Vehicle-mounted atmosphere lamp control method and system based on environment perception

Abstract

The invention relates to the technical field of intelligent control, and discloses a vehicle-mounted atmosphere lamp control method and system based on environmental perception. The method comprises the following steps: acquiring in-vehicle image data and vehicle-mounted sensor data, and inputting the data into a double-branch Mama information fusion module for feature extraction and fusion processing to obtain a fusion feature map; performing situation recognition and analysis on the fused feature map to obtain a driving situation feature vector and a situation incidence matrix; according to the driving situation feature vector and the situation incidence matrix, vehicle-mounted atmosphere lamp parameters are decoupled, and decoupled illumination parameters are obtained; and performing situation self-adaptive lighting regulation and control and dynamic situation weighted loss calculation on the decoupled lighting parameters and the driving situation feature vectors to obtain target lighting control parameters. According to the invention, the problem of unbalanced illumination requirements in different driving scenes is solved, and the adjustment of the vehicle-mounted atmosphere lamp is more accurate.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a vehicle-mounted ambient light control method and system based on environment perception. Background Art

[0002] As an important component to enhance the user's driving experience, in-vehicle ambient lighting has been widely used in the field of smart cars in recent years. With the rapid development of intelligent connected car technology, users' demand for personalized and intelligent in-vehicle environment is growing. In-vehicle ambient lighting is no longer just a simple decorative lighting, but has multiple functions such as adjusting the driving atmosphere, indicating driving status, and optimizing the driving experience. However, the existing in-vehicle ambient lighting control system is mostly based on simple sensor data or preset lighting modes for adjustment. It lacks deep modeling between different driving scenarios and is difficult to accurately capture the complex changes and subtle features of driving situations, resulting in the inability of ambient lighting to adapt to the dynamically changing driving environment in real time.

[0003] Traditional in-vehicle ambient light controllers generally have color-brightness coupling problems when adjusting the lighting effects of different areas in the car, that is, when adjusting a certain parameter, it will inevitably affect other parameters, making precise control difficult. For example, adjusting the hue may cause unexpected changes in brightness, or increasing the brightness may affect the color saturation, which not only reduces the user experience, but also limits the application effect of in-vehicle ambient lights in specific driving scenarios. At the same time, the existing system is difficult to provide differentiated lighting control for different driving areas (such as the driving area, the co-pilot area, and the rear area), and cannot meet the personalized lighting needs of different occupants in different driving scenarios. Summary of the invention

[0004] The present invention provides a vehicle-mounted ambient light control method and system based on environmental perception, which solves the problem of unbalanced lighting requirements in different driving scenarios and makes the adjustment of the vehicle-mounted ambient light more precise.

[0005] In a first aspect, the present invention provides a method for controlling a vehicle-mounted ambient light based on environment perception, and the method for controlling a vehicle-mounted ambient light based on environment perception comprises: Collect in-vehicle image data and on-board sensor data and input them into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map; Performing situation recognition and analysis on the fused feature graph to obtain a driving situation feature vector and a situation association matrix; Decoupling the vehicle ambient light parameters according to the driving scenario feature vector and the scenario association matrix to obtain decoupled lighting parameters; Context-adaptive lighting regulation and dynamic context-weighted loss calculation are performed on the decoupled lighting parameters and the driving context feature vector to obtain target lighting control parameters.

[0006] In a second aspect, the present invention provides a vehicle-mounted ambient light control system based on environment perception, and the vehicle-mounted ambient light control system based on environment perception includes: The acquisition module is used to collect in-vehicle image data and vehicle sensor data and input them into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map; A situation recognition module, used to perform situation recognition and analysis on the fused feature map to obtain a driving situation feature vector and a situation association matrix; A decoupling processing module, used for performing decoupling processing on vehicle ambient light parameters according to the driving scenario feature vector and the scenario association matrix to obtain decoupled lighting parameters; The loss calculation module is used to perform situation-adaptive lighting regulation and dynamic situation weighted loss calculation on the decoupled lighting parameters and the driving situation feature vector to obtain target lighting control parameters.

[0007] In a third aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on a computer, the computer executes the above-mentioned vehicle ambient light control method based on environmental perception.

[0008] In the technical solution provided by the present invention, by using a dual-branch Mamba information fusion module to process the in-vehicle image data and the vehicle-mounted sensor data, the effective fusion of multi-source heterogeneous data is achieved, and the perception ability of the driving situation is improved. The module uses an image processing branch and a sensor data processing branch to extract spatial features and temporal features respectively, and adaptively fuses them through a cross-modal feature fusion layer, overcoming the problem of insufficient information from a single data source. A multi-scale driving situation attention module is adopted to extract multi-scale feature representations through convolution kernels of different sizes, and the self-attention mechanism and cross-scale attention fusion are applied to achieve accurate identification and analysis of different driving situations, so that the system can focus on the features most relevant to the adjustment of the atmosphere light. In view of the color-brightness coupling problem existing in the traditional vehicle-mounted atmosphere light controller, the present invention proposes a parameter decoupling processing mechanism. By establishing a LAFC parameter coupling relationship model and a dual-input dual-output disturbance observation mechanism, the independent and precise control of the four parameters of hue, saturation, brightness and brightness is achieved, making the adjustment of the vehicle-mounted atmosphere light more accurate. Based on the situational adaptive lighting control method, the system can dynamically adjust the lighting parameters according to the identified driving situation, and perform differentiated control on different areas in the car, meeting the personalized lighting needs of different occupants in different driving scenarios. The innovative introduction of the dynamic situational weighted loss function solves the problem of unbalanced lighting requirements in different driving scenarios. By adjusting the situational weight factor, the system can adjust the priority for specific driving situations and optimize the overall lighting effect. In addition, the system also continuously accumulates driving experience through the lighting strategy self-learning module to continuously improve the lighting control effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0010] Figure 1 A schematic diagram of an embodiment of a vehicle ambient light control method based on environment perception in an embodiment of the present invention; Figure 2 The figure is a schematic diagram of an embodiment of a vehicle ambient light control system based on environment perception in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] An embodiment of the present invention provides a method and system for controlling a vehicle-mounted ambient light based on environmental perception. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a vehicle ambient light control method based on environment perception in an embodiment of the present invention includes: Step S101, collecting in-vehicle image data and vehicle-mounted sensor data and inputting them into a dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map; It is understandable that the execution subject of the present invention may be a vehicle ambient light control system based on environment perception, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0013] Specifically, a data acquisition system is installed in the car, in which a high-resolution camera is used as the main source of visual data. The installation position is selected at the rearview mirror inside the car to ensure full coverage of the driver's face and the main areas inside the car. The acquisition frequency of the camera is set to 30 frames per second, and the resolution reaches 1920×1080 pixels to ensure the clarity and real-time nature of the collected images. At the same time, it is equipped with an on-board sensor array, which includes a light sensor, a temperature sensor, an acceleration sensor and a GPS positioning module. Among them, the light sensor is used to measure the light intensity inside and outside the car, with a sampling frequency of 10Hz and a measurement range of 0 to 100,000 lux to ensure that the changes in light can be perceived in time; the temperature sensor collects the temperature data inside the car, and its sampling frequency is set to 1Hz, and the measurement accuracy reaches ±0.5℃ to meet the requirements of on-board environmental monitoring; the acceleration sensor is used to detect the dynamic state of the vehicle, with a range set to ±2g and a sampling frequency of 50Hz to accurately capture the acceleration changes of the vehicle; the GPS module is responsible for obtaining the vehicle's geographic location, speed and driving direction data, and the update frequency is set to 1Hz to ensure that the vehicle's driving trajectory can be accurately recorded. The in-vehicle camera image and the vehicle's surrounding environment parameter data are packaged and processed according to the predefined data packet format. Each data packet contains a timestamp, sensor ID, data value, and status identifier. In order to improve the efficiency of data transmission and processing, the formatted data packets are classified according to different priorities, and all data are divided into real-time high-priority data streams and timed low-priority data streams. The real-time data stream contains the in-vehicle camera image data and the sensor data with fast dynamic changes of the vehicle, such as acceleration information, while the low-priority data stream mainly contains relatively slow-changing data such as ambient temperature and light intensity. On the basis of data stream stratification, the data stream is dynamically adjusted and processed in combination with the vehicle's driving status to ensure that the system can collect the most critical information in different driving scenarios. For example, when the vehicle is stationary or driving at low speed, the frame rate of image acquisition is reduced or the sampling frequency of certain environmental parameters is reduced to reduce the computational burden, while in high-speed driving or special driving situations, the data acquisition frequency is increased to ensure the integrity and timeliness of the data. The in-vehicle image data and vehicle-mounted sensor data are input into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map.

