A camera-aware LED lighting adaptive dimming method and system

Abstract

This invention discloses a camera-sensing adaptive dimming method and system for LED lighting, comprising: controlling an LED light source to output at least two different detection illumination states; acquiring environmental images of the same scene under each detection illumination state; preprocessing the environmental images to obtain a target area image; extracting a subset of reliable pixels from the target area image after interference suppression processing; constructing a multi-source response vector of the target area based on the reliable pixel subsets under each detection illumination state; selecting a locally optimal sample subset from a preset sample library based on the multi-source response vector; calculating reflection characteristic characterization parameters; correcting the illuminance and color temperature estimation results of the target area to obtain the target brightness, target color temperature, and dimming parameters; and adjusting the LED light source using the generated dimming parameters. This invention can improve the accuracy of target area light environment perception under complex reflection environments and enhance the accuracy and scene adaptability of LED brightness and color temperature adjustment.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting control technology, and in particular to a camera-sensing method and system for adaptive dimming of LED lighting. Background Technology

[0002] With the development of LED lighting technology, lighting systems have gradually evolved from traditional fixed switch control and tiered dimming control to intelligent and adaptive control. In existing technologies, a common approach is to collect environmental information using photoresistors, illuminance sensors, infrared sensors, or human presence sensors, and then adjust the brightness and color temperature of the LED light source according to preset thresholds, control rules, or simple feedback strategies to meet the lighting needs of different indoor scenarios such as offices, classrooms, and residences. However, lighting control methods based on single photosensitive devices or a small number of environmental sensors typically only acquire local illuminance information, making it difficult to reflect the true lighting conditions of different areas within an indoor scene. This is especially true in situations with complex surface materials, partial obstructions, reflective objects, strong light near windows, and shadows in dark areas, easily leading to discrepancies between the collected results and the actual visual environment. Furthermore, these methods often focus on adjusting the overall brightness and darkness, with weak perception capabilities for spatial distribution characteristics, color change characteristics, and the reflective properties of scene surfaces, making it difficult to achieve more precise joint adjustment of brightness and color temperature.

[0003] To improve environmental perception capabilities, existing technologies have developed solutions that utilize cameras to capture scene images and combine them with image processing techniques for lighting adjustment. While these solutions can obtain richer environmental information than traditional point sensors, in practical applications, the images captured by cameras are easily affected by high-intensity reflection, specular reflection, low-light noise, changes in exposure conditions, and differences in the reflectivity of different surfaces. This results in inconsistencies between the image brightness, color, and other representations and the actual illuminance and color temperature of the target area. When judging directly based on pixel values ​​of the entire image or a partial image, inaccurate identification of the true lighting environment of the scene can easily occur. Furthermore, many existing image-sensing lighting control solutions analyze image information based on single acquisition results or under a single lighting condition, making it difficult to fully distinguish between image response differences caused by changes in external lighting and those caused by the surface reflectivity of the target area. They also struggle to reliably obtain effective pixel information that accurately reflects the optical response of the target area in complex indoor environments. This leads to insufficient accuracy in subsequent illuminance estimation, color temperature estimation, and dimming parameter generation, consequently affecting the stability, precision, and scene adaptability of LED lighting adjustment.

[0004] Therefore, how to acquire image information that can more accurately represent the real light environment of the target area based on camera perception, and on this basis, realize adaptive adjustment of the brightness and color temperature of LED light source, remains a problem that needs further research in current intelligent lighting control technology. Summary of the Invention

[0005] The purpose of this invention is to address the problems of insufficient accuracy in light environment perception, brightness, and color temperature adjustment in existing LED lighting adjustment methods by providing a camera-sensing adaptive dimming method and system for LED lighting.

[0006] To achieve the above objectives, a first aspect of the present invention provides a camera-sensing adaptive dimming method for LED lighting. The method includes: controlling an LED light source to output at least two different detection lighting states; acquiring environmental images of the same scene under each detection lighting state using a camera; preprocessing the environmental images to obtain a target region image; extracting a subset of reliable pixels from the target region image, wherein the subset of reliable pixels is a set of effective pixels in the target region obtained after interference suppression processing; constructing a multi-source response vector for the target region based on the subset of reliable pixels under each detection lighting state; selecting a locally optimal sample subset from a preset sample library based on the multi-source response vector to calculate a reflection characteristic characterization parameter for the target region; correcting the illuminance and color temperature estimation results of the target region using the reflection characteristic characterization parameter to obtain the target brightness and target color temperature, generating corresponding dimming parameters, and adjusting the LED light source using the dimming parameters.

[0007] In one embodiment, the at least two different detection illumination states are illumination states that differ in at least one of the following parameters: color temperature parameter, brightness parameter, LED sub-light source combination, and spectral power distribution.

[0008] In one embodiment, the preprocessing of the environmental image includes: restoring the brightness of the environmental image based on the brightness response curve of the camera; performing multi-exposure fusion on environmental images acquired under different exposure conditions to expand the brightness dynamic range of the target area; and performing low-light enhancement, dark area compensation, and local brightness correction on the fused environmental image to obtain the target area image.

[0009] In one embodiment, extracting a subset of reliable pixels from the target region image includes: identifying specular reflection pixels, specular reflection pixels, and low-light noise pixels in the target region; filtering out specular reflection pixels, specular reflection pixels, and low-light noise pixels; and retaining stable pixels with consistent responses under different detection illumination conditions among the diffuse reflection feature pixels as the subset of reliable pixels.

[0010] In one embodiment, the calculation of the reflectance characteristic characterization parameters of the target region includes: extracting response features of the target region based on a subset of reliable pixels under each detection illumination state, and combining the extracted response features under each detection illumination state to form a multi-source response vector; calculating the similarity between the multi-source response vector and the response features of each training sample in a preset sample library, selecting at least two training samples whose similarity meets a preset condition to form a local optimal sample subset; assigning corresponding weights to the training samples in the local optimal sample subset according to the similarity of each training sample, and calculating the reflectance characteristic characterization parameters based on the weighted local optimal sample subset.

[0011] In one embodiment, the reflectance characteristic characterization parameters include at least one of the following: a regional reflectance characterization parameter for characterizing the surface reflectance characteristics of the target area, a classification parameter for characterizing the reflectance type of the target area, a specular reflection intensity parameter for characterizing the specular reflection degree of the target area, a diffuse reflection proportion parameter for characterizing the diffuse reflection contribution of the target area, a reflectance compensation coefficient for correcting the color temperature estimation result, and a color temperature compensation coefficient for correcting the color temperature estimation result.

[0012] In one embodiment, obtaining the target brightness and target color temperature includes: extracting brightness response features and color response features based on a subset of reliable pixels in the target area under various detection illumination states, generating an illuminance estimation result based on the brightness response features, and generating a color temperature estimation result based on the color response features; matching the reflection characteristic characterization parameters of the target area with preset correction rules to determine illuminance correction parameters and color temperature correction parameters respectively; using the illuminance correction parameters to perform weighted correction or bias compensation on the illuminance estimation results to obtain corrected illuminance, and determining the target brightness based on the corrected illuminance according to a preset brightness mapping relationship; using the color temperature correction parameters to perform weighted correction or bias compensation on the color temperature estimation results to obtain corrected color temperature, and determining the target color temperature based on the corrected color temperature according to a preset color temperature mapping relationship.

[0013] In one embodiment, generating the corresponding dimming parameters includes: normalizing the target brightness and target color temperature and then inputting them into a lightweight neural network model, wherein the lightweight neural network model includes an input layer, a low-dimensional feature mapping layer, and a parameter output layer connected in sequence; the low-dimensional feature mapping layer is used to perform joint feature extraction and nonlinear mapping on the normalized target brightness and target color temperature, and the parameter output layer is used to output dimming parameters corresponding to the brightness adjustment and color temperature adjustment of the LED light source.

[0014] In one embodiment, adjusting the LED light source using dimming parameters includes: determining the brightness adjustment amount and the color temperature adjustment amount according to the dimming parameters; limiting the brightness adjustment amount and the color temperature adjustment amount; and gradually updating the brightness driving state and the color temperature driving state of the LED light source according to a preset time window.

[0015] A second aspect of the present invention provides a camera-sensing adaptive dimming system for LED lighting, the system comprising: The detection lighting control module is used to control the LED light source to output at least two different detection lighting states; The image acquisition module is used to acquire environmental images of the same scene under various detection lighting conditions using a camera; The image preprocessing module is used to preprocess the environmental image to obtain the target area image; The trusted pixel extraction module is used to extract a subset of trusted pixels from the target region image. The trusted pixel subset is the set of effective pixels in the target region obtained after interference suppression processing. The reflection characteristic parameter calculation module is used to construct the multi-source response vector of the target area based on the reliable pixel subset under each detection illumination state, and to select the local optimal sample subset from the preset sample library according to the multi-source response vector in order to calculate the reflection characteristic characterization parameters of the target area. The correction and parameter generation module is used to correct the illuminance and color temperature estimation results of the target area using the reflection characteristic characterization parameters, so as to obtain the target brightness and target color temperature, and generate the corresponding dimming parameters. The dimming execution module is used to adjust the LED light source using dimming parameters.

[0016] Compared with the prior art, the dimming method and system provided by the present invention have the following beneficial effects: 1. This invention actively detects and acquires images of the same scene by setting at least two different detection lighting states. Compared with single image acquisition based on only a single lighting state, it is more effective in distinguishing the differences in image response caused by changes in lighting and the differences in image response caused by the surface reflection characteristics of the target area, thereby improving the accuracy of subsequent reflection characteristic estimation.

[0017] 2. This invention reduces the impact of high light reflection, specular reflection and low illumination noise on subsequent image analysis by extracting a subset of reliable pixels after interference suppression processing from the target area image. Furthermore, it further filters stable pixels with consistent responses under different detection illumination conditions from the diffuse reflection feature pixels, which helps to improve the reliability of multi-source response vector construction.

[0018] 3. This invention selects a locally optimal sample subset from a preset sample library based on the multi-source response vector, and calculates the reflection characteristic characterization parameters based on the weighted locally optimal sample subset. This makes the reflection characteristic estimation of the target area closer to the actual situation of the current scene, thereby improving the correction accuracy of the illuminance and color temperature estimation results.

[0019] 4. This invention corrects the illuminance and color temperature estimation results of the target area by using reflection characteristic characterization parameters, and obtains the target brightness and target color temperature on this basis, thereby generating dimming parameters, which can improve the accuracy and scene adaptability of LED light source brightness adjustment and color temperature adjustment.

[0020] 5. This invention achieves low-dimensional mapping from target brightness and target color temperature to dimming parameters through a lightweight neural network model, and performs LED light source adjustment by combining amplitude limiting processing and gradual updates according to a preset time window, which is beneficial to balance computational efficiency, adjustment stability and practical deployment feasibility. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the adaptive dimming method for LED lighting sensed by a camera in an embodiment of the present invention; Figure 2 This is a schematic diagram of the camera-sensing LED lighting adaptive dimming system in an embodiment of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, equivalent substitutions, modifications or improvements made without departing from the concept of the present invention should all fall within the scope of protection of the present invention.

