A lighting brightness self-adaptive adjusting system based on visual detection

Abstract

The application discloses a kind of illumination brightness self-adaptive adjustment systems based on visual detection, comprising: image acquisition and pre-processing module, for obtaining environmental image and pre-processing;Illumination feature extraction and modeling module, for the environmental image after pre-processing is input into improved Neural ODE network, and illumination feature data is obtained;Spatial illumination distribution generation module, for forming spatial illumination distribution data according to illumination feature data;Brightness deviation calculation module, for calculating illumination brightness deviation according to spatial illumination distribution data;Adaptive control module, for inputting illumination brightness deviation into MRAC control model, and generating illumination brightness adjustment instruction;Drive execution module, for driving illumination driving circuit according to illumination brightness adjustment instruction, modulating drive current, adaptive adjustment illumination lamp brightness, real-time perception and intelligent adjustment to illumination environment brightness are realized.

Description

Technical Field

[0001] This invention relates to the fields of visual inspection and intelligent lighting, and more particularly to an adaptive lighting brightness adjustment system based on visual inspection. Background Technology

[0002] With the development of smart buildings, smart homes, and smart cities, lighting systems have gradually evolved from traditional manual switch control to intelligent and adaptive control methods. Existing lighting control methods mainly include timer-based control, sensor-based control, and human-machine interaction-based control. For example, photosensitive or infrared sensors are used to detect ambient brightness and human activity, and then simple logic is used to control the switching on / off or brightness adjustment of the lights. While these methods can achieve energy saving and convenience to some extent, they often suffer from insufficient perception of complex spatial light distribution, low adjustment accuracy, and poor adaptability because they rely on single-point sensors to obtain limited environmental information.

[0003] In existing technologies, sensor-based lighting control methods typically only reflect the brightness of local points and cannot comprehensively reflect the lighting distribution of the entire space. When there are multiple light sources or shaded areas in the environment, the information output by a single sensor may differ significantly from actual visual perception, causing the adjustment results to deviate from actual needs. Furthermore, sensors are costly and complex to install, increasing the difficulty of system deployment. On the other hand, some research attempts to introduce image processing-based lighting analysis methods, acquiring environmental images through cameras and extracting brightness information to assist in judging spatial lighting. However, traditional image processing methods mostly rely on static feature extraction and threshold segmentation, which are easily affected by noise, reflection, and dynamic environmental changes, exhibiting poor robustness and failing to achieve high-precision and real-time lighting control. Summary of the Invention

[0004] One objective of this invention is to propose a visual detection-based adaptive lighting brightness adjustment system. This invention fully utilizes technologies such as environmental image acquisition, improved Neural ODE networks, spatial illumination distribution analysis, and MRAC control models, and describes in detail the overall method of intelligent adaptive adjustment of lighting brightness by extracting illumination features through image perception, calculating global and local brightness deviations, and generating adjustment commands.

[0005] An adaptive lighting brightness adjustment system based on visual detection according to an embodiment of the present invention includes:

[0006] The image acquisition and preprocessing module is used to acquire environmental images and preprocess them.

[0007] The illumination feature extraction and modeling module is used to input the preprocessed environmental image into the improved Neural ODE network, and obtain illumination feature data through input encoding, continuous-time dynamic modeling and numerical integration.

[0008] A spatial illumination distribution generation module is used to obtain global brightness information and local brightness information based on the illumination feature data to form spatial illumination distribution data;

[0009] The brightness deviation calculation module is used to compare spatial illumination distribution data with preset brightness reference values ​​and calculate the amount of illumination brightness deviation.

[0010] The adaptive control module is used to input the lighting brightness deviation into the MRAC control model and generate lighting brightness adjustment commands.

[0011] The drive execution module is used to drive the lighting drive circuit according to the lighting brightness adjustment command, modulate the drive current, and realize the adaptive adjustment of the lighting brightness through photoelectric conversion.

[0012] Optionally, modules can be integrated using the following methods:

[0013] Environmental images are acquired using an image acquisition device and then preprocessed.

[0014] The preprocessed environmental images are dynamically processed using an improved Neural ODE network. Illumination feature data is obtained through feature extraction and continuous illumination change modeling.

