LED intelligent control system for scene recognition

Abstract

The application discloses an LED intelligent control system for scene recognition, and relates to the technical field of LED lighting control. The system comprises a multi-modal scene data acquisition unit, a scene matching unit and a decision output unit, the multi-modal scene data acquisition unit is connected with the scene matching unit, and the scene matching unit is connected with the decision output unit. According to different illumination conditions, the weight proportion of each modal data is flexibly adjusted, the working characteristics of each sensor under different illumination are adapted, the rationality and effectiveness of multi-modal data fusion are improved, the normalization processing and error compensation are used to realize the targeted calibration of multi-modal data, the data precision is ensured, the multi-modal data is subjected to light-weight processing, the data processing efficiency is improved under the premise of ensuring the data correlation degree, reliable support is provided for the real-time regulation and control of the LED, different illumination conditions and various use scenes can be adapted, and the application range is wider.

Description

Technical Field

[0001] This invention relates to the field of LED lighting control technology, specifically to an LED intelligent control system for scene recognition. Background Technology

[0002] LED lights are luminaires that use light-emitting diodes as their light source. They are now widely used in the lighting industry. However, current LED lighting control either relies on active human control or on single-modal parameters (such as relying solely on light sensors or human body sensors) for lighting control. This makes it difficult to adapt to complex application scenarios. Some existing solutions have introduced the concept of multimodal data acquisition, attempting to improve scene recognition capabilities by integrating multiple types of data, but the following shortcomings still exist:

[0003] I. Existing LED lighting control based on multimodal data often adopts a fixed weight allocation mode, which does not consider the impact of light intensity environment on the reliability of different modal data. For example, visual sensors are prone to distortion in strong light environment, and the accuracy of human feature acquisition decreases in low light environment. Fixed weight cannot adapt to this difference, resulting in poor multimodal data fusion effect and obvious scene recognition deviation.

[0004] Second, the lack of a targeted error calibration mechanism in the multimodal data fusion process makes it easy for sensor inherent errors and environmental interference signals to mix into the effective data, causing data distortion and affecting scene positioning accuracy.

[0005] Third, the data processing flow is crude, and the multimodal data has not been refined and lightweighted. There is too much redundant information and invalid data, which increases the system's computing load and makes it difficult to meet the real-time control requirements.

[0006] Fourth, the scenario determination relies solely on the currently collected data without considering historical behavioral patterns, making it prone to erroneous adjustments due to occasional data fluctuations, resulting in insufficient stability.

[0007] Fifth, LED parameter adjustments are mostly abrupt switching, lacking a smooth transition mechanism, resulting in a poor user experience. Summary of the Invention

[0008] In view of the deficiencies in the existing technology, the technical problem to be solved by this application is: how to achieve accurate identification of LED lamp application scenarios and real-time control of LED lamps.

[0009] To achieve the above objectives, embodiments of this application provide an LED intelligent control system for scene recognition, comprising:

[0010] A multimodal scene data acquisition unit is provided, which is connected to a scene matching unit. The multimodal scene data acquisition unit is used to acquire multimodal scene data and send the multimodal scene data to the scene matching unit.

[0011] A scene matching unit is connected to a decision output unit. The scene matching unit is used to determine the suspected scene category based on multimodal scene data and transmit the suspected scene category determination information to the decision output unit.

[0012] The decision output unit is used to determine the final scene category based on the suspected scene category judgment information, and outputs and executes the LED target control parameters when the LED control command is triggered.

[0013] In one embodiment, the multimodal scene data acquisition unit includes a multimodal scene data acquisition module, a weight allocation module, and an error compensation module;

[0014] The multimodal scene data acquisition module is used to acquire multimodal scene data, which includes human feature parameters, environmental parameters, and audio feature parameters.

[0015] The weight allocation module is used to determine the light intensity environment based on environmental parameters, allocate weights to the multimodal scene data acquisition sources based on the light intensity environment, and obtain the fused feature vector based on the weight allocation result and the multimodal scene data.