[0023] This invention provides a camera-sensing adaptive dimming method for LED lighting, such as... Figure 1 As shown, the method specifically includes the following implementation steps: controlling the LED light source to output at least two different detection illumination states, and using a camera to acquire environmental images of the same scene under each detection illumination state; preprocessing the environmental images to obtain target area images; extracting a reliable pixel subset from the target area images, wherein the reliable pixel subset is the set of effective pixels in the target area obtained after interference suppression processing; constructing a multi-light source response vector of the target area based on the reliable pixel subset under each detection illumination state, and selecting a locally optimal sample subset from a preset sample library based on the multi-light source response vector to calculate the reflection characteristic characterization parameters of the target area; using the reflection characteristic characterization parameters to correct the illuminance and color temperature estimation results of the target area to obtain the target brightness and target color temperature, and generating corresponding dimming parameters, and using the dimming parameters to adjust the LED light source.

[0024] The detection illumination state refers to the different output states of the LED light source when used for scene detection. Different detection illumination states differ in at least one optical output dimension, resulting in distinguishable image responses for the same target area under different detection illumination states. The detection illumination state is a short-term detection state, its purpose being to provide multiple sets of inputs for subsequent optical response analysis, rather than being directly maintained as the final illumination state for a long period of time.

[0025] The environmental image refers to scene image data acquired by a camera under various detection lighting conditions of the same scene. The environmental image reflects the optical performance of different areas in the scene under the corresponding detection lighting conditions, including brightness distribution, color distribution, and response differences caused by surface reflection characteristics. Because the environmental image is simultaneously affected by lighting conditions, surface material, reflection, and imaging noise, it cannot be directly used as a basis for dimming and requires subsequent processing.

[0026] The preprocessing refers to the basic normalization and analyzability processing of the environmental image to extract the target region image for subsequent calculations. The preprocessing should achieve at least one or more of the following: improving the overall usability of the image, suppressing the influence of irrelevant background areas, unifying the comparability of image data under different detection illumination conditions, and improving the clarity of the target region boundaries. After preprocessing, the target region image is obtained. The target region image refers to the region image determined from the environmental image for subsequent optical response analysis and dimming decisions. It can correspond to the illuminated work area, activity area, desktop area, ground area, or other areas requiring lighting control in an indoor scene. The purpose of the target region image is to limit subsequent analysis to the area directly related to lighting control, avoiding interference from irrelevant backgrounds.

[0027] The reliable pixel subset refers to the set of effective pixels in the target region image that, after interference suppression processing, can realistically represent the optical response of the target region. The interference suppression processing refers to weakening, eliminating, or filtering adverse factors that affect the determination of the true response of the target region. These adverse factors include at least reflection interference and noise interference. Reflection interference includes abnormal bright spots or localized distortion areas formed by specular reflection and high light intensity, while noise interference includes unstable pixel information introduced due to low lighting conditions, acquisition disturbances, or image quality degradation. By extracting the reliable pixel subset, the pixels subsequently used in response analysis can be derived from a more stable and representative effective region, rather than directly using all pixels in the target region.

[0028] After obtaining a subset of reliable pixels under each illumination condition, a multi-source response vector for the target region is constructed based on this subset. Here, the multi-source response vector refers to a vectorized representation formed by orderly combining the response features of the same target region under different illumination conditions. In other words, the same target region corresponds to a set of response information under each illumination condition. After extraction and arrangement, this response information forms a unified representation describing the overall response pattern of the target region across multiple illumination conditions. This multi-source response vector is not a single pixel value in a single image, but rather a feature expression derived from multiple illumination conditions that comprehensively reflects the trend of response changes in the target region.

[0029] The preset sample library is a pre-established set of reference samples. It stores response features corresponding to multiple samples and associated reflectance characteristic reference information. The purpose of the sample library is to provide a reference basis for the current target area, allowing the multi-source response vectors formed by the current target area under different detection illumination conditions to be compared with existing reference samples. The locally optimal sample subset refers to a subset of samples in the preset sample library that have a high similarity to the multi-source response vectors of the current target area, rather than all samples in the sample library. By selecting the locally optimal sample subset from the sample library, the subsequent calculation of reflectance characteristic parameters can more closely reflect the actual response of the current target area, reducing the influence of irrelevant samples on the calculation results.

[0030] The reflection characteristic characterization parameters refer to a set of parameters used to characterize the surface reflection properties of a target area. This parameter set describes the reflection behavior characteristics of the target area when it is illuminated, reflecting how the surface of the target area responds to incident light and the degree of influence it has on image brightness and color formation. The purpose of setting the reflection characteristic characterization parameters is to distinguish between changes in the target area image caused by the actual lighting conditions and changes caused by the surface reflection properties, thereby avoiding misjudging image deviations caused by reflection as changes in the lighting itself.

[0031] The illuminance and color temperature estimation results are estimates of the light environment state of the target area based on the current image response of the target area. Specifically, the illuminance estimation result represents the estimated light intensity of the target area, and the color temperature estimation result represents the estimated warm or cool properties of the current light environment of the target area. Since these estimation results are derived from the image response, they may still be affected by surface reflection characteristics before correction; therefore, correction is required in conjunction with reflection characteristic parameters.

[0032] By utilizing reflection characteristics as a characterization parameter, the original illuminance and color temperature estimates are corrected to more closely approximate the actual lighting environment of the target area. This correction process reduces the bias caused by high surface reflectivity, specular reflection, or other reflectivity types in illuminance and color temperature judgments. After correction, target luminance and target color temperature are obtained for controlling the lighting output. Target luminance represents the target brightness value that the LED light source should achieve, and target color temperature represents the target color temperature value that the LED light source should achieve. Both are target output quantities in the dimming control stage, no longer merely estimates of the current scene state, but control targets for subsequent dimming execution.

[0033] These dimming parameters are control parameters generated based on the target brightness and color temperature, which can be directly used to drive and adjust the LED light source. The function of these dimming parameters is to convert the target brightness and color temperature into adjustment commands that the LED light source can execute, enabling it to adjust brightness and color temperature according to the target requirements. When adjusting the LED light source using these dimming parameters, the output state of the LED light source switches from a detection illumination state to an actual adjustment state, thereby providing the target area with an illumination effect corresponding to the target brightness and color temperature.

[0034] The aforementioned adaptive dimming implementation method is geared towards indoor lighting scenarios. It controls the LED light source to output different detection illumination states within a short period and uses a camera to capture images of the same scene, obtaining environmental images that characterize the differences in the optical response of the target area. Based on this, the environmental images are processed to extract highly reliable and effective pixel information. This allows for the construction of a multi-source response representation of the target area under different detection illumination states. Reflectance characteristic parameters are calculated based on sample matching results. These parameters are then used to correct the illuminance and color temperature estimation results of the target area, yielding the target brightness and target color temperature for subsequent dimming control. Finally, corresponding dimming parameters are generated to adjust the brightness and color temperature of the LED light source. Through this implementation method, this embodiment can achieve multi-state perception of the target area using a camera without relying on local measurement results from a single sensor. Combined with reliable pixel extraction, sample matching, and reflection correction, it achieves a more accurate characterization of the light environment state of the target area, thereby improving the targeting and accuracy of LED light source brightness and color temperature adjustment.

[0035] To ensure comparable optical response differences for the same target area under different lighting conditions, multi-state probing illumination is applied to the scene before formal lighting adjustments. By setting at least two different probing illumination states for the LED light source and acquiring environmental images of the same scene under each probing illumination state, multiple sets of input information can be provided for subsequent analysis of the optical response of the target area. The focus of this approach is not on maintaining different lighting states indefinitely, but on utilizing the differences between different probing illumination states to enable the target area to form distinctive response results under each state, thereby improving the effectiveness of subsequent analysis. Based on this, the technical solution in this embodiment is as follows: the at least two different probing illumination states are illumination states that differ in at least one of the following parameters: color temperature parameter, brightness parameter, LED sub-light source combination, and spectral power distribution.

[0036] In this embodiment, the detection lighting state refers to the output state of the LED light source during scene detection and acquisition. This output state is not a stable state for long-term illumination, but a short-term operating state set up to acquire the response changes of the target area under different lighting conditions. The phrase "at least two different detection lighting states" means that during the same scene detection process, the LED light source outputs at least two different lighting output forms sequentially, enabling the camera to acquire environmental images of the same scene under different states.

[0037] The difference in at least one of the parameters—color temperature, brightness, LED sub-light source combination, and spectral power distribution—means that it is not required for all four types of parameters to change simultaneously between the two detection illumination states. A change in at least one type of parameter is sufficient to constitute a different detection illumination state. In other words, the difference between different detection illumination states can stem from a single parameter difference or from a combination of multiple parameters.

[0038] The color temperature parameter characterizes the warm or cool properties of the light output from the light source under the detection illumination state. When the color temperature parameter changes, the overall color representation of the same target area in the camera image will change accordingly. For example, one detection illumination state can be set to a warmer illumination output state, and another to a cooler illumination output state, so that the target area forms different color responses under the two different illumination conditions. The brightness parameter characterizes the set amount of light source output intensity under the detection illumination state. When the brightness parameter changes, the brightness representation of the same target area in the ambient image will change accordingly. For example, the first detection illumination state can be set to a lower brightness output state, and the second detection illumination state can be set to a higher brightness output state, so that the target area forms different brightness responses under different light intensities. LED sub-light source combination refers to the combination of LED light-emitting units used to constitute the current detection illumination state. In multi-channel, multi-sub-light source, or multi-chip LED light sources, different sub-light sources can have different light-emitting characteristics. By switching different sub-light sources to participate in light emission, different detection illumination states can be formed. For example, in a group of LED light sources, a first group of sub-light sources can be activated under a first detection illumination state, and a second group of sub-light sources can be activated under a second detection illumination state, or the number and combination of sub-light sources participating in emission can be changed under different detection illumination states. The spectral power distribution refers to the energy distribution of the light source within different wavelength ranges. When the spectral power distribution changes, even if the overall brightness of the scene does not change significantly, the color response and reflection response of the target area in camera imaging may still differ. Therefore, in this embodiment, different detection illumination states can also be formed by changing the spectral distribution of the output light from the LED light source. For example, the first detection illumination state can correspond to a set of reference spectral distributions, while the second detection illumination state can correspond to another set of spectral distributions with different energy distributions in certain wavelength bands.

[0039] In practical implementation, the source of difference in detection illumination states can be flexibly selected according to system configuration conditions. In one implementation, only the color temperature parameter is changed to create a warm / cool difference between the two detection illumination states; in another implementation, only the brightness parameter is changed to create a light / dark difference; alternatively, both the color temperature and brightness parameters can be changed simultaneously, resulting in differences in both warm / cool attributes and light / darkness between the two detection illumination states; alternatively, the combination of LED sub-light sources or the spectral power distribution can be changed to make the target area exhibit different image responses under different detection illumination states. All of the above methods belong to at least two different detection illumination state implementation forms described in this embodiment.

[0040] This embodiment sets at least two different detection illumination states to enable comparable response results for the same target area under multiple illumination conditions. Since the image performance of a target area depends not only on the current illumination conditions but also on its surface reflectivity, introducing different detection illumination states allows the target area to exhibit response differences under different illumination inputs. This response difference provides a basis for subsequent identification and analysis of the optical features of the target area. Therefore, the difference settings between detection illumination states are not arbitrarily switched but rather used as a pre-operation to construct multi-state scene acquisition conditions.