[0015] Global and local brightness information are obtained from illumination feature data to generate spatial illumination distribution data.

[0016] The spatial illumination distribution data is compared with the preset brightness reference value to calculate the illumination brightness deviation.

[0017] The lighting brightness deviation is input into the MRAC control model, and the MRAC control model calculates and generates a lighting brightness adjustment command.

[0018] The lighting drive circuit is driven by the lighting brightness adjustment command to adaptively adjust the brightness of the lighting lamp.

[0019] Optionally, the preprocessing includes image denoising, image enhancement, illumination correction, scene segmentation, and feature extraction.

[0020] Optionally, obtaining the illumination feature data specifically includes:

[0021] The preprocessed environmental image is input into an improved Neural ODE network, which includes an input encoding module, a dynamic modeling module, a numerical integration module, and a feature output module. The input encoding module converts the preprocessed environmental image into an initial illumination feature vector. The dynamic modeling module performs continuous-time modeling on the initial illumination feature vector. The numerical integration module integrates the output of the dynamic modeling module within a time interval. The feature output module converts the final illumination feature vector into illumination feature data with adaptive brightness adjustment.

[0022] The preprocessed environmental image is constructed into an environmental image data matrix. This matrix is ​​then converted into an initial illumination feature vector via an input encoding module. The construction process involves extracting brightness information through a weighted RGB linear combination, normalizing the extracted brightness values ​​to ensure pixel intensity is limited to the [0,1] interval. The normalized brightness image is then spatially resampled, and bilinear interpolation is used to map the original image to a vector of size [0,1]. The grid structure is used to construct an environmental image data matrix. The transformation process is to flatten the environmental image data matrix into a vector form by rows, and then use the weight matrix and the bias vector to perform a linear mapping on the flattened vector. Combined with the activation function, the initial illumination feature vector is output.

[0023] The initial illumination feature vector is input into the dynamic modeling module, which introduces the image feature vector, time sine and cosine basis functions and nonlinear mapping functions to perform continuous modeling and obtain dynamic equations.

[0024] The target illumination feature vector is obtained by numerically integrating the dynamic equation over the time interval using the numerical integration module.

[0025] Input the target illumination feature vector into the feature output module to output illumination feature data.

[0026] Optionally, the generation of the spatial illumination distribution data specifically includes:

[0027] The brightness vector in the illumination feature data is reconstructed into an illumination brightness matrix according to a preset number of rows and columns;

[0028] The global average brightness is obtained by averaging the brightness values ​​of all pixels in the illumination brightness matrix, and the global brightness variance is obtained by averaging the squared differences between the brightness values ​​of all pixels and the global average brightness, thus forming the global brightness information.

[0029] Set the local window size, calculate the average of the pixel brightness values ​​in the neighborhood within the local window size, centered on the brightness value of each pixel in the illumination matrix, to obtain the local average brightness, and calculate the weighted squared average of the differences between the pixel brightness values ​​in the neighborhood and the local average brightness to obtain the local brightness variance, thus forming local brightness information.

[0030] The illumination matrix, global brightness information, and local brightness information are combined to generate spatial illumination distribution data.

[0031] Optionally, the calculation of the lighting brightness deviation specifically includes:

[0032] Set a preset brightness reference value, which includes a global brightness reference and a local brightness reference;

[0033] Based on the global brightness reference, the global brightness deviation is calculated, whereby the global brightness deviation refers to the difference between the global brightness information and the global brightness reference.

[0034] Based on a local brightness reference, a local brightness deviation matrix is ​​constructed. The construction of the local brightness deviation matrix refers to the element-wise subtraction of the local brightness information and the local brightness reference at corresponding positions.

[0035] Based on the local brightness deviation matrix, a local brightness deviation scalar is calculated, wherein the local brightness deviation scalar is the average of the squares of each element in the local brightness deviation matrix.

[0036] The illumination brightness deviation is obtained by weighting the square of the global brightness deviation with the local brightness deviation scalar using a non-negative weighting coefficient.

