Multi-parameter eyelid dynamic driving screen brightness adjusting method and system

Abstract

The invention discloses a multi-parameter eyelid dynamic driving screen brightness adjusting method and system. The method comprises the following steps: acquiring an eyelid movement track of a user in real time through a front-facing camera; extracting blink frequency, duration and amplitude as multi-dimensional dynamic parameters; dynamically distributing the weight according to the real-time change rate of each parameter, and when the change rate exceeds a preset threshold value, improving the weight of the corresponding parameter; inputting the weighted parameters into a Sigmoid type nonlinear function to calculate the target brightness; and finally, the screen brightness is adjusted in a stepped smooth mode. The system comprises a collaboration module for realizing the functions. According to the method, the real-time physiological state of the user is used as a unique core drive, accurate response and rapid intervention of the nonlinear accumulation process of visual fatigue are achieved through dynamic weight distribution and nonlinear mapping, and the visual comfort degree of use of the mobile terminal is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of mobile terminal display control and human-computer interaction technology, specifically to a method and system for adjusting screen brightness using multi-parameter eyelid dynamic driving. Background Technology

[0002] In the mobile internet era, smartphones, tablets, and other mobile devices have become core carriers for people's work, entertainment, and information access. Users' daily screen time is constantly increasing, leading to increasingly prominent problems such as visual fatigue and dry eye syndrome. Screen brightness, as a key parameter directly affecting visual comfort and energy consumption, is crucial for ensuring user visual health and improving the human-computer interaction experience through intelligent adjustment. Therefore, developing screen brightness adjustment technology that can adapt to the user's actual state has become an important research direction in the fields of display control and human-computer interaction.

[0003] Currently, technological exploration in this field has mainly formed several typical routes, but none of them have fundamentally achieved adaptive adjustment based on the user's real-time physiological state. Firstly, the adjustment scheme based on ambient light sensors is the most common. It adjusts screen brightness by sensing the external ambient light intensity. However, the fundamental flaw of this approach is that its adjustment logic is entirely dependent on the external environment, completely disconnected from the user's own visual load and fatigue state, and cannot meet the differentiated screen brightness needs of users at different fatigue levels under the same lighting conditions. Secondly, some studies have attempted to introduce user biometrics, such as detecting a single parameter (e.g., blinking frequency or pupil diameter) through a camera and adjusting based on a fixed threshold. While such methods address the user's state, due to the single parameter dimension and linear judgment logic, they cannot characterize and respond to the complex physiological processes of nonlinear accumulation and abrupt changes in visual fatigue. For example, they cannot effectively distinguish between accidental blinking and persistent eyelid closure, which indicates deep fatigue, resulting in insufficient adjustment accuracy. Thirdly, there are adjustment strategies based on preset scenario rules (such as reading mode, video mode) or specific interactive behaviors (such as stylus usage). These methods are essentially switching predefined strategies and lack the ability to respond in real time to continuous and dynamic changes in the user's physiological state. In addition, although existing technologies involve multi-source data fusion or nonlinear mapping, the weights of their data fusion are often fixed values ​​or set according to environmental factors, and their nonlinear relationships are not specifically designed to simulate the physiological response of human fatigue, but rather to optimize display consistency or power consumption performance.

[0004] In summary, existing technologies have limitations in terms of adjustment basis, parameter dimensions, response logic, and model specificity. Therefore, there is an urgent need in this field for a new screen brightness adjustment method and system that can be driven by real-time multidimensional physiological signals of users and has a built-in mechanism to simulate fatigue nonlinear response. Summary of the Invention

[0005] The purpose of this invention is to provide a screen brightness adjustment method and system with multi-parameter eyelid dynamic driving, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A multi-parameter eyelid-dynamically driven screen brightness adjustment method includes the following steps: Step S1: The user's facial video stream is captured in real time by the front-facing camera of the mobile terminal, and the user's eyelids are identified and tracked from the video stream based on a lightweight neural network model to generate eyelid movement trajectory data; Step S2: Based on the eyelid movement trajectory data, extract multidimensional eyelid dynamic parameters in real time. The multidimensional eyelid dynamic parameters include at least blinking frequency, blinking duration, and blinking amplitude. Step S3: Based on the real-time change rate of each of the multidimensional eyelid dynamic parameters, dynamically calculate and allocate the contribution weight corresponding to each parameter. When the real-time change rate of any parameter exceeds its corresponding preset mutation threshold, the contribution weight of that parameter is increased. Step S4: Input the multidimensional eyelid dynamic parameters after the contribution weighting process into a preset nonlinear brightness mapping function to calculate the target screen brightness value; Step S5: Adjust the screen brightness of the mobile terminal according to the target screen brightness value.

[0007] Furthermore, in step S3, the real-time rate of change The calculation formula is: The historical parameter mean and historical parameter standard deviation are calculated based on historical parameter data within a preset time period.