[0014] The in-vehicle image data and on-board sensor data are preprocessed to ensure the consistency of data format and optimize computational efficiency. During the data acquisition phase, the in-vehicle camera records 1920×1080 pixel high-resolution images at a frequency of 30 frames per second, while the on-board sensor array, including light sensor, temperature sensor, acceleration sensor and GPS module, provides environmental parameter data at sampling frequencies of 10Hz, 1Hz, 50Hz and 1Hz respectively. After acquisition, these data will be standardized, where the image data is cropped and scaled to a unified 256×256 pixel format to meet the input requirements of the neural network, followed by histogram equalization to enhance image contrast, and normalization to map RGB channel pixel values ​​to the [0,1] interval to make image features more standardized. For the on-board sensor data, outlier detection is used to remove noise data, and the sliding window averaging method is used for smoothing to ensure the continuity and stability of the data. Then, time series alignment is performed to ensure that all sensor data have a unified timestamp, and finally converted into a standardized format for subsequent feature extraction. After data preprocessing, the standardized image is input to the image processing branch of the dual-branch Mamba information fusion module, in which the first convolution layer is used for preliminary feature extraction. The convolution layer uses a 3×3 convolution kernel, a step size of 1, a padding of 1, and a channel number of 64. ReLU is used as the activation function, so that low-level edge and texture features are effectively extracted; the second convolution layer continues to extract more complex local features based on the previous layer, with the same parameter settings, but the number of channels is increased to 128 to enhance the feature representation capability; the third and fourth convolution layers further extract high-level semantic features, and a maximum pooling layer is added after each convolution layer. The pooling kernel size is set to 2×2 and the step size is set to 2 to reduce the dimension of the feature map and retain the most representative key information. The final image spatial features contain key visual information such as the illumination distribution, color information, and driver status of the in-vehicle environment. At the same time, the standardized sensor data is input into the sensor data processing branch of the dual-branch Mamba information fusion module, which uses the Mamba state space model for temporal feature extraction. The key parameters of the Mamba state space model are set to state dimension E8, expansion step size 4, expansion coefficient 2, and bidirectional scanning to ensure that the time dependency of sensor data can be effectively captured. In the specific calculation process, the model performs state modeling on the input data and maps the sensor data to a high-dimensional state space to better describe its trend over time. Then, the sequence is expanded by expanding the window with a step size of 4, so that the system can simultaneously consider the change information in adjacent time slices and increase the data expression ability by using the expansion coefficient 2. At the same time, a bidirectional scanning mechanism is adopted, that is, not only the current state is predicted based on past data, but also the future information is used for reverse adjustment to improve the integrity and accuracy of the time series characteristics.Through this processing method, the system accurately extracts key dynamic features such as vehicle speed changes, light intensity fluctuations, temperature trends, and acceleration changes to form sensor time series features. The image spatial features and sensor time series features are cross-modally fused to fully utilize the complementarity of the two data types. These features are input into the cross-modal feature fusion layer of the dual-branch Mamba information fusion module, where the attention weight matrix is ​​calculated using the attention mechanism. The dimension of the matrix is ​​set to 128×128 to measure the importance relationship between different modal features. The weights are then normalized using the Softmax function so that the features of different modalities can adaptively adjust their influence during the fusion process to ensure balanced fusion of information. Since image features are mainly used to describe spatial information, while sensor data are mainly used to describe time evolution features, the attention mechanism can ensure that in different driving scenarios, the system pays more attention to those features that contribute most to the current state. For example, in the case of drastic changes in light, more attention is paid to light sensor data, while in the case of abnormal driver status, more attention is paid to image information. In order to prevent the loss of original feature information during the fusion process and improve the training stability of the model, residual connections are added to the adaptive fusion features to retain some of the original information, and layer normalization technology is used to prevent the deviation of feature distribution and ensure that the data can maintain a stable numerical range during different batches of training. The fused and optimized feature data is output as a fused feature map, the size of which is set to 128×64×64, which contains multi-dimensional information such as in-vehicle visual information, environmental parameters, and vehicle dynamic status.

[0015] Step S102: performing situation recognition and analysis on the fused feature graph to obtain a driving situation feature vector and a situation association matrix; Specifically, the fused feature map passes through a multi-scale feature extraction module, which uses three different sizes of convolution kernels for convolution operations, using convolution kernels of sizes 3×3, 5×5, and 7×7, respectively, to ensure that feature information of different scales can be captured. Small-sized convolution kernels extract local detail information such as edges and textures, while medium-sized convolution kernels are used to identify medium-range feature patterns such as object contours and lighting distribution, and larger-sized convolution kernels help extract global information, including overall brightness trends and spatial structures. Through the extraction process of three different-sized convolution kernels, feature representations of three different scales are obtained. In order to enhance the correlation between features and mine the intrinsic patterns of the data, the self-attention mechanism is applied to the feature representations of these three different scales, and the scaled dot product attention algorithm is used to calculate the spatial correlation within the feature map. The algorithm obtains query, key and value matrices through linear transformation, then calculates the dot product of query and key matrices, and uses the square root of the feature dimension as a scaling factor to ensure numerical stability. Then, the calculated attention weights are normalized by applying the Softmax function to ensure that the sum of all attention distributions is 1, and the value matrix is ​​weighted summed using the normalized weights to generate an attention weighted feature map. Since feature maps of different scales have different spatial distribution characteristics, the scaled dot product attention algorithm can ensure that the system can find the most important feature areas at different scales, thereby improving the recognition ability of driving scenarios and generating three attention weighted feature maps of different scales. The three attention weighted feature maps of different scales are input into the cross-scale attention fusion module to achieve adaptive fusion of information. In this process, a learnable weight parameter is introduced so that the fusion process automatically adjusts the contribution ratio of features of different scales to ensure the optimality of the final fused features. The cross-scale attention fusion module maps feature vectors of different scales to the same feature space, and then uses weight parameters to perform weighted summation on them. These weight parameters are constantly adjusted during the training process to minimize the error of driving situation classification, so that the fused features can fully express multi-scale information and obtain fused feature vectors. The fused feature vectors are fully connected to output the driving situation category distribution. At this stage, the fused feature vectors are sent to a fully connected neural network, which contains multiple hidden layers and uses nonlinear activation functions to enhance the nonlinear expression ability of the data. The Softmax function is used in the output layer to map the feature vectors to the probability distribution of driving situation categories. The probability distribution represents the classification possibility of different driving situations, such as high-speed cruising, urban congestion, night driving, rainy driving, emergency braking, etc., and the current driving situation is determined according to the highest probability category.After obtaining the driving scenario category, the key feature indicators related to the scenario are extracted to generate a driving scenario feature vector. The feature vector contains four key indicators, including lighting conditions, driver status, vehicle speed change rate and environmental complexity. The lighting conditions are measured by the in-car light sensor to quantify the brightness level in the car. The driver status is analyzed by the in-car camera to analyze the driver's facial expression and attention level to judge the driver's mental state. The vehicle speed change rate is measured by GPS and acceleration sensors to reflect the vehicle's driving mode. The environmental complexity is comprehensively evaluated by combining vehicle speed changes, external lighting conditions and driver status to judge the stability of the current driving scene. Through these four key features, the current driving scenario is characterized. Based on the driving scenario feature vector, the correlation coefficient between different driving scenarios and ambient light parameters is calculated to generate a scenario association matrix. The matrix is ​​used to quantify the weight of the ambient light adjustment requirements under different scenarios. The calculation method is based on statistical analysis and regression analysis of historical data. The size of the matrix is ​​E×4, where E represents the number of driving scenarios, and 4 corresponds to the four parameters of the ambient light, including hue, saturation, brightness and brightness. In order to calculate the correlation coefficient, a large amount of historical data was collected, and the changing trend of each ambient light parameter under each driving scenario was evaluated through a regression analysis model. Then, the Pearson correlation coefficient was calculated to quantify the dependency between these parameters. Finally, a complete situational association matrix was formed. Each element of the matrix represents the degree of influence of a certain driving scenario on a certain ambient light parameter, and is used to guide subsequent adaptive lighting control.

[0016] Step S103, decoupling the vehicle ambient light parameters according to the driving scenario feature vector and the scenario association matrix to obtain decoupled lighting parameters; Specifically, the LAFC parameter coupling relationship model is constructed by experimentally measuring the visual effects under different RGB color combinations and brightness levels. In the experimental process, a standard lighting environment is used to analyze the color visual perception under different RGB ratio combinations, and the degree of influence of different lighting conditions on hue, saturation, brightness and brightness is recorded through a combination of subjective experiments and objective measurements to form a coupling coefficient matrix, which is used to describe the mutual influence relationship between the four parameters, so that the system can quantitatively evaluate the degree of influence of a certain parameter change on other parameters. Based on the coupling coefficient matrix, the HSV color space is divided into multiple grids, and the corresponding relationship between the visual effect and the RGB brightness is calculated at each grid point to form a four-dimensional parameter mapping table. The grid division method of the mapping table is based on the statistical results of experimental data. The hue, saturation and brightness are divided with a resolution of 36×10×10, and a complete four-dimensional parameter mapping table is constructed with a 100-level brightness distribution, in which the hue ranges from 0 to 359 degrees and is divided at 10 degree intervals, the saturation and brightness ranges from 0 to 100% and are divided at 10% intervals, and the brightness parameter is finely divided at 1% intervals. Based on the four-dimensional parameter mapping table, a dual-input dual-output perturbation observation mechanism is designed. This mechanism measures the influence relationship between various parameters through small perturbation experiments, and performs preset perturbations within the error range to observe the impact of different parameter changes on the overall visual effect. Based on the current parameter settings, the system applies a small perturbation of ±5% on the values ​​of hue, saturation, lightness and brightness, and records the visual effect change data before and after the perturbation. These data can help the system quantify the interactive influence between parameters. According to the visual effect change data before and after the perturbation, the local Jacobian matrix between parameters is calculated. This matrix is ​​used to describe the sensitivity relationship of various lighting parameters near the current working point. Through the differential approximation method, the changes in different parameters are compared with the changes in visual effects to form a parameter sensitivity matrix. The size of the matrix is ​​4×4, and each element represents the partial derivative relationship between hue, saturation, lightness and brightness, reflecting the degree of influence of each parameter on the overall visual effect when it is slightly adjusted. The parameter compensator is constructed based on the inverse matrix of the parameter sensitivity matrix. The compensator offsets the cross-effects between different parameters through matrix transformation to achieve independent control. To this end, the inverse matrix of the parameter sensitivity matrix is ​​calculated and used to construct the decoupling transformation matrix. The value range of the transformation matrix is ​​limited to [-0.5, 0.5] to ensure the stability of the adjustment process and prevent the sudden change of visual effects caused by excessive parameter changes. The initial lighting parameters are determined according to the driving situation feature vector and the situation association matrix, and are adjusted using the decoupling transformation matrix to ensure that the final output lighting parameters can adapt to the current driving environment.According to the driving scenario category, the corresponding ambient light parameter adjustment weights are extracted from the scenario association matrix, and the initial lighting parameter settings are determined in combination with the lighting conditions, driver status, vehicle speed change rate, and environmental complexity in the driving scenario feature vector. These settings include the default values ​​of basic hue, saturation, brightness, and brightness, and are dynamically adjusted in combination with the scenario information to obtain the initial lighting parameters. The initial lighting parameters are input into the decoupling transformation matrix for processing to automatically calculate and adjust the mutual influence between the parameters to obtain the final decoupled lighting parameters.