[0037] Optionally, the generation of the lighting brightness adjustment command specifically includes:

[0038] The lighting brightness deviation is defined as an error signal. The reference model in the MRAC control model is used to set the state variables and output of the reference model.

[0039] A regression vector is constructed based on the state variables, output, and error signal of the reference model, and an adaptive parameter vector is preset. The construction process involves scaling the state variables, output, and error signal of the reference model, generating cross features by multiplying the output and error signal of the reference model, and concatenating the scaled state variables, output, error signal, cross features, and constant terms of the reference model in a fixed order to form a regression vector.

[0040] The control law is generated by a linear combination of the regression vector and the adaptive parameter vector, and the control quantity is calculated. The calculation process refers to arranging the constructed regression vector into a column vector in a fixed order, multiplying the corresponding components of the adaptive parameter vector and the regression vector one by one, and summing all the products to obtain the output result of the control law.

[0041] The control quantity is associated with the error signal, the adaptive parameter vector is updated based on the product of the control quantity and the error signal, and the control law is corrected using the updated adaptive parameter vector to obtain the updated control quantity.

[0042] Apply amplitude constraints to the updated control quantity, limiting it to between a preset lower limit and an upper limit, to obtain the constrained control quantity;

[0043] The constraint control quantity is converted into a lighting brightness adjustment command and output. The conversion process involves normalizing the constraint control quantity and then converting it into a voltage signal through level mapping to form a lighting brightness adjustment command.

[0044] Optionally, the adaptive adjustment of the lighting brightness specifically includes:

[0045] The lighting brightness adjustment command is input to the lighting drive circuit in the form of a duty cycle signal;

[0046] A drive current reference value is generated based on the duty cycle signal. The drive current reference value is obtained by proportionally mapping between a preset lower current limit value and a preset upper current limit value.

[0047] In the lighting drive circuit, an actual drive current is generated, and the difference between the drive current reference value and the actual drive current is calculated to obtain the current error.

[0048] Based on the current error, the drive control quantity is generated using proportional-integral calculation.

[0049] The driving control quantity is applied to the lighting driving circuit to modulate the actual driving current, and the brightness of the lighting lamp is obtained through the photoelectric conversion relationship. The brightness of the lighting lamp is adaptively adjusted. The photoelectric conversion relationship refers to the quantitative relationship between the actual driving current and the brightness of the lighting lamp.

[0050] The beneficial effects of this invention are:

[0051] First, this invention introduces environmental image acquisition and preprocessing, and combines it with an improved Neural ODE network to model illumination features, enabling the system to simultaneously acquire global and local brightness information in complex dynamic environments. Compared with traditional methods that rely solely on single-point photosensors or simple image processing algorithms, the method of this invention can more comprehensively and realistically reflect the spatial illumination distribution, effectively overcoming the problems of insufficient overall illumination perception and neglect of local illumination differences caused by limited detection points in existing technologies, thus ensuring high-precision perception of the real environment.

[0052] Secondly, in the process of calculating brightness deviation, this invention no longer limits itself to using a single global average brightness as the criterion, but comprehensively considers both the global brightness reference value and the local brightness reference matrix. Through methods such as difference calculation, square averaging, and weighted fusion, the amount of illumination brightness deviation can be obtained more accurately. This design not only improves the precision of deviation calculation, but also allows the adjustment strategy to fully consider the actual needs of the human eye for uniform illumination, transition between light and dark, and visual comfort, thereby avoiding the problems of unreasonable adjustment, abrupt brightness, or uneven illumination caused by overly simplistic calculations in existing methods.

[0053] Finally, this invention introduces the MRAC control model into the lighting control process, enabling the dynamic generation of lighting brightness adjustment commands based on real-time calculated brightness deviations. Unlike traditional PID control or fixed threshold control, the MRAC control model can adaptively adjust to rapid changes in ambient light and quickly act on the lighting drive circuit through the drive execution module, thereby achieving real-time modulation of the lighting fixture drive current. This process not only improves the response speed and stability of lighting brightness adjustment but also effectively avoids overshoot and hysteresis, ensuring the lighting system always remains in optimal operating condition. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is an overall flowchart of a vision-based adaptive lighting brightness adjustment system proposed in this invention.