[0041] In one example of this embodiment, the LED light source can first be set to a first detection illumination state, such as a medium brightness, warm-toned output state, and a corresponding first environmental image can be captured using a camera. Then, the LED light source can be switched to a second detection illumination state, such as a cool-toned output state at the same brightness, or a higher brightness output state at the same color temperature, and a corresponding second environmental image can be captured using a camera. In this way, the same target area forms corresponding environmental images under different detection illumination states, and subsequent processing stages can analyze the response differences of the target area under different detection illumination states based on these environmental images.

[0042] This embodiment explicitly defines at least two different detection illumination states, meaning that the differences between the different detection illumination states originate from changes in at least one of the following parameters: color temperature parameter, brightness parameter, LED sub-light source combination, and spectral power distribution. Therefore, multi-state detection illumination conditions can be provided for the same scene without limiting the specific hardware structure, providing a clear implementation basis for subsequent scene image acquisition and response analysis.

[0043] The environmental image does not directly enter the subsequent analysis process, but first undergoes a continuous preprocessing chain. This is because the raw image output by the camera is usually affected by factors such as the response characteristics of the imaging device, differences in exposure conditions, loss of detail in low-light areas, and uneven local brightness. If the raw image is directly used as the basis for subsequent analysis, it can easily lead to insufficient representation of the true optical information in the target area, thus affecting the subsequent response feature extraction and parameter estimation results. Therefore, this embodiment performs layer-by-layer normalization and analyzable processing on the environmental image through brightness restoration, multi-exposure fusion, and enhancement compensation, making the final target area image closer to an effective image representation that can be used for subsequent calculations. Specifically, the preprocessing of the environmental image includes: brightness restoration of the environmental image based on the camera's brightness response curve; multi-exposure fusion of environmental images acquired under different exposure conditions to expand the dynamic range of brightness in the target area; and low-light enhancement, dark area compensation, and local brightness correction of the fused environmental image to obtain the target area image.

[0044] In this embodiment, the brightness response curve refers to the correspondence between the camera's imaging output value and the actual brightness response of the scene. Since the pixel output of a camera is usually not an ideal linear expression of scene brightness, the same real brightness may correspond to different pixel performances under different imaging conditions. Brightness restoration of the environmental image based on the camera's brightness response curve aims to correct the pixel brightness values ​​in the environmental image to a form that more closely approximates the actual brightness response of the scene, providing a unified basis for subsequent fusion and enhancement.

[0045] The brightness response curve can be obtained through pre-calibration. For example, before system deployment, images can be acquired at multiple brightness levels using a standard reference board with known brightness distribution, a grayscale target, or a reference area under uniform illumination conditions, and a correspondence between the camera output value and the reference brightness can be established, thereby forming the camera's brightness response curve. Alternatively, during the device initialization phase, the camera exposure can be varied under fixed illumination conditions to acquire multiple sets of images of the same reference area, thereby fitting the relationship between the camera's imaging output and the brightness response. The brightness recovery can be achieved through table lookup, function mapping, or curve correction, that is, converting the original brightness expression in the environmental image into the recovered brightness expression.

[0046] Environmental images acquired under different exposure conditions refer to multiple scene images obtained by changing the camera's exposure time, gain, or other exposure control parameters under the same detection illumination condition. The purpose of setting different exposure conditions is to ensure that bright and dark areas in the same scene are adequately represented in at least one image. For example, short-exposure, medium-exposure, and long-exposure images can be acquired consecutively for the same scene under the same detection illumination condition. The short-exposure image is used to preserve details in bright areas, the long-exposure image to preserve details in dark areas, and the medium-exposure image to capture overall tonal gradation. In this way, even if both bright reflective areas and darker shadow areas exist simultaneously in the target area, effective information can be obtained from each exposure image.

[0047] Multi-exposure fusion refers to combining environmental images acquired under different exposure conditions to generate a fused image with a wider dynamic range of brightness. The core of multi-exposure fusion lies in not directly using data from a single image for different locations within the target area, but rather selecting or fusing pixel information based on the effectiveness of each exposure image at the corresponding location. Effectiveness here can be understood as whether the pixels at that location exhibit significant overexposure, underexposure, loss of saturation, or a low signal-to-noise ratio. For areas with higher brightness, effective pixel information from shorter exposure images can be prioritized; for areas with lower brightness, effective pixel information from longer exposure images can be prioritized; for areas with transitional brightness, information from intermediate exposure images can be used, or corresponding pixels from multiple images can be weighted and fused. In this way, details in both bright and dark areas of the target area can be preserved more completely, thereby expanding the dynamic range of brightness in the target area.

[0048] It should be noted that before or during multi-exposure fusion, the spatial correspondence of environmental images acquired under different exposure conditions can be ensured. If the image acquisition time interval is short and the scene is basically static, the images can be considered naturally aligned; if there is slight displacement, alignment processing can be used to make the target areas in each image correspond to each other, thus avoiding edge misalignment or local distortion during the fusion process. After multi-exposure fusion, the resulting image reflects the overall brightness distribution of the target area under complex lighting conditions better than a single image.

[0049] After obtaining the fused environmental image, further low-light enhancement, dark area compensation, and local brightness correction are performed. Low-light enhancement refers to boosting areas with low brightness and indistinct details in the fused image, making textures, edges, or color information that were originally submerged in the low-light background clearer. Low-light enhancement is not simply overall brightening, but rather a targeted boost to low-brightness areas to avoid excessive magnification of high-brightness areas. For example, based on pixel brightness distribution, a higher enhancement gain can be applied to low-brightness areas, while lower or no enhancement can be applied to medium- and high-brightness areas, thus achieving a zoned enhancement effect.

[0050] Dark area compensation refers to the process of supplementing the brightness of dark areas in a fused image, caused by occlusion, insufficient local incident light, or uneven brightness distribution in the scene itself. The purpose of dark area compensation is to reduce brightness banding caused by uneven local illumination within the same target area, ensuring that the dark information in the target area is not too dark and affects subsequent image analysis. Dark area compensation can be performed on a regional gain basis for the entire dark area, or it can be performed pixel-by-pixel or region-by-region compensation based on the pixel distribution within the dark area.

[0051] Local brightness correction refers to the further balancing of brightness distribution among different local regions in a fused image after low-light enhancement and dark area compensation, resulting in smoother brightness transitions and more consistent local brightness expression within the target area. The focus of local brightness correction is not on changing the overall brightness trend, but rather on mitigating abrupt changes in local brightness caused by imaging non-uniformity, differences in local enhancement, or differences in the magnitude of local compensation. Through local brightness correction, the target area image can achieve a more uniform and stable brightness expression in space.

[0052] In one specific implementation of this embodiment, the original environmental image output by the camera is first subjected to brightness restoration according to a pre-calibrated brightness response curve to obtain a brightness restored image. Then, short-exposure, medium-exposure, and long-exposure images are acquired under the same detection illumination condition, and these three images are fused according to pixel validity to obtain a fused image with expanded brightness dynamic range. Next, low-light enhancement is performed on the fused image to reveal details in low-brightness areas, compensation is applied to dark areas to improve their discernibility, and local brightness correction is used to reduce brightness abrupt changes between regions. Finally, the target region image for subsequent analysis is determined from the image after the above processing. Here, the target region image is a region image representation that, after the above continuous preprocessing, can be used for subsequent reliable pixel extraction and response analysis. Because this image has undergone brightness restoration, multi-exposure fusion, and enhancement compensation correction processing, it is superior to the original environmental image in terms of brightness representation integrity, dark area visibility, and local brightness consistency, and is therefore more suitable as input for subsequent analysis.

[0053] The above embodiments do not interpret preprocessing as a single image enhancement step, but rather as a continuous process that integrates brightness restoration, multi-exposure fusion, and enhancement compensation correction. First, the brightness response curve is used to eliminate the deviation between the camera output and the actual brightness representation. Then, multi-exposure fusion expands the dynamic range of the target area's brightness. Subsequently, low-light enhancement, dark area compensation, and local brightness correction further improve the local usability of the fused image, ultimately obtaining an image of the target area that can be used in subsequent steps. This provides a more stable and sufficient image foundation for subsequent optical response analysis.

[0054] Although the target area image has undergone preliminary processing, it may still contain pixel information that is detrimental to subsequent analysis. For example, some locations may exhibit localized abnormal bright spots due to surface reflection, while others may receive strong reflection information that deviates from the true optical state of the target area due to specular reflection. Still others may exhibit unstable pixel responses due to noise disturbances under low-light conditions. If these pixels are used together with other pixels for subsequent analysis, the accuracy of the optical characterization of the target area can be easily reduced. To address this, in another embodiment of the present invention, instead of directly using all pixels in the target area image, a reliable subset of pixels suitable for subsequent analysis is extracted from the target area image through a combination of identification, filtering, and retention. Specifically, extracting the reliable subset of pixels from the target area image includes: identifying high-reflection pixels, specular reflection pixels, and low-light noise pixels in the target area; filtering out high-reflection pixels, specular reflection pixels, and low-light noise pixels; and retaining stable pixels among the diffuse reflection feature pixels that respond consistently under different detection illumination conditions as the reliable subset of pixels.

[0055] In this embodiment, high-brightness reflection pixels are pixels in the target area image that exhibit abnormally increased brightness due to strong local reflection. These pixels typically have a brightness value significantly higher than the surrounding area and are prone to forming local bright spots, saturation points, or abrupt edge changes. The image response corresponding to high-brightness reflection pixels often does not accurately reflect the actual lighting state of the target area itself, but rather reflects the abnormal enhancement caused by local reflection. Therefore, these pixels are not suitable for direct participation in subsequent analysis. Specular reflection pixels refer to pixels formed due to the specular reflection characteristics of the target area surface, causing incident light to be concentrated and reflected in a specific direction towards the camera's imaging direction. Compared to general diffuse reflection pixels, specular reflection pixels are more likely to exhibit strong directionality, local concentration, and significant changes in image response characteristics with varying viewing angles. These pixels do not reflect the uniform scattering of light by the target area surface, but rather a strong directional reflection result, and are therefore also unsuitable as effective pixels for characterizing the general optical response of the target area. Low-light noise pixels refer to abnormal pixels caused by sensor noise, quantization errors, or dark current disturbances in low-light areas or areas with weak imaging signals. These types of pixels typically exhibit unstable brightness fluctuations, significant color shifts, discontinuity with surrounding pixels, or a lack of consistency across different acquired images. The information contained in low-light noise pixels is of low reliability, which can affect the stable extraction of image features from the target area.

[0056] Identifying specular reflection pixels, specular reflection pixels, and low-light noise pixels in a target area can be understood as classifying and judging pixels in the target area image according to their imaging performance. This identification is not limited to a single method; any method that can distinguish these three types of undesirable pixels from the target area image is sufficient. For example, specular reflection pixels can be identified based on characteristics such as significantly higher local brightness than their surrounding neighborhood, abrupt changes in local gradients, or a saturation trend. Specular reflection pixels can be identified based on their concentrated brightness, directional reflection, or differences in brightness and color distribution compared to surrounding diffuse reflection areas. Low-light noise pixels can be identified based on their random fluctuations in dark areas, signal discontinuities, or cross-image instability. The purpose of this identification process is to separate pixels that are unsuitable as a basis for judging the true optical response from the target area image.

[0057] After identifying pixels with high light reflectivity, specular reflectivity, and low-light noise, these pixels are filtered out. This filtering process ensures that these pixels are no longer considered as reliable sources for subsequent analysis. Filtering can be implemented by directly removing the identified pixels from the candidate pixel set, or by marking them as invalid pixels during the subsequent selection of valid pixels, thus preventing them from being included in the reliable pixel subset. The core principle is that the responses corresponding to these three types of pixels are no longer included in the subsequent set of valid pixels.

