Vehicle color-changing glass control method, device, apparatus, and medium

Abstract

This application relates to the technical field of vehicle control, and in particular to a method, device, equipment, and medium for controlling vehicle tinted windows. The method includes: acquiring multidimensional cognitive load data of a target user and multimodal data of the vehicle; inputting the multidimensional cognitive load data and multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within a future time window; constructing a dynamic game objective function based on the cognitive vulnerability window probability distribution, and obtaining the optimal transmittance change trajectory by solving the objective function; the objective function includes at least an environment matching loss term and a cognitive interference loss term, and the weight coefficients of the two terms are dynamically adjusted according to the cognitive vulnerability window probability distribution; and controlling the vehicle tinted windows to adjust transmittance according to the optimal transmittance change trajectory. This method can respond promptly to environmental changes, pre-adjust the transmittance of the vehicle windows, and reduce the occurrence of safety accidents.

Description

Technical Field

[0001] This application relates to the field of vehicle control, and in particular to methods, apparatus, equipment and media for controlling vehicle tinted glass. Background Technology

[0002] With the development of vehicle technology, tinted glass control technology is no longer satisfied with simple light response and is developing towards intelligence and systematization. Most systems generally adjust the light transmittance after detecting environmental changes through cameras, or achieve gradual changes in light transmittance by recognizing landmarks along the route. When sudden environmental changes occur, such as strong light at the tunnel exit, the driver's eyes need a short period of time to complete physiological adaptation. Even if the existing system can respond instantly, it cannot compensate for the physiological blind spot. More seriously, if the driver is already fatigued, their adaptability will further decline, potentially leading to safety accidents. Therefore, there is an urgent need for a method, device, equipment, and medium for controlling vehicle tinted glass. Summary of the Invention

[0003] This application provides a method, apparatus, device, and medium for controlling vehicle tinted glass, which can respond promptly to environmental changes, pre-adjust the light transmittance of vehicle glass, and reduce the occurrence of safety accidents.

[0004] On one hand, embodiments of this application provide a method for controlling tinted glass in a vehicle, the method comprising: Acquire multidimensional cognitive load data of target users and multimodal data of vehicles; The multidimensional cognitive load data and the multimodal data are input into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows; the probability distribution of cognitive vulnerability windows is used to characterize the probability that the driver's cognitive load will exceed the personalized load baseline due to environmental changes at various future time points; Based on the probability distribution of the cognitive vulnerability window, a dynamic game objective function is constructed, and the optimal transmittance change trajectory is obtained by solving the objective function; the objective function includes at least an environment matching loss term and a cognitive interference loss term, and the weight coefficients of the two terms are dynamically adjusted according to the probability distribution of the cognitive vulnerability window; The vehicle's tinted glass is controlled to adjust its light transmittance based on the optimal light transmittance change trajectory.

[0005] Optionally, the step of inputting the multidimensional cognitive load data and the multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows includes: Principal component analysis is performed on the multidimensional cognitive load data to reduce dimensionality and extract the load principal components that characterize the driver's comprehensive cognitive state. At the same time, the multimodal data is subjected to time-series alignment and normalization to generate an environmental state vector. The principal component of the load and the environmental state vector are concatenated according to timestamps to form a multidimensional temporal input sequence. The multidimensional temporal input sequence is then input into the bidirectional long short-term memory network layer of the cognitive load-environment spatiotemporal coupling model to extract the temporal dependency features between environmental changes and cognitive load along the forward and reverse directions, respectively. The temporal dependency features are input into the attention mechanism layer of the cognitive load-environment spatiotemporal coupling model. The attention mechanism layer automatically calculates the importance weight of each historical moment to the current prediction and sums them up to obtain a context vector. The context vector represents the historical environmental change information that has the greatest impact on cognitive load. The context vector is fused with the input at the current moment through a fully connected layer to predict the cognitive load prediction sequence for the next N moments, and at the same time output the sensitivity amplification coefficient sequence for the next N moments. The sensitivity amplification coefficient is used to characterize the driver's cognitive response intensity multiple to a unit change in the environment under the current load state. The sequence of predicted cognitive load values ​​is compared one by one with a pre-established personalized load baseline. The degree to which the predicted value exceeds the baseline at each time point is calculated. The degree to exceed the baseline is then nonlinearly amplified by the sensitivity amplification factor to generate the probability distribution of the cognitive vulnerability window within the future time window.

[0006] Optionally, before comparing the cognitive load prediction sequence one by one with a pre-established personalized load baseline and calculating the degree to which the prediction value exceeds the baseline at each time point, the method further includes: Collect data on the user's gaze dispersion, pupil diameter change rate, and braking reaction time offset during daily driving. A Gaussian mixture model was used to fit the probability distribution of each index; The baseline threshold is the standard deviation of each indicator plus a preset multiple. Personalized load limits are constructed based on the baseline threshold.

[0007] Optionally, constructing the dynamic game objective function based on the cognitive vulnerability window probability distribution includes: Based on the probability distribution of the cognitive vulnerability window at the current moment, calculate the first weight coefficient of the environment matching loss term and the second weight coefficient of the cognitive interference loss term, respectively, wherein the first weight coefficient is inversely proportional to the second weight coefficient and the second weight coefficient is directly proportional to the probability distribution of the cognitive vulnerability window; The ideal transmittance for environmental matching at each future moment is calculated based on the multimodal data. Calculate the comprehensive cognitive load index based on the multidimensional cognitive load data; The dynamic game objective function is constructed based on the first weighting coefficient, the second weighting coefficient, the ideal light transmittance for environmental matching, and the comprehensive cognitive load index.

[0008] Optionally, the multidimensional cognitive load data includes fixation dispersion, pupil diameter change rate, and braking reaction time offset. The calculation of the comprehensive cognitive load index based on the multidimensional cognitive load data includes: The gaze dispersion, pupil diameter change rate, and braking reaction time offset were normalized to preset ranges. Principal component analysis was used to extract the first principal component as the initial loading index. The preliminary load index is multiplied by the sensitivity amplification factor output from the cognitive load-environment spatiotemporal coupling model to obtain the final comprehensive index. The final comprehensive index is determined as the cognitive load comprehensive index.

[0009] Optionally, the step of calculating the ideal transmittance for environmental matching at future times based on the multimodal data includes: Obtain the driving intent of the target user; Based on the light intensity, light angle, weather conditions and road type in the multimodal data, the basic value is obtained by querying the preset transmittance mapping table; The baseline value is adjusted based on the driving intention at a future moment to obtain the ideal transmittance for environmental matching; The correction coefficients are obtained through offline reinforcement learning training, which ensures that the light transmittance has the greatest correlation with the driver's subjective comfort score under different driving intentions.

[0010] Optionally, controlling the light transmittance adjustment of the vehicle's photochromic glass according to the optimal light transmittance change trajectory includes: Based on the cognitive vulnerability window probability distribution, the control parameters of the controller are determined; The controller is adjusted based on the control parameters to obtain an optimized controller; The target value at the current moment in the optimal transmittance change trajectory is used as the set value of the controller, and the current actual transmittance is used as the feedback value. The driving voltage control quantity is calculated by the optimized controller and output to the electrochromic glass driving circuit to realize the gradual tracking of transmittance to the target trajectory.

