Auto-induction lamplight angle adjustment control method and device and lighting equipment

Abstract

The invention relates to the technical field of automatic induction, and discloses an automatic induction lamplight angle adjustment control method and device and lighting equipment. The method comprises the following steps: acquiring ambient light parameters in real time to obtain ambient light sensing data; performing multi-scale space division and feature fusion on the ambient light sensing data to obtain a three-dimensional ambient light distribution diagram; inputting the three-dimensional ambient light distribution diagram into a light angle calculation network for light change characteristic analysis to obtain a light optimal angle parameter group; performing smooth transition light sequence calculation according to the light optimal angle parameter group and the current light configuration state to obtain a sectional light adjustment control instruction; and performing motion path planning on the end effector of the lamp based on the sectional light adjustment control instruction to obtain an effector motion instruction set. According to the invention, light change characteristics of different spatial scales can be captured at the same time, and the adjustment accuracy and adaptability of the actuator are improved.

Description

Technical Field

[0001] The present invention relates to the field of automatic sensing technology, and particularly to an automatic sensing lighting angle adjustment control method, device and lighting equipment. Background Art

[0002] Traditional lighting systems usually adopt fixed-angle lighting or simple on-off control modes, and cannot adaptively adjust according to ambient light changes and actual lighting requirements, resulting in problems such as unsatisfactory lighting effects, energy waste and poor user experience. With the progress of artificial intelligence and sensor technology, intelligent lighting systems have begun to develop towards automatic sensing and precise control, but the existing technologies still face many challenges.

[0003] Existing automatic sensing-based lighting angle control methods generally have problems such as high computational cost, large model complexity, and slow response speed, and are difficult to be deployed and operated in resource-constrained actual application environments. Traditional methods usually rely on single-dimensional light perception data and cannot comprehensively capture multi-dimensional light change characteristics in complex environments. Especially in scenes with rapid light changes, problems such as inaccurate lighting angle adjustment and visual discomfort caused by sudden light changes are likely to occur. In addition, the existing methods lack fine control of the end effector, resulting in unsmooth mechanical movement and insufficient precision in the actual lighting angle adjustment process, and it is difficult to achieve precise lighting positioning and smooth transition. Summary of the Invention

[0004] The main object of the present invention is to provide an automatic sensing lighting angle adjustment control method, device and lighting equipment, which can simultaneously capture light change characteristics at different spatial scales and improve the accuracy and adaptability of actuator adjustment.

[0005] To achieve the above object, the present invention provides an automatic sensing lighting angle adjustment control method, including the following steps: Collect ambient light parameters in real time to obtain ambient light perception data; Perform multi-scale spatial partitioning and feature fusion on the ambient light perception data to obtain a three-dimensional ambient light distribution map; Input the three-dimensional ambient light distribution map into a lighting angle calculation network for light change feature analysis to obtain a set of optimal lighting angle parameters; Execute a smooth transition light sequence calculation according to the set of optimal lighting angle parameters and the current lighting configuration state to obtain a segmented lighting adjustment control instruction; Based on the segmented lighting adjustment control instruction, perform a motion path planning for the end effector of the lamp to obtain an actuator motion instruction set.

[0006] The present invention also provides an automatic sensing lighting angle adjustment control device, including: The acquisition module is used to collect environmental light parameters in real time to obtain environmental light perception data; The feature fusion module is used to perform multi-scale spatial partitioning and feature fusion on the environmental light perception data to obtain a three-dimensional environmental light distribution map; The feature analysis module is used to input the three-dimensional environmental light distribution map into a lighting angle calculation network for analyzing the characteristics of light change to obtain an optimal lighting angle parameter set; The sequence calculation module is used to perform a smooth transition light sequence calculation according to the optimal lighting angle parameter set and the current lighting configuration state to obtain a segmented lighting adjustment control instruction; The path planning module is used to perform a motion path planning on the end effector of the lighting fixture based on the segmented lighting adjustment control instruction to obtain an actuator motion instruction set.

[0007] The present invention also provides a lighting device, and the lighting device executes the steps of any one of the above methods.

[0008] In summary, the technical solution provided by the present invention uses a multi-sensor fusion acquisition technology to achieve a full-dimensional perception of environmental light parameters. By integrating light intensity, color temperature, multi-spectral, depth, and motion information, it comprehensively captures the characteristics of the light environment, solves the problem of insufficient acquisition dimensions of traditional single sensors, and makes light perception more comprehensive and accurate. It can simultaneously capture the characteristics of light change at different spatial scales, and through the aggregation processing and hierarchical analysis of different neighborhood sizes, effectively characterize the light environment changes in local, medium, and large ranges, enabling the system to better understand and adapt to complex light distributions. The lightweight lighting angle calculation network adopts a luminance-time attention module structure, significantly reducing the computational complexity and the number of model parameters, while maintaining a high-precision angle calculation ability, enabling the system to operate efficiently in resource-constrained environments. The light dynamic buffering and transition optimization mechanism realizes smooth light transition by designing various transition functions and dynamic buffer parameter adjustments, effectively avoiding visual discomfort caused by sudden changes in lighting, and improving the user experience and lighting comfort. The end effector motion planning method based on a conditional generative adversarial network realizes the decoupling of lighting configuration and actuator control. Through the lighting enhancement module and diversity loss optimization, it generates an optimal lighting angle configuration that meets multi-dimensional constraint conditions, improving the accuracy and adaptability of actuator adjustment. Description of the Drawings

[0009] Figure 1 is a schematic diagram of the steps of an automatic sensing lighting angle adjustment control method in an embodiment of the present invention; Figure 2 is a block diagram of the structure of an automatic sensing lighting angle adjustment control device in an embodiment of the present invention.

[0010] The realization, functional features, and advantages of the object of the present invention will be further described in conjunction with embodiments with reference to the accompanying drawings. Detailed implementation manners

[0011] In order to make the object, technical solutions, and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0012] Referring to Figure 1 , this embodiment provides an automatic sensing light angle adjustment control method, including the following steps: S1, collect ambient light parameters in real time to obtain ambient light perception data; Among them, light intensity sensors are evenly deployed in the space to be adjusted. Such sensors use high-sensitivity photodiodes or photovoltaic cell components, with a response sensitivity above 0.5 lx and a dynamic measurement range up to 100,000 lx. Combined with a high-frequency sampling rate of 100 Hz, they can achieve a rapid response to changes in light intensity. By collecting the light intensity distribution in each area in a grid-like manner, a basic illuminance data field is constructed. Color temperature sensors are configured to collect the color temperature data emitted by the main light sources in the space. Such sensors use a three-color photosensitive structure or a color temperature conversion model based on CIE standard chromaticity calculation, with a measurement range of 1000 K to 10,000 K and an accuracy of ±50 K, ensuring that the obtained color temperature information can effectively reflect the hue trend of the current light source. At the same time, multi-spectral sensors are used to scan the light energy in different bands in the lighting environment. Such sensors cover the visible spectral range from 380 nm to 780 nm and have a spectral resolution of 5 nm, capturing the energy distribution of specific wavelength components and identifying the spectral characteristic differences between different light sources. To comprehensively perceive the three-dimensional geometric information in the space and the influence of the occlusion structure, depth sensors based on the principle of structured light or TOF technology are deployed to obtain the depth information of each key point in the space scene. This information includes the distance distribution of fixed structures such as walls, floors, and ceilings, and also includes the relative positions of various movable or semi-fixed devices in the space, with a measurement accuracy better than ±1 cm and a ranging range covering 0.5 m to 10 m. On the basis of completing the perception of the above light environment and structural environment information, motion sensors are introduced, such as dynamic target detection devices based on the principle of pyroelectric infrared detection, to capture the activity trajectories and state changes of people in the space in real time. Such sensors have the ability to be sensitive to weak heat changes and can distinguish the presence and movement direction of objects, thus supplementing the response requirements of the light environment to people. Time synchronization and multi-modal data fusion operations are performed on the light intensity data, color temperature distribution data, spectral distribution data, spatial depth information data, and target motion trajectory data. All sensor data are accessed into the central data processing module through high-speed communication interfaces such as RS485. At the moment of access, the system assigns a high-precision timestamp to each group of data, then performs noise reduction processing on the collected data through the Kalman filter algorithm, completes data smoothing by using the sliding weighted average method, uses the three-sigma discrimination criterion to eliminate outliers, and fills in the existing missing items by linear interpolation to ensure that the obtained various light environment information is consistent in the time dimension. During the data fusion process, a feature layer fusion mechanism is adopted to uniformly map the light intensity, color temperature, multi-spectral, depth, and motion trajectory data into a three-dimensional space coordinate system, and complete the superposition of multiple parameters under a unified grid framework to form structured ambient light perception data with position attributes, time tags, spectral dimensions, and dynamic information.