[0058] In this embodiment, after filtering out the aforementioned undesirable pixels, not all remaining pixels are retained. Instead, stable pixels with consistent responses under different detection illumination conditions are further retained from the remaining pixels, focusing on diffuse reflection feature pixels. Diffuse reflection feature pixels, unlike specular reflection pixels, more uniformly reflect the scattering characteristics of the target area surface to incident light. The responses corresponding to diffuse reflection feature pixels typically do not exhibit localized abnormal bright spots or strong directional reflections, but rather are closer to the pixel representation of the general illumination state of the target area surface. Therefore, compared to specular reflection pixels, diffuse reflection feature pixels are more suitable for subsequent analysis.

[0059] Furthermore, among the diffuse reflection feature pixels, it is necessary to screen for stable pixels with consistent responses under different detection illumination conditions. Consistent response here does not mean that the brightness or color values ​​of the corresponding pixels are exactly the same under different detection illumination conditions, but rather that the response changes of the pixel under each detection illumination condition are continuous, reasonable, and comparable, without irregular jumps, sudden abnormal enhancements, or obvious unstable fluctuations. That is, if a diffuse reflection feature pixel exhibits a response change consistent with the illumination change pattern under the first detection illumination condition, the second detection illumination condition, and other detection illumination conditions, it can be considered a stable pixel with consistent response; conversely, if a pixel, although not a specular or specular reflection pixel, still exhibits abnormal fluctuations under different detection illumination conditions, it will not be retained as a reliable pixel.

[0060] In practical implementation, the brightness, color, or overall color response of pixels at the same location under different detection illumination conditions can be compared. For example, for a candidate diffuse reflection feature pixel in the target area, its imaging value can be read under the first detection illumination condition, the second detection illumination condition, and other detection illumination conditions. If the response change trend of the pixel is consistent with the change of the detection illumination condition and there is no abnormal deviation from the overall pixel change pattern, it is identified as a stable pixel. If the pixel suddenly shows abnormal brightness, abnormal color cast, or significantly deviates from the change trend of surrounding similar pixels under a certain detection illumination condition, it will not be retained.

[0061] Therefore, the reliable pixel subset in this embodiment is not simply the set of remaining pixels after removing a small number of outliers, but rather an effective pixel set formed after two layers of filtering: the first layer identifies and filters out pixels with high light reflection, specular reflection, and low-light noise; the second layer further retains stable pixels with diffuse reflection characteristics and consistent responses under different detection illumination conditions from the remaining pixels. After the above two layers of processing, the final reliable pixel subset can more realistically represent the effective optical response of the target area.

[0062] For example, in an indoor desktop scene, there may be white paper, a glass, and darker shadow areas on the table. High-brightness and specular-brightness pixels are likely to form near the glass, while low-light noise pixels may appear in the shadow areas of the table, and most areas of the paper surface are more likely to exhibit a relatively stable diffuse response. In this case, by identifying and filtering out high-brightness and specular-brightness pixels near the glass, filtering out low-light noise pixels in the shadow areas, and then retaining stable pixels with consistent response changes under different detection illumination conditions from the paper surface and other diffuse areas, a reliable subset of pixels can be formed for subsequent analysis.

[0063] Through the above processing, this embodiment can exclude pixels in the target area image that are significantly affected by reflection interference and low-light noise, and further filter out stable and representative effective pixels from the remaining pixels. The resulting reliable pixel subset is more suitable as a basic expression of the effective optical response of the target area, providing more reliable pixel input for subsequent analysis.

[0064] After obtaining the reliable pixel subset, the reflectance characteristics are not directly output from this pixel set. Instead, the reliable pixel subset is first converted into a feature representation that can be compared, combined, and matched. Then, based on this feature representation, a set of samples that are closer to the current target region is selected from the reference samples. Finally, this set of samples is used to jointly participate in the calculation of reflectance characteristics. In other words, this embodiment does not simply rely on a single training sample to determine the reflectance attributes of the current target region. Instead, it first forms a multi-state response representation, then finds several reference objects in the sample space that are closer to this representation, and assigns different influence weights according to the degree of proximity, so that the obtained reflectance characteristics are closer to the actual situation of the current target region. Therefore, in another embodiment provided by the present invention, the calculation of the reflection characteristic characterization parameters of the target area specifically includes: extracting the response features of the target area based on the reliable pixel subset under each detection illumination state, and combining the response features extracted under each detection illumination state to form a multi-source response vector; calculating the similarity between the multi-source response vector and the response features of each training sample in the preset sample library, and selecting at least two training samples whose similarity meets the preset conditions to form a local optimal sample subset; assigning corresponding weights to the training samples in the local optimal sample subset according to the similarity of each training sample, and calculating the reflection characteristic characterization parameters based on the weighted local optimal sample subset.

[0065] In this embodiment, the response feature refers to the feature information extracted from the trusted pixel subset corresponding to the target area under each detection illumination state, which can characterize the optical response characteristics of the target area. This response feature is not limited to a single form and can be a brightness response feature, color response feature, RGB response feature, pixel statistical feature, or a combination of two or more of the above features. Specifically, the brightness response feature is used to characterize the brightness and darkness response of the target area under the corresponding detection illumination state, such as the average brightness, brightness distribution range, brightness change gradient, or brightness statistics of the trusted pixel subset; the color response feature is used to characterize the comprehensive color response of the target area under the corresponding detection illumination state, such as the comprehensive color value, color shift, or comprehensive color distribution feature; the RGB response feature is used to characterize the response of the trusted pixel subset in the red, green, and blue channels, such as the RGB three-channel mean, RGB channel ratio, or RGB channel joint distribution; the pixel statistical feature can be used to summarize the overall statistical state of the trusted pixel set within the target area, such as the mean, variance, skewness, kurtosis, or interval distribution. Any feature that can reflect the response performance of the target area under the corresponding detection illumination state can be used as the response feature in this embodiment.

[0066] Extracting response features of a target region based on a subset of trusted pixels under different illumination states refers to extracting features from each set of trusted pixels formed by the same target region under different illumination states. Since the trusted pixel representation of the same target region differs under different illumination states, the response features extracted under each illumination state have state correspondence. For example, if the system has a first illumination state and a second illumination state, the first response feature can be extracted from the subset of trusted pixels corresponding to the first illumination state, and the second response feature can be extracted from the subset of trusted pixels corresponding to the second illumination state. If three or more illumination states are set, the corresponding features can be extracted one by one according to each state.

[0067] In this embodiment, combining the response features extracted under various detection illumination states to form a multi-source response vector means concatenating, arranging, or jointly encoding the response features extracted from the same target area under multiple detection illumination states in a predetermined order to form a unified vectorized representation. This multi-source response vector is used to comprehensively characterize the overall response pattern of the target area under multiple detection illumination inputs. The combination is not limited to simple addition, but only needs to maintain the correspondence and overall expressive power of the response features under each detection illumination state. For example, in one implementation, the brightness response features and color response features under the first detection illumination state can be concatenated with the brightness response features and color response features under the second detection illumination state in sequence to form a joint feature vector; in another implementation, RGB response sub-vectors under each state can be formed first, and then the sub-vectors can be combined sequentially into a total vector. The resulting multi-source response vector is no longer a local description under a single illumination state, but an overall expression of the response performance of the target area under multiple illumination conditions.

[0068] The preset sample library is a pre-established set of reference samples, which stores multiple training samples and their corresponding response features. These training samples can be understood as previously acquired and organized reference region samples, corresponding to different optical response modes. The purpose of the sample library is to provide comparison objects for the current target region, enabling the multi-source response vector formed by the current target region to be compared with existing samples for similarity.

[0069] Calculating the similarity between the multi-source response vector and the response features of each training sample in a pre-defined sample library involves comparing the multi-source response vector of the current target region with the response features corresponding to each training sample in the sample library to determine the degree of similarity in their response performance. Similarity represents the closeness between the current target region and the training samples; the higher the similarity, the closer the response performance of the training samples is to the current target region. Similarity can be calculated using distance or correlation metrics, as long as they can distinguish the closeness between the training samples and the current target region. For example, similarity can be obtained by comparing the degree of difference between two vectors, or by comparing the degree of consistency between two vectors.

[0070] After similarity calculation, at least two training samples that meet preset similarity criteria are selected to form a locally optimal sample subset. This locally optimal sample subset is not all samples in the sample library, but rather a group of training samples that are closer to the current target region and to the current multi-source response vector. The reason for selecting at least two training samples, rather than just one, is to avoid subsequent reflectance characteristic calculations relying excessively on a single sample, which could lead to accidental result bias. The preset criteria can be understood as the judgment conditions used to select samples, such as similarity exceeding a preset threshold, similarity ranking within a certain range, or meeting a preset sorting range. The goal is to select several training samples from the sample library that are closer to the current target region.

[0071] In this embodiment, after selecting a locally optimal sample subset, corresponding weights are assigned to the training samples in the locally optimal sample subset based on the similarity of each training sample. These weights reflect the degree of contribution of different training samples to the calculation of the reflectance characteristics of the current target region. Generally, training samples with higher similarity have larger weights; training samples with relatively low similarity but still meeting preset conditions have relatively smaller weights. This approach allows the subsequent calculation of reflectance characteristic parameters to be more influenced by more similar samples, rather than treating all samples equally. For example, in one implementation, the similarity values ​​of each training sample in the locally optimal sample subset can be directly converted into weights; in another implementation, the similarity can be normalized first, and then the normalized result can be used as the corresponding weight. Regardless of the method used, the core is to establish a weighting relationship where the more similar the samples, the higher their participation.

[0072] Calculating reflectance characteristic parameters based on a weighted local optimal sample subset refers to obtaining the reflectance characteristic parameters of the current target region by combining the reference role of each training sample in the reflectance characteristics of the current target region after assigning different weights to the local optimal sample subset. Using a weighted local optimal sample subset means that the reflectance characteristic parameters are not directly determined by the multi-source response vector of the current target region, nor are they simply the ready-made result of a single most similar sample. Instead, they are calculated by combining several similar samples and their weights. This ensures that the calculation result reflects the proximity relationship between the current target region and the closest sample, while avoiding the amplification of bias caused by a single sample. For example, if luminance response features and RGB response features are extracted from the current target area under three different detection illumination states, the features obtained under the three states can be combined sequentially into a multi-source response vector. Then, this vector is compared with the response features of each training sample in the sample library, and the three training samples with the highest similarity ranking and meeting the preset conditions are selected as the local optimal sample subset. Then, different weights are assigned to these three training samples according to the similarity between them and the current target area. Finally, the reflection information corresponding to these three training samples is comprehensively calculated based on the weighting results to obtain the reflection characteristic characterization parameters of the current target area.

[0073] This embodiment effectively transforms a set of reliable pixels into reflectance characteristic parameters. First, response features reflecting the target region's response are extracted from the subset of reliable pixels under various illumination conditions. Second, multi-source response vectors are formed by combining response features from different illumination conditions, giving the target region's multi-state responses a unified expression. Third, a locally optimal sample subset is selected from a pre-defined sample library through similarity comparison, and training samples within this subset are assigned weights corresponding to their similarity. Finally, reflectance characteristic parameters are calculated based on the weighted locally optimal sample subset. Thus, the reflectance characteristic calculation results for the current target region are based on multi-state response comparisons and local sample references, improving the result's relevance to the actual target region.

