A multi-scene adaptive lighting cluster control system and method

Abstract

The present invention relates to the field of lighting control technology, and specifically to a multi-scene adaptive lighting cluster control system and method. The method includes processing monitoring data, actor data and activity data to generate an initial lighting control strategy for various lamps in a commercial block; simulating the control strategy in a scene simulation model, and locally adjusting the control strategy so that the obtained target lighting control strategy fits the actual situation, thereby improving the overall adaptive setting of scene lighting control effect; when unrecorded behavior data is detected from the monitoring data during an activity, a dynamic adjustment model is used to analyze the unrecorded actor or audience behavior data, and the target control strategy is adjusted to ensure that the lighting effect of the lamp cluster fits the actual situation, which helps to solve the problem that traditional lighting systems rely on preset schedules and cannot dynamically adjust lights according to real-time activity needs, resulting in lighting lag after the activity starts or excessive lighting when no one is around.

Description

Technical Field

[0001] The present invention relates to the field of lamp control technology, and in particular to a multi-scene adaptive lamp cluster control system and method. Background Art

[0002] When it comes to lighting control in commercial districts, most still rely on traditional manual control. This requires frequent manual switching and adjustment of lighting modes as events and scene settings change, a method that is inefficient and prone to errors. Furthermore, traditional lighting systems rely on preset schedules (such as timed on / off switches and fixed scene switching) and are unable to dynamically adjust lighting based on real-time event needs. This can lead to lighting lags after an event begins or excessive lighting when no one is around. Furthermore, due to the large scale and diverse types of lighting clusters (such as floodlights, washlights, and lasers), traditional centralized control is prone to communication delays and command conflicts, which can reduce the audience experience.

[0003] Therefore, the present invention provides a multi-scene adaptive lighting cluster control system and method to solve the above problems. Summary of the Invention

[0004] In response to the above situation and to overcome the shortcomings of the existing technology, the present invention provides a multi-scene adaptive lighting cluster control system and method to solve the problem that the above-mentioned traditional lighting system relies on preset schedules (such as timed switches and fixed scene switching) and cannot dynamically adjust the lighting according to real-time activity needs, resulting in lighting lag after the activity starts or excessive lighting when no one is around.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] In a first aspect, a multi-scenario adaptive lighting cluster control method includes:

[0007] Acquire activity data, actor data, and monitoring data; process the activity data and actor data to determine text feature data; analyze scene information in the monitoring data to obtain scene feature data;

[0008] Matching scene simulation models based on scene feature data, text feature data and a preset scene feature library;

[0009] A method combining multimodal fusion and attention mechanism is used to fuse text feature data and scene feature data to determine the fused feature data. A preset strategy matching model constructed by random forest is used to analyze the fused feature data to determine the initial lighting control strategy.

[0010] Control the scene simulation model according to the initial lighting control strategy to obtain lighting simulation state data; use a multi-objective optimization algorithm to optimize the lighting simulation state data to obtain the target lighting control strategy;

[0011] When unrecorded behavior data is detected from the monitoring data during the activity, a dynamic adjustment model is used to analyze the unrecorded behavior data to adjust the target control strategy.

[0012] Preferably, the processing of the activity data and actor data to determine the text feature data includes: suppressing the activity data and actor data using the TF-IDF algorithm to generate a text feature matrix; and processing the text feature matrix using a word vector model to obtain text feature data.

[0013] Preferably, the analysis of scene information in the monitoring data to obtain scene feature data includes: splicing the scene layout data in the monitoring data to obtain a scene layout diagram; and using a target detection algorithm to extract features from the monitoring data and the scene layout diagram to obtain scene feature data.

[0014] Preferably, the matching of the scene simulation model based on the scene feature data, the text feature data and the preset scene feature library includes: using cosine similarity to respectively calculate the similarity between the scene feature data and the preset scene feature library, and the text feature data and the preset scene simulation model template in the preset scene feature library to obtain text matching and scene matching; performing weighted summation on the preset weights, text matching and scene matching to obtain the target matching, and using the preset scene simulation model template corresponding to the target matching ranked first as the scene simulation model.