[0011] On the other hand, embodiments of this application provide a vehicle tinted glass control device, the device comprising: The acquisition module is used to acquire multidimensional cognitive load data of target users and multimodal data of vehicles; The input module is used to input the multidimensional cognitive load data and the multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows; the probability distribution of cognitive vulnerability windows is used to characterize the probability that the driver's cognitive load will exceed the personalized load baseline due to environmental changes at various future time points; A construction module is used to construct a dynamic game objective function based on the probability distribution of the cognitively vulnerable window, and to obtain the optimal transmittance change trajectory by solving the objective function; the objective function includes at least an environment matching loss term and a cognitive interference loss term, and the weight coefficients of the two terms are dynamically adjusted according to the probability distribution of the cognitively vulnerable window; The control module is used to control the vehicle's tinted glass to perform light transmittance adjustment according to the optimal light transmittance change trajectory.

[0012] In another aspect, embodiments of this application provide an electronic device, the device including: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the vehicle tinted glass control method as described in the first aspect.

[0013] In another aspect, embodiments of this application provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the vehicle tinted glass control method as described in the first aspect.

[0014] The vehicle tinted glass control method, apparatus, device, and medium of this application can predict the driver's cognitive vulnerability window within a future time window. Therefore, this method can construct a dynamic game objective function and dynamically adjust the weights of environmental matching and cognitive interference according to the driver's cognitive vulnerability, thereby generating an optimal transmittance change trajectory that balances environmental adaptability and driver cognitive load. This method effectively avoids the excessive cognitive load or visual discomfort that may occur to the driver during sudden environmental changes in traditional systems. Especially when the driver is fatigued, it can proactively optimize the in-vehicle lighting environment and reduce driving risks. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a vehicle tinted glass control method provided in an embodiment of this application; Figure 2 This is a structural block diagram of a vehicle tinted glass control device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0016] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0018] For ease of understanding, the following explains some key terms in this embodiment: Multidimensional cognitive load data refers to a collection of various physiological or behavioral indicators that reflect a driver's current cognitive state. This data may include, but is not limited to, driver eye-tracking data, physiological signals, and driving behavior data. By comprehensively analyzing this multidimensional data, the driver's cognitive load level can be assessed.

[0019] Multimodal data refers to various types of data acquired from the vehicle's internal and external environment. This data may include, but is not limited to, ambient lighting information, weather information, road information, and in-vehicle environmental parameters collected by vehicle sensors. By integrating these different modalities of data, a comprehensive characterization of the vehicle's driving environment can be achieved.

[0020] The cognitive load-environment spatiotemporal coupling model is a trained predictive model designed to analyze the complex temporal and spatial relationships between driver cognitive load and environmental changes. This model can accept multidimensional and multimodal cognitive load data as input and predict trends in driver cognitive load over specific future time windows, with particular focus on the impact of environmental changes on cognitive load.

[0021] The probability distribution of the cognitive vulnerability window refers to the distribution of the probability at which a driver's cognitive load exceeds their personalized load baseline due to environmental changes within a future period. This distribution quantifies the cognitive challenges a driver may face at each future moment in probabilistic form; a higher probability value indicates that the driver is more likely to experience cognitive overload at that moment.

[0022] A personalized cognitive load baseline is a reference standard established for each driver to measure their normal cognitive load level. This baseline is customized based on individual driver differences and historical driving data. When a driver's real-time cognitive load exceeds this baseline, it indicates that they may be in a state of excessive or unhealthy cognitive load.

[0023] The dynamic game objective function is a mathematical optimization function designed to balance the relationship between environmental comfort and driver cognitive load. This function comprehensively considers environmental matching loss and cognitive interference loss, and dynamically adjusts the weights of each component based on the driver's cognitive vulnerability window probability distribution, in order to minimize the risk of cognitive overload while ensuring driver comfort.

[0024] The optimal transmittance variation trajectory refers to the ideal sequence of light transmittance changes over time within a future time window, obtained by solving the objective function of a dynamic game. This trajectory represents the best adjustment path for the transmittance of the tinted glass while balancing environmental adaptability and driver cognitive load.

[0025] The environment matching loss term is a component of the dynamic game objective function, used to quantify the difference between the actual transmittance of the vehicle's tinted glass and the ideal transmittance corresponding to the current environmental conditions. This term aims to enable the tinted glass to better adapt its transmittance adjustment to the external environment, providing better visual comfort.

[0026] The cognitive interference loss term is another component of the dynamic game objective function, used to quantify the potential impact of changes in the transmittance of tinted glass on driver cognitive load. This term aims to minimize the increase in driver cognitive load that may be caused by transmittance adjustment, especially during periods when the driver's cognition is vulnerable.

[0027] Automotive photochromic glass refers to glass that can automatically adjust its light transmittance according to external environmental conditions or control signals. This glass typically employs electrochromic, thermochromic, or photochromic technologies to control the amount of light entering the vehicle by altering its optical properties, thereby improving driving comfort and safety.

[0028] To address the problems of existing technologies, this application provides a method, apparatus, device, and medium for controlling tinted glass in vehicles. In this application, the method can predict the driver's cognitive vulnerability window within a future timeframe. Therefore, the method can construct a dynamic game-theoretic objective function and dynamically adjust the weights of environmental matching and cognitive interference based on the driver's cognitive vulnerability, thereby generating an optimal transmittance change trajectory that balances environmental adaptability and driver cognitive load. This method effectively avoids the excessive cognitive load or visual discomfort that may occur in traditional systems during sudden environmental changes, especially when the driver is fatigued. It can proactively optimize the in-vehicle lighting environment and reduce driving risks.

[0029] The method for controlling vehicle tinted glass provided in the embodiments of this application will be described below.

[0030] Figure 1 A schematic flowchart of a vehicle tinted glass control method according to an embodiment of this application is shown. Figure 1 As shown, the vehicle tinted glass control method may include S101-S104: S101, acquire multidimensional cognitive load data of the target user and multimodal data of the vehicle.

[0031] In this embodiment, multidimensional cognitive load data can be acquired in various ways. For example, drivers can be required to periodically assess their subjective load during driving, or heart rate data can be collected by wearing simple physiological sensors (such as heart rate monitors) to indirectly reflect their load status. Vehicle multimodal data may include external ambient brightness information acquired through onboard sensors, and cabin temperature data acquired through in-vehicle sensors. The acquisition of this data forms the basis for subsequent analysis.

[0032] S102, inputting multidimensional cognitive load data and multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows.

[0033] In this embodiment, the cognitive load-environment spatiotemporal coupling model can be a statistical prediction model trained on historical data. For example, the model can be a multiple linear regression model, whose inputs are the current cognitive load data and environmental data, and whose output is a predicted value of the driver's cognitive load at a fixed future time point. This predicted value is then compared with a preset general load threshold to determine whether the driver is likely to exceed the load, thereby generating a simple binary vulnerability indicator. The cognitive vulnerability window probability distribution is used to characterize the probability that the driver's cognitive load will exceed the personalized load baseline at various future time points due to environmental changes. The personalized load baseline can be a fixed load upper limit set according to the average level of the driver group.

[0034] S103, based on the probability distribution of the cognitively vulnerable window, constructs a dynamic game objective function, and obtains the optimal transmittance change trajectory by solving the objective function.

[0035] In this embodiment, the objective function includes at least an environment matching loss term and a cognitive interference loss term. For example, the environment matching loss term can be simply defined as the difference between the current transmittance and a preset environmental comfort transmittance. The cognitive interference loss term can be defined as the square of the rate of change of transmittance, penalizing drastic transmittance changes. The weight coefficients of these two terms can be dynamically adjusted according to the probability distribution of the cognitive vulnerability window. For example, when the probability distribution of the cognitive vulnerability window indicates a high probability of excessive cognitive load on the driver at some future moment, the weight of the cognitive interference loss term can be manually increased to prioritize the driver's cognitive comfort. The objective function can be solved using a simple iterative search algorithm to find the transmittance sequence that minimizes the objective function value within a preset time window.

[0036] S104 controls the vehicle's tinted glass to adjust its light transmittance based on the optimal light transmittance change trajectory.

[0037] In this embodiment, the control process can employ an open-loop control method, that is, directly sending the target transmittance value in the optimal transmittance change trajectory to the drive circuit of the photochromic glass, causing it to adjust according to a preset trajectory. For example, a fixed voltage corresponding to the target transmittance value can be applied to the photochromic glass to achieve the corresponding transmittance. This adjustment can be segmented, that is, directly jumping to the target transmittance at each time point.

[0038] By comprehensively considering the driver's multidimensional cognitive load data and the vehicle's multimodal environmental data, this method can predict the driver's cognitive vulnerability window within future timeframes. Therefore, it can construct a dynamic game-theoretic objective function and dynamically adjust the weights of environmental matching and cognitive interference based on the driver's cognitive vulnerability, thereby generating an optimal transmittance change trajectory that balances environmental adaptability and driver cognitive load. This method effectively avoids the excessive cognitive load or visual discomfort that traditional systems may cause during sudden environmental changes, especially when the driver is fatigued. It can proactively optimize the in-vehicle lighting environment and reduce driving risks.

[0039] In some other embodiments, S102 may include: Principal component analysis was performed on the multidimensional cognitive load data to reduce dimensionality and extract the principal components of the load that represent the driver’s comprehensive cognitive state. At the same time, the multimodal data were time-series aligned and normalized to generate an environmental state vector. The principal components of the load and the environmental state vector are concatenated according to the timestamp to form a multidimensional temporal input sequence. The multidimensional temporal input sequence is then input into the bidirectional long short-term memory network layer of the cognitive load-environment spatiotemporal coupling model to extract the temporal dependency features between environmental changes and cognitive load along the forward and backward directions, respectively. The temporal-dependent features are input into the attention mechanism layer of the cognitive load-environment spatiotemporal coupling model. The attention mechanism layer automatically calculates the importance weight of each historical moment to the current prediction and sums them up to obtain the context vector. The context vector represents the historical environmental change information that has the greatest impact on cognitive load. The context vector and the input at the current moment are fused through a fully connected layer to predict the sequence of cognitive load prediction values ​​for the next N moments, and at the same time output the sequence of sensitivity amplification coefficients for the next N moments. The sensitivity amplification coefficient is used to characterize the multiple of the driver's cognitive response intensity to a unit change in the environment under the current load state. The cognitive load prediction sequence is compared one by one with the pre-established personalized load baseline. The extent to which the prediction exceeds the baseline at each time point is calculated. The extent of the exceedance is then nonlinearly amplified by a sensitivity amplification factor to generate the probability distribution of the cognitive vulnerability window within the future time window.

[0040] In this embodiment, multidimensional cognitive load data typically includes various physiological and behavioral indicators, such as heart rate variability, electroencephalogram (EEG), eye-tracking data, and driving operation frequency. Principal Component Analysis (PCA), as a statistical dimensionality reduction technique, can transform this high-dimensional, potentially redundant data into a few mutually orthogonal "load principal components." These principal components can retain the information in the original data to the greatest extent and effectively represent the driver's comprehensive cognitive state, thereby simplifying model input and highlighting key features. Simultaneously, temporal alignment and normalization are performed on the multimodal data to generate an environmental state vector. The multimodal data may cover information such as vehicle speed, acceleration, steering angle, external light intensity, light angle, weather conditions, and road type. Temporal alignment ensures the consistency of data collected from different sensors or data sources over time, while normalization eliminates differences between data of different dimensions, bringing all environmental features to a uniform numerical scale, which is beneficial for stable model training and accurate learning.

[0041] Subsequently, the principal components of the load and the environmental state vector are concatenated according to timestamps to form a multidimensional temporal input sequence. This sequence integrates the continuous evolution information of the driver's cognitive state and the environmental state over time. This multidimensional temporal input sequence is then fed into the bidirectional long short-term memory (Bi-LSTM) network layer of the cognitive load-environment spatiotemporal coupling model to extract the temporal dependency features between environmental changes and cognitive load along both the forward and backward directions. The Bi-LSTM network is a recurrent neural network capable of simultaneously processing both forward and backward information in a sequence. It can capture the lagged impact of an event in the environment on the driver's cognitive load and identify the potential impact of changes in cognitive load on subsequent environmental perception, thus providing a more comprehensive and in-depth understanding of the complex, cross-time-step dynamic relationship between the environment and cognitive load.

[0042] Building upon this, temporally dependent features are input into the attention mechanism layer of the cognitive load-environment spatiotemporal coupling model. This layer automatically calculates the importance weights of each historical moment for the current prediction and sums these weighted features to obtain a context vector. This context vector highly condenses and represents the historical environmental changes that have the greatest impact on the driver's current cognitive load. Through this mechanism, the model can intelligently focus on the historical events most relevant to the current cognitive load prediction, effectively filtering out irrelevant or secondary historical information, and significantly improving the model's ability to identify key influencing factors.

[0043] Furthermore, the context vector and the current input are fused through a fully connected layer to predict a sequence of cognitive load forecasts for the next N time steps. This fully connected layer is responsible for deeply integrating historical context information with the current real-time input, thereby accurately predicting the driver's future cognitive load based on comprehensive spatiotemporal information. Simultaneously, the model also outputs a sequence of sensitivity amplification coefficients for the next N time steps. The sensitivity amplification coefficient is a key indicator that characterizes the driver's cognitive response intensity to a unit change in the environment under the current load state. For example, when the driver is already under high cognitive load, even a small environmental change can cause a sharp increase in their cognitive load, resulting in a larger sensitivity amplification coefficient, and vice versa.

[0044] Finally, the sequence of predicted cognitive load values ​​is compared one by one with a pre-established personalized load baseline to calculate the degree to which the predicted value exceeds the baseline at each time point. The personalized load baseline is preset based on the individual differences of each driver and represents their upper limit of cognitive load under normal driving conditions. If the predicted value is higher than this baseline, the degree of exceedance is calculated. More importantly, the degree of exceedance is nonlinearly amplified by a sensitivity amplification factor to generate a probability distribution of cognitive vulnerability windows within future time windows. This nonlinear amplification mechanism fully considers the differences in drivers' sensitivity to environmental stimuli at different cognitive load levels, so that even a slight exceedance when the driver's cognitive load is already high will be amplified into a higher level of cognitive vulnerability, thus reflecting the driver's cognitive risk more realistically and dynamically.