[0013] S2. Perform multi-scale spatial partitioning and feature fusion on the ambient light perception data to obtain a three-dimensional ambient light distribution map; Specifically, the ambient light perception data is uniformly processed by coordinate grid division according to the actual functional areas in space. The target space is divided into multiple equivalent grid cells, and multi-dimensional perception data such as the light intensity, color temperature, multi-spectral components, spatial depth, and dynamic trajectory of the sampling points are mapped into the corresponding grids, forming a basic ambient light distribution matrix with spatial positions (x, y, z) and corresponding light parameter vectors. Each element in this matrix not only represents the local light environment state at a certain spatial position but also embeds a time tag to support subsequent dynamic analysis. A multi-scale aggregation mechanism is introduced, that is, the basic matrix is aggregated based on different neighborhood scales, and the first, second, and third scale distribution maps are constructed respectively. The first scale distribution map is generated by performing a weighted average operation on the light parameters of 2×2 grid neighborhood units. This scale highlights the fine differences in local illumination. The second scale distribution map is formed by 4×4 neighborhood aggregation, emphasizing the regional illumination trend. The third scale extracts the global trend with an 8×8 neighborhood, showing the distribution characteristics of the large-scale illumination structure. The weight coefficients used in the aggregation operations at each scale are adaptively allocated according to the signal-to-noise ratio and sampling credibility of the original perception data to ensure the reliability and physical consistency of the data fusion process. After aggregation, the ambient light distribution maps at the three scales respectively constitute spatial perception expressions at different scales. The system unifies and fuses them on this basis to construct a multi-scale spatial light distribution tensor structure. This tensor not only contains light parameters in different spatial dimensions but also retains the context structures at each scale level. Through convolution transformation or weight mapping mechanisms, the multi-scale information is spatially fused to obtain a light distribution field with multi-resolution perception capabilities. At the same time, to enhance the dynamic adaptability and time sensitivity of the three-dimensional distribution map, time series modeling is combined with historical light change data. This historical data is obtained by performing exponential weighted moving average, periodic trend extraction, and wavelet transform analysis on the original ambient light perception data sequence, and is used to model the short-term fluctuations, medium-term trends, and long-term stable structures of illumination respectively. The multi-scale spatial light distribution tensor and the historical light change data are jointly correlated in time series. By constructing a sliding time window of the tensor sequence, the change rate, fluctuation trend, and frequency domain characteristics of the light parameters at the same position under different time scales are analyzed to establish a stable space-time coupling model. Based on the above fusion and analysis results, a three-dimensional ambient light distribution map is generated, which is expressed in the form of tensor T(x, y, z, λ, t), expressing the distribution states of parameters such as light intensity, color temperature, and multi-spectral at spatial positions, and integrating the spectral response in the wavelength λ dimension and the change trend in the time t dimension to form a comprehensive expression of the multi-faceted characteristics of the light environment in terms of spatial structure, light source type, time evolution, dynamic response, etc.

[0014] S3. Input the three-dimensional ambient light distribution map into the lighting angle calculation network for analyzing the characteristics of light changes, and obtain the optimal lighting angle parameter group; It should be noted that the three-dimensional ambient light distribution map is input into the lighting angle calculation network. This distribution map is represented in the form of a multi-dimensional tensor T(x, y, z, λ, t), which contains information on multiple dimensions such as the spatial position of the light, the spectral wavelength, and the temporal variation. In the encoder part of the network, a layer-by-layer downsampling process is performed on the input tensor. Each layer uses a convolutional operation with a PReLU activation function and a max-pooling structure to gradually compress the spatial resolution and enhance the feature abstraction ability, obtaining multi-level light change features. These features capture the pattern differences between local and global lighting distributions at the spatial level and retain the dynamic change trends of illuminance, color temperature, and spectrum in the time dimension. The output feature map of each layer retains the original position index information for subsequent skip connections and spatial restoration. The feature maps of each layer are respectively input into a specially designed luminance attention branch in the lighting angle calculation network. This branch uses a channel attention mechanism to extract the importance weights of spatial features. The luminance attention branch extracts local spatial luminance features through a 3×3 standard convolution, and then performs global average pooling and max-pooling operations respectively to generate two feature vectors describing the channel responses. These two vectors are then input into a fully connected network with shared parameters (including two fully connected layers, the output dimension of the first layer is 16, and the second layer restores to the original number of channels), and a normalized spatial feature weight vector is output through a Sigmoid activation function, thereby realizing the evaluation and weighted processing of the contribution of different channels to the perception of lighting changes at the spatial level. At the same time, the same batch of multi-level feature maps is fed into the temporal attention branch in parallel. This branch uses a temporal convolutional network to capture the patterns of light evolution over time. The temporal convolutional network module uses a dilated convolution structure, the kernel size is set to 1×5, and the dilation rate is set to 2. The effective receptive field covers 9 consecutive time points, which helps to identify periodic lighting fluctuations or mutation trends. After temporal feature extraction, a self-attention mechanism is introduced to calculate the similarity and influence weights between different time points in the time dimension, generating a temporal feature weight matrix to measure the dynamic importance of each time point in lighting adjustment. The spatial feature weight vector corresponding to each layer is weighted and fused with the temporal feature weight matrix to construct multi-level spatio-temporal fusion features. This fusion process is realized by tensor multiplication to achieve cross-dimensional interaction. The temporal weights are broadcast by expanding the channel dimension, and the activation intensity of each feature channel at different times is adjusted proportionally to form a more dynamically adaptable feature expression result. The multi-level spatio-temporal fusion features are input into the decoder part of the lighting angle calculation network. The decoder consists of four symmetrically structured transposed convolutional layers, with the kernel size of each layer set to 4×4 and the stride to 2. The original input spatial resolution is gradually restored through layer-by-layer upsampling, and the features of the corresponding layer in the encoder are concatenated with the features of the current decoding layer through a skip connection mechanism to retain high-resolution position information and local details, obtaining the target lighting angle feature map.The target light angle feature map is input into the fully connected layer at the end of the network, compressed into a target angle parameter vector through a set of linear mapping operations, and the output range is normalized to [-1, 1] using the Tanh activation function. Then, through inverse normalization transformation, it is mapped to the physically adjustable angle range of the lamp, and the optimal light angle parameter group θ = (α, β, γ) is output, where α is the horizontal rotation angle, β is the vertical tilt angle, and γ is the axial rotation angle.

[0015] Taking a three-dimensional ambient light distribution map composed of multi-dimensional data such as spatial coordinates, light intensity, color temperature, multi-spectral components, depth information, and time tags as the input, this high-dimensional tensor is input into the encoder part of the light angle calculation network. At the first layer of the encoder structure, an initial local receptive field feature extraction is performed on the input tensor through a standard 3×3 two-dimensional convolutional layer. This convolutional operation performs a sliding calculation in the spatial dimension to extract underlying features such as the edges, gradients, and spot structures of the light distribution. The output result of the convolutional layer is then non-linearly transformed through the PReLU activation function to enhance the model's ability to express local brightness changes, obtaining an initial feature map with preliminary discrimination ability. The initial feature map is input into the first luminance-time attention module to perform joint feature enhancement. This module contains a set of spatial channel attention paths and a set of temporal convolutional paths, which process the lighting characteristics in two dimensions of space and time respectively. Among them, the channel attention path uses pooling and fully connected structures to extract the significant regions of the spatial distribution, while the time path uses dilated convolution and self-attention mechanisms to enhance the sensitivity to dynamic lighting changes. The two branches are weighted and fused to obtain the first layer of light change features. This feature map is then downsampled in size through a max pooling layer to compress redundancy and enhance the feature aggregation ability. After processing, the first layer feature map is input into the second luminance-time attention module, repeating the above spatial and temporal feature fusion process, performing feature enhancement again and completing max pooling to generate the second layer of light change features. This process continues in sequence, passing the second layer output to the third luminance-time attention module. After passing through the same feature enhancement structure and downsampling mechanism, the third layer of light change features is formed, and finally, it enters the fourth luminance-time attention module. After completing feature extraction and max pooling, the fourth layer of light change features is output. The four-layer feature maps respectively represent the analytical ability of the lighting structure under different spatial receptive fields: the first layer focuses on local details, the second layer expands to small-scale regional structures, the third layer further covers the lighting change trends in medium-sized regions, and the fourth layer has the comprehensive expression ability for the global lighting situation. At the end of the encoder, these four layers of light change feature maps are concatenated in the channel dimension, that is, the feature maps of each layer are directly spliced along the channel axis to form a fused tensor with expanded dimensions, constituting multi-level light change features.

[0016] S4, performing smooth transition light sequence calculation according to the lighting optimal angle parameter group and the current lighting configuration state to obtain a segmented lighting adjustment control instruction; Specifically, the optimal lighting angle parameter group and the current lighting angle are differentially analyzed to identify the specific changes and directions of the two in each rotation dimension. A stability threshold is set as the basis for trigger judgment. If the change in any angle dimension exceeds the threshold, the smooth transition calculation logic is immediately triggered to activate the subsequent dimming action to avoid abrupt changes in the lighting state causing visual discomfort. When the trigger condition is met, the system uniformly constructs the current lighting state and the target lighting state into a standardized vector expression structure, representing the parameter sets of the current configuration and the target configuration respectively. These vectors contain three-axis angle parameters and integrate supplementary lighting attributes such as light intensity, color temperature, and spatial distribution uniformity. According to the functional positioning of the lighting space and the requirements of the environment's response to light changes, the transition method that best suits the current scene is selected, including a linear progressive method, a smooth mode of slow start and slow stop, and a dynamic form that tends to stabilize after an initial rapid response, and the static state parameters are mapped to a light transition path that continuously evolves over time. In order to convert the continuous transition process into a specific instruction that can be executed by the controller, the transition path is discretized. Combined with the total amplitude of the angle change, it is determined how many adjustment stages the transition process is decomposed into, that is, the number of segments of the transition process is set. According to the speed characteristics of the illumination change in the current space environment, the time interval required between each two transition stages, that is, the speed of the adjustment rhythm, is dynamically calculated. If the ambient light changes smoothly, the interval is lengthened. If the change is drastic, the rhythm is accelerated to enhance the response speed. The obtained segmented control sequence is a dynamic instruction set consisting of multiple intermediate states with specific time points, target angles, and illumination configurations. The segmented control sequence is evaluated for the smoothness of illumination changes. By continuously checking the illumination change trend between each adjustment stage, it is calculated whether there are abrupt jumps in the overall change process, and it is determined whether it meets the comfort standard. The parameterized intermediate state information is integrated to output formatted segmented lighting adjustment control instructions.

[0017] S5, planning the motion path of the lamp end effector based on the segmented light adjustment control instruction to obtain an actuator motion instruction set.

[0018] Among them, kinematic modeling is carried out on the three-axis end effector structure adopted by the lamp, the degrees of freedom, maximum stroke range, angular velocity and acceleration limits of each rotating axis are clarified, and the motion coupling relationship between each axis is constructed to form a motion parameter model covering three dimensions of horizontal rotation, vertical pitch and axial rotation. The motion parameter model and the angle change point sequence in the segmented lighting adjustment control instruction are input into the lighting configuration sampler. This sampler is designed based on the conditional generative adversarial network structure, and its core is an adversarial system composed of a generator and a discriminator. The generator continuously tries to generate possible angle configuration samples according to the constraint conditions, while the discriminator judges and screens the validity of the samples according to the specific limitations of the lighting task, such as angle range, spatial occlusion, energy consumption limitation, etc., so as to output a set of lighting angle configuration sample sets that meet the control constraints and conform to the lighting target in the continuous game and optimization. Implement a dynamic weight adjustment mechanism for this sample set, adjust the importance weight of each sample in the decision-making according to the degree of adaptation between the sample and the task target, and introduce an illumination enhancement module to reorganize the sample space distribution, so that the optimal solution not only meets the basic requirements, but also is more accurate and natural in terms of detail performance. After the weight adjustment is completed, perform sample diversity analysis on all configuration samples, exclude redundancy by comparing the differences between different samples in the angle space, and retain a candidate set of solutions with strong representativeness and large differences, and select the optimal lighting angle configuration parameters from them. Integrate the optimal configuration scheme with the segmented lighting adjustment control instruction, construct a control path that evolves from the current angle state according to the established adjustment steps, and construct an actuator motion path point sequence through the time sequence distribution of the trajectory points. Considering the continuity requirement of the actuator in the real motion process and the mechanical motion inertia, apply the quintic spline interpolation algorithm to optimize the trajectory of this path point sequence, so that the changes in angular velocity and angular acceleration between each path segment are continuous and smooth, thus avoiding mechanical shocks and actuator jitters caused by sudden changes at the trajectory turning points. Combine the trajectory optimization results with the drive characteristics of the actuator body to generate a structured actuator motion instruction set. This instruction set is organized based on the time series and specifically includes the target angle values of the three axes, the expected angular velocity, necessary dynamic control parameters, and safety redundancy detection signals corresponding to each time point.

[0019] In one example, ambient light parameters are collected in real time to obtain ambient light perception data, including: The light intensity distribution of each area in the space is collected through a light intensity sensor to obtain light intensity data; the color temperature parameter of the light source in the space is measured through a color temperature sensor to obtain color temperature distribution data; the spectral characteristics in a preset wavelength range are scanned through a multispectral sensor to obtain spectral distribution data; the three-dimensional structure of the space and the position of the object are measured through a depth sensor to obtain spatial depth information data; the moving objects in the space are detected and tracked through a motion sensor to obtain target motion trajectory data; Perform time synchronization and data fusion processing on light intensity data, color temperature distribution data, spectral distribution data, spatial depth information data, and target motion trajectory data to obtain ambient light perception data.

[0020] In this example, an optical intensity sensor network is deployed, with high-sensitivity optoelectronic components as the core units. Such sensors have a large dynamic response range and a fast sampling period, and can perform high-frequency acquisition of the local illuminance states in different regions of the space. They are hierarchically arranged according to the spatial functional areas in their distribution, and higher-density sensors are deployed in areas with high lighting demand accuracy to ensure the integrity of spatial coverage of sampling and the balance of regional illuminance resolution. Each sampling unit returns the light intensity changes it captures to the central processing module in numerical form. After preliminary filtering, error compensation, and normalization processing, a structurally unified light intensity data matrix is formed, which constitutes a basic description of the current spatial lighting distribution. At the same time, color temperature sensors are deployed to monitor the luminous characteristics of different light sources in the space in real time. These sensors are based on three-channel photosensitive units, and the corresponding color temperature values are calculated by comparing the light intensity ratios in the red, green, and blue bands. Their measurement range is sufficient to cover the full color temperature range from warm white light to cold white light. The system constructs a two-dimensional or three-dimensional color temperature distribution map through color temperature sampling at multiple points. This map intuitively reflects the hue changes brought by light sources in different regions of the space, and provides a basis for color distribution for the subsequent formulation of lighting angle adjustment strategies by marking its balance degree, mutation regions, etc. To obtain more detailed information on the composition of the light spectrum, a multispectral sensor module is introduced. This module uses a grating spectrometer or an interference filter array as the core optical structure, covers the entire visible light band, and samples the light energy at fixed wavelength intervals to form a complete spectral reflection curve at each observation point. The collected data can not only identify the spectral differences between different light source types but also effectively identify the combined characteristics under mixed lighting of natural light and artificial light, and is applicable to spaces with high requirements for color consistency. The spectral distribution data is obtained through multi-point scanning to form a regional spectral change map, providing an adjustment basis for the intelligent lighting system based on light quality optimization. On the basis of the completion of light characteristic acquisition, to establish a physical relationship model between the space and lighting, a depth sensor based on structured light or TOF technology is introduced to measure the three-dimensional geometric relationship and relative position of each surface in the space in real time. The depth map output by this type of sensor can describe the distance information of each corner, identify the positional relationship of the light-blocking structure and the influence of the spatial structure on the light propagation path, thus supporting the system to establish a more realistic light field projection model. The continuously updated depth information can also dynamically track the lighting area changes caused by the movement of furniture or equipment in the space. To endow the lighting system with the ability to perceive the activity state of people and achieve lighting prediction and response based on behavior patterns, motion sensors of the pyroelectric infrared or millimeter-wave radar type are deployed. Using their detection ability for human thermal radiation or motion reflection signals, the appearance, movement path, and staying behavior of people in the space are recorded and analyzed to form target motion trajectory data.Perform time synchronization on the light intensity data, color temperature distribution data, spectral distribution data, spatial depth information data, and target motion trajectory data. Based on a high-precision clock, timestamp all data streams, and adopt a time window synchronization mechanism to crop and align multi-dimensional data to make it meet spatial coordination within the same time segment. To eliminate sampling errors and random interferences, adopt a multi-filtering strategy. Use Kalman filtering for light intensity and color temperature data to enhance time consistency, use principal component analysis for spectral data to reduce redundant dimensions while retaining representative wavelength components, use a boundary compensation algorithm for depth data to repair occluded areas, and combine a prediction model for motion trajectory data to smooth the human movement curve. After preprocessing, standardize and convert all data. Through a fusion processing module, map all perceptual data to a unified spatial coordinate system, and adopt a fusion mechanism based on weight assignment and the minimum criterion of information entropy to perform multi-dimensional fusion on the five types of data: light intensity, color temperature, spectrum, depth, and motion. This fusion not only realizes information complementarity at the feature level but also forms a unified ambient light field perception model at the decision-making level, enabling the illumination state at any position to be completely defined as a composite parameter vector, including multiple elements such as illuminance level, light color attribute, spectral composition, spatial openness, and dynamic activity, and finally output ambient light perception data.

[0021] In one example, perform multi-scale spatial partitioning and feature fusion on the ambient light perception data to obtain a three-dimensional ambient light distribution map, including: Grid partition the ambient light perception data according to spatial functions to obtain a basic ambient light distribution matrix; Perform 2×2 neighborhood aggregation processing on the basic ambient light distribution matrix to obtain a first-scale distribution map; perform 4×4 neighborhood aggregation processing on the basic ambient light distribution matrix to obtain a second-scale distribution map; perform 8×8 neighborhood aggregation processing on the basic ambient light distribution matrix to obtain a third-scale distribution map; Perform weighted fusion on the first-scale distribution map, the second-scale distribution map, and the third-scale distribution map to obtain a multi-scale spatial light distribution tensor, and calculate historical light change data based on the ambient light perception data; Perform temporal correlation analysis on the multi-scale spatial light distribution tensor and the historical light change data to obtain a three-dimensional ambient light distribution map.

[0022] In this example, during the spatial modeling stage, the overall space is logically partitioned by combining the functional attributes of the actual scene. Different areas are divided into multiple functional units according to their usage purposes, lighting requirements, and personnel activity density, such as work areas, rest areas, display areas, or passage areas, etc. On this basis, the entire three-dimensional space is mapped into a three-dimensional grid structure composed of regular grid cells according to the spatial size and lighting accuracy requirements. Each grid cell is assigned a unique spatial position identifier and serves as a carrier for lighting data. By performing spatial projection on the position of each sensing point in the ambient light perception data, it is made to fall into the corresponding grid, and the lighting parameter information carried by it, such as light intensity, color temperature, spectral components, depth coordinates, and dynamic activity level, etc., is aggregated, averaged, and normalized. Thus, a basic ambient light distribution matrix is constructed. This matrix uses spatial coordinates as the index and composite lighting vectors as the content to form the expression of spatial lighting information. Through a multi-scale analysis mechanism, different levels of neighborhood aggregation processing are performed on the basic ambient light distribution matrix. A 2×2 neighborhood aggregation operation is executed for the smallest local area. Every four adjacent grid cells are aggregated into a new feature cell, and its value is calculated by local weighted average or maximum response method, thereby constructing the first-scale distribution map. This scale map reflects the detailed response ability of the lighting system in the microscopic space and is suitable for detecting brightness changes or local spot features in a small range. A 4×4 neighborhood aggregation process is performed on the basic matrix, and sixteen adjacent units are combined and calculated to generate the second-scale distribution map. This scale is used to extract regional lighting features, such as the distribution relationship between light source groups, the smoothness of color transition, and its mesoscale mapping with the spatial structure. An 8×8 neighborhood aggregation operation is executed to integrate a larger range of grid cells and generate the third-scale distribution map. This level focuses on presenting the macroscopic characteristics of the large-scale lighting pattern, including the overall illuminance tendency of the space, the dominant spectral direction, and the main structural trend of the light field distribution. The first-scale distribution map, the second-scale distribution map, and the third-scale distribution map are weighted and fused to construct a spatial light distribution tensor structure with multi-scale perception ability. Weighted superposition is performed according to the weight coefficients of each scale in the spatial structure analysis. The weight coefficients are jointly determined by the confidence of the ambient light perception data, the severity of lighting changes, and the spatial importance. The higher the weight, the greater the contribution of this scale to the current scene. The fusion process not only compresses and fuses multi-scale features at the spatial level but also performs channel merging operations in dimensions such as spectrum and color temperature to reduce the computational complexity and enhance the structural compactness of the expression, generating a multi-scale spatial light distribution tensor that covers multi-layer spatial features and compresses full-dimensional perception information. To enable this tensor to have time dynamic response ability, the time dimension in the ambient light perception data is introduced into the analysis process to establish the linkage relationship between spatial features and time evolution.Based on the timestamp information in the perception data, a time series window divided at fixed time intervals is established, and the illumination change trend of each spatial grid cell within consecutive time segments is statistically analyzed. By constructing an illuminance change curve, a color temperature shift trajectory, and a spectral component amplitude change, the temporal features of the illumination state are extracted. During this process, an exponential weighting method is adopted to give higher weights to recent data to reflect the dominant change trend of the current environment, and methods such as wavelet transform or moving average are introduced to capture periodic and mutational features. Combining these time-evolving data with the previously generated multi-scale spatial tensors, the system uses the spatial position as the indexing unit, maps and superimposes the feature change vectors within each time period onto the corresponding spatial units, forming a joint representation structure that includes spatial dimension, scale dimension, and time dimension. By performing temporal correlation analysis on the fused tensor structure, the illumination state change trajectories of each region in the space at multiple scales and multiple time periods are identified, thereby constructing a three-dimensional ambient light distribution map.

[0023] In one example, the three-dimensional ambient light distribution map is input into the lighting angle calculation network for ray change feature analysis to obtain the optimal lighting angle parameter group, including: The three-dimensional ambient light distribution map is input into the encoder part of the lighting angle calculation network to perform layer-by-layer downsampling processing to obtain multi-level ray change features; Each layer of ray change features in the multi-level ray change features is input into the luminance attention branch of the lighting angle calculation network for channel attention processing to obtain a spatial feature weight vector; Each layer of ray change features in the multi-level ray change features is input into the temporal attention branch of the lighting angle calculation network for temporal convolutional processing to obtain a temporal feature weight matrix; The spatial feature weight vector and the temporal feature weight matrix corresponding to each layer of ray change features are weighted and fused to obtain multi-level spatio-temporal fusion features; The multi-level spatio-temporal fusion features are input into four transposed convolutional layers in the decoder part of the lighting angle calculation network for feature map size restoration processing to obtain the target lighting angle feature map; The target lighting angle feature map is input into the fully connected layer for mapping and Tanh activation processing to obtain the optimal lighting angle parameter group.

[0024] In this example, a three-dimensional ambient light distribution map is input into the encoder part of the lighting angle calculation network. This distribution map has structurally integrated spatial coordinate information, light intensity changes, color temperature distribution, multi-spectral components, and the lighting state at each time point in the time series, and is expressed in tensor form, which has complete expressive power in the spatial dimension, optical feature dimension, and time dimension. After the input, the encoder part of the network performs layer-by-layer dimensionality reduction processing on the input data through multiple standard convolutional layers with non-linear activation functions and downsampling operations. Each convolutional layer extracts higher-order feature expressions, and the spatial dimension of the feature map is gradually reduced through stride or pooling operations, so as to remove redundant content and compress the computational scale while maintaining the core structural information, generating a multi-level set of light change features. These light change features extracted layer by layer by the encoder respectively correspond to the description of lighting states at different scales and different abstraction levels. The first layer retains the detail changes, and the second to fourth layers gradually abstract the regional change trends and global structural information. The multi-level light features are respectively input into the brightness attention branch for channel attention processing. This branch adopts a global pooling and channel weighting mechanism. First, global average pooling and max pooling are performed on each layer of feature map to compress the spatial information and obtain two sets of channel description vectors, and then the response feature weights of each channel are extracted through a shared fully connected network. The weights reflect the contribution of each channel to the perception of brightness changes in the current feature map, and the normalized result is output through an activation function, thereby generating a spatial feature weight vector for enhancing the response ability to important lighting channels in the feature map. At the same time, the same light change feature of each layer is parallelly input into the time attention branch for convolutional processing based on the temporal structure. This branch takes the temporal convolutional module as the core, extracts the temporal evolution pattern of the lighting features through multiple one-dimensional convolutional kernels along the time axis direction, and models the long-term dependence relationship through the expansion of the temporal receptive field. The self-attention unit connected after the convolutional layer is further used to capture the interdependence degree and response weights between time points, and outputs a temporal feature weight matrix, which describes the importance of features at each time point and depicts the dynamic association degree between different time nodes, enhancing the model's response ability to lighting fluctuations, periodic changes, and sudden illuminance events. The spatial feature weight vector and the temporal feature weight matrix corresponding to the light change feature of each layer are weighted and fused to obtain a spatio-temporal fusion feature with joint modeling. The fusion operation adopts a weighted superposition and broadcasting mechanism, maps the spatial weights in the channel dimension to the dimension matching the temporal weights, and performs element-wise multiplication at the corresponding positions, thereby constructing an enhanced feature map with spatio-temporal linkage relationship. The multi-level spatio-temporal fusion features are input into the decoder part of the lighting angle calculation network. The decoder consists of four consecutive transposed convolutional layers. Each layer performs an upsampling operation on the upper-layer feature map and cascades the original feature map at the corresponding encoder stage through skip connections, so as to restore the original spatial resolution while maintaining the consistency of the structural hierarchy.During the transposed convolution process, the size of the feature map is restored layer by layer, and the structures such as the lighting direction, distribution pattern, and time trend contained therein are further refined and restored until the target light angle feature map is reconstructed. This feature map retains all the semantic information extracted during the process from input to output in terms of structure, and has a complete spatial structure, lighting channel intensity, and time variation trend. The target light angle feature map is input into the fully connected layer at the end of the network, and the fully connected layer maps the high-dimensional feature map into a low-dimensional vector. Each dimension in this vector corresponds to the control quantity for the angle adjustment of the lamp, including the horizontal rotation angle, vertical pitch angle, axial rotation angle, etc., and the output is normalized through the Tanh activation function, making the output range stably fall within the controllable interval, facilitating the subsequent execution by the actuator, and at the same time suppressing the interference caused by numerical anomalies. The finally output optimal light angle parameter group serves as the optimal configuration instruction under the current lighting state.

[0025] In one example, the three-dimensional ambient light distribution map is input into the encoder part of the light angle calculation network to perform layer-by-layer downsampling processing, obtaining multi-level light change features, including: The three-dimensional ambient light distribution map is input into the encoder part of the light angle calculation network, and initial feature extraction is performed through the 3×3 convolutional layer and PReLU activation function in the encoder part to obtain the initial feature map; The initial feature map is input into the first luminance-time attention module in the encoder part for feature enhancement and max-pooling processing to obtain the first-level light change feature; The first-level light change feature is input into the second luminance-time attention module in the encoder part for feature enhancement and max-pooling processing to obtain the second-level light change feature; The second-level light change feature is input into the third luminance-time attention module in the encoder part for feature enhancement and max-pooling processing to obtain the third-level light change feature; The third-level light change feature is input into the fourth luminance-time attention module in the encoder part for feature enhancement and max-pooling processing to obtain the fourth-level light change feature; The first-level light change feature, the second-level light change feature, the third-level light change feature, and the fourth-level light change feature are subjected to channel dimension concatenation processing to obtain multi-level light change features.

