Light supplement control method and monitoring system

Abstract

The invention provides a light supplement control method and a monitoring system, and relates to the technical field of imaging. The light supplementing control comprises the following steps: determining the light supplementing intensity for supplementing light to the light supplementing device based on the collected environment data; the light supplementing device comprises an infrared light assembly and a visible light assembly; and performing light supplementing control on the light supplementing device according to the light supplementing intensity. The monitoring system comprises a processor and a light supplementing device. The processor is connected with the light supplementing device; the processor is used for determining the light supplementing intensity for supplementing light to the light supplementing device based on the collected environment data; performing light supplementing control on the light supplementing device according to the light supplementing intensity; the light supplementing device comprises an infrared light assembly and a visible light assembly. The light supplementing intensity of the infrared light assembly and the visible light assembly is determined based on the actual environment condition, and light supplementing control is carried out, so that the infrared light assembly and the visible light assembly can adapt to the environment to carry out cooperative work, the light supplementing effect is effectively optimized, the power consumption of light supplementing is reduced, and the quality of an image obtained based on light supplementing is improved.

Description

Technical Field

[0001] The present application relates to the field of imaging technology, and in particular to a fill light control method and a monitoring system. Background Art

[0002] In applications such as intelligent security and computer vision, image quality in low-light environments is crucial. Existing fill-light technologies have exposed numerous issues in actual use, including red exposure when the infrared fill-light wavelength is 850nm or less, significant differences in effective distance and fill-light efficiency between different wavelengths of infrared fill-light, and high power consumption required for fill-light. Furthermore, existing visible light fill-light technologies suffer from single-band color shift in low-light environments. For example, the lack of blue light in sodium lamp lighting results in a yellowish image, making accurate color reproduction impossible. These issues severely limit the application scope and effectiveness of monitoring devices, resulting in poor image quality and, consequently, poor monitoring effectiveness, making them unable to meet existing monitoring and management needs. Summary of the Invention

[0003] In view of this, an object of the embodiments of the present application is to provide a fill light control method and a monitoring system to improve the problem of poor monitoring effect in the prior art.

[0004] In order to solve the above problems, in a first aspect, an embodiment of the present application provides a fill light control method, the method comprising:

[0005] Determining the fill light intensity of a fill light device based on the collected environmental data; wherein the fill light device includes an infrared light component and a visible light component;

[0006] The fill light device is controlled according to the fill light intensity.

[0007] In the above implementation process, the actual environmental conditions can be determined based on the collected environmental data, so as to determine the fill light intensity required for fill light according to the actual environmental conditions, and the infrared light component and visible light component in the fill light device can be controlled according to the fill light intensity, so that the infrared light component and the visible light component can adapt to the environment and work together to fill light. The ability to perform fill light for different bands effectively optimizes the fill light effect of the fill light device, and the ability to adjust the fill light intensity according to actual needs, reducing the power consumption required for fill light. It is suitable for fill light in a variety of different lighting scenarios, effectively improving the quality of images or videos obtained by monitoring or shooting based on the fill light device, thereby optimizing the monitoring effect and meeting the monitoring and management needs in a variety of different application scenarios.

[0008] Optionally, determining the fill light intensity for the fill light device based on the collected environmental data includes:

[0009] The environmental data is collected by a sensor device; wherein the environmental data includes: target distance and spectral information;

[0010] Determining a first fill light factor corresponding to the infrared light component and a second fill light factor corresponding to the visible light component based on the trained target model;

[0011] determining the fill light intensity for performing fill light on the fill light device based on the environmental data, the first fill light factor, and the second fill light factor;

[0012] The fill light intensity includes: a first fill light intensity corresponding to the infrared light component and a second fill light intensity corresponding to the visible light component.

[0013] In the above implementation process, the environmental data of the current environment can be collected in real time according to the set sensing device. The environmental data may include the target distance to the target object and the spectral information in the current environment. Taking into account the differences and mutual influence relationship between the two different types of components in the fill light device, as well as the correlation between power consumption and fill light intensity, the fill light factors of the two components can be determined based on the trained target model, so as to combine the environmental data and the fill light factors to determine the first fill light intensity corresponding to the infrared light component and the second fill light intensity corresponding to the visible light component. The fill light intensities of the infrared light component and the visible light component can be determined based on the actual environmental conditions and the fill light factors that affect the fill light effect and power consumption, so as to perform coordinated fill light control processing on the infrared light component and the visible light component, effectively optimizing the fill light effect, improving the quality of the obtained image or video, and reducing the power consumption of the fill light device.

[0014] Optionally, the determining the fill light intensity for filling light for the fill light device based on the environmental data, the first fill light factor, and the second fill light factor includes:

[0015] determining the first fill light intensity of the infrared light assembly based on the environmental data, the first fill light factor, and a first fill light formula;

[0016] The second fill light intensity of the visible light component is determined based on the first fill light intensity, the second fill light factor, and the second fill light formula.

[0017] In the above implementation process, the corresponding first fill light formula and second fill light formula can be determined based on the fill light influencing factors of the infrared light component and the visible light component. The first fill light intensity of the infrared light component can be determined by combining the environmental data, the first fill light factor, and the first fill light formula. Moreover, considering the correlation between the fill light effects of the two components, the second fill light intensity of the visible light component can be determined based on the first fill light intensity, the second fill light factor, and the second fill light formula. The corresponding fill light formula can be designed and calculated based on the actual operating conditions of different components, and the second fill light intensity of the visible light component can be calculated based on the first fill light intensity of the infrared light component, effectively improving the effectiveness of the two fill light intensities.

[0018] Optionally, the first fill light formula includes:

[0019]

[0020] Among them, I IR is the first fill light intensity, K is the first fill light factor, D is the target distance, S / N target is the target signal-to-noise ratio, α is the atmospheric attenuation coefficient determined based on the spectral information and environmental physical parameters;

[0021] The second fill light formula includes:

[0022]

[0023] in, is the second fill light intensity, η is the second fill light factor, C req The total amount of color required to achieve color reproduction, C IR is the contribution of infrared light to the overall color during color restoration determined based on the spectral information, C max The maximum color contribution provided to visible light.

[0024] In this implementation, the actual operating conditions and hardware characteristics of the infrared light component are combined, and the hardware and algorithm are coupled to determine the first fill light formula based on the corresponding attenuation compensation algorithm. This enables the infrared light component to achieve long-distance, covert fill light and noise control without red exposure. Furthermore, the actual operating conditions and hardware characteristics of the visible light component are combined, and the second fill light formula is determined based on the corresponding color compensation algorithm. This allows the visible light component to break through the limitations of single-band fill light and fixed color correction, achieving dynamic color balance through hardware-level spectral perception.

[0025] Optionally, the target model is trained in the following manner:

[0026] Collect historical data of different lighting conditions; wherein the lighting conditions include at least one of: light intensity, lighting scene, lighting distance, and weather conditions;

[0027] Performing data preprocessing on the historical data to obtain training data; wherein the data preprocessing includes: data cleaning, data normalization, image processing and labeling;

[0028] Determining a reward function shared by the first fill light factor and the second fill light factor; wherein the reward function is constructed based on a mapping relationship between the environmental data, the first fill light factor, and the second fill light factor, and the reward function is used to balance a peak signal-to-noise ratio (PSNR) of an image, a color reproduction error, and fill light power consumption;

[0029] Based on the lightweight model, training is performed in combination with the training data and the reward function to obtain the target model.

[0030] In the above implementation process, in order to determine the fill light factor that affects the fill light intensity, the corresponding target model can be trained for processing. Historical data under different lighting conditions are collected and preprocessed accordingly to obtain training data used for model training. The training data is then processed in a lightweight model in combination with the reward function shared by the two fill light factors to obtain the corresponding target model. The ability to train based on data under different lighting conditions improves the effectiveness of the fill light factor determined based on the target model, thereby improving the effectiveness of the fill light intensity and the adaptability and performance of the fill light device in complex scenarios. The use of a lightweight model for processing effectively reduces the amount of computation and cost required for training and calculation, increases the processing speed of the model, and thus improves the efficiency of the fill light device in performing fill light.

[0031] Optionally, controlling the fill light device according to the fill light intensity includes:

[0032] Processing the first fill light intensity based on signal conversion to obtain a first control signal for controlling the fill light of the infrared light component;

[0033] Processing the second fill light intensity based on pulse width modulation to obtain a second control signal for controlling the fill light of the visible light component;

[0034] The first control signal and the second control signal share a common clock source.

[0035] In the above implementation process, taking into account the device differences between the infrared light component and the visible light component, when performing fill light control, the infrared light component and the visible light component can be controlled and processed separately according to the device characteristics and the corresponding fill light intensity. The first fill light intensity can be processed by signal conversion technology to directly obtain a first control signal for fill light control of the infrared light component, and the second fill light intensity can be processed by pulse width modulation technology to obtain a second control signal for fill light control of the visible light component. In addition, in order to achieve synchronous collaborative fill light, the first control signal and the second control signal share the same clock source. According to the actual control logic of the device, the components can be controlled accordingly in combination with the fill light intensity, effectively improving the efficiency of the collaborative fill light of the fill light device and the fill light effect.

[0036] Optionally, the processing of the second fill light intensity based on pulse width modulation to obtain a second control signal for controlling the fill light of the visible light component includes:

[0037] Determine the operating frequency of pulse width modulation according to the application scenario;

[0038] Obtaining an adjustable duty cycle of pulse width modulation based on the linear mapping conversion of the second fill light intensity;

[0039] Determining duty cycle offsets of multiple bands of the visible light component based on the adjustable duty cycle and the spectral information;

[0040] The second control signal is determined according to the duty cycle offset.

[0041] In the above implementation process, the operating frequency of the pulse width modulation can be first determined based on the actual application scenario, and a linear mapping conversion process can be performed based on the second fill light intensity to obtain an adjustable duty cycle of the pulse width modulation. The adjustable duty cycle and the spectral information are combined to determine the duty cycle offsets of the multiple bands corresponding to the visible light component, and then the corresponding pulse width modulation signal is determined as the corresponding second control signal based on the duty cycle offsets. Based on the actual scenario and the fill light intensity, the second control signal capable of achieving dynamic color correction can be determined, thereby achieving dynamic fill light adjustment of the visible light component and further optimizing the fill light effect of the fill light device.

[0042] Optionally, the sensing device includes: a distance sensor covering close-range, macro-range and long-range monitoring ranges, and a multispectral sensor covering the full-band spectral range of ultraviolet light, visible light and near-infrared light;

[0043] The environmental data collected by the sensor device includes:

[0044] Acquiring initial distance data collected by the distance sensor;

[0045] Performing compensation processing on the initial distance data to obtain the target distance;

[0046] The spectral information obtained by the multispectral sensor based on environmental detection is acquired.

[0047] In the above implementation process, the sensing device may include a distance sensor capable of monitoring multiple distances and a multispectral sensor covering multiple different spectral bands, thereby effectively improving the validity of the environmental data collected by the sensing device. After obtaining the initial distance data collected by the distance sensor, the initial distance data can be compensated to improve the accuracy of the data, taking into account the influencing factors during detection. The initial distance data can then be combined with the spectral information detected by the multispectral sensor based on the current environment as the final collected environmental data. The ability to collect data using a device with a wider detection range effectively improves the authenticity and accuracy of the environmental data, thereby increasing the effectiveness of the fill light intensity determined based on the environmental data.

[0048] In a second aspect, an embodiment of the present application further provides a monitoring system, the monitoring system comprising: a processor and a fill light device;

[0049] The processor is connected to the fill light device;

[0050] The processor is used to determine the fill light intensity of the fill light device based on the collected environmental data; and perform fill light control on the fill light device according to the fill light intensity; wherein the fill light device includes an infrared light component and a visible light component.

[0051] In the above implementation process, the processor can determine the actual environmental conditions based on the collected environmental data, determine the fill light intensity required for fill light according to the actual environmental conditions, and control the infrared light component and visible light component in the fill light device according to the fill light intensity, so that the infrared light component and visible light component can adapt to the environment to perform fill light work.

[0052] Optionally, in the fill light device, the infrared light component and the visible light component are packaged on the same substrate with the same aperture, and the infrared light emitted by the infrared light component and the visible light emitted by the visible light component are emitted through the same aperture;

[0053] The wavelength range of the infrared light component is 940nm±10nm, and the wavelength range of the visible light component is 400nm-700nm;

[0054] The infrared light component includes a plurality of infrared light LEDs, and the visible light component includes a plurality of visible light LEDs;

[0055] The plurality of infrared light LEDs and the plurality of visible light LEDs are staggered and arranged in a matrix structure or a ring shape;

[0056] The light emitting centers of all the infrared light LEDs and the visible light LEDs are in the same plane and have the same emission direction; wherein the deviation of the emission optical axes of the infrared light and the visible light is less than or equal to 1°, and the flatness error of the light emitting centers of the infrared light and the visible light is less than or equal to 0.05mm.

[0057] In the above implementation process, the infrared light component and the visible light component can be packaged on the same substrate with the same aperture. The infrared light emitted by the infrared light component and the visible light emitted by the visible light component can be emitted through the same aperture to ensure that the infrared light and visible light can be emitted from the same position, reducing the deviation of the light during the propagation process, and making the fill light more uniform and accurate. In addition, in order to achieve the fill light function of different bands, the appropriate wavelength range of the infrared light component and the visible light component can be selected respectively. While filling the light, the red exposure of the infrared light component can also be adjusted by wavelength selection. Taking into account the different fill light requirements of different scenes, the infrared light component and the visible light component can include multiple corresponding LED devices, and the multiple infrared LEDs and the multiple visible light LEDs are staggered to ensure that visible light and infrared light are emitted in the same area at the same time, and the light emission centers of all infrared LEDs and visible light LEDs are in the same plane and the emission directions are consistent, further optimizing the uniformity and reliability of the fill light.

[0058] In summary, the embodiments of the present application provide a fill light control method and a monitoring system, which can determine the fill light intensity of the infrared light component and the visible light component based on the actual environmental conditions and perform fill light control, so that the infrared light component and the visible light component can adapt to the environment and work together, effectively optimizing the fill light effect, reducing the power consumption of the fill light, and thus improving the quality of the image obtained based on the fill light. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 A flow chart of a fill light control method provided in an embodiment of the present application;

[0061] Figure 2 A detailed flowchart of step S100 provided in an embodiment of the present application;

[0062] Figure 3 A detailed flowchart of step S130 provided in an embodiment of the present application;

[0063] Figure 4 A schematic flow chart of another fill light control method provided in an embodiment of the present application;

[0064] Figure 5 A detailed flowchart of step S200 provided in an embodiment of the present application;

[0065] Figure 6 A detailed flowchart of step S220 provided in an embodiment of the present application;

[0066] Figure 7 A detailed flowchart of step S110 provided in an embodiment of the present application;

[0067] Figure 8 A schematic diagram of the structure of a monitoring system provided in an embodiment of the present application.

[0068] Icon: 400-processor; 500-fill light device; 510-infrared light component; 520-visible light component. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of them. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the embodiments of the present application.

[0070] Existing fill light technology has exposed numerous practical issues. For example, when the infrared fill light wavelength is 850nm or less, the human eye can see red light. In security surveillance scenarios, this phenomenon can easily reveal the camera's location, compromising the equipment's concealment. This makes it unsuitable for applications requiring covert surveillance, such as secret surveillance areas and privacy-protected locations. This not only reduces security but can also prematurely alert the target, rendering the surveillance ineffective. Different infrared fill light wavelengths vary significantly in their effective range and efficiency. For example, while an 850nm wavelength offers a relatively high efficiency of 120lm / W, its effective range is only 30 meters and suffers from red light exposure. While a 940nm wavelength overcomes this red light exposure issue and is invisible to the human eye, its efficiency drops to 95lm / W, shortening its effective range to 20 meters. While a 1550nm wavelength offers an effective range of up to 50 meters, its efficiency drops further to 60lm / W. These limitations on fill light efficiency and distance make it difficult to select the appropriate infrared fill light solution for different scenarios, failing to meet the dual requirements of long-distance and efficient fill light. Furthermore, traditional fill lights typically consume high power in constant-on mode. In battery-powered applications, such as field surveillance equipment and portable monitoring devices, high power consumption means frequent battery replacement and charging, increasing maintenance costs and difficulty. It can also cause the device to malfunction due to low power at critical moments, severely impacting its usability and stability.

[0071] Therefore, due to the red exposure problem when the infrared fill light wavelength is less than or equal to 850nm, the significant differences in the effective distance and fill light efficiency of infrared fill lights of different wavelengths, and the high power consumption required for fill light, in addition, the existing technology has a single-band color shift problem in low-light environments. For example, the lack of blue light in the sodium lamp lighting scene causes the image to be yellowish, and accurate color reproduction cannot be achieved. These problems seriously limit the application scope and effect of the monitoring device, resulting in poor image quality obtained by the monitoring device, and thus poor monitoring effect, which cannot meet the existing monitoring management needs.

[0072] In order to solve the above problems, the embodiments of the present application provide a fill light control method and a monitoring system. The fill light control method is applied in a processor set in the monitoring system, and can determine the fill light intensity of the infrared light component and the visible light component based on the actual environmental conditions and perform fill light control, so that the infrared light component and the visible light component can adapt to the environment and work together, effectively optimizing the fill light effect, reducing the power consumption of the fill light, and thus improving the quality of the image obtained based on the fill light.

[0073] See also Figure 1 , Figure 1A flow chart of a fill light control method provided in an embodiment of the present application, the method may include steps S100-S200.

[0074] Step S100: determining the fill light intensity for the fill light device based on the collected environmental data.

[0075] The actual environmental conditions can be determined based on the collected environmental data, so as to determine the fill light intensity required for the fill light according to the actual environmental conditions.

[0076] Optionally, in order to realize the fill light function of different bands, the fill light device may include an infrared light component and a visible light component. The infrared light component may be an infrared light LED array composed of a corresponding plurality of infrared light LEDs, and the visible light component may be a visible light LED array composed of a corresponding plurality of visible light LEDs.

[0077] Step S200: Controlling the fill light device according to the fill light intensity.

[0078] The infrared light component and the visible light component in the fill light device can be controlled according to the fill light intensity, so that the infrared light component and the visible light component can adapt to the environment and perform coordinated fill light work.

[0079] exist Figure 1 In the embodiment shown, fill light can be performed for different bands, effectively optimizing the fill light effect of the fill light device, and the fill light intensity can be adjusted according to actual needs, reducing the power consumption required for fill light. It is suitable for fill light in a variety of different lighting scenarios, effectively improving the quality of images or videos obtained by monitoring or shooting based on the fill light device, thereby optimizing the monitoring effect and meeting the monitoring and management needs in a variety of different application scenarios.

[0080] Optionally, see Figure 2 , Figure 2 A detailed flowchart of step S100 is provided in an embodiment of the present application. Step S100 may include steps S110-S130.

[0081] Step S110: collecting environmental data through a sensor device.

[0082] Among them, environmental data of the current environment can be collected in real time according to the set sensing device. The sensing device can include various types of detection devices such as distance sensors for measuring distance and spectral sensors for detecting spectra. The environmental data can include the target distance and spectral information obtained by detection. The target distance is the distance data between the camera and the target object detected by the distance sensor, for example, the distance between the camera and the person appearing in the picture is 10m, etc. The spectral information is the intensity distribution of different wavelengths and lighting characteristics obtained by the spectral sensor for environmental analysis of different scenes, such as the ratio of infrared light and visible light in the ambient light, whether there is interference from special light sources, etc.

[0083] Step S120 , determining a first fill light factor corresponding to the infrared light component and a second fill light factor corresponding to the visible light component based on the trained target model.

[0084] Step S130 : determining fill light intensity for fill light provided to the fill light device based on the environmental data, the first fill light factor, and the second fill light factor.

[0085] The fill light intensity can include a first fill light intensity corresponding to the infrared component and a second fill light intensity corresponding to the visible light component. Taking into account the differences and mutual influence between the two different types of components in the fill light device, as well as the correlation between power consumption and fill light intensity, the fill light factors of the two components can be determined based on the trained target model. This, combined with the environmental data and the fill light factors, can determine the first fill light intensity corresponding to the infrared component and the second fill light intensity corresponding to the visible light component.

[0086] Optionally, the first fill light factor and the second fill light factor may both be corresponding weight factors.

[0087] exist Figure 2 In the embodiment shown, the fill light intensities of the infrared light component and the visible light component can be determined respectively based on the actual environmental conditions and the fill light factors that affect the fill light effect and power consumption, so as to perform coordinated fill light control processing on the infrared light component and the visible light component, effectively optimizing the fill light effect, improving the quality of the obtained image or video, and reducing the power consumption of the fill light device.

[0088] Optionally, see Figure 3 , Figure 3 A detailed flowchart of step S130 is provided in an embodiment of the present application. Step S130 may include steps S131-S132.

[0089] Step S131 : determining a first fill light intensity of the infrared light component based on environmental data, a first fill light factor, and a first fill light formula.

[0090] Among them, the corresponding first fill light formula and second fill light formula can be determined according to the fill light influencing factors of the infrared light component and the visible light component, and the first fill light intensity of the infrared light component can be determined in combination with the environmental data, the first fill light factor and the first fill light formula.

[0091] Step S132 : determining a second fill light intensity of the visible light component based on the first fill light intensity, the second fill light factor, and the second fill light formula.

[0092] In which, considering the correlation effect of the two components when filling light, the second fill light intensity of the visible light component can be determined according to the first fill light intensity, the second fill light factor and the second fill light formula.

[0093] It should be noted that the first fill light formula may include:

[0094]

[0095] Among them, I IR is the first fill light intensity, K is the first fill light factor, D is the target distance, S / N target is the target signal-to-noise ratio, and α is the atmospheric attenuation coefficient determined based on spectral information and environmental physical parameters. target It determines the ratio of signal intensity to noise intensity required for imaging and is one of the key indicators to ensure imaging quality. The target distance D represents the distance between the camera and the target object. As the distance increases, the infrared light will attenuate during the propagation process. Therefore, the square of the distance D 2 The formula reflects the effect of distance on fill light intensity. The atmospheric attenuation coefficient α reflects the intensity attenuation of infrared light due to absorption, scattering and other factors when it propagates in the atmosphere. -αD This describes the exponential relationship between attenuation and distance. The first fill light factor, K, is a comprehensive adjustment factor used to calibrate the calculation results based on actual hardware parameters and scene characteristics. This method combines the actual operating conditions and hardware characteristics of infrared light components, couples hardware with algorithms, and determines the first fill light formula based on the corresponding attenuation compensation algorithm. This solves the conflict between long-distance covert fill light (without red exposure) and noise control, which is unattainable with existing technologies. This enables infrared light components to achieve long-distance covert fill light and noise control without red exposure.

[0096] It should be noted that the second fill light formula may include:

[0097]

[0098] in, is the second fill light intensity, η is the second fill light factor, C req The total amount of color required to achieve color reproduction, C IRis the contribution of infrared light to the overall color during color restoration based on spectral information, C max The maximum color contribution provided to visible light. Second fill light intensity According to C req with C IR Dynamic adjustment. C req Refers to the total amount of color required to achieve an ideal color reproduction effect, which is determined based on the color calibration standard of the imaging device and the color requirements of the actual scene. IR Indicates the contribution of infrared light to the overall color during color restoration. Since infrared light itself does not carry color information, it may affect the color balance of the image during the fill light process, so it needs to be taken into account when calculating the visible light fill light intensity. max The second fill light factor, η, is the maximum color contribution that visible light can provide, depending on the spectral characteristics and luminous intensity of the visible light component. This factor is used to adjust the calculated visible light fill light intensity to accommodate varying hardware performance and scene conditions. This second fill light formula can be determined based on a corresponding color compensation algorithm, taking into account the actual operating conditions and hardware characteristics of the visible light component. This allows the visible light component to break through the limitations of single-band fill light and fixed color correction in existing technologies, achieving dynamic color balance through hardware-level spectral perception.

[0099] Optionally, the mapping relationship between the spectral information collected by the multispectral sensor and the parameters in the formula exists in the following situations: 1. The spectral information includes the light intensity distribution of the entire band of 300-1600nm; the corresponding formula parameter is the atmospheric attenuation coefficient α; the physical meaning and calculation logic: taking the wavelength range of the infrared light component as 940nm as an example, the attenuation rate of the infrared light intensity at 940nm is reversed: α=-ln(I measured / I reference ) / D(I measured : Measured 940nm light intensity; I reference : initial light intensity without attenuation); in foggy and rainy days, α increases with the increase of particle concentration (e.g., α = 0.1 / m in foggy and smoggy weather, α = 0.01 / m on sunny days); 2. Spectral information includes spectral components of the visible light band (400-700nm); corresponding formula parameter: the total color volume C required to achieve color restoration req ; Physical meaning and calculation logic: Calculate the ideal color ratio: C req =[R req , G req , B req Example: C under standard daylight req =[30%, 50%, 20%]; in sodium lamp environment, due to lack of blue light, C req=[25%, 45%, 30%] (forced increase in the proportion of blue light); 3. Spectral information includes the intensity of light in the near-infrared band (900-1000nm); corresponding formula parameters: the contribution of infrared light to the overall color during color reproduction determined based on spectral information C IR ; Physical meaning and calculation logic: The 940nm infrared light contribution of the fill light device itself: (k is the response coefficient of the visible light sensor to 940nm, obtained through calibration); the infrared contribution of ambient natural light: The proportion of light intensity to the total color of visible light (e.g. infrared light accounts for 40% of the sunset, That is, total 4. Spectral information includes ambient light intensity (lux); corresponding formula parameters: formula input constraints; physical meaning and calculation logic: when light intensity > 50 lux, the fill light is forced to turn off When the light intensity is less than 0.1lux and less than or equal to 50lux, the dual-band coordinated fill light is activated; when the light intensity is less than or equal to 0.1lux, the infrared power is increased to 8-10W first (the first fill light formula is dominant).

[0100] Optionally, parameter distances are described in a variety of different scenarios:

[0101] Scenario 1: Road illuminated by sodium lamps (light intensity = 2 lux, distance D = 20 m): Multispectral sensor data: Spectral distribution: yellow light (550-600 nm) accounts for 80%, blue light (400-500 nm) accounts for only 5%, and near infrared (900-1000 nm) accounts for 10% (ambient infrared + sodium lamp stray light); 940 nm light intensity: I measured =1lux (D=20m, α=0.1 / m, I reference =2.718lux, e^(-0.1×20)=0.135); calculate the first fill light intensity I by the first fill light formula IR :α=0.1 / m,S / N target =25dB; Substitute into the first fill light formula: The K value is obtained through target model training. During training, the optimization targets are PSNR ≥ 30dB and power consumption ≤ 5W. The Adam optimizer is used to iterate for multiple rounds. For example, it converges after 50 rounds. For example, when K = 2, I IR =4.32W, medium power infrared fill light; calculate the second fill light intensity through the second fill light formula The standard blue light accounts for 20%, which needs to be compensated to 20% (currently only 5%), so C req =[30%, 50%, 20%] (forced blue light to 20%); C IR Fill light infrared contribution (k=0.1, visible light sensor has low response to 940nm), ambient infrared contribution [80%, 5%, 5%] = [8%, 0.5%, 0.5%]; Total C IR =[8.432%, 0.5%, 0.5%]; Substitute into the second fill light formula: (When η=1, the proportion of blue light LEDs increases to 19.5% / total visible light = about 30%, compensating for the blue light deficiency problem of sodium lamps, ΔE<10).

[0102] Scene 2: Outdoor monitoring at sunset (light intensity = 5 lux, distance D = 10m): Multispectral sensor data: Spectral distribution: red light (600-700nm) accounts for 70%, green light (500-600nm) accounts for 15%, blue light (400-500nm) accounts for 10%, near infrared (900-1000nm) accounts for 5% (ambient natural light infrared); 940nm light intensity: measured I measured =0.5lux (due to atmospheric attenuation, initial I reference =5lux, D=10m); calculate the first fill light intensity I by the first fill light formula IR : Atmospheric attenuation coefficient α = -ln(0.5 / 5) / 10 = 0.23 / m (simulating light haze at sunset); target signal-to-noise ratio S / N target =20dB (imaging clarity requirement), substitute into the first fill light formula: (K is the hardware calibration coefficient, assuming K = 2, I IR =1W, turn on low-power infrared fill light); calculate the second fill light intensity using the second fill light formula The standard color ratio is [30%, 50%, 20%], but the sunset red light is excessive, so it needs to be corrected to [20%, 45%, 35%] (reduce red light and increase blue light); C IR Ambient infrared contribution (When the fill light device does not turn on infrared, );Substitute into the second fill light formula: (When η=1, the brightness of each visible light band increases proportionally, and the proportion of blue light LEDs increases to 34.5%, correcting the reddish cast problem).

[0103] Scene 3: Rainy and foggy environment (light intensity = 0.5 lux, distance D = 30m): Multispectral sensor data: Spectral distribution: scattering is enhanced across all bands, visible light is uniformly attenuated, near infrared (900-1000nm) accounts for 30% (ambient scattered infrared); 940nm light intensity: I measured =0.3lux(D=30m,α=0.3 / m, strong attenuation); calculate the first fill light intensity I by the first fill light formula IR:α=0.3 / m;S / N target =15dB (the noise is allowed to be slightly higher under low illumination), substitute into the first fill light formula: Due to the extremely long distance and strong attenuation, the K value needs to be greatly increased to 1000, and I IR =7.4W, close to the upper limit of infrared power); calculate the second fill light intensity through the second fill light formula Colors are blurred in rain and fog, and the contrast needs to be enhanced. req =[35%, 45%, 20%] (enhance red-green contrast); C IR Fill light infrared contribution Infrared scattering in rain and fog causes the visible light sensor response to increase), and the ambient infrared contribution (Diffuse light is warmer); Total C IR =[10.48%, 9.9%, 13.48%]; Substitute into the second fill light formula:

[0104] (When η = 1, the proportion of green LEDs increases to 35.1%, enhancing object outline recognition, while reducing blue light to 6.52% to reduce scattering interference in foggy weather).

[0105] Optionally, based on the Beer-Lambert Law, α can be calculated by the attenuation difference between 940nm and 850nm infrared light to distinguish the effects of atmospheric attenuation and target reflectivity. The spectral data is converted from the wavelength space to the CIEXYZ color space, and the ΔE value (color restoration error) is calculated, which is directly used as the optimization target of the second fill light formula (ΔE<10). For example, when the multispectral sensor detects a sudden drop in ambient light intensity (such as clouds blocking moonlight), and the ToF sensor measures a sudden increase in target distance from 10m to 20m: The first fill light formula: D 2 From 100 to 400, e -αD Attenuates with increasing distance, I IR Need to be increased to 4 times the original value (assuming α remains unchanged); the second fill light formula: I IR Increased C IR Increase, visible light needs to enhance complementary colors (such as enhancing blue when infrared is warm), through real-time C IR Adjustment This "spectrum-distance" dual trigger mechanism enables the fill light device to always maintain the optimal fill light effect in a dynamic environment.

[0106] exist Figure 3In the embodiment shown, corresponding fill light formulas can be designed for calculation according to the actual working conditions of different devices, and the second fill light intensity of the visible light component can be calculated in combination with the first fill light intensity of the infrared light component, effectively improving the effectiveness of the two fill light intensities.

[0107] Optionally, see Figure 4 , Figure 4 This is a flow chart of another fill light control method provided in an embodiment of the present application. The method may further include steps S310-S340.

[0108] Step S310: collecting historical data under different lighting conditions.

[0109] To determine the fill light factor that affects fill light intensity, a corresponding target model can be trained for processing. Historical data under different lighting conditions can be collected to improve the effectiveness of model training.

[0110] Optionally, lighting conditions may include one or more of light intensity, lighting scenarios, lighting distance, and weather conditions. When selecting historical data, considering that data volume is crucial for model training accuracy and generalization, a minimum of 100,000 sets of historical data can be selected. The lighting conditions for these historical data should cover an illumination range of 0.001-10 lux, with stratified sampling used to ensure a balanced representation of each scenario. This large-scale data set, encompassing a wide range of illumination, can fully train the model to learn the principles of supplemental lighting under different lighting conditions, improving the model's adaptability and accuracy in various real-world scenarios. For example, lighting scenarios can include simulated indoor scenes, such as conference rooms with varying lighting conditions or warehouses with lighting intensities of 0.1-10 lux. Environmental parameters can also be annotated in detail based on the characteristics and needs of each scenario. For outdoor scenes, data can be collected under different weather conditions, times, and environmental types, with meteorological conditions annotated. When collecting data in special environments, different road sections can be selected for rainy and foggy weather, and areas with different orientations and lighting angles can be selected for data collection and annotation in backlit environments. Alternatively, collaboration with security monitoring systems can be used to acquire data from surveillance cameras in different areas. Covering urban streets, residential areas, parking lots and other places, recording environmental parameters and fill light parameters, screening data and annotating the ideal fill light parameter combination. We also collected public computer vision-related data sets, screened data related to fill light and imaging, and re-annotated the data, supplemented it with environmental parameters and ideal fill light parameter combinations, and obtained the corresponding historical data.

[0111] Optionally, historical data can also be subjected to scene classification and labeling processing. In order to make the trained model have wide adaptability, a historical data set covering a variety of scenarios can be constructed. The historical data set can include indoor and outdoor scenes (for example, 40% and 60% each), special environments (such as 15% rain and fog, 10% backlight). The labeling can generate ideal fill light parameters based on the first fill light formula and the second fill light formula, and ensure consistency through cross-validation. At the same time, special environments such as rain, fog and backlight are also considered. These environments pose higher challenges to fill light and imaging. In each scene, environmental parameters can be annotated in detail, such as light intensity, accurately measuring the light intensity distribution of each band in the environment; distance, recording the distance between the camera and the main target object; meteorological conditions, including weather conditions, humidity, visibility and other information. In addition, the ideal fill light parameter combination is also annotated, that is, the infrared and visible light fill light intensity settings that can achieve the best imaging effect for the current scene.

[0112] Step S320: pre-process the historical data to obtain training data.

[0113] Among them, considering that the amount of historical data is large and contains a lot of irrelevant information, in order to improve the effectiveness of the data, various types of preprocessing can be performed on the historical data to obtain training data that can be substituted into the model for training. Preprocessing can include: data cleaning, data normalization, image processing, labeling processing and other processing methods.

[0114] Optionally, considering that the collected historical data may contain noise, outliers, and duplicate data, which will interfere with model training, it is necessary to clean it. The IQR (Interquartile Range) method can be used to remove outliers in environmental parameters such as light intensity, distance, and humidity. For example, if the light intensity is far beyond the normal range, it can be regarded as an outlier and deleted. At the same time, a reasonable threshold is set through MD5 hash deduplication to reduce the impact of redundant data on training.

[0115] Optionally, in order to speed up model training and improve training results, the environmental parameter data in the historical data can be normalized. Mapping data such as light intensity, distance, and humidity to the 0-1 range makes data of different magnitudes comparable. Using the normalization formula Where x is the original data, x min and x max are the minimum and maximum values ​​of the data feature. For example, if the minimum value of the light intensity data is 0.001 lux and the maximum value is 10 lux, for a light intensity value of 5 lux, after normalization, (5-0.001) / (10-0.001)≈0.5.

[0116] Optionally, the collected historical data images can be cropped, scaled, and processed to unify the data format. Based on the model's input requirements, the images can be cropped to a fixed size, for example, scaled to 224×224, normalized to [0, 1], and augmented with random flipping.

[0117] Optionally, historical data can be annotated with environmental parameters and ideal fill light parameter combinations. Environmental parameters include information such as light intensity, distance, and weather conditions. The ideal fill light parameter combination is the infrared and visible light fill light intensity settings that achieve the best imaging effect for the current scene. The annotated data is integrated with the original historical data to form a complete training dataset. For example, in an indoor scene, the annotated light intensity is 2 lux, the distance is 5m, the ideal infrared fill light intensity is 3W, and the visible light fill light intensity is 800 lux. This information is integrated with the corresponding image data.

[0118] It's important to note that after performing various preprocessing steps on historical data, the resulting training data can be partitioned into training, validation, and test sets according to specific ratios. Typically, the training set accounts for 60%-80%, the validation set 10%-20%, and the test set 10%-20%. When partitioning, ensure that the data in each subset is evenly distributed, covering a variety of scenarios and data characteristics. For example, from a dataset containing multiple scenarios, randomly sample data according to the aforementioned ratios to form the training, validation, and test sets, ensuring that each subset includes data from various scenarios, such as indoor and outdoor scenarios, and data under varying lighting conditions.

[0119] Step S330: Determine a reward function shared by the first fill light factor and the second fill light factor.

[0120] The reward function is constructed based on the mapping relationship between the environmental data, the first fill light factor and the second fill light factor. The reward function is used to balance the peak signal-to-noise ratio (PSNR), color reproduction error (ΔE) and fill light power consumption (P consumed The first fill light factor and the second fill light factor are both trained and optimized through the target model, and they share the same reward function R = ω1·PSNR+ω2·ΔC+ω3·(1 / P consumed), when power consumption needs to be reduced (ω3 weight increases), the first fill light factor and the second fill light factor are reduced simultaneously to suppress the dual-band intensity; when color restoration is emphasized (ω2 weight increases), the second fill light factor is increased to enhance visible light compensation, and color deviation can be forced to be corrected even if the first fill light intensity is high. The first fill light factor and the second fill light factor can be used to constrain the energy allocation of the two fill light intensities. The product of the first fill light intensity and the second fill light intensity is limited by the hardware power consumption upper limit (such as total power consumption ≤ 5W). The first fill light formula and the second fill light formula are used to collaboratively achieve the optimal solution of "infrared-visible light energy increase and decrease".

[0121] Step S340: Based on the lightweight model, training is performed in combination with the training data and the reward function to obtain a target model.

[0122] Among them, the lightweight model can be trained according to the obtained training model and combined with the reward function to obtain the corresponding target model.

[0123] Optionally, during model training, a reward function can be defined first. The reward function is used to evaluate the quality of the actions selected by the model and is defined as: R = ω1·PSNR+ω2·ΔC+ω3·(1 / P consumed ), where PSNR is the peak signal-to-noise ratio of image quality, which is used to measure the clarity and quality of imaging. The higher the PSNR value, the smaller the image distortion and the clearer the details. In the fill light process, by adjusting the fill light intensity, maximizing the PSNR value of the imaging helps to improve the visual effect and recognizability of the image. ΔC is the color reproduction error (ΔE<2000 standard), which is used to evaluate the color reproduction accuracy of the imaging. The smaller the ΔE value, the smaller the difference between the color of the imaging and the color of the real scene, and the better the color reproduction effect. In fill light control, try to reduce ΔC to ensure that the color of the image is true and reliable. consumed The power consumption of the fill light reflects the energy consumption of the fill light device during operation. While ensuring image quality, the fill light intensity is optimized to reduce the total power consumption and improve energy efficiency. Weight coefficients ω1, ω2, and ω3 are used to balance the importance of different evaluation indicators in the reward function. The initial settings are ω1 = 0.5, ω2 = 0.3, and ω3 = 0.2, and can be adjusted later based on actual application requirements and training conditions. During training, the model selects actions based on the current state, calculates the reward value based on the reward function, and uses a deep reinforcement learning algorithm to update the model parameters. The validation set is used regularly to evaluate model performance. If performance does not improve, the hyperparameters are adjusted to continue training.

[0124] After defining the reward function, the model training process can be started using the deep reinforcement learning framework (DRL). The state space can be defined as environmental parameters such as spectrum, distance, and weather. These parameters fully describe the environmental state of the fill light device; the action space is the first fill light intensity I IR and the second fill light intensity The model can affect the environment (imaging effect) by selecting different actions (i.e., fill light intensity settings). During the training process, the model can select actions based on the current state, calculate the reward value based on the reward function, and update the model parameters using the deep reinforcement learning algorithm. The Adam optimizer is used for training, the learning rate is set to 1e-4, the batch size is 32, and the training cycle is set to 50. The weight is adjusted every 5 cycles. The specific operations may include: calculating the R value and each sub-indicator (PSNR, ΔE, power consumption) under the current weight on the verification set. If ΔE>10, increase ω2 (+0.1) and reduce ω3 (-0.1); if P consumed If the validation set R value is >5W, increase ω3 (+0.1) and decrease ω1 (-0.1); otherwise, keep the weights stable. Ensure that the weight range is constrained to ω1, ω2, and ω3 ∈ [0, 1], and that ω1 + ω2 + ω3 = 1. To prevent overfitting, an early stopping mechanism is used. If the validation set R value does not improve for 10 consecutive cycles, training is terminated. L2 regularization is used, and the weight decay coefficient is set to 0.001. During training, the validation set is regularly used to evaluate model performance. After each 5-cycle training, the current optimal model parameters (based on the validation set R value) are saved. Finally, from the multiple saved model checkpoints, the model with the best overall performance on the validation set and test set is selected as the target model, and its parameters are saved to a designated file or storage medium for subsequent fill light control.

[0125] It should be noted that after obtaining the target model, its performance in different scenarios can be analyzed. Based on the actual application scenario, the dataset can be divided into multiple scene categories, such as indoor low-light, outdoor rain and fog, and backlight. For each scene category, the difference between the fill light intensity predicted by the target model and the actual ideal fill light intensity is statistically analyzed. For example, metrics such as mean absolute error (MAE) and mean square error (MSE) are calculated. In indoor low-light scenarios, the MAE is calculated between the predicted fill light intensity and the labeled ideal fill light intensity. A large MAE value indicates that the target model's prediction accuracy in this scenario is low. Similarly, in outdoor rain and fog and backlight scenarios, corresponding evaluation metrics are calculated to measure the target model's performance in different scenarios. The target model's evaluation metrics for different scenarios can be visualized graphically, with bar charts showing the MAE values ​​for each scenario. This facilitates intuitive comparison of the target model's performance across different scenarios and allows for quick identification of underperforming scenarios.

[0126] Optionally, for scenes with poor performance, the target model can be retrained by increasing the amount of data. After determining the scenes with poor performance, develop a targeted data collection plan. If the target model performs poorly in outdoor rain and fog scenes, increase the collection of rain and fog scene data under different rain and fog concentrations, different time periods, and different road environments to ensure that the newly collected historical data covers a wider range of changes. Detailed annotations are made to the newly collected historical data, including environmental parameters and ideal fill light parameter combinations, to obtain new training data, which is then integrated with the original training data. Use the integrated data set to retrain the target model, keeping the training algorithm and other hyperparameters unchanged, and observe the performance changes of the target model after retraining. During training, save target model checkpoints regularly for backtracking when training is interrupted or the results are unsatisfactory.

[0127] Optionally, the weight coefficients of the reward function can be adjusted to optimize the performance of the target model. The direction of adjustment of the reward function weight coefficients can be determined based on the actual application requirements. For example, in security monitoring scenarios with extremely high image quality requirements, the weight coefficient ω1 corresponding to PSNR can be appropriately increased. In portable device scenarios with power sensitivity, the weight coefficient ω3 corresponding to the power consumption term can be increased. A gradual adjustment approach can be adopted, adjusting one weight coefficient at a time and observing the performance changes of the target model on the validation set. For example, first increase ω1 by 0.1, retrain the target model, and evaluate the performance. If performance improves, continue fine-tuning; if performance degrades, revert the adjustment and try adjusting other weight coefficients. Through multiple rounds of weight coefficient adjustment and model training, the optimal weight combination is continuously sought, ensuring that the target model achieves optimal performance while meeting the actual application requirements. After each weight coefficient adjustment, a comprehensive evaluation should be conducted on the validation and test sets to ensure the generalization ability of the target model.

[0128] Optionally, the parameters of the optimal target model can be saved for subsequent fill light control. During the training process, the convergence of the model is determined by observing the changes in the performance indicators of the model on the validation set. When the performance indicators (such as MAE, MSE, etc.) on the validation set no longer change significantly over multiple consecutive training cycles, the model can be considered to have converged. From the multiple saved model checkpoints, the model with the best comprehensive performance on the validation set and the test set is selected as the target model. Compare the evaluation indicators of different models in various scenarios and select the model with the best overall performance. Save the parameters of the target model to a specified file or storage medium. For example, store the target model in a processor. In an actual fill light scenario, these parameters can be loaded so that the target model can accurately calculate and output the first fill light intensity and the second fill light intensity based on the data obtained by the sensing device, thereby achieving precise fill light control.

[0129] It should be noted that while the original model (such as ResNet-50) has powerful performance in image recognition and processing, it has high computational complexity and requires high hardware computing power, making it unsuitable for direct deployment on edge devices. Therefore, this application uses a lightweight model, such as a lightweight MobileNetV3, for computational processing. Knowledge distillation technology can be used to transfer the knowledge of a large teacher model (such as ResNet-50) to a small student model, MobileNetV3. For example, the knowledge distillation temperature T = 4, the loss function is 0.5×soft cross entropy + 0.5×hard cross entropy, and quantization is INT8. The MobileNetV3 model has a lightweight structure and low computational complexity, and can maintain high model performance while meeting the computing power requirements of edge devices (≤1TOPS). During the knowledge distillation process, a specific loss function is constructed to enable MobileNetV3 to learn the output results of ResNet-50. For example, MobileNetV3 is allowed to imitate ResNet-50's judgment method on the relationship between various environmental parameters (such as spectrum, distance, meteorology, etc.) and fill light intensity, significantly reducing the complexity of the model while ensuring certain performance. ResNet-50 has a large network structure and numerous parameters, resulting in high computational complexity. MobileNetV3, on the other hand, uses knowledge distillation to retain only key feature representations and computational paths, significantly reducing model size. The MobileNetV3 model itself incorporates a series of lightweight designs, such as the depthwise separable convolution architecture. Traditional convolution performs a full convolution on every channel of every pixel when processing images, which is computationally intensive. Depthwise separable convolution, on the other hand, decomposes the convolution operation into depthwise convolution and pointwise convolution. Depthwise convolution only performs convolution on each channel individually, while pointwise convolution integrates channel information. This separable convolution approach significantly reduces computational complexity. For example, for an image with a resolution of m×n and the number of channels c, the computational complexity of traditional convolution is approximately m×n×c×k×k×c (k is the kernel size), while the computational complexity of depthwise separable convolution is approximately m×n×c×k×k+m×n×c×c, significantly reducing the computational complexity. In addition, MobileNetV3 also introduces the SE (Squeeze-and-Excitation) module, which models the relationship between channels and adaptively adjusts channel weights, improving model performance without increasing excessive computational complexity, thereby further optimizing the model structure.

[0130] Optionally, the model's data precision can be adjusted by quantizing it from a higher precision (such as 32-bit floating point) to a lower precision (such as 8-bit integer). While quantization sacrifices a small amount of precision, it significantly reduces memory usage and computational complexity. For example, if a parameter occupies 4 bytes of memory when represented as a 32-bit floating point, quantizing it to an 8-bit integer will only take up 1 byte, reducing the memory footprint by a factor of four. Integer calculations are faster and require fewer hardware resources than floating-point calculations. This makes the model more efficient when storing and processing data, laying the foundation for reducing real-time inference latency. After implementing these lightweight methods, the computational complexity of the MobileNetV3 model is significantly reduced, significantly shortening the computation time required for real-time inference. For example, on the Jetson Nano platform, a ResNet-50 model performing inference on a set of environmental parameter data may require billions of floating-point operations, resulting in a significant delay of up to 150ms. After knowledge distillation, structural optimization, and quantization, the MobileNetV3 model reduces the amount of computation to a fraction of the original amount or even lower, requiring only tens of millions of operations, significantly shortening the inference time and reducing the real-time inference latency to less than 20ms. The lightweight model not only reduces the amount of computation, but also significantly reduces memory usage. During the actual inference process, the target model needs to read data from the memory for processing. Less memory usage means faster data reading speed. For example, when the processor reads the parameters and input data of the MobileNetV3 model from the memory, due to model quantization and structural optimization, the amount of data becomes smaller, and the reading time is shortened from tens of milliseconds to a few milliseconds. In addition, during data transmission, such as from the storage module to the computing core, the reduction in the amount of data also greatly reduces the transmission time, further speeding up the inference speed and effectively reducing the real-time inference latency.

[0131] For example, in security monitoring scenarios, when people or objects move quickly, the fill light device needs to adjust the fill light intensity according to environmental changes in a very short time to ensure clear and accurate images are captured. Due to the lightweight design of the MobileNetV3 model, the real-time inference latency can meet this requirement. For example, in a monitoring area with frequent entry and exit, when someone walks quickly, the fill light device can complete the analysis of environmental parameters and the decision of the fill light strategy within 20ms, and promptly adjust the fill light intensity of infrared and visible light to ensure that the captured image of the person is clear and details such as facial features and clothing color are accurately presented, effectively improving the monitoring effect and image quality.

[0132] exist Figure 4In the embodiment shown, training can be performed based on data under different lighting conditions, thereby improving the effectiveness of the fill light factor determined based on the target model, thereby improving the effectiveness of the fill light intensity, and improving the adaptability and performance of the fill light device in complex scenes. The use of a lightweight model for processing effectively reduces the amount of computation and cost required for training and calculation, improves the processing speed of the model, and thus improves the efficiency of the fill light device in filling light.

[0133] Optionally, see Figure 5 , Figure 5 A detailed flow chart of step S200 is provided in an embodiment of the present application. Step S200 may include steps S210-S220.

[0134] Step S210 : Processing the first fill light intensity based on the signal conversion to obtain a first control signal for controlling the fill light of the infrared light component.

[0135] Considering the differences between the infrared and visible light components, when performing fill light control, the infrared and visible light components can be controlled separately based on their device characteristics and corresponding fill light intensities. The first fill light intensity can be processed using signal conversion technology to directly obtain a first control signal for fill light control of the infrared component.

[0136] Optionally, an analog voltage signal or a digital signal + DAC conversion control algorithm can be used to convert the first fill light intensity into a first control signal. The signal range of the first control signal can be 0-3.3V (corresponding to power 0-10W). The output of the digital potentiometer is controlled by the processor's DAC channel or SPI / I2C bus, and converted into the current of the infrared light component through the constant current source drive circuit to achieve continuous power adjustment.

[0137] Step S220 : Processing the second fill light intensity based on pulse width modulation to obtain a second control signal for performing fill light control on the visible light component.

[0138] The second fill light intensity may be processed by pulse width modulation technology to obtain a second control signal for controlling the fill light of the visible light component.

[0139] Optionally, a PWM (pulse width modulation) signal can be used as the second control signal for control. The second control signal is a digital pulse type with a default frequency of 500Hz (dynamically adjustable within the range of 200Hz-1kHz) and an adjustable duty cycle of 0%-100% (resolution ≥10 bits). The processor can generate a second control signal through a built-in PWM timer module, which is amplified by the driver chip to control the MOSFET switch to achieve precise adjustment of the brightness of the visible light component. Each visible light band (such as red / green / blue / white) has an independent PWM channel, which supports single-band independent adjustment or full-band synchronous control.

[0140] It should be noted that, in order to achieve synchronous coordinated fill light, the first control signal and the second control signal share a common clock source.

[0141] Optionally, the first control signal can be updated in a 50ms cycle, triggered by real-time data collected by the sensor data. The first control signal and the second control signal share the system clock to ensure synchronous adjustment. For close-range low-light scenes (D<5m, light intensity<0.1lux): infrared power ≤2W, visible light PWM duty cycle ≥60%, with visible light as the main fill light; for long-range extremely dark scenes (D>20m, light intensity<0.01lux): infrared power ≥8W, visible light PWM duty cycle ≤30%, switch to infrared dominant mode, and visible light is only used for color correction. When the total power consumption of the fill light device is close to the hardware upper limit (such as 5W), the intensity of the first control signal and the second control signal can be automatically compressed proportionally to ensure Multiple protection mechanisms are also built in. For example, the duty cycle of the second control signal is monitored in real time, with a tolerance of ±5%. A hardware reset is triggered when an anomaly occurs. If the infrared light component's current exceeds 3.5A, a hardware comparator is used to forcibly shut down the signal and report a fault. This application overcomes the limitations of existing technologies that rely on single signal types or fixed parameter control. By integrating visible light PWM with infrared analog signals in dual modes, it achieves dynamic control across multiple dimensions: spectrum, distance, and power consumption. According to the first and second fill light formulas, the first fill light intensity is adjusted based on factors such as the target signal-to-noise ratio, distance, and atmospheric attenuation coefficient. An increase in the first fill light intensity indicates an increase in infrared light content in the environment, potentially affecting the color balance of the image. To maintain good color reproduction, the processor adjusts the second control signal of the visible light component accordingly. If the first fill light intensity increases with increasing target distance, the PWM duty cycle of the blue and green visible light components is appropriately increased to adjust the color balance and ensure natural and accurate image colors. Compared with the second-level response delay of traditional fill light technology, this solution shortens the control cycle to milliseconds (50ms). Through multi-channel independent adjustment and color balancing algorithms, it significantly improves the fill light accuracy and imaging quality in complex environments, providing a more efficient low-light full-color imaging solution for scenarios such as smart security and computer vision.

[0142] exist Figure 5 In the illustrated embodiment, the fill light control of the components can be performed accordingly according to the actual control logic of the device and in combination with the fill light intensity, thereby effectively improving the efficiency of the coordinated fill light performed by the fill light device and the fill light effect.

[0143] Optionally, see Figure 6 , Figure 6 A detailed flow chart of step S220 is provided in an embodiment of the present application. Step S220 may include steps S221-S224.

[0144] Step S221: Determine the operating frequency of the pulse width modulation according to the application scenario.

[0145] The pulse width modulation operating frequency can be determined based on the actual application scenario. For example, a default frequency of 500Hz is used to balance the human eye's persistence of vision and the hardware load. This frequency can be switched in real time in special scenarios. For example, in high-speed dynamic scenes (such as shooting moving objects), the operating frequency can be increased to 1kHz to reduce flicker. In low-power scenarios (such as battery-powered), the operating frequency can be reduced to 200Hz to reduce the energy consumption of the drive circuit.

[0146] Optionally, the operating frequency can be adjusted by the processor modifying the clock division coefficient of the PWM timer, with a response time of ≤1ms. The PWM operating frequency determines the smoothness and visual quality of the visible light component's brightness changes. Taking into account the human eye's persistence of vision and hardware performance, the PWM frequency is set between 200Hz and 1kHz. When the frequency is lower than 200Hz, the human eye may perceive LED flicker, affecting imaging stability and visual comfort. While a frequency that is too high can further reduce the perception of flicker, it increases the burden on the hardware driver circuit, resulting in increased power consumption and possible electromagnetic interference. Extensive experimental verification has shown that a frequency of around 500Hz provides the best balance between flicker suppression and hardware performance. This ensures a smooth transition of visible light component brightness changes without placing excessive stress on the hardware, ensuring stable control of visible light brightness under varying lighting conditions.

[0147] Step S222: Obtain an adjustable duty cycle of pulse width modulation based on the linear mapping conversion of the second fill light intensity.

[0148] A linear mapping conversion process may be performed based on the second fill light intensity to obtain an adjustable duty cycle of pulse width modulation.

[0149] For example, the second fill light intensity calculated by the second fill light formula can be Converted to PWM adjustable duty cycle through linear mapping (in ). For example, when The duty cycle is automatically adjusted to 75%.

[0150] Step S223 : determining duty cycle offsets of multiple bands of the visible light component based on the adjustable duty cycle and the spectral information.

[0151] The adjustable duty cycle and the spectral information can be combined to determine the duty cycle offsets of the multiple bands corresponding to the visible light component. Based on the spectral information, the processor can calculate the duty cycle offset (ΔD) of each band to achieve dynamic color correction: For example, in sodium lamp lighting scenarios, when blue light is missing, the blue light channel duty cycle is automatically increased by 40%-60% to compensate for spectral deviation and ensure that the color reproduction error ΔE<10.

[0152] Step S224: Determine a second control signal according to the duty cycle offset.

[0153] A corresponding pulse width modulation signal may be determined according to the duty cycle offset as the corresponding second control signal.