[0074] The image response of a target area under different illumination conditions does not necessarily correspond directly to the actual illumination state of that area. This is because the image performance of the target area is affected not only by the incident illumination but also by factors such as surface material, surface roughness, specular component, diffuse reflection component, and overall color shift. Therefore, in order to express the reflection behavior of the target area in a calculable and retrievable form, this embodiment summarizes the reflection information of the target area into a set of reflection characteristic characterization parameters. This set of parameters can be used to summarize the overall reflection pattern of the target area, describe the strength relationship of different reflection components, and further serve as the basis for subsequent illuminance and color temperature correction. Based on this, in this embodiment, the reflection characteristic characterization parameters include at least one of the following: a regional reflectivity characterization parameter for characterizing the surface reflection characteristics of the target area, a classification parameter for characterizing the reflection type of the target area, a specular reflection intensity parameter for characterizing the specular reflection degree of the target area, a diffuse reflection proportion parameter for characterizing the diffuse reflection contribution of the target area, a reflection compensation coefficient for correcting the illuminance estimation results, and a color temperature compensation coefficient for correcting the color temperature estimation results.

[0075] In this embodiment, the reflection characteristic characterization parameter is not a single numerical value, but a set of parameters used to describe the reflection behavior of the target area. In specific implementations, only one of the reflection characteristic characterization parameters may be used, or two, three, or more parameters may be used in combination; it is not required that all parameters exist simultaneously. That is, in different application scenarios, one or more suitable parameters can be selected to constitute the characterization result of the reflection characteristics of the target area, based on the reflection complexity of the target area, computational requirements, or system configuration.

[0076] Regional reflectivity characterization parameters are used to characterize the overall reflectivity of a target region's surface. This parameter focuses on the overall reflectivity of the target region to incident light, providing a generalized expression of the surface reflectivity level. The characterization parameter does not necessarily need to be an absolute reflectivity curve in the full spectral sense; it can also be a parameter representation of the target region's surface reflectivity after normalization, compression, or statistical analysis. In other words, anything that reflects the overall reflectivity level of the target region can be used as a regional reflectivity characterization parameter. For example, in one implementation, the comprehensive color response change of the target region under various detection illumination conditions can be converted into a comprehensive reflectivity parameter to indicate whether the region is more of a high-reflectivity or low-reflectivity surface. In another implementation, a set of reflectivity-related statistics can be extracted and used as a regional reflectivity characterization parameter.

[0077] The reflection type classification parameter is used to characterize the overall reflection pattern classification of the target area. Reflection type does not simply indicate intensity, but rather which type of reflection pattern the target area is closer to. That is, this parameter answers what type of reflection the area belongs to, not how strong the reflection is. In this embodiment, the reflection type classification parameter can be used to distinguish whether the target area belongs to diffuse-dominant, specular-dominant, mixed-reflection, or other preset reflection types. For example, paper, wall, and fabric surfaces are generally closer to diffuse-dominant; glass, polished metal surfaces, and display screen surfaces may be closer to specular-dominant; while some plastic tabletops or semi-gloss coated surfaces may exhibit mixed-reflection. Using this classification parameter, the overall reflection pattern can be used to determine which type of reflection behavior the target area belongs to.

[0078] Specular reflection intensity parameters characterize the strength of specular reflection components in a target area. The emphasis here is on the strength of the specular components, i.e., the relative intensity of directional reflection components in the target area, rather than classifying the overall reflection pattern. If an area exhibits significant specular reflection, its specular reflection intensity parameter will be relatively large; if an area mainly exhibits uniform scattering and lacks strong directional reflection, its specular reflection intensity parameter will be relatively small. Specular reflection intensity parameters can be determined by the concentration of bright spots in the target area, the proportion of locally bright areas, the degree of brightness deviation in strongly reflective areas, or the directional reflection performance. For example, under similar lighting conditions, the surface of a glass often shows sharp and concentrated bright areas, and the specular reflection intensity parameter corresponding to these areas is usually high; while the surface of ordinary paper, although exhibiting brightness variations, typically does not show strongly directional bright spots, and its specular reflection intensity parameter is usually low.

[0079] The diffuse reflection proportion parameter characterizes the contribution of the diffuse reflection component in a target area. It emphasizes the proportion of diffuse reflection in the overall reflection, rather than simply indicating the presence of diffuse reflection in the area. In other words, the reflection behavior of a target area is often not composed of a single component, but may simultaneously include specular and diffuse reflection. The diffuse reflection proportion parameter describes the percentage or contribution of the diffuse reflection portion. For example, in rougher surfaces like walls, fabric, or paper, diffuse reflection typically constitutes the majority, resulting in a higher diffuse reflection proportion parameter. In high-gloss glass or polished metal surfaces, while some diffuse reflection may exist, its proportion is relatively low, leading to a smaller parameter. This parameter can further explain why a target area exhibits a relatively smooth and uniform response under different detection lighting conditions, or why a superimposed strong reflection effect occurs.

[0080] In this embodiment, the reflection compensation coefficient is used to correct the illumination estimation result. This parameter functions as follows: when the image brightness of the target area is determined not only by the actual illumination state but also by the intensity of surface reflection, the reflection compensation coefficient is used to correct the illumination estimation result, thereby reducing the deviation introduced by surface reflection characteristics. For example, for high-reflectivity areas, directly inferring illumination based on image brightness may overestimate the actual illumination level of the target area; this can be corrected using the reflection compensation coefficient. Similarly, for low-reflectivity areas, if the image appears dark, it may underestimate the actual illumination; this can also be compensated for using this coefficient. In other words, the reflection compensation coefficient is a parameter that corrects the difference between how bright the image appears and the actual intensity of illumination received by the area.

[0081] The color temperature compensation coefficient is used to correct the color temperature estimation results. This parameter's function is that the color representation of the target area image is affected not only by the color temperature of the illumination itself, but also by factors such as surface material and the overall color reflectance characteristics of the surface. Therefore, the color shift observed in the image may not be entirely equivalent to the color temperature change of the actual lighting environment. The color temperature compensation coefficient is used to correct this color temperature deviation caused by reflectance characteristics. For example, some warm-colored surfaces may still exhibit a warm image response under neutral color temperature illumination, while some cool-colored, highly reflective surfaces may exhibit a cool response under the same illumination conditions. In this case, the color temperature compensation coefficient can be used to reduce the offset caused by surface reflectance characteristics on the color temperature estimation, making the color temperature estimation results closer to the actual illumination state of the target area.

[0082] In this embodiment, the various parameters are not redundant but rather characterize the reflective behavior of the target area from different perspectives. The regional reflectivity parameter is biased towards a generalization of overall reflectivity; the reflection type classification parameter is biased towards a classification of overall reflection patterns; the specular reflection intensity parameter is used to represent the strength of the specular component; the diffuse reflection proportion parameter is used to represent the contribution of the diffuse reflection component; and the reflection compensation coefficient and color temperature compensation coefficient are further geared towards the subsequent correction process, used to map the reflective characteristics to the correction of illuminance and color temperature estimation results. Therefore, in specific implementations, only one parameter can be selected for characterization, or multiple parameters can be combined as needed. For example, in scenarios with simpler reflection relationships, only the regional reflectivity parameter and the reflection compensation coefficient are needed; while in scenarios with more complex reflection relationships, the reflection type classification parameter, specular reflection intensity parameter, and diffuse reflection proportion parameter can be further introduced to characterize the reflective characteristics of the target area from multiple angles.

[0083] In an indoor office setting, a white wall might exhibit dominant diffuse reflection, with a high diffuse reflection ratio and a low specular reflection intensity. A glass partition might exhibit dominant specular reflection, with a high specular reflection intensity and a low diffuse reflection ratio. A wooden tabletop might exhibit mixed reflectivity, possessing both a certain regional reflectivity characteristic and a moderate specular reflection component. In such a scenario, corresponding reflectivity characteristic parameters can be obtained for each target area, and these parameters can then be used to correct subsequent illuminance and color temperature estimation results.

[0084] The image information of the target area cannot be directly equated with the final dimming target. This is because the brightness and color representation of the same target area in the camera image are affected by both the current lighting conditions and the surface reflection properties of the target area. Therefore, in another embodiment provided by this invention, brightness response features and color response features reflecting the current image response state are first extracted from a subset of trusted pixels to form illuminance estimation results and color temperature estimation results, respectively. Subsequently, the estimation results are corrected by combining the reflection characteristic characterization parameters of the target area. After obtaining the corrected lighting environment state, the target brightness and target color temperature for subsequent dimming control are further determined according to preset brightness mapping relationships and preset color temperature mapping relationships. The specific technical solution adopted is as follows: Brightness response features and color response features are extracted from a subset of reliable pixels in the target area under various detection illumination conditions. Illuminance estimation results are generated based on the brightness response features, and color temperature estimation results are generated based on the color response features. The reflectance characteristic parameters of the target area are matched with preset correction rules to determine illuminance correction parameters and color temperature correction parameters, respectively. The illuminance correction parameters are used to perform weighted correction or bias compensation on the illuminance estimation results to obtain corrected illuminance, and the target illuminance is determined from the corrected illuminance according to a preset illuminance mapping relationship. The color temperature correction parameters are used to perform weighted correction or bias compensation on the color temperature estimation results to obtain corrected color temperature, and the target color temperature is determined from the corrected color temperature according to a preset color temperature mapping relationship.

[0085] In this embodiment, the brightness response feature is the input basis for generating the illuminance estimation result. That is, the illuminance estimation result is not given out of thin air, but is derived from the brightness-related information exhibited by a subset of reliable pixels in the target area under various detection illumination states. The brightness response feature can be a feature quantity that reflects the brightness state of the target area, such as the average brightness value of the reliable pixel subset, the brightness distribution range, the brightness uniformity index, the brightness gradient statistics, the brightness variation range under different detection illumination states, or a combination of the above quantities. For example, in one implementation, the average brightness of reliable pixels in the target area under the first and second detection illumination states can be statistically analyzed, and the change between the two can be calculated. This, combined with the brightness distribution range within the target area, forms the brightness response feature. Therefore, the brightness response feature essentially characterizes the brightness response pattern of the target area under various detection illumination inputs.

[0086] The color response features serve as the input for generating the color temperature estimation results. The color temperature estimation results are derived from the color response performance of a subset of trusted pixels in the target area, rather than relying solely on a single color value. Color response features can be comprehensive color distribution features, comprehensive color shift features, RGB channel ratios, comprehensive color coordinate parameters, or other features that reflect the color response trend of the target area. For example, in one implementation, the mean and channel ratio of the trusted pixel subset in the red, green, and blue channels can be extracted separately, or comprehensive color coordinates can be extracted and the color shift direction and magnitude under different detection illumination conditions can be compared to form color response features. Therefore, the color response features express the comprehensive color response pattern exhibited by the target area under the current acquisition conditions.

