AI-based elderly-friendly indoor environment personalized map construction method

Abstract

The invention relates to an AI-based old-age-friendly indoor environment personalized map construction method. The method comprises the following steps: acquiring indoor temperature and humidity, air quality and illumination intensity parameters through an environment sensor deployed in an indoor key area; a camera with a privacy protection function and an infrared sensor combined device are used for monitoring indoor movement tracks and data of sleeping, getting up and walking behaviors of old people. Through deep analysis of the individual data of the old people, the system can accurately meet the unique demands of each old people for the indoor environment, improves the living comfort and life quality of the old people, achieves the automatic and intelligent regulation and control of the indoor environment, reduces the inconvenience of manual operation of equipment by the old people, improves the accuracy and timeliness of environment regulation and control, and improves the user experience. The indoor environment is associated with the health condition of the old people, the influence of environmental factors on health can be found in time, early warning and intervention are carried out, diseases can be prevented, and the health and safety of the old people can be guaranteed.

Description

Technical Field

[0001] This invention relates to the fields of interior environment design and artificial intelligence technology, specifically to a method for constructing personalized maps of age-friendly indoor environments based on AI. Background Technology

[0002] With the accelerating pace of global aging, the quality of living environments for the elderly is receiving increasing attention. Due to the decline in physical functions, the elderly population has unique needs for their indoor environment, such as suitable temperature and humidity, good lighting, and a safe spatial layout.

[0003] Traditional indoor environment design often adopts general standards and specifications, making it difficult to meet the individualized needs of each elderly person. At the same time, existing indoor environmental monitoring and control systems lack in-depth analysis and intelligent response to factors such as the elderly's behavioral habits and health conditions. The rapid development of artificial intelligence technology offers a new approach to solving these problems.

[0004] Therefore, an AI-based method for constructing personalized maps of elderly-friendly indoor environments is proposed to address the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-based method for constructing personalized maps of age-friendly indoor environments. By using AI technology to analyze a large amount of environmental data and individual data of the elderly, a personalized indoor environment map is constructed, enabling precise design and intelligent control of age-friendly indoor environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: 1. A method for constructing a personalized map of an age-friendly indoor environment based on AI, characterized by comprising the following steps: (1) Data collection: Deploy environmental sensors in key indoor areas to collect indoor temperature, humidity, air quality and light intensity parameters; Use a combination device consisting of a camera with privacy protection function and an infrared sensor to monitor the activity trajectory of the elderly in the room, as well as sleep, getting up and walking behavior data; Use wearable devices or smart home terminals to collect physiological health data such as heart rate, blood pressure and sleep quality of the elderly in real time. (2) AI data analysis: The multi-dimensional data collected in step (1) are preprocessed by outlier removal, noise filtering and normalization. The principal component analysis (PCA) algorithm or wavelet transform algorithm is used to extract environmental parameter features, behavioral time distribution features and health data fluctuation features from the preprocessed multi-dimensional data. A deep learning model based on long short-term memory network (LSTM) is constructed. The dataset after feature extraction is divided into training set and test set in a 7:3 ratio. The deep learning model is iteratively trained by Adam optimizer to obtain a correlation model that can associate environmental parameters, behavioral patterns and health status. (3) Personalized map construction: Based on the output results of the association model trained in step (2), combined with the feedback information obtained from the subjective needs interviews with the elderly and their families, the personalized indoor environment demand thresholds for different regions and time periods are determined; 3D modeling software is used to load the basic model of indoor space layout, and the appropriate environmental parameter range, behavioral hotspot area labeling information and health risk environment warning thresholds for each region are superimposed on the basic model of indoor space layout in the form of a visual layer to generate an elderly-friendly indoor environment personalized map containing spatial coordinates, parameter thresholds and warning signs; (4) Intelligent control and feedback: Establish a communication connection between the personalized map generated in step (3) and the smart home control system, and automatically send control instructions to the air conditioner, lights, humidifier and air purifier according to the parameters in the personalized map; continuously collect the elderly’s adaptation feedback data to the indoor environment and the elderly’s health index change data, and input the collected feedback data and health index change data into the correlation model every 24 hours, iteratively optimize the parameters of the correlation model, and dynamically update the personalized map and the corresponding environmental control strategy according to the optimized model parameters.

[0007] Furthermore, the environmental sensor mentioned in step (1) uses Bluetooth Low Energy (BLE) or LoRa wireless communication protocol and is deployed in the living room, bedroom, bathroom, and kitchen areas. The sampling frequency is set to 5-15 minutes / time, and the sensor sampling accuracy meets the following requirements: temperature ±0.5℃, humidity ±3%RH, light intensity ±10lx, PM2.5 ±5μg / m³. 3 . Furthermore, the camera described in step (1) has human contour recognition and facial blurring functions. The infrared sensor adopts passive infrared detection (PIR) technology. After the data of the two are fused, behavior recognition is performed through the edge computing module. The recognition delay is no more than 100ms, and the original image data is not stored locally or in the cloud. Furthermore, the data preprocessing in step (2) also includes data time-series alignment processing, which synchronizes environmental data, behavioral data, and health data to millisecond-level timestamps; during feature extraction, attention mechanism is used to enhance the weight ratio of key features such as "frequency of getting up" and "change in walking speed" in behavioral data. Furthermore, the LSTM model described in step (2) includes an input layer, three hidden layers and an output layer. The number of neurons in the hidden layers is set to 128, 64 and 32 respectively. The Dropout technique is used to suppress overfitting. The Dropout probability is set to 0.2. The number of training iterations of the model is no less than 50 rounds. The loss function is the mean squared error (MSE). Furthermore, the personalized environmental requirement thresholds mentioned in step (3) include: bedroom temperature of 22-26℃, humidity of 40-60%, and light intensity of ≤50lx during sleep; living room temperature of 24-28℃, humidity of 35-55%, and light intensity of 150-300lx during activity; and the map supports touch operation and voice interaction query functions, and the warning signs use red, yellow, and green to distinguish high, medium, and low health risk levels. Furthermore, the smart home control described in step (1) adopts the MQTT communication protocol. When the elderly's sleep quality score decreases by ≥20% and the indoor humidity is below 40%, the system automatically increases the humidifier's working power by 30-50% and records the changes in sleep data within 2 hours after the control. During the model iteration and optimization process, incremental learning technology is used to update the model parameters without retraining all the data. Furthermore, privacy protection measures are included: the collected behavioral and health data are encrypted and stored on the blockchain, and can only be accessed by authorized users after biometric verification (fingerprint / face); environmental sensors and wearable devices are equipped with local data de-identification processing functions, and data transmission adopts end-to-end encryption algorithms. Furthermore, the 3D modeling software mentioned in step (3) is SketchUp or Blender. After the map is generated, it can be exported as PDF or PNG format and can be synchronized to the mobile terminal APP of the elderly's children to realize remote viewing and feedback submission functions. