An intelligent control system for lighting lamps in environmental monitoring image processing

Abstract

The present invention discloses an intelligent control system for lighting lamps in environmental monitoring image processing, which relates to the field of image processing. The key points of its technical solution include: a mode construction module designs the mode of the lighting lamp according to the rule of expanding the network, and obtains the currently selected mode; a data acquisition module is used to collect comprehensive environmental data, splice and process the comprehensive environmental data, and obtain a comprehensive panoramic atlas; a data aggregation and processing module is used to adaptively segment the comprehensive panoramic atlas, use the K-U clustering method to perform feature aggregation on the segmented image, and obtain a feature data set; a control module analyzes the environment of the sub-region based on the feature data set, generates a feature scheme, and uses PID control to control the lighting lamp according to the feature scheme, so as to realize the intelligent control of the lighting lamp based on environmental monitoring image processing.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and more specifically, it relates to an intelligent lighting control system for environmental monitoring image processing. Background Art

[0002] In some public scenarios, traditional lighting systems often can only provide fixed lighting modes and are difficult to dynamically switch according to different scenario requirements (such as energy conservation, high efficiency, and free adjustment). Traditional lighting control systems may rely on a single sensor (such as a light or occupancy detection sensor) to collect environmental information, resulting in a single source of information and insufficient accuracy and comprehensiveness of regulation. As a result, traditional lighting systems often adopt a unified brightness control method, leading to energy waste.

[0003] In addition, when collecting images of the lighting environment, especially for continuous and uninterrupted collection, multi-camera monitoring devices will generate a large amount of image data, causing difficulties in in-depth mining during the image processing process. Moreover, when analyzing images, due to the characteristics of light source diffusion attenuation and the superposition and interlacing of multiple light sources, the colors and luminances of the acquired images are different, which greatly increases the difficulty of image analysis and also makes lamp source control more difficult.

[0004] Therefore, based on the above problems, it is necessary to design an intelligent lighting control for environmental monitoring image processing. Summary of the Invention

[0005] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide an intelligent lighting control system for environmental monitoring image processing to achieve intelligent control of lighting lamps.

[0006] To achieve the above purpose, the present invention provides the following technical solutions: The intelligent lighting control system for environmental monitoring image processing includes:

[0007] Mode construction module: Design the mode of the lighting lamp according to rule network expansion and obtain the currently selected mode;

[0008] Data acquisition module: Used to collect comprehensive environmental data and splice and process the comprehensive environmental data to obtain a comprehensive panoramic atlas;

[0009] Data aggregation processing module: Used to adaptively segment the comprehensive panoramic atlas, use the K-U clustering method to aggregate the features of the segmented images, and obtain a feature data set;

[0010] Regulation module: Analyze the environment of the sub-region based on the feature data set, generate a feature scheme, and use PID control to regulate the lighting lamp according to the feature scheme.

[0011] Preferably, the construction method of the rule network expansion includes:

[0012] The rule expansion network includes first-level nodes and second-level nodes, and the first-level nodes and the second-level nodes are connected by directed edges;

[0013] The first-level nodes are simulated from the modes of the lighting lamps, and the modes of the lighting lamps are set as the normal mode, the energy-saving mode, the free mode, and the follow mode;

[0014] The second-level nodes are simulated from the rule name, rule description, trigger condition, and processing method. The rule name is set to be the same as the mode of the lighting lamp. The rule description is for handling the regulation of the lighting lamp in different modes. The trigger condition is the selection of the mode, and the processing method is the processing method for the environmental monitoring images based on the mode selection;

[0015] The rule description of the normal mode is that all lighting lamps illuminate with a fixed brightness and color. The energy-saving mode is to distinguish and adjust the brightness and color of the lighting lamps in unoccupied and occupied areas according to the personnel flow. The rule description of the free mode is manual adjustment; the follow mode is to turn off the lighting lamps in unoccupied areas according to the personnel flow and adjust the brightness and color of the lighting lamps in occupied areas;

[0016] The direction of the directed edge is from the first-level node to the second-level node.

[0017] Preferably, the method for collecting comprehensive environmental data includes:

[0018] Obtain the complete space that needs lighting lamp regulation, and divide the complete space into R_K sub-regions;

[0019] Collect images of the environment through the monitoring devices installed in each sub-region, and set the acquisition frequency within the period time to be K_U times;

[0020] The images collected at the same time point in each sub-region form an image data group, and the image data groups within the period time are collected to form an image data set;

[0021] Use one-hot encoding to perform numerical conversion on the sub-regions, and use the numerical codes corresponding to each sub-region as labels to mark the image data set;

[0022] Pool the image data sets of all the marked sub-regions to form comprehensive environmental data.

[0023] Preferably, the method for splicing and processing the comprehensive environmental data to obtain a comprehensive panoramic atlas includes:

[0024] Step A1: Extract the image data sets of each sub-region in the comprehensive environmental data. For each image in each image data group in the image data set, scale it to the same size and convert it into a grayscale image;

[0025] Step A2: Detect the key feature points in each image using the SIFT algorithm, and generate a description vector for each key feature point;

[0026] Step A3: For each key feature point in any one image, calculate its Euclidean distance from the key feature points in other images, search for the key feature points with the Euclidean distance less than the preset segmentation threshold, and record them as corresponding feature points. The key feature points and the corresponding feature points are used as correct matching point pairs;

[0027] Step A4: Based on the correct matching point pairs, use the direct linear transformation algorithm or the least squares method to calculate the homography matrix between the images. Select one image as the reference image, and transform the other images to the coordinate system of the reference image through the homography matrix. Then fuse the transformed images to generate the final panoramic image;

[0028] Step A5: Collect the panoramic images of each sub-region to form a comprehensive panoramic image set.