[0026] In this example, the three-dimensional ambient light distribution map is input into the encoder part of the lighting angle calculation network. The encoder part is designed as a layer-by-layer downsampling architecture. Its starting processing layer is a convolutional layer composed of 3×3 convolutional kernels. This layer serves as the initial feature extraction channel and scans the input tensor in the spatial and temporal dimensions through a sliding window mechanism to effectively identify underlying structures such as illuminance gradient boundaries, sudden changes in illumination, and spot distribution patterns. At the same time, it extracts the combined pattern of spectral components and color temperature in the local area, completes a round of feature channel expansion and information reconstruction operations, and then performs a non-linear transformation on the convolutional output through the PReLU activation function to enhance the network's expression ability for data in the negative value domain and the gradient circulation efficiency, obtaining the initial feature map. The initial feature map is input into the first group of luminance-time attention modules in the encoder structure. This module simultaneously focuses on the luminance distribution differences in the spatial dimension and the dynamic change characteristics in the temporal dimension. Its internal structure consists of two branches. The first branch enhances the selectivity of the luminance channels through a channel attention mechanism. The specific operations include global average pooling and max pooling of the feature map to extract channel statistical information at different scales, and then outputting the response weight coefficients of each channel through a group of fully connected layers with shared weights to emphasize the channel features that play a dominant role in the current lighting state. The second branch captures the change trend of light in continuous time segments through a temporal convolutional network, identifies short-term periodic fluctuations, sudden change events, and key points of gradual changes through convolutional operations and attention mechanisms unfolded in the temporal dimension, and maps the temporal weight matrix to the dimension fused with the spatial channels. Finally, the results of the two branches form a fused feature map through weighted superposition and broadcast mechanisms, enhancing the model's perception ability of the spatial-temporal changes in the current lighting state. The fused feature map is subjected to max pooling processing to achieve the aggregation of local responses and compress the size of the feature map through size reduction, obtaining the first-layer light change feature map. The first-layer light change feature is input into the second luminance-time attention module. Structurally, it continues the dual-branch strategy of the previous module. However, through the expansion of the parameter sharing range and the increase in the convolutional receptive field, the attention mechanism of the second module can identify lighting patterns within a larger spatial range and a longer time span than the previous layer, enhancing the modeling ability for regional lighting distribution changes and long-time sequence evolution laws. Similarly, after attention enhancement and pooling processing, the second-layer light change feature map is obtained. The second-layer feature map is input into the third luminance-time attention module, which aims to further expand the spatial reception range and the depth of temporal analysis. By increasing the number of convolutional layers and channels and introducing residual connections to improve the information transmission efficiency, a deeper spatial-temporal feature fusion structure is constructed. This layer incorporates larger-scale spatial lighting structures, such as the layout of light source arrays, the distribution of bright and dark areas caused by spatial occlusion, and the long-term illuminance attenuation trend, into the modeling scope. After completing the third-layer attention enhancement and downsampling processing, the output third-layer light change feature map reflects the global lighting structure with a higher semantic abstraction level.Input the third-layer light change features into the fourth luminance-temporal attention module in the encoder part. This module captures the long-term change patterns of the ambient lighting system and the global configuration structure of the lighting environment through deeper convolutional stacking, wide-field receptive field settings, and sequential memory mechanisms in the temporal structure. Combining with the cross-channel fusion mechanism, it realizes the encoding expression of complex lighting states, maximally compresses the original information, and retains the key pattern structure. After completing the last layer of attention enhancement and max pooling, it outputs the fourth-layer light change feature map. Perform channel dimension concatenation processing on the first-layer light change features, the second-layer light change features, the third-layer light change features, and the fourth-layer light change features, and splice all the feature maps in the channel direction to form a multi-level feature fusion tensor, obtaining multi-level light change features.

[0027] In an example, perform smooth transition light sequence calculation according to the optimal light angle parameter group and the current light configuration state, and obtain segmented light adjustment control instructions, including: Perform difference analysis on the optimal light angle parameter group and the current light angle to obtain the angle change amount and the change direction parameter; Compare and judge the angle change amount with a preset stability threshold. When the angle change amount is greater than the preset stability threshold, trigger the light transition calculation to obtain a light transition trigger signal; Perform parameterization processing on the current light configuration state and the target light configuration state according to the light transition trigger signal to obtain a starting state vector and a target state vector; Perform light transition function calculation based on the starting state vector and the target state vector to obtain a continuous light transition function; Discretize the continuous light transition function according to the angle change amount and the ambient light change rate, and generate a segmented transition control point sequence by dynamically adjusting the buffer depth and buffer width parameters; Calculate the variance of the illuminance change gradient at adjacent time points according to the segmented transition control point sequence to control visual comfort, and generate segmented light adjustment control instructions.

[0028] In this example, the optimal angle parameter group of the light output by the angle calculation network is obtained in real time after each ambient light state update, and a one-to-one comparison is made with the current actual angle state of the lamp. Through vector difference analysis, the change trend and amplitude difference of each angle component in the direction are identified, and an angle change vector containing three components: horizontal rotation angle, vertical tilt angle and axial rotation angle is obtained. At the same time, the direction label that the angle rotation should follow in each dimension is marked for reference in the subsequent actuator path planning. The angle change vector is compared and judged dimension by dimension with the stability threshold. The threshold is determined based on the experience of the human eye's perceptual sensitivity to changes in light direction. Its function is to avoid frequent movement and visual interference of lamps caused by small angle changes. When the judgment result shows that at least one-dimensional angle change exceeds the threshold, the system immediately generates a light transition trigger signal, which will be transmitted to the light transition control module as a logical flag to activate the subsequent progressive adjustment calculation process of the lighting state. When the trigger signal is generated, the current lighting configuration state and the target lighting configuration state are structured and parameterized, and the current angle parameters and lighting properties such as brightness, color temperature, spot distribution and other information are constructed as a starting state vector. At the same time, the target angle and the lighting output characteristics it should achieve are encapsulated as a target state vector. According to the lighting task requirements, environmental constraints and target change patterns, a continuous light transition function is established between the starting state and the target state. The essence of this function is a path trajectory that describes how the lighting state evolves over time. Its shape is fitted using different curve models according to different scenarios. For example, linear interpolation functions are used in conventional places, slow-start and slow-stop smooth functions are suitable for areas sensitive to visual transitions, and fast-responding exponential curves are suitable for dynamic environments that urgently need to adjust the lighting direction. The continuous light transition function is discretized according to the angle change and the ambient light change rate to meet the timing requirements of the control instructions. The buffer depth and buffer width required for light adjustment are comprehensively determined based on the two dynamic parameters, the current angle change and the overall change rate of ambient light. The buffer depth indicates how many stages the light adjustment process will be divided into. The greater the angle change and the more stable the environment, the more stages there will be. The buffer width represents the time span of each stage. The more drastic the ambient light change and the more urgent the system's lighting needs, the shorter the buffer width will be. By dynamically setting these two parameters, the continuous transition function is converted into a segmented transition control point sequence consisting of a series of discrete control points. Each control point in the sequence corresponds to the angle state and lighting output value that the lamp should reach at a specific moment, ensuring that the action is controllable, the rhythm is clear, and the transition is coherent during the physical execution process.To improve the user's visual experience during the light adjustment process, the illuminance change smoothness of the entire segmented control point sequence is evaluated. By statistically analyzing the change in illuminance gradient between every two adjacent time points, the variance of the illuminance change gradient in the entire sequence is calculated and compared with the human eye comfort standard. If the gradient change is too large, the system automatically fine-tunes the change rhythm or target illuminance between control points to make the increase or decrease process of the light intensity smoother and avoid visual impact on the user caused by sudden enhancement or weakening. After completing the illuminance comfort correction, the final control point sequence is formatted into a segmented lighting adjustment control instruction, which includes the angle parameter corresponding to each moment, as well as information such as execution time, illuminance target, color temperature target, and dynamic buffer settings.