[0015] Preferably, the method of combining multimodal fusion and attention mechanism to fuse text feature data and scene feature data to determine fused feature data includes: using a fully connected layer with shared parameters to project the text feature data and scene feature data to the same dimension to obtain text projection features and scene projection features; using a cross-modal attention calculation method to process the text projection features and scene projection features to determine the correlation strength between the text features and the scene features to obtain attention weights; performing weighted calculation on the attention weights, text feature data and scene feature data to obtain fused feature data.

[0016] Preferably, the preset strategy matching model constructed using random forest analyzes the fused feature data to determine the initial lighting control strategy, including: each tree splits the path according to the fused feature data and outputs the prediction results of the sub-path; integrating the various prediction results to obtain the majority voting results; and determining the initial lighting control strategy based on the mapping relationship between the majority voting results and the preset strategy library.

[0017] Preferably, the controlling of the scene simulation model according to the initial lamp control strategy to obtain the lighting simulation state data includes: processing the initial lamp control strategy using a Lambertian model to simulate the spatial illumination distribution; integrating the illumination distribution results of each space to obtain the lighting simulation state data; the spatial illumination calculation formula for each position is:

[0018] ,

[0019] in, For location The illumination, is the intensity of the i-th lamp, is the total number of lamps, is the incident angle of the light, For the lamp to the position distance, For location reflectivity.

[0020] Preferably, the multi-objective optimization algorithm is used to optimize the lighting simulation state data to obtain the target lighting control strategy, including: sequentially calculating the uniformity index, color difference index, dynamic effect index and shadow and occlusion index of the lighting simulation state data; determining the corresponding adjustment value based on the difference between each index and the preset standard interval; and adjusting the corresponding parameters in the initial lighting control strategy based on the adjustment value to obtain the target lighting control strategy.

[0021] Preferably, the dynamic adjustment model is used to analyze the unrecorded behavior data to adjust the target control strategy, including: using the isolation forest algorithm to detect the unrecorded behavior data, and when the abnormality score is greater than a preset threshold, triggering the dynamic adjustment mode; the dynamic adjustment mode uses the elastic network regression algorithm to analyze the unrecorded data to obtain a dynamic target control strategy containing correction parameters.

[0022] In a second aspect, the present invention further provides a multi-scene adaptive lighting cluster control system, which is used to execute the multi-scene adaptive lighting cluster control method described in any one of the above technical solutions.

[0023] The beneficial effects of the present invention are:

[0024] 1. The present invention combines scene layout data, actor data and activity data in monitoring data to generate adaptive lighting control methods for various lamps in commercial blocks, and simulates the lighting conditions under the control mode in a scene simulation model, and performs local adaptive adjustments to make the obtained target lamp control strategy more in line with the actual situation, thereby improving the overall adaptive setting of scene lighting control effect; and when an activity is in progress, when unrecorded behavior data is detected from the monitoring data, a dynamic adjustment model is used to analyze the unrecorded actor or audience behavior data to adjust the target control strategy to ensure that the lighting effect of the lamp cluster is in line with the actual situation, which helps to solve the problem that traditional lighting systems rely on preset schedules (such as timed switches, fixed scene switching), cannot dynamically adjust lights according to real-time activity needs, and cause lighting to lag after the start of the activity or over-lighting when no one is around.

[0025] 2. The present invention adopts the attention mechanism to automatically capture the fine-grained association between text and scene features, and adopts a multimodal complementary approach to enable text features to provide semantic information (such as activity type) and scene features to provide spatial information (such as lamp coordinates), thereby obtaining fused feature data, which helps to subsequently improve the accuracy of the output control strategy.

[0026] 3. The present invention can optimize some parameters of the initial lighting control strategy from the perspective of uniformity index, color difference index, dynamic effect index and shadow and occlusion index, thereby improving the adaptability of the lighting control strategy to the activity scenes of commercial blocks, so that the lighting effects controlled by the target lighting control strategy are suitable for current commercial block activities.