[0029] Preferably, the method of using the SIFT algorithm to detect the key feature points in each image and generate a description vector for each key feature point includes:

[0030] Set the parameter range of the image as , and discretize it into u_l parameters at fixed intervals within the parameter range . One parameter controls one Gaussian filter. Represent the image in the scale space through the Gaussian filter, obtain the scale images blurred by the Gaussian filter at different scales, and calculate the difference between the scale images at adjacent parameters to obtain the DoG pyramid;

[0031] Set the image , then the blurred image in the scale space representation of the image is , and , where represents the Gaussian filter controlled by the parameter , and , is the pixel point coordinate on the image, represents the abscissa, ordinate.

[0032] Compare each pixel with its surrounding pixels between each layer and the adjacent layer of the DoG pyramid to obtain the local extreme points, and record them as candidate feature points;

[0033] Calculate the principal curvature and the secondary curvature of each candidate feature point. The principal curvature , represents the second-order derivative of the image in the direction, represents the second-order derivative of the image in the direction. The secondary curvature , represents the mixed second-order derivative of the image in the x and y directions. Set a boundary threshold. If both the principal curvature and the secondary curvature of the candidate feature points are greater than the boundary threshold, they are retained; otherwise, they are discarded.

[0034] Collect all the retained candidate feature points and denote them as the key feature points of the image. For each key feature point, take it as the center, select a 16x16 neighborhood window, and divide it into 4x4 sub-regions of 4x4. In each 4x4 sub-region, calculate the sum of the gradient magnitudes in 8 directions. The description vector consists of the sum of the gradient magnitudes in 8 direction intervals, which is a 128-dimensional feature vector.

[0035] Preferably, the method of scaling each image in each image data group in the image dataset to the same size includes:

[0036] Step B1: Set the size of the panoramic image , identify the number of images in the image data group , obtain the average image size as , compare the size of each image in the image data group with the average image size, and take the image with the closest size as the reference image;

[0037] Step B2: In the reference image, randomly select a reference block using the reference frame. According to the image selected by the reference block, identify the same parts in other images in the image data group, and circle them using the similar frame;

[0038] Step B3: The image is enlarged or reduced in the same proportion following the similar frame until the size of the similar frame is the same as the reference frame, and then stop. At this time, the scaled image and the reference image have the same size;

[0039] Step B4: Repeat Step B2 and B3 until all images in the image data group have the same size as the reference image and then stop.

[0040] Preferably, the method of adaptively segmenting the comprehensive panoramic image set includes:

[0041] Step C1: For each panoramic image in the comprehensive panoramic image set, extract the maximum pixel value and the minimum pixel value , set the interval threshold , where and represent the proportion coefficients, represents the set of all pixel points in the image, represents the position in the image and the pixel value at that position;

[0042] Step C2: Randomly select the pixel value at a pixel point as the reference point. Taking the reference point as the center, calculate the pixel difference between it and the adjacent pixel points. Incorporate the pixel points with pixel differences less than or equal to the interval threshold into the cutting range. Continuously calculate until a pixel point with a pixel difference greater than the interval threshold appears, and then stop. Collect all the pixel points incorporated into the cutting range as a small tile.

[0043] Step C3: Repeat Step C2 until all the pixel points in the panoramic image are completely segmented, and then divide a panoramic image into Q_W small tiles to form a tile set.

[0044] Step C4: Repeat Step C3 until all the panoramic images in the comprehensive panoramic image set are completely segmented.

[0045] Preferably, the method for using the K-U clustering method to perform feature aggregation on the segmented image and obtain the feature data set includes:

[0046] For each tile set formed by segmenting each panoramic image in the comprehensive panoramic image set;

[0047] Step D1: Preset the number of clusters K_B value, randomly select K_B small tiles, and calculate the pixel average value of each small tile as the clustering center.

[0048] Step D2: Calculate the distance between the pixel average value of each small tile and the clustering center, and assign the small tile to the nearest clustering center, dividing the small tiles into K_B groups.

[0049] Step D3: Calculate the pixel average value of all small tiles in each group and update it as the new clustering center. Repeat Step D2 until all the clustering centers no longer change.

[0050] Step D4: Mark the small tiles in each group as a cluster group, and collect all the cluster groups as the feature data set.

[0051] Step D5: Repeat Steps D1 to D4 to obtain the feature data sets of all the panoramic images in the comprehensive panoramic image set.

[0052] Preferably, the method for analyzing the environment of the sub-region based on the feature data set and generating the feature scheme includes:

[0053] Construct a machine learning model. The machine learning model is designed based on a deep neural network. Set the input of the model as the feature data set of all sub-regions in the complete space and the lighting layout, and the output as the feature scheme. Use the sample set to train the model until the iteration times are reached, and then obtain the trained model.