[0154] Optionally, the duty cycle can be adjusted according to the ambient light, and the multispectral sensor monitors the spectral distribution and intensity information of the ambient light in real time. When the ambient light is dim, in order to ensure the brightness and color reproduction of the imaging, the brightness of the visible light component needs to be increased. At this time, the processor calculates the required second fill light intensity through the second fill light equation based on the data from the multispectral sensor, and at the same time derives the color contribution (C IR ), thereby increasing the duty cycle of the PWM signal in the second control signal. In a low-illuminance environment (such as an indoor scene with 0.1lux), the duty cycle may be increased to 80%-90% after calculation, so that the visible light component emits light at a higher power, ensuring that the image is clear, bright and the color is accurate. On the contrary, when the ambient light is strong, in order to avoid excessively bright imaging resulting in loss of details and color distortion, the duty cycle of the PWM signal in the second control signal will be reduced. In bright outdoor daytime scenes, the duty cycle may be reduced to 10%-20%, reducing the luminous intensity of the visible light component to ensure that the imaging effect conforms to the actual environment. It can be dynamically adjusted in combination with imaging requirements. In scenes where color details need to be highlighted, such as shooting color logos, artworks, etc., in order to achieve the best color reproduction effect, it can be adjusted according to the color reproduction requirements C req With infrared contribution C IR The processor fine-tunes the PWM duty cycle to detect differences in image saturation. If the saturation of certain colors in the image is insufficient, the processor can appropriately increase the PWM duty cycle of the corresponding LED to enhance the intensity of that color, making the image more vivid and rich, while ensuring a low color reproduction error (ΔE) of <10. When capturing dynamic scenes, the PWM dimming response speed can be optimized to ensure real-time imaging and stability, and to avoid smearing caused by delayed LED brightness changes. By adjusting the drive circuit parameters, the PWM signal can quickly respond to the processor's control commands. When objects move quickly, the brightness of the visible light component can be quickly adjusted to ensure clear, smear-free dynamic images.

[0155] exist Figure 6 In the illustrated embodiment, the second control signal capable of implementing dynamic color correction can be determined based on the actual scene and combined with the fill light intensity, thereby achieving dynamic fill light adjustment of the visible light component and further optimizing the fill light effect of the fill light device.

[0156] It should be noted that the sensing device includes distance sensors that cover close-range, macro-range, and long-range monitoring ranges, as well as multispectral sensors that cover the full spectrum of ultraviolet, visible, and near-infrared light. In other words, the sensing device can include distance sensors capable of monitoring multiple distances and multispectral sensors covering multiple different spectral bands, effectively improving the effectiveness of the environmental data collected by the sensing device.

[0157] Optionally, see Figure 7 , Figure 7 A detailed flowchart of step S110 is provided in an embodiment of the present application. Step S110 may include steps S111-S113.

[0158] Step S111: acquiring initial distance data collected by the distance sensor.

[0159] Step S112: Compensate the initial distance data to obtain the target distance.

[0160] After obtaining the initial distance data collected by the distance sensor, the initial distance data may be compensated to obtain the desired target distance in order to improve the accuracy of the data, taking into account the influencing factors existing during detection.

[0161] For example, a distance sensor can include a ToF distance sensor with a range of 0.1-50m, covering a variety of detection distances with a detection accuracy of up to ±0.1m. ToF (Time of Flight) technology calculates the distance of a target object by measuring the time it takes for a light pulse to be transmitted and received. Regarding measurement parameter settings, the measurement frequency can be determined and adjusted based on the target object's motion state. In static scenarios, such as monitoring fixed objects indoors, to reduce power consumption and data processing, the measurement frequency can be set to 1-3Hz, ensuring stable distance acquisition without overloading the system. In dynamic scenarios, such as traffic intersections or public places with frequent traffic, to capture distance changes in real time, the measurement frequency should be increased to 10-20Hz. This allows the fill light device to respond promptly to target movement and adjust the fill light to ensure clear imaging. When measuring at close range (0.1-5m), to avoid saturation at the receiver, the transmitted optical power can be reduced, using a lower power setting to minimize energy consumption and interference. For long-distance measurements (30-50m), the transmitted optical power needs to be increased to a higher level to ensure effective propagation and reflection of the optical pulses for accurate distance measurement. For data processing, a combination of median filtering and Kalman filtering can be used to improve data accuracy. Median filtering removes sudden outliers, sorts continuously collected data and takes the median value, suppressing spike noise caused by environmental interference. Kalman filtering, based on a target motion model, uses historical distance data to predict the current distance and optimizes it with the latest measured values. This method is highly effective for tracking the distance of dynamic targets. Working together, these two methods can smooth and accurately generate initial distance data, providing a reliable basis for fill light control.

[0162] For example, the ToF distance sensor can be calibrated regularly to ensure accuracy. Using a standard reflector of known distance, measurements are taken at different distances and compared with the actual distance. The resulting calibration coefficient is stored in the device memory, and the measurement data is corrected in real time to ensure that the error is within the range of ±0.1m. At the same time, environmental factors such as temperature, humidity, and atmospheric particulate matter concentration can affect the speed of light propagation, thereby affecting measurement accuracy. Therefore, by using built-in temperature and humidity sensors and atmospheric particulate matter monitoring modules to monitor environmental parameters in real time, the measurement data can be compensated using pre-established environmental parameters and measurement error models. In harsh environments such as high temperature, high humidity, or haze, the measurement results can be automatically adjusted based on the environmental parameters to improve the accuracy of the target distance.

[0163] Optionally, when performing fill light control, the fill light intensity can be automatically adjusted based on the first fill light formula and the second fill light formula according to the measured target distance. When the distance is close, the fill light intensity is reduced to prevent overexposure of the image; when the distance is far, the fill light intensity is increased to ensure that the target is fully illuminated. For example, when the distance is 10m, the first fill light intensity and the second fill light intensity are adjusted accordingly based on the calculation to achieve the best imaging effect. In addition, the fill light mode can be switched based on the ambient light conditions and distance information. For example, when the ambient light is dark and the target distance is close, visible light is used as the main fill light to ensure color reproduction; when the ambient light is extremely dark and the target distance is far, the proportion of infrared fill light is increased, or even switched to pure infrared fill light mode to achieve covert fill light and ensure clear imaging.

[0164] Step S113: Acquire spectral information obtained by the multispectral sensor based on environmental detection.

[0165] The spectral information detected by the multispectral sensor based on the current environment can be used as the final collected environmental data.

[0166] Optionally, the raw data collected by the multispectral sensor can be subjected to median filtering to remove impulse noise, and then the main spectral components (such as the proportion of red light and the amount of blue light missing) can be extracted through principal component analysis to obtain the corresponding spectral information.

[0167] Optionally, the multispectral sensor can cover the 300-1600nm band and has a high resolution of ±5nm. It can accurately perceive the spectral information in the environment, including the intensity distribution of light of different wavelengths. By analyzing the spectral data, the lighting characteristics of the environment can be obtained, such as the ratio of infrared light and visible light in the ambient light, whether there is interference from special light sources, etc., which provides an important basis for subsequent fill light control and provides rich and accurate spectral information for fill light adjustment. In terms of fill light adjustment based on ambient light intensity, when the ambient light intensity is extremely low, such as in a dark room (light intensity less than 0.1lux) or outdoors late at night, the light intensity value detected by the multispectral sensor is far lower than normal. At this time, to ensure clear imaging, the fill light device will significantly increase the fill light intensity. For example, the power of the infrared light component is increased to 8-10W to provide basic lighting, and the brightness of the visible light component is adjusted to 1500-2000lux to improve color reproduction. As ambient light intensity gradually increases but still falls short of the required brightness for normal vision (e.g., 2-5 lux on a cloudy day), the fill light device appropriately reduces the fill light intensity according to the first and second fill light formulas based on the spectral information fed back by the multispectral sensor. For example, the infrared light component power is adjusted to 3-5W, and the visible light component brightness is adjusted to 500-1000 lux to avoid imaging issues. When the ambient light intensity reaches normal levels (e.g., during daytime when the light intensity is greater than 10 lux outdoors in bright sunlight), if the light intensity exceeds 50 lux, the multispectral sensor detects sufficient light intensity and significantly reduces the fill light intensity or even stops it. For example, it turns off the infrared and visible light components, relying on ambient light for imaging, saving energy and avoiding fill light interference. The fill light color can be adjusted based on spectral composition. For example, in a sunset scene, the multispectral sensor detects that the red spectral component in the ambient light accounts for 70%-80%, while the blue and green spectral components are relatively weak, accounting for 10%-15% and 10%-15%, respectively, resulting in a reddish image. To correct for color deviation, the fill light device can increase the brightness of the blue and green LEDs in the visible light module by 30%-50% by increasing their PWM duty cycle, while also appropriately reducing the primary fill light intensity. In road scenes illuminated by sodium lamps, blue light only accounts for 5%-10% of the lamp's spectrum, far below the normal spectral distribution. The fill light device can specifically increase the PWM duty cycle of the blue LEDs in the visible light module by 40%-60% to compensate for the missing spectral components and enhance image color. Considering dynamic adjustment over time, for example, during early morning sunrise, the ambient light intensity gradually increases from near darkness, and the spectral composition also changes. Half an hour before sunrise, the ambient light intensity is between 0.01-0.1 lux, with infrared light accounting for approximately 80% of the spectrum. At this time, the infrared module power is maintained at 5-8W, while the visible light module brightness is between 200-500 lux.As the sun rises, the fill-light device can proactively adjust its fill-light strategy based on the spectral information detected by the multispectral sensor. For example, it can gradually reduce the power of the infrared light component and adjust the spectral distribution of the visible light component. At night, while the ambient light intensity and spectral composition are relatively stable, they change when clouds block moonlight. When cloud cover reduces the light intensity from 1-2 lux to 0.1-0.5 lux, the proportion of blue light in the spectrum drops by approximately 20%-30%. The multispectral sensor detects these changes in real time, and the fill-light device immediately increases the fill-light intensity. For example, it increases the power of the infrared light component to 3-5W and increases the brightness of the blue LED in the visible light component, ensuring that imaging quality is not affected and that monitoring and imaging requirements are always met.

[0168] Optionally, the ToF distance sensor can also be integrated with a multispectral sensor to fuse the acquired distance information with data such as ambient light intensity and spectral composition detected by the multispectral sensor. For example, in low illumination and when the target is close, visible light fill light is prioritized; if the target is far away, the infrared fill light intensity is appropriately increased. This allows for intelligent and precise fill light adjustment, allowing the fill light device to more accurately adjust the fill light intensity and method based on the target distance and spectral information.

[0169] exist Figure 7 In the illustrated embodiment, a device with a wide detection range can be used to collect data, which effectively improves the authenticity and accuracy of the environmental data, thereby improving the effectiveness of the fill light intensity determined based on the environmental data.

[0170] See also Figure 8 , Figure 8 This is a structural diagram of a monitoring system provided in an embodiment of the present application. The monitoring system may include: a processor 400 and a fill light device 500;

[0171] The processor 400 and the fill light device 500 can be connected through a driving circuit. The processor 400 is used to determine the fill light intensity of the fill light device 500 based on the collected environmental data; and control the fill light of the fill light device 500 according to the fill light intensity; wherein the fill light device 500 includes an infrared light component 510 and a visible light component 520.