[0008] Specifically, the real-time rate of change of the blink duration parameter is calculated. ,like If the value exceeds the first preset threshold, the weight vector will be assigned as: blink frequency weight. Within the first low value range, the weight of blink duration Within the highest value range, the weight of blink amplitude Located in the second lowest value range; and / or calculating the real-time rate of change of the blink frequency parameter. ,like If the value exceeds the second preset threshold, the weight vector will be assigned as: blink frequency weight. The second highest value range, blink duration weight Located in the third lowest value range, blink amplitude weighting It is in the fourth low value range; the first preset threshold is 1.5, the first high value range is not less than 0.7; the second preset threshold is 2.0, the second high value range is not less than 0.6.

[0009] Furthermore, the preset nonlinear brightness mapping function is a Sigmoid type function, and its expression is: ,in, The target screen brightness value. The base brightness reference value is given, and k is a shape adjustment factor ranging from 0.6 to 1.0. , respectively, represent the contribution weights of blink frequency, blink duration, and blink amplitude, and f, d, and a are the normalized blink frequency, blink duration, and blink amplitude values, respectively.

[0010] Furthermore, step S5 specifically includes: calculating the difference between the target screen brightness value and the current screen brightness value; if the absolute value of the difference exceeds 10%, the brightness adjustment process is divided into at least 3 steps, each step is executed once every 50 milliseconds, so as to adjust the current screen brightness value to the target screen brightness value step by step.

[0011] Preferably, the method further includes: when a single blink lasts for more than 100 milliseconds and the blink amplitude exceeds 80%, it is determined to be a deep fatigue event; in response to the deep fatigue event, the screen brightness of the mobile terminal is controlled to decrease to a level not exceeding 30% of the basic brightness reference value within 200 milliseconds.

[0012] A multi-parameter eyelid-dynamically driven screen brightness adjustment system, integrated into a mobile terminal, includes: The eyelid dynamic recognition module is used to collect the user's facial video stream in real time through the front-facing camera of the mobile terminal, and to identify and track the user's eyelids from the video stream based on a lightweight neural network model to generate eyelid movement trajectory data. The multi-parameter extraction module is used to extract multi-dimensional eyelid dynamic parameters in real time based on the eyelid movement trajectory data. The multi-dimensional eyelid dynamic parameters include at least blinking frequency, blinking duration, and blinking amplitude. An adaptive weight allocation module is used to dynamically calculate and allocate contribution weights corresponding to each parameter based on the real-time change rate of each of the multidimensional eyelid dynamic parameters. When the real-time change rate of any parameter exceeds its corresponding preset mutation threshold, the contribution weight of that parameter is increased. The nonlinear brightness mapping module is used to input the multidimensional eyelid dynamic parameters after the contribution weighting process into a preset nonlinear brightness mapping function to calculate the target screen brightness value. The screen brightness adjustment execution module is used to generate a control signal to adjust the screen brightness of the mobile terminal based on the target screen brightness value.

[0013] Furthermore, the adaptive weight allocation module is deployed in the AI ​​coprocessor of the mobile terminal or a dedicated processing unit that supports direct memory access (DMA); the nonlinear brightness mapping module is integrated in the screen driver controller firmware of the mobile terminal.

[0014] Furthermore, the eyelid dynamic recognition module is configured to process video streams with a frame rate of not less than 30fps in real time; and the multi-parameter extraction module, the adaptive weight allocation module, and the nonlinear brightness mapping module are configured to operate in a collaborative pipeline manner.

[0015] Furthermore, the eyelid dynamic recognition module and the multi-parameter extraction module interact with each other via a shared memory buffer; the multi-parameter extraction module and the adaptive weight allocation module are connected via a DMA channel; the adaptive weight allocation module and the nonlinear brightness mapping module are connected via I... 2 C bus connection.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Achieved precise nonlinear adjustment in accordance with physiological laws: By integrating multidimensional eyelid parameters and using a Sigmoid function for mapping, the brightness adjustment curve can match the nonlinear cumulative characteristics of visual fatigue, solving the problem of the disconnect between traditional linear or threshold adjustment and physiological response.

[0017] 2. Possesses keen perception and rapid response capabilities to sudden changes in fatigue state: Innovatively, the system dynamically allocates weights based on the real-time rate of change of each parameter, enabling it to prioritize responses to deep fatigue signals such as prolonged eye closure and complete brightness intervention within hundreds of milliseconds, significantly improving adjustment sensitivity and timeliness.

[0018] 3. Provides a smooth and imperceptible user experience: Through a step-by-step smooth adjustment algorithm, visual flicker and interference caused by abrupt changes in brightness are avoided, resulting in a natural brightness transition and improved user comfort.

[0019] 4. An efficient and reliable low-latency processing architecture was constructed: through memory sharing, DMA transfer and modular collaborative design, the end-to-end processing latency of the system is controlled at the millisecond level, ensuring the real-time perception of brightness adjustment by physiological signals, and making the technical solution highly feasible and practical. Attached Figure Description

[0020] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart illustrating a multi-parameter eyelid dynamic driving screen brightness adjustment method provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the composition of a multi-parameter eyelid dynamic driving screen brightness adjustment system provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0024] Example 1

[0025] like Figure 1 As shown, Embodiment 1 of the present invention discloses a screen brightness adjustment method based on multi-parameter eyelid dynamic driving, comprising the following steps: Step S1: The user's facial video stream is captured in real time by the front-facing camera of the mobile terminal, and the user's eyelids are identified and tracked from the video stream based on a lightweight neural network model to generate eyelid movement trajectory data.