[0045] By performing principal component analysis to reduce the dimensionality of multidimensional cognitive load data and temporal alignment and normalization of multimodal data, key features of the driver's comprehensive cognitive state and environmental state are effectively extracted, reducing data complexity. A bidirectional long short-term memory network layer can comprehensively capture the complex positive and negative temporal dependencies between environmental changes and cognitive load, overcoming the limitations of traditional unidirectional models in understanding time-series data. The introduction of an attention mechanism layer enables the model to intelligently identify and focus on historical environmental information that has the greatest impact on current cognitive load, avoiding interference from irrelevant information and significantly improving the accuracy and interpretability of predictions. More importantly, by predicting future cognitive load value sequences and simultaneously outputting sensitivity amplification coefficient sequences, this scheme not only predicts the trend of cognitive load but also quantifies the driver's response intensity to environmental changes under different load states. Finally, the degree exceeding the baseline is nonlinearly amplified by the sensitivity amplification coefficient, which can more realistically and dynamically reflect the driver's cognitive vulnerability and generate a high-precision probability distribution of the cognitive vulnerability window. This enables the subsequent construction of the dynamic game objective function and the solution of the optimal light transmittance change trajectory to be based on a more reliable and forward-looking cognitive risk assessment, thereby achieving smarter, more personalized and more effective control over vehicle tinted glass, significantly improving driver comfort and safety, and avoiding cognitive overload caused by environmental changes.

[0046] In other embodiments, before comparing the cognitive load prediction sequence one by one with a pre-established personalized load baseline and calculating the extent to which the prediction exceeds the baseline at each time point, the method further includes: Collect data on the user's gaze dispersion, pupil diameter change rate, and braking reaction time offset during daily driving. A Gaussian mixture model was used to fit the probability distribution of each index; The baseline threshold is the standard deviation of each indicator plus a preset multiple. Personalized load limits are constructed based on baseline thresholds.

[0047] In this embodiment, firstly, the driver's gaze dispersion, pupil diameter change rate, and braking reaction time offset are collected during daily driving. Gazing dispersion quantifies the driver's visual attention concentration. Real-time monitoring of the driver's eye movements is achieved using an in-vehicle eye-tracking device or camera, analyzing the distribution and frequency of their gaze points within the visual field. Higher cognitive load often leads to more dispersed gaze points, resulting in increased gaze dispersion. The pupil diameter change rate serves as an objective indicator of the driver's physiological arousal level and cognitive load. Continuous measurement of the driver's pupil diameter is performed using devices such as infrared eye trackers, calculating its rate of change over a short period. Increased cognitive load typically alters the pupil diameter's expansion or contraction pattern, causing fluctuations in its change rate. Braking reaction time offset measures the deviation in time required from perceiving the need to brake to actually pressing the brake pedal in the face of a sudden situation. Driver braking behavior data is recorded using vehicle sensors and compared with standard or historical average reaction times to calculate the offset. Excessive cognitive load significantly prolongs the driver's reaction time, increasing the braking reaction time offset. The emphasis on collecting this data in everyday driving environments is to obtain the driver's physiological and behavioral patterns in real, natural conditions, thereby ensuring that the established personalized load baseline can accurately reflect their individual characteristics and habits.

[0048] Secondly, a Gaussian mixture model (GMM) was used to fit the probability distributions of each indicator. A GMM is a statistical model used to represent the probability distribution of several subpopulations within a larger population. Here, the GMM was used to model historical data for indicators such as gaze dispersion, pupil diameter change rate, and braking reaction time offset. Since a driver's cognitive load can exhibit multiple patterns under different situations, a single Gaussian distribution is insufficient to accurately describe it. The GMM can decompose these complex, multimodal distributions into a weighted sum of several Gaussian components, thereby capturing the statistical characteristics of each indicator under different cognitive states more precisely, providing more accurate probability density information for setting subsequent baseline thresholds.

[0049] Next, the mean of each indicator is added to the standard deviation by a preset multiple as the baseline threshold. After fitting the probability distribution of each indicator, the statistical characteristics of each indicator, including the mean and standard deviation, can be obtained. To construct a personalized load baseline, this application adopts a statistical method: the mean of each indicator is used as the center value of its normal state, and a preset multiple of the standard deviation is added to this value. This "preset multiple" is an adjustable parameter, such as 1, 2, or 3 times the standard deviation, which determines the leniency of the baseline threshold. In this way, a dynamic and personalized upper limit can be set for each indicator. When the indicator value exceeds this threshold, it is considered that a state of high cognitive load may have been entered. This method takes into account the individual differences of drivers and their tolerance for load fluctuations.

[0050] Finally, a personalized load limit is constructed based on baseline thresholds. After setting baseline thresholds for various cognitive load-related indicators such as gaze dispersion, pupil diameter change rate, and braking reaction time offset, these independent thresholds are combined to construct a multi-dimensional "personalized load limit." This limit represents the upper limit of cognitive load that a driver can tolerate in daily driving without significantly affecting driving safety and comfort. The construction method can be a simple logical combination, for example, considering any indicator exceeding a threshold as exceeding the load limit, or a more complex weighted summation or machine learning model that fuses the thresholds of multiple indicators to form a comprehensive load limit judgment standard. This personalized load limit is a key reference for subsequent assessments of whether the predicted cognitive load exceeds the baseline.

[0051] This application's embodiments, based on physiological and behavioral data collected from drivers during daily driving and combined with advanced statistical modeling methods, establish a highly personalized and dynamically adjusted cognitive load baseline for each driver. This baseline is no longer a uniform, static threshold, but fully considers individual driver differences and their load fluctuation characteristics under different driving situations. Therefore, when subsequent cognitive load predictions are compared with this personalized load baseline, it can more accurately identify when a driver is truly in a cognitively vulnerable state, avoiding misjudgments or omissions caused by using a universal baseline. This makes the generated cognitive vulnerability window probability distribution more accurate, thus providing a more reliable input for constructing the dynamic game objective function. Ultimately, it optimizes the transmittance adjustment strategy of the vehicle's tinted glass, enabling it to respond more promptly and accurately to the driver's actual cognitive state, effectively reducing the driver's cognitive load and improving driving safety and comfort.

[0052] In some other embodiments, S103 may include: Based on the probability distribution of the cognitive vulnerability window at the current moment, the first weight coefficient of the environment matching loss term and the second weight coefficient of the cognitive interference loss term are calculated respectively, wherein the first weight coefficient is inversely proportional to the second weight coefficient and the second weight coefficient is directly proportional to the probability distribution of the cognitive vulnerability window. Calculate the ideal transmittance for the environment at each future moment based on multimodal data; Calculate the comprehensive cognitive load index based on multidimensional cognitive load data; A dynamic game objective function is constructed based on the first weighting coefficient, the second weighting coefficient, the ideal light transmittance for environmental matching, and the comprehensive index of cognitive load.

[0053] In this embodiment, the calculation of the first and second weighting coefficients aims to dynamically adjust the priority of environmental matching and cognitive interference in the objective function based on the driver's cognitive vulnerability. When the probability distribution of the cognitive vulnerability window indicates a high probability that the driver's cognitive load will exceed the personalized load baseline at a future point in time, the system will correspondingly increase the second weighting coefficient to increase the penalty for cognitive interference, thereby prioritizing the driver's cognitive safety; simultaneously, the first weighting coefficient will decrease accordingly, reducing the emphasis on environmental matching. Conversely, when the probability of cognitive vulnerability is low, the system will place more emphasis on environmental matching. This inverse and direct proportional relationship can be achieved through various functional forms. For example, a total weight can be set, and then allocated according to the value of the probability distribution of the cognitive vulnerability window through linear interpolation, the sigmoid function, or an exponential function, ensuring the smoothness and rationality of the weight adjustment.