[0016] The error compensation module is used to perform error calibration on the fused feature vector and obtain calibration data.

[0017] In one embodiment, the process of determining the light intensity environment based on environmental parameters includes:

[0018] Preset light intensity thresholds for low-light, normal, and high-light environments;

[0019] The light intensity is obtained from environmental parameters, and the current light intensity environment is determined based on the light intensity.

[0020] In one embodiment, the process of obtaining the fused feature vector includes:

[0021] Pre-establish the weights of each dimension parameter under the corresponding light intensity environment;

[0022] After obtaining the signal strength of the multimodal scene data acquisition source, normalization processing is performed to obtain the normalized value of the multimodal scene data acquisition source.

[0023] After determining the weights of the corresponding multimodal scene data based on the light intensity environment, the fused feature vector is obtained by combining the normalized value.

[0024] In one embodiment, the scene matching unit includes a feature preprocessing module, a parameter optimization module, and a scene matching module;

[0025] The feature preprocessing module is used to perform noise reduction and normalization on the calibration data to obtain preprocessed calibration data;

[0026] The parameter optimization module is used to perform lightweight processing on the preprocessed calibration data to obtain the final calibration data.

[0027] The scene matching module is used to match the final calibration data with a preset scene feature library to obtain the suspected scene category corresponding to the final calibration data.

[0028] In one embodiment, the lightweighting process includes:

[0029] Preset a variance threshold for the corresponding light intensity environment, and remove preprocessed calibration data with variance lower than the variance threshold;

[0030] Preset a signal strength threshold for human feature parameters, and remove human feature parameters that are below the threshold;

[0031] Preset signal strength thresholds for environmental parameters and audio feature parameters, and remove environmental parameters and audio feature parameters that are below these thresholds;

[0032] Based on the weights of parameters in each dimension under light intensity conditions and the remaining preprocessed calibration data, the final calibration data is obtained.

[0033] In one embodiment, the decision output unit includes a confidence response module and a smoothing control module;

[0034] The confidence response module is used to determine the final scene category based on the suspected scene category judgment information and to determine whether to trigger the LED control command.

[0035] The smooth control module is used to receive LED control commands, generate LED target control parameters, and control the LED according to the LED target control parameters.

[0036] In one embodiment, the process of determining whether an LED control command is triggered includes:

[0037] Calculate the confidence score for the suspected scene category. The confidence score is calculated using the following formula:

[0038] C = α·M + β·F;

[0039] In the formula, C is the confidence level, M is the similarity between the current final calibration data and the suspected sample, F is the frequency of the current product group within the historical set time period, α and β are both weight coefficients, and α+β=1;

[0040] Determine if the confidence level exceeds a preset threshold. If so, determine the suspected scene category with the highest confidence level as the final scene category and trigger the LED control command; otherwise, determine the suspected scene category with the highest confidence level as the final scene category and do not trigger the LED control command.

[0041] In one embodiment, the process of obtaining the frequency of occurrence of the current product group within a historical set time period includes:

[0042] After determining the normalized values ​​of each dimension parameter and the corresponding products of the weights of each dimension parameter in the current final calibration data, the current product group is obtained. The similarity between the historical product group and the current product group detected within the historical set time period is determined, and the number of times the historical product group with similarity ≥ preset standard threshold appears is recorded as the frequency of the current product group within the historical set time period.

[0043] In one embodiment, the formula for calculating the target control parameters of the generated LED is:

[0044] ;

[0045] In the formula, For LED target control parameters, The current parameters of the LED. These are the LED parameters for the corresponding light intensity environment under the final scene category. For the transition period, For adjustment intervals.