[0056] Figure 2 This is a schematic diagram of the module structure of an improved Neural ODE network for a vision-based adaptive lighting brightness adjustment system proposed in this invention. Detailed Implementation

[0057] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0058] refer to Figure 1-2 A vision-based adaptive lighting brightness adjustment system includes:

[0059] The image acquisition and preprocessing module is used to acquire environmental images and preprocess them.

[0060] The illumination feature extraction and modeling module is used to input the preprocessed environmental image into the improved Neural ODE network, and obtain illumination feature data through input encoding, continuous-time dynamic modeling and numerical integration.

[0061] A spatial illumination distribution generation module is used to obtain global brightness information and local brightness information based on the illumination feature data to form spatial illumination distribution data;

[0062] The brightness deviation calculation module is used to compare spatial illumination distribution data with preset brightness reference values ​​and calculate the amount of illumination brightness deviation.

[0063] The adaptive control module is used to input the lighting brightness deviation into the MRAC control model and generate lighting brightness adjustment commands.

[0064] The drive execution module is used to drive the lighting drive circuit according to the lighting brightness adjustment command, modulate the drive current, and realize the adaptive adjustment of the lighting brightness through photoelectric conversion.

[0065] In this embodiment, the modules are interconnected using the following method:

[0066] Environmental images are acquired using an image acquisition device and then preprocessed.

[0067] The preprocessed environmental images are dynamically processed using an improved Neural ODE network. Illumination feature data is obtained through feature extraction and continuous illumination change modeling.

[0068] Global and local brightness information are obtained from illumination feature data to generate spatial illumination distribution data.

[0069] The spatial illumination distribution data is compared with the preset brightness reference value to calculate the illumination brightness deviation.

[0070] The lighting brightness deviation is input into the MRAC control model, and the MRAC control model calculates and generates a lighting brightness adjustment command.

[0071] The lighting drive circuit is driven by the lighting brightness adjustment command to adaptively adjust the brightness of the lighting lamp.

[0072] In this embodiment, the preprocessing includes image denoising, image enhancement, illumination correction, scene segmentation, and feature extraction.

[0073] In this embodiment, obtaining the illumination feature data specifically includes:

[0074] The preprocessed environmental image is input into an improved Neural ODE network, which includes an input encoding module, a dynamic modeling module, a numerical integration module, and a feature output module. The input encoding module converts the preprocessed environmental image into an initial illumination feature vector. The dynamic modeling module performs continuous-time modeling on the initial illumination feature vector. The numerical integration module integrates the output of the dynamic modeling module within a time interval. The feature output module converts the final illumination feature vector into illumination feature data with adaptive brightness adjustment.

[0075] The preprocessed environmental image is constructed into an environmental image data matrix. This matrix is ​​then converted into an initial illumination feature vector via an input encoding module. The construction process involves extracting brightness information through a weighted RGB linear combination, normalizing the extracted brightness values ​​to ensure pixel intensity is limited to the [0,1] interval. The normalized brightness image is then spatially resampled, and bilinear interpolation is used to map the original image to a vector of size [0,1]. The grid structure is used to construct an environmental image data matrix. The transformation process is to flatten the environmental image data matrix into a vector form by rows, and then use the weight matrix and the bias vector to perform a linear mapping on the flattened vector. Combined with the activation function, the initial illumination feature vector is output.

[0076] The initial illumination feature vector is input into the dynamic modeling module, which uses image feature vectors, time sine and cosine basis functions, and nonlinear mapping functions for continuous modeling to obtain the dynamic equation:

[0077]

[0078] in, Lighting feature vector Regarding time The first derivative, For time The illumination feature vector under the following conditions This is a vectorized form of the environmental image data matrix. These are the weighting coefficients of the illumination feature vector. Weighting coefficients of the environmental image data matrix. Indicates model parameters, Let be the bias vector of the nonlinear mapping function. Let be the bias vector of the nonlinear mapping function. The coefficients of the time weight matrix are... For time frequency factor, For time bias vector, This is the sine operator. This is the cosine operator;

[0079] The target illumination feature vector is obtained by numerically integrating the dynamic equation over the time interval using the numerical integration module.