[0054] The calculation of ideal transmittance for environmental matching aims to determine the optimal transmittance for vehicle tinted glass from the perspective of external environmental comfort, without considering driver cognitive load. This calculation can be based on multimodal vehicle data, such as light intensity, light angle, weather conditions (e.g., sunny, cloudy, rainy, snowy), and road type (e.g., tunnel, elevated road, ordinary road) obtained through onboard sensors. The system can pre-store a transmittance mapping table, which presets corresponding ideal transmittance values ​​for different environmental conditions. In practical applications, the system queries this mapping table in real time and combines it with current and predicted external environmental data to calculate the ideal transmittance for environmental matching at various future times. For example, the ideal transmittance may be lower under strong direct sunlight, and higher under cloudy conditions or in tunnels. Furthermore, for further optimization, the baseline values ​​can be adjusted based on the driver's driving intentions, ensuring that the transmittance has the highest correlation with the driver's subjective comfort score under different driving intentions.

[0055] The calculation of the comprehensive cognitive load index aims to quantify the driver's current overall cognitive load level. Multidimensional cognitive load data can include, but is not limited to, physiological indicators (such as heart rate variability, skin conductance, and pupil diameter change rate) and behavioral indicators (such as gaze dispersion, braking reaction time deviation, and steering wheel operation frequency). To obtain a unified comprehensive index, data fusion techniques can be employed. For example, after normalizing the raw data, principal component analysis (PCA) can be used to extract the principal components, or a weighted average method can be used, assigning different weights to each indicator based on its contribution to cognitive load, and then performing a weighted summation. This comprehensive index can intuitively reflect the driver's cognitive state at the current moment, providing a basis for subsequent decision-making. For example, gaze dispersion, pupil diameter change rate, and braking reaction time deviation can be normalized to preset intervals, and the first principal component can be extracted using PCA as a preliminary load index. This preliminary load index is then multiplied by a sensitivity amplification coefficient output from the cognitive load-environment spatiotemporal coupling model to obtain the final comprehensive index.

[0056] The construction of the dynamic game objective function integrates the calculated weight coefficients, the ideal transmittance for environmental matching, and the comprehensive cognitive load index to form a mathematical model that can be optimized. This objective function is typically designed as a minimization, with its core being the balancing of environmental matching loss and cognitive interference loss. The environmental matching loss term can be represented as the difference between the current transmittance and the ideal transmittance for environmental matching (e.g., the squared difference), aiming to make the glass transmittance as close as possible to the environmental comfort requirements. The cognitive interference loss term can be represented as the impact of transmittance changes on the driver's comprehensive cognitive load index, aiming to avoid increasing the driver's cognitive burden due to drastic or inappropriate changes in transmittance. By multiplying the first and second weight coefficients by their respective loss terms, the objective function achieves a dynamic trade-off between these two conflicting objectives under different levels of cognitive vulnerability, thus obtaining the optimal transmittance change trajectory by solving this function.

[0057] In this embodiment, the priority between environmental comfort and driver cognitive load in the vehicle's tinted glass control strategy can be dynamically adjusted based on the driver's real-time cognitive vulnerability level. Specifically, when the system predicts that the driver is in a cognitively vulnerable state, it significantly increases the weight of the cognitive interference loss term, making the light transmittance adjustment of the tinted glass more inclined to reduce additional stimulation to the driver's cognitive load, even if this means sacrificing some environmental matching perfection. Conversely, when the driver's cognitive state is good, the system will focus more on optimizing environmental matching and providing the best visual comfort. This dynamic weight adjustment mechanism based on the probability distribution of the cognitive vulnerability window means that the control of the vehicle's tinted glass is no longer solely pursuing environmental comfort, but can intelligently balance and trade off between environmental comfort and driver cognitive safety, thereby maximizing the comfort of the driving experience while ensuring driver cognitive safety, significantly improving the system's intelligence and humanization level.

[0058] In other embodiments, the multidimensional cognitive load data includes fixation dispersion, pupil diameter change rate, and braking reaction time offset. A comprehensive cognitive load index is calculated based on the multidimensional cognitive load data, including: The gaze dispersion, pupil diameter change rate, and braking reaction time offset were normalized to preset ranges. Principal component analysis was used to extract the first principal component as the initial loading index. The final comprehensive index is obtained by multiplying the preliminary load index by the sensitivity amplification factor output from the cognitive load-environment spatiotemporal coupling model. The final composite index was determined to be the cognitive load composite index.

[0059] In this embodiment, when calculating the comprehensive cognitive load index, the original multidimensional cognitive load data, such as gaze dispersion, pupil diameter change rate, and braking reaction time offset, first need to be normalized. This step aims to eliminate differences in the dimensions and numerical ranges between different cognitive load indicators, ensuring that they have comparable and fair weights in subsequent calculations. For example, gaze dispersion may be measured in angles or pixels, pupil diameter change rate in millimeters per second, and braking reaction time offset in seconds; these original data may have numerical differences of orders of magnitude. Through normalization, these indicators can be uniformly mapped to a preset numerical range, such as [0, 1] or [-1, 1], thereby avoiding the dominance of certain indicators with larger values ​​in the calculation and ensuring that the contribution of each indicator to the comprehensive index is balanced and effective. Common normalization methods include min-max normalization and Z-score normalization.

[0060] Based on this, Principal Component Analysis (PCA) was used to extract the first principal component as a preliminary cognitive load index. PCA is a commonly used multivariate statistical analysis method that transforms a set of potentially correlated variables into a set of linearly uncorrelated variables through orthogonal transformation; these new variables are called principal components. In this case, multidimensional cognitive load data (such as normalized gaze dispersion, pupil diameter change rate, and braking reaction time offset) may exhibit some correlation and contain redundant information. PCA can identify the most important direction of variation in the data, i.e., the principal components. The first principal component typically explains the largest variance in the data, preserving the information of the original data to the greatest extent and compressing it into a single dimension. Therefore, using the first principal component as a preliminary cognitive load index can effectively extract the core information that best represents the driver's overall cognitive state from multidimensional data, while reducing the dimensionality and noise of the data.

[0061] Subsequently, the preliminary load index is multiplied by the sensitivity amplification factor output from the cognitive load-environment spatiotemporal coupling model to obtain the final comprehensive index. The preliminary load index reflects the driver's baseline cognitive load level at a given moment. However, the driver's cognitive response to environmental changes is not constant but dynamically adjusts according to their current cognitive state and environmental conditions. The sensitivity amplification factor is a dynamic parameter predicted by the cognitive load-environment spatiotemporal coupling model based on current multidimensional cognitive load data and multimodal data. It characterizes the driver's cognitive response intensity to a unit change in the environment under the current load state. By multiplying the preliminary load index by this sensitivity amplification factor, the comprehensive cognitive load index can be dynamically adjusted so that it not only reflects the driver's baseline load but, more importantly, reflects the driver's cognitive vulnerability or sensitivity to environmental stimuli in a specific environment. For example, when a driver is highly fatigued or distracted, their sensitivity amplification factor may be high. Even if the preliminary load index is not high, the final comprehensive index will be amplified, thus more accurately reflecting their cognitive risk. Finally, the value obtained after the above normalization, principal component analysis and sensitivity amplification coefficient adjustment is determined as the cognitive load comprehensive index.

