Monitoring light supplement automatic adjustment system based on ambient light detection

Abstract

The present application relates to the technical field of monitoring light supplement adjustment, and discloses a monitoring light supplement automatic adjustment system based on ambient light detection. The ambient light dynamic perception module of the system captures visible light and infrared spectral components of a monitoring area in real time through a multispectral sensor array, constructs a light intensity distribution model in combination with historical light intensity data, and outputs a theoretical light distribution map; the spectral feature analysis module compares the theoretical distribution map with measured light field data in multiple dimensions, performs spectral deviation detection, and generates a spectral difference feature tensor; the light supplement demand mapping module inputs the tensor into a spatial light analysis network, combines the field of view angle of a monitoring device and the position coordinates of a light supplement lamp, generates a light supplement coverage demand heat map, and calibrates the low-illumination area boundary; the adaptive light supplement strategy execution module enables a multi-waveband mixed light supplement mode for high-demand areas and implements phase collaborative control for adjacent light supplement lamp groups according to the heat map. The system realizes dynamic and accurate adjustment of light supplement, and can effectively adapt to complex ambient light changes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of monitoring light supplement adjustment, in particular to a monitoring light supplement automatic adjustment system based on ambient light detection. BACKGROUND

[0002] In the running process of the monitoring system, the stability of the ambient light directly affects the imaging quality, and the dynamic changes of natural and artificial light sources often bring challenges to the monitoring light supplement. The traditional monitoring light supplement system mostly adopts a fixed parameter mode, for example, triggering the light supplement switch according to a preset time threshold, or executing simple brightness adjustment after sensing the ambient light intensity through a single photosensitive resistor. This mode is difficult to cope with complex and variable ambient light scenes, and has obvious limitations.

[0003] In actual monitoring scenes, the ambient light distribution often presents significant spatial heterogeneity. For example, there may be building shadows, tree obstructions and other situations in the monitoring area, resulting in large differences in light intensity at different positions in the same area. The traditional system only relies on single-point light detection and cannot capture this spatial distribution characteristic, which easily causes insufficient light supplement in some areas and increased picture noise, and excessive light supplement in another part of the area, resulting in overexposure. In night monitoring, such problems are more prominent, and a fixed-power light supplement lamp may make the near-range area too bright, resulting in loss of details, while the far-range area is still in a low-illumination state, affecting the target recognition effect.

[0004] When multiple light supplement lamps work together, the traditional system lacks an effective linkage mechanism. Adjacent light supplement lamps often operate according to uniform parameters, which easily forms strong light interference in the light overlap area, producing light spots or halos, while the edge coverage area is dark due to insufficient light supplement. In addition, the influence of the spectral characteristics of ambient light on monitoring imaging quality has not been fully considered, and the traditional light supplement only focuses on light intensity adjustment, ignoring the spectral matching problem of visible light and infrared light. When the spectrum of natural light source changes due to weather and time changes, the mismatch between the spectrum of the light supplement and the spectrum of the ambient light will cause color distortion and contrast reduction of the picture, increasing the difficulty of subsequent image processing.

[0005] The dynamic changes of the ambient light also test the adaptability of the light supplement system. For example, the frequent fluctuations in light intensity caused by moving clouds on a cloudy day, and the sudden change in light intensity when street lights are turned on at dusk, the traditional light supplement system often has a lag in adjustment due to the lack of a fast response mechanism. Fixed light supplement parameters cannot make precise adjustments in response to real-time changes in ambient light, resulting in unstable monitoring picture quality, which makes it difficult to meet the demand for continuous clear imaging in high-definition monitoring. These problems greatly reduce the practicality of the traditional monitoring light supplement system in complex environments, and there is an urgent need for a technical solution that can dynamically perceive the characteristics of ambient light and accurately adjust the light supplement strategy. SUMMARY

[0006] The present application aims to provide a monitoring light supplement automatic adjustment system based on ambient light detection to solve the problems raised in the background.

[0007] To achieve the above-mentioned purpose, the present application provides a monitoring light supplement automatic adjustment system based on ambient light detection, which comprises:

[0008] The ambient light dynamic perception module: real-time capture of visible light and infrared spectral components of the monitoring area through a multi-spectral sensor array, construction of a light intensity distribution model based on illumination intensity historical data, synchronous acquisition of ambient illuminance value, shadow coverage coefficient and natural light source azimuth, and output of a theoretical light distribution map through the light intensity distribution model;

[0009] The spectral feature analysis module: multi-dimensional comparison of the theoretical light distribution map with the image sensor measured light field data, execution of spectral deviation detection, the spectral deviation detection including visible light band attenuation gradient, infrared spectrum shift and shadow area coverage difference, and generation of a spectral difference feature tensor of the monitoring point;

[0010] The light supplement demand mapping module: input of the spectral difference feature tensor into a spatial light analysis network, combination of the monitoring device field of view angle parameter and the light supplement lamp position coordinates, generation of a light supplement coverage demand heat map of the target area, and demarcation of the low-illumination physical area boundary;

[0011] The adaptive light supplement strategy execution module: configuration of light supplement parameters according to the light supplement coverage demand heat map, including:

[0012] Enabling a multi-band mixed light supplement mode for high demand areas;

[0013] Implementing phase coordinated control for adjacent light supplement lamp groups.