[0026] Specifically, step S1 includes the following sub-steps to achieve high-precision, low-latency eyelid positioning: Step S11: Hardware configuration and video stream capture.

[0027] This invention can be implemented on a tablet computer with typical configurations such as running the Android operating system, equipped with an octa-core ARM Cortex application processor, a dedicated NPU (Neural Processing Unit), and a high-resolution front-facing camera. The front-facing camera module is configured to output an RGB video stream with a resolution of 640 pixels × 480 pixels and a frame rate of 30 frames per second. This specification effectively controls data transmission and processing overhead while ensuring sufficient image detail, laying the foundation for real-time processing.

[0028] Step S12: Eyelid key point recognition based on lightweight convolutional neural network.

[0029] The acquired raw video frames first undergo a preprocessing stage, where a Haar cascade classifier or a lightweight face detector based on MobileNet-SSD is used to quickly locate the user's facial bounding box. Subsequently, the system crops the region of interest (ROI) including the eyes based on facial geometric features. This ROI is then scaled to the standard input size of the neural network model using bilinear interpolation: 224 pixels × 224 pixels × 3 channels (RGB).

[0030] One of the core components of this invention is a lightweight eyelid keypoint detection neural network optimized for real-time inference depth on mobile devices. The network is structured as an encoder-regressor architecture, with the input being the aforementioned 224×224×3 eye ROI image and the output being a 24-dimensional vector representing the normalized coordinates of 12 keypoints in both eyes.

[0031] S121: Feature-encoding backbone network (graph convolutional layer).

[0032] The first part of the network is the feature encoding backbone network, which extracts deep semantic features from the input image. This invention uses MobileNetV2 as the backbone network. The MobileNetV2 core consists of multiple stacked inverted residual bottleneck layers, including depthwise separable convolutions, which can significantly reduce the number of model parameters and computational cost while maintaining high feature extraction capabilities. Specifically, the input image passes through a standard convolutional layer and multiple bottleneck layers sequentially, ultimately outputting a 7×7×1280-dimensional high-order feature map. This feature map It contains rich spatial and contextual information needed to locate key points on the eyelids and can be regarded as a “shallow” or “basic” feature representation of the image.

[0033] S122: Key point coordinate regression head (fully connected regression layer).

[0034] S122: Key point coordinate regression head (fully connected regression layer).

[0035] The second part of the network is the keypoint coordinate regression head, which is responsible for extracting high-dimensional features from the backbone network. This is mapped to specific coordinate values. First, the feature map... After passing through a global average pooling layer, its spatial dimension is compressed to 1×1 while retaining the channel dimension (1280 dimensions), resulting in a global feature vector. This operation enhances the translation invariance of the features and reduces the number of parameters required for subsequent calculations.

[0036] Subsequently, It is fed into a regression module consisting of two fully connected layers: The first fully connected layer reduces the 1280-dimensional input to 512 dimensions and introduces nonlinearity using the ReLU activation function.

[0037] The second fully connected layer (output layer): reduces the 512-dimensional input to a final 24-dimensional output, with no activation function. This 24-dimensional output corresponds to... , representing the normalized coordinates (range [0,1]) of 6 key points for each of the left and right eyes. These 6 key points are carefully designed and typically include: the inner corner of the upper eyelid, the apex (highest point) of the upper eyelid, the outer corner of the upper eyelid, the outer corner of the lower eyelid, the bottom (lowest point) of the lower eyelid, and the inner corner of the lower eyelid. These points are sufficient to accurately delineate the eyelid contour and calculate the opening and closing angle.

[0038] To meet the stringent requirements of mobile deployment, the above-mentioned basic model has been specifically optimized: Channel pruning: After model training converges, by evaluating the importance of channels in each convolutional layer of the backbone network (e.g., using the L1 norm of the channels), approximately 40% of redundant or low-contribution channels are removed. This reduces the model computation from approximately 300 MMACs (multiply-accumulate operations) to approximately 180 MMACs with an accuracy loss of less than 0.5%.

[0039] INT8 quantization: The weights and activation values ​​of the trained FP32 (32-bit floating-point) model are quantized to INT8 (8-bit integer) precision after calibration. This process compresses the model size from approximately 9.8MB to approximately 2.5MB and can fully utilize the integer computing units of mobile CPUs / GPUs / NPUs, improving inference speed by approximately 2-3 times.

