Smart home lighting system based on feature recognition

By integrating user recognition, situation recognition, lighting adjustment, personalized customization and abnormal alarm modules in the smart home lighting system, combined with sensor data, the problem that traditional systems cannot provide personalized lighting is solved, and the depth of user characteristics and environment is achieved Analysis and dynamic adjustment of lighting parameters improves lighting experience and user satisfaction.

CN118632407BActive Publication Date: 2024-11-15SHANXI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410769487.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-14
Publication Date
2024-11-15
Estimated Expiration
2044-06-14

AI Technical Summary

Technical Problem

Traditional smart home lighting systems cannot automatically identify the user's population types and roles, cannot provide personalized lighting solutions, and cannot deeply analyze user's characteristic information, resulting in the inability to provide users with a truly personalized lighting experience.

Method used

A smart home lighting system based on feature recognition is designed, including user recognition module, situation recognition module, lighting adjustment module, personalized customization module, abnormal alarm module and learning optimization module to obtain user's facial, body shape and behavior characteristics through sensors. Information, combined with environmental data, independently learn and identify usage scenarios, dynamically adjust lighting parameters, provide personalized lighting solutions, and trigger alarms if necessary.

Benefits of technology

It realizes accurate identification of user population types and roles, automatically adjusts lighting parameters, provides personalized lighting solutions for different users, improves the lighting experience of users, meets the needs of different users, and continuously optimizes lighting strategies through learning optimization modules.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118632407B_ABST
    Figure CN118632407B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of smart home technology, and in particular to a smart home lighting system based on feature recognition. In order to solve the problems that the existing smart home lighting system cannot automatically identify the user's crowd type and role, cannot provide customized lighting solutions according to the user's personalized needs, cannot deeply analyze the user's feature information, and cannot provide the user with a truly personalized lighting experience; first, the user's feature information is obtained and recognized to determine the user's role and crowd type as well as the user's expression characteristics and behavior characteristics, and then the user's current usage scenario is identified through a usage scenario model based on a usage scenario mode, and finally, based on the scenario recognition result, user feature information and user preference, the optimal lighting parameters are calculated to provide personalized lighting solutions for different users, meet the lighting needs of different users, and improve the user's lighting experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of smart home technology, and in particular to a smart home lighting system based on feature recognition. Background Art

[0002] With the rapid development of science and technology, people's pursuit of quality of life is getting higher and higher. Smart home can provide people with a comfortable and convenient living environment and improve their quality of life. Smart home integrates home devices and systems using technologies including but not limited to automatic control technology, network communication technology and security technology, so that family members can remotely control home appliances.

[0003] Smart home consists of multiple subsystems, and smart home lighting system is a subsystem in smart home. It can also be used as a separate unit in smart home with other systems. However, traditional smart home lighting systems usually use simple preset lighting schemes, which cannot automatically identify the user's type and role, cannot provide customized lighting schemes according to the user's personalized needs, and cannot deeply analyze the user's characteristic information to provide users with a truly personalized lighting experience.

[0004] Therefore, there is an urgent need to develop a smart home lighting system based on feature recognition to overcome the shortcomings of the existing technology. Summary of the invention

[0005] In order to solve the problem that the existing smart home lighting system cannot be intelligently used by users and cannot provide personalized lighting solutions, the present invention provides such a smart home lighting system based on feature recognition, including a user recognition module, a scene recognition module, a lighting adjustment module, a personalized customization module, an abnormal alarm module and a learning optimization module;

[0006] The user identification module is used to obtain characteristic information of people entering the smart home through sensors installed at necessary locations in the room, determine the role and type of people through the characteristic information, and identify the user's current psychological state and activity intensity through micro-expressions and micro-behavior characteristics contained in the characteristic information;

[0007] The scenario recognition module is used to obtain the environment where the people in the smart home are located through sensors installed at necessary locations in the smart home, and to autonomously learn and recognize different usage scenarios in the smart home environment;

[0008] The lighting adjustment module automatically adjusts lighting parameters according to the identification result provided by the user identification module to provide the best lighting solution for people in the smart home; and adjusts the lighting solution according to the lighting strategies in other modules;

[0009] The personalized customization module is used to provide users with personalized lighting solutions, including adjusting the initial lighting solution and subsequent adjustment strategies;

[0010] The abnormal alarm module immediately triggers an alarm and adopts a backup lighting strategy when the user identification module cannot identify the information of the person entering the smart home or there is a fault in the smart home that affects the lighting; at the same time, based on the characteristic information in the user identification module, it is judged whether the user has a dangerous behavior tendency or situation. If so, an alarm message is issued and the lighting strategy is adjusted;

[0011] The learning optimization module is used to record the user's feedback and modifications to the lighting effect, and optimize the lighting strategy based on the recorded content.

[0012] Furthermore, the specific execution process of the user identification model is as follows:

[0013] S111: Acquire characteristic information of a person through sensors installed at necessary locations of the smart home, the characteristic information including facial features, micro-expressions, behavioral features and identity features;

[0014] S112: Fusing the acquired personnel feature information to generate a high-dimensional feature vector;

[0015] S113: According to the fused high-dimensional feature vector, the user's identity, psychological state, and activity intensity are matched and recognized in a pre-set feature model library;

[0016] S114: Transmit the recognition result to other modules in the smart home lighting system to drive subsequent lighting adjustment and personalized customization strategies.

[0017] Furthermore, the specific steps of the scenario recognition module for autonomously learning and recognizing different usage scenarios are as follows:

[0018] S121: various environmental parameters in the smart home are collected in real time through various sensors installed in the smart home, and the collected environmental data are preprocessed and feature extracted;

[0019] S122: Comprehensively analyzing the pre-processed environment characteristics and user characteristics, and identifying the user's current usage scenario through a pre-deployed scenario recognition model;

[0020] S123: The scenario recognition model outputs the type of usage scenario the user is currently in, and the scenario recognition module transmits the recognition result to the lighting adjustment module;

[0021] S124: When the context recognition module detects a context transfer event, it promptly recognizes the new usage context;

[0022] S125: In subsequent use, the scenario recognition module continues to receive new environmental data, autonomously learns different usage scenarios to adapt to new usage scenarios, and continuously optimizes the accuracy of scenario recognition.

[0023] Furthermore, the specific process of the scenario recognition model to identify different usage scenarios is as follows:

[0024] S1221: fusing different feature vectors into a high-dimensional feature vector z, wherein the different feature vectors include an environment feature vector a and a user identity feature vector b;

[0025] S1222: First, perform hierarchical clustering on the user identity feature vector b to obtain cluster centers μ for different identity categories. k ; Then, for each user identity category k, fine-grained clustering is performed based on the high-dimensional feature vector z to obtain the corresponding

[0026] S1223: Define clustering loss function L:

[0027]

[0028] in, i is the sample size, is the sample set of the jth cluster under the kth identity; K is the number of different categories divided according to user identity information, J is the number of clusters subdivided under the kth identity category, and z i is the high-dimensional feature vector of the i-th sample, is the jth cluster center under the kth identity category;

[0029] S1224: Update the cluster center by iteratively optimizing the cluster loss function L Until convergence, the final clustering model parameters are obtained

[0030] S1225: For the new environment features and user identities (a, b), the distances to each cluster center are calculated using the Euclidean distance:

[0031]

[0032] Classify new environment features and user identities (a, b) to the nearest cluster center That is, identify the current usage scenario; where [a, b] is a high-dimensional feature vector concatenated from the newly input environment feature vector a and the user identity feature vector b. is the jth cluster center under the kth identity category, is the distance between (a,b) and each cluster center.

[0033] Furthermore, the lighting adjustment module establishes a lighting optimization model based on machine learning to dynamically adjust the lighting, wherein the lighting optimization model obtains the recognition result of the scenario recognition model, the user feature information obtained in the user recognition module, and the lighting strategies in the personalized customization module, the abnormal alarm module, and the learning optimization module to calculate the optimal lighting parameters;

[0034] The specific calculation formula of the lighting optimization model is:

[0035] f(x)=W L *σ(W L-1 *…*σ(W1*x+b1)+b L-1 )+b L

[0036] Where f(x) is the predicted optimal lighting parameter, x is the input feature vector of the lighting optimization model, and W L is the weight matrix of the Lth hidden layer, b L is the bias vector of the Lth hidden layer; through multiple layers of nonlinear transformation, the lighting optimization model predicts the optimal lighting parameters.

[0037] Furthermore, when adjusting lighting parameters, the lighting adjustment module identifies multiple users in the same usage scenario in the smart home, assigns different weights to different users based on their identity characteristics, gives priority to the preferences of family members, and provides personalized lighting solutions for users of different age groups;

[0038] When the smart home lighting system detects two or more child users, the lighting decision will be made based on the youngest child.

[0039] When the smart home lighting system identifies two or more elderly or adult users, the lighting decision will be made based on the oldest user.

[0040] Furthermore, the personalized customization module adopts the default initial lighting scheme of the smart home lighting system when it is used for the first time, and gradually adjusts the lighting scheme by recording user feedback during use;

[0041] For people whose identity tags are owners in the feature model library: the next time they use the device, the personalized customization module will formulate a personalized lighting plan based on the previous usage feedback and collected lighting data;

[0042] For persons whose identity tags are guests in the feature model library: when the person has feature data that can match the person's identity in the feature model library, the owner can modify the person's identity tag. After the person's identity tag is modified to guest, the personalized customization module formulates a personalized lighting plan for the guest based on the feature data in the feature model library; if the person has feature data that can match the person's identity in the feature model library, and the person's identity tag has not been modified to guest by the owner, the person is determined to be a stranger and the alarm mechanism is immediately triggered.

[0043] Furthermore, the personalized customization module formulates a personalized lighting plan in advance when the user does not start the lighting according to the user's historical usage data, user feedback information and personal preferences, and the personalized lighting plan is the initial lighting plan when the user starts the lighting;

[0044] The personalized customization module presets personalized lighting schemes for multiple scenarios according to the usage scenarios. When the user starts the lighting in the corresponding scenario, the personalized lighting scheme of the scenario is used as the initial lighting scheme;

[0045] After the lighting is turned on, the lighting adjustment module dynamically adjusts the lighting based on the personalized lighting plan.