[0087] Based on the brightness and darkness of the target area as reflected by the brightness response features, an estimate or quantity of the current illuminance level of the target area is generated. This illuminance estimation result represents the system's judgment of the current light intensity received by the target area. This result can be expressed numerically, as well as as an interval quantity or a graded quantity, as long as it can characterize the current illuminance state of the target area. For example, by using a pre-established correspondence between brightness response features and illuminance, the current illuminance estimate of the target area can be obtained by looking up a table, substituting into a function, or calling an estimation model from the input brightness response features. Similarly, based on the comprehensive color change of the target area as reflected by the color response features, an estimate or quantity of the current color temperature state of the target area is generated. This color temperature estimate represents the system's judgment of the current warm or cool properties of the light environment in the target area. For example, by using a pre-established correspondence between color response features and comprehensive color temperature, the color temperature estimate can be obtained by mapping the color response features. Therefore, brightness response features and color response features correspond to the basis for generating illuminance estimation results and color temperature estimation results, respectively. In this embodiment, both are preliminary input bases, not subsequent additional descriptions.

[0088] In this embodiment, after obtaining the illuminance estimation result and the color temperature estimation result, it is necessary to match the reflectance characteristic parameters of the target area with preset correction rules to determine the illuminance correction parameters and color temperature correction parameters respectively. The preset correction rules refer to a pre-established rule system used to describe the correspondence between reflectance characteristics and correction methods. These preset correction rules can be implemented using rule sets, parameter correspondence tables, compensation models, functional relationships, or any equivalent form thereof. For example, in one implementation, it can be pre-set that: when the specular reflection intensity parameter of the target area is higher than a certain threshold, a stronger suppression correction is applied to the illuminance estimation result; when the diffuse reflection ratio parameter of the target area is high, a weaker correction or maintenance of the original estimate is applied to the illuminance estimation result; when the comprehensive color reflectance characteristics of the target area surface have a significant shift in the comprehensive color temperature, a color temperature correction parameter with corresponding direction and amplitude is introduced to the color temperature estimation result. Through this matching method, different correction strategies can be applied to target areas with different reflectance characteristics, rather than uniformly correcting all areas. The illuminance correction parameters are parameters used to correct the illuminance estimation results. It can be expressed as a weighting coefficient, compensation bias, scaling factor, or other corrections that can alter the illuminance estimation result. The color temperature correction parameter is used to correct the color temperature estimation result; it can also be expressed as a weighting coefficient, compensation bias, scaling factor, or other corrections that can alter the color temperature estimation result. For example, when the target area is of a high reflectance type, the illuminance correction parameter can correct the original illuminance estimation result to a lower level; when the overall color reflectance of the target area tends to be warm, the color temperature correction parameter can correct the original color temperature estimation result to a higher level to offset the influence of the overall surface color deviation.

[0089] Corrected illuminance is obtained by applying weighted corrections or bias compensation to the illuminance estimation results using illuminance correction parameters. This involves adjusting the original illuminance estimation result based on predetermined illuminance correction parameters to make it closer to the true lighting conditions of the target area. Weighted correction involves changing the original illuminance estimation result through multiplicative or proportional methods; bias compensation involves correcting the original illuminance estimation result through incremental or subtractive methods. These two methods can be used individually or in combination. For example, if a target area appears too bright in an image due to high surface reflectivity, the illuminance estimation result can be multiplied by a weighting coefficient less than 1, or a certain compensation amount can be subtracted from the original estimate to obtain the corrected illuminance.

[0090] After obtaining the corrected illuminance, the target luminance is determined based on a preset luminance mapping relationship. This preset luminance mapping relationship refers to a pre-established correspondence used to convert the corrected illuminance into the target luminance. This mapping relationship can be at least one of the following: a rule, a table, a model, or a function. For example, in one implementation, the preset luminance mapping relationship can be an illuminance-luminance lookup table: when the corrected illuminance is within a certain range, a corresponding target luminance range is output. In another implementation, the preset luminance mapping relationship can be a function model, where the target luminance is obtained through function calculation based on the corrected illuminance. A rule-based model can also be used, for example, increasing the target luminance when the corrected illuminance is below a certain threshold, and decreasing the target luminance when the corrected illuminance is above a certain threshold. Therefore, the target luminance is no longer a simple restatement of the current luminance state, but rather a lighting output target determined based on the corrected illuminance state.

[0091] Accordingly, the color temperature estimation result is weighted or offset using color temperature correction parameters to obtain the corrected color temperature. This involves adjusting the original color temperature estimation result based on the determined color temperature correction parameters to make it closer to the overall color temperature state of the actual lighting environment of the target area. For example, if the overall color reflection characteristics of a target area's surface cause the image to appear warm, the color temperature estimation result can be compensated using color temperature correction parameters to improve the overall color temperature estimate; conversely, if a target area's overall color is cool, causing the image to appear cool, the color temperature estimation result can be corrected in the opposite direction. After this step, the obtained corrected color temperature better reflects the actual warm or cool lighting state of the target area than the original color temperature estimation result.

[0092] After obtaining the corrected color temperature, the target color temperature is determined based on a preset color temperature mapping relationship. This preset color temperature mapping relationship is also a pre-established correspondence, which can take the form of rules, tables, models, or functions. For example, in one implementation, a color temperature mapping table can be used to map different ranges of corrected color temperatures to different ranges of target color temperatures; in another implementation, a mapping function can be used to directly convert the corrected color temperature into a suitable target color temperature; rules can also be used, such as maintaining or slightly increasing the target color temperature when the corrected color temperature is too low, and decreasing the target color temperature when the corrected color temperature is too high. In this way, the determination of the target color temperature is based on the corrected overall color state, rather than directly using the uncorrected image color representation.

[0093] In one example of this embodiment, the average brightness, brightness variation amplitude, RGB channel mean, and comprehensive color shift can be extracted from a subset of reliable pixels in the target area under various detection illumination conditions to form brightness response features and color response features, respectively. Then, based on a pre-established estimation model, the illuminance estimation result and color temperature estimation result of the current target area are obtained. Subsequently, the reflection characteristic characterization parameters corresponding to the current target area are input into a preset correction rule to output illuminance correction parameters and color temperature correction parameters, respectively. Then, the illuminance correction parameters are used to proportionally correct the illuminance estimation result to obtain the corrected illuminance, and the target brightness is obtained through a illuminance mapping table. At the same time, the color temperature correction parameters are used to offset the color temperature estimation result to obtain the corrected color temperature, and the target color temperature is obtained through a color temperature mapping function.

[0094] After obtaining the target brightness and target color temperature, instead of directly converting them into driving commands, a lightweight neural network model is introduced to perform joint mapping processing on the two to generate dimming parameters corresponding to the LED light source adjustment. The reason for this approach is that while target brightness and target color temperature can characterize the desired lighting output state, the relationship between them and the actual dimming parameters is not always a one-to-one linear one. There may also be joint coupling effects between different brightness targets and different color temperature targets. Therefore, this embodiment uses a neural network model with a small parameter scale, simple structure, and short computation chain to achieve the joint mapping from target brightness and target color temperature to dimming parameters while ensuring computational efficiency. Specifically, the following technical solution is adopted: the target brightness and target color temperature are normalized and then input into the lightweight neural network model. The lightweight neural network model includes a sequentially connected input layer, a low-dimensional feature mapping layer, and a parameter output layer. The low-dimensional feature mapping layer is used to perform joint feature extraction and nonlinear mapping on the normalized target brightness and target color temperature. The parameter output layer is used to output the dimming parameters corresponding to the LED light source brightness and color temperature adjustment.

[0095] In this embodiment, the target brightness and target color temperature represent the desired brightness and color temperature values ​​for the target region, respectively. Since their numerical ranges, dimensional characteristics, and variation intervals are not entirely consistent, to facilitate unified processing in subsequent models, the target brightness and target color temperature are normalized before being input into the lightweight neural network model. Normalization refers to converting the original target brightness and original target color temperature to a preset numerical range, making them comparable and jointly mappable inputs. For example, the target brightness can be mapped to 0 to 1 according to preset upper and lower brightness limits, or the target color temperature can be mapped to 0 to 1 according to a preset color temperature range; alternatively, they can be mapped to other preset ranges, as long as the differences caused by different dimensions and numerical ranges are eliminated, making them suitable as inputs to a unified model. After normalization, the resulting input no longer directly represents the original physical quantity, but rather a standardized input suitable for network computation.

[0096] In this embodiment, the lightweight neural network model refers to a neural network model with controlled parameter size, simplified hierarchical structure, low computational complexity, and suitability for execution in lighting control terminals or edge devices. Lightweight here does not mean that only one type of fixed network can be used, but rather emphasizes that the model, in achieving the mapping from target brightness and target color temperature to dimming parameters, does not rely on large-scale deep networks, but uses fewer layers, fewer parameters, and shorter computational paths to complete the mapping process. Therefore, it balances model expressiveness and deployment efficiency, enabling the model to complete the joint calculation from target values ​​to dimming parameters without placing an excessive computational burden on local hardware. The model adopts a hierarchical structure that sequentially passes data in a forward-looking order; that is, input data first enters the input layer, then is sent to the low-dimensional feature mapping layer, and finally outputs from the low-dimensional feature mapping layer to the parameter output layer. The entire processing chain is completed sequentially in a unidirectional flow. This structure does not require the introduction of deeper, more complex stacks, nor does it require the use of loop structures or multi-branch loops, thus ensuring a simple and clear model structure. In one implementation, the model can consist of three functional layers: an input layer that receives the normalized target brightness and target color temperature, a low-dimensional feature mapping layer that performs intermediate feature transformations, and a parameter output layer that generates the final dimming parameters. In another implementation, even if each functional layer contains several neurons or parallel computing units, as long as the overall structure still exhibits the sequential transmission relationship of input layer—low-dimensional feature mapping layer—parameter output layer, it still belongs to the sequential connection structure described in this embodiment.

[0097] In this embodiment, the input layer receives the normalized target brightness and target color temperature. The input layer itself does not perform complex calculations; its main function is to receive, organize, and transmit the input data. Since the input objects in this embodiment are target brightness and target color temperature, the input layer can use two input units to receive these two input quantities respectively, or it can use an equivalent form to input them as a group. The function of the input layer is to provide standardized input data for the subsequent low-dimensional feature mapping layer.

[0098] The low-dimensional feature mapping layer is the core processing layer in this embodiment used to achieve the joint mapping of target brightness and target color temperature. Low-dimensionality does not mean compressing the input dimension to a fixed value, but rather emphasizes that this layer uses intermediate feature representations of a controlled scale within the network. It avoids complex calculations through too many intermediate layers or excessively high-dimensional expansion, instead completing joint feature extraction and mapping within a relatively concise feature space. This joint feature extraction means that this layer does not process target brightness and target color temperature independently, but treats them as a whole input, extracting the correlation features between them in a unified feature space. In other words, this layer can reflect the joint relationship of how the corresponding dimming parameters should change when a certain target brightness and a certain target color temperature are combined, rather than completely separating brightness and color temperature.

[0099] The nonlinear mapping refers to the layer's processing not being limited to simple linear transformations, but allowing for a nonlinear expression of the relationship between the input target brightness and target color temperature through activation functions or other nonlinear calculation methods. Its significance lies in the fact that the relationship between target brightness and target color temperature and dimming parameters may not be a simple proportional one; nonlinear mapping can improve the model's ability to fit such relationships.

[0100] To provide more adequate support for the low-dimensional feature mapping layer, this embodiment can be further understood as follows: After the input layer receives the normalized target brightness and target color temperature, the low-dimensional feature mapping layer uses a controlled number of hidden nodes or feature units to perform weighted combination and nonlinear transformation on the input data, forming a set of intermediate mapping features. This set of intermediate mapping features is not the final dimming parameters, but rather an intermediate result after compressing and expressing the combined relationship between the brightness target and the color temperature target. For example, in one implementation, this layer can use a hidden layer structure with fewer nodes than a preset upper limit to maintain a lightweight model; in another implementation, this layer can also use a set of low-dimensional feature units to form a small number of intermediate feature values ​​by weighted superposition and nonlinear processing of the input variables. The common point of the above methods is that they do not rely on deep, complex networks, but complete the joint mapping through a controlled-scale intermediate feature space.