[0046] Compared with the prior art, the advantages of this application are:

[0047] (1) Based on the light intensity environment, dynamic weight allocation is carried out to break through the limitation of fixed weight in the existing multimodal system. According to different lighting conditions, the weight ratio of each modal data is flexibly adjusted to adapt to the working characteristics of each sensor under different lighting conditions, which improves the rationality and effectiveness of multimodal data fusion. Through normalization processing and error compensation, targeted calibration of multimodal data is achieved to ensure data accuracy. Furthermore, by performing lightweight processing on multimodal data, data processing efficiency can be improved while ensuring data correlation, providing reliable support for real-time control of LEDs. It can not only adapt to different lighting conditions and various usage scenarios, but also achieve a balance between reliability, real-time performance and comfort, making it more widely applicable.

[0048] (2) By combining similarity calculation with confidence determination mechanism based on historical behavior frequency, the accuracy of scene recognition is guaranteed, effectively reducing the probability of scene misjudgment and miscontrol.

[0049] (3) The LED parameters are gradually switched by using a linear smooth control algorithm to avoid visual discomfort caused by sudden control and improve the lighting experience. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of the architecture of an LED intelligent control system for scene recognition in an embodiment of this application;

[0052] Figure 2 This is a flowchart illustrating an LED intelligent control method for scene recognition in an embodiment of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0056] Reference Figure 1 , Figure 1 This is a schematic diagram of the architecture of an LED intelligent control system for scene recognition, as described in an embodiment of this application. Figure 1 As shown in the figure, this application provides an LED intelligent control system for scene recognition, including a multimodal scene data acquisition unit, a scene matching unit, and a decision output unit. The multimodal scene data acquisition unit is connected to the scene matching unit, and the scene matching unit is connected to the decision output unit.

[0057] The multimodal scene data acquisition unit is used to collect multimodal scene data (human feature parameters, environmental parameters, and audio feature parameters) and send the multimodal scene data to the scene matching unit; the scene matching unit is used to determine the suspected scene category based on the multimodal scene data and transmit the suspected scene category determination information to the decision output unit; the decision output unit is used to determine the final scene category based on the suspected scene category determination information, and outputs and executes the LED target control parameters when the LED control command is triggered.

[0058] In one embodiment, the multimodal scene data acquisition unit includes a multimodal scene data acquisition module, which is used to acquire multimodal scene data including human feature parameters, environmental parameters, and audio feature parameters; wherein the human feature parameters include parameters such as position, action, and micro-motion posture, such as position and action parameters acquired by a 1080P high-definition visual sensor, and micro-motion posture parameters acquired by a 24GHz millimeter-wave radar (such as obtaining human micro-motion posture parameters with an amplitude ≥0.1mm by analyzing the amplitude through phase difference).

[0059] Environmental parameters include light intensity, temperature and humidity parameters, such as light intensity parameters collected by a 1080P high-definition visual sensor, and temperature and humidity parameters collected by a temperature and humidity sensor.

[0060] Audio characteristic parameters include parameters such as frequency, decibels, and rhythm, which can be acquired by a full-frequency domain audio sensor.

[0061] Based on the above-mentioned method of acquiring multimodal scene data, the data acquisition source and the multimodal scene data acquisition module are distinguished and referred to as the multimodal scene data acquisition source. The multimodal scene data acquisition source includes the 1080P high-definition visual sensor, 24GHz millimeter-wave radar, temperature and humidity sensor, and full-frequency domain audio sensor mentioned in the above parameter acquisition.

[0062] In one embodiment, the multimodal scene data acquisition unit further includes a weight allocation module and an error compensation module;

[0063] The weight allocation module is used to determine the light intensity environment based on environmental parameters, and to allocate weights to the multimodal scene data acquisition sources based on the light intensity environment. Based on the weight allocation results and the multimodal scene data containing human feature parameters, environmental parameters and audio feature parameters, a fused feature vector is obtained.

[0064] The above process for determining the light intensity environment based on environmental parameters includes:

[0065] Preset light intensity thresholds for low-light, normal, and high-light environments;

[0066] Determine the current light intensity environment based on the light intensity parameters in the environmental parameters.