[0046] Furthermore, when the user identification module fails to identify the person entering the smart home, the abnormal alarm module classifies the person as a stranger, and the abnormal module immediately triggers an alarm; the abnormal alarm module determines whether the user has dangerous behavior through the user feature information in the user identification module, and when dangerous behavior is detected, the abnormal module immediately triggers an alarm;

[0047] The abnormal alarm module reminds users in the smart home through a preset alarm lighting scheme. If no one is in the smart home, the abnormal alarm module will notify the set emergency contacts of the alarm information through the mobile phone APP.

[0048] Furthermore, the learning optimization module optimizes the lighting decision process. The specific steps are as follows:

[0049] S131: the learning optimization module represents the current state of the smart home lighting system as s, where the state representation includes the current scene recognition result, user feature information, and the current lighting parameter setting; represents the action of the smart home lighting system as a, where the action representation includes the adjustment of the lighting parameters;

[0050] S132: defining a reward function r(s,a) for evaluating the benefit of taking action a in state s; the reward function comprehensively considers user feedback, energy consumption factors, and lighting comfort factors, and is used to evaluate the benefit of taking a certain action in a certain state;

[0051] S133: The learning optimization module defines the Q value function Q(s,a), the Q value function Q C (s,a) represents the long-term expected reward of taking action a in state s;

[0052] The calculation formula of the Q value function is:

[0053] Q C (s,a)←Q o1 (s,a)+α[r(s,a)+γmaxQ n (s′,a′)-Q o2 (s,a)]

[0054] Among them, Q C (s,a) represents the long-term expected reward of taking action a in state s, Q o1 (s,a) represents the original Q value of taking action a in the current state s, Q o2 (s,a) represents the new Q(s,a) obtained after updating, Q n (s′, a′) represents the long-term expected reward of taking action a′ in the next state s′, α is used to control the update speed of the Q-value function, r(s, a) is the immediate reward, γ represents the importance of future rewards, s′ represents the next state, and a′ represents the next optional action in state s′;

[0055] S134: The learning optimization module generates an optimized lighting strategy using the Q-value function, and sends the lighting strategy to the lighting adjustment module, which adjusts the lighting parameters according to the lighting strategy and feeds back the adjusted state s′ and the immediate reward r to the learning optimization module;

[0056] S135: The learning optimization module continues to generate an optimized lighting strategy based on the state s′ and the immediate reward r obtained through feedback, and repeats steps S3 and S4 to continuously optimize the lighting strategy of the lighting system until the lighting is turned off.

[0057] Beneficial effects:

[0058] 1. The present invention can accurately identify the user's crowd type and role by comprehensively analyzing the user's facial features, body features and behavioral characteristics; at the same time, based on the identification results of the crowd type and role, the smart home lighting system automatically adjusts the lighting parameters to provide personalized lighting solutions for different users, meet the lighting needs of different users, and improve the user's lighting experience.

[0059] 2. The present invention analyzes the home environment, establishes a model based on usage scenarios, autonomously learns and identifies different usage scenarios, and the smart home lighting system dynamically adjusts lighting parameters according to the identified usage scenarios to provide users with the most appropriate lighting effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a system architecture diagram of the present invention,

[0061] Figure 2 This is a specific flow chart of the present invention for identifying a user.

[0062] Figure 3 This is a specific flow chart of the present invention for identifying the user's situation. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0064] In order to solve the problems of being unable to automatically identify the user's crowd type and role, being unable to provide customized lighting solutions according to the user's personalized needs, being unable to deeply analyze the user's characteristic information, and being unable to provide users with a truly personalized lighting experience, such as Figure 1 As shown, the present invention provides a smart home lighting system based on feature recognition, including a user recognition module, a scenario recognition module, a lighting adjustment module, a personalized customization module, an abnormal alarm module and a learning optimization module. These modules cooperate with each other to form a complete smart home lighting system.

[0065] like Figure 2-Figure 3 As shown, in this embodiment, the user identification module is used to obtain characteristic information of people entering the smart home through sensors installed at necessary locations in the room, determine the role and crowd type of the people through the characteristic information, and identify the user's current psychological state and activity intensity through micro-expressions and micro-behavior characteristics contained in the characteristic information.

[0066] Specifically, the acquired personal characteristic information includes facial characteristics, body characteristics and behavioral characteristics.

[0067] Facial features are extracted by extracting facial images, extracting facial key points (corners of the eyes and mouth, etc.) and facial contour features, and extracting micro-expression features of the person. Micro-expression features include but are not limited to emotional states such as pleasure, tension, concentration, and frustration, so as to more accurately judge the user's current psychological state and adjust the lighting parameters accordingly.

[0068] Body posture characteristics include the user's standing posture, sitting posture, walking posture, etc. By analyzing the user's body posture characteristics, it is possible to determine whether the user's current activity status is resting, working, or entertaining, etc.