[0040] Training the neural network is a separate but crucial stage. Training data comes from publicly available facial landmark datasets (such as 300-W and WFLW) and a self-built dataset of eye images containing different ethnicities, lighting conditions, poses, and blink states, totaling over 50,000 images. Each image is meticulously annotated by experts with the aforementioned 12 landmarks (6 for each eye). During training, mean squared error (MSE) is used as the loss function, optimized using the Adam optimizer, with an initial learning rate set to 10. -3 The model is equipped with a cosine annealing strategy. Data augmentation techniques are widely used, including random rotation (±15 degrees), scaling, translation, brightness and contrast adjustment, and simulated motion blur, to greatly improve the robustness of the model. The fully trained model has an average keypoint localization error of less than 2.5 pixels (relative to a 224×224 input size) on the independent test set, meeting the high accuracy requirements.

[0041] During deployment, the optimized model is converted to a mobile-specific format using inference frameworks such as TensorFlowLite or ONNXRuntime and integrated into the camera driver layer. On the aforementioned hardware platform, the average inference time for a single frame can be stably controlled within 8 milliseconds, achieving true real-time processing.

[0042] Step S13: Trajectory generation and data buffering.

[0043] The normalized coordinates output by the model are transformed back to the pixel coordinates of the original eye ROI, and then mapped back to the global camera coordinate system. By temporally associating the predicted eyelid keypoint coordinates in consecutive video frames and using a Kalman filter for smoothing and prediction, jitter in single-frame detection is effectively suppressed, resulting in a stable eyelid motion trajectory with high temporal resolution. The state vector of the Kalman filter contains the position and velocity of the keypoints, and its process noise and measurement noise matrices are set empirically to balance the smoothness of tracking and response speed.

[0044] The processed trajectory data (i.e., the 24-dimensional coordinate vector of each frame along with a timestamp) is written in real-time to a shared memory buffer managed by the operating system kernel. This buffer is designed as a 128KB ring structure, employing a dual-mapping technique to exist simultaneously in kernel space and user space, thereby achieving zero-copy data exchange and significantly reducing data transmission latency. The buffer follows a first-in, first-out (FIFO) principle, automatically overwriting the oldest data when it is full, ensuring that it always retains the eyelid trajectory data from the most recent approximately 2 seconds (corresponding to 60 frames), providing a time window for subsequent analysis.

[0045] Through step S1 and its sub-steps, the system establishes a high-precision eyelid motion sensing channel updated at a frequency of 30Hz, providing a reliable raw data source for subsequent fatigue analysis, and with extremely low end-to-end processing latency.

[0046] Step S2: Based on the eyelid movement trajectory data, extract multidimensional eyelid dynamic parameters in real time. The multidimensional eyelid dynamic parameters include at least blink frequency, blink duration, and blink amplitude.

[0047] Specifically, a background service process (multi-parameter extraction module) running independently in user space accesses the aforementioned shared buffer via memory mapping and is woken up at fixed intervals (e.g., 100 milliseconds) to execute step S2, which includes the following parameter calculation logic: Step S21: Calculate the eyelid opening and closing degree.

[0048] Eyelid opening / closing is a core intermediate variable. For each frame and each eye, physical eyelid opening / closing is defined by calculating the vertical distance (Y-axis direction) between the midpoint of the upper eyelid (averaged by the medial and lateral points of the upper eyelid) and the midpoint of the lower eyelid (averaged by the medial and lateral points of the lower eyelid) based on the coordinates of six key points. To eliminate the influence of individual differences in anatomical structure (such as different eye sizes), the system dynamically maintains a short-term (e.g., 1 second) baseline value for full eyelid opening. Normalized eyelid opening / closing (EOD) is calculated as the ratio of the current frame's physical eyelid opening / closing to this baseline value, with the result constrained to the [0,1] interval, where 1 represents a fully open state. The baseline value is updated slowly to adapt to minor changes in the user's pose.

[0049] Step S22: Blink event detection and parameter extraction.

[0050] The system monitors the EOD curve in real time and defines a complete blink event as meeting the following conditions: the EOD sequence first decreases from a high value (judgment threshold, e.g., >0.85), passes through a local minimum significantly below the closure threshold (e.g., <0.15), and then successfully rises back to a high value threshold (>0.85). This three-stage judgment can effectively distinguish blinking from other micro-eye movements.

[0051] Blink Duration (BD): Measured in milliseconds. Precisely recorded as the time interval during a single blink event from when the EOD first falls below a high threshold (0.85) to when it rises above that threshold again. This reflects the length of time the eye muscles remain closed.

[0052] Blink amplitude (BA): Expressed as a percentage. Calculated as the difference between the local minimum of EOD and the current fully open baseline value during a single blink event, divided by the baseline value. The formula is: ,in This represents the local minimum of EOD during a single blink. The larger the value, the more completely the eyelid closes.

[0053] Blink frequency (BF): The unit is blinks per minute. The system maintains a sliding time window of 60 seconds, counts the number of confirmed complete blink events within the window in real time, and obtains an estimate of the number of blinks per minute through linear interpolation.

[0054] Step S23: Parameter packaging and transmission.