[0067] This consideration stems from the following: In the bedroom scenario, visual sensors are prone to distortion under strong light. Increasing the weight of micro-motion data collected by millimeter-wave radar can accurately identify whether a person is active, avoiding misjudgments caused by visual bias. When millimeter-wave radar detects a human micro-motion amplitude ≤0.05mm (breathing signal), combined with a light intensity ≤0.3 (weak light), it can assist in determining the sleep scenario, providing a basis for subsequent light-off control. When the audio data is ≤0.2 decibels, the sleep scenario can be verified, thus eliminating misjudgments caused by external interference.

[0068] Based on this, the process of obtaining the fused feature vector includes:

[0069] The weights of parameters for each dimension are pre-established for different light intensity environments (e.g., in the home scene category, the total weight of human feature parameters is 0.5 in strong light environment, with the weight ratio of position, action, and micro-motion posture parameters being 3:2:1, the total weight of environmental parameters being 0.2, and the total weight of audio feature parameters being 0.3; in normal environment, the total weight of human feature parameters is 0.3, the total weight of environmental parameters is 0.4, and the total weight of audio feature parameters is 0.3; in low light environment, the total weight of human feature parameters is 0.2, the total weight of environmental parameters is 0.6, and the total weight of audio feature parameters is 0.2).

[0070] The signal strength of the multimodal scene data acquisition source is normalized to obtain the normalized value of the multimodal scene data acquisition source. That is, the signal strength of the multimodal scene data acquisition source is uniformly mapped to the range of 0-1, where 0 represents no signal and 1 represents the strongest signal.

[0071] After determining the weights of the corresponding multimodal scene data based on the light intensity environment, a weighted fusion is performed using normalized values ​​to obtain the fused feature vector; the weighted fusion calculation formula is as follows:

[0072] ;

[0073] In the formula, To fuse feature vectors, This represents the total number of dimensions of the parameters in the multimodal scene data. Let be the normalized value of the i-th dimension parameter in the multimodal scene data. represents the weight of the normalized value of the i-th dimension parameter in the multimodal scene data.

[0074] Furthermore, the error compensation module is used to perform error calibration on the fused feature vector and obtain calibration data. The methods for obtaining calibration data include:

[0075] ;

[0076] In the formula, B represents the calibration data. This is the error correction value, and .

[0077] In one embodiment, the scene matching unit includes a feature preprocessing module, a parameter optimization module, and a scene matching module; wherein the feature preprocessing module is used to perform noise reduction and normalization processing on the calibration data to obtain preprocessed calibration data.

[0078] This secondary normalization process avoids intermodal signal deviations that occur during feature vector fusion, thus supporting improved scene matching accuracy.

[0079] In one embodiment, the parameter optimization module is used to perform lightweight processing on the preprocessed calibration data to obtain the final calibration data; the lightweight processing flow includes:

[0080] Preset the variance threshold for the corresponding light intensity environment (e.g., 0.06 in low light environment to adapt to millimeter wave signal fluctuations, 0.05 in normal light environment, and 0.04 in strong light environment to suppress visual distortion interference), and remove preprocessed calibration data whose variance (variance of the single-dimensional parameter itself) is lower than the variance threshold. For example, in strong light environment, the variance of temperature and humidity data is 0.03 < 0.04, so the static data is directly removed.

[0081] Preset signal strength thresholds and minimum retention levels for human feature parameters (e.g., retain at least three sets of position parameters, at least three sets of motion parameters, and at least three sets of micro-motion posture parameters), and remove human feature parameters whose signal strength is below the preset human feature parameter thresholds;

[0082] When the remaining human feature parameters are lower than the corresponding minimum retention level, the removed human feature parameters are sorted in descending order of signal strength, and the human feature parameters ranked at the top (e.g., ranked 1st to 3rd) are selected as supplements to the minimum retention level.