[0069] Behavioral characteristics include the detection of the intensity of user activities, such as relaxing play, strenuous exercise, and sitting and resting. By identifying the intensity of user activities, the lighting solution can be further optimized to meet the lighting needs of users in different activity states.

[0070] Specifically, the specific execution process of the user identification module includes:

[0071] S111: Through high-definition cameras and sensors installed in key locations in the smart home, multi-dimensional feature data such as facial features, micro-expressions, behavioral features, body features, and identity features of people entering the smart home are collected in real time.

[0072] S112: The user identification module analyzes the extracted feature information, finds the internal connection and mutual influence between different feature information, and quantifies the correlation and association between different feature information, and then fuses the acquired personnel feature information to generate a high-dimensional feature vector.

[0073] S113: According to the fused high-dimensional feature vector, the user's identity, psychological state and activity intensity are matched and recognized in a pre-set feature model library, wherein the pre-set feature model library stores different feature information of people.

[0074] S114: The user identification module then transmits the identification result to other modules of the smart home lighting system to drive subsequent lighting adjustment and personalized customization strategies.

[0075] In this embodiment, the scenario recognition module is used to obtain the environment of the location of people in the smart home through sensors installed in key locations in the smart home, and to autonomously learn and recognize different usage scenarios in the smart home environment.

[0076] Specifically, the specific steps for the scenario recognition module to autonomously learn and recognize different usage scenarios are as follows:

[0077] S121: The scenario recognition module collects various environmental data of the smart home in real time through various sensors installed in key positions in the smart home, including environmental factors such as light brightness, temperature and humidity, and noise, and preprocesses and extracts features from the collected environmental data.

[0078] S122: Comprehensively analyze the pre-processed environmental features and user features, and identify the usage scenario of the current room through the pre-deployed scenario recognition model. The specific execution process of the scenario recognition model identifying the usage scenario of the current room is as follows:

[0079] S1221: Fusing different feature vectors to form a high-dimensional feature vector z, where the different feature vectors include an environment feature vector a and a user identity feature vector b, where a = [a1, a2, …, a n ],b=[b1,b2,…,b n ], and a i is the i-th environmental feature, b j is the jth user identity feature;

[0080] S1222: First, perform hierarchical clustering on the user identity feature vector b to obtain cluster centers μ for different identity categories. k ; Then, for each user identity category k, fine-grained clustering is performed based on the high-dimensional feature vector z to obtain the corresponding

[0081] S1223: Define clustering loss function L:

[0082]

[0083] in, i is the sample size, is the sample set of the jth cluster under the kth identity; K is the number of different categories divided according to user identity information, J is the number of clusters subdivided under the kth identity category, and z i is the high-dimensional feature vector of the i-th sample, is the jth cluster center under the kth identity category;

[0084] S1224: Update the cluster center by iteratively optimizing the cluster loss function L Until convergence, the final clustering model parameters are obtained

[0085] S1225: For new environmental features and user identity features (a, b), the distance to each cluster center is calculated using the Euclidean distance:

[0086]

[0087] Classify the new environment features and user identity features (a, b) to the nearest cluster center That is, identify the current usage scenario; where [a, b] is a high-dimensional feature vector concatenated from the newly input environment feature vector a and the user identity feature vector b. is the jth cluster center under the kth identity category, is the distance between (a,b) and each cluster center.

[0088] S123: The scenario recognition model outputs the currently determined usage scenario type of the room, and the scenario recognition module transmits the recognition result to the lighting adjustment module.

[0089] S124: If the scenario recognition module detects a scenario transfer event, the scenario recognition module can identify the usage scenario in a timely manner, where the scenario transfer event is specifically an event in which the user switches from reading to rest or walks from indoors to outdoors, etc. The scenario recognition module identifies the usage scenario in a timely manner to ensure that the transition of the lighting scheme is smooth, avoiding the user feeling abrupt or uncomfortable under different lighting conditions.

[0090] S125: In subsequent use, the scenario recognition module continues to receive new environmental data, autonomously learns different usage scenarios to adapt to new usage situations, and continuously optimizes the accuracy of scenario recognition.

[0091] In this embodiment, the lighting adjustment module automatically adjusts the lighting parameters according to the data provided by the user identification module to provide the best lighting solution for the users in the smart home; at the same time, the lighting solution is adjusted according to the lighting strategies in other modules.

[0092] Specifically, the lighting adjustment module establishes a lighting optimization model based on machine learning to adjust the lighting, wherein the lighting optimization model obtains the scenario recognition results of the scenario recognition module and the user feature information obtained in the user recognition module, as well as the lighting strategies provided in the personalized optimization module, the abnormal alarm module and the learning optimization module, to calculate the optimal lighting parameters, determine the optimal brightness value, color temperature value and lighting area range, and control the lighting equipment to illuminate within the specified lighting area according to the determined brightness value and color temperature value.

[0093] The analysis results of other modules as input features respectively describe the current environmental conditions, user needs and user feedback information. Taking these data as input features of the lighting optimization model can predict the optimal existing lighting parameters.