[0055] At the end of each calculation cycle (100ms), the calculated blink frequency, blink duration, blink amplitude, and the current frame EOD value generated in step S21 are packaged into a compact, byte-aligned data structure (e.g., a 64-byte data packet). Simultaneously, the module generates status flags based on preset physiological thresholds; for example, when a single instance of "BD > 100ms and BA > 80%" is detected, the "deep fatigue event" flag is set. This data packet is then sent to the coprocessor responsible for weight calculation via an efficient direct memory access (DMA) channel. This DMA channel is reserved and initialized by the device tree at system startup, with a bandwidth locked at 50Mbps, and the measured end-to-end transmission latency is consistently within 2-5 milliseconds.

[0056] Step S3: Based on the real-time change rate of each of the multidimensional eyelid dynamic parameters, dynamically calculate and allocate the contribution weight corresponding to each parameter. When the real-time change rate of any parameter exceeds its corresponding preset mutation threshold, the contribution weight of that parameter is increased.

[0057] Specifically, step S3 is executed by a module deployed on the NPU coprocessor or a dedicated real-time processing unit to ensure microsecond-level decision-making speed, and it includes the following core sub-steps: Step S31: Calculate the rate of change.

[0058] The module maintains an independent sliding history window of 5 seconds (with 50 samples at 100ms sampling points) for each dynamic parameter (blink frequency, blink duration, blink amplitude). When a new parameter packet is received, the corresponding history window is first updated, and the arithmetic mean of the data within that window is calculated. and standard deviation The calculation of the real-time rate of change Δ strictly follows the formula: Standard deviation This measures the inherent volatility of the parameter. If the data within the historical window is completely stable ( =0), and to avoid division by zero errors, Δ is forcibly defined as 0. This standardized rate of change effectively measures the deviation of current behavior from the recent "baseline" or "normality" and is extremely sensitive to sudden changes. Based on the real-time rate of change of each multidimensional eyelid dynamic parameter, the contribution weights corresponding to each parameter are dynamically calculated and assigned. ,in These represent the dynamic weights of blink frequency, blink duration, and blink amplitude in the comprehensive fatigue index, respectively, and satisfy the following conditions: .

[0059] Step S32: Threshold-based dynamic weight decision.

[0060] The system presets a critical threshold for sudden changes in the rate of change. This threshold, calibrated based on extensive experimental data, reflects the critical point at which a parameter undergoes a "qualitative change." The weighted decision logic is a rule combining hard switching and smooth switching. Deep fatigue response (highest priority): the rate of change in blink duration. Greater than the threshold (If the empirical value is set to 1.5), it is determined that the user may be experiencing deep fatigue (e.g., starting to doze off). The system immediately sets the original weight vector to... This configuration significantly increases the decision weight of the blink duration parameter to 70%, making the system extremely sensitive to "long eye closure" signals and driving subsequent modules to make a large and rapid reduction in brightness.

[0061] Acute discomfort response: the rate of change in blink frequency Greater than the threshold (If the empirical value is set to 2.0), it is determined that it may be acute visual discomfort (such as rapid blinking caused by screen glare, dry eyes, etc.). In this case, the original weight vector is set to... They shifted their decision-making focus to blink frequency.

[0062] Normal smooth adjustment: If none of the above mutations occur, the default baseline weight vector is used. This reflects the balanced contribution of the three parameters in the accumulation of normal fatigue.

[0063] Step S33: Weight smoothing and distribution.

[0064] To prevent sudden changes in weights, the original weights determined in the decision are... Apply first-order exponential smoothing filter: .in, This represents the smoothed weight vector output in the current calculation period t, which is ultimately used for brightness mapping; This represents the smoothing weight vector output in the previous calculation cycle (t-1), which serves as the historical benchmark for this smoothing calculation. This represents the original weight vector generated by the threshold decision logic in the current period t; The smoothing coefficient is a configurable hyperparameter with a value range of (0,1). In this embodiment, it is set to... =0.95, its function is to act as a low-pass filter, so that the weight vector of the final output changes continuously and smoothly, avoiding high-frequency jitter in the adjustment command.

[0065] The final smoothed weight vector is encapsulated into an instruction packet and sent to the screen driver controller via the I²C bus (configured in fast mode+, clock frequency 1MHz). The communication protocol is custom-defined and includes a start bit, device address, weight data payload (typically each weight is represented by 10-12 bits), and stop bit, ensuring a single transmission latency of less than 2 milliseconds.

[0066] Step S4: Input the multidimensional eyelid dynamic parameters after the contribution weighting process into a preset nonlinear brightness mapping function to calculate the target screen brightness value.

[0067] Specifically, step S4 is implemented in the internal microcontroller firmware of the screen driver controller to ensure low latency and determinism of brightness control at the hardware level, including: Step S41: Parameter normalization and basic brightness acquisition.

[0068] The screen driver controller receives weights from the coprocessor via the I²C slave interface. Meanwhile, the driver IC periodically (e.g., every second) obtains normalized eyelid parameters (f, d, a) from the application processor via a low-speed asynchronous serial port (UART, baud rate 115200). The normalization rule aims to unify parameters of different dimensions into the [0,1] interval: 30 times / minute is the upper limit of the typical fatigue threshold determined by clinical studies.

[0069] 500 milliseconds is the empirical upper limit for extreme eye-closed duration.