[0083] Preset the signal strength threshold and minimum retention level corresponding to the preset environmental parameters and audio feature parameters (e.g., retain at least three sets of light intensity parameters, at least three sets of temperature and humidity parameters, at least three sets of frequency parameters, at least three sets of decibel parameters, and at least three sets of rhythm parameters), and remove environmental parameters and audio feature parameters that are lower than the signal strength threshold corresponding to the preset environmental parameters and audio feature parameters.

[0084] When the remaining environmental parameters and audio feature parameters are lower than the corresponding minimum retention level, the removed environmental parameters and audio feature parameters are sorted in descending order of signal strength, and the environmental parameters and audio feature parameters ranked at the top (e.g., ranked 1st to 3rd) are selected as supplements to the minimum retention level.

[0085] The final calibration data is obtained by weighting and fusing the parameters of each dimension under light intensity conditions and the remaining preprocessed calibration data (see the weighted fusion calculation formula above).

[0086] At this point, the lightweight processing of the data has been completed.

[0087] In one embodiment, the scene matching module is used to match the final calibration data with a preset scene feature library to obtain the suspected scene category corresponding to the final calibration data; wherein the process of establishing the scene feature library includes:

[0088] Determine the application scenario category, such as the living room scenario category or the bedroom scenario category, which are mainly based on people;

[0089] Within each scene category, the light intensity environment is divided. The logic for dividing the light intensity environment includes:

[0090] Low-light environment: light intensity ≤50 lux;

[0091] Typical environment: light intensity 50-500 lux;

[0092] Strong light environment: light intensity ≥ 500 lux;

[0093] For each light intensity environment under each scene category, a quantization threshold group containing human features, environmental features, and audio features is established (the acquisition method is the same as the acquisition process of the final calibration data mentioned above. The data of each component corresponding to the final calibration data obtained in this way is the quantization threshold. The quantization threshold of all dimensional parameters under the light intensity environment is the quantization threshold group. For example, under the bedroom scene category, the quantization threshold of micro-motion posture in a low light environment is 0.05mm). The statistical sample is the effective sample data under each light intensity environment under each scene category, and each group of effective sample data is no less than 100 groups. Through statistical analysis, the mean value of the quantization threshold corresponding to the same type of effective sample data is taken as the initial threshold. After removing abnormal samples that deviate from the mean ±3σ, the corresponding quantization threshold group is obtained.

[0094] After establishing the correlation between scene categories, light intensity environment, and LED parameters, a scene feature library is formed.

[0095] Based on this, the process for obtaining suspected scene categories includes:

[0096] First, extract the quantization threshold group corresponding to the current light intensity environment from the scene feature library as a suspected sample;

[0097] Calculate the similarity between the current final calibration data and the suspected sample. If the similarity is ≥0.8, a successful match is determined, and the suspected scene category of the corresponding suspected sample is output. The similarity calculation formula is:

[0098] ;

[0099] In the formula, Sim represents the similarity score, and j represents the total number of dimensions of the parameters in the final calibration data. This represents the normalized value of the i-th dimension parameter in the final calibration data. The weights of the normalized values ​​of the i-th dimension parameter in the final calibration data. is the quantization threshold for the i-th dimension parameter of the suspected sample.

[0100] In one embodiment, the decision output unit includes a confidence response module and a smoothing control module; wherein the historical feature statistics module is used to determine the normalized value of each dimension parameter and the corresponding product of the weight of each dimension parameter in the current final calibration data, obtain the current product group, determine the similarity (e.g., cosine similarity) between the historical product group (the product group corresponding to the historically collected final calibration data) and the current product group within a historical set time period (e.g., within 7 days), and record the number of times the historical product group appears when the similarity is ≥ a preset standard threshold (e.g., 0.95), as the frequency of the current product group within the historical set time period;

[0101] The confidence response module is used to obtain the confidence level of the suspected scene category based on the frequency of occurrence of the product group corresponding to the suspected scene category of the corresponding suspected sample and the current final calibration data within a historical set time period, and to determine whether to trigger the LED control command based on the confidence level; the confidence level is calculated as follows:

[0102] C = α·M + β·F;

[0103] In the formula, C is the confidence level, M is the similarity between the current final calibration data and the suspected sample, F is the frequency of the current product group within the historical set time period, α and β are both weighting coefficients, and α+β=1.