[0027] 4. The present invention can use the isolation forest algorithm to detect anomalies in real-time monitoring data. When an abnormal situation occurs and triggers the dynamic adjustment mode, the dynamic adjustment mode uses the elastic network regression algorithm to update some parameters in the target control strategy, so that the adjusted parameters are more in line with the on-site environment, enhancing the experience of actors and audiences. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic flow chart of a multi-scene adaptive lighting cluster control method of the present invention;

[0029] Figure 2 This is a schematic structural diagram of a multi-scene adaptive lighting cluster control system of the present invention. DETAILED DESCRIPTION

[0030] The following will refer to the attached Figure 1 To the attached Figure 2 The embodiments of the present invention are described in detail. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0031] A multi-scene adaptive lighting cluster control method, as shown in the attached Figure 1 As shown, the following steps are included:

[0032] Step S11: Acquire activity data, actor data, and monitoring data.

[0033] Step S12: Process the activity data and actor data to determine text feature data; analyze the scene information in the monitoring data to obtain scene feature data.

[0034] Step S13: Matching a scene simulation model based on the scene feature data, the text feature data and the preset scene feature library.

[0035] Step S14: adopt a method combining multimodal fusion and attention mechanism to fuse the text feature data and the scene feature data to determine the fused feature data.

[0036] Step S15: Analyze the fused feature data using the preset strategy matching model constructed by random forest to determine the initial lighting control strategy.

[0037] Step S16: controlling the scene simulation model according to the initial lamp control strategy to obtain lighting simulation state data.

[0038] Step S17: Use a multi-objective optimization algorithm to optimize the lighting simulation state data to obtain a target lighting control strategy; then adjust the on-site lighting state according to the target lighting control strategy so that the lighting effect adapts to the current commercial block activity scene.

[0039] Step S18: When unrecorded behavior data is detected from the monitoring data during the activity, the unrecorded behavior data is analyzed using a dynamic adjustment model to adjust the target control strategy.

[0040] Among them, activity data includes activity name, activity type, activity time, and activity description data; actor data includes actor name, actor popularity, role, etc.; monitoring data includes scene layout data and lighting layout data.

[0041] The preset scene feature library is provided with a mapping relationship between a feature list formed by features of different attention levels and a preset scene simulation model template. It can match the appropriate scene simulation model according to the attention level of the feature to simulate the lighting control effect.

[0042] Specifically, after acquiring activity data, actor data and monitoring data, the present invention performs text analysis and key feature extraction on the activity data and actor data to obtain text feature data; analyzes the scene information in the monitoring data, and analyzes the scene layout, lamp type, lamp position, prop information, etc. of the commercial block to obtain corresponding scene feature data; then, based on the scene feature data, text feature data and the preset scene feature library, the preset scene simulation model template with the highest matching degree is used as the scene simulation model; after analyzing the fusion feature data using the preset strategy matching model, the initial lamp control strategy of the lamp is determined; then, in the scene simulation model, the lighting effect is simulated according to the initial lamp control strategy to obtain lighting simulation state data; then, a multi-objective optimization algorithm is used to optimize the lighting simulation state data to obtain a target lamp control strategy, and the lamp state on site is adjusted according to the target lamp control strategy, so that the lighting effect adapts to the current commercial block activity scene; when unrecorded behavior data is detected from the monitoring data during the activity, a dynamic adjustment model is used to analyze the unrecorded behavior data to adjust the target control strategy so that the lighting effect conforms to the actual situation on site.

[0043] Through the above-mentioned method, the present invention combines the scene layout data, actor data and activity data in the monitoring data to generate an adaptive lighting control mode for various lamps in the commercial district, and simulates the lighting conditions under the control mode in the scene simulation model, and performs local adaptive adjustments to make the obtained target lamp control strategy more in line with the actual situation, thereby improving the overall adaptive setting of the scene lighting control effect; and when the activity is in progress, when unrecorded behavior data is detected from the monitoring data, a dynamic adjustment model is used to analyze the unrecorded actor or audience behavior data to adjust the target control strategy to ensure that the lighting effect of the lamp cluster is in line with the actual situation, which helps to solve the problem that the traditional lighting system relies on preset schedules (such as timed switches, fixed scene switching), cannot dynamically adjust the lighting according to real-time activity needs, and causes the lighting to lag after the activity starts or over-lighting when no one is there.