[0014] Preferably, the ambient light dynamic perception module specifically comprises:

[0015] Historical light feature extraction: spatio-temporal decoupling processing of the monitoring area historical light data, including: using adaptive filtering to separate the energy proportion of direct light and scattered light, establishing a mapping matrix of infrared components and ambient temperature through spectral correlation analysis, and aligning the light distribution patterns of different time periods using a spatio-temporal registration algorithm;

[0016] Light intensity distribution model construction: input of the processed historical light data into a hybrid modeling unit, the hybrid modeling unit comprising:

[0017] A time series prediction unit based on the sun's trajectory, for generating a basic illuminance prediction curve,

[0018] A feature fusion network embedded with a band attention mechanism, for correcting prediction bias caused by atmospheric transmittance,

[0019] Based on the shadow feature compensator, the model weight is dynamically adjusted according to the real-time collected shadow coverage coefficient;

[0020] The transient fluctuation amplitude of the ambient illumination value, the intensity distribution of the 3-12 mu band of the infrared spectrum, the change rate of the azimuth angle of the natural light source and the elevation angle gradient are synchronously captured by the distributed light-sensitive unit.

[0021] Preferably, in the generation of the theoretical light distribution map, the following steps are performed:

[0022] The dynamic calibration processing based on the meteorological conditions eliminates the measurement noise caused by cloud cover shielding;

[0023] The correlation features of visible light, infrared and shadow are integrated through a multi-source feature fusion algorithm;

[0024] The theoretical light distribution map containing the normal light fluctuation range is output, and the theoretical light distribution map is dynamically updated with the change of seasons.

[0025] Preferably, the spectral feature analysis module specifically includes:

[0026] Visible light attenuation gradient calculation: regional comparison of theoretical values and measured values is performed with a preset spatial grid, non-uniformly sampled illumination sequences are aligned using an intensity gradient matching algorithm, visible light attenuation in each grid is calculated, and a visible light deviation vector is generated;

[0027] Infrared spectrum shift detection: the infrared spectrum components of the theoretical values and the measured values are decomposed in the waveband domain, the energy ratio of the characteristic waveband is calculated, the spectrum shift index of each sub-waveband is extracted, and an infrared shift vector is constructed;

[0028] Shadow coverage rate evaluation: based on the structural similarity algorithm, the overlap degree of the theoretical values and the measured shadow distribution is matched, the position coincidence error of the shadow boundary is calculated, the Jaccard coefficient of the coverage rate difference is quantified, and a shadow difference vector is generated;

[0029] Spectral difference feature tensor generation: the visible light deviation vector, the infrared shift vector and the shadow difference vector are stacked in three dimensions, the dimension difference is eliminated through feature saliency weighted normalization operation, and a three-order feature tensor with dimensions of [monitoring point × time stamp × spectral feature] is output.

[0030] Preferably, the light compensation requirement mapping module specifically includes:

[0031] Monitoring topology modeling: a device spatial relationship topology map is constructed according to the light compensation lamp position coordinates, the field of view angle overlap parameters between devices are labeled, the obstacle shielding coefficient is superimposed in the topology map, and an optical topology model containing a field of view coverage matrix and a light attenuation matrix is generated;

[0032] Lighting demand simulation: mapping the spectral difference feature tensor to the corresponding nodes of the optical topology model; performing lighting demand deduction based on a graph convolution network, the calculation of the lighting demand deduction including:

[0033] Calculating the lighting coverage decay factor according to the field of view angle overlap parameter;

[0034] Capturing cross-regional lighting correlation features through a multi-head graph attention mechanism;

[0035] Simulating the transmission path of the lighting demand in the topology network using the Monte Carlo method;

[0036] Heat map generation: statistically analyzing the lighting demand frequency of each region in the simulation transmission, combining the lighting decay matrix to calculate the lighting intensity loss probability value, and generating a lighting coverage demand heat map covering the entire region, marking the set of regions to be enhanced whose probability value exceeds the preset demand threshold;

[0037] Physical boundary calibration: performing spatial clustering segmentation on the lighting coverage demand heat map to identify high demand aggregation areas; according to the location of monitoring equipment and the distribution of obstacles, the low-illumination physical region boundary coordinates are drawn.

[0038] Preferably, the adaptive lighting strategy execution module specifically includes:

[0039] When the loss probability value of a region in the lighting coverage demand heat map exceeds the preset demand threshold, a mode switching instruction is issued to the lighting equipment belonging to the region, and the following is performed:

[0040] Increase the visible light intensity to 3-5 times the reference value;

[0041] Synchronously enable the infrared light enhancement mode to capture the 700-1200nm band intensity mutation;

[0042] Deploy a shadow compensator on the edge to record the light recovery delay of the blocked area;

[0043] Phase coordination control execution: for the adjacent equipment group with the largest demand gradient change in the lighting coverage demand heat map, implement multi-band phase synchronization operation.

[0044] Preferably, the multi-band hybrid lighting mode includes:

[0045] Injecting 380-780nm visible light and 850-1550nm infrared light combination through adjustable spectrum light source;

[0046] Theoretical lighting response calculation: based on the optical topology model and the lighting decay matrix, the theoretical lighting response curve is calculated;

[0047] Measured light response acquisition: record the illuminance value of each area after the injection of the light compensation signal, and obtain the measured light compensation response curve;

[0048] Response deviation determination: calculate the Manhattan distance between the theoretical light compensation response curve and the measured light compensation response curve;

[0049] Compare the Manhattan distance between the measured response curve and the theoretical response curve, calculate the light compensation missing area offset, and mark it as a light compensation node to be optimized when the offset exceeds the preset tolerance threshold.

[0050] Preferably, the phase coordination control execution specifically includes:

[0051] Send a phase synchronization signal to the adjacent light compensation lamp group through the central controller;

[0052] Theoretical phase difference calculation: based on the device distance and the speed of light parameters, calculate the theoretical phase delay time;

[0053] Measured phase synchronization acquisition: record the time stamp of each device responding to the synchronization signal, and obtain the measured phase deviation;

[0054] Coordination error determination: calculate the absolute error value of the theoretical phase delay time and the measured phase deviation;

[0055] When the absolute error value exceeds the preset synchronization tolerance threshold, recalibrate the device clock source and feed back the clock calibration parameters to the light analysis network.