[0172] For example, the processor 400 can be connected to the fill light device 500 via the SPI bus to transmit the fill light control signal. The first control signal for fill light control by the infrared light component 510 and the second control signal for fill light control by the visible light component 520 in the fill light device 500 share a common clock source, and the synchronization accuracy of the clock source is ≤10ns to achieve synchronous collaborative fill light.

[0173] In an optional embodiment, in the fill light device 500, the infrared light component 510 and the visible light component 520 are packaged on the same substrate with the same aperture, and the infrared light emitted by the infrared light component 510 and the visible light emitted by the visible light component 520 are emitted through the same aperture;

[0174] The wavelength range of the infrared light component 510 is 940nm±10nm, and the wavelength range of the visible light component 520 is 400nm-700nm;

[0175] The infrared light component 510 includes a plurality of infrared light LEDs, and the visible light component 520 includes a plurality of visible light LEDs;

[0176] Multiple infrared LEDs and multiple visible light LEDs are staggered and arranged in a matrix structure or a ring shape;

[0177] The light emission centers of all infrared LEDs and visible light LEDs are in the same plane and have the same emission direction. The deviation of the emission optical axes of infrared and visible light is less than or equal to 1°, and the flatness error of the light emission centers of infrared and visible light is less than or equal to 0.05mm.

[0178] For example, infrared light assembly 510 can be a near-infrared LED array, or infrared LED array, comprised of multiple near-infrared LEDs with a wavelength of 940nm ± 10nm. This wavelength selection effectively avoids red-blurring issues, ensuring that visible light leakage does not reveal the camera's location during fill-lighting. Visible light assembly 520 can be a visible light LED array. For example, the visible light LED array can include 12-24 visible light LEDs, while the infrared light LED array can include 8-16 infrared LEDs. For short-range fill-light scenarios, such as indoor surveillance, a configuration of 12 visible light LEDs and 8 infrared LEDs is recommended. For large-scale outdoor surveillance, a combination of 24 visible light LEDs and 16 infrared LEDs is recommended to ensure rich color information while meeting the fill-light requirements of various scenarios. Specifically, the visible light LEDs and infrared light LEDs are arranged in a staggered matrix on the same substrate of the fill-light device 500. Taking a two-dimensional plane as an example, the horizontal and vertical spacing is arranged at a certain distance (such as a spacing of x millimeters), and the visible light LEDs and infrared light LEDs appear alternately in the matrix. For example, a 3×5 staggered matrix is ​​used on the same plane, with the horizontal spacing of the LED beads being 15mm, the vertical spacing being 20mm, and the offset spacing between two adjacent rows of beads being 7.5mm. On the same substrate of the fill light device 500, the visible light LEDs and infrared light LEDs are arranged in a staggered manner to form a matrix structure. This distribution method ensures that each visible light LED is surrounded by an infrared light LED, and vice versa, ensuring that visible light and infrared light are emitted simultaneously in the same area.

[0179] For example, multiple infrared LEDs and multiple visible light LEDs can be stacked and nested in a ring shape to achieve a ring distribution. With the center of the light-emitting surface of the substrate as the center of the circle, multiple concentric rings are designed, and visible light LEDs and infrared light LEDs are alternately distributed on different rings. A certain number of visible light LEDs are arranged in the innermost ring, infrared LEDs are arranged in the adjacent outer ring, and visible light LEDs are arranged in the outer ring, and so on. The number of LEDs on each ring is reasonably set according to the circumference of the ring to ensure that the light is evenly distributed in the circumferential direction. For example, 8 visible light LEDs are arranged in a ring with an inner radius of 5 mm, and 12 infrared LEDs are arranged in a ring with an outer radius of 10 mm, and the spacing between adjacent rings is 2 mm to ensure the uniformity of the light in the radial direction. Taking into account the LED light-emitting angle (typical value is 120°) and the superposition effect of the light spot, it is recommended that the number of rings does not exceed 5 layers, and the spacing between single-ring LEDs is controlled in the range of 3-5 mm.

[0180] Optionally, the infrared light component 510 is designed to have an adjustable power range of 0-10W, enabling flexible adjustment of the infrared fill light intensity based on actual scene requirements to meet fill light requirements at different distances and lighting conditions. The visible light component 520 has a wavelength range of 400-700nm, which encompasses all colors visible to the human eye and can provide rich color information for imaging. The brightness of the visible light component 520 is set to the middle value (1000lux) in the range of 0-2000lux, entering the standby state.

[0181] Optionally, the visible light LED array and infrared light LED array share the same substrate and packaging structure, achieving a co-aperture package. The substrate is made of a material with good thermal conductivity and mechanical strength (such as aluminum), providing a stable mounting base for the two LED arrays. The two LEDs can be soldered to the substrate using surface mount technology (SMT). Their positioning accuracy is strictly controlled. During installation, the deviation between the emission optical axes of infrared and visible light is maintained within 1°, and the flatness error of the emission center is maintained within 0.05mm. This ensures that the emission centers of all LEDs are in the same plane and have consistent emission directions. During packaging, a unified optical lens or diffuser is used to cover the entire array, ensuring that infrared and visible light are emitted from the same aperture. The lens design takes into account the wavelength characteristics of both light sources, providing excellent transmittance and collimation for both visible and infrared light, reducing light scattering and deviation during emission. Because the two LEDs are closely adjacent in the array structure and their emission positions are aligned, the infrared and visible light originate from the same location and travel the same path during transmission. This ensures that the illumination areas of the two types of light are substantially aligned when they reach the surface of the illuminated object, reducing light propagation deviation caused by different exit positions. For example, when the fill light device 500 is 5 meters away from the illuminated object, the center deviation of the illumination spots of the visible light and infrared light is less than 2 mm, thereby achieving uniformity and accuracy of the fill light.

[0182] Alternatively, the two LED arrays can be integrated into a single unit using a common-aperture packaging structure, reducing the size of the fill light device 500. Compared to traditional separate visible and infrared fill light modules, the fill light device 500 of this application is 40% smaller and 60% more integrated. Furthermore, the compact structure makes the fill light device 500 easier to install, allowing it to be directly integrated into various cameras, monitoring devices, and other systems without requiring additional installation space or complex calibration procedures.

[0183] For example, the processor 400 can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The various methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor.

[0184] Optionally, in terms of hardware, a SoC (system-on-chip) with an integrated NPU (neural network processing unit) can be used as processor 400. The computing power requirement is ≥1TOPS. This processor incorporates specialized neural network acceleration hardware, enabling efficient AI model execution and rapid image data processing and fill-light strategy decision-making. The NPU enables processor 400 to achieve higher computational efficiency and lower power consumption than traditional CPUs when processing complex deep learning algorithms, meeting real-time and low-power requirements. Memory requirements include ≥2GB of RAM, supporting multispectral data caching. Since multispectral sensors collect large amounts of spectral data in real time, sufficient memory is required to cache this data for subsequent analysis and processing by processor 400. Furthermore, during model execution, a certain amount of memory is required to store intermediate calculation results and model parameters. 2GB of RAM ensures smooth and stable data processing, preventing data loss or processing interruptions due to insufficient memory. Optical devices utilize coated filters with an infrared transmittance of ≥90% and a visible light cutoff of ≥99%. Coated filters screen and filter light, ensuring that only specific wavelengths of light pass through. High infrared transmittance ensures that infrared light can effectively propagate to the target object and reflect back, improving the effect of infrared fill light; high visible light cutoff rate avoids the interference of visible light in the environment on infrared imaging, ensuring the accuracy and stability of infrared imaging.

[0185] The monitoring system's operating temperature range is optionally set to -30°C to 70°C. In low-temperature environments, such as -30°C, the performance of electronic components can change, potentially leading to slower response speeds, increased power consumption, or even failure. To address this, the hardware design incorporates specialized low-temperature-compatible materials and circuit designs. For example, components such as capacitors and resistors with excellent low-temperature characteristics are selected to ensure stable capacitance and resistance values ​​at low temperatures, ensuring proper circuit operation. Furthermore, an optimized heat dissipation structure ensures that the system quickly reaches normal operating temperature in low-temperature environments, avoiding startup errors caused by excessively low temperatures. In high-temperature environments, such as 70°C, excessive heat can cause electronic components to generate excessive heat, impacting their lifespan and performance. Therefore, efficient heat dissipation measures, such as heat sinks and thermal grease, are employed to quickly dissipate heat, keeping the temperature of key components, such as chips, within normal ranges and ensuring stable system operation. The system achieves an IP67 rating, ensuring dust and water resistance, ensuring complete protection against dust intrusion and resisting the harmful effects of water even when briefly immersed in water at a certain depth. The structural design utilizes sealing features such as rubber seals and waterproof vents to effectively isolate the system from the external environment. Sealing rubber seals are installed at various interfaces and gaps in the system to prevent dust and moisture from entering, while waterproof vents balance the air pressure inside and outside the system while preventing moisture from seeping in. In practical applications, whether in dusty industrial environments or rainy outdoor locations, the system's internal electronic components are protected from external environmental corrosion, extending system life and improving reliability.

[0186] Optionally, to fully verify the performance of the fill light control method, comparative experiments can be conducted in a variety of representative scenarios. The experiments use standardized processes to quantitatively analyze core indicators such as color error (ΔE), image clarity (PSNR), power consumption, and dynamic range. Specific methods may include:

[0187] Experimental Design and Measurement Standards: Definition and Measurement Principle of Color Error (ΔE): ΔE is an indicator that measures the difference between imaged colors and real scene colors. The smaller the value, the more accurate the color reproduction. This application uses the CIE1976Lab color space to calculate ΔE; Where L is the brightness difference (ranging from 0-100, with 0 being black and 100 being white); a is the difference between the red and green axes (positive values ​​indicate redder, negative values ​​indicate greener); and b is the difference between the blue and yellow axes (positive values ​​indicate yellower, negative values ​​indicate bluish). A true color benchmark can be established using an X-Rite ColorChecker standard color chart. In the experiment, a spectrophotometer was used to measure the Lab values ​​of each color block on the color chart. Then, an imaging device was used to capture the image of the color chart after supplemental lighting. The image was converted from RAW format to sRGB and mapped to Lab space. The pixel-by-pixel ΔE was calculated and averaged. For example, under standard daylight conditions, the ideal ΔE should be within 10, with smaller values ​​indicating closer color reproduction to the real scene. Multimodal data acquisition is simultaneous: The target distance measured by the distance sensor and spectral information measured by the multispectral sensor can be simultaneously recorded, such as the full-band light intensity distribution from 300-1600nm (including the infrared light intensity attenuation rate at 940nm), ambient light intensity (lux), and temperature and humidity parameters. This ensures that the input conditions of the supplemental lighting solution are consistent with the actual scene. Comparison solution settings: The existing solution uses 850nm infrared fill light (with red exposure) or single visible light fill light (fixed power 5W), without a dynamic adjustment mechanism.