[0070] The magnitude itself is already a percentage.

[0071] Meanwhile, the driver IC uses another dedicated I / O pin. 2 Channel C reads the illuminance value (in lux) from the ambient light sensor and maps it linearly or non-linearly to the base brightness value by consulting a pre-set lookup table of 256 entries in the firmware. This lookup table is calibrated based on the characteristics of human vision (Weber-Fechner law) and screen characteristics, and is the basis for the screen's automatic brightness function.

[0072] Step S42: Calculate the Sigmoid nonlinear mapping.

[0073] Target brightness Calculated by the following Sigmoid function:

[0074] Among them, the comprehensive fatigue index , is a linear combination of weighted eyelid features; k is a device calibration coefficient, an adjustable shape factor used to control the "steepness" of the Sigmoid function. In a preferred embodiment, k can be dynamically fine-tuned according to ambient light: in low light environments (<50 Lux), where the human eye is more sensitive to changes in brightness, k=1.0 is set to produce a smoother adjustment; in high light environments (>1000 Lux), k=0.6 is set to make the adjustment response steeper; and in moderate light environments, k=0.8 is used. The "slow-steep-slow" characteristics of this function perfectly match the nonlinearity of human eye's perception of fatigue: in the early stages of fatigue (small S), the adjustment is gentle to avoid interference; when fatigue accumulates to the critical range (medium S), the adjustment sensitivity is highest; and in extreme fatigue (large S), the brightness approaches the lower limit of saturation, avoiding excessive dimming that affects viewing.

[0075] Step S5: Adjust the screen brightness of the mobile terminal according to the target screen brightness value.

[0076] Specifically, step S5 is executed by the brightness control unit of the screen driver controller, and its core is a stepped smooth adjustment algorithm to ensure visual comfort and avoid a jarring effect: Step S51: Calculate and judge the difference.

[0077] Calculate target brightness With the current actual screen brightness absolute difference .

[0078] Step S52: Stepped smooth adjustment (normal mode).

[0079] like Exceeding the adjustment threshold (e.g., 0.1) If the brightness is reduced to 10% of the base brightness, the change will not be executed immediately. The system divides the total adjustment into N equal steps (typically N=3). Then, a high-precision hardware timer (such as SysTick) is started, and the brightness is changed by one step at fixed intervals T (e.g., T=50 milliseconds). After NT time, it smoothly transitions to the target value. This gradual change completely eliminates the visual flicker and discomfort caused by sudden changes in brightness.

[0080] Step S53: Rapid response to deep fatigue (priority mode).

[0081] In a preferred embodiment, if the "deep fatigue event" flag is parsed from the data packet sent by the coprocessor, the system immediately enters a fast response mode. This mode performs two operations: first, it calculates... The value is forcibly limited to no more than 0.3. Within the safe range; secondly, take over the step adjustment controller and significantly shorten the step interval time (e.g., from 50 milliseconds to 20 milliseconds), and the number of steps may be increased (e.g., 5 steps), so as to complete the rapid and smooth transition from the current brightness to the safe brightness within a total of about 200 milliseconds, and realize timely and powerful intervention for deep fatigue states such as drowsiness.

[0082] In summary, through the closed-loop process of steps S1 to S5 described above, this invention achieves a complete technical chain from precise physiological signal perception, intelligent weight decision-making, physiologically consistent nonlinear mapping to final smooth light output. This method surpasses traditional methods based on ambient light or simple thresholds, achieving truly adaptive and personalized brightness adjustment centered on the user's real-time physiological state.

[0083] Example 2 like Figure 2 As shown, Embodiment 2 of the present invention discloses a multi-parameter eyelid dynamic-driven screen brightness adjustment system, comprising the following five main functional modules: Module M10, Eyelid Dynamic Recognition Module: Used to collect user facial video stream in real time through the front-facing camera of the mobile terminal, and to identify and track user eyelids from the video stream based on a lightweight neural network model to generate eyelid movement trajectory data.

[0084] Specifically, module M10 is a deeply coupled hardware and software component. The hardware core consists of a front-facing camera sensor and its associated image signal processor. The software portion is a kernel-loadable module deeply integrated into the operating system (such as Linux) V4L2 camera driver framework. This module handles the raw image data from the camera, and its core is the optimized and transformed neural network model file described in Example 1. This model is driven by an NPU or GPU for high-speed inference. The module runs with the highest hardware interrupt priority, ensuring that each frame is processed promptly without frame loss. Its output keypoint coordinate data is directly written to a dual-mapped circular buffer shared with module M20 through the kernel's memory management mechanism, achieving efficient zero-copy data transfer across the kernel / user space boundary—a key design feature for reducing overall system latency.

[0085] Module M20, multi-parameter extraction module: used to extract multi-dimensional eyelid dynamic parameters in real time based on the eyelid movement trajectory data.