[0101] The parameter output layer is used to convert the intermediate mapping features output by the low-dimensional feature mapping layer into dimming parameters corresponding to the brightness and color temperature adjustment of the LED light source. The output does not refer to any arbitrary result, but rather to a parameter form that can be directly used in subsequent dimming execution. In this embodiment, the dimming parameters can be expressed as brightness adjustment parameters and color temperature adjustment parameters, or as a set of parameters corresponding to the brightness driving state and color temperature driving state. As long as the output parameter can be called by the subsequent dimming execution process to control the brightness and color temperature changes of the LED light source, it belongs to the dimming parameters in this embodiment. For example, in one implementation, the parameter output layer can output a parameter value corresponding to brightness adjustment and a parameter value corresponding to color temperature adjustment respectively; in another implementation, it can also output a set of parameters, where some parameters are used for brightness adjustment and others for color temperature adjustment. The function of the parameter output layer is to complete the transformation from intermediate mapping features to final control parameters.

[0102] In one specific processing step of this embodiment, the target brightness and target color temperature are first normalized according to a preset range to form two normalized input quantities. These two normalized input quantities are then input to the input layer, and passed to the low-dimensional feature mapping layer. The low-dimensional feature mapping layer performs joint feature extraction and nonlinear mapping on the two, outputting a small number of intermediate mapping features. The parameter output layer then performs a linear or nonlinear transformation on these intermediate mapping features, outputting dimming parameters corresponding to the LED light source brightness and color temperature adjustments. This completes the process of generating dimming parameters from target brightness and target color temperature.

[0103] For example, in an indoor lighting scenario, after preliminary processing, the target brightness has been determined to be one target value, and the target color temperature to be another. At this point, both values ​​can be normalized separately before being fed into a lightweight neural network model consisting of an input layer, a low-dimensional feature mapping layer, and a parameter output layer. The low-dimensional feature mapping layer extracts intermediate joint mapping features based on this brightness-color temperature target combination, and the parameter output layer further outputs a set of dimming parameters corresponding to brightness and color temperature adjustments. Because this model uses a structure with fewer layers and controlled parameters, it can quickly complete calculations within local lighting control devices.

[0104] Once the target brightness, target color temperature, and corresponding dimming parameters are determined, the dimming process is not equivalent to directly writing the parameters into the LED driver at once. Instead, it requires a conversion and update process directed to the execution layer. This is because the brightness and color temperature outputs of LED light sources typically exhibit dynamic changes. If switched instantaneously based on the final values, it can easily cause sudden changes in brightness, color, or localized visual discomfort. Therefore, in another embodiment of this invention, before the dimming parameters are actually applied to the LED light source, they are first decomposed into brightness adjustment and color temperature adjustment amounts. Then, the amplitude of each change is constrained through limiting processing, and the driving state of the LED light source is updated step-by-step according to a preset time window, making the entire dimming process executable and gradual. Specifically, the following technical solution is adopted: determine the brightness adjustment amount and color temperature adjustment amount according to the dimming parameters; perform limiting processing on the brightness adjustment amount and color temperature adjustment amount; and gradually update the brightness driving state and color temperature driving state of the LED light source according to a preset time window.

[0105] In this embodiment, the dimming parameters are a set of parameters generated in the preprocessing stage that can be used to control changes in the output of the LED light source. These dimming parameters are not directly equivalent to the final output state of the LED light source, but rather serve as input instructions for subsequent adjustment processes. The dimming parameters may include parameter information characterizing the trend of brightness change, parameter information characterizing the trend of color temperature change, or a combination of both.

[0106] Based on the dimming parameters, the brightness adjustment amount and color temperature adjustment amount are determined separately. The dimming parameters can be decomposed into two execution directions: one part controls the brightness change of the LED light source, forming the brightness adjustment amount; the other part controls the color temperature change of the LED light source, forming the color temperature adjustment amount. The brightness adjustment amount refers to the change in the LED light source's brightness output relative to the current brightness driving state within the current adjustment cycle; the color temperature adjustment amount refers to the change in the LED light source's color temperature output relative to the current color temperature driving state within the current adjustment cycle. Neither of these quantities is an absolute target value, but rather an incremental or variable expression directed towards the execution layer.

[0107] In one implementation, the brightness-related portion of the dimming parameters can be converted into a brightness increment value, and the color temperature-related portion of the dimming parameters can be converted into a color temperature increment value. For example, if the current LED light source brightness driving state corresponds to a certain reference output, and the dimming parameters indicate that the brightness needs to be increased, then the brightness adjustment amount can be determined as a positive increment; if the dimming parameters indicate that the brightness needs to be decreased, then the brightness adjustment amount can be determined as a negative increment. Similarly, if the current color temperature driving state needs to be adjusted towards a higher color temperature, then the color temperature adjustment amount can be expressed as a positive change; if it needs to be adjusted towards a lower color temperature, then the color temperature adjustment amount can be expressed as a negative change.

[0108] In this embodiment, the limiting process refers to restricting the range of change in brightness and color temperature adjustment, ensuring that the change in each update does not exceed a preset allowable range. The purpose of this limiting process is to prevent excessive jumps in the LED light source output state caused by the dimming parameters during a single execution. This limiting can be understood as setting maximum allowable change values ​​for both brightness and color temperature adjustment, or as setting upper and lower limits for each update. For example, when the calculated brightness adjustment exceeds a preset upper limit, it is truncated to that limit; similarly, when the color temperature adjustment exceeds a preset color temperature range, it is also constrained within the allowable range. In this way, the LED light source can transition smoothly from its current state to the target state, rather than undergoing a sudden, large jump.

[0109] The brightness and color temperature driving states of the LED light source are updated gradually within a preset time window. This means that instead of applying all the adjustments at once, the limited brightness and color temperature adjustments are applied to the driver in stages, steps, or sequentially over a pre-defined time period. The preset time window can be understood as a time interval used to control the dimming rhythm; its length can be set according to actual application needs. For example, in scenarios requiring gentler lighting changes, the time window can be set relatively long; in scenarios requiring faster response, the time window can be set relatively short. Gradual updates mean updating the driving state multiple times or sequentially within the same time window, rather than switching directly to the final state at the beginning of the window. Brightness drive state refers to the current drive control state of the LED light source in the brightness adjustment dimension, such as the current setting state, duty cycle setting state, brightness level control state, or equivalent drive state used to drive brightness output. Color temperature drive state refers to the current drive control state of the LED light source in the color temperature adjustment dimension, such as the control state used to adjust the output ratio of warm light channels and cool light channels, the overall color channel ratio state, or the equivalent color temperature control state. Therefore, gradually updating the brightness drive state and color temperature drive state essentially means adjusting the drive control input of the LED light source in both brightness and color temperature dimensions step by step, so that it gradually approaches the target output state corresponding to the dimming parameters from its current state.

[0110] For example, in one implementation, if the dimming parameters indicate that the current LED light source needs to increase brightness and slightly decrease color temperature, the brightness adjustment amount and color temperature adjustment amount can be determined first based on the dimming parameters. If the brightness adjustment amount is greater than the preset maximum single brightness change amount, the brightness adjustment amount is first limited to an allowable range through limiting processing. If the color temperature adjustment amount exceeds the single color temperature change threshold, limiting is also applied. Subsequently, within a preset time window, the brightness driving state of the LED light source is updated several times to gradually increase the brightness, while the color temperature driving state is updated according to a predetermined step size to gradually decrease the overall color temperature. This allows the light source output to smoothly transition to the new brightness and color temperature states.

[0111] For example, in another implementation, when the environment needs to be adjusted from a darker, warmer state to a brighter, more neutral color temperature state, the system can first decompose the corresponding dimming parameters into a larger brightness adjustment amount and a smaller color temperature adjustment amount, and then perform limiting processing on each. Afterwards, the brightness drive state and color temperature drive state can be updated sequentially over multiple consecutive time windows, allowing the LED light source to first achieve a significant brightness increase, and then simultaneously perform a more gradual color temperature adjustment, thereby avoiding abrupt visual changes.

[0112] Furthermore, in some implementations, the above adjustment process can be combined with iterative closed-loop updates. Specifically, after updating the brightness-driven state and color-temperature-driven state within a time window, the current scene image can be acquired again, and the actual light environment state of the target area can be reassessed. The subsequent adjustment amounts are then adjusted based on the new acquisition results. In this way, the adjustment of the LED light source is not performed only once, but can gradually approach a better state over multiple adjustment cycles. This closed-loop update method can be used as an optional extended implementation to further improve adjustment accuracy and scene adaptability.

[0113] To achieve an adaptive dimming method for indoor scenes, this invention also provides a camera-sensing LED lighting adaptive dimming system, such as... Figure 2As shown, the system includes: a detection lighting control module for controlling the LED light source to output at least two different detection lighting states; an image acquisition module for acquiring environmental images of the same scene under each detection lighting state using a camera; an image preprocessing module for preprocessing the environmental images to obtain a target area image; a reliable pixel extraction module for extracting a reliable pixel subset from the target area image, wherein the reliable pixel subset is a set of effective pixels in the target area obtained after interference suppression processing; a reflection characteristic parameter calculation module for constructing a multi-source response vector of the target area based on the reliable pixel subset under each detection lighting state, and selecting a locally optimal sample subset from a preset sample library based on the multi-source response vector to calculate the reflection characteristic characterization parameters of the target area; a correction and parameter generation module for correcting the illuminance and color temperature estimation results of the target area using the reflection characteristic characterization parameters to obtain the target brightness and target color temperature, and generating corresponding dimming parameters; and a dimming execution module for adjusting the LED light source using the dimming parameters.

[0114] In this embodiment, the system may include a lighting terminal body, an image sensing unit, and a control processing unit. The lighting terminal body may be installed on the indoor ceiling, wall, ceiling track, inside the lamp housing, or in an installation position adjacent to the lamp; the image sensing unit may be a camera assembly facing the target area; the control processing unit may be located on the control board inside the lamp, an independent control box, an edge control gateway, or a main control board electrically connected to the lamp. Each module may be implemented by the processor in the control processing unit calling program instructions in the memory, or may be partially implemented by dedicated circuits, driver chips, or image processing chips.

[0115] The detection lighting control module can be embedded in the control processing unit and electrically connected to the LED driver circuit. The LED driver circuit can be mounted on the luminaire driver board to receive control commands from the control processing unit and change the output state of the LED light source during the detection phase. Specifically, the detection lighting control module can switch the brightness parameters, color temperature parameters, LED sub-light source combinations, and spectral power distribution of the LED light source through PWM control signals, constant current drive adjustment signals, channel duty cycle control signals, or bus control commands. For example, when the LED light source is a dual-channel color temperature adjustable luminaire, the detection lighting control module can form at least two detection lighting states by adjusting the driving ratio of the warm light channel and the cool light channel respectively; when the LED light source is a multi-sub-light source combination structure, the detection lighting control module can form different detection lighting states by switching the start / stop states of different sub-light source channels.