[0017] Step S104: performing situation-adaptive lighting control and dynamic situation weighted loss calculation on the decoupled lighting parameters and driving situation feature vector to obtain target lighting control parameters.

[0018] Specifically, a scenario-lighting mapping database is constructed for E typical driving scenarios. The database contains the ideal lighting parameters of the vehicle ambient light under different driving scenarios, and defines the recommended values ​​and adjustable ranges of the four basic parameters of hue, saturation, brightness and brightness. The setting of these parameters is based on long-term experimental measurements and user feedback data, and takes into account driving safety, comfort and the psychological feelings of the driver. Fuzzy logic control is performed according to the current driving scenario feature vector to calculate the initial lighting parameter adjustment amount. The input variables of the fuzzy logic control include driving scenario category, scenario confidence, vehicle speed, ambient light intensity and driver status, while the output variables include hue adjustment amount, saturation adjustment amount, brightness adjustment amount and brightness adjustment amount. Each input variable is divided into multiple fuzzy sets, for example, the vehicle speed is divided into low speed, medium speed and high speed, and the ambient light intensity is divided into weak light, normal light and strong light, and the output is derived by setting a series of fuzzy rules. After calculating the initial adjustment of the lighting parameters, the parameters are optimized to ensure the smoothness of the lighting changes. Therefore, the initial adjustment is dynamically optimized based on the real-time monitoring of the driving situation change rate. When the driving situation change rate is small, a smaller adjustment step is used to make the change of the ambient light smoother. When the driving situation change rate is large, a larger adjustment step is used to ensure that the lighting can quickly adapt to the new driving situation. The adaptive optimization strategy can effectively improve the stability of lighting control and make the change of the ambient light more in line with the driver's physiological and psychological perception. In order to optimize the lighting control, the interior space is divided into 4 independent lighting areas, including the driving area, the co-pilot area, the left side of the rear row and the right side of the rear row, and the calculated smooth adjustment parameters are regionally differentiated to meet the needs of occupants in different areas. Specifically, the lighting parameters are adjusted independently for different areas according to the activity areas and positions of the driver and passengers. After completing the calculation of the regionalized lighting control parameters, they are matched with the lighting template parameters in the situation-lighting mapping database to evaluate the closeness of the current lighting scheme to the preset optimal lighting scheme, calculate the Euclidean distance between the regionalized lighting control parameters and the lighting template parameters, and evaluate the lighting effect matching degree based on the calculation results. The Euclidean distance is calculated by normalizing the four parameters of hue, saturation, lightness and brightness to the same scale range, and calculating their distance in four-dimensional space. The higher the matching degree, the closer the current lighting scheme is to the optimal setting. When the lighting effect matching degree is lower than the preset target score, the system will automatically enter the optimization mode to further fine-tune the lighting parameters.During the optimization process, the regional lighting control parameters are fine-tuned by binary search, the adjustment direction of the current lighting parameters is calculated, and a better parameter combination is tried in the parameter space. Each time the adjustment is made, a new Euclidean distance is calculated, and it is determined whether the matching degree has been improved. The system will continue to perform this process until the matching degree reaches the preset threshold or the number of adjustments reaches the maximum allowable value. With the support of this optimization strategy, the system can ensure that the final lighting parameters are as close to the optimal settings as possible to improve the adaptability and consistency of the in-vehicle lighting experience. After the optimization of the lighting parameters is completed, the dynamic situational weighted loss is calculated to ensure that the final target lighting control parameters can adapt to the specific needs of different driving situations. A driving experience evaluation model is constructed, and the current lighting effect is scored based on four dimensions: visual comfort, driving assistance, emotional regulation, and situational matching. The weight of each dimension is initially set to 0.25, and is dynamically adjusted based on the driver's facial expressions, eye tracking, and operating behaviors captured by the in-vehicle camera. The scoring range is set between 0 and 100 to reflect the adaptability of the lighting solution in different driving situations. The final target lighting control parameters are calculated using a dynamic context-weighted loss function, which is weighted by the importance of different driving contexts to optimize the lighting control strategy. The gradient descent optimization algorithm is used to iteratively calculate the loss function and adjust the lighting parameters to make them as close to the target setting as possible. To ensure the stability of the optimization process, the learning rate is set to 0.01, the maximum number of iterations is 100, and the convergence threshold is set to 0.001 to ensure that the loss function can converge within a reasonable range. The target lighting control parameters are calculated and transmitted to the vehicle-mounted ambient light hardware execution unit, so that it can automatically adjust the lighting scheme in different driving contexts, thereby improving the driving experience and enhancing driving safety.

[0019] An evaluation model is built based on the driver's facial expressions, eye movements and operating behaviors. The model uses an in-car camera to capture the driver's facial expressions in real time, and combines deep learning algorithms to classify expression features to determine the driver's emotional state. Eye tracking technology is used to analyze the driver's line of sight and blinking frequency to detect attention levels and potential fatigue states. At the same time, the on-board control unit monitors the driver's operating behaviors, such as steering angle, braking force, and throttle control method, to evaluate the driver's driving style and behavior patterns. After fusion, these data are used to calculate four key evaluation dimensions, namely visual comfort, driving assistance, emotional regulation, and situational matching. Visual comfort measures whether the current lighting environment interferes with the driver's visual perception, driving assistance evaluates whether the lighting conditions help improve driving safety, emotional regulation measures the impact of the lighting solution on the driver's psychological state, and situational matching reflects whether the current lighting meets the expected driving situation requirements. The scores of these four dimensions together constitute the experience score, which is set between 0 and 100 to reflect the adaptability and user experience of different lighting parameters. After obtaining the experience score, the dynamic context weighted loss function is applied to calculate the initial lighting control parameters to quantify the deviation between the lighting parameters and the optimal experience. The loss function is weighted based on the importance of the driving context to optimize the lighting control strategy. Based on the calculated initial loss value, the gradient descent optimization algorithm is executed to iteratively optimize the lighting parameters, setting the learning rate to 0.01, the maximum number of iterations to 100, and the convergence threshold to 0.001. At each iteration, the partial derivative of the loss function with respect to the lighting parameters is calculated, and the parameter value is adjusted according to the direction of the gradient descent to minimize the loss, ensuring that the final optimized lighting parameters can fit the optimal lighting solution to the greatest extent possible, thereby improving the driving experience and safety. After the optimization is completed, the optimized lighting parameters are subjected to a safety constraint test to ensure that they meet road safety standards and human visual comfort requirements. The test process includes brightness limitation, color temperature control and hue change constraints. The brightness limitation ensures that the light intensity of the ambient light does not exceed the maximum brightness specified by the regulations to avoid affecting the driver's or other road users' sight, while the color temperature control ensures that the lighting scheme is in line with the human eye's visual comfort range to avoid discomfort caused by overly cold or overly warm color temperatures. At the same time, the hue change constraint ensures that the hue adjustment rate of the ambient light does not exceed the preset threshold to avoid sudden color changes affecting the driver's attention. The lighting parameters that have passed this safety test are considered to meet safety standards and are used for final control execution.The lighting parameters that meet safety standards are compared with the initial lighting control parameters through loss function evaluation to quantify the actual effect of the optimization process. This evaluation method evaluates the improvement of the optimization strategy by calculating the difference in loss values ​​before and after optimization. If the loss value after optimization is significantly reduced, it means that the optimization strategy has effectively improved the adaptability of the lighting scheme. If the loss value does not change significantly, it indicates that the current optimization strategy needs further adjustment. The difference in experience scores before and after optimization is calculated to verify whether the optimization has improved the driver's actual experience. The impact of different lighting parameters on the driving experience is evaluated through statistical analysis. All these evaluation data will be stored as optimization effect data. In order to ensure that the system can continuously improve and adapt to changes in different driving environments, the scenario-lighting mapping database is updated using an incremental learning method based on optimization effect data and experience scores. The update process adopts a batch update method, and a parameter adjustment is performed after accumulating 100 driving data each time. During the update process, the optimal lighting parameters for different scenarios are recalculated in combination with the newly collected data, and the lighting template is updated. At the same time, during the incremental learning process, historical data is used for regression analysis to ensure that the updated database will not be over-adjusted due to short-term data deviations, thereby maintaining long-term stability and adaptability, outputting the optimized target lighting control parameters, and transmitting them to the on-board ambient light hardware execution unit via the CAN bus to achieve precise lighting control, thereby ensuring that the ambient light can always provide the best visual experience and safety in different driving scenarios.

[0020] In an embodiment of the present invention, by using a dual-branch Mamba information fusion module to process in-vehicle image data and vehicle-mounted sensor data, effective fusion of multi-source heterogeneous data is achieved, and the perception of driving situations is improved. The module uses image processing branches and sensor data processing branches to extract spatial features and temporal features respectively, and adaptively fuses through a cross-modal feature fusion layer, overcoming the problem of insufficient information from a single data source. A multi-scale driving situation attention module is used to extract multi-scale feature representations through convolution kernels of different sizes, and the self-attention mechanism and cross-scale attention fusion are applied to achieve accurate identification and analysis of different driving situations, so that the system can focus on the features most relevant to the adjustment of the atmosphere light. In view of the color-brightness coupling problem existing in the traditional vehicle-mounted atmosphere light controller, the present invention proposes a parameter decoupling processing mechanism. By establishing a LAFC parameter coupling relationship model and a dual-input dual-output disturbance observation mechanism, the independent and precise control of the four parameters of hue, saturation, brightness and brightness is achieved, making the adjustment of the vehicle-mounted atmosphere light more accurate. Based on the situational adaptive lighting control method, the system can dynamically adjust the lighting parameters according to the identified driving situation, and perform differentiated control on different areas in the car, meeting the personalized lighting needs of different occupants in different driving scenarios. The innovative introduction of the dynamic situational weighted loss function solves the problem of unbalanced lighting requirements in different driving scenarios. By adjusting the situational weight factor, the system can adjust the priority for specific driving situations and optimize the overall lighting effect. In addition, the system also continuously accumulates driving experience through the lighting strategy self-learning module to continuously improve the lighting control effect.

[0021] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The image of the interior environment of the vehicle is collected by a high-resolution camera installed at the rearview mirror inside the vehicle to obtain an in-vehicle camera image; The vehicle-mounted sensor array is used to collect environmental parameters. The vehicle-mounted sensor array includes a light sensor, a temperature sensor, an acceleration sensor and a GPS positioning module to obtain the vehicle's surrounding environmental parameter data; Packing the in-vehicle camera image and the vehicle surrounding environment parameter data according to a predefined data packet format to obtain a formatted data packet, where the predefined data packet format includes a timestamp, a sensor ID, a data value, and a status identifier; Divide the formatted data packets into real-time high-priority data streams and scheduled low-priority data streams to obtain hierarchical data streams, and dynamically adjust and process the hierarchical data streams according to the vehicle driving status to obtain in-vehicle image data and vehicle-mounted sensor data; The in-vehicle image data and on-board sensor data are input into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fused feature map.

[0022] Specifically, a high-resolution camera is installed at the rearview mirror inside the vehicle. The camera shoots a high-definition picture of 1920×1080 pixels at a rate of 30 frames per second, ensuring full coverage of the driver's face and the main areas inside the vehicle, and obtaining an accurate image of the vehicle's environment. Under the coordinated work of the camera, the vehicle-mounted sensor array is used to collect environmental parameters. The sensor array includes a light sensor, a temperature sensor, an acceleration sensor and a GPS positioning module. The light sensor is used to measure the light intensity inside and outside the vehicle. Its sampling frequency is set to 10Hz and the measurement range is 0 to 100,000 lux to ensure that the light changes in the vehicle environment can be accurately sensed. The temperature sensor is used to record the temperature inside the vehicle. Its accuracy reaches ±0.5℃ and the sampling frequency is set to 1Hz to ensure the real-time nature of the data. The acceleration sensor is used to measure the dynamic acceleration of the vehicle. The range is set to ±2g and the sampling frequency reaches 50Hz, so as to ensure that the acceleration changes of the vehicle can be accurately captured. The GPS module obtains the vehicle's location information, driving speed and direction in real time. Its update frequency is 1Hz to ensure the accuracy of the positioning information. The in-vehicle camera image and the vehicle's surrounding environment parameter data are formatted to ensure that data from different sources can be efficiently stored and calculated. All data are uniformly packaged according to the predefined data packet format. Each data packet contains a timestamp, sensor ID, data value, and status identifier. This formatting method helps to standardize the data storage and ensure that the data is accurately aligned during the multi-source information fusion process, thereby reducing the possibility of information loss. In order to improve the efficiency of data transmission and calculation, the formatted data packets are divided into two types of data streams, namely real-time high-priority data streams and timed low-priority data streams. Among them, the real-time high-priority data stream mainly includes the in-vehicle camera image data and the sensor data with fast vehicle dynamic changes, such as acceleration information, while the timed low-priority data stream includes environmental parameter data with slow changes, such as ambient temperature and light intensity. Based on the division of data streams, the data stream is dynamically adjusted in combination with the vehicle's driving status to ensure that the most critical information is collected in different driving scenarios. For example, when the vehicle is driving at a low speed or stationary, the frame rate of image acquisition is reduced, and the sampling frequency of some sensors is reduced to reduce the calculation burden. In the case of high-speed driving or complex driving conditions, the data collection frequency is increased to ensure the integrity and timeliness of the data. A dynamic adjustment strategy based on adaptive information entropy is set up. This strategy measures the uncertainty of current information by calculating data entropy and adjusts the data collection frequency according to the changing trend of uncertainty. Specifically, data entropy is defined as for: in, Representative The normalized probability of class data, Represents the total number of data categories. When the data entropy is high, it means that the environment has changed a lot. The system automatically increases the data sampling frequency to obtain more information. When the data entropy is low, it means that the environment is relatively stable. The system appropriately reduces the data sampling frequency to reduce the computational burden and improve processing efficiency. After completing the dynamic adjustment of the data flow, the obtained in-vehicle image data and vehicle sensor data are input into the dual-branch Mamba information fusion module for feature extraction and fusion processing. In this module, the data is sent to two independent processing branches respectively, where the image data enters the image processing branch. This branch uses a multi-layer convolutional neural network for feature extraction. The convolution kernel size of the first convolution layer is set to 3×3, the step size is 1, the number of channels is set to 64, and ReLU is used as the activation function to extract basic edge features. Then the number of channels of the second convolution layer is increased to 128. , to extract higher-level image semantic information. Each convolutional layer is followed by a maximum pooling layer with a pooling kernel size of 2×2 and a step size of 2 to reduce computational complexity and retain the most critical feature information. The sensor data enters the Mamba state space model, which is specifically used to capture time series features. In this model, all sensor data are time-aligned and a bidirectional scanning method with a state dimension of 128, an expansion step of 4, and an expansion coefficient of 2 is adopted to ensure that time series information can be effectively extracted, thereby accurately modeling the dynamic state of the vehicle and the trend of environmental changes. After completing the feature extraction of the two data branches, the two types of features are cross-modally fused to fully utilize the complementarity of image and sensor data. These features are input into the cross-modal feature fusion layer, which uses the attention mechanism to calculate the importance weights of the features and normalizes the weights through the Softmax function, so that the features of different modalities can adaptively adjust their influence during the fusion process, thereby ensuring that the final fusion feature map can fully express the key feature information of the current driving situation. In order to optimize the feature fusion effect, an optimization strategy based on feature mutual information is introduced. This strategy defines feature mutual information. for: in, and Represent the characteristic data of two different modes respectively. represents the joint probability distribution, and Respectively represent their marginal probability distributions. By maximizing the mutual information of features, the system ensures that the fused features retain the key information of the image and sensor data as much as possible without losing important details. The output fused feature map contains the visual information of the in-vehicle environment and integrates the on-board sensor data, thereby fully reflecting the driving status of the vehicle and environmental changes.

[0023] In a specific embodiment, the execution step inputs the in-vehicle image data and the vehicle sensor data into the dual-branch Mamba information fusion module for feature extraction and fusion processing, and the process of obtaining the fusion feature map can specifically include the following steps: Performing image preprocessing on in-vehicle image data to obtain standardized images, and performing data preprocessing on-vehicle sensor data to obtain standardized sensor data; The standardized image is input into the image processing branch of the dual-branch Mamba information fusion module, and the spatial features are extracted through 4 convolutional layers to obtain the image spatial features; The standardized sensor data is input into the sensor data processing branch of the dual-branch Mamba information fusion module, and the time series features are extracted through the Mamba state space model. The Mamba state space model sets the state dimension to E8, the expansion step size to 4, the expansion coefficient to 2, and the bidirectional scanning mode to obtain the sensor time series features; The image spatial features and sensor temporal features are input into the cross-modal feature fusion layer of the dual-branch Mamba information fusion module. The attention weight matrix is ​​calculated using the attention mechanism and normalized by the Softmax function to obtain the adaptive fusion features. Residual connections are added to the adaptive fusion features to retain the original feature information, and layer normalization is used to prevent feature distribution shift to obtain a fused feature map.