[0086] Specifically, module M20 is a user-space software entity existing as an independent system daemon. It is initialized at system startup and maps the shared buffer created by module M10 into its own process address space via the mmap() system call. Internally, it contains logical components such as an event detection state machine, a sliding window statistician, and a parameter calculation unit. This process is periodically woken up (every 100ms) by a high-precision timer (e.g., timerfd) to execute a feature extraction algorithm. After calculation, instead of communicating via conventional sockets or pipes, it calls a low-level IOCTL command to trigger the DMA controller, directly writing the prepared parameter data packet into a pre-allocated, physically contiguous reserved memory area (also mapped into module M30's address space), thus completing the high-speed, low-overhead data delivery to the next-level module.

[0087] Module M30, Adaptive Weight Allocation Module: Used to dynamically calculate and allocate contribution weights corresponding to each parameter based on the real-time change rate of each of the multidimensional eyelid dynamic parameters.

[0088] Specifically, module M30 primarily runs on an NPU coprocessor or a dedicated real-time microcontroller. This processor has independent SRAM, firmware storage, and a rich set of peripherals. The firmware of module M30 includes a DMA receive engine, a rate-of-change calculation unit (with a floating-point unit), threshold comparison logic circuitry, and a weighted smoothing filter. It automatically retrieves data from the output area of ​​module M20 via a hardware DMA controller; the calculation process is entirely local and unaffected by the host operating system's scheduling. After the calculation is complete, the processor's internal I / O logic is configured... 2The C controller peripheral is set to master mode, sending weight instruction packets to the specified slave address (screen driver controller). Deployed on a dedicated processor, it ensures time determinism and low latency in algorithm execution, meeting the requirements of real-time decision-making.

[0089] Module M40, Nonlinear Brightness Mapping Module: Used to input the multidimensional eyelid dynamic parameters after the contribution weighting process into a preset nonlinear brightness mapping function to calculate the target screen brightness value.

[0090] Specifically, module M40 is an embedded functional unit within the screen driver integrated circuit, typically implemented by an ARM Cortex-M series microcontroller embedded in the driver IC and its running firmware. This firmware comprises several parts: I 2 The system includes a slave communication protocol stack (for receiving weights), a UART driver (for receiving normalized parameters), an ambient light sensor interface driver, a fast Sigmoid function calculation library (possibly implemented using a combination of lookup tables and linear interpolation), a brightness mapping lookup table, and a step-wise adjustment state machine. Module M40 is the execution center of the hardware adjustment; it integrates all input information and ultimately generates the target brightness value and the corresponding PWM control sequence.

[0091] Module M50, screen brightness adjustment execution module: used to generate a control signal to adjust the screen brightness of the mobile terminal based on the target screen brightness value.

[0092] Specifically, module M50 is the backlight driving circuit or the pixel current driving circuit of the OLED located inside the screen module. It receives PWM signals or digital dimming commands from module M40, and through a sophisticated power management chip, converts them into precisely controlled current or voltage, thereby physically driving the LED backlight strip of the LCD or each pixel of the OLED to achieve the final change in screen brightness.

[0093] The system's connectivity follows a clearly defined data pipeline: modules M10 and M20 are connected via a 128KB shared memory buffer for zero-copy data transfer; modules M20 and M30 are connected via a dedicated 50Mbps DMA channel for high-speed, stable parameter delivery; modules M30 and M40 are connected via a 1Mbps I²C bus for reliable control command issuance; and modules M40 and M50 are connected via a PWM signal line for final light output control. Additionally, an ambient light sensor is connected to module M40 via an auxiliary I²C bus.

[0094] Through the aforementioned highly collaborative, hierarchical modular design and meticulously optimized data pathways, this system achieves extremely low end-to-end latency. From the moment the camera captures a frame to the screen backlight responding to a brightness change signal, the total latency of the entire process, as tested on a real hardware platform, can be strictly controlled to less than 150 milliseconds. This performance makes the entire adjustment process almost instantaneous for the user, achieving a seamless yet effective improvement in comfort.

[0095] In addition, the system includes a global anomaly handling and monitoring unit (not shown in the figure). This unit continuously monitors the camera's working status (e.g., whether it is obstructed or malfunctioning), the continuity of the data stream, and the heartbeat signals of each module. When a camera is detected to be continuously obstructed (e.g., by a user's hand), this unit triggers the system to switch to a robust degradation mode: In this mode, module M30 suspends weight updates based on eyelid data, and module M40 ignores input from the eyelid analysis path, only performing very conservative, small-amplitude brightness adjustments based on ambient light sensor data and historical eyelid parameter characteristics within the most recent valid time period (e.g., the past 30 seconds). Simultaneously, the system resets its internal dynamic weights to default values ​​every 5 minutes to prevent irreversible drift of the algorithm state under abnormal input conditions, thus ensuring the overall robustness and reliability of the system in complex real-world environments.

[0096] In summary, the system described in Example 2 fully and comprehensively embodies the method flow of Example 1 through its specific and feasible modular division, clear deployment strategy, clear interface design, and extreme performance optimization.