[0054] Input the feature datasets of all sub-regions and the layout of lighting fixtures in the current complete space into the trained model to obtain the current feature scheme;

[0055] The sample set consists of the feature datasets of all sub-regions and the layout of lighting fixtures in the complete space, as well as the corresponding feature schemes;

[0056] The feature scheme is presented by horizontally expanding the lighting fixtures of each sub-region and vertically expanding the brightness and color to form a two-dimensional table, and then forming R_K two-dimensional tables to compose the feature scheme.

[0057] Preferably, the method for regulating the lighting fixtures using PID control according to the feature scheme includes:

[0058] Extract the brightness at each power of the lighting fixtures, and use PID control to regulate the brightness and color of each sub-region according to the feature scheme.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] By designing four modes (conventional, energy-saving, free, and follow), combined with the node design and trigger conditions of the rule-expanded network, the system can flexibly adapt to different scenario requirements. In the energy-saving mode and follow mode, the brightness and color are automatically adjusted based on the dynamic detection of personnel flow, saving energy while improving the user experience.

[0061] By dividing sub-regions and using monitoring devices to collect multi-sub-region images, combined with image stitching technology to generate a comprehensive panoramic atlas, the environmental state of the complete space can be comprehensively perceived. Use one-hot encoding to label the sub-regions to achieve regional management of data. It improves the coverage rate and comprehensiveness of environmental information collection, providing data support for subsequent intelligent regulation. It solves the problems of single information source and insufficient space coverage in traditional systems.

[0062] Adopt the SIFT algorithm to extract key feature points, combined with image stitching technology, to efficiently and accurately extract the key features in the environmental images and generate high-quality panoramic atlases. Use image segmentation and K-U clustering methods to refine the feature information in the panoramic images and extract accurate feature datasets. It improves the perception ability of environmental state changes, especially for data processing in complex spaces (such as multi-regions and multi-dynamic scenarios). It provides a more accurate decision-making basis for the regulation of lighting fixtures.

[0063] By constructing a deep neural network model, based on the design of environmental feature datasets and the layout of lighting fixtures, generate real-time feature schemes to achieve intelligent regulation of lighting fixtures. Description of the Drawings

[0064] Figure 1This is a schematic structural diagram of an intelligent lighting control system for environmental monitoring image processing proposed by the present invention;

[0065] Figure 2 This is a schematic method diagram of an intelligent lighting control system for environmental monitoring image processing applied in the present invention. Specific embodiments

[0066] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0067] Embodiment 1

[0068] Referring to Figure 1 and Figure 2 , Embodiment 1 further illustrates an intelligent lighting control system for environmental monitoring image processing proposed by the present invention.

[0069] The present invention is applicable to the following scenarios:

[0070] Intelligent building lighting: Intelligent lighting management for large buildings such as office buildings, shopping malls, and hotels.

[0071] Public facilities: Dynamic lighting adjustment in areas such as subway stations, airports, and parking lots.

[0072] Home intelligent lighting: Combining free mode and follow mode to achieve personalized and energy-saving home lighting.

[0073] Industrial scenarios: Regional lighting management in scenarios such as factory workshops and warehouses.

[0074] Through the implementation of the above technical solutions, the system can achieve efficient, intelligent, and energy-saving lighting solutions in various scenarios, specifically including:

[0075] Mode construction module: Design the mode of the lighting lamp according to the rule network expansion, and obtain the currently selected mode;

[0076] The construction method of the rule network expansion includes:

[0077] The rule network expansion includes a first-level node and a second-level node, and the first-level node and the second-level node are connected by a directed edge;

[0078] The first-level node is simulated from the mode of the lighting lamp, and the modes of the lighting lamp are set as normal mode, energy-saving mode, free mode, and follow mode;

[0079] The secondary nodes are simulated by rule names, rule descriptions, trigger conditions, and processing methods. The rule name is set to be the same as the mode of the lighting fixture. The rule description is for handling the regulation of the lighting fixture in different modes. The trigger condition is the selection of the mode, and the processing method is a method for processing the environmental monitoring images based on the mode selection;

[0080] The rule description for the normal mode is that all lighting fixtures illuminate with a fixed brightness and color. The energy-saving mode is to distinguish and adjust the brightness and color of the lighting fixtures in unoccupied and occupied areas according to the personnel flow. The rule description for the free mode is manual adjustment; the follow mode is to turn off the lighting fixtures in unoccupied areas according to the personnel flow and adjust the brightness and color of the lighting fixtures in occupied areas;

[0081] The direction of the directed edge is from the primary node to the secondary node;

[0082] For example, if the selected mode is the energy-saving mode, the primary node in the rule topology network will be activated according to the mode selection, and the secondary node will be activated through the directed edge. In the secondary node, the rule name is the energy-saving mode, and the rule description is to distinguish and adjust the brightness and color of the lighting fixtures in unoccupied and occupied areas according to the personnel flow. That is, for the entire environment, or in the house, the lighting fixtures in areas where no personnel activities are recognized are turned off, and the brightness and color of the lighting fixtures in the environment with personnel activities are adjusted. The color is the light color, and the processing method is to selectively process the environmental monitoring images according to the selected energy-saving mode.

[0083] The rule description for the free mode is manual adjustment. Specifically, it means that personnel can freely adjust the lights and brightness according to their own behaviors, not limited to the brightness and color of the lighting fixtures in the other three modes.