[0080] ;

[0081] in, For the target illumination feature vector, Let T be the initial illumination feature vector, and T be the upper limit time for integration. For the integrand over the time interval [0,T], by variable Perform definite integrals;

[0082] Input the target illumination feature vector into the feature output module to output illumination feature data.

[0083] In this embodiment, the generation of the spatial illumination distribution data specifically includes:

[0084] The brightness vector in the illumination feature data is reconstructed into an illumination brightness matrix according to a preset number of rows and columns;

[0085] The global average brightness is obtained by averaging the brightness values ​​of all pixels in the illumination brightness matrix, and the global brightness variance is obtained by averaging the squared differences between the brightness values ​​of all pixels and the global average brightness, thus forming the global brightness information.

[0086] Set the local window size, calculate the average of the pixel brightness values ​​in the neighborhood within the local window size, centered on the brightness value of each pixel in the illumination matrix, to obtain the local average brightness, and calculate the weighted squared average of the differences between the pixel brightness values ​​in the neighborhood and the local average brightness to obtain the local brightness variance, thus forming local brightness information.

[0087] The illumination matrix, global brightness information, and local brightness information are combined to generate spatial illumination distribution data.

[0088] In this embodiment, the calculation of the lighting brightness deviation specifically includes:

[0089] Set a preset brightness reference value, which includes a global brightness reference and a local brightness reference. The global brightness reference is a single scalar, and the local brightness reference is a two-dimensional matrix with m rows and n columns.

[0090] Based on the global brightness reference, the global brightness deviation is calculated, whereby the global brightness deviation refers to the difference between the global brightness information and the global brightness reference.

[0091] Based on a local brightness reference, a local brightness deviation matrix is ​​constructed. The construction of the local brightness deviation matrix refers to the element-wise subtraction of the local brightness information and the local brightness reference at corresponding positions.

[0092] Based on the local brightness deviation matrix, a local brightness deviation scalar is calculated, wherein the local brightness deviation scalar is the average of the squares of each element in the local brightness deviation matrix.

[0093] The illumination brightness deviation is obtained by weighting the square of the global brightness deviation with the local brightness deviation scalar using a non-negative weighting coefficient.

[0094] In this embodiment, the generation of the lighting brightness adjustment command specifically includes:

[0095] The lighting brightness deviation is defined as an error signal. The reference model in the MRAC control model is used to set the state variables and output of the reference model.

[0096] A regression vector is constructed based on the state variables, output, and error signal of the reference model, and an adaptive parameter vector is preset. The construction process involves scaling the state variables, output, and error signal of the reference model, generating cross features by multiplying the output and error signal of the reference model, and concatenating the scaled state variables, output, error signal, cross features, and constant terms of the reference model in a fixed order to form a regression vector.

[0097] The control law is generated by a linear combination of the regression vector and the adaptive parameter vector, and the control quantity is calculated. The calculation process refers to arranging the constructed regression vector into a column vector in a fixed order, multiplying the corresponding components of the adaptive parameter vector and the regression vector one by one, and summing all the products to obtain the output result of the control law.

[0098] The control quantity is associated with the error signal, the adaptive parameter vector is updated according to the product of the control quantity and the error signal, and the control law is corrected using the updated adaptive parameter vector to obtain the updated control quantity. The update rate of the adaptive parameter update law is controlled by a symmetric positive definite gain matrix.

[0099] Apply amplitude constraints to the updated control quantity, limiting it to between a preset lower limit and an upper limit, to obtain the constrained control quantity;

[0100] The constraint control quantity is converted into a lighting brightness adjustment command and output. The conversion process involves normalizing the constraint control quantity and then converting it into a voltage signal through level mapping to form a lighting brightness adjustment command.

[0101] In this embodiment, the adaptive adjustment of the lighting brightness specifically includes:

[0102] The lighting brightness adjustment command is input to the lighting drive circuit in the form of a duty cycle signal;

[0103] A drive current reference value is generated based on the duty cycle signal. The drive current reference value is obtained by proportionally mapping between a preset lower current limit value and a preset upper current limit value.