[0044] In one embodiment of the present invention, the processing of activity data and actor data to determine text feature data includes: suppressing the activity data and actor data using the TF-IDF algorithm to generate a text feature matrix; and processing the text feature matrix using a word vector model to obtain text feature data.

[0045] Specifically, the TF-IDF algorithm is used to suppress the activity data and actor data. By suppressing high-frequency words (such as "de" and "shi") and highlighting low-frequency keywords, the semantic information related to the current activity theme in the activity data and actor data is extracted. For example, "concert" in the activity name or "lead singer" in the actor role will be given higher weights. Furthermore, the Word2Vec model is used to process the associated words in the text feature matrix to compensate for the deficiencies of TF-IDF in terms of synonyms and context sensitivity, and the text is transformed into text feature data in the form of 300-dimensional dense vectors, which is convenient for subsequent fusion with the scene feature vectors.

[0046] In an embodiment of the present invention, the analysis of the scene information in the monitoring data to obtain scene feature data includes: splicing the scene layout data in the monitoring data to obtain a scene layout map; using a target detection algorithm to extract features from the monitoring data and the scene layout map to obtain scene feature data.

[0047] Specifically, the method of splicing the scene layout data in the monitoring data to obtain a scene layout map is as follows: (1) Establish a three-dimensional coordinate system, unify the format of the monitoring data, unify the scale of the monitoring data to the same standard, and then crop the unified standard monitoring data to obtain a local scene map; splice the local scene map in the three-dimensional coordinate system to obtain a scene layout map. (2) Obtain the scene layout map planned by the technical personnel.

[0048] The process of using a target detection algorithm to extract features from the monitoring data and the scene layout map to obtain scene feature data is as follows: Use the YOLOv5 target detection algorithm to detect the positions and categories of lamps and props (such as "LED lights" and "neon light strips") in real time, provide spatial parameters for the scene layout, directly describe the physical positions of the lamps with bounding box coordinates, provide spatial constraints for the lighting simulation, and display them in the scene layout map. Furthermore, use the PSPNet semantic segmentation method to process the scene layout map, segment the scene image, and identify functional areas (such as the audience area and the stage area) for subsequent auxiliary strategy formulation. Then, integrate the target positions, categories, and area labels to generate a scene feature vector and obtain the final scene feature data.

[0049] In an embodiment of the present invention, the matching of the scene simulation model based on the scene feature data, text feature data, and preset scene feature library includes: using cosine similarity to calculate the similarity between the scene feature data and the preset scene simulation model template in the preset scene feature library, and between the text feature data and the preset scene simulation model template in the preset scene feature library to obtain the text matching degree and the scene matching degree; performing weighted summation on the preset weight, text matching degree, and scene matching degree to obtain the target matching degree, and using the preset scene simulation model template corresponding to the target matching degree ranked first as the scene simulation model.

[0050] Through the above method, the present invention uses cosine similarity and weighted summation to calculate the matching degree of scene feature data, text feature data and preset scene simulation model template, and can objectively and reasonably determine the preset scene simulation model template with the highest matching degree as the scene simulation model, so as to subsequently simulate the lighting control effect of the initial control strategy in the scene simulation model.

[0051] In one embodiment of the present invention, the method of combining multimodal fusion and attention mechanism to fuse text feature data and scene feature data to determine fused feature data includes: using a fully connected layer with shared parameters to project the text feature data and scene feature data to the same dimension to obtain text projection features and scene projection features; using a cross-modal attention calculation method to process the text projection features and scene projection features to determine the correlation strength between the text features and the scene features to obtain attention weights; performing weighted calculation on the attention weights, text feature data and scene feature data to obtain fused feature data.