[0056] Preferably, the system further comprises:

[0057] Model iterative optimization module: receive the updated parameters of the light compensation coverage demand heat map and the measured light field data, and execute:

[0058] Extract the residual vector of the theoretical light distribution map and the measured data;

[0059] Adjust the convolution kernel weight of the light intensity distribution model through the back propagation algorithm;

[0060] Update the attenuation coefficient of the shadow feature compensator;

[0061] Output the optimized model parameters to the ambient light dynamic perception module.

[0062] Preferably, the model iterative optimization module specifically includes:

[0063] Residual convergence detection: calculate the Euclidean distance of the residual vector in the continuous iteration period;

[0064] When the Euclidean distance change rate is lower than the preset convergence threshold, freeze the model parameter update;

[0065] When the rate of change of the Euclidean distance exceeds a preset convergence threshold, a learning rate self-adaptive regulator is activated.

[0066] The model update state identifier is sent to the spectral feature analysis module.

[0067] Compared with the prior art, the present application has the following beneficial effects:

[0068] The system can capture visible light and infrared spectral components in the monitoring area in real time through the multispectral sensor array of the ambient light dynamic perception module, construct a light intensity distribution model combined with historical light intensity data, synchronously collect ambient illuminance values, shadow coverage coefficients and natural light source azimuth angles, and output a theoretical light distribution map that can more comprehensively reflect the spatial distribution characteristics of ambient light. This multi-dimensional ambient light perception method breaks through the limitations of traditional light supplement systems that rely on single light intensity detection, can accurately capture lighting details in the monitoring area, including the distribution of shadow areas, the orientation of different light sources and the spectral composition, and provides more abundant ambient light information for subsequent light supplement adjustment.

[0069] The spectral feature analysis module performs multi-dimensional comparison between the theoretical light distribution map and the actual light field data measured by the image sensor, performs spectral deviation detection, covers visible light band attenuation gradient, infrared spectrum offset and shadow area coverage difference, and generates a spectral difference feature tensor that can accurately identify the deviation of ambient light from ideal lighting conditions. This process makes up for the shortcomings of traditional light supplement adjustment, which only focuses on light intensity and ignores spectral characteristics, can specifically find lighting problems in different spectral bands, such as visible light attenuation during propagation, spectral matching degree of infrared light supplement and ambient light, etc., so that light supplement adjustment is more in line with the needs of spectral characteristics of monitoring imaging.

[0070] The light supplement demand mapping module inputs the spectral difference feature tensor into the spatial light analysis network, and combines the field of view angle parameters of the monitoring device and the light supplement lamp position coordinates, to generate a light supplement coverage demand heat map that can intuitively present the light supplement demand distribution of the target area. The calibrated low-illumination physical area boundary provides clear guidance for the accurate deployment of light supplement resources. Compared with the traditional system of uniform light supplement or fixed area light supplement, this light supplement planning based on spatial demand can avoid waste of light supplement resources, ensure that light supplement energy is concentrated in the truly needed low-illumination area, and reduce the interference of invalid light supplement on imaging quality.

[0071] The adaptive light supplement strategy execution module implements an adjustment strategy for the light supplement coverage demand heat map, enables a multi-waveband mixed light supplement mode for high demand areas, can flexibly adjust the proportion of visible light and infrared light supplement, and adapts to the imaging demand under different environmental light spectrum characteristics; phase coordination control is implemented on adjacent light supplement lamp groups, which can effectively coordinate the working states of multiple lamps, reduce the strong light interference in the light supplement overlapping area, and at the same time guarantee the light supplement intensity of the edge area. This fine light supplement control method enables the light supplement system to dynamically respond to the complex changes of the environmental light, and can maintain the stable quality of the monitoring picture and improve the applicability of the monitoring system under various environmental conditions through real-time adjustment, regardless of the strength fluctuation of the natural light source, the movement change of the shadow, or the difference in spectrum characteristics at different time periods. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 The working principle diagram of the monitoring light supplement automatic adjustment system based on environmental light detection is described.

[0073] Figure 2 The specific process diagram of the environmental light dynamic perception module is described.

[0074] Figure 3 The specific process diagram of the spectrum feature analysis module is described.

[0075] Figure 4 The process diagram of the multi-waveband mixed light supplement mode is described. DETAILED DESCRIPTION

[0076] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0077] Please refer to Figure 1 The present application provides a monitoring light supplement automatic adjustment system based on environmental light detection, which comprises:

[0078] The ambient light dynamic perception module captures visible light and infrared spectral components of the monitoring area in real time through a multispectral sensor array, while collecting ambient illumination values, shadow coverage coefficients, and natural light source azimuth angles. A light intensity distribution model is constructed based on historical light intensity data, and a theoretical light distribution map is output. The spectral feature analysis module compares the theoretical light distribution map with the measured light field data of the image sensor in multiple dimensions, performs spectral deviation detection, including visible light band attenuation gradient, infrared spectral band offset, and shadow area coverage rate difference, and generates a spectral difference feature tensor of the monitoring point. The light supplement demand mapping module inputs the spectral difference feature tensor into a spatial light analysis network, combines the monitoring device field of view angle parameters and the light supplement lamp position coordinates, generates a light supplement coverage demand heat map of the target area, and calibrates the low-illumination physical region boundary. The adaptive light supplement strategy execution module configures light supplement parameters according to the light supplement coverage demand heat map, enables a multi-band mixed light supplement mode for high demand areas, and implements phase coordinated control for adjacent light supplement lamp groups.

[0079] Embodiment 1: see Figure 2 The operation of the ambient light dynamic perception module includes multiple coordinated links, which together realize accurate capture and model construction of the light conditions of the monitoring area.

[0080] In the historical light feature extraction stage, the historical light data accumulated in the monitoring area is first processed by spatio-temporal decoupling. An adaptive filtering algorithm is used to process the original data, and by analyzing the change frequency and amplitude of the light intensity, the energy proportion of direct light and scattered light is separated. In this process, the algorithm automatically adjusts the filtering parameters to adapt to the feature differences of the two light components under different weather conditions. At the same time, through spectral correlation analysis, the correlation pattern between infrared spectral components and environmental temperature is explored, forming a mapping matrix that can reflect the mutual relationship between the two. The construction of this matrix is based on the statistical analysis of the corresponding relationship between infrared intensity and temperature in a large amount of historical data, without relying on a pre-set physical model. In addition, a spatio-temporal registration algorithm is used to process the light distribution patterns collected at different times, and by identifying feature points in the light field, the light distribution at different times is aligned in a unified spatial coordinate system, eliminating the spatial position deviation caused by time difference, so that the light data at different times is comparable.

[0081] The construction of the light intensity distribution model relies on the collaborative work of hybrid modeling units. Among them, the time series prediction unit based on the sun's trajectory will combine the geographical location information of the monitoring area, including latitude and longitude coordinates, and seasonal variation parameters, to calculate the position parameters of the sun at different times, and then generate the basic illumination prediction curve. This curve reflects the trend of the light intensity of the monitoring area changing with time under ideal weather conditions. The feature fusion network embedded with the waveband attention mechanism is responsible for processing information of different spectral wavebands, dynamically adjusting the weights of different wavebands in the model by analyzing the influence of each waveband light on the monitoring imaging quality. When the atmospheric conditions change, such as changes in humidity and particulate matter concentration leading to changes in atmospheric transmittance, the network will automatically correct the prediction results to reduce the impact of such changes on light prediction. The shadow feature compensator will adjust the weight factor related to the shadow in the model based on the real-time acquisition of the shadow coverage coefficient. When the shadow coverage range expands, the calculation weight of the corresponding area in the model will be correspondingly increased to enhance the sensitivity to the light changes in the shadow area.

[0082] The distributed light-sensitive units play a role in real-time data acquisition in the perception process. These sensors distributed in different positions of the monitoring area collect environmental illumination values in a high-frequency manner, which can capture the transient fluctuations of light intensity in a short time and record the amplitude and duration of the light mutation. At the same time, for the infrared spectrum, the sensor will focus on collecting the intensity distribution of the 3-12 μm waveband, and through the subdivision detection of this waveband, it can obtain the infrared energy information of different sub-wavebands. In addition, the sensor will continuously track the azimuth angle change rate and elevation angle gradient of the natural light source, which reflects the real-time changes of the sun's position, providing real-time input data for the light intensity distribution model, so that the model can respond to changes in the light source position in a timely manner.

[0083] The generation process of the theoretical illumination distribution map includes multiple processing steps. First, dynamic calibration based on meteorological conditions is performed. By accessing external meteorological data, information such as current cloud coverage is obtained to correct the collected illumination data. When cloud cover causes fluctuations in light measurement, the calibration algorithm eliminates the impact of these noises on the measurement results based on the thickness and distribution of the cloud cover. Subsequently, through a multi-source feature fusion algorithm, visible light, infrared, and shadow-related feature information are integrated. During the fusion process, different weights are assigned to visible light and infrared features based on the sensitivity of the monitoring device to different spectra, while considering the impact of shadow features on the overall illumination distribution. The final theoretical illumination distribution map includes the normal range of illumination fluctuations, which is not fixed but dynamically updated with seasonal changes. By analyzing historical illumination data and the overall impact of seasonal changes on illumination, the system redefines the upper and lower limits of the illumination fluctuation range every quarter, allowing the theoretical illumination distribution map to adapt to changes in illumination characteristics in different seasons.

[0084] Embodiment 2: Refer to Figure 3 The operation process of the spectral feature analysis module covers multiple interconnected operation links. By comparing the theoretical and measured illumination data in multiple dimensions, a feature tensor reflecting the spectral differences in the monitoring area is generated.

[0085] In the visible light attenuation gradient calculation link, the monitoring area is first divided according to the preset spatial grid. The size of the grid is determined according to the spatial resolution requirements of the monitoring scene, usually considering both computational efficiency and analysis accuracy. After division, the theoretical illumination distribution map output by the ambient light dynamic perception module and the measured light field data collected by the image sensor are regionally compared according to the grid units. Due to the possible differences in the distribution of sampling points between the theoretical and measured values, a light intensity gradient matching algorithm is used to process the non-uniformly sampled illumination sequence, aligning the two at the spatial sampling points through interpolation operations. After the alignment operation is completed, the visible light attenuation amount is calculated grid by grid, i.e., the difference between the theoretical and measured illumination values within the grid. After integrating the attenuation amounts of all grids, a visible light deviation vector containing the attenuation information of each point in space is formed.

[0086] The infrared spectrum deviation detection link focuses on the difference analysis of the theoretical and measured infrared spectrum components. First, the infrared spectrum data of the two is decomposed in the waveband domain, and the infrared waveband of 3-12 μm is divided into several sub-wavebands. The division of the sub-wavebands is determined according to the sensitive range of the monitoring equipment to the infrared spectrum. For each sub-waveband, the ratio of the measured energy to the theoretical energy is calculated to reflect the energy deviation degree of the sub-waveband. At the same time, the spectrum deviation index of each sub-waveband is extracted. The index is obtained by comparing the position difference between the measured waveband center wavelength and the theoretical waveband center wavelength, and can quantify the deviation degree of the spectrum on the wavelength axis. The energy ratio and deviation index of these sub-wavebands are arranged in order to construct an infrared deviation vector that can comprehensively reflect the difference of the infrared spectrum.