[0084] Through the energy-saving mode and the follow mode, the system can dynamically sense the personnel flow and the regional status, reduce the brightness or turn off the lighting fixtures in unoccupied areas, and reduce unnecessary energy waste. In occupied areas, the brightness and color are adjusted in combination with the environmental status to ensure the lighting effect while saving energy to the greatest extent. The overall energy consumption is significantly reduced, and the energy utilization efficiency of the system is improved. In large-scale scenarios (such as shopping malls, office buildings, communities), the operating costs can be effectively reduced.

[0085] Data acquisition module: used to collect comprehensive environmental data and perform splicing processing on the comprehensive environmental data to obtain a comprehensive panoramic atlas;

[0086] The methods for collecting comprehensive environmental data include:

[0087] Obtain the complete space that requires lighting control, and divide the complete space into R_K sub-regions; for example, the complete space can be the entire house, and the entire house can be divided into multiple sub-regions such as a work area, a rest area, a training area, and a dining area according to functional or other custom division requirements.

[0088] Collect images of the environment through monitoring devices installed in each sub-region, and set the acquisition frequency within the cycle time to K_U times;

[0089] The images collected at the same time point in each sub-region form an image data group, and the image data groups within the collection cycle time are collected to form an image data set;

[0090] Use one-hot encoding to perform numerical conversion on the sub-regions, and use the numerical codes corresponding to each sub-region as labels to label the image data set;

[0091] Collect the image data sets of all labeled sub-regions to form comprehensive environmental data.

[0092] Suppose there are 4 monitoring devices (such as cameras) installed in the work area, and data is collected at a frequency of K_U = 5 times within a cycle time of 30 seconds. In the present invention, it is default that all monitoring devices perform one collection at the same time point, that is, the time stamps of the images are the same. In this way, 5 image data groups are obtained, and the 5 image data groups within these 30 seconds are collected to form comprehensive environmental data. Therefore, when performing real-time monitoring of the environment, only continuous image collection in the cycle time is required to continuously monitor the entire house.

[0093] The method for splicing and processing the comprehensive environmental data to obtain a comprehensive panoramic atlas includes:

[0094] Step A1: Extract the image data sets of each sub-region in the comprehensive environmental data, scale each image in each image data group in the image data set to the same size, and convert it into a grayscale image;

[0095] If there are obvious color differences between the images, color correction can be performed first, such as using histogram equalization or color transfer algorithm, and then convert the color image into a grayscale image, that is, separate the color image into three channels of red, green, and blue, calculate the grayscale value of each pixel according to the selected conversion method (such as weighted average method), and combine the calculated grayscale values into a new grayscale image.

[0096] Step A2: Use the SIFT algorithm to detect the key feature points in each image, and generate a description vector for each key feature point;

[0097] Step A3: For each key feature point in any one image, calculate its Euclidean distance from the key feature points in other images, search for the key feature points whose Euclidean distance is less than the preset segmentation threshold, and denote them as corresponding feature points. The key feature points and the corresponding feature points are used as correct matching point pairs;

[0098] The preset segmentation threshold can be set according to experimental data analysis or historical data experience analysis.

[0099] Step A4: Based on the correct matching point pairs, use the direct linear transformation algorithm or the least squares method to calculate the homography matrix between the images. Select one image as the reference image, and transform the other images to the coordinate system of the reference image through the homography matrix. Methods such as bilinear interpolation or Lanczos interpolation can be used to resample the transformed images, fill the pixel vacancies generated by the transformation, and fuse the transformed images to generate the final panoramic image;

[0100] Image fusion can use weighted average fusion or feathering. Weighted average fusion is to perform weighted averaging on the pixel values in the overlapping area to eliminate the stitching seam. Commonly used weight functions include linear weights, multi-band fusion weights, etc. Feathering is to perform a smooth transition at the edge of the overlapping area to eliminate obvious stitching traces.

[0101] Step A5: Collect the panoramic images of each sub-region to form a comprehensive panoramic image set.

[0102] The method of using the SIFT algorithm to detect the key feature points in each image and generating a description vector for each key feature point includes:

[0103] Set the parameter range of the image to , and discretize it into u_l parameters at a fixed interval within the parameter range , one parameter controls one Gaussian filter, represent the image in the scale space through the Gaussian filter, obtain the scale maps blurred by the Gaussian at different scales, and calculate the difference between the scale maps at adjacent parameters to obtain the DoG pyramid;

[0104] Set the image , then the blurred image in the scale space representation of the image is , and , where represents the Gaussian filter controlled by the parameter , and , is the pixel point coordinate on the image, represents the abscissa, ordinate.

[0105] The scale of the image is achieved by convolving the original image with Gaussian filters of multiple different parameters (scales). The scale of the Gaussian filter is controlled by a parameter. The larger the value of

[0106] For each layer in the DoG pyramid and between adjacent layers, compare the magnitude of each pixel with its surrounding pixels (including pixels at the same scale and adjacent scales) to obtain local extreme points (maxima and minima), denoted as candidate feature points.

[0107] For the local maximum and local minimum of each pixel point, it is necessary to detect whether the determinant of the Hessian matrix of each pixel point is positive or negative. If the determinant is positive, the point is a local maximum; if the determinant is negative, the point is a local minimum.

[0108] Since there may be some unstable points among the detected candidate feature points, these points need to be filtered. Specifically, it is necessary to calculate the principal curvature and the secondary curvature of each point and filter out the unstable points based on these values.