[0029] In one example, based on the segmented lighting adjustment control instruction, the motion path of the end effector of the lamp is planned to obtain an actuator motion instruction set, including: Perform kinematic modeling on the three-axis actuator structure of the lamp according to the segmented lighting adjustment control instruction to obtain an actuator motion parameter model; Input the actuator motion parameter model and the lighting task constraint conditions into the lighting configuration sampler, and perform effective configuration analysis through the conditional generative adversarial network in the lighting configuration sampler to obtain a set of lighting angle configuration samples; Perform dynamic weight adjustment on the set of lighting angle configuration samples to obtain a target lighting angle configuration scheme, and perform sample diversity evaluation based on the target lighting angle configuration scheme to obtain the optimal lighting angle configuration parameters; Generate a motion path point sequence of the actuator based on the optimal lighting angle configuration parameters and the segmented lighting adjustment control instruction, and perform quintic spline interpolation trajectory optimization based on the actuator motion path point sequence to obtain an actuator motion instruction set.

[0030] In this example, kinematic modeling is performed on the three-axis actuator used in the luminaire according to the segmented lighting adjustment control instruction. The three-axis mechanism includes a horizontal rotation axis, a vertical tilt axis, and an axial rotation axis. Each axis corresponds to an independent rotational degree of freedom and can achieve omnidirectional positioning control of the light source. Multiple parameter items such as the rotation range, maximum angular velocity, maximum angular acceleration, response inertia, and drive power are established for each axis, and the coupling relationship, structural limitations, and redundancy compensation mechanism between the axes are unified and integrated into a set of solvable motion parameter models. This model expresses the mapping relationship between the angle input and the drive output at any time point in the form of a vector function. The actuator motion parameter model and the segmented lighting adjustment control instruction are input into the lighting configuration sampler. The core logic of this sampler is constructed based on the conditional generative adversarial network structure and includes a generator module and a discriminator module. Among them, the generator synthesizes multiple candidate angle configuration samples under the constraint conditions, and the discriminator identifies the validity of the generated samples and feedbacks the optimization direction. During the operation of the network, each target state in the segmented control instruction is used as a conditional input, and at the same time, the constraint conditions set in the lighting task are superimposed, including angle boundaries, safety limitations, occlusion avoidance, energy-saving strategies, etc. These information are jointly encoded into a conditional vector to guide the generator to generate a series of lighting angle configuration samples according to physical reachability and lighting target rationality. Through continuous iteration of adversarial training, the system gradually improves the quality and diversity of the generated angle combinations in the sample space, forming a set of lighting angle configuration sample sets with high feasibility and adjustment value under the current lighting task and actuator structure conditions. Dynamic weight adjustment is performed on the lighting angle configuration sample set. Corresponding optimization weights are assigned to each sample based on the difference in the matching degree between the samples and the lighting target. At the same time, a lighting distribution response model is introduced to evaluate key performance indicators such as the spot coverage range, energy consumption efficiency, and illuminance uniformity. According to the strength of the performance response, the influence of the sample in the next round of selection is dynamically increased or weakened. By continuously adjusting the weight distribution, a set of angle configuration schemes that are closer to the target state and take into account controllability and energy conservation are gradually converged. To avoid the selected samples being highly concentrated or falling into local optima, sample diversity evaluation is performed on the adjusted target configuration scheme. This evaluation calculates the relative distance, distribution balance degree, and structural dispersion index between the samples in the angle space, identifies redundant samples, and preferentially retains representative configurations with strong differences and wide adaptability, and screens out the optimal lighting angle configuration parameters as the input target for path planning. The optimal angle configuration parameters and the segmented lighting adjustment control instruction are coupled and analyzed. According to each segmented target time point and its corresponding angle state, combined with the current actuator structure model, a complete three-axis angle change sequence, that is, the actuator motion path point sequence, is derived. This sequence describes a series of angle states required to start from the current lighting direction, pass through each transition stage until the final target direction.To improve the continuity of the movement process and the smoothness of the controller response, trajectory optimization is performed on this sequence of path points. The quintic spline interpolation method is used to insert smooth transition curves between each segment of the path, enabling the path to have continuous derivatives in the three-dimensional parameters of angle, angular velocity, and angular acceleration, and avoiding phenomena such as actuator oscillation, response lag, or control errors caused by angle jumps or sudden speed changes. The continuous angle trajectory after spline optimization is segmented and encoded, and combined with the sampling time interval and the actuator response period, it is converted into a structured instruction sequence to form the final actuator motion instruction set. This instruction set contains the three-axis angle values required at each moment and embeds speed buffering, motion look-ahead adjustment, limit angle verification, and exception handling flag information.

[0031] Referring to Figure 2 , this embodiment provides an automatically sensing light angle adjustment control device, including: Acquisition module 1, used to collect environmental light parameters in real time to obtain environmental light perception data; Feature fusion module 2, used to perform multi-scale spatial partitioning and feature fusion on the environmental light perception data to obtain a three-dimensional environmental light distribution map; Feature analysis module 3, used to input the three-dimensional environmental light distribution map into a light angle calculation network for ray change feature analysis to obtain an optimal light angle parameter set; Sequence calculation module 4, used to perform smooth transition light sequence calculation according to the optimal light angle parameter set and the current light configuration state to obtain a segmented light adjustment control instruction; Path planning module 5, used to perform motion path planning on the end actuator of the lamp based on the segmented light adjustment control instruction to obtain an actuator motion instruction set.

[0032] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to that described in the above method embodiment, and details will not be elaborated here.

[0033] This invention embodiment also provides an illumination device that executes the above method.

[0034] It should be noted that in this article, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, device, article, or method including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, device, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, device, article, or method including that element.

[0035] The above are only the preferred embodiments of the present invention, and do not thereby limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall similarly be included within the patent protection scope of the present invention.

Claims

1. An automatic induction-based lighting angle adjustment control method, characterized in that Including: Collecting environmental light parameters in real time to obtain environmental light perception data; Performing multi-scale spatial partitioning and feature fusion on the environmental light perception data to obtain a three-dimensional environmental light distribution map; Inputting the three-dimensional environmental light distribution map into a lighting angle calculation network for analyzing light change characteristics to obtain an optimal lighting angle parameter set; Performing a smooth transition light sequence calculation according to the optimal lighting angle parameter set and the current lighting configuration state to obtain a segmented lighting adjustment control instruction; Performing a motion path planning on the end effector of the lighting fixture based on the segmented lighting adjustment control instruction to obtain an actuator motion instruction set.