[0087] The shadow coverage evaluation link realizes the quantitative analysis of the difference between the theoretical and measured shadow distributions through a variety of algorithm combinations. First, the structural similarity algorithm is used to compare the shadow distributions of the two. This algorithm calculates the similarity of the brightness, contrast and structural features of the shadow area to obtain a structural similarity index that reflects the overall overlap degree. Then, the edge detection algorithm is used to extract the boundary profiles of the theoretical shadow and the measured shadow respectively. By calculating the spatial distance of the corresponding points on the boundary, the position coincidence error of the shadow boundary is obtained. On this basis, the Jaccard coefficient of the coverage difference is calculated. The coefficient is the ratio of the intersection area to the union area of the theoretical shadow and the measured shadow, and can directly reflect the coincidence degree of the two in the coverage range. The structural similarity index, boundary position coincidence error and Jaccard coefficient are integrated to generate a shadow difference vector.

[0088] The generation of the spectral difference feature tensor is the final output link of the spectral feature analysis module. First, the visible light deviation vector, the infrared deviation vector and the shadow difference vector generated in the previous text are stacked in three dimensions to form an initial three-order tensor. Since the physical dimensions of the three vectors are different, direct stacking will affect the accuracy of subsequent analysis, so normalization operation with feature saliency weighting is needed. In the weighting process, different weights are assigned to different vectors according to the influence of the features contained in each vector on the monitoring imaging quality. For example, the visible light deviation vector, which has a greater impact on the imaging clarity, may be assigned a higher weight. After weighting, the normalization algorithm is used to map the numerical values of the elements in the tensor to the [0, 1] interval, eliminating the influence of the dimension difference. The final output spectral difference feature tensor has a dimension of [monitoring point × time stamp × spectral feature], where the monitoring point dimension corresponds to the installation position of each monitoring equipment in the monitoring area, the time stamp dimension records the difference at different sampling times, and the spectral feature dimension contains various feature parameters related to visible light, infrared and shadow, forming a structured data that can comprehensively reflect the spectral difference of the monitoring area.

[0089] The operation of the light supplement demand mapping module starts with the analysis of the spatial relationship of the devices in the monitoring area. Based on the actual installation of the three-dimensional coordinates of the light supplement lamp, the device correlation topology structure is constructed in the digital space model. Each light supplement lamp is an independent node, and the connection relationship between the nodes is determined according to the actual physical distance. The distance value is directly marked on the connection line, and the density of the device distribution is intuitively presented. In this process, the overlapping area of the field of view angle of each device needs to be counted. The proportion of the overlapping area to the total area of the single device field of view is obtained through pixel comparison, which is used as the field of view angle overlap parameter. At the same time, combined with the three-dimensional modeling data of the obstacles, the shielding area of the light at different angles is analyzed, which is converted into a shielding coefficient between 0 and 1. The larger the coefficient, the more significant the shielding effect. After integrating these parameters, a matrix containing the device field of view coverage and a matrix of light propagation attenuation parameters are formed, which together constitute the optical topology model, providing a spatial framework for subsequent light supplement demand analysis.

[0090] The spectral difference feature tensor output by the spectral feature analysis module is assigned to the corresponding node of the optical topology model one by one, so that each node loads the spectral deviation information of the area. When deducing the light supplement demand based on the graph convolution network, the correlation weight of adjacent nodes is adjusted according to the field of view angle overlap parameter. The nodes with higher overlap parameters have stronger information transmission weight between them. Through the multi-head graph attention mechanism, the cross-regional lighting correlation rules are mined from multiple dimensions such as device location, light intensity, and spectral features. Each attention branch focuses on the analysis of one type of correlation. The Monte Carlo method is used to simulate the transmission path of the light supplement demand in the topology network. Each simulation randomly selects a transmission route from the high deviation area to the surrounding area. After repeating multiple times, the frequency of each area being covered by the path is counted, which reflects the diffusion trend of the light supplement demand.

[0091] In the process of generating the heat map, the total number of times each area is marked as needing light supplement in the simulation transmission, i.e., the light supplement demand frequency, is first summarized. Combined with the attenuation parameter corresponding to the area in the light attenuation matrix, the light intensity loss probability value is calculated, and the calculation formula is:

[0092]

[0093] Among them, represents the light intensity loss probability value, is the light supplement demand frequency, is the corresponding attenuation coefficient in the light attenuation matrix, is the total number of simulations. According to the calculation result, the light supplement coverage demand heat map covering the whole domain is drawn in the form of color gradient change. Different colors correspond to different probability value ranges, and the areas with probability values exceeding the set threshold are circled in the map to form a list of areas to be enhanced. The threshold is set according to the imaging light sensitivity parameters of the monitoring device.

[0094] When performing spatial clustering on the heatmap of supplemental lighting coverage demand, a density clustering algorithm is used to merge areas that are spatially close and have similar probability values ​​into high-demand clusters. Each cluster is defined by its minimum bounding polygon. Combining the installation coordinates of the monitoring equipment and the spatial distribution of obstacles, an edge tracking algorithm is used to determine the boundaries of low-light areas. Boundary points are recorded in three-dimensional coordinates to form a closed area outline, thus clarifying the spatial range of supplemental lighting operations.