[0024] Specifically, the in-vehicle image data is preprocessed. The image is cropped to focus on the driver's facial area and the main environmental area in the car to ensure the integrity of key information and minimize background noise. The image is scaled to 256×256 pixels to reduce the computational burden and adapt to subsequent neural network processing. Histogram equalization technology is applied during scaling to enhance image contrast and reduce the impact of illumination changes. At the same time, in order to eliminate the deviation caused by uneven illumination, the RGB channels of the image are normalized to map the pixel values ​​to the [0,1] interval to improve the stability of the model and prevent feature distribution drift caused by different illumination conditions. While preprocessing the image data, the vehicle sensor data is preprocessed to ensure that the data of different types of sensors can be fused and analyzed under the same time reference. Since different vehicle sensors have different sampling frequencies and data formats, time series alignment is performed, and interpolation is performed according to the timestamps of each sensor to ensure that all data points correspond to the same time step. Outlier detection is performed on sensor data. For example, for light sensors, if the change of multiple consecutive data points exceeds the set physical limit, these points are marked as outliers and smoothed using the sliding window averaging method to reduce the impact of environmental noise on the data. All sensor data are standardized to have a mean of zero and a variance of one, thereby ensuring the consistency of data distribution and preventing the difference in the range of different sensor data from having an uneven impact on model learning. After completing data preprocessing, the standardized image is input into the image processing branch of the dual-branch Mamba information fusion module, which uses a four-layer convolutional neural network for spatial feature extraction. The convolution kernel size of the first convolution layer is set to 3×3, the step size is 1, the number of channels is 64, and ReLU is used as the activation function to extract basic edge features; the number of channels of the second convolution layer is increased to 128, and higher-level local pattern features are continued to be extracted; the third and fourth convolution layers further extract more complex spatial features, and a maximum pooling layer is attached after each convolution layer, with a pooling kernel size of 2×2 and a step size of 2 to reduce computational complexity and retain the most representative spatial information. After four layers of convolution processing, image spatial features including in-vehicle environment information, illumination change features, and driver status are extracted. At the same time, the standardized sensor data is input into the sensor data processing branch of the dual-branch Mamba information fusion module, which uses the Mamba state space model for temporal feature extraction, where the state dimension is set to E8, the expansion step is 4, the expansion coefficient is 2, and a bidirectional scanning method is used to ensure that the temporal dependency of the sensor data can be effectively captured.The model performs state modeling on the input data and maps the sensor data to a high-dimensional state space to better describe its trend over time. Then, the sequence is expanded by expanding the window with a step size of 4, so that the system can simultaneously consider the change information in adjacent time slices and increase the data expression ability by using the expansion factor 2. At the same time, a bidirectional scanning mechanism is adopted, that is, not only predicting the current state based on past data, but also using future information for reverse adjustment to improve the integrity and accuracy of time series features. In order to improve the time series modeling ability of the Mamba state space model, an optimization method based on an adaptive memory gating mechanism is defined, which adjusts the weights of past time steps to adapt to the data change patterns in different environments. This mechanism ensures that the system adaptively adjusts the degree of attention to different historical data and maintains stable time series modeling capabilities in complex driving environments. After extracting the image spatial features and sensor temporal features, these two types of features are input into the cross-modal feature fusion layer of the dual-branch Mamba information fusion module. This layer uses the attention mechanism to calculate the importance weights of the features and normalizes them through the Softmax function, so that the features of different modalities can adaptively adjust their influence during the fusion process, thereby ensuring that the final fusion features can fully express the key information of the current driving situation. After completing the feature fusion, in order to prevent the loss of original feature information during the fusion process and improve the training stability of the model, residual connections are added to the adaptive fusion features to retain some of the original information, and layer normalization technology is used to prevent the offset of feature distribution and ensure that the data can still maintain a stable numerical range during different batches of training. The fused and optimized feature data is output as a fusion feature map, the size of which is set to 128×64×64, which contains multi-dimensional information such as in-vehicle visual information, environmental parameters, and vehicle dynamic status.

[0025] In a specific embodiment, the process of executing step S102 may specifically include the following steps: The fused feature map is extracted using three convolution kernels of different sizes to obtain feature representations of three different scales. The self-attention mechanism is applied to the feature representations of three different scales respectively, and the scaled dot product attention algorithm is used to calculate the spatial correlation within the feature map to obtain the attention-weighted feature maps of three different scales; The attention weighted feature maps of three different scales are input into the cross-scale attention fusion module, and adaptively fused through learnable weight parameters to obtain the fused feature vector; Perform full connection processing on the fused feature vector and output the driving situation category distribution; Based on the driving situation category distribution, the target feature indicators of each driving situation are extracted to obtain the driving situation feature vector. The target feature indicators include lighting conditions, driver status, vehicle speed change rate and environmental complexity. The driving situation feature vectors were quantitatively analyzed, and the correlation coefficients between E driving situations and four ambient light parameters were calculated to generate a situational correlation matrix. The four ambient light parameters included hue, saturation, brightness, and luminance.

[0026] Specifically, multi-scale feature extraction is performed on the fused feature map to capture environmental information at different scales. The fused feature map is convolved with three different sizes of convolution kernels, 3×3, 5×5, and 7×7, respectively, to ensure that local detail features, medium-scale structural information, and global patterns can be extracted. Small-sized convolution kernels focus on extracting local edge information, such as the boundaries of objects in the car, the driver's eye details, or the brightness distribution of the dashboard, while medium-sized convolution kernels are used to identify medium-scale features such as the illuminated area and the cabin layout, while larger-sized convolution kernels extract overall spatial relationships, such as the global distribution of the interior lighting and the overall sitting posture characteristics of the driver. Through this multi-scale convolution strategy, feature representations of three different scales are obtained. After obtaining feature representations of different scales, the spatial correlation within the feature map is calculated, so the self-attention mechanism is applied to these three feature representations respectively, and the scaled dot product attention algorithm is used for calculation. The algorithm transforms the input features into queries, keys and values, and calculates the dot product of the query and key matrices to measure the similarity between the features. To ensure numerical stability, the square root of the feature dimension is used for scaling. The calculated attention weights are then normalized using the Softmax function so that the sum of the attention distribution is 1. Finally, the value matrix is ​​weighted and summed using the normalized attention weights to generate an attention-weighted feature map. Since feature maps of different scales have different spatial distribution characteristics, the scaled dot product attention algorithm can ensure that the system can find the most important feature areas at different scales, thereby improving the recognition ability of driving scenarios and generating three attention-weighted feature maps of different scales. Each feature map contains feature information redistributed based on the attention mechanism, making it more consistent with the actual characteristics of the driving scenario. The three attention-weighted feature maps of different scales are input into the cross-scale attention fusion module, in which a learnable weight parameter is introduced so that the fusion process automatically adjusts the contribution ratio of features of different scales to ensure the optimality of the final fused features. The cross-scale attention fusion module maps feature vectors of different scales to the same feature space, and then uses weight parameters to perform weighted summation on them. These weight parameters are constantly adjusted during the training process to minimize the error of driving situation classification, so that the fused features can fully express multi-scale information, thereby obtaining a fused feature vector containing rich environmental information. The fused feature vector is fully connected to output the driving situation category distribution. At this stage, the fused feature vector is sent to a fully connected neural network, which contains multiple hidden layers and uses nonlinear activation functions to enhance the nonlinear expression ability of the data. The Softmax function is used in the output layer to map the feature vector to the probability distribution of the driving situation category. The probability distribution represents the classification possibility of different driving situations, such as high-speed cruising, urban congestion, night driving, rainy driving, emergency braking, etc., and the current driving situation is determined according to the highest probability category.After obtaining the driving scenario category, the key feature indicators related to the scenario are extracted to generate a driving scenario feature vector. The feature vector contains four key indicators, including lighting conditions, driver status, vehicle speed change rate and environmental complexity. The lighting conditions are measured by the in-car light sensor to quantify the brightness level in the car. The driver status is analyzed by the in-car camera to analyze the driver's facial expression and attention level to judge the driver's mental state. The vehicle speed change rate is measured by GPS and acceleration sensors to reflect the vehicle's driving mode. The environmental complexity is comprehensively evaluated by combining vehicle speed changes, external lighting conditions and driver status to judge the stability of the current driving scene. Through these four key features, the current driving scenario is accurately portrayed. Based on the driving scenario feature vector, the correlation coefficients between different driving scenarios and ambient light parameters are calculated to generate a scenario association matrix. The matrix is ​​used to quantify the weight of the ambient light adjustment requirements under different scenarios. The calculation method is based on statistical analysis and regression analysis of historical data. The size of the matrix is ​​E×4, where E represents the number of driving scenarios, and 4 corresponds to the four parameters of the ambient light, including hue, saturation, brightness and brightness. In order to calculate the correlation coefficient, a large amount of historical data is collected, and the changing trend of each ambient light parameter under each driving scenario is evaluated through the regression analysis model. Then, the Pearson correlation coefficient is calculated to quantify the dependency between these parameters. The calculation formula is as follows: in, Indicates Driving situation and The correlation coefficients between the parameters of the atmosphere lights are Representative in the The driving situation feature value in observations, is the mean of the situational feature values, Representative The ambient light parameter values ​​in the observation, is the mean value of the ambient light parameter, Represents the number of samples. By calculating the correlation coefficient, the situational correlation matrix is ​​obtained. Each element of the matrix represents the degree of influence of a certain driving situation on a certain ambient light parameter.