[0116] The image acquisition module can consist of a camera assembly and its interface circuitry. The camera assembly can be fixedly mounted on the edge of the lamp housing, an indoor ceiling bracket, a wall mount, or an independent sensing mounting base, with its acquisition direction facing the illuminated target area. The camera assembly can be connected to the control processing unit via a serial camera interface, USB interface, MIPI interface, parallel image interface, or equivalent image transmission interface. After receiving a trigger signal from the control processing unit, the image acquisition module acquires environmental images of the same scene under each detection illumination state and transmits the image data to the image preprocessing module. In one implementation, the image data can be transmitted in a frame image data format, where each image frame may contain a timestamp, detection illumination status identifier, and pixel matrix data. In another implementation, the image frames and status numbers can be packaged and transmitted to the control processing unit via an internal bus.

[0117] The image preprocessing module, the reliable pixel extraction module, the reflectance characteristic parameter calculation module, and the correction and parameter generation module can be executed sequentially by the same processor in the same software program, or they can be implemented by different functional threads in the control processing unit, the image coprocessing unit, or the AI ​​computing unit. The control processing unit can be equipped with a memory to store program instructions, preset sample libraries, preset correction rules, preset brightness mapping relationships, preset color temperature mapping relationships, and intermediate processing data.

[0118] The image preprocessing module receives the environmental image from the image acquisition module, preprocesses it, and outputs the target region image. Subsequently, the reliable pixel extraction module extracts a subset of reliable pixels from the target region image and outputs the corresponding data structure to the reflection characteristic parameter calculation module. The output data here can be a set of pixel coordinates, a pixel mask matrix, a feature buffer block, or a compressed set of valid pixel indices.

[0119] The reflection characteristic parameter calculation module can be deployed in the main processor of the control processing unit, or partially executed using a DSP, NPU, or image accelerator. When the control processing unit itself has edge AI computing capabilities, the preset sample library can be directly stored in local memory; when the system adopts a split structure, the preset sample library can also be stored in the edge gateway or local area server, and the reflection characteristic parameter calculation module can access the sample library data via wired or wireless means. For example, the lighting control board and the edge gateway can exchange data via RS485, CAN, Ethernet, Wi-Fi, Bluetooth Mesh, ZigBee, or other local communication methods. In this case, the environmental image or its intermediate features acquired by the image acquisition module can first be preprocessed and reliable pixel extracted locally, and then the multi-source response vector can be uploaded to the edge gateway. The edge gateway then performs local optimal sample subset selection and reflection characteristic parameter calculation based on the preset sample library, and finally returns the calculation results to the control processing unit.

[0120] The correction and parameter generation module can also be deployed in the control processing unit or edge gateway. It interacts with the reflection characteristic parameter calculation module via internal function calls, shared buffers, message queues, or bus data packets. After receiving the reflection characteristic parameters, this module combines them with the correction rules, brightness mapping relationships, and color temperature mapping relationships stored in memory to generate the target brightness, target color temperature, and corresponding dimming parameters. In one implementation, the target brightness, target color temperature, and dimming parameters can be stored and transmitted using a structured data format, such as recording the target brightness value, target color temperature value, brightness adjustment parameters, and color temperature adjustment parameters in the form of a parameter table. In another implementation, parameter data frames containing field identifiers can also be used for transmission to facilitate subsequent calls by the dimming execution module.

[0121] The dimming execution module can be directly connected to the LED driver circuit to receive dimming parameters output by the calibration and parameter generation module and convert them into update commands for brightness and color temperature drive states. This module can be located in the main control board of the luminaire or in a separate driver board. If the dimming execution module and the LED driver circuit are set separately, they can communicate via PWM control lines, I²C bus, SPI bus, UART serial port, DALI bus, or other drive control interfaces. For example, when the dimming parameters include brightness adjustment and color temperature adjustment, the dimming execution module can calculate the corresponding PWM duty cycle update value and channel current ratio update value respectively, and then send these update values ​​to the LED driver circuit. After receiving the update values, the LED driver circuit gradually changes the brightness and color temperature drive states of the LED light source to achieve the target lighting effect.

[0122] In a more specific hardware layout example, the camera assembly can be installed at the edge of the ceiling light fixture. The control processing unit and LED driver circuit can be integrated into the same light fixture control board. The memory and processor are soldered onto the control board, forming an integrated lighting terminal. The detection lighting control module, image preprocessing module, reliable pixel extraction module, reflectivity parameter calculation module, correction and parameter generation module, and dimming execution module are all implemented by the processor on the control board calling stored programs. The camera assembly transmits environmental images to the control board via a flexible cable or serial interface. The control board completes image analysis and parameter generation according to the program flow, and then controls the LED light source output through the driver circuit.

[0123] In another example, the camera component and the light fixture are set up separately. The camera component is installed in the center of the top of the room, and the light fixture is installed on the ceiling or wall. The camera component transmits image data to the edge controller wirelessly. The edge controller performs image analysis and parameter calculation, and then sends the dimming parameters to the light fixture driver board via a wired bus. The light fixture driver board then performs dimming control.

[0124] Therefore, the functional modules in this embodiment are not abstract logical units, but can be embedded into actual hardware structures such as the camera component, control processing unit, memory, LED driver circuit, and communication interface. The modules interact through image data, feature data, parameter data, and drive control data, and work collaboratively via wired or wireless communication. The entire adaptive dimming system not only has a complete software processing flow but also a clear hardware structure and interaction link, enabling camera-sensing LED adaptive dimming functionality in actual indoor lighting equipment.

[0125] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail in conjunction with the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the above embodiments, or make equivalent substitutions for some of the technical features; any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A camera-aware LED lighting adaptive dimming method, characterized in that, The method includes: Control the LED light source to output at least two different detection lighting states, and use a camera to capture environmental images of the same scene under each detection lighting state; The environmental image is preprocessed to obtain the target area image; Extract a subset of reliable pixels from the target region image, wherein the subset of reliable pixels is the set of effective pixels in the target region obtained after interference suppression processing; Based on the reliable pixel subsets under each detection illumination state, a multi-source response vector of the target area is constructed, and a local optimal sample subset is selected from the preset sample library according to the multi-source response vector to calculate the reflection characteristic characterization parameters of the target area. The illuminance and color temperature estimation results of the target area are corrected by using the reflection characteristic characterization parameters to obtain the target brightness and target color temperature, and the corresponding dimming parameters are generated. The LED light source is then adjusted using the dimming parameters. 2.The camera-aware LED lighting adaptive dimming method of claim 1, wherein, The at least two different detection illumination states are illumination states that differ in at least one of the following parameters: color temperature parameter, brightness parameter, LED sub-light source combination, and spectral power distribution. 3.The camera-aware LED lighting adaptive dimming method of claim 1, wherein, The preprocessing of the environmental image includes: restoring the brightness of the environmental image based on the brightness response curve of the camera; performing multi-exposure fusion on environmental images acquired under different exposure conditions to expand the dynamic range of brightness in the target area; and performing low-light enhancement, dark area compensation, and local brightness correction on the fused environmental image to obtain the target area image.

4. The camera-sensing adaptive dimming method for LED lighting according to claim 1, characterized in that, The extraction of a reliable pixel subset from the target region image includes: identifying specular reflection pixels, specular reflection pixels, and low-light noise pixels in the target region; filtering out specular reflection pixels, specular reflection pixels, and low-light noise pixels; and retaining stable pixels with consistent responses under different detection illumination conditions among the diffuse reflection feature pixels as the reliable pixel subset.

5. The camera-sensing adaptive dimming method for LED lighting according to claim 1, characterized in that, The calculation of the reflection characteristic characterization parameters of the target area includes: extracting the response features of the target area based on the reliable pixel subset under each detection illumination state, and combining the extracted response features under each detection illumination state to form a multi-source response vector; calculating the similarity between the multi-source response vector and the response features of each training sample in the preset sample library, selecting at least two training samples whose similarity meets the preset conditions to form a local optimal sample subset; assigning corresponding weights to the training samples in the local optimal sample subset according to the similarity of each training sample, and calculating the reflection characteristic characterization parameters based on the weighted local optimal sample subset.

6. The camera-sensing adaptive dimming method for LED lighting according to claim 1, characterized in that, The reflection characteristic characterization parameters include at least one of the following: a regional reflectance characterization parameter for characterizing the surface reflection characteristics of the target area, a classification parameter for characterizing the reflection type of the target area, a specular reflection intensity parameter for characterizing the specular reflection degree of the target area, a diffuse reflection ratio parameter for characterizing the diffuse reflection contribution of the target area, a reflection compensation coefficient for correcting the color temperature estimation results, and a color temperature compensation coefficient for correcting the color temperature estimation results.

7. The camera-sensing adaptive dimming method for LED lighting according to claim 1, characterized in that, The process of obtaining the target brightness and target color temperature includes: extracting brightness response features and color response features based on a subset of reliable pixels in the target area under various detection illumination conditions; generating illuminance estimation results based on the brightness response features and color temperature estimation results based on the color response features; matching the reflection characteristic characterization parameters of the target area with preset correction rules to determine illuminance correction parameters and color temperature correction parameters respectively; using the illuminance correction parameters to perform weighted correction or bias compensation on the illuminance estimation results to obtain corrected illuminance, and determining the target brightness based on the corrected illuminance according to a preset brightness mapping relationship; using the color temperature correction parameters to perform weighted correction or bias compensation on the color temperature estimation results to obtain corrected color temperature, and determining the target color temperature based on the corrected color temperature according to a preset color temperature mapping relationship.

8. The camera-sensing adaptive dimming method for LED lighting according to claim 1, characterized in that, The generation of corresponding dimming parameters includes: normalizing the target brightness and target color temperature and then inputting them into a lightweight neural network model. The lightweight neural network model includes an input layer, a low-dimensional feature mapping layer, and a parameter output layer connected in sequence. The low-dimensional feature mapping layer is used to perform joint feature extraction and nonlinear mapping on the normalized target brightness and target color temperature. The parameter output layer is used to output dimming parameters corresponding to the brightness adjustment and color temperature adjustment of the LED light source.

9. The camera-sensing adaptive dimming method for LED lighting according to claim 1, characterized in that, The method of adjusting the LED light source using dimming parameters includes: determining the brightness adjustment amount and color temperature adjustment amount according to the dimming parameters; limiting the brightness adjustment amount and color temperature adjustment amount; and gradually updating the brightness driving state and color temperature driving state of the LED light source according to a preset time window.

10. A camera-sensing adaptive dimming system for LED lighting, characterized in that, The system includes: The detection lighting control module is used to control the LED light source to output at least two different detection lighting states; The image acquisition module is used to acquire environmental images of the same scene under various detection lighting conditions using a camera; The image preprocessing module is used to preprocess the environmental image to obtain the target area image; The trusted pixel extraction module is used to extract a subset of trusted pixels from the target region image. The trusted pixel subset is the set of effective pixels in the target region obtained after interference suppression processing. The reflection characteristic parameter calculation module is used to construct the multi-source response vector of the target area based on the reliable pixel subset under each detection illumination state, and to select the local optimal sample subset from the preset sample library according to the multi-source response vector in order to calculate the reflection characteristic characterization parameters of the target area. The calibration and parameter generation module is used to correct the illuminance and color temperature estimation results of the target area using the reflection characteristic characterization parameters, so as to obtain the target brightness and target color temperature, and generate the corresponding dimming parameters. The dimming execution module is used to adjust the LED light source using dimming parameters.

Images