[0094] The specific calculation formula of the lighting optimization model is:

[0095] f(x)=W L *σ(W L-1 *…*σ(W1*x+b1)+b L-1 )+b L

[0096] Wherein, f(x) is the predicted optimal lighting parameter, that is, the output result of this lighting optimization model is one or more optimal lighting parameters, including but not limited to light intensity and color temperature; x is the input feature vector of the lighting optimization model, including vectors such as scene recognition results, user feature information and preferences; W L is the weight matrix of the Lth hidden layer, bL is the bias vector of the Lth hidden layer, and σ is the activation function, which is used to enable the lighting optimization model to learn the nonlinear relationship between input features and outputs. Through multiple layers of nonlinear transformation, the lighting optimization model predicts the optimal lighting parameters.

[0097] Specifically, the lighting adjustment module automatically adjusts the brightness and color temperature of the light according to the identified micro-expressions of fatigue to reduce visual fatigue and enhance the user's comfort and relaxation; according to the identified micro-expressions of excitement, it automatically increases the lighting brightness and improves the vitality of the ambient light to match the user's happy mood and enhance the effect of the activity; according to the identified sitting and resting behavior, it adjusts the brightness and color temperature of the light to create a comfortable resting atmosphere; according to the identified strenuous exercise behavior, it provides brighter and wider coverage lighting to ensure the safety of exercise. From the above description, it can be seen that the lighting adjustment module adopts different lighting parameters according to the identified micro-expressions and behavioral characteristics. The lighting adjustment module will modify the lighting parameters from subtle changes in micro-expressions to the diversity of behavioral characteristics, so as to more accurately obtain the user's status and needs and provide users with better lighting services.

[0098] In this embodiment, the personalized customization module is used to provide users with personalized customized lighting solutions, including adjusting the initial lighting solution and subsequent lighting strategy adjustments.

[0099] Specifically, the personalized customization module formulates a personalized lighting plan in advance based on the user's historical data, user feedback information and personal preferences when the user does not start the lighting, and this personalized lighting plan will be used as the initial lighting plan when the user starts the lighting.

[0100] The personalized customization module presets multiple personalized lighting schemes for different usage scenarios according to the usage scenarios that the user may be in. When the user starts the lighting in the corresponding usage scenario, the personalized lighting scheme corresponding to the usage scenario is used as the initial lighting scheme.

[0101] When the user turns on the lighting, the personalized lighting solution adopted by the personalized customization module will be used by the lighting adjustment module as the basis for adjusting the lighting parameters to dynamically adjust the lighting.

[0102] Specifically, the personalized customization module sets specific execution strategies according to the different types and ages of people, including:

[0103] 1. For first-time users, the personalized customization module will adopt a general initial lighting solution. During subsequent use, based on the user's feature information continuously collected by the smart home lighting system, including facial, body shape, and behavioral feature information, the lighting solution will be gradually personalized to meet the user's personalized needs.

[0104] 2. For people whose types are identified as "owners", that is, the user identification module matches the identified user features in the feature model library, and the matched person identity label is the owner. The personalized customization module will customize a unique personalized lighting plan for them based on the last usage feedback and collected personal data. This personalized plan is continuously optimized as user preferences change.

[0105] 3. For people whose crowd type is identified as "guests", that is, after the user identification module matches the identified user features in the feature model library, the matched person identity tag is a guest. The personalized customization module will customize a temporary personalized lighting plan for the guest after the host confirms the guest's identity. If the identity of the person is not confirmed by the host or the confirmation record is cancelled by the host, and the feature data of the person in the feature model library has not been deleted by the host, the person will be judged as a stranger and the alarm mechanism will be triggered immediately. When the user identification module does not recognize the identity feature data of the person in the matching feature model library, the user identification module will establish a corresponding person table in the feature model library to store the identity features of the person, and within a fixed period of time, such as 5 minutes, the person whose identity tag in the feature model library is the host can modify the identity tag in the person table. If the identity tag in the person table is not modified within this time period, the person will be judged as a stranger.

[0106] 4. The personalized module can also adjust the lighting scheme according to the age characteristics of users, such as the elderly, adults and children, to meet the needs of different user groups. When multiple users are identified in the same usage scenario of the smart home, the owner's preference will be given priority, and the weight will be adjusted according to the age characteristics of family members to calculate the best comprehensive lighting scheme; when the smart home lighting system detects that there are two or more child users, the lighting decision will be made based on the youngest child; when the smart home lighting system identifies two or more elderly or adult users, the lighting decision will be made based on the oldest user; when the smart home lighting system identifies that there are two or more children and the elderly together, the best lighting scheme will be calculated based on the respective weight coefficients of the personnel. The specific calculation formula is:

[0107]

[0108] Among them, Best is the best lighting solution calculated, argmax is to select the solution that maximizes the value of the entire expression, and w oi is the weight coefficient of elderly person i, w cj is the weight coefficient of child j, i represents the number of the elderly, j represents the number of the child, n represents the total number of elderly people, and m represents the total number of children.