[0027] In a specific embodiment, the process of executing step S103 may specifically include the following steps: By experimentally measuring the visual effects under different RGB color combinations and brightness levels, the LAFC parameter coupling relationship model is established to obtain the coupling coefficient matrix, which represents the degree of mutual influence between the four parameters of hue, saturation, lightness and brightness. The HSV color space is divided into multiple grids based on the coupling coefficient matrix, and the correspondence between the visual effect and the RGB brightness is calculated in each grid to obtain a four-dimensional parameter mapping table; Based on the four-dimensional parameter mapping table, a small perturbation within the preset error range is performed, and a dual-input and dual-output perturbation observation mechanism is constructed to obtain the visual effect change data before and after the perturbation; According to the visual effect change data before and after the disturbance, the local Jacobian matrix between the parameters is calculated to obtain the parameter sensitivity matrix, and the parameter compensator is constructed based on the inverse matrix of the parameter sensitivity matrix to obtain the decoupling transformation matrix; The initial lighting parameters are determined according to the driving situation feature vector and the situation association matrix, and the initial lighting parameters are input into the decoupling transformation matrix for processing. The decoupled lighting parameters are obtained by automatically calculating the mutual influence between the adjustment parameters.

[0028] Specifically, the visual effects under different RGB color combinations and brightness levels are experimentally measured to establish the LAFC parameter coupling relationship model. The core of this model is to quantitatively analyze the mutual influence between the four parameters of hue, saturation, lightness and brightness. During the experiment, a high-precision photometer and a standardized test environment are used to record the color perception data under different RGB combinations. Through psychophysical experiments, the user's subjective evaluation of different lighting conditions is collected. At the same time, the spectrum analyzer is combined to obtain objective data to quantify the comprehensive impact of different lighting parameters on human visual perception. Through regression analysis of a large amount of experimental data, a coupling coefficient matrix is ​​constructed. The matrix is ​​used to describe the dependence between hue, saturation, lightness and brightness. For example, under a specific hue, increasing brightness will lead to a decrease in saturation, while certain specific brightness changes will cause hue shifts. These nonlinear coupling relationships are accurately quantified for subsequent decoupling calculations. The HSV color space is gridded based on the coupling coefficient matrix to construct a color mapping table. The mapping table discretizes the HSV space to form multiple high-dimensional data points. Each grid point represents a fixed combination of hue, saturation and brightness, and calculates the correspondence between the visual effect and the RGB brightness at each grid point. The HSV color space is divided into grids, where is the number of hue divisions, is the number of divisions of saturation, is the number of divisions of brightness, and the RGB brightness at each grid point is fitted by experimentally measured data to ensure accurate description of the impact of different color parameters on visual effects. Through this mapping table, the corresponding RGB brightness values ​​are found under different input conditions to match the human eye's perception characteristics of light, and used for reference and adjustment in subsequent control. After constructing a complete parameter mapping table, the parameter adjustment strategy is optimized to ensure that the adjustments between different parameters do not interfere with each other. Based on the mapping table, small perturbations within the preset error range are made, and a dual-input and dual-output perturbation observation mechanism is constructed. This mechanism applies small perturbations to the lighting parameters and observes the changes in visual effects to measure the interaction between parameters. For example, based on the currently set hue and saturation, a small perturbation amount is added or subtracted to the brightness and brightness respectively. and , then measure the change of visual effects and record the data difference before and after the disturbance. These data will be used to calculate the sensitivity between parameters, so as to establish a more accurate parameter decoupling model. After obtaining the visual effect change data before and after the disturbance, the local Jacobian matrix between parameters is calculated based on these data. This matrix is ​​used to describe the sensitivity relationship of the current lighting parameters. The calculation formula of this matrix is ​​as follows: in, Representative The first visual effect parameter (such as brightness, hue) is relative to the The partial derivatives of the control parameters (such as RGB values), Represents a small change in visual effect, while Represents a small perturbation value of the input lighting parameter. Each element of the Jacobian matrix reflects the degree of influence of a control variable on the visual effect. When the value of the matrix is ​​large, it means that the parameter has a strong influence on visual perception, and when the matrix value is small, it means that the adjustment of the parameter has a small influence on the visual effect. Since there is a coupling effect between the parameters in this matrix, its inverse matrix is ​​calculated to construct a parameter compensator to achieve the decoupling of the lighting parameters. The role of the compensator is to offset the cross-influence between different parameters through matrix transformation, so as to ensure that when adjusting a certain parameter, it will not interfere with other parameters. The decoupling transformation matrix By the Jacobian matrix The inverse matrix of is calculated, that is: The matrix is ​​used to modify the control signal so that the input lighting parameter adjustment can act independently on hue, saturation, lightness and brightness without affecting each other. After the calculation of the decoupling transformation matrix is ​​completed, the initial lighting parameters are determined according to the driving situation feature vector and the situation association matrix, and adjusted using the decoupling transformation matrix to ensure that the final output lighting parameters can adapt to the current driving environment. According to the driving scenario category, the corresponding ambient light parameter adjustment weights are extracted from the scenario association matrix, and the setting values ​​of the initial lighting parameters are determined in combination with the lighting conditions, driver status, vehicle speed change rate and environmental complexity in the driving scenario feature vector. These setting values ​​include the default values ​​of basic hue, saturation, lightness and brightness, and are dynamically adjusted in combination with the scenario information. The initial lighting parameters are input into the decoupling transformation matrix for processing to automatically calculate and adjust the mutual influence between the parameters to obtain the final decoupled lighting parameters. In this process, the decoupling transformation matrix is ​​used to perform a linear transformation on the initial lighting parameters to eliminate the dependency between the parameters and ensure that the hue, saturation, lightness and brightness are adjusted independently. The decoupled lighting parameters finally generated can provide the best visual experience in the current driving environment, and are adaptively adjusted according to subsequent driving scenario changes to ensure the vehicle ambient light control effect based on environmental perception.

[0029] In a specific embodiment, the process of executing step S104 may specifically include the following steps: For E typical driving scenarios, a scenario-lighting mapping database is constructed to obtain lighting template parameters including four basic parameters of hue, saturation, brightness, and brightness and their range of variation; Fuzzy logic control is performed according to the driving situation feature vector to obtain the initial adjustment amount of the lighting parameter, and the initial adjustment amount of the lighting parameter is dynamically optimized based on the driving situation change rate monitored in real time to obtain a smooth adjustment parameter; The interior space of the vehicle is divided into four lighting areas, including the driving area, the co-pilot area, the left side of the rear seat and the right side of the rear seat. The smooth adjustment parameters are processed by regional differentiation to obtain regional lighting control parameters. By comparing the Euclidean distance between the regionalized lighting control parameters and the lighting template parameters, the lighting effect matching degree is obtained; When the lighting effect matching degree is lower than the preset target score, the regional lighting control parameters are fine-tuned by binary search until the matching degree reaches the threshold, and the initial lighting control parameters are obtained; The dynamic context-weighted loss is calculated for the initial lighting control parameters to obtain the target lighting control parameters.