[0104] Based on this, the process for determining whether to trigger an LED control command according to the confidence level includes:

[0105] When C≥G (G is a preset threshold, taken as 0.9), the suspected scene category with the highest confidence is determined as the final scene category, and the LED control command is triggered;

[0106] When C < G, the suspected scenario category with the highest confidence level is determined as the final scenario category (this category is only for display purposes to facilitate system maintenance by operation and maintenance personnel), and no LED control command is triggered.

[0107] The smoothing control module receives LED control commands and generates target LED control parameters (such as brightness). It then controls the LED according to these target control parameters, which are calculated using the following formula:

[0108] ;

[0109] In the formula, For LED target control parameters, The current parameters of the LED. Here are the LED parameters for the corresponding light intensity environment under the final scene category, where t is the transition time in seconds. , For the adjustment interval, The unit is seconds (s).

[0110] In one embodiment, reference is made to Figure 2 , Figure 2 This is a flowchart illustrating an LED intelligent control method for scene recognition, as described in an embodiment of this application. Figure 2 As shown, an LED intelligent control method for scene recognition is applied to the aforementioned LED intelligent control system for scene recognition. The method includes the following steps:

[0111] S1. Collect multimodal scene data including human feature parameters (position, movement, micro-motion posture parameters), environmental parameters (light intensity, temperature and humidity parameters), and audio feature parameters (frequency, decibels, rhythm parameters);

[0112] S2. Extract light intensity data from the collected environmental parameters to determine the current light intensity environment (based on the above-mentioned light intensity environment classification logic).

[0113] S3. After obtaining the fusion feature vector based on the multimodal scene data (see the above process for obtaining the fusion feature vector), perform error calibration on the fusion feature vector (see the above method for obtaining calibration data) to obtain calibration data.

[0114] S4. After performing noise reduction and normalization on the calibration data, preprocessed calibration data is obtained.

[0115] S5. Perform lightweight processing on the preprocessed calibration data (see the lightweight processing flowchart above) to obtain the final calibration data:

[0116] S6. Preset scene feature library (see the scene feature library establishment process above), determine whether there is a suspected scene category corresponding to the final calibration data (the similarity between the final calibration data and the suspected sample is determined to exist if the similarity is ≥0.8; otherwise, it is determined not to exist). If yes, go to S7; otherwise, end.

[0117] S7. After obtaining the confidence level of the suspected scene category (see the confidence level calculation formula above), determine the suspected scene category with the highest confidence level as the final scene category, and determine whether to trigger the LED control command (see the process of determining whether to trigger the LED control command based on the confidence level above). If yes, proceed to S8; otherwise, end.

[0118] S8. Generate LED target control parameters (see the calculation formula for LED target control parameters above), and control the LED according to the LED target control parameters.

[0119] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0120] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0121] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0122] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0124] The above are merely specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the scope of the claims.