[0109] In this embodiment, the abnormal alarm module is used to immediately trigger an alarm and adopt a preset alarm strategy to remind the user when the user identification module cannot identify the information of the person entering the smart home or there is a fault in the smart home that affects the lighting; at the same time, based on the characteristic information in the user identification module, it is judged whether the user has a dangerous behavior tendency or situation. If so, an alarm is triggered and a preset alarm strategy is adopted to remind the user.

[0110] Specifically, the abnormal alarm module monitors the working status of each sensor in real time. Once an abnormal situation is found, it analyzes and diagnoses the abnormal situation, gives corresponding troubleshooting suggestions, and promptly notifies the user of the abnormal situation through mobile phone APP or other communication methods, and provides fault diagnosis and emergency measures. When the abnormal alarm module detects an abnormal situation that affects personalized lighting, it adopts strategies such as emergency lighting or minimum lighting to ensure that indoor lighting is not completely interrupted.

[0111] When the abnormal alarm module detects possible dangerous behaviors, such as the user being stationary for a long time or falling, it will immediately conduct further analysis and judgment, including:

[0112] 1. If the user is detected to be stationary for a long time and falls to the ground or stops breathing, the user is judged to be in danger, the alarm mechanism is activated immediately, and the environmental safety is enhanced by increasing the brightness or flashing the warning light;

[0113] 2. If the user is monitored to be stationary for a long time, but the vital signs such as heart rate or breathing are monitored and found to be normal, it is judged that the user is in a state of sleep or rest. It is not judged as a dangerous behavior, but the lighting brightness is appropriately reduced and the alarm mechanism is not triggered.

[0114] Specifically, when the user identification module fails to identify the person entering the smart home, the abnormal alarm module classifies the person as a stranger and the abnormal module immediately triggers an alarm; the abnormal alarm module determines whether the user has dangerous behavior through the user feature information in the user identification module. When dangerous behavior is detected, the abnormal alarm module immediately triggers an alarm and starts high-brightness lighting to increase visibility; the abnormal alarm module reminds users in the smart home through a preset alarm lighting scheme. If no one is in the smart home, the abnormal alarm module notifies the set emergency contact of the alarm information through the mobile phone APP.

[0115] In this embodiment, the learning optimization module is used to record the user's feedback and modifications on the lighting effect, and use this as a basis to optimize the lighting strategy; and interact with other subsystems in the smart home to obtain the historical usage records of users in other subsystems to optimize the lighting solutions provided to users.

[0116] Specifically, the learning optimization module optimizes the lighting decision process. The specific steps are:

[0117] S131: The learning optimization module represents the current state of the smart home lighting system as s, which includes the current scene recognition result, user feature information and the current lighting parameter settings; and represents the action of the smart home lighting system as a, which includes the adjustment of the lighting parameters.

[0118] S132: Define a reward function r(s,a) for evaluating the benefit of taking action a in state s; the reward function comprehensively considers user feedback, energy consumption factors, and lighting comfort factors, and is used to evaluate the benefit of taking a certain action in a certain state.

[0119] S133: The learning optimization module defines a Q value function Q(s, a), wherein the Q value function Q C (s,a) represents the long-term expected reward of taking action a in state s.

[0120] The calculation formula of the Q value function is:

[0121] Q C (s,a)←Q o1 (s,a)+α[r(s,a)+γmaxQ n (s′,a′)-Q o2 (s,a)]

[0122] Among them, Q C (s,a) represents the long-term expected return Q obtained by taking action a in state s o1 (s,a) represents the original Q value of taking action a in the current state s; Q o2 (s,a) represents the new Q(s,a) obtained after updating; Q n (s′,a′) represents the long-term expected reward of taking action a′ in the next state s′; α is the learning rate, which is used to control the update speed of the Q-value function; r(s,a) is the immediate reward; γ is the discount factor, which indicates the importance of future rewards; s′ represents the next state; a′ represents the next optional action in state s′; ← is the assignment operation; By continuously updating the Q-value function, the learning optimization module can learn the optimal lighting strategy, thereby improving the user's lighting comfort and energy saving effect.

[0123] S134: The learning optimization module generates an optimized lighting strategy using the Q value function, and sends the lighting strategy to the lighting adjustment module, which adjusts the lighting parameters according to the lighting strategy and sends the adjusted state s ′ And the immediate reward r is fed back to the learning optimization module.

[0124] S135: The learning optimization module obtains the state s based on the feedback ′ and the instant reward r, continue to generate the optimized lighting strategy, repeat steps S133 and S134, and continuously optimize the lighting strategy of the lighting system until the lighting is turned off.

[0125] At the same time, the learning optimization module is also equipped with an identity library to store personal identity information, including user information of different types of people such as hosts and guests. It also stores the user's lighting preferences in different usage scenarios to establish user portraits. The learning optimization module combines user portraits to provide personalized lighting solution recommendations for different users.

[0126] The above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications, improvements and substitutions can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the attached claims.