[0062] By normalizing the original multidimensional cognitive load data, such as gaze dispersion, pupil diameter change rate, and braking reaction time offset, the dimensional differences between different indicators were eliminated, ensuring data consistency and comparability. Based on this, principal component analysis was used to extract the first principal component as a preliminary load index. This effectively extracted the single indicator most representative of the driver's core cognitive state from redundant and potentially correlated multidimensional data, significantly reducing data dimensionality and noise interference. Furthermore, this preliminary load index was multiplied by the sensitivity amplification coefficient output from the cognitive load-environment spatiotemporal coupling model. This resulted in a final comprehensive cognitive load index that not only reflected the driver's baseline load level but also dynamically incorporated the driver's cognitive response intensity to environmental changes under the current load state, thus more accurately characterizing the driver's cognitive vulnerability. This provides a more reliable and refined input for subsequently constructing a dynamic game objective function, enabling the light transmittance adjustment of the vehicle's tinted glass to more accurately respond to the driver's cognitive needs, effectively reducing cognitive interference and improving driving comfort and safety.

[0063] In other embodiments, calculating the ideal transmittance for future environmental conditions based on multimodal data includes: Obtain the driving intent of the target user; Based on the light intensity, light angle, weather conditions and road type in the multimodal data, the basic value is obtained by querying the preset transmittance mapping table; The baseline value is adjusted based on the driving intention at future moments to obtain the ideal light transmittance for environmental matching; The correction coefficients are obtained through offline reinforcement learning training, which ensures that the light transmittance has the greatest correlation with the driver's subjective comfort score under different driving intentions.

[0064] In this embodiment, the driver's intention is first acquired. Driving intention refers to the driver's planning and expectations for the vehicle's future behavior in a specific driving situation, such as overtaking, changing lanes, turning, decelerating, or maintaining the current speed. Acquiring driving intention can be achieved in various ways, such as by analyzing the driver's operational behavior (e.g., turn signal signals, accelerator / brake pedal depth, steering wheel angle change rate), eye-tracking data (e.g., gaze point, saccade patterns), voice commands, gesture recognition, or by combining route information and traffic condition predictions provided by the navigation system. Accurately acquiring driving intention is crucial for predicting the driver's reaction to environmental changes and their light transmittance requirements, as it directly reflects the driver's expectations of the road conditions ahead and their attention allocation.

[0065] Based on this, a baseline value is obtained by querying a pre-defined transmittance mapping table using the light intensity, illumination angle, weather conditions, and road type data from the multimodal data. This step aims to determine an initial, universally recommended transmittance value based on objective environmental factors. Light intensity in the multimodal data can be obtained from external light sensors on the vehicle, reflecting the overall brightness of the environment; illumination angle can be obtained from the vehicle's attitude sensors and a solar position calculation model, indicating the direction of direct or reflected sunlight; weather conditions can be obtained from onboard sensors (such as rain sensors and fog sensors) or external meteorological data interfaces, such as sunny, cloudy, rainy, snowy, and foggy; road types can be identified using high-precision map data or visual recognition technology, such as highways, urban roads, tunnels, and rural roads. The pre-defined transmittance mapping table is a database storing recommended transmittance values ​​for different environmental combinations. This mapping table is pre-established through extensive driving experiments, expert experience, or statistical analysis to ensure that reasonable baseline transmittance values ​​are provided under typical environmental conditions.

[0066] Subsequently, the baseline value is adjusted based on the driver's intended driving at future moments to obtain the ideal transmittance for the environment. After obtaining the baseline transmittance value based on objective environmental factors, it is necessary to further consider the driver's personalized and real-time needs. The adjustment process refers to adjusting the above baseline value according to the driver's intended driving at future moments. For example, when the driver intends to overtake, a higher transmittance may be needed to obtain a clearer field of vision; when the driver intends to slow down or stop, a lower transmittance may be preferred to reduce glare. This adjustment ensures that the transmittance adjustment not only adapts to the environment but also to the driver's driving behavior and subjective feelings, thereby improving driving comfort and safety.

[0067] The correction coefficient is obtained through offline reinforcement learning training, maximizing the correlation between transmittance and driver's subjective comfort score under different driving intentions. The correction coefficient is a key parameter used to adjust the base transmittance. Offline reinforcement learning is a machine learning method that trains a strategy by analyzing historical data (e.g., driver transmittance choices, subjective comfort scores, and physiological data under different driving intentions and environments). This strategy outputs a correction coefficient that maximizes the driver's subjective comfort score under a specific driving intention. During training, driving intention and environmental factors can be considered as states, transmittance correction as an action, and driver comfort score as a reward signal. Through offline learning, the system can learn how to intelligently adjust transmittance according to driving intentions in various complex situations, ensuring that the final environmentally matched ideal transmittance best meets the driver's personalized comfort needs, avoiding the limitations of simple rules or fixed parameters.

[0068] In this embodiment, the driver's real-time driving intention is first obtained, and a basic transmittance value is obtained from a preset mapping table by combining multimodal environmental data such as light intensity, light angle, weather conditions, and road type. Based on this, a correction coefficient obtained through offline reinforcement learning training is used to dynamically correct this basic value according to the driving intention at future moments. This correction mechanism ensures that the final environmentally matched ideal transmittance is correlated to the driver's subjective comfort score to the greatest extent, thus avoiding the driver discomfort or poor visibility problems that may result from adjusting transmittance solely based on environmental factors. Compared to schemes that calculate the environmentally matched ideal transmittance based solely on multimodal data, this scheme, by introducing driving intention and correction coefficients from reinforcement learning training, makes the calculation of the environmental matching loss term more refined and personalized. This not only improves the accuracy of the environmental matching loss term in the dynamic game objective function but also enables the entire vehicle tinted glass control system to more intelligently predict and meet the driver's actual needs, thereby significantly improving driver comfort and safety while ensuring environmental adaptability and effectively reducing the risk of cognitive interference caused by unsuitable transmittance.

[0069] In some other embodiments, S104 may include: The control parameters of the controller are determined based on the probability distribution of the cognitive vulnerability window; By adjusting the controller based on the control parameters, an optimized controller is obtained. The target value at the current moment in the optimal transmittance change trajectory is used as the set value of the controller, and the current actual transmittance is used as the feedback value. The controller is optimized to calculate the driving voltage control quantity and output it to the electrochromic glass driving circuit to realize the gradual tracking of transmittance to the target trajectory.

[0070] In this embodiment, firstly, the control parameters of the controller are determined based on the probability distribution of the cognitive vulnerability window. The probability distribution of the cognitive vulnerability window dynamically reflects the probability that the driver's cognitive load will exceed the personalized load baseline due to environmental changes at various future time points. Therefore, when performing transmittance adjustment, the controller's internal parameters need to be adaptively adjusted according to this dynamically changing cognitive vulnerability state to ensure that the control strategy can accurately respond to the driver's real-time cognitive needs and environmental risks. Specifically, a set of parameters can be preset, and the corresponding parameter set can be selected according to a specific threshold or range of the probability distribution; alternatively, a machine learning model can be used, inputting the feature vector of the probability distribution and outputting the optimal control parameters. For example, when the probability distribution of the cognitive vulnerability window shows a high probability that the driver's cognitive load will exceed the baseline at a certain future time period, parameters such as the controller's response speed and overshoot limit can be adjusted to make the transmittance adjustment more stable and gradual, avoiding sudden changes that could cause driver discomfort.

[0071] Secondly, the controller is adjusted based on the control parameters to obtain an optimized controller. After determining the control parameters adapted to the current cognitively vulnerable state, these parameters need to be practically applied to the controller, transforming it from a general controller into one optimized for the current driving environment and the driver's cognitive state. This optimized controller can better perform the transmittance adjustment task to adapt to the driver's real-time needs. For example, for a PID controller, its proportional coefficient, integral coefficient, and derivative coefficient can be adjusted in real time; for a fuzzy controller, its membership function or rule base can be dynamically adjusted; and for a model predictive controller, its prediction model, constraints, or objective function weights can be adjusted.