[0030] Specifically, a scenario-lighting mapping database is constructed for E typical driving scenarios, and the optimal lighting parameter combination under different driving scenarios is established, including four basic parameters of hue, saturation, brightness and brightness and their adjustable ranges, to ensure that the system dynamically adjusts the lighting scheme of the vehicle atmosphere lamp according to different environmental requirements. The construction of the mapping database is based on a large number of driving experiments and user feedback data. During the experiment, the in-vehicle lighting environment under typical scenarios such as normal driving, high-speed cruising, urban congestion, night driving, and rainy driving is collected, and the driver's subjective comfort evaluation and visual fatigue status are recorded. By analyzing these data, the optimal lighting parameter range for each scenario is summarized. Fuzzy logic control is performed based on the current driving scenario feature vector to calculate the initial adjustment amount of the lighting parameters. The input variables of fuzzy logic control include driving scenario category, ambient light intensity, driver status and vehicle speed change rate, while the output variables include hue adjustment, saturation adjustment, brightness adjustment and brightness adjustment. Each input variable is divided into multiple fuzzy sets. For example, the vehicle speed change rate is divided into stable, fluctuating and drastic changes, while the ambient light intensity is divided into weak light, normal light and strong light, and reasoning is performed in combination with preset fuzzy rules. Based on the real-time monitoring of the driving scenario change rate, the initial adjustment amount of the lighting parameters is dynamically optimized. When the driving scenario change rate is small, a smaller adjustment step size is used to make the change of the ambient light smoother to avoid sudden lighting changes causing discomfort to the driver. When the driving scenario change rate is large, such as when the vehicle enters a tunnel from daytime or enters a city congestion from a highway, the adjustment speed is accelerated to ensure that the ambient light can quickly adapt to the new driving scenario, thereby improving the adaptability and consistency of lighting adjustment. In order to optimize the lighting control effect, the interior space is divided into four independent lighting areas, including the driving area, the co-pilot area, the left side of the rear row and the right side of the rear row, and the calculated smooth adjustment parameters are processed by regional differentiation to meet the needs of occupants in different areas. The lighting parameters of each area are adjusted independently according to the activity areas and needs of the driver and passengers. After completing the calculation of the regionalized lighting control parameters, the matching degree between the current lighting scheme and the preset optimal lighting scheme is evaluated, the Euclidean distance between the regionalized lighting control parameters and the lighting template parameters is calculated, and the lighting effect matching degree is evaluated based on the calculation results. The higher the matching degree, the closer the current lighting scheme is to the optimal setting. When the lighting effect matching degree is lower than the preset target score, the system will automatically enter the optimization mode to further fine-tune the lighting parameters.During the optimization process, the regional lighting control parameters are fine-tuned by binary search, the adjustment direction of the current lighting parameters is calculated, and a better parameter combination is tried in the parameter space. Each time the adjustment is made, a new Euclidean distance is calculated, and it is determined whether the matching degree has been improved. The system will continue to perform this process until the matching degree reaches the preset threshold or the number of adjustments reaches the maximum allowable value. With the support of the optimization strategy, the system can ensure that the final lighting parameters are as close to the optimal settings as possible to improve the adaptability and consistency of the in-vehicle lighting experience. After the optimization of the lighting parameters is completed, the dynamic situational weighted loss is calculated to ensure that the final target lighting control parameters are adapted to the specific needs of different driving situations. To this end, a driving experience evaluation model is constructed, and the current lighting effect is scored based on four dimensions: visual comfort, driving assistance, emotional regulation, and situational matching. The weight of each dimension is initially set to 0.25, and is dynamically adjusted based on the driver's facial expressions, eye tracking, and operating behaviors captured by the in-vehicle camera. The scoring range is set between 0 and 100 to reflect the adaptability of the lighting solution in different driving situations. After obtaining the driving experience score, the final target lighting control parameters are calculated using a dynamic scenario weighted loss function. This loss function is weighted by the importance of different driving scenarios to optimize the lighting control strategy. During the calculation of the loss function, the adjustment of different lighting parameters is optimized according to the importance weight of the current driving environment. Through adaptive adjustment, it is ensured that the final lighting control parameters not only meet the standard lighting template, but also can be optimized according to the driver's real-time status and external environment changes to provide the best driving experience. Through the above steps, the target lighting control parameters are calculated and transmitted to the vehicle-mounted ambient light hardware execution unit, so that it can automatically adjust the lighting scheme under different driving scenarios.

[0031] In a specific embodiment, the execution step performs dynamic context weighted loss calculation on the initial lighting control parameters to obtain the target lighting control parameters may specifically include the following steps: An evaluation model is built based on the driver's facial expressions, eye movements and operating behaviors, and four evaluation dimensions are set: visual comfort, driving assistance, emotional regulation, and situation matching, to obtain an experience score. Applying a dynamic context weighted loss function to calculate the initial lighting control parameters to obtain an initial loss value, and executing a gradient descent optimization algorithm based on the initial loss value to obtain optimized lighting parameters; Conduct safety constraint inspection on the optimized lighting parameters to obtain lighting parameters that meet safety standards; Compare the lighting parameters that meet safety standards with the initial lighting control parameters through loss function evaluation to obtain optimization effect data; Based on the optimization effect data and experience scores, the situation-lighting mapping database is updated using an incremental learning method to output the target lighting control parameters.

[0032] Specifically, a comprehensive evaluation model is constructed to analyze the driver's state, and on this basis, the control parameters of the vehicle-mounted ambient light are adjusted to provide a more intelligent and personalized lighting solution. In this model, the driver's expression is captured in real time by the in-car camera, and a deep learning algorithm is used to classify the expression to identify the driver's current emotional state, such as tension, relaxation, fatigue or concentration. At the same time, the system integrates eye tracking technology to analyze key indicators such as the driver's line of sight, blinking frequency and fixation time to evaluate his attention level. The driver's operating behavior, including steering wheel rotation angle, braking force and throttle control method, is monitored through the on-board control unit to further judge the driver's driving style and state. After the information is integrated, the driving experience is divided into four core evaluation dimensions, namely visual comfort, driving assistance, emotional regulation and situation matching. Among them, visual comfort is used to measure whether the current lighting environment interferes with the driver's visual perception, driving assistance is used to evaluate whether the lighting conditions help improve driving safety, emotional regulation reflects the impact of the lighting scheme on the driver's emotional state, and situational matching measures whether the current lighting scheme is adapted to the driving situation. The scores of these four dimensions together constitute the driver's overall experience score, and the score range is set between 0 and 100 to quantitatively evaluate the adaptability of different lighting parameters. After obtaining the experience score, the dynamic situational weighted loss function is applied to calculate the initial lighting control parameters to quantify the deviation between the lighting parameters and the optimal experience. The loss function optimizes the adjustment of lighting parameters according to the importance of different driving situations. The difference between the current lighting scheme and the optimal lighting scheme is calculated and mapped to the loss function, where the loss value is calculated as follows: The system executes a gradient descent optimization algorithm based on the loss value, calculates the partial derivative of the loss function with respect to each lighting parameter in each iteration, and adjusts the parameter value according to the gradient direction to make it approach the optimal solution. In the specific implementation process, the system sets the learning rate to 0.01, the maximum number of iterations to 100, and the convergence threshold to 0.001 to ensure that the optimization process converges to the optimal solution at a reasonable computational cost. After this optimization process, the optimized lighting parameters are generated. The optimized lighting parameters are subjected to safety constraint tests to ensure that the final lighting control scheme meets road safety standards and driver visual comfort requirements. The safety constraint tests mainly include brightness limitation, color temperature control, and hue change rate constraints. Among them, the brightness limitation ensures that the light intensity of the vehicle ambient light does not exceed the maximum brightness specified by the regulations to prevent affecting the driver's or other road users' line of sight, while the color temperature control is used to ensure that the lighting scheme meets the visual comfort range of the human eye and avoids visual fatigue or glare problems caused by too high or too low color temperature.In addition, a hue change rate constraint is set to ensure that the color change of the ambient light does not exceed the set threshold within the specified time, so as to avoid sudden changes in illumination affecting the driver's attention. After this series of safety tests, lighting parameters that meet safety standards are obtained. The lighting parameters that meet safety standards are compared with the initial lighting control parameters for loss function evaluation to quantify the actual effect of the optimization process. The difference in loss values ​​before and after optimization is calculated to evaluate the improvement of the optimization strategy. If the loss value after optimization is significantly reduced, it means that the optimization strategy has effectively improved the adaptability of the lighting scheme. If the loss value does not change significantly, it indicates that the current optimization strategy needs further adjustment. In addition, the difference in experience scores before and after optimization is calculated to verify whether the optimization improves the actual experience of the driver, and the impact of different lighting parameters on the driving experience is evaluated through statistical analysis. All these evaluation data will be stored as optimization effect data for subsequent system learning and optimization. In order to ensure that the system can continuously learn and evolve to adapt to changes in different driving environments, the context-lighting mapping database is updated using an incremental learning method based on the optimization effect data and experience score. The update process adopts a batch update method, and the database is adjusted once after accumulating 100 driving data each time. During the database update process, the optimal lighting parameters in different scenarios are recalculated in combination with the newly collected data, and regression analysis is performed based on the historical optimization data to ensure that the updated database will not be over-adjusted due to short-term data deviations. During the incremental learning process, an adaptive weight adjustment strategy is used to assign higher weights to long-term stable data and lower weights to data with large recent changes to ensure the stability and adaptability of database updates. The system outputs the optimized target lighting control parameters and transmits them to the vehicle ambient light hardware execution unit via the CAN bus to achieve precise lighting control.

[0033] The above describes the vehicle atmosphere light control method based on environment perception in the embodiment of the present invention. The following describes the vehicle atmosphere light control system based on environment perception in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an embodiment of a vehicle-mounted ambient light control system based on environment perception includes: The acquisition module 201 is used to collect in-vehicle image data and vehicle-mounted sensor data and input them into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map; A situation recognition module 202 is used to perform situation recognition and analysis on the fused feature graph to obtain a driving situation feature vector and a situation association matrix; A decoupling processing module 203 is used to perform decoupling processing on the vehicle ambient light parameters according to the driving situation feature vector and the situation association matrix to obtain decoupled lighting parameters; The loss calculation module 204 is used to perform situation-adaptive lighting control and dynamic situation weighted loss calculation on the decoupled lighting parameters and driving situation feature vector to obtain target lighting control parameters.

[0034] Through the collaborative cooperation of the above-mentioned components, the dual-branch Mamba information fusion module is used to process the in-vehicle image data and the vehicle-mounted sensor data, so as to realize the effective fusion of multi-source heterogeneous data and improve the perception of driving situations. The module uses the image processing branch and the sensor data processing branch to extract spatial features and temporal features respectively, and adaptively fuses them through the cross-modal feature fusion layer, thus overcoming the problem of insufficient information from a single data source. The multi-scale driving situation attention module is adopted to extract multi-scale feature representations through convolution kernels of different sizes, and the self-attention mechanism and cross-scale attention fusion are applied to realize accurate identification and analysis of different driving situations, so that the system can focus on the features most relevant to the adjustment of the atmosphere light. In view of the color-brightness coupling problem existing in the traditional vehicle-mounted atmosphere light controller, the present invention proposes a parameter decoupling processing mechanism. By establishing the LAFC parameter coupling relationship model and the dual-input dual-output disturbance observation mechanism, the independent and precise control of the four parameters of hue, saturation, brightness and brightness is realized, so that the adjustment of the vehicle-mounted atmosphere light is more accurate. Based on the situational adaptive lighting control method, the system can dynamically adjust the lighting parameters according to the identified driving situation, and perform differentiated control on different areas in the car, meeting the personalized lighting needs of different occupants in different driving scenarios. The innovative introduction of the dynamic situational weighted loss function solves the problem of unbalanced lighting requirements in different driving scenarios. By adjusting the situational weight factor, the system can adjust the priority for specific driving situations and optimize the overall lighting effect. In addition, the system also continuously accumulates driving experience through the lighting strategy self-learning module to continuously improve the lighting control effect.