Claims

1. A smart home lighting system based on feature recognition, characterized in that: It includes user identification module, scenario identification module, lighting adjustment module, personalized customization module, abnormal alarm module and learning optimization module; The user identification module is used to obtain characteristic information of people entering the smart home through sensors installed at necessary locations in the room, determine the role and type of people through the characteristic information, and identify the user's current psychological state and activity intensity through micro-expressions and micro-behavior characteristics contained in the characteristic information; The scenario recognition module is used to obtain the environment of the location of the person in the smart home through the sensors installed at the necessary locations in the smart home, and autonomously learn and recognize different usage scenarios in the smart home environment in combination with the recognition results of the user recognition module; The lighting adjustment module automatically adjusts lighting parameters according to the recognition result provided by the user recognition module to provide the best lighting solution for people in the smart home; and adjusts the lighting solution according to the lighting strategies in the user recognition module, the scenario recognition module, the personalized customization module, the abnormal alarm module and the learning optimization module; The personalized customization module is used to provide users with personalized lighting solutions, including adjusting the initial lighting solution and subsequent adjustment strategies; The abnormal alarm module immediately triggers an alarm and adopts a backup lighting strategy when the user identification module cannot identify the information of the person entering the smart home or there is a fault in the smart home that affects the lighting; at the same time, based on the characteristic information in the user identification module, it is judged whether the user has a dangerous behavior tendency or situation. If so, an alarm message is issued and the lighting strategy is adjusted; When the abnormal alarm module detects possible dangerous behaviors, such as the user being stationary for a long time or falling, it will immediately conduct further analysis and judgment, including: 1: If the user is detected to be stationary for a long time and falls to the ground or stops breathing, the system will determine that the user is in danger and immediately activate the alarm mechanism, and enhance environmental safety by increasing the brightness or flashing the warning light; 2: If the user is monitored to be stationary for a long time, but the vital signs such as heart rate or breathing of the user are monitored and found to be normal, it is judged that the user is in a state of sleep or rest, and it is not judged as a dangerous behavior, but the lighting brightness is appropriately reduced, and the alarm mechanism is not triggered; the learning optimization module is used to record the user's feedback and modification on the lighting effect, and optimize the lighting strategy based on the recorded content; The specific execution process of the user identification module is as follows: S111: Acquire characteristic information of a person through sensors installed at necessary locations of the smart home, the characteristic information including facial features, micro-expressions, behavioral features, and identity features; S112: Fusing the acquired personnel feature information to generate a high-dimensional feature vector; S113: According to the fused high-dimensional feature vector, the user's identity, psychological state, and activity intensity are matched and recognized in a pre-set feature model library, wherein the pre-set feature model library stores different feature information of people; S114: Transmitting the recognition result to the scenario recognition module, the lighting adjustment module, the personalized customization module, the abnormal alarm module and the learning optimization module in the smart home lighting system to drive subsequent lighting adjustment and personalized customization strategies; The specific steps for the scenario recognition module to autonomously learn and recognize different usage scenarios are as follows: S121: various environmental parameters in the smart home are collected in real time through various sensors installed in the smart home, and the collected environmental data are preprocessed and feature extracted; S122: Comprehensively analyzing the pre-processed environment characteristics and user characteristics, and identifying the user's current usage scenario through a pre-deployed scenario recognition model; S123: The scenario recognition model outputs the type of usage scenario the user is currently in, and the scenario recognition module transmits the recognition result to the lighting adjustment module; S124: When the context recognition module detects a context transfer event, it promptly recognizes the new usage context; S125: In subsequent use, the scenario recognition module continuously receives new environmental data, autonomously learns different usage scenarios to adapt to new usage scenarios, and continuously optimizes the accuracy of scenario recognition; The learning optimization module optimizes the lighting decision process. The specific steps are: S131: the learning optimization module represents the current state of the smart home lighting system as s, where the state representation includes the current scene recognition result, user feature information, and the current lighting parameter setting; represents the action of the smart home lighting system as a, where the action representation includes the adjustment of the lighting parameters; S132: defining a reward function r(s,a) for evaluating the benefit of taking action a in state s; the reward function comprehensively considers user feedback, energy consumption factors, and lighting comfort factors, and is used to evaluate the benefit of taking a certain action in a certain state; S133: The learning optimization module defines the Q value function Q(s,a), the Q value function Q C (s,a) represents the long-term expected reward of taking action a in state s; The calculation formula of the Q value function is: Q C (s,a)←Q o1 (s,a)+α[r(s,a)+γmaxQ n (s′,a′)-Q o2 (s,a)] Among them, Q C (s,a) represents the long-term expected reward of taking action a in state s, Q o1 (s,a) represents the original Q value of taking action a in the current state s, Q o2 (s,a) represents the new Q(s,a) obtained after updating, Q n (s′, a′) represents the long-term expected reward of taking action a′ in the next state s′, α is used to control the update speed of the Q-value function, r(s, a) is the immediate reward, γ represents the importance of future rewards, s′ represents the next state, and a′ represents the next optional action in state s′; S134: The learning optimization module generates an optimized lighting strategy using the Q value function, and sends the lighting strategy to the lighting adjustment module, which adjusts the lighting parameters according to the lighting strategy and sends the adjusted state s ′ And the immediate reward r is fed back to the learning optimization module; S135: The learning optimization module obtains the state s based on the feedback ′ and the instant reward r, continue to generate the optimized lighting strategy, repeat steps S133 and S134, and continuously optimize the lighting strategy of the lighting system until the lighting is turned off.