[0104] In the lighting drive circuit, an actual drive current is generated, and the difference between the drive current reference value and the actual drive current is calculated to obtain the current error.

[0105] Based on the current error, the drive control quantity is generated using proportional-integral calculation.

[0106] The driving control quantity is applied to the lighting driving circuit to modulate the actual driving current, and the brightness of the lighting lamp is obtained through the photoelectric conversion relationship. The brightness of the lighting lamp is adaptively adjusted. The photoelectric conversion relationship refers to the quantitative relationship between the actual driving current and the brightness of the lighting lamp.

[0107] Example 1:

[0108] In an office building scenario, natural light enters the interior through glass windows during the day, and the building is also equipped with multiple LED lighting fixtures. Since the intensity of external light varies significantly with time, weather, and building orientation, traditional lighting systems typically rely on photosensitive sensors or manual adjustments for control. This approach has the following problems: sensors can only detect the brightness of local points and cannot reflect the global and local distribution differences across the entire space; manual adjustments lack timeliness, easily leading to energy waste or uneven lighting. The embodiments of this invention introduce an environmental image acquisition device to capture real-time images of the office area and utilize an improved neural differential equation network to model the illumination features in the images, thereby simultaneously acquiring global brightness information and local brightness distribution, solving the problems of incomplete perception and inaccurate adjustment inherent in traditional methods.

[0109] In this scenario, the system first acquires environmental images using a wide-angle camera mounted on the ceiling. After preprocessing, the images are input into an improved Neural ODE network. Illumination features are encoded and integrated within a continuous-time dynamic framework to obtain spatial illumination distribution data. The system further compares global and local brightness reference values ​​to calculate the illumination brightness deviation. Subsequently, the deviation is input to the model reference adaptive control module, which generates dynamic adjustment commands and transmits them to the drive circuit to modulate the drive current of the LED luminaires in real time. In this way, the illumination brightness can automatically adjust according to changes in external light, keeping indoor lighting in a comfortable and energy-efficient state.

[0110] To verify the beneficial effects of this invention, three typical lighting conditions were set up in the experiment: cloudy morning (external brightness approximately 250 lx), sunny noon (external brightness approximately 1200 lx), and evening (external brightness approximately 80 lx). Under each lighting condition, the traditional photosensor control method and the visual detection-based adaptive adjustment method proposed in this invention were compared. The main indicators examined during the comparison included indoor average illuminance, illuminance uniformity, system response time, and energy consumption.

[0111] Experimental results show that the method of this invention outperforms traditional methods in terms of indoor illuminance stability, visual comfort, and energy saving. For example, under sunny midday conditions, traditional sensor methods are prone to over-adjustment due to strong light interference, resulting in localized areas that are too bright or too dark indoors. In contrast, this invention, through global image modeling, can more accurately calculate brightness deviations, maintaining indoor illuminance at a comfortable level of approximately 500 lx. Simultaneously, the system response time is significantly shortened, enabling stable lighting even with rapid changes in external light.

[0112] Table 1 Comparison of lighting adjustment effects under different lighting conditions

[0113]

[0114] As can be seen from the experimental data in Table 1, the present invention exhibits significant advantages under different lighting conditions. On a cloudy morning, while traditional sensor control can raise indoor illuminance to a usable level, the average illuminance remains below the comfort level, and the illuminance uniformity is poor, resulting in some areas being too dark. The method of the present invention, however, can better balance the overall lighting, bringing the average indoor illuminance close to 500 lx and improving the uniformity to 0.81, significantly improving visual comfort.

[0115] On sunny midday days, with strong external sunlight, traditional methods suffer from sensor saturation leading to unstable adjustments, resulting in localized indoor illuminance exceeding 600 lx and causing glare. This invention, however, stabilizes the average indoor illuminance at 498 lx through global image modeling and local brightness deviation correction, approaching standard office illuminance levels while maintaining high uniformity. Simultaneously, the system response time is reduced to 3.1 seconds, enabling faster adaptation to sudden changes in external lighting.