[0109] Calculate the principal curvature and the secondary curvature of each candidate feature point. The principal curvature is the average of the second-order derivatives in the x and y directions. It reflects the average rate of change of the image in two orthogonal directions. represents the second-order derivative of the image in the direction and is related to the rate of change in the direction. represents the second-order derivative of the image in the direction and is related to the rate of change in the direction. The secondary curvature is half of the square root of the sum of the squares of the differences of the second-order derivatives in the x and y directions and the square of the mixed second-order derivative. It reflects the difference in the rate of change of the image in two orthogonal directions. represents the mixed second-order derivative of the image in the x and y directions and is related to the interaction of the rates of change in the x and y directions. Set a boundary threshold. If both the principal curvature and the secondary curvature of the candidate feature point are greater than the boundary threshold, it is retained; otherwise, it is discarded.

[0110] Collect all the retained candidate feature points and denote them as the key feature points of the image. For each key feature point, take a 16x16 neighborhood window centered on it and divide it into 4x4 sub-regions. In each 4x4 sub-region, calculate the sum of the gradient magnitudes in 8 directions. The description vector consists of the sum of the gradient magnitudes in 8 direction intervals, which is a 4x4x8 = 128-dimensional feature vector.

[0111] For each point, the descriptive vector represents the gradient information of the area around the point. Specifically, it is necessary to calculate the gradient magnitude and direction of the area around the key point. The gradient magnitude is represented as and the gradient direction is represented as , where and respectively represent the partial derivatives in the x and y directions. The gradient direction is divided into 8 direction intervals, and the range of each interval is 45°. For each interval, calculate the sum of the gradient magnitudes within that interval, that is , where . The descriptive vector consists of the sums of the gradient magnitudes of these 8 direction intervals, and this is the descriptive vector generated by the SIFT algorithm.

[0112] This solution uses the SIFT algorithm to extract image feature points and generate 128-dimensional descriptive vectors, effectively enhancing the anti-interference ability of feature point matching. Using image stitching technology, the sub-region images are precisely fused through the homography matrix to generate high-quality panoramic image sets, ensuring the accuracy of data processing. Using an adaptive segmentation method, the panoramic image is flexibly segmented, which can adapt to different image features and avoid errors that may be caused by fixed segmentation methods. The system still has high processing accuracy and robustness in complex environments (such as light changes, dynamic scenes). The accurate panoramic images and feature data provide a reliable basis for subsequent analysis and regulation.

[0113] The method of scaling each image in each image data group in the image dataset to the same size includes:

[0114] Step B1: Set the size of the panoramic image , which can be preset according to experience or actual requirements, or can be preset according to historical or experimental data analysis. Identify the number of images in the image data group, obtain the average image size as , compare the size of each image in the image data group with the average image size, and use the image with the closest size as the reference image;

[0115] Step B2: In the reference image, randomly select a reference block using a reference frame (a circular or square frame). According to the image selected by the reference block, identify the same parts in other images in the image data group and circle them with a similar frame; the similar frame has the same shape as the reference frame, but the size can be different;

[0116] Step B3: The image is enlarged or reduced in the same proportion following the similar frame until the size of the similar frame is the same as the reference frame, and then stop. At this time, the scaled image and the reference image have the same size;

[0117] Step B4: Repeat Step B2 and B3 until all the images in the image data group have the same size as the reference image, then stop.

[0118] For example, assume there are 5 images in the image data group. Select one image as the reference image. If a 1×1 square box is used as the reference box, randomly enclose a piece of image content on the reference image with the reference box as the reference block. Based on the content in the reference box, identify the same parts in other images as the content in the reference box and then enclose them. At this time, the enclosed box is the similar box. Scale the similar box to the same size as the reference box, and the image in the similar box will be scaled accordingly, so that the content sizes of the two images are the same.

[0119] Data aggregation processing module: used for adaptively segmenting the comprehensive panoramic map set, using the K-U clustering method to perform feature aggregation on the segmented images, and obtaining the feature data set;

[0120] The method for adaptively segmenting the comprehensive panoramic map set includes:

[0121] Step C1: For each panoramic map in the comprehensive panoramic map set, extract the maximum pixel value and the minimum pixel value , and set the interval threshold , where and represent the proportion coefficients, represents the set of all pixel points in the image, represents the position in the image and the pixel value at that position;

[0122] Step C2: Randomly select the pixel value at a pixel point as the reference point. Taking the reference point as the center, calculate the pixel difference between it and the adjacent pixel points. Incorporate the pixel points with pixel differences less than or equal to the interval threshold into the cutting range. Continuously calculate until a pixel point with a pixel difference greater than the interval threshold appears, then stop. Collect all the pixel points incorporated into the cutting range as a small tile.

[0123] Step C3: Repeat Step C2 until all the pixel points in the panoramic map are completely segmented, and then divide a panoramic map into Q_W small tiles to form a tile set;

[0124] Step C4: Repeat Step C3 until all the panoramic maps in the comprehensive panoramic map set are completely segmented, then stop.