[0052] Specifically, due to the inconsistent dimensions of text feature data and scene feature data, they need to be mapped to the same space when performing feature fusion for fusion processing. The present invention uses a fully connected layer with shared parameters to project text feature data and scene feature data into the same dimension. It then uses a cross-modal attention calculation method to dynamically measure the correlation strength of different parts of the text feature data and scene feature data in the same dimension to obtain the attention weights of the text features and scene features. It then uses a weighted fusion method to process the attention weights of the text features and scene features, as well as the text feature data and scene feature data, to obtain fused feature data.

[0053] Among them, the attention formula for calculating text projection features and scene projection features through cross-modal attention is:

[0054] ,

[0055] in, is the attention weight of the i-th modality, is a learnable query vector used to measure modality importance, is the feature splicing operation, is the text projection feature, is the scene projection feature, i represents the i-th text projection feature, the total number is n, j represents the j-th scene projection feature, is the total number of projected features in the scene.

[0056] Through the above method, the present invention uses an attention mechanism to automatically capture fine-grained associations between text and scene features, such as the mapping of "stage" locations to corresponding lamps. Furthermore, a multimodal approach is employed to combine text features with semantic information (such as activity type) and scene features with spatial information (such as lamp coordinates). This fused feature data helps improve the accuracy of the output control strategy.

[0057] In one embodiment of the present invention, the preset strategy matching model constructed using random forest analyzes the fused feature data to determine the initial lighting control strategy, including: each tree splits the path according to the fused feature data and outputs the prediction results of the sub-path; the prediction results are integrated to obtain the majority voting results; and the initial lighting control strategy is determined based on the mapping relationship between the majority voting results and the preset strategy library.

[0058] Specifically, the process of building the preset strategy matching model includes:

[0059] Determine the corresponding preset strategy label according to the preset lighting control strategy, and then generate the corresponding result set according to the preset strategy label; obtain a number of artificially or randomly generated fusion feature samples, each sample corresponds to a preset strategy label; the collection of these samples forms a data set, and then divide the data set into a training set and a test set according to a preset ratio (7:3) to ensure a uniform distribution of different scene types, so as to construct a preset strategy matching model.

[0060] When building the preset strategy matching model, the model is initialized first, and the parameters of the random forest are initialized. For example, the number of trees is 100, the maximum depth is 10 (to prevent overfitting), the minimum sample split is 2, and the feature random subset ratio is 24.

[0061] During training, during bootstrap sampling, m samples (e.g., 70% of the original number of samples) are randomly sampled from the training set with replacement, and this is repeated n times to generate multiple sub-datasets.

[0062] During decision tree training, a single decision tree is trained for each subset of the dataset. The training process includes: feature selection: randomly selecting a 24-dimensional subset from the 600-dimensional fused features; node splitting: selecting the optimal splitting feature and threshold based on the Gini index; and recursive growth: repeating the splitting until the maximum depth is reached or the samples become inseparable. All decision trees form a forest, with each tree independently voting (classification) or averaging (regression). Feature importance analysis is then performed, using the Gini importance calculation formula to calculate feature importance and identify key features corresponding to the voting results. A matching strategy is determined based on the mapping between the key features corresponding to the majority voting results and the pre-set matching strategy samples. After hyperparameter tuning and evaluation, the matching model for the pre-set strategy is constructed.

[0063] Through the above method, the present invention uses a preset strategy matching model constructed by random forest to analyze the fused feature data and determine the initial lighting control strategy, which can preliminarily ensure the adaptability of the initial lighting control strategy to the commercial block, so that the adjusted lighting conforms to the actual situation of the commercial block scene.

[0064] Furthermore, by combining distributed control of various lamps, the present invention solves the problem that traditional centralized control is prone to communication delays or command conflicts when the lamp cluster is large in scale and diverse in types (such as floodlights, wash lamps, and lasers), which can easily reduce the audience's experience.