[0097] The foregoing has shown and described the basic principles, main features, and advantages of this invention. Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for adjusting screen brightness using multi-parameter eyelid dynamic driving, characterized in that, Includes the following steps: Step S1: The user's facial video stream is captured in real time by the front-facing camera of the mobile terminal, and the user's eyelids are identified and tracked from the video stream based on a lightweight neural network model to generate eyelid movement trajectory data; Step S2: Based on the eyelid movement trajectory data, extract multidimensional eyelid dynamic parameters in real time. The multidimensional eyelid dynamic parameters include at least blinking frequency, blinking duration, and blinking amplitude. Step S3: Based on the real-time change rate of each of the multidimensional eyelid dynamic parameters, dynamically calculate and allocate the contribution weight corresponding to each parameter. When the real-time change rate of any parameter exceeds its corresponding preset mutation threshold, the contribution weight of that parameter is increased. Step S4: Input the multidimensional eyelid dynamic parameters after the contribution weighting process into a preset nonlinear brightness mapping function to calculate the target screen brightness value; Step S5: Adjust the screen brightness of the mobile terminal according to the target screen brightness value.

2. The method according to claim 1, characterized in that, In step S3, the real-time rate of change The calculation formula is: The historical parameter mean and historical parameter standard deviation are calculated based on historical parameter data within a preset time period.

3. The method according to claim 2, characterized in that, Step S3 specifically includes: calculating the real-time rate of change of the blink duration parameter. ,like If the value exceeds the first preset threshold, the weight vector will be assigned as: blink frequency weight. Within the first low value range, the weight of blink duration Within the highest value range, the weight of blink amplitude Located in the second lowest value range; and / or calculating the real-time rate of change of the blink frequency parameter. ,like If the value exceeds the second preset threshold, the weight vector will be assigned as: blink frequency weight. The second highest value range, blink duration weight Located in the third lowest value range, blink amplitude weighting It is in the fourth low value range; the first preset threshold is 1.5, the first high value range is not less than 0.7; the second preset threshold is 2.0, the second high value range is not less than 0.

6.

4. The method according to claim 1, characterized in that, In step S4, the preset nonlinear brightness mapping function is a Sigmoid function, and its expression is: ,in, The target screen brightness value. The base brightness reference value is given, and k is a shape adjustment factor ranging from 0.6 to 1.

0. , respectively, represent the contribution weights of blink frequency, blink duration, and blink amplitude, and f, d, and a are the normalized blink frequency, blink duration, and blink amplitude values, respectively.

5. The method according to claim 1, characterized in that, Step S5 specifically includes: calculating the difference between the target screen brightness value and the current screen brightness value; if the absolute value of the difference exceeds 10%, the brightness adjustment process is divided into at least 3 steps, with each step executed once every 50 milliseconds, so as to adjust the current screen brightness value to the target screen brightness value step by step.

6. The method according to claim 1, characterized in that, The method further includes: when a single blink lasts for more than 100 milliseconds and the blink amplitude exceeds 80%, it is determined as a deep fatigue event; in response to the deep fatigue event, the screen brightness of the mobile terminal is controlled to decrease to a level not exceeding 30% of the basic brightness reference value within 200 milliseconds.

7. A multi-parameter eyelid dynamic-driven screen brightness adjustment system, integrated into a mobile terminal, characterized in that, include: The eyelid dynamic recognition module is used to collect the user's facial video stream in real time through the front-facing camera of the mobile terminal, and to identify and track the user's eyelids from the video stream based on a lightweight neural network model to generate eyelid movement trajectory data. The multi-parameter extraction module is used to extract multi-dimensional eyelid dynamic parameters in real time based on the eyelid movement trajectory data. The multi-dimensional eyelid dynamic parameters include at least blinking frequency, blinking duration, and blinking amplitude. An adaptive weight allocation module is used to dynamically calculate and allocate contribution weights corresponding to each parameter based on the real-time change rate of each of the multidimensional eyelid dynamic parameters. When the real-time change rate of any parameter exceeds its corresponding preset mutation threshold, the contribution weight of that parameter is increased. The nonlinear brightness mapping module is used to input the multidimensional eyelid dynamic parameters after the contribution weighting process into a preset nonlinear brightness mapping function to calculate the target screen brightness value. The screen brightness adjustment execution module is used to generate a control signal to adjust the screen brightness of the mobile terminal based on the target screen brightness value.

8. The system according to claim 7, characterized in that, The adaptive weight allocation module is deployed in the AI ​​coprocessor of the mobile terminal or a dedicated processing unit that supports direct memory access (DMA); the nonlinear brightness mapping module is integrated in the screen driver controller firmware of the mobile terminal.

9. The system according to claim 7 or 8, characterized in that, The eyelid dynamic recognition module is configured to process video streams with a frame rate of not less than 30fps in real time; and the multi-parameter extraction module, the adaptive weight allocation module, and the nonlinear brightness mapping module are configured to operate in a collaborative pipeline manner.

10. The system according to claim 7, characterized in that, The eyelid dynamic recognition module and the multi-parameter extraction module interact with each other through a shared memory buffer; the multi-parameter extraction module and the adaptive weight allocation module are connected through a DMA channel; the adaptive weight allocation module and the nonlinear brightness mapping module are connected through an I²C bus.

Images