[0125] The method for using the K-U clustering method to perform feature aggregation on the segmented images and obtaining the feature data set includes:

[0126] For the tile set formed by segmenting each panoramic map in the comprehensive panoramic map set;

[0127] Step D1: Preset the number of clusters \(K_B\), randomly select \(K_B\) small patches, and calculate the pixel average value of each small patch as the clustering center;

[0128] Step D2: Calculate the distance between the pixel average value of each small patch and the clustering center, and assign the small patch to one of the clustering centers with the closest distance, dividing the small patches into \(K_B\) groups;

[0129] Step D3: Calculate the pixel average value of all small patches in each group, update it as the new clustering center, repeat Step D2 until all clustering centers no longer change and then stop;

[0130] Step D4: Mark the small patches in each group as a cluster group, and collect all cluster groups as the feature data set;

[0131] Step D5: Repeat Steps D1 to D4 to obtain the feature data sets of all panoramic images in the comprehensive panoramic image set.

[0132] Regulation module: Analyze the environment of the sub-region based on the feature data set, generate a feature scheme, and use PID control to regulate the lighting lamps according to the feature scheme;

[0133] The method for analyzing the environment of the sub-region based on the feature data set and generating a feature scheme includes:

[0134] Construct a machine learning model. The machine learning model is designed based on a deep neural network. Set the input of the model as the feature data sets of all sub-regions in the complete space and the lighting lamp layout, and the output as the feature scheme. Use the sample set to train the model until the iteration times are reached and then stop to obtain the trained model;

[0135] Input the feature data sets of all sub-regions in the current complete space and the lighting lamp layout into the trained model to obtain the current feature scheme;

[0136] The sample set consists of the feature data sets of all sub-regions in the complete space, the lighting lamp layout, and the corresponding feature scheme;

[0137] The presentation method of the feature scheme is to horizontally expand the lighting lamps in each sub-region and vertically expand them in terms of brightness and color to form a two-dimensional table, and then form \(R_K\) two-dimensional tables to constitute the feature scheme;

[0138] The lighting lamp layout refers to the distribution positions of the lighting lamps in each sub-region, and each lighting lamp can also be manually numbered. It is also possible to preset the brightness and color of the lighting lamps in the sub-region in the unmanned and manned environments in the normal mode, energy-saving mode, and following mode to form a standard specification table;

[0139] The method of regulating the lighting lamp using PID control according to the feature scheme includes:

[0140] Extract the brightness of the lighting lamp at each power, and according to the feature scheme, use PID control to regulate the brightness and color of each sub-region.

[0141] By introducing a deep neural network model, the system can generate a dynamic regulation scheme based on a multi-dimensional feature dataset, improving the intelligence level of regulation. Using the PID control algorithm to adjust the brightness and color of the lighting lamp in each sub-region in real time can effectively reduce the adjustment delay and improve the control accuracy. It improves the response speed of the system to environmental changes and can quickly adapt to dynamically changing scenarios. It realizes refined control of regions, avoids the "one-size-fits-all" control method of traditional systems, and further improves the flexibility of the system.

[0142] For example, first, the user selects the mode of the house, and the selection flows through the rule network to intelligently control the lighting lamps in the whole house.

[0143] When the selected mode is the normal mode, according to the preset brightness and color of the lighting lamps in each sub-region in the normal mode, where the preset can be directly set by the program or can be default. For example, in the normal mode, the brightness and color of the lighting lamp in the rest area are set to 30 and orange respectively, then the lighting lamp is directly adjusted to this value according to the preset.

[0144] When the selected mode is the energy-saving mode, according to the image of the monitoring device, first identify the sub-region where the person is located, and distinguish the sub-region where the person is located from other sub-regions. For example, in the energy-saving mode, when there is a person in the rest area, the brightness and color of the lighting lamp are set to 50 and bright white respectively, and when there is no person, the brightness and color of the lighting lamp are set to 10 and orange respectively, then the lighting lamp at the corresponding position is adjusted and controlled according to the real-time brightness.

[0145] Generally speaking, multiple lighting lamps are set in a sub-region, and each lighting lamp controls the brightness and color at a position. Therefore, by identifying image blocks with the same pixels and clustering the positions with the same brightness together, and then regulating the brightness and color through the lighting lamp at this position, the color and brightness in the space are unified.

[0146] Embodiment 2

[0147] Refer to Figure 1 and Figure 2 , Embodiment 2 further illustrates an intelligent regulation system for lighting lamps for environmental monitoring image processing proposed by the present invention.

[0148] An intelligent regulation method for lighting lamps for environmental monitoring image processing includes the following steps:

[0149] Step S1: Obtain the currently selected mode of the lighting lamp according to the rule-based network expansion.

[0150] Step S2: Collect comprehensive environmental data, perform splicing processing, and obtain a comprehensive panoramic atlas.

[0151] Step S3: Perform adaptive segmentation on each panoramic image in the comprehensive panoramic atlas, divide each panoramic image into multiple small image blocks, and form an image block set.

[0152] Step S4: Use the K-U clustering method to perform feature aggregation on each small image block in each image block set, aggregate the small image blocks with similar pixel values together, and form a feature data set.

[0153] Step S5: Analyze the environment of the sub-region based on the feature data set, and generate a feature solution.

[0154] Step S6: Use PID control to adjust the lighting lamp according to the feature solution.

[0155] An intelligent control system for lighting lamps in environmental monitoring image processing, which is applied to an intelligent control method for lighting lamps in environmental monitoring image processing, includes:

[0156] Mode construction module: Design the mode of the lighting lamp according to the rule-based network expansion, and obtain the currently selected mode.

[0157] Data acquisition module: Used to collect comprehensive environmental data, and perform splicing processing on the comprehensive environmental data to obtain a comprehensive panoramic atlas.