[0116] In the evening, traditional methods often involve excessive lighting to ensure brightness, leading to energy consumption of up to 3.7 kWh / hour and low illuminance uniformity. The method of this invention, however, achieves reasonable supplemental lighting through precise deviation calculation, maintaining illuminance at a comfortable level while reducing energy consumption to 2.9 kWh / hour, demonstrating significant energy savings.

[0117] In summary, this invention not only solves the shortcomings of traditional sensor methods in spatial illumination perception and control accuracy, but also demonstrates excellent performance in visual comfort, response speed and energy saving, showing strong practical value and promising prospects for widespread application.

Claims

1. A visual detection-based adaptive lighting brightness adjustment system, characterized in that, include: The image acquisition and preprocessing module is used to acquire environmental images and preprocess them. The illumination feature extraction and modeling module is used to input the preprocessed environmental image into the improved Neural ODE network, and obtain illumination feature data through input encoding, continuous-time dynamic modeling and numerical integration. A spatial illumination distribution generation module is used to obtain global brightness information and local brightness information based on the illumination feature data to form spatial illumination distribution data; The brightness deviation calculation module is used to compare spatial illumination distribution data with preset brightness reference values ​​and calculate the amount of illumination brightness deviation. The adaptive control module is used to input the lighting brightness deviation into the MRAC control model and generate lighting brightness adjustment commands. The drive execution module is used to drive the lighting drive circuit according to the lighting brightness adjustment command, modulate the drive current, and realize the adaptive adjustment of the lighting brightness through photoelectric conversion. The generation of the spatial illumination distribution data specifically includes: The brightness vectors in the illumination feature data are reconstructed into an illumination brightness matrix according to a preset number of rows and columns. The reconstruction process is to fill the brightness vectors into the two-dimensional matrix in a row-major manner according to the preset number of rows and columns to form an illumination brightness matrix. The global average brightness is obtained by averaging the brightness values ​​of all pixels in the illumination brightness matrix, and the global brightness variance is obtained by averaging the squared differences between the brightness values ​​of all pixels and the global average brightness, thus forming the global brightness information. Set the local window size, calculate the average of the pixel brightness values ​​in the neighborhood within the local window size, centered on the brightness value of each pixel in the illumination matrix, to obtain the local average brightness, and calculate the weighted squared average of the differences between the pixel brightness values ​​in the neighborhood and the local average brightness to obtain the local brightness variance, thus forming local brightness information. The illumination luminance matrix, global luminance information, and local luminance information are combined to generate spatial illumination distribution data; The calculation of the lighting brightness deviation specifically includes: Set a preset brightness reference value, which includes a global brightness reference and a local brightness reference; Based on the global brightness reference, the global brightness deviation is calculated, whereby the global brightness deviation refers to the difference between the global brightness information and the global brightness reference. Based on a local brightness reference, a local brightness deviation matrix is ​​constructed. The construction of the local brightness deviation matrix refers to the element-wise subtraction of the local brightness information and the local brightness reference at corresponding positions. Based on the local brightness deviation matrix, a local brightness deviation scalar is calculated, wherein the local brightness deviation scalar is the average of the squares of each element in the local brightness deviation matrix. The illumination brightness deviation is obtained by weighting the square of the global brightness deviation with the local brightness deviation scalar using a non-negative weighting coefficient.

2. The adaptive lighting brightness adjustment system based on visual detection according to claim 1, characterized in that, The modules are connected in the following way: Environmental images are acquired using an image acquisition device and then preprocessed. The preprocessed environmental images are dynamically processed using an improved Neural ODE network. Illumination feature data is obtained through feature extraction and continuous illumination change modeling. Global and local brightness information are obtained from illumination feature data to generate spatial illumination distribution data. The spatial illumination distribution data is compared with the preset brightness reference value to calculate the illumination brightness deviation. The lighting brightness deviation is input into the MRAC control model, and the MRAC control model calculates and generates a lighting brightness adjustment command. The lighting drive circuit is driven by the lighting brightness adjustment command to adaptively adjust the brightness of the lighting lamp.