Claims

1. An LED intelligent control system for scene recognition, characterized in that, include: A multimodal scene data acquisition unit is provided, which is connected to a scene matching unit. The multimodal scene data acquisition unit is used to acquire multimodal scene data and send the multimodal scene data to the scene matching unit. A scene matching unit is connected to a decision output unit. The scene matching unit is used to determine the suspected scene category based on multimodal scene data and transmit the suspected scene category determination information to the decision output unit. The decision output unit is used to determine the final scene category based on the suspected scene category judgment information, and outputs and executes the LED target control parameters when the LED control command is triggered. The multimodal scene data acquisition unit includes a multimodal scene data acquisition module, a weight allocation module, and an error compensation module; The multimodal scene data acquisition module is used to acquire multimodal scene data, which includes human feature parameters, environmental parameters, and audio feature parameters. The weight allocation module is used to determine the light intensity environment based on environmental parameters, allocate weights to the multimodal scene data acquisition sources based on the light intensity environment, and obtain the fused feature vector based on the weight allocation result and the multimodal scene data. The error compensation module is used to perform error calibration on the fused feature vector and obtain calibration data. The scene matching unit includes a feature preprocessing module, a parameter optimization module, and a scene matching module; The feature preprocessing module is used to perform noise reduction and normalization on the calibration data to obtain preprocessed calibration data; The parameter optimization module is used to perform lightweight processing on the preprocessed calibration data to obtain the final calibration data. The scene matching module is used to match the final calibration data with a preset scene feature library to obtain the suspected scene category corresponding to the final calibration data. The process for determining whether an LED control command has been triggered includes: Calculate the confidence score for the suspected scene category. The confidence score is calculated using the following formula: ； In the formula, C represents the confidence level, M represents the similarity between the current final calibration data and the suspected samples, the suspected samples are the quantization threshold groups corresponding to the current light intensity environment in the scene feature library, F represents the frequency of the current product group within the historical set time period, α and β are both weighting coefficients, and α + β = 1; the process of obtaining the frequency of the current product group within the historical set time period includes: After determining the normalized values ​​of each dimension parameter and the corresponding products of the weights of each dimension parameter in the current final calibration data, the current product group is obtained. The similarity between the historical product group and the current product group detected within the historical set time period is determined, and the number of times the historical product group with similarity ≥ preset standard threshold appears is recorded as the frequency of the current product group within the historical set time period. Determine if the confidence level exceeds a preset threshold. If so, determine the suspected scene category with the highest confidence level as the final scene category and trigger the LED control command; otherwise, determine the suspected scene category with the highest confidence level as the final scene category and do not trigger the LED control command.

2. The LED intelligent control system for scene recognition according to claim 1, characterized in that, The process of determining the light intensity environment based on environmental parameters includes: Preset light intensity thresholds for low-light, normal, and high-light environments; The light intensity is obtained from environmental parameters, and the current light intensity environment is determined based on the light intensity.

3. The LED intelligent control system for scene recognition according to claim 2, characterized in that, The process of obtaining the fused feature vector includes: Pre-establish the weights of each dimension parameter under the corresponding light intensity environment; After obtaining the signal strength of the multimodal scene data acquisition source, normalization processing is performed to obtain the normalized value of the multimodal scene data acquisition source. After determining the weights of the corresponding multimodal scene data based on the light intensity environment, the fused feature vector is obtained by combining the normalized value.

4. The LED intelligent control system for scene recognition according to claim 3, characterized in that, The lightweighting process includes: Preset a variance threshold for the corresponding light intensity environment, and remove preprocessed calibration data with variance lower than the variance threshold; Preset a signal strength threshold for human feature parameters, and remove human feature parameters that are below the threshold. Preset signal strength thresholds for environmental parameters and audio feature parameters, and discard environmental parameters and audio feature parameters that are below these thresholds; Based on the weights of parameters in each dimension under light intensity conditions and the remaining preprocessed calibration data, the final calibration data is obtained.

5. The LED intelligent control system for scene recognition according to claim 4, characterized in that, The decision output unit includes a confidence response module and a smoothing control module; The confidence response module is used to determine the final scene category based on the suspected scene category judgment information, and to determine whether to trigger the LED control command; The smooth control module is used to receive LED control commands, generate LED target control parameters, and control the LED according to the LED target control parameters.

6. The LED intelligent control system for scene recognition according to claim 5, characterized in that, The formula for calculating the target control parameters of the generated LED is: ； In the formula, For LED target control parameters, The current parameters of the LED. Here are the LED parameters for the corresponding light intensity environment under the final scene category, and t is the transition time. For the adjustment interval, and , .

Images