[0188] Experimental conditions and data collection of typical scenarios: Indoor low-light scene (light intensity 0.1lux, distance 5m): The environment is characterized by extremely weak light, and it is necessary to balance infrared concealment and visible light color reproduction; traditional data collection solution: 850nm infrared fill light is fully turned on, power consumption is 5.2W, imaging is reddish, and the ΔE of the red block of the color card reaches 28.3; this application solution: infrared power is 2.1W (calculated and determined by the first fill light formula), visible light PWM duty cycle is 80% (the second fill light formula compensates for the lack of blue), power consumption is 2.1W, ΔE = 8.2. Outdoor rain and fog scene (light intensity 0.5lux, distance 30m, α=0.3 / m): The environmental characteristics are strong atmospheric attenuation (α=0.3 / m), and spectral scattering causes color blur; traditional data acquisition solution: only visible light fill light (power 5W), due to scattering, it cannot be focused, the imaging is grayscale, and ΔE cannot be calculated; the solution of this application: infrared power 7.4W (the first fill light formula corrects for atmospheric attenuation), visible light enhances green light 35.1%, and the second fill light formula improves contrast), power consumption 4.8W, ΔE=12.7. Backlit scene (light intensity 500 lux, distance 10m, red light accounts for 70%): The environmental characteristics are strong background light, which leads to insufficient exposure of the dark part of the target, and the traditional solution is prone to overexposure; traditional data acquisition solution: dynamic range <80dB, ΔE of the white block of the bright color card>50 (overexposure distortion); this application solution: the visible light duty cycle is reduced to 15% (the second fill light formula suppresses overexposure), the infrared power is 1W (the first fill light formula enhances the dark part), the dynamic range is increased to 120dB, ΔE=9.1.

[0189] Standardization of experimental data: Error correction: The raw data can be subjected to median filtering to remove impulse noise, distance data can be optimized through Kalman filtering, and the atmospheric attenuation coefficient α can be inverted based on the Beer-Lambert law to ensure data accuracy; Normalization comparison: Indicators such as power consumption and dynamic range are uniformly converted to the same environmental conditions (such as light intensity of 0.1 lux and distance of 10m) to eliminate the interference of scene differences on the results; Statistical significance: Each scene is tested 20 times, the median is taken, and the standard deviation is calculated to ensure the reliability of the experimental results (such as the standard deviation of ΔE in indoor scenes is <0.5).

[0190] Through the above experimental design, the advantages of this application in terms of color reproduction accuracy, power consumption control, and adaptability to complex environments have been systematically verified, providing a scientific basis for subsequent performance comparisons. Therefore, the experimental results (one of which is used for illustration) are as follows: Indoor low-light environment (0.1lux): In indoor low-light environment, at a distance of 5m, the color error (ΔE) of the traditional fill light solution reaches 28.3, which means that there is a large deviation between the color of the image and the color of the real scene, and the image quality is poor. However, this application reduces the color error to 8.2 through the synergistic effect of the fill light device and the fill light control method, effectively improving the accuracy of color reproduction and making the image more realistically reflect the color information of the scene. At the same time, the power consumption of the fill light device of this application is only 2.1W, which is about 60% lower than the power consumption of more than 5W in the normal on mode of the traditional fill light, significantly improving energy utilization efficiency. While ensuring imaging quality, it is more suitable for battery-powered and other scenarios with strict requirements on power consumption.

[0191] In addition, considering the outdoor rain and fog environment (for example, 5 meters away), it poses a great challenge to fill light and imaging. Due to severe environmental interference, the traditional fill light solution cannot achieve full-color imaging and can only output grayscale images, losing a lot of color information. This seriously affects the monitoring effect in scenes where the color and logo of objects need to be identified. However, this application relies on the precise perception of the environmental spectrum by the multi-spectral sensor and the accurate measurement of the distance by the distance sensor, combined with the fill light collaborative control algorithm, it can still maintain a color error of 12.7 in the same rain and fog environment, and achieve full-color imaging. Although the rain and fog environment will increase the scattering and absorption of light, making fill light more difficult, the fill light device of this application consumes only 4.8W of power, which maintains a low power consumption level while overcoming environmental interference. In a backlight environment, the traditional fill light solution has insufficient dynamic range (<80dB), resulting in overexposure of the image, which cannot clearly present the details and color information of the target object. This application dynamically adjusts the fill light intensity, leveraging the multi-band fill light capabilities of the fill light device to increase the dynamic range to over 120dB. This effectively resolves the issue of backlight overexposure, ensuring clear visibility of both bright and dark details in the image and normal color reproduction. During this process, the fill light device consumes only 3.5W, achieving low power consumption while ensuring high-quality imaging.

[0192] In summary, the embodiments of the present application can achieve the following technical effects: In terms of imaging quality: traditional fill light technology is difficult to achieve high-quality full-color imaging in low-light environments. Under the influence of the red exposure problem, the use of short-wavelength infrared fill light will interfere with the normal imaging color; and the limitations of different wavelengths of infrared fill light in effective distance and fill light efficiency make the imaging effect poor at long distances or in complex environments, with low color reproduction and poor clarity. This application uses a dual-band fill light device and dynamic fill light control technology to achieve full-color imaging with a color error (ΔE) <10 in a variety of complex low-light scenes, greatly improving the image quality and providing a more accurate data basis for subsequent image analysis, target recognition and other tasks. In security monitoring, clear full-color images help to more accurately identify information such as facial features of people, color and details of objects. In terms of concealment: traditional infrared fill light with a wavelength of 850nm and below has a red exposure problem, which seriously affects the concealment of monitoring. This application uses 940nm invisible infrared fill light to fundamentally solve the red exposure problem, making the fill light device difficult to detect when filling light, meeting the needs of special scenarios such as secret monitoring and privacy protection, and expanding the application scope of the fill light device. In terms of power consumption: the power consumption of traditional fill lights in the normal on mode usually exceeds 5W, and the battery life is poor and the maintenance cost is high in battery-powered scenarios. This application reduces power consumption by more than 40% while ensuring imaging quality by optimizing the control algorithm and hardware design. Taking the indoor low-light environment as an example, the power consumption is only 2.1W, which makes the monitoring system more practical in scenarios that rely on battery power, such as field monitoring and portable devices, reducing the trouble of frequent battery replacement or charging, and improving the convenience and stability of the equipment. In terms of environmental adaptability: the imaging effect of traditional fill light technology will be seriously affected in complex environments such as rain, fog, and backlight, and may even fail to work properly. This application uses a sensor device, combined with a fill light collaborative control algorithm, to accurately perceive environmental changes in a variety of complex environments and dynamically adjust the fill light strategy. It can still maintain full-color imaging with a color error of 12.7 in outdoor rainy and foggy environments, and increase the dynamic range to more than 120dB in backlit environments, effectively solving the problem of environmental interference and greatly improving the environmental adaptability of the fill light device, enabling it to work stably in various harsh environments.

[0193] In the several embodiments provided in this application, it should be understood that the disclosed devices can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices according to the multiple embodiments of the present application. In this regard, each box in the block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram, and the combination of the block diagrams, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0194] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0195] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0196] The foregoing is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included within the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures.

[0197] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

[0198] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

Claims

1. A fill light control method, characterized in that: The method comprises: Determining the fill light intensity of a fill light device based on the collected environmental data; wherein the fill light device includes an infrared light component and a visible light component; The fill light device is controlled according to the fill light intensity.

2. The method according to claim 1, characterized in that The step of determining the fill light intensity for the fill light device based on the collected environmental data includes: The environmental data is collected by a sensor device; wherein the environmental data includes: target distance and spectral information; Determining a first fill light factor corresponding to the infrared light component and a second fill light factor corresponding to the visible light component based on the trained target model; determining the fill light intensity for performing fill light on the fill light device based on the environmental data, the first fill light factor, and the second fill light factor; The fill light intensity includes: a first fill light intensity corresponding to the infrared light component and a second fill light intensity corresponding to the visible light component.

3. The method according to claim 2, characterized in that The determining, based on the environmental data, the first fill light factor, and the second fill light factor, the fill light intensity for performing fill light on the fill light device includes: determining the first fill light intensity of the infrared light assembly based on the environmental data, the first fill light factor, and a first fill light formula; The second fill light intensity of the visible light component is determined based on the first fill light intensity, the second fill light factor, and the second fill light formula.

4. The method according to claim 3, characterized in that in, The first fill light formula includes: Among them, I IR is the first fill light intensity, K is the first fill light factor, D is the target distance, S / N target is the target signal-to-noise ratio, α is the atmospheric attenuation coefficient determined based on the spectral information and environmental physical parameters; The second fill light formula includes: in, is the second fill light intensity, η is the second fill light factor, C req The total amount of color required to achieve color reproduction, C IR is the contribution of infrared light to the overall color during color restoration determined based on the spectral information, C max The maximum color contribution provided to visible light.

5. The method according to claim 2, characterized in that in, The target model is trained in the following way: Collect historical data of different lighting conditions; wherein the lighting conditions include at least one of: light intensity, lighting scene, lighting distance, and weather conditions; Performing data preprocessing on the historical data to obtain training data; wherein the data preprocessing includes: data cleaning, data normalization, image processing and labeling; Determining a reward function shared by the first fill light factor and the second fill light factor; wherein the reward function is constructed based on a mapping relationship between the environmental data, the first fill light factor, and the second fill light factor, and the reward function is used to balance a peak signal-to-noise ratio (PSNR) of an image, a color reproduction error, and fill light power consumption; Based on the lightweight model, training is performed in combination with the training data and the reward function to obtain the target model.

6. The method according to claim 2, characterized in that The step of controlling the fill light device according to the fill light intensity includes: Processing the first fill light intensity based on signal conversion to obtain a first control signal for controlling the fill light of the infrared light component; Processing the second fill light intensity based on pulse width modulation to obtain a second control signal for controlling the fill light of the visible light component; The first control signal and the second control signal share a common clock source.

7. The method according to claim 6, characterized in that The processing of the second fill light intensity based on pulse width modulation to obtain a second control signal for controlling the fill light of the visible light component includes: Determine the operating frequency of pulse width modulation according to the application scenario; Obtaining an adjustable duty cycle of pulse width modulation based on the linear mapping conversion of the second fill light intensity; Determining duty cycle offsets of multiple bands of the visible light component based on the adjustable duty cycle and the spectral information; The second control signal is determined according to the duty cycle offset.

8. The method according to claim 2, characterized in that in, The sensing device includes: a distance sensor covering close-range, macro-range and long-range monitoring ranges, and a multispectral sensor covering the full spectrum range of ultraviolet light, visible light and near-infrared light; The environmental data collected by the sensor device includes: Acquiring initial distance data collected by the distance sensor; Performing compensation processing on the initial distance data to obtain the target distance; The spectral information obtained by the multispectral sensor based on environmental detection is acquired.

9. A monitoring system, characterized in that: The monitoring system includes: a processor and a fill light device; The processor is connected to the fill light device; The processor is used to determine the fill light intensity of the fill light device based on the collected environmental data; and perform fill light control on the fill light device according to the fill light intensity; wherein the fill light device includes an infrared light component and a visible light component.

10. The monitoring system according to claim 9, characterized in that: in, In the fill light device, the infrared light component and the visible light component are packaged on the same substrate with the same aperture, and the infrared light emitted by the infrared light component and the visible light emitted by the visible light component are emitted through the same aperture; The wavelength range of the infrared light component is 940nm±10nm, and the wavelength range of the visible light component is 400nm-700nm; The infrared light component includes a plurality of infrared light LEDs, and the visible light component includes a plurality of visible light LEDs; The plurality of infrared light LEDs and the plurality of visible light LEDs are staggered and arranged in a matrix structure or a ring shape; The light emitting centers of all the infrared light LEDs and the visible light LEDs are in the same plane and have the same emission direction; wherein the deviation of the emission optical axes of the infrared light and the visible light is less than or equal to 1°, and the flatness error of the light emitting centers of the infrared light and the visible light is less than or equal to 0.05mm.

Images