[0072] Finally, the target value at the current moment in the optimal transmittance change trajectory is used as the controller's setpoint, with the current actual transmittance as the feedback value. The controller calculates the driving voltage control quantity and outputs it to the electrochromic glass driving circuit, achieving gradual tracking of the transmittance along the target trajectory. This is the core of the entire control loop. The optimization controller receives the real-time target value from the optimal transmittance change trajectory and compares it with the actual transmittance of the electrochromic glass to calculate the voltage or current control quantity required to drive the electrochromic glass. The goal is to ensure that the actual transmittance smoothly and accurately follows the optimal trajectory, avoiding abrupt changes, thereby ensuring driver comfort and safety. The optimization controller calculates the voltage or current signal to drive the electrochromic glass based on the error between the setpoint (target transmittance) and the feedback value (actual transmittance), combined with its internal control algorithm and adjusted parameters. After receiving this control quantity, the electrochromic glass driving circuit changes the voltage applied to the electrochromic glass accordingly, thereby changing its transmittance. Gradual tracking means that the control process is smooth, without abrupt changes, which is crucial for avoiding visual discomfort and cognitive interference for the driver.

[0073] In this embodiment, the controller parameters are dynamically adjusted based on the driver's real-time cognitive vulnerability window probability distribution, thereby obtaining a controller optimized for the current driving state. This optimized controller uses the current target value in the optimal transmittance change trajectory as a setpoint and, combined with feedback from the actual transmittance, accurately calculates the voltage control amount required to drive the electrochromic glass. This adaptive closed-loop control mechanism ensures that the transmittance of the vehicle's electrochromic glass smoothly and gradually tracks the optimal trajectory, avoiding visual shock and cognitive interference to the driver caused by sudden changes in transmittance. This not only significantly improves the driver's visual comfort but, more importantly, effectively reduces the driver's cognitive load in complex environments through refined and personalized transmittance adjustment, thereby improving driving safety.

[0074] Based on the vehicle tinted glass control method provided in the above embodiments, this application also provides a specific implementation of the vehicle tinted glass control device 200. Please refer to the following embodiments.

[0075] First see Figure 2 The vehicle tinted glass control device 200 provided in this application embodiment may include: The acquisition module 201 is used to acquire multidimensional cognitive load data of the target user and multimodal data of the vehicle; The input module 202 is used to input multidimensional cognitive load data and multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows. The probability distribution of cognitive vulnerability windows is used to characterize the probability that the driver's cognitive load will exceed the personalized load baseline due to environmental changes at various future time points. Module 203 is used to construct a dynamic game objective function based on the probability distribution of the cognitively vulnerable window, and to obtain the optimal transmittance change trajectory by solving the objective function; the objective function includes at least an environmental matching loss term and a cognitive interference loss term, and the weight coefficients of the two terms are dynamically adjusted according to the probability distribution of the cognitively vulnerable window; The control module 204 is used to control the vehicle's tinted glass to perform light transmittance adjustment according to the optimal light transmittance change trajectory.

[0076] As an alternative implementation, the input module 202 can be specifically used for: Principal component analysis was performed on the multidimensional cognitive load data to reduce dimensionality and extract the principal components of the load that represent the driver’s comprehensive cognitive state. At the same time, the multimodal data were time-series aligned and normalized to generate an environmental state vector. The principal components of the load and the environmental state vector are concatenated according to the timestamp to form a multidimensional temporal input sequence. The multidimensional temporal input sequence is then input into the bidirectional long short-term memory network layer of the cognitive load-environment spatiotemporal coupling model to extract the temporal dependency features between environmental changes and cognitive load along the forward and backward directions, respectively. The temporal-dependent features are input into the attention mechanism layer of the cognitive load-environment spatiotemporal coupling model. The attention mechanism layer automatically calculates the importance weight of each historical moment to the current prediction and sums them up to obtain the context vector. The context vector represents the historical environmental change information that has the greatest impact on cognitive load. The context vector and the input at the current moment are fused through a fully connected layer to predict the sequence of cognitive load prediction values ​​for the next N moments, and at the same time output the sequence of sensitivity amplification coefficients for the next N moments. The sensitivity amplification coefficient is used to characterize the multiple of the driver's cognitive response intensity to a unit change in the environment under the current load state. The cognitive load prediction sequence is compared one by one with the pre-established personalized load baseline. The extent to which the prediction exceeds the baseline at each time point is calculated. The extent of the exceedance is then nonlinearly amplified by a sensitivity amplification factor to generate the probability distribution of the cognitive vulnerability window within the future time window.

[0077] As an alternative implementation, the input module 202 can be specifically used for: Collect data on the user's gaze dispersion, pupil diameter change rate, and braking reaction time offset during daily driving. A Gaussian mixture model was used to fit the probability distribution of each index; The baseline threshold is the standard deviation of each indicator plus a preset multiple. Personalized load limits are constructed based on baseline thresholds.

[0078] As an alternative implementation, the input module 202 can be specifically used for: Based on the probability distribution of the cognitive vulnerability window at the current moment, the first weight coefficient of the environment matching loss term and the second weight coefficient of the cognitive interference loss term are calculated respectively, wherein the first weight coefficient is inversely proportional to the second weight coefficient and the second weight coefficient is directly proportional to the probability distribution of the cognitive vulnerability window. Calculate the ideal transmittance for the environment at each future moment based on multimodal data; Calculate the comprehensive cognitive load index based on multidimensional cognitive load data; A dynamic game objective function is constructed based on the first weighting coefficient, the second weighting coefficient, the ideal light transmittance for environmental matching, and the comprehensive index of cognitive load.

[0079] As an alternative implementation, the input module 202 can be specifically used for: The gaze dispersion, pupil diameter change rate, and braking reaction time offset were normalized to preset ranges. Principal component analysis was used to extract the first principal component as the initial loading index. The final comprehensive index is obtained by multiplying the preliminary load index by the sensitivity amplification factor output from the cognitive load-environment spatiotemporal coupling model. The final composite index was determined to be the cognitive load composite index.

[0080] As an alternative implementation, the input module 202 can be specifically used for: Obtain the driving intent of the target user; Based on the light intensity, light angle, weather conditions and road type in the multimodal data, the basic value is obtained by querying the preset transmittance mapping table; The baseline value is adjusted based on the driving intention at future moments to obtain the ideal light transmittance for environmental matching; The correction coefficients are obtained through offline reinforcement learning training, which ensures that the light transmittance has the greatest correlation with the driver's subjective comfort score under different driving intentions.

[0081] As an alternative implementation, the control module 204 can also be used for: The control parameters of the controller are determined based on the probability distribution of the cognitive vulnerability window; By adjusting the controller based on the control parameters, an optimized controller is obtained. The target value at the current moment in the optimal transmittance change trajectory is used as the set value of the controller, and the current actual transmittance is used as the feedback value. The controller is optimized to calculate the driving voltage control quantity and output it to the electrochromic glass driving circuit to realize the gradual tracking of transmittance to the target trajectory.

[0082] Figure 3 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0083] An electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0084] Specifically, the processor 301 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0085] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 302 may include removable or non-removable (or fixed) media, or memory 302 may be non-volatile solid-state memory. Memory 302 may be internal or external to the integrated gateway disaster recovery device.

[0086] In one instance, memory 302 may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0087] Memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the vehicle tinted glass control method according to the first aspect of this disclosure.