3. The adaptive lighting brightness adjustment system based on visual detection according to claim 2, characterized in that, The preprocessing includes image denoising, image enhancement, illumination correction, scene segmentation, and feature extraction.

4. The adaptive lighting brightness adjustment system based on vision detection according to claim 2, characterized in that, The acquisition of the illumination feature data specifically includes: The preprocessed environmental image is input into an improved Neural ODE network, which includes an input encoding module, a dynamic modeling module, a numerical integration module, and a feature output module. The input encoding module converts the preprocessed environmental image into an initial illumination feature vector. The dynamic modeling module performs continuous-time modeling on the initial illumination feature vector. The numerical integration module integrates the output of the dynamic modeling module within a time interval. The feature output module converts the final illumination feature vector into illumination feature data with adaptive brightness adjustment. The preprocessed environmental image is constructed into an environmental image data matrix. This matrix is ​​then converted into an initial illumination feature vector via an input encoding module. The construction process involves extracting brightness information through a weighted RGB linear combination, normalizing the extracted brightness values ​​to ensure pixel intensity is confined to the [0,1] interval, and spatially resampling the normalized brightness image. Bilinear interpolation is then used to map the original image to a vector of size […]. The grid structure is used to construct an environmental image data matrix. The transformation process is to flatten the environmental image data matrix into a vector form by rows, and then use the weight matrix and the bias vector to perform a linear mapping of the flattened vector. Combined with the activation function, the initial illumination feature vector is output. The initial illumination feature vector is input into the dynamic modeling module, which introduces the image feature vector, time sine and cosine basis functions and nonlinear mapping functions to perform continuous modeling and obtain dynamic equations. The target illumination feature vector is obtained by numerically integrating the dynamic equation over the time interval using the numerical integration module. Input the target illumination feature vector into the feature output module to output illumination feature data.

5. The adaptive lighting brightness adjustment system based on visual detection according to claim 2, characterized in that, The generation of the lighting brightness adjustment command specifically includes: The lighting brightness deviation is defined as an error signal. The reference model in the MRAC control model is used to set the state variables and output of the reference model. A regression vector is constructed based on the state variables, output, and error signal of the reference model, and an adaptive parameter vector is preset. The construction process involves scaling the state variables, output, and error signal of the reference model, generating cross features by multiplying the output and error signal of the reference model, and concatenating the scaled state variables, output, error signal, cross features, and constant terms of the reference model in a fixed order to form a regression vector. The control law is generated by a linear combination of the regression vector and the adaptive parameter vector, and the control quantity is calculated. Specifically, the constructed regression vector is arranged into a column vector in a fixed order, the corresponding components of the adaptive parameter vector and the regression vector are multiplied one by one, and all products are summed to obtain the output result of the control law. The control quantity is associated with the error signal, the adaptive parameter vector is updated based on the product of the control quantity and the error signal, and the control law is corrected using the updated adaptive parameter vector to obtain the updated control quantity. Apply amplitude constraints to the updated control quantity, limiting it to between a preset lower limit and an upper limit, to obtain the constrained control quantity; The constraint control quantity is converted into a lighting brightness adjustment command and output. The conversion process involves normalizing the constraint control quantity and then converting it into a voltage signal through level mapping to form a lighting brightness adjustment command.

6. The adaptive lighting brightness adjustment system based on visual detection according to claim 2, characterized in that, The adaptive adjustment of the lighting brightness specifically includes: The lighting brightness adjustment command is input to the lighting drive circuit in the form of a duty cycle signal; A drive current reference value is generated based on the duty cycle signal. The drive current reference value is obtained by proportionally mapping between a preset lower current limit value and a preset upper current limit value. In the lighting drive circuit, an actual drive current is generated, and the difference between the drive current reference value and the actual drive current is calculated to obtain the current error. Based on the current error, the drive control quantity is generated using proportional-integral calculation. The driving control quantity is applied to the lighting driving circuit to modulate the actual driving current, and the brightness of the lighting lamp is obtained through the photoelectric conversion relationship. The brightness of the lighting lamp is adaptively adjusted. The photoelectric conversion relationship refers to the quantitative relationship between the actual driving current and the brightness of the lighting lamp.

Images