[0065] In one embodiment of the present invention, controlling the scene simulation model according to the initial lighting control strategy to obtain lighting simulation state data includes: processing the initial lighting control strategy using a Lambertian model to simulate spatial illumination distribution; integrating the illumination distribution results of each space to obtain lighting simulation state data; the spatial illumination calculation formula for each position is:

[0066] ,

[0067] in, For location The illumination, is the intensity of the i-th lamp, is the total number of lamps, is the incident angle of the light, For the lamp to the position distance, For location reflectivity.

[0068] In the above manner, the present invention can truly reflect the lighting distribution effect of light in the real world by adopting the Lambertian model as the calculation method of the lighting at each position in the scene simulation model.

[0069] In one embodiment of the present invention, the multi-objective optimization algorithm is used to optimize the lighting simulation state data to obtain a target lighting control strategy, including: sequentially calculating the uniformity index, color difference index, dynamic effect index, and shadow and occlusion index of the lighting simulation state data; determining the corresponding adjustment value based on the difference between each index and a preset standard interval; and adjusting the corresponding parameters in the initial lighting control strategy based on the adjustment value to obtain the target lighting control strategy.

[0070] Specifically, the calculation formula of the uniformity index is: ,

[0071] in, is the uniformity index, is the standard deviation, is the average illumination value of the area; the uniformity index is used to ensure uniform illumination of the target area and avoid excessive differences in brightness and darkness.

[0072] The calculation formula of color difference index is:

[0073] ,

[0074] in, is the color difference index, 、 、 is a parameter of CIE Lab chromaticity value, 、 、 It is another parameter of the CIE Lab chromaticity value. The parameters in the CIE Lab chromaticity value (L∗, a∗, b∗) represent the brightness layer, red-green axis, and yellow-blue axis respectively. The color difference index is used to ensure the consistency of light color of different lamps and avoid color difference.

[0075] The calculation formula for dynamic effect index is:

[0076] ,

[0077] in, is a dynamic effect indicator, indicating smoothness. is the upper limit of the integral, It is a function of the light intensity change curve. The dynamic effect index is used to verify the smoothness and visual effects of dynamic modes (such as gradient and flashing).

[0078] The shadow and occlusion indicators are calculated as follows: a shadow map is generated by the rendering engine, and the ratio of the number of shadow pixels to the total number of pixels is determined based on the shadow map to detect whether the shadow area affects the visual experience.

[0079] Then, the corresponding adjustment value is determined according to the difference between each indicator and the preset standard interval; the corresponding parameters in the initial lamp control strategy are adjusted according to the adjustment value to obtain the target lamp control strategy.

[0080] Through the above method, the present invention can optimize some parameters of the initial lighting control strategy from the uniformity index, color difference index, dynamic effect index and shadow and occlusion index, improve the adaptability of the lighting control strategy to the commercial block activity scene, so that the lighting effect controlled by the target lighting control strategy is suitable for the current commercial block activities.

[0081] In one embodiment of the present invention, the use of a dynamic adjustment model to analyze unrecorded behavior data to adjust the target control strategy includes: using an isolation forest algorithm to detect unrecorded behavior data, and when the anomaly score is greater than a preset threshold, triggering a dynamic adjustment mode; the dynamic adjustment mode uses an elastic network regression algorithm to analyze the unrecorded data to obtain a dynamic target control strategy containing correction parameters.

[0082] Specifically, the formula for anomaly detection using the isolation forest algorithm is:

[0083] ,

[0084] in, is the abnormality score, New monitoring data is used to detect unrecorded behavioral data, such as abnormal data such as sudden changes in crowd density. is the length of the data path in the tree, is the average path length of the training set; when the anomaly score is greater than the preset threshold, the dynamic adjustment mode is triggered.

[0085] The formula for updating the lighting control parameters using the elastic network regression algorithm in dynamic adjustment mode is:

[0086] ,

[0087] in, are the updated model parameters, is the solution to the optimization problem, is the model parameter vector, which represents the strategy parameters that need to be optimized. is a new feature matrix, including real-time data or abnormal data, is the new target strategy vector, 、 are the regularization coefficients of L1 and L2 respectively.