[0088] The processor 301 reads and executes computer program instructions stored in the memory 302 to achieve... Figure 1 A method for controlling tinted glass in a vehicle, as shown in the embodiment.

[0089] In one example, the electronic device may also include a communication interface 303 and a bus 304. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0090] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0091] Bus 304 includes hardware, software, or both, that couples components of an electronic device together. For example, and not as a limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0092] The electronic device can execute the vehicle tinted glass control method in the embodiments of this application, thereby achieving the combination of Figures 1-2 The method and apparatus for controlling tinted glass in vehicles are described.

[0093] Furthermore, in conjunction with the vehicle tinted glass control method in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the vehicle tinted glass control methods in the above embodiments.

[0094] In an optional embodiment, in conjunction with the vehicle tinted glass control method in the above embodiments, this application embodiment can provide a computer program product to implement it. The instructions in the computer program product are executed by the processor of an electronic device, enabling the electronic device to implement any of the vehicle tinted glass control methods in the above embodiments.

[0095] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0096] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0097] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0098] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0099] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A vehicle tinted glass control method characterized by, include: Acquire multidimensional cognitive load data of target users and multimodal data of vehicles; The multidimensional cognitive load data and the multimodal data are input into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows. The probability distribution of the cognitive vulnerability window is used to characterize the probability that the driver's cognitive load will exceed the personalized load baseline due to environmental changes at various future time points. Based on the probability distribution of the cognitive vulnerability window, a dynamic game objective function is constructed, and the optimal transmittance change trajectory is obtained by solving the objective function; the objective function includes at least an environment matching loss term and a cognitive interference loss term, and the weight coefficients of the two terms are dynamically adjusted according to the probability distribution of the cognitive vulnerability window; The vehicle's tinted glass is controlled to adjust its light transmittance based on the optimal light transmittance change trajectory.

2. The method of claim 1, wherein, The step of inputting the multidimensional cognitive load data and the multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows includes: Principal component analysis is performed on the multidimensional cognitive load data to reduce dimensionality and extract the load principal components that characterize the driver's comprehensive cognitive state. At the same time, the multimodal data is subjected to time-series alignment and normalization to generate an environmental state vector. The principal component of the load and the environmental state vector are concatenated according to timestamps to form a multidimensional temporal input sequence. The multidimensional temporal input sequence is then input into the bidirectional long short-term memory network layer of the cognitive load-environment spatiotemporal coupling model to extract the temporal dependency features between environmental changes and cognitive load along the forward and reverse directions, respectively. The temporal dependency features are input into the attention mechanism layer of the cognitive load-environment spatiotemporal coupling model. The attention mechanism layer automatically calculates the importance weight of each historical moment to the current prediction and sums them up to obtain a context vector. The context vector represents the historical environmental change information that has the greatest impact on cognitive load. The context vector is fused with the input at the current moment through a fully connected layer to predict the cognitive load prediction sequence for the next N moments, and at the same time output the sensitivity amplification coefficient sequence for the next N moments. The sensitivity amplification coefficient is used to characterize the driver's cognitive response intensity multiple to a unit change in the environment under the current load state. The sequence of predicted cognitive load values ​​is compared one by one with a pre-established personalized load baseline. The degree to which the predicted value exceeds the baseline at each time point is calculated. The degree to exceed the baseline is then nonlinearly amplified by the sensitivity amplification factor to generate the probability distribution of the cognitive vulnerability window within the future time window.

3. The method of claim 2, wherein, Before comparing the cognitive load prediction sequence one by one with a pre-established personalized load baseline and calculating the degree to which the prediction value exceeds the baseline at each time point, the method further includes: Collect data on the user's gaze dispersion, pupil diameter change rate, and braking reaction time offset during daily driving. A Gaussian mixture model was used to fit the probability distribution of each index; The baseline threshold is the standard deviation of each indicator plus a preset multiple. Personalized load limits are constructed based on the baseline threshold.

4. The method of claim 2, wherein, The construction of the dynamic game objective function based on the cognitive vulnerability window probability distribution includes: Based on the probability distribution of the cognitive vulnerability window at the current moment, calculate the first weight coefficient of the environment matching loss term and the second weight coefficient of the cognitive interference loss term, respectively, wherein the first weight coefficient is inversely proportional to the second weight coefficient and the second weight coefficient is directly proportional to the probability distribution of the cognitive vulnerability window; The ideal transmittance for environmental matching at each future moment is calculated based on the multimodal data. Calculate the comprehensive cognitive load index based on the multidimensional cognitive load data; The dynamic game objective function is constructed based on the first weighting coefficient, the second weighting coefficient, the ideal light transmittance for environmental matching, and the comprehensive cognitive load index.

5. The method according to claim 4, characterized in that, The multidimensional cognitive load data includes fixation dispersion, pupil diameter change rate, and braking reaction time offset. The calculation of the comprehensive cognitive load index based on the multidimensional cognitive load data includes: The gaze dispersion, pupil diameter change rate, and braking reaction time offset were normalized to preset ranges. Principal component analysis was used to extract the first principal component as the initial loading index. The preliminary load index is multiplied by the sensitivity amplification factor output from the cognitive load-environment spatiotemporal coupling model to obtain the final comprehensive index. The final comprehensive index is determined as the cognitive load comprehensive index.

6. The method according to claim 4, characterized in that, The calculation of the ideal transmittance for environmental matching at future times based on the multimodal data includes: Obtain the driving intent of the target user; Based on the light intensity, light angle, weather conditions, and road type in the multimodal data, the base value is obtained by querying the preset transmittance mapping table; The baseline value is adjusted based on the driving intention at a future moment to obtain the ideal transmittance for environmental matching; The correction coefficients are obtained through offline reinforcement learning training, which ensures that the light transmittance has the greatest correlation with the driver's subjective comfort score under different driving intentions.

7. The method according to claim 1, characterized in that, The step of controlling the light transmittance adjustment of the vehicle's photochromic glass according to the optimal light transmittance change trajectory includes: Based on the cognitive vulnerability window probability distribution, the control parameters of the controller are determined; The controller is adjusted based on the control parameters to obtain an optimized controller; The target value at the current moment in the optimal transmittance change trajectory is used as the set value of the controller, and the current actual transmittance is used as the feedback value. The driving voltage control quantity is calculated by the optimized controller and output to the electrochromic glass driving circuit to realize the gradual tracking of transmittance to the target trajectory.

8. A vehicle tinted glass control device, characterized in that, The device includes: The acquisition module is used to acquire multidimensional cognitive load data of target users and multimodal data of vehicles; The input module is used to input the multidimensional cognitive load data and the multimodal data into a pre-trained cognitive load-environment spatiotemporal coupling model to generate a probability distribution of cognitive vulnerability windows within future time windows; the probability distribution of cognitive vulnerability windows is used to characterize the probability that the driver's cognitive load will exceed the personalized load baseline due to environmental changes at various future time points; A construction module is used to construct a dynamic game objective function based on the probability distribution of the cognitively vulnerable window, and to obtain the optimal transmittance change trajectory by solving the objective function; the objective function includes at least an environment matching loss term and a cognitive interference loss term, and the weight coefficients of the two terms are dynamically adjusted according to the probability distribution of the cognitively vulnerable window; The control module is used to control the vehicle's tinted glass to perform light transmittance adjustment according to the optimal light transmittance change trajectory.

9. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the vehicle tinted glass control method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the vehicle tinted glass control method as described in any one of claims 1-7.

Images