[0035] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the vehicle-mounted ambient light control method based on environmental perception.

[0036] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0037] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an environment-aware vehicle ambient light control device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0038] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle ambient light control method based on environment perception, characterized in that: include: Collect in-vehicle image data and on-board sensor data and input them into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map; Performing situation recognition and analysis on the fused feature graph to obtain a driving situation feature vector and a situation association matrix; Decoupling the vehicle ambient light parameters according to the driving scenario feature vector and the scenario association matrix to obtain decoupled lighting parameters; Context-adaptive lighting regulation and dynamic context-weighted loss calculation are performed on the decoupled lighting parameters and the driving context feature vector to obtain target lighting control parameters.

2. The vehicle-mounted ambient light control method based on environment perception according to claim 1, characterized in that: The in-vehicle image data and the vehicle-mounted sensor data are collected and input into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map, including: The image of the interior environment of the vehicle is collected by a high-resolution camera installed at the rearview mirror inside the vehicle to obtain an in-vehicle camera image; The vehicle-mounted sensor array is used to collect environmental parameters, wherein the vehicle-mounted sensor array includes a light sensor, a temperature sensor, an acceleration sensor and a GPS positioning module to obtain the vehicle surrounding environmental parameter data; Packaging the in-vehicle camera image and the vehicle surrounding environment parameter data according to a predefined data packet format to obtain a formatted data packet, wherein the predefined data packet format includes a timestamp, a sensor ID, a data value, and a status identifier; Dividing the formatted data packets into a real-time high-priority data stream and a scheduled low-priority data stream to obtain a hierarchical data stream, and dynamically adjusting the hierarchical data stream according to the driving state of the vehicle to obtain in-vehicle image data and vehicle-mounted sensor data; The in-vehicle image data and the on-board sensor data are input into a dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map.

3. The vehicle-mounted ambient light control method based on environment perception according to claim 2 is characterized in that: The in-vehicle image data and the vehicle-mounted sensor data are input into a dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map, including: Performing image preprocessing on the in-vehicle image data to obtain a standardized image, and performing data preprocessing on the in-vehicle sensor data to obtain standardized sensor data; Input the standardized image into the image processing branch of the dual-branch Mamba information fusion module, extract spatial features through 4 convolutional layers, and obtain image spatial features; Input the standardized sensor data into the sensor data processing branch of the dual-branch Mamba information fusion module, extract the timing features through the Mamba state space model, and set the state dimension to E8, the expansion step to 4, the expansion coefficient to 2, and the bidirectional scanning mode to obtain the sensor timing features; Input the image spatial features and the sensor temporal features into the cross-modal feature fusion layer of the dual-branch Mamba information fusion module, calculate the attention weight matrix using the attention mechanism and normalize it through the Softmax function, and obtain the adaptive fusion features; A residual connection is added to the adaptive fusion feature to retain the original feature information, and layer normalization is used to prevent feature distribution deviation to obtain a fusion feature map.

4. The vehicle-mounted ambient light control method based on environment perception according to claim 1, characterized in that: The performing of situation identification and analysis on the fused feature graph to obtain a driving situation feature vector and a situation association matrix includes: Extracting features from the fused feature map using three convolution kernels of different sizes to obtain feature representations of three different scales; Applying the self-attention mechanism to the feature representations of the three different scales respectively, using the scaled dot product attention algorithm, calculating the spatial correlation within the feature map, and obtaining the attention-weighted feature maps of three different scales; Inputting the three attention weighted feature maps of different scales into the cross-scale attention fusion module, performing adaptive fusion through learnable weight parameters, and obtaining a fused feature vector; Performing full connection processing on the fused feature vector and outputting driving situation category distribution; Based on the driving situation category distribution, extracting target feature indicators of each driving situation to obtain a driving situation feature vector, wherein the target feature indicators include lighting conditions, driver status, vehicle speed change rate, and environmental complexity; The driving scenario feature vectors are quantitatively analyzed, correlation coefficients between E driving scenarios and four ambient light parameters are calculated, and a scenario association matrix is ​​generated. The four ambient light parameters include hue, saturation, brightness, and luminance.

5. The vehicle-mounted ambient light control method based on environment perception according to claim 1, characterized in that: The decoupling process of the vehicle ambient light parameters according to the driving scenario feature vector and the scenario association matrix to obtain the decoupled lighting parameters includes: By experimentally measuring the visual effects under different RGB color combinations and brightness levels, a LAFC parameter coupling relationship model is established to obtain a coupling coefficient matrix, which represents the degree of mutual influence between the four parameters of hue, saturation, lightness and brightness; The HSV color space is divided into a plurality of grids based on the coupling coefficient matrix, and the corresponding relationship between the visual effect and the RGB brightness is calculated in each grid to obtain a four-dimensional parameter mapping table; Based on the four-dimensional parameter mapping table, a small disturbance within a preset error range is performed to construct a dual-input dual-output disturbance observation mechanism to obtain visual effect change data before and after the disturbance; According to the visual effect change data before and after the disturbance, the local Jacobian matrix between the parameters is calculated to obtain a parameter sensitivity matrix, and a parameter compensator is constructed based on the inverse matrix of the parameter sensitivity matrix to obtain a decoupling transformation matrix; Initial lighting parameters are determined according to the driving scenario feature vector and the scenario association matrix, and the initial lighting parameters are input into the decoupling transformation matrix for processing, and the decoupled lighting parameters are obtained by automatically calculating the mutual influence between the adjustment parameters.

6. The vehicle-mounted ambient light control method based on environment perception according to claim 1, characterized in that: The performing situation-adaptive lighting control and dynamic situation weighted loss calculation on the decoupled lighting parameters and the driving situation feature vector to obtain target lighting control parameters includes: For E typical driving scenarios, a scenario-lighting mapping database is constructed to obtain lighting template parameters including four basic parameters of hue, saturation, brightness, and brightness and their range of variation; Executing fuzzy logic control according to the driving situation feature vector to obtain an initial adjustment amount of the lighting parameter, and dynamically optimizing the initial adjustment amount of the lighting parameter based on a driving situation change rate monitored in real time to obtain a smooth adjustment parameter; The interior space of the vehicle is divided into four lighting areas, including a driving area, a co-pilot area, a rear left side, and a rear right side, and the smooth adjustment parameters are processed by regional differentiation to obtain regional lighting control parameters; Obtaining a lighting effect matching degree by comparing the Euclidean distance between the regionalized lighting control parameter and the lighting template parameter; When the lighting effect matching degree is lower than the preset target score, the regionalized lighting control parameters are fine-tuned by binary search until the matching degree reaches a threshold value, thereby obtaining initial lighting control parameters; Dynamic situational weighted loss calculation is performed on the initial lighting control parameters to obtain target lighting control parameters.

7. The vehicle-mounted ambient light control method based on environment perception according to claim 6, characterized in that: The performing dynamic situation weighted loss calculation on the initial lighting control parameter to obtain the target lighting control parameter includes: An evaluation model is built based on the driver's facial expressions, eye movements and operating behaviors, and four evaluation dimensions are set: visual comfort, driving assistance, emotional regulation, and situation matching, to obtain an experience score. Applying a dynamic context weighted loss function to calculate the initial lighting control parameters to obtain an initial loss value, and executing a gradient descent optimization algorithm based on the initial loss value to obtain optimized lighting parameters; Performing a safety constraint check on the optimized lighting parameters to obtain lighting parameters that meet safety standards; Performing loss function evaluation and comparison on the lighting parameters that meet the safety standards and the initial lighting control parameters to obtain optimization effect data; Based on the optimization effect data and the experience score, the scenario-lighting mapping database is updated using an incremental learning method, and target lighting control parameters are output.

8. A vehicle-mounted ambient light control system based on environmental perception, characterized in that: Used to implement the vehicle-mounted ambient light control method based on environment perception as described in any one of claims 1 to 7, the vehicle-mounted ambient light control system based on environment perception includes: The acquisition module is used to collect in-vehicle image data and vehicle sensor data and input them into the dual-branch Mamba information fusion module for feature extraction and fusion processing to obtain a fusion feature map; A situation recognition module, used to perform situation recognition and analysis on the fused feature map to obtain a driving situation feature vector and a situation association matrix; A decoupling processing module, used for performing decoupling processing on vehicle ambient light parameters according to the driving scenario feature vector and the scenario association matrix to obtain decoupled lighting parameters; The loss calculation module is used to perform situation-adaptive lighting regulation and dynamic situation weighted loss calculation on the decoupled lighting parameters and the driving situation feature vector to obtain target lighting control parameters.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the vehicle ambient light control method based on environment perception as claimed in any one of claims 1 to 7.

Images