[0088] Through the above method, the present invention can use the isolation forest algorithm to detect anomalies in real-time monitoring data. When an abnormal situation occurs and triggers the dynamic adjustment mode, the dynamic adjustment mode uses the elastic network regression algorithm to update some parameters in the target control strategy, so that the adjusted parameters are more in line with the on-site environment, thereby enhancing the experience of the actors and the audience.

[0089] In one embodiment of the present invention, as shown in the attached Figure 2 As shown, the present invention also provides a multi-scene adaptive lighting cluster control system, which is used to execute the multi-scene adaptive lighting cluster control method described in any one of the above embodiments.

[0090] Specifically including: data acquisition module, obtaining activity data, actor data and monitoring data.

[0091] The feature extraction module processes the activity data and actor data to determine the text feature data; and analyzes the scene information in the monitoring data to obtain the scene feature data.

[0092] The scene model matching module matches the scene simulation model based on scene feature data, text feature data and a preset scene feature library.

[0093] The strategy determination module uses a method combining multimodal fusion and attention mechanism to fuse text feature data and scene feature data to determine the fused feature data; it uses a preset strategy matching model constructed by random forest to analyze the fused feature data and determine the initial lighting control strategy.

[0094] The strategy optimization module controls the scene simulation model according to the initial lighting control strategy to obtain lighting simulation state data; and uses a multi-objective optimization algorithm to optimize the lighting simulation state data to obtain the target lighting control strategy.

[0095] The dynamic adjustment module uses a dynamic adjustment model to analyze the unrecorded behavior data when unrecorded behavior data is detected from the monitoring data during the activity to adjust the target control strategy.

[0096] Specifically, after acquiring activity data, actor data and monitoring data, the present invention performs text analysis and key feature extraction on the activity data and actor data to obtain text feature data; analyzes the scene information in the monitoring data, and analyzes the scene layout, lamp type, lamp position, prop information, etc. of the commercial block to obtain corresponding scene feature data; then, based on the scene feature data, text feature data and the preset scene feature library, the preset scene simulation model template with the highest matching degree is used as the scene simulation model; after analyzing the fusion feature data using the preset strategy matching model, the initial lamp control strategy of the lamp is determined; then, in the scene simulation model, the lighting effect is simulated according to the initial lamp control strategy to obtain lighting simulation state data; then, a multi-objective optimization algorithm is used to optimize the lighting simulation state data to obtain a target lamp control strategy, and the lamp state on site is adjusted according to the target lamp control strategy, so that the lighting effect adapts to the current commercial block activity scene; when unrecorded behavior data is detected from the monitoring data during the activity, a dynamic adjustment model is used to analyze the unrecorded behavior data to adjust the target control strategy so that the lighting effect conforms to the actual situation on site.

[0097] Through the mutual cooperation between the above-mentioned modules, the present invention combines the scene layout data, actor data and activity data in the monitoring data to generate an adaptive lighting control mode for various lamps in the commercial district, and simulates the lighting conditions under the control mode in the scene simulation model, and performs local adaptive adjustments to make the obtained target lamp control strategy more in line with the actual situation, thereby improving the overall adaptive setting of the scene lighting control effect; and when the activity is in progress, when unrecorded behavior data is detected from the monitoring data, a dynamic adjustment model is used to analyze the unrecorded actor or audience behavior data to adjust the target control strategy to ensure that the lighting effect of the lamp cluster is in line with the actual situation, which helps to solve the problem that traditional lighting systems rely on preset schedules (such as timed switches, fixed scene switching), cannot dynamically adjust lights according to real-time activity needs, and cause lighting lags after the activity starts or excessive lighting when no one is around.

