhancement analysis in the high illuminance jump regions to determine the adjustment reference position, and compare the adjustment reference position with the brightness perception data frame to determine the backlight adjustment node. Backlight control module 204 is used to determine backlight driving parameters by adjusting PWM driver parameters according to the backlight adjustment node, and to generate a backlight optimization sequence by controlling the driving circuit using the backlight driving parameters, and to generate an adjustment configuration scheme by performing advanced multi-scene brightness fusion on the backlight optimization sequence. The modulation output module 205 is used to perform illuminance modulation analysis based on the intensity change components in the ambient illuminance data to obtain modulation amplitude data, align the visual masking window of the modulation amplitude data with the adjustment configuration scheme to determine the adjustment period, and execute the PWM closed-loop modulation output brightness adjustment command according to the adjustment period.

[0065] The aforementioned display screen environment adaptive brightness adjustment device can implement a display screen environment adaptive brightness adjustment method according to the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0066] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. A method for adaptive brightness adjustment of a display screen based on environmental conditions, characterized in that, include: Acquire ambient illuminance data and user facial brightness data, and perform multi-source light feedback fusion based on the ambient illuminance data and the user facial brightness data to form a brightness perception data frame; A scene response sequence is constructed using the brightness perception data frame. A brightness sampling point set is extracted by blink cycle masking scene type detection of the scene response sequence. An illuminance distribution matrix is ​​constructed based on the brightness sampling point set. Illuminance abrupt change feature analysis is performed on the illuminance distribution matrix to identify high illuminance abrupt change regions. In the high illuminance abrupt change regions, continuous abrupt change attenuation priority enhancement analysis is performed to determine the adjustment reference position. The adjustment reference position is compared with the brightness perception data frame to determine the backlight adjustment node. Based on the backlight adjustment node, the PWM driver parameters are adjusted to determine the backlight driving parameters. The backlight driving parameters are used to control the generation of a backlight optimization sequence through the driving circuit. The backlight optimization sequence is then used to generate an adjustment configuration scheme by performing advanced multi-scene brightness fusion. Illumination modulation analysis is performed on the intensity change components in the ambient illuminance data to obtain modulation amplitude data. The adjustment period is determined by aligning the visual masking window of the modulation amplitude data with the adjustment configuration scheme. The brightness adjustment command is then executed according to the adjustment period using PWM closed-loop modulation.

2. The method according to claim 1, characterized in that, The step of performing multi-source light feedback fusion based on the ambient illuminance data and the user's facial brightness data to form a brightness perception data frame includes: The ambient illuminance data is decomposed into direct light and scattered light components. The direct light component is used to weight the user's facial brightness data with pupil area priority illuminance to form a composite facial brightness. The synthesized facial brightness and the scattered light component are temporally correlated and fused to generate fused light sensing parameters; Based on the fused light-sensing parameters, a frame structure is grouped to form a brightness-sensing data frame.

3. The method according to claim 1, characterized in that, The step of detecting the scene type by blink cycle masking and extracting the brightness sampling point set for the scene response sequence includes: Facial action recognition is performed on the scene response sequence to obtain blink timing markers; Based on the blink timing marker, the scene response sequence is subjected to high-frequency blink fatigue state removal to generate an effective sampling sequence; The valid sampled sequences are classified into scene types to obtain scene classification results; Based on the scene classification results, sampling points are extracted to form a brightness sampling point set.

4. The method according to claim 1, characterized in that, The step of analyzing the illuminance abrupt change characteristics of the illuminance distribution matrix to identify high illuminance abrupt change regions includes: The illuminance distribution matrix is ​​used to establish the illuminance time-series variation distribution and generate time-series illuminance data. Based on the aforementioned time-series illuminance data, an autoluminescence suppression abrupt gradient analysis was performed to determine the illuminance abrupt change region. The illumination abrupt change region is subjected to cross-frame continuous verification to form a continuous jump band; Based on the continuous transition zone, high-illuminance transition regions are extracted and identified.