2. The automatic sensing light angle adjustment control method according to claim 1, characterized in that The collecting environmental light parameters in real time to obtain environmental light perception data includes: Collecting the light intensity distribution in each area of the space through a light intensity sensor to obtain light intensity data; measuring the color temperature parameter of the light source in the space through a color temperature sensor to obtain color temperature distribution data; scanning the spectral characteristics in a preset wavelength range through a multi-spectral sensor to obtain spectral distribution data; measuring the three-dimensional structure of the space and the position of objects through a depth sensor to obtain spatial depth information data; detecting and tracking moving objects in the space through a motion sensor to obtain target motion trajectory data; Performing time synchronization and data fusion processing on the light intensity data, the color temperature distribution data, the spectral distribution data, the spatial depth information data, and the target motion trajectory data to obtain environmental light perception data.

3. The automatic induction light angle adjustment control method according to claim 1, characterized in that The performing multi-scale spatial partitioning and feature fusion on the environmental light perception data to obtain a three-dimensional environmental light distribution map includes: Partitioning the environmental light perception data into a grid according to spatial functions to obtain a basic environmental light distribution matrix; Performing 2×2 neighborhood aggregation processing on the basic environmental light distribution matrix to obtain a first-scale distribution map; performing 4×4 neighborhood aggregation processing on the basic environmental light distribution matrix to obtain a second-scale distribution map; performing 8×8 neighborhood aggregation processing on the basic environmental light distribution matrix to obtain a third-scale distribution map; Performing weighted fusion on the first-scale distribution map, the second-scale distribution map, and the third-scale distribution map to obtain a multi-scale spatial light distribution tensor, and calculating historical light change data according to the environmental light perception data; Performing a temporal correlation analysis on the multi-scale spatial light distribution tensor and the historical light change data to obtain a three-dimensional environmental light distribution map.

4. The automatic induction light angle adjustment control method according to claim 1, characterized in that The inputting the three-dimensional environmental light distribution map into a lighting angle calculation network for analyzing light change characteristics to obtain an optimal lighting angle parameter set includes: Inputting the three-dimensional environmental light distribution map into the encoder part of the lighting angle calculation network to perform layer-by-layer downsampling processing to obtain multi-level light change characteristics; Inputting each layer of the multi-level light change characteristics into the brightness attention branch of the lighting angle calculation network for channel attention processing to obtain a spatial feature weight vector; Inputting each layer of the multi-level light change characteristics into the temporal attention branch of the lighting angle calculation network for temporal convolutional processing to obtain a temporal feature weight matrix; Perform weighted fusion on the spatial feature weight vector and the temporal feature weight matrix corresponding to the light change characteristics of each layer to obtain multi-layer spatio-temporal fusion features; Input the multi-layer spatio-temporal fusion features into the four transposed convolutional layers in the decoder part of the light angle calculation network to perform feature map size restoration processing to obtain the target light angle feature map; Input the target light angle feature map into the fully connected layer for mapping and Tanh activation processing to obtain the optimal light angle parameter group.

5. The automatic sensing light angle adjustment control method according to claim 4, wherein The input of the three-dimensional ambient light distribution map into the encoder part of the light angle calculation network performs layer-by-layer downsampling processing to obtain multi-level light change characteristics, including: Input the three-dimensional ambient light distribution map into the encoder part of the light angle calculation network, and perform initial feature extraction through the 3×3 convolutional layer and the PReLU activation function in the encoder part to obtain the initial feature map; Input the initial feature map into the first luminance-time attention module of the encoder part for feature enhancement and max pooling processing to obtain the first-layer light change characteristics; Input the first-layer light change characteristics into the second luminance-time attention module of the encoder part for feature enhancement and max pooling processing to obtain the second-layer light change characteristics; Input the second-layer light change characteristics into the third luminance-time attention module of the encoder part for feature enhancement and max pooling processing to obtain the third-layer light change characteristics; Input the third-layer light change characteristics into the fourth luminance-time attention module of the encoder part for feature enhancement and max pooling processing to obtain the fourth-layer light change characteristics; Perform channel dimension concatenation processing on the first-layer light change characteristics, the second-layer light change characteristics, the third-layer light change characteristics, and the fourth-layer light change characteristics to obtain multi-level light change characteristics.

6. The automatic induction light angle adjustment control method according to claim 1, wherein The calculation of the smooth transition light sequence according to the optimal light angle parameter group and the current light configuration state to obtain the segmented light adjustment control instruction includes: Perform difference analysis on the optimal light angle parameter group and the current light angle to obtain the angle change amount and the change direction parameter; Compare and judge the angle change amount with the preset stability threshold. When the angle change amount is greater than the preset stability threshold, trigger the light transition calculation to obtain the light transition trigger signal; Perform parameterization processing on the current light configuration state and the target light configuration state according to the light transition trigger signal to obtain the starting state vector and the target state vector; Perform light transition function calculation based on the starting state vector and the target state vector to obtain the continuous light transition function; Perform discretization processing on the continuous light transition function according to the angle change amount and the ambient light change rate, and generate a segmented transition control point sequence by dynamically adjusting the buffer depth and buffer width parameters; Calculate the variance of the illuminance change gradient at adjacent time points according to the segmented transition control point sequence to control the visual comfort, and generate a segmented light adjustment control instruction.

7. The automatic induction light angle adjustment control method according to claim 1, characterized in that Perform motion path planning on the end actuator of the lamp based on the segmented light adjustment control instruction to obtain the actuator motion instruction set, including: Perform kinematic modeling on the lamp three-axis actuator structure according to the segmented lighting adjustment control instruction to obtain an actuator motion parameter model; Input the actuator motion parameter model and the lighting task constraint conditions into the lighting configuration sampler, and perform effective configuration analysis through the conditional generative adversarial network in the lighting configuration sampler to obtain a lighting angle configuration sample set; Perform dynamic weight adjustment on the lighting angle configuration sample set to obtain a target lighting angle configuration scheme, and perform sample diversity evaluation based on the target lighting angle configuration scheme to obtain optimal lighting angle configuration parameters; Generate a path for the optimal lighting angle configuration parameters and the segmented lighting adjustment control instruction to obtain a sequence of actuator motion path points, and perform quintic spline interpolation trajectory optimization based on the sequence of actuator motion path points to obtain an actuator motion instruction set.

8. An automatically inductive light angle adjustment control device, characterized in that, For implementing the steps of the automatic sensing lighting angle adjustment control method according to any one of claims 1 to 7, the automatic sensing lighting angle adjustment control device includes: An acquisition module for real-time acquisition of ambient light parameters to obtain ambient light perception data; A feature fusion module for performing multi-scale spatial partitioning and feature fusion on the ambient light perception data to obtain a three-dimensional ambient light distribution map; A feature analysis module for inputting the three-dimensional ambient light distribution map into a lighting angle calculation network for analysis of light change features to obtain a set of optimal lighting angle parameters; A sequence calculation module for performing smooth transition light sequence calculation according to the set of optimal lighting angle parameters and the current lighting configuration state to obtain a segmented lighting adjustment control instruction; A path planning module for performing motion path planning on the lamp end actuator based on the segmented lighting adjustment control instruction to obtain an actuator motion instruction set.

9. A lighting device, characterized in that, The lighting device is used to execute the steps of the automatic sensing lighting angle adjustment control method according to any one of claims 1 to 7.

Images