[0095] When the probability of a missing area in the supplemental lighting coverage demand heatmap exceeds a set threshold, the system sends a mode switching command to the supplemental lighting device in that area. After the command is executed, the output power of the visible light supplemental lighting module is adjusted to 3-5 times the baseline value, which is the average measurement value of that area under normal lighting conditions during the same historical period. Simultaneously, the infrared supplemental lighting module activates its enhancement mode, monitoring real-time intensity changes in the 700-1200nm band. When fluctuations in intensity exceeding a set amplitude are detected in this band, the system records the time point of the fluctuation and the magnitude of the intensity change. Shadow compensators installed at the edge of the monitored area continuously monitor illumination values ​​using built-in high-sensitivity light sensors. After the obstruction is removed, the system records the time interval required for the illumination to rise and stabilize within the normal range. For adjacent device groups with the highest demand change rate in the supplemental lighting coverage demand heatmap, the central controller sends a synchronization signal containing a time stamp to each device. The devices adjust their own emission start time according to the signal to ensure that the supplemental lighting pulses of adjacent devices are consistent in time, avoiding uneven illumination superposition due to time differences. Throughout the process, data is exchanged in real time at each stage to ensure that the supplementary lighting strategy can accurately respond to changes in lighting conditions in the monitored area.

[0096] Example 4: See Figure 4 The operation of the multi-band hybrid supplementary lighting mode begins with the spectral configuration of the light source. The adjustable spectral light source can simultaneously output visible light (380-780nm) and infrared light (850-1550nm), and the output ratio of the two spectra can be adjusted through the control module. For example, in monitoring areas with predominantly human activity, the proportion of visible light can be increased to make the details of people in the image clearer; while in unattended warehouse areas at night, the output of infrared light can be increased to ensure monitoring effectiveness without affecting the surrounding environment. After the light source is activated, the internal spectral detection unit monitors the output spectral composition in real time to ensure that the actual output matches the set ratio.

[0097] The analysis of the theoretical light response is based on the constructed optical topology model and the light attenuation matrix. First, the installation position of the light supplement device is located in the model, and then the propagation path of the light in space is simulated according to the illumination angle and power parameters of the device, including the reflection and refraction when encountering obstacles such as walls and shelves. Through this simulation, the illumination values that should be theoretically reached at each position in the monitoring area can be obtained. Arranging these values according to the spatial position forms the theoretical light supplement response curve. The horizontal axis of the curve is the horizontal distance of the monitoring area, and the vertical axis is the corresponding illumination value, which intuitively presents the distribution trend of light in space.

[0098] The acquisition of the actual light response needs to rely on multiple light sensors distributed in the monitoring area. These sensors are uniformly installed at different heights and positions, and can fully capture the actual light conditions. When the light supplement device starts and stabilizes the output, the sensor collects the illumination value at the position at a fixed time interval, for example, recording data every 0.5 seconds. After continuous collection for a period of time, the actual measurement data is arranged according to the corresponding spatial position to form the actual light supplement response curve. In the monitoring scene of a large parking lot, sensors may be installed at ground level, the middle of a column, and the ceiling, etc. to record the light intensity at different heights.

[0099] The determination process of response deviation is a comparative analysis of the theoretical and actual curves. By calculating the illumination difference of the corresponding spatial points of the two curves, and then adding the absolute values of these differences, the Manhattan distance is obtained to measure the overall deviation. When this distance exceeds the set deviation threshold, further analysis of the offset of the light supplement missing area is needed. For example, in a rectangular workshop monitoring, the light supplement should theoretically uniformly cover the entire workshop, but the actual measurement found that the area near the north wall has a significantly low illumination. At this time, the coordinate difference, i.e. the offset, between the center position of the low-illumination area and the center position of the theoretical low-illumination area (if exists) needs to be calculated. When the offset exceeds the pre-set tolerance threshold, the corresponding light supplement node of the area will be marked as to be optimized for subsequent adjustment.

[0100] Phase coordination control is mainly applied to adjacent light supplement lamp groups. For example, in the atrium area of a large shopping mall, multiple light supplement lamps are usually installed to illuminate from different angles. At this time, it is necessary to ensure the consistent light supplement rhythm of these lamp groups. The central controller sends a phase synchronization signal to each group of light supplement lamps, which contains accurate time markers and phase instructions. After receiving the signal, each group of devices records its response time, i.e. the time interval from receiving the signal to starting to adjust the light state. By comparing the response times of different devices, the actual phase deviation can be obtained.

[0101] In a collaborative system composed of three groups of fill light, after the central controller sends a synchronization signal, the first group of devices responds after 10 milliseconds, the second group responds after 12 milliseconds, and the third group responds after 9 milliseconds. These time data are transmitted back to the control center in real time. The control center analyzes the actual distance between devices and the speed of light propagation, compares the actual phase deviation with the expected response time difference of each group of devices, and calculates the absolute error value. When the absolute error value of two groups of devices exceeds the preset synchronization tolerance threshold, the system recalibrates the clock source of the devices, adjusts the internal timing module to keep the response time of each group of devices consistent, and feeds back the parameter changes generated during the calibration process to the light analysis network for subsequent analysis.

[0102]

[0103] The above table presents some data of three groups of fill light in a phase synchronization control, where the synchronization signal receiving time is the unified time point when each group of devices receives the signal from the central controller, the response synchronization signal time is the time when the device starts to adjust the light state, and the measured phase deviation is the difference between the actual response time and the preset response time. Through such data recording and analysis, the coordination of each group of devices can be clearly understood, providing a basis for subsequent adjustment.

[0104] In example 5, the model iterative optimization module continuously receives updated parameters of fill light coverage demand heat maps and measured light field data feedback from image sensors, and realizes dynamic adjustment of the light intensity distribution model through multi-step data processing. First, residual vectors are extracted from the theoretical light distribution map and the measured data. Specifically, the values at the same spatial coordinates and time stamps are compared one by one, and the difference at each corresponding point is calculated. These differences are arranged in spatial position and time order to form a residual vector that reflects the model prediction bias.