2. The feature recognition-based smart home lighting system according to claim 1, characterized in that: The specific process of the scenario recognition model to identify different usage scenarios is as follows: S1221: fusing different feature vectors into a high-dimensional feature vector z, wherein the different feature vectors include an environment feature vector a and a user identity feature vector b; S1222: First, perform hierarchical clustering on the user identity feature vector b to obtain cluster centers μ for different identity categories. k ; Then, for each user identity category k, fine-grained clustering is performed based on the high-dimensional feature vector z to obtain the corresponding S1223: Define clustering loss function L: in, i is the sample size, is the sample set of the jth cluster under the kth identity; K is the number of different categories divided according to user identity information, J is the number of clusters subdivided under the kth identity category, and z i is the high-dimensional feature vector of the i-th sample, is the jth cluster center under the kth identity category; S1224: Update the cluster center by iteratively optimizing the cluster loss function L Until convergence, the final clustering model parameters are obtained S1225: For new environmental features and user identity features (a, b), the distance to each cluster center is calculated using the Euclidean distance: Classify new environment features and user identities (a, b) to the nearest cluster center That is, identify the current usage scenario; where [a, b] is a high-dimensional feature vector concatenated from the newly input environment feature vector a and the user identity feature vector b. is the jth cluster center under the kth identity category, is the distance between (a,b) and each cluster center.

3. The feature recognition-based smart home lighting system according to claim 1, characterized in that: The lighting adjustment module establishes a lighting optimization model based on machine learning to dynamically adjust the lighting. The lighting optimization model obtains the recognition result of the scenario recognition model, the user feature information obtained in the user recognition module, and the lighting strategies in the personalized customization module, the abnormal alarm module and the learning optimization module to calculate the optimal lighting parameters; The specific calculation formula of the lighting optimization model is: f(x)=W L *σ(W L-1 *…*σ(W1*x+b1)+b L-1 )+b L Where f(x) is the predicted optimal lighting parameter, x is the input feature vector of the lighting optimization model, and W L is the weight matrix of the Lth hidden layer, b L is the bias vector of the Lth hidden layer; through multiple layers of nonlinear transformation, the lighting optimization model predicts the optimal lighting parameters.

4. The feature recognition-based smart home lighting system according to claim 3, characterized in that: When adjusting lighting parameters, the lighting adjustment module identifies multiple users in the same usage scenario in the smart home and assigns different weights to different users based on their identity characteristics, giving priority to the preferences of family members and providing personalized lighting solutions for users of different age groups; When the smart home lighting system detects two or more child users, the lighting decision will be made based on the youngest child. When the smart home lighting system identifies two or more elderly or adult users, the lighting decision will be made based on the oldest user.

5. The feature recognition-based smart home lighting system according to claim 1, characterized in that: The personalized customization module adopts the default initial lighting scheme of the smart home lighting system when it is used for the first time, and gradually adjusts the lighting scheme by recording user feedback during use; For people whose identity tags are owners in the feature model library: the next time they use the device, the personalized customization module will formulate a personalized lighting plan based on the previous usage feedback and collected lighting data; For persons whose identity tags are guests in the feature model library: when the person has feature data that can match the person's identity in the feature model library, the owner can modify the person's identity tag. After the person's identity tag is modified to guest, the personalized customization module formulates a personalized lighting plan for the guest based on the feature data in the feature model library; if the person has feature data that can match the person's identity in the feature model library, and the person's identity tag has not been modified to guest by the owner, the person is determined to be a stranger and the alarm mechanism is immediately triggered.

6. The feature recognition-based smart home lighting system according to claim 5, characterized in that: The personalized customization module formulates a personalized lighting plan in advance when the user does not start the lighting according to the user's historical usage data, user feedback information and personal preferences. The personalized lighting plan is the initial lighting plan when the user starts the lighting. The personalized customization module presets personalized lighting schemes for multiple scenarios according to the usage scenarios. When the user starts the lighting in the corresponding scenario, the personalized lighting scheme of the scenario is used as the initial lighting scheme; After the lighting is turned on, the lighting adjustment module dynamically adjusts the lighting based on the personalized lighting plan.

7. The feature recognition-based smart home lighting system according to claim 1, characterized in that: When the user identification module fails to identify the person entering the smart home, the abnormal alarm module classifies the person as a stranger and the abnormal module immediately triggers an alarm; the abnormal alarm module determines whether the user has dangerous behavior through the user feature information in the user identification module. When dangerous behavior is detected, the abnormal module immediately triggers an alarm; the abnormal alarm module reminds the user in the smart home through the preset alarm lighting scheme. If no one is in the smart home, the abnormal alarm module notifies the set emergency contact of the alarm information through the mobile phone APP.