5. The method according to claim 1, characterized in that, The step of determining the adjustment reference position by performing continuous abrupt attenuation preferential enhancement analysis in the high illuminance abrupt change region includes: A continuous jump sequence is obtained by performing consistency statistics on the jump direction of the high-illuminance jump region. A jump priority distribution is generated by performing cumulative decay weight analysis based on the continuous jump sequence; Peak location is performed on the jump priority distribution to obtain the priority extreme value coordinates; The adjustment reference position is determined based on the priority extreme value coordinates.

6. The method according to claim 1, characterized in that, The advanced multi-scene brightness fusion generation and adjustment configuration scheme for the backlight optimization sequence includes: The scene switching trend is predicted by performing scene optimization sequence to obtain the predicted scene identifier; Based on the predicted scene identifier, delay compensation and advance brightness estimation are performed to generate advance brightness parameters; The advanced brightness parameters and the backlight optimization sequence are weighted and fused in multiple scenarios to form a fused brightness sequence; An adjustment configuration scheme is generated based on the fused brightness sequence.

7. The method according to claim 1, characterized in that, The step of aligning the visual masking window based on the modulation amplitude data with the adjustment configuration scheme to determine the adjustment period includes: Eye-tracking event detection is performed on the modulation amplitude data to generate fast eye-tracking components and slow eye-tracking components. The fast eye movement component is used to perform masking depth modulation on the slow eye movement component to form a masking depth timing sequence. The masking window boundary distribution is generated by reconstructing the window boundary using the masking depth time series. The adjustment period is determined by aligning the duration based on the distribution of the masking window boundaries and the adjustment configuration scheme.

8. The method according to claim 3, characterized in that, The step of classifying the scene type of the valid sampled sequence to obtain the scene classification result includes: Illuminance dynamic range statistics are performed on the effective sampling sequence to generate dynamic range distribution characteristics; Based on the dynamic range distribution characteristics, cross-frame illuminance interpolation is performed to identify three types of scene identifiers: strong light direct irradiance section, low light section at night, and rapid light change section. Confidence assessment is performed on the three types of scene identifiers to generate scene confidence sequences; The optimal scene classification is performed based on the scene confidence sequence to obtain the scene classification result.

9. The method according to claim 5, characterized in that, The step of generating a jump priority distribution based on the cumulative decay weight analysis of the continuous jump sequence includes: Interval distribution analysis of the continuous jump sequence generates short-interval dense components and long-interval sparse components. The attenuation intensity of the long-interval sparse component is modulated using the short-interval dense component to form a modulation attenuation sequence. The attenuation weight distribution feature is generated by reconstructing the weight distribution using the modulation attenuation sequence. A jump intensity hierarchical analysis is performed on the attenuation weight distribution characteristics to generate a jump priority distribution.

10. A display screen environmental adaptive brightness adjustment device, characterized in that, include: The data fusion module is used to acquire ambient illuminance data and user facial brightness data, and perform multi-source light feedback fusion based on the ambient illuminance data and the user facial brightness data to form a brightness perception data frame; The scene detection module is used to construct a scene response sequence using the brightness perception data frame, perform blink cycle masking scene type detection on the scene response sequence to extract a brightness sampling point set, and construct an illuminance distribution matrix based on the brightness sampling point set. The reference positioning module is used to perform illuminance jump feature analysis on the illuminance distribution matrix to identify high illuminance jump regions, perform continuous jump attenuation priority enhancement analysis in the high illuminance jump regions to determine the adjustment reference position, and compare the adjustment reference position with the brightness perception data frame to determine the backlight adjustment node. The backlight control module is used to determine the backlight driving parameters by adjusting the PWM driver parameters according to the backlight adjustment node, and to generate a backlight optimization sequence by controlling the driving circuit through the backlight driving parameters. The backlight optimization sequence is then used to generate an adjustment configuration scheme by performing advanced multi-scene brightness fusion. The modulation output module is used to perform illuminance modulation analysis based on the intensity change components in the ambient illuminance data to obtain modulation amplitude data, align the adjustment period with the adjustment configuration scheme based on the visual masking window of the modulation amplitude data, and execute PWM closed-loop modulation output brightness adjustment commands according to the adjustment period.

Images