[0105] When adjusting the convolution kernel weights of the light intensity distribution model, the backpropagation algorithm is used to correct the parameters inside the model. The algorithm adjusts the weight values of the convolution kernel in the direction indicated by the residual vector, so that the model can reduce similar deviations in subsequent predictions. In this process, the adjustment amplitude of the weights of different convolution layers is determined according to their influence on the output results. Convolution kernels close to the output layer are usually given larger adjustment amplitudes to quickly respond to residual changes.

[0106] The attenuation coefficient update of the shadow feature compensator is closely related to the deviation of the shadow area in the residual vector. By analyzing the numerical distribution of the corresponding shadow area in the residual vector, the adjustment direction and amplitude of the attenuation coefficient are determined. When the actual measured shadow coverage of a certain area is greater than the theoretical prediction, the attenuation coefficient of the area is correspondingly increased; otherwise, the coefficient is decreased, so that the compensator can more accurately reflect the influence of the actual shadow on the light distribution. The adjusted model parameters are fed back to the ambient light dynamic perception module in real time to replace the original parameter settings, ensuring that the subsequent generated theoretical light distribution map is based on the optimized model.

[0107] Residual convergence detection is a key step to determine whether the model needs to continue optimization. The Euclidean distance of the residual vectors in two consecutive iteration periods is calculated, which is obtained by taking the square root of the sum of the square of the difference between the corresponding elements of the two residual vectors, and can reflect the overall change of the deviation in two iterations. When the change rate of the Euclidean distance is at a low level, it means that the prediction deviation of the model has entered a stable state, at which time the parameter update is stopped and the current model configuration is maintained.

[0108] If the change rate of the Euclidean distance exceeds the set threshold, the system will activate the learning rate adaptive adjuster. The adjuster dynamically changes the learning rate according to the specific value of the change rate. When the change rate is large, a larger learning rate is used to speed up the adjustment of the model parameters; when the change rate gradually decreases, the learning rate is automatically reduced, making the parameter adjustment more precise and avoiding model instability caused by excessive adjustment.

[0109] During the entire iterative optimization process, the model update state identifier is sent to the spectral feature analysis module in real time. The identifier contains information such as whether the model is in an update state, the time of the last update, and the current iteration number. The spectral feature analysis module adjusts its analysis parameters according to this information, for example, when the model is in an update state, the sampling frequency of spectral comparison is appropriately increased to obtain more detailed measured data, providing more abundant input for model optimization; when the model parameters are frozen, the normal sampling frequency is restored to balance the analysis accuracy and system resource consumption.

[0110] This continuous iterative optimization mechanism enables the light intensity distribution model to continuously adapt to long-term changes in the lighting conditions of the monitoring area. Whether it is a change in lighting patterns caused by seasonal changes or differences in shadow distribution caused by changes in the surrounding environment, the model can maintain accurate prediction of the actual lighting conditions through parameter adjustment, providing reliable basic data for subsequent spectral feature analysis and light compensation strategy formulation.

[0111] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0112] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A monitoring and automatic adjustment system of light supplement based on ambient light detection, characterized in that, Comprise: Ambient light dynamic perception module: real-time capture visible light and infrared spectral components of the monitoring area through a multispectral sensor array, construct a light intensity distribution model based on historical light intensity data, simultaneously collect ambient illuminance value, shadow coverage coefficient and natural light source azimuth angle, output theoretical light distribution map through the light intensity distribution model; The ambient light dynamic perception module specifically comprises: Historical light feature extraction: spatio-temporal decoupling processing of historical light data of the monitoring area, including: using adaptive filtering to separate the energy proportion of direct light and scattered light, establishing a mapping matrix of infrared components and ambient temperature through spectral correlation analysis, aligning light distribution patterns at different times using a spatio-temporal registration algorithm; Light intensity distribution model construction: input the processed historical light data into a hybrid modeling unit, the hybrid modeling unit includes: Time series prediction unit based on solar trajectory, for generating a basic illuminance prediction curve, Feature fusion network embedded with waveband attention mechanism, to correct prediction bias caused by atmospheric transmittance, Shadow feature-based compensator, dynamically adjusting model weights according to the real-time collected shadow coverage coefficient; Synchronously capture the transient fluctuation amplitude of ambient illuminance value, the 3-12 μm waveband intensity distribution of infrared spectrum, the change rate of natural light source azimuth angle and the elevation angle gradient through a distributed photosensitive unit; Spectral feature analysis module: multi-dimensional comparison of the theoretical light distribution map and the measured light field data of the image sensor, performing spectral deviation detection, the spectral deviation detection includes visible light waveband attenuation gradient, infrared spectrum offset and shadow area coverage difference, generating spectral difference feature tensor of the monitoring point; Light supplement demand mapping module: input the spectral difference feature tensor into a spatial light analysis network, combine the monitoring device field of view angle parameters and the light supplement lamp position coordinates, generate the light supplement coverage demand heat map of the target area, and calibrate the low-illumination physical area boundary; Adaptive light supplement strategy execution module: configure light supplement parameters according to the light supplement coverage demand heat map, including: Enable multi-waveband mixed light supplement mode for high demand areas; Implement phase coordinated control for adjacent light supplement lamp groups.

2. The system according to claim 1, wherein, In the generation of the theoretical light distribution map, including: Dynamic calibration processing based on weather conditions, eliminating measurement noise caused by cloud cover; Integrate the correlation features of visible light, infrared and shadow through a multi-source feature fusion algorithm; Output the theoretical light distribution map containing the normal light fluctuation range, which is dynamically updated with the change of seasons.