[0158] Data aggregation processing module: Used to perform adaptive segmentation on the comprehensive panoramic atlas, use the K-U clustering method to perform feature aggregation on the segmented image, and obtain a feature data set.

[0159] Regulation module: Analyze the environment of the sub-region based on the feature data set, generate a feature solution, and use PID control to adjust the lighting lamp according to the feature solution.

[0160] Each module is connected by wired and / or wireless means.

[0161] The technical advantages of the present invention are as follows

[0162] Multi-mode flexible control: Support four modes of normal, energy-saving, free, and follow, meeting the needs of multiple scenarios.

[0163] Comprehensive information collection: Through image monitoring equipment and panoramic stitching technology, realize the environmental perception of the complete space.

[0164] Fine-grained image processing: Use the SIFT algorithm to extract feature points, combine image stitching and adaptive segmentation to ensure the accuracy of data processing.

[0165] Intelligent regulation: Based on the feature generation scheme of deep neural network, combined with PID control to achieve precise dynamic adjustment.

[0166] Energy conservation: Significantly reduces energy consumption and improves energy utilization efficiency in energy-saving mode and follow-up mode.

[0167] High robustness and flexibility: The system has strong adaptability to complex scene changes, flexible regulation, and rapid response.

[0168] Modularity and scalability: Facilitate system maintenance and optimization, and are applicable to scenarios of different scales and requirements.

[0169] In addition, according to the embodiments of the present application, the process described in the accompanying drawings of an intelligent lighting regulation system for environmental monitoring image processing can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present application. Of course, the architecture shown in the accompanying drawings of an intelligent lighting regulation system for environmental monitoring image processing is only exemplary. When implementing different devices, adaptive selection or adjustment can be made according to actual needs.

[0170] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0171] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An intelligent control system for lighting lamps in environmental monitoring image processing, characterized in that, The intelligent lighting control system for environmental monitoring image processing includes: Mode construction module: Design the lighting mode according to the rule network expansion, and obtain the currently selected mode; The construction method of the rule network expansion includes: The rule network expansion includes a first-level node and a second-level node, and the first-level node and the second-level node are connected by a directed edge; The first-level node is simulated from the lighting mode, and the lighting modes are set as normal mode, energy-saving mode, free mode, and following mode; The second-level node is simulated from the rule name, rule description, trigger condition, and processing method. The rule name is set to be the same as the lighting mode, the rule description is used to handle the regulation of the lighting in different modes, the trigger condition is the selection of the mode, and the processing method is the processing method of the environmental monitoring image based on the mode selection; Data acquisition module: Used to collect comprehensive environmental data, and splice and process the comprehensive environmental data to obtain a comprehensive panoramic atlas; Data aggregation processing module: Used to adaptively segment the comprehensive panoramic atlas, and use the K-U clustering method to aggregate the features of the segmented images to obtain a feature data set; Regulation module: Analyze the environment of the sub-region based on the feature data set, generate a feature scheme, and use PID control to regulate the lighting according to the feature scheme; Specifically include: Input the feature data sets and lighting layouts of all sub-regions in the current complete space into the trained model to obtain the current feature scheme; Extract the brightness at each power of the lighting, and use PID control to regulate the brightness and color of each sub-region according to the feature scheme.

2. The intelligent lighting control system for environmental monitoring image processing according to claim 1, wherein The construction method of the rule network expansion further includes: The rule description of the normal mode is that all lights illuminate with a fixed brightness and color, the energy-saving mode is to distinguish and adjust the brightness and color of the lights in unoccupied and occupied areas according to the personnel flow, the rule description of the free mode is manual adjustment; The following mode is to turn off the lights in unoccupied areas according to the personnel flow, and adjust the brightness and color of the lights in occupied areas; The direction of the directed edge is from the first-level node to the second-level node.

3. The intelligent lighting control system for environmental monitoring image processing according to claim 2, wherein, The method for collecting comprehensive environmental data includes: Obtain the complete space that needs lighting regulation, and divide the complete space into R_K sub-regions; Collect images of the environment through the monitoring devices installed in each sub-region, and set the acquisition frequency within the cycle time to be K_U times; The images collected at the same time point in each sub-region form an image data group, and the image data groups within the cycle time are collected to form an image data set; Use one-hot encoding to perform numerical conversion on the sub-regions, and use the numerical codes corresponding to each sub-region as labels to mark the image data set; Collect the image data sets of all the marked sub-regions to form comprehensive environmental data.

4. The intelligent lighting control system for environmental monitoring image processing according to claim 3, wherein The method for splicing and processing the comprehensive environmental data to obtain a comprehensive panoramic atlas includes: Step A1: Extract the image data sets of each sub-region in the comprehensive environmental data, scale each image in each image data group in the image data set to the same size, and convert it into a grayscale image; Step A2: Use the SIFT algorithm to detect the key feature points in each image, and generate a description vector for each key feature point; Step A3: For each key feature point in any one image, calculate its Euclidean distance from the key feature points in other images, search for the key feature points whose Euclidean distance is less than the preset segmentation threshold, and denote them as corresponding feature points. The key feature points and the corresponding feature points are used as correct matching point pairs; Step A4: Based on the correct matching point pairs, use the direct linear transformation algorithm or the least squares method to calculate the homography matrix between the images. Select one image as the reference image, and transform the other images to the coordinate system of the reference image through the homography matrix, and fuse the transformed images to generate the final panoramic image; Step A5: Collect the panoramic images of each sub-region to form a comprehensive panoramic image set.