[0098] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0099] It should be noted that, in the description of the present invention, the terms "first", "second" and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0100] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0101] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0102] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0103] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0104] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0105] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A multi-scene adaptive lighting cluster control method, characterized in that: include: Obtain activity data, actor data, and monitoring data; Processing activity data and actor data to determine text feature data; analyzing scene information in monitoring data to obtain scene feature data; Matching scene simulation models based on scene feature data, text feature data and a preset scene feature library; A method combining multimodal fusion and attention mechanism is used to fuse text feature data and scene feature data to determine fused feature data, including: using a fully connected layer with shared parameters to project text feature data and scene feature data into the same dimension to obtain text projection features and scene projection features; using a cross-modal attention calculation method to process text projection features and scene projection features, determine the correlation strength between text features and scene features, and obtain attention weights; the attention weight formula for cross-modal attention calculation of text projection features and scene projection features is: ， in, is the attention weight of the i-th modality, is a learnable query vector used to measure modality importance, is the feature splicing operation, is the text projection feature, is the scene projection feature, i represents the i-th text projection feature, the total number is n, j represents the j-th scene projection feature, is the total number of scene projection features; A weighted calculation is performed on the attention weights, text feature data, and scene feature data to obtain fused feature data. The fused feature data is analyzed using a preset strategy matching model constructed using random forests to determine the initial lighting control strategy. This includes: each tree splits the path based on the fused feature data and outputs the prediction results of the sub-paths; the prediction results are integrated to obtain the majority vote result; and the initial lighting control strategy is determined based on the mapping relationship between the majority vote result and the preset strategy library. The scene simulation model is controlled according to the initial lighting control strategy to obtain lighting simulation status data, including: using the Lambertian model to process the initial lighting control strategy to simulate the spatial illumination distribution; integrating the illumination distribution results of each space to obtain lighting simulation status data; the spatial illumination calculation formula for each position is: ， in, For location The illumination, is the intensity of the i-th lamp, is the total number of lamps, is the incident angle of the light, For the lamp to the position distance, For location reflectivity; optimize the lighting simulation state data using a multi-objective optimization algorithm to obtain a target lighting control strategy, including: sequentially calculating the uniformity index, color difference index, dynamic effect index, and shadow and occlusion index of the lighting simulation state data; determining the corresponding adjustment value based on the difference between each index and a preset standard interval; and adjusting the corresponding parameters in the initial lighting control strategy based on the adjustment value to obtain the target lighting control strategy; When unrecorded behavior data is detected from the monitoring data during an activity, a dynamic adjustment model is used to analyze the unrecorded behavior data to adjust the target control strategy, including: using the isolation forest algorithm to detect unrecorded behavior data, and when the anomaly score is greater than the preset threshold, triggering the dynamic adjustment mode; the dynamic adjustment mode uses the elastic network regression algorithm to analyze the unrecorded data to obtain a dynamic target control strategy containing correction parameters.

2. The multi-scene adaptive lighting cluster control method according to claim 1, characterized in that: The processing of the activity data and the actor data to determine the text feature data includes: suppressing the activity data and the actor data using the TF-IDF algorithm to generate a text feature matrix; and processing the text feature matrix using a word vector model to obtain the text feature data.

3. The multi-scene adaptive lighting cluster control method according to claim 1, characterized in that: The analysis of scene information in the monitoring data to obtain scene feature data includes: splicing the scene layout data in the monitoring data to obtain a scene layout diagram; and extracting features from the monitoring data and the scene layout diagram using a target detection algorithm to obtain scene feature data.

4. The multi-scene adaptive lighting cluster control method according to claim 1, characterized in that: The matching of scene simulation models based on scene feature data, text feature data and a preset scene feature library includes: using cosine similarity to respectively calculate the similarity between the scene feature data and the preset scene feature library, and between the text feature data and the preset scene simulation model template in the preset scene feature library, to obtain a text matching degree and a scene matching degree; performing weighted summation on the preset weight, the text matching degree and the scene matching degree to obtain a target matching degree, and using the preset scene simulation model template corresponding to the target matching degree ranked first as the scene simulation model.

5. A multi-scene adaptive lighting cluster control system, characterized in that: The multi-scene adaptive lighting cluster control system includes a processor, and the processor is used to execute the multi-scene adaptive lighting cluster control method according to any one of claims 1 to 4.

Images