3. The system according to claim 1, wherein, The spectral feature analysis module specifically comprises: Visible light attenuation gradient calculation: regionalize comparison of theoretical value and measured value with a preset spatial grid, align non-uniformly sampled illuminance sequence using light intensity gradient matching algorithm, calculate visible light attenuation in each grid, and generate visible light deviation vector; Infrared spectrum offset detection: waveband domain decomposition of the infrared spectral components of the theoretical value and the measured value, calculate the energy ratio of the characteristic waveband, extract the spectrum offset index of each sub-waveband, and construct an infrared offset vector; Shadow coverage evaluation: based on the structural similarity algorithm, the matching value and the overlap degree of the measured shadow distribution are calculated, the position coincidence error of the shadow boundary is calculated, the Jacard coefficient of the coverage difference is quantified, and the shadow difference vector is generated; Spectral difference feature tensor generation: the visible light deviation vector, infrared offset vector and shadow difference vector are stacked in three dimensions, the dimension difference is eliminated through feature significance weighted normalization operation, and a three-order feature tensor with dimensions [monitoring point × timestamp × spectral feature] is output.

4. The system according to claim 3, wherein the system further comprises a light sensor for detecting the ambient light and a light source for providing the light to the object. The light supplement demand mapping module specifically includes: Monitoring topology modeling: construct a device spatial relationship topology graph according to the light supplement lamp position coordinates, label the field of view angle overlap parameters between each device, superimpose the obstacle blocking coefficient in the topology graph, and generate an optical topology model containing a field of view coverage matrix and an illumination attenuation matrix; Light supplement demand simulation: map the spectral difference feature tensor to the corresponding nodes of the optical topology model; based on the graph convolution network, the light supplement demand deduction is performed, and the calculation of the light supplement demand deduction includes: Calculate the light supplement coverage attenuation factor according to the field of view angle overlap parameter; Capture cross-regional illumination correlation features through multi-head graph attention mechanism; Adopt Monte Carlo method to simulate the transmission path of light supplement demand in the topology network; Heat map generation: count the light supplement demand frequency of each region in the simulation transmission, calculate the light supplement intensity loss probability value combined with the illumination attenuation matrix, generate a light supplement coverage demand heat map covering the whole domain, and label the to-be-enhanced region set whose probability value exceeds the preset demand threshold; Physical boundary calibration: perform spatial clustering segmentation on the light supplement coverage demand heat map to identify high-demand aggregation areas; according to the positions of monitoring devices and the distribution of obstacles, the low-illumination physical region boundary coordinates are drawn.

5. The system according to claim 4, wherein, The adaptive light supplement strategy execution module specifically includes: When the loss probability value of a certain region in the light supplement coverage demand heat map exceeds the preset demand threshold, a mode switching instruction is issued to the light supplement device to which the region belongs, and the following is performed: Increase the visible light supplement intensity to 3-5 times the reference value; Synchronously enable the infrared light supplement enhancement mode to capture the 700-1200nm band intensity mutation; Deploy a shadow compensator on the edge to record the light recovery delay of the blocked area; Phase coordination control execution: for the adjacent device group with the largest demand gradient change in the light supplement coverage demand heat map, implement multi-band phase synchronization operation.

6. The system according to claim 5, wherein the system further comprises a light sensor for detecting the ambient light and a light source for providing the light to the object. The multi-band mixed light supplement mode includes: Inject 380-780nm visible light and 850-1550nm infrared light combination through adjustable spectral light source; Theoretical illumination response calculation: based on the optical topology model and the illumination attenuation matrix, the theoretical light supplement response curve is calculated; Measured illumination response acquisition: record the illumination value of each region after light supplement signal injection to obtain the measured light supplement response curve; Response deviation determination: calculate the Manhattan distance between the theoretical light supplement response curve and the measured light supplement response curve; Compare the Manhattan distance between the measured response curve and the theoretical response curve, calculate the offset amount of the light supplement loss area, and mark it as a to-be-optimized light supplement node when the offset amount exceeds the preset tolerance threshold.

7. The system according to claim 6, wherein the system further comprises a light sensor for detecting the ambient light and a light source for providing the light to the object. The phase coordination control execution specifically includes: Send a phase synchronization signal to the adjacent light supplement lamp group through the central controller; Theoretical phase difference calculation: based on the distance between devices and the speed of light parameters, calculate the theoretical phase delay time; Measured phase synchronization acquisition: record the time stamp of each device responding to the synchronization signal to obtain the measured phase deviation; Synergistic error determination: calculate the absolute error value of the theoretical phase delay time and the measured phase deviation; When the absolute error value exceeds the preset synchronization tolerance threshold, recalibrate the device clock source and feedback the clock calibration parameters to the light analysis network.

8. The system of claim 1, wherein the system is configured to automatically adjust the intensity of the light source based on the intensity of the ambient light detected by the ambient light sensor. Also includes: Model iterative optimization module: receive the update parameters of the light supplement coverage demand heat map and the measured light field data, execute: Extract the residual vector of the theoretical light distribution map and the measured data; Adjust the convolution kernel weight of the light intensity distribution model through the back propagation algorithm; Update the attenuation coefficient of the shadow feature compensator; Output the optimized model parameters to the ambient light dynamic perception module.

9. The system according to claim 8, wherein, The model iterative optimization module specifically includes: Residual convergence detection: calculate the Euclidean distance of the residual vector in the continuous iteration period; When the Euclidean distance change rate is lower than the preset convergence threshold, freeze the model parameter update; When the Euclidean distance change rate exceeds the preset convergence threshold, activate the learning rate adaptive adjuster; Send the model update state identifier to the spectral feature analysis module.

Images