5. The intelligent lighting control system for environmental monitoring image processing according to claim 4, characterized in that, The method of using the SIFT algorithm to detect the key feature points in each image and generate a description vector for each key feature point includes: Set the parameter range of the image to be , and discretize it into u_l parameters at fixed intervals within the parameter range . One parameter controls one Gaussian filter. Through the Gaussian filter, the image is represented in the scale space to obtain scale images blurred by Gaussian at different scales. Calculate the difference between scale images at adjacent parameters to obtain the DoG pyramid; Set image , then the blurred image in the scale space representation of the image is , and , where represents the Gaussian filter controlled by the parameter , and , is the pixel point coordinate on the image, represents the abscissa, ordinate; Between each layer and the adjacent layer of the DoG pyramid, compare the size of each pixel with its surrounding pixels to obtain the local extreme points, which are denoted as candidate feature points; Calculate the principal curvature and the secondary curvature of each candidate feature point. The principal curvature , represents the second-order derivative of the image in the direction, and represents the second-order derivative of the image in the direction. The secondary curvature , represents the mixed second-order derivative of the image in the x and y directions; Set the boundary threshold. If both the principal curvature and the secondary curvature of the candidate feature point are greater than the boundary threshold, keep it; otherwise, discard it; Collect all the remaining candidate feature points as the key feature points of the image. For each key feature point, select a 16x16 neighborhood window centered on it and divide it into 4x4 sub-regions. In each 4x4 sub-region, calculate the sum of the gradient magnitudes in 8 directions. The description vector consists of the sum of the gradient magnitudes in 8 direction intervals, which is a 128-dimensional feature vector.

6. The intelligent lighting control system for environmental monitoring image processing according to claim 5, characterized in that, The method of scaling each image in each image data group in the image dataset to the same size includes: Step B1: Set the size of the panoramic image , identify the number of images in the image data group , obtain the average image size as , compare the size of each image in the image data group with the average image size, and use the image with the closest size as the reference image; Step B2: In the reference image, randomly select a reference block using the reference frame. According to the image selected by the reference block, identify the same parts in the other images in the image data group and circle them using a similar frame; Step B3: The image is scaled up or down proportionally following the similar frame until the size of the similar frame is the same as that of the reference frame, and then stop. At this time, the scaled image has the same size as the reference image; Step B4: Repeat steps B2 and B3 until all the images in the image data group have the same size as the reference image and then stop.

7. The intelligent lighting control system for environmental monitoring image processing according to claim 6, characterized in that, The method of adaptively segmenting the comprehensive panoramic image set includes: Step C1: For each panoramic image in the comprehensive panoramic atlas, extract the maximum pixel value and the minimum pixel value , and set the interval threshold , where and represent the proportion coefficients, represents the set of all pixel points in the image, represents the position in the image and the pixel value at that position; Step C2: Randomly select the pixel value at a pixel point as the reference point. Taking the reference point as the center, calculate the pixel difference between it and the adjacent pixel points. Incorporate the pixel points with pixel differences less than or equal to the interval threshold into the cutting range, and continuously calculate until a pixel point with a pixel difference greater than the interval threshold appears and then stop. Collect all the pixel points incorporated into the cutting range as a small tile; Step C3: Repeat step C2 until all the pixel points in the panoramic image are completely segmented, and then divide a panoramic image into Q_W small tiles to form a tile set; Step C4: Repeat step C3 until all the panoramic images in the comprehensive panoramic image set are completely segmented and then stop.

8. The intelligent lighting control system for environmental monitoring image processing according to claim 7, characterized in that, The method of using the K-U clustering method to perform feature aggregation on the segmented images to obtain a feature dataset includes: For the tile set formed by segmenting each panoramic image in the comprehensive panoramic image set; Step D1: Preset the number of clusters \(K_B\) value, randomly select \(K_B\) small patches, and calculate the pixel average value of each small patch as the clustering center; Step D2: Calculate the distance between the pixel average value of each small patch and the clustering center, and assign the small patch to the nearest clustering center, dividing the small patches into \(K_B\) groups; Step D3: Calculate the pixel average value of all small patches in each group, update it as the new clustering center, repeat Step D2 until all clustering centers no longer change and then stop; Step D4: Mark the small patches in each group as a cluster group, and collect all cluster groups as the feature dataset; Step D5: Repeat Steps D1 to D4 to obtain the feature datasets of all panoramas in the comprehensive panorama set.

9. The intelligent control system for lighting lamps in environmental monitoring image processing according to claim 8, wherein The method for analyzing the environment of sub-regions based on the feature dataset and generating a feature scheme includes: Construct a machine learning model, which is designed based on a deep neural network. Set the input of the model as the feature datasets and lighting layouts of all sub-regions in the complete space, and the output as the feature scheme. Use the sample set to train the model until the iteration number is reached and then stop to obtain the trained model; The sample set consists of the feature datasets and lighting layouts of all sub-regions in the complete space, and the corresponding feature schemes; The presentation method of the feature scheme is to horizontally expand the lighting lamps of each sub-region and vertically expand them by brightness and color to form a two-dimensional table, and then form \(R_K\) two-dimensional tables to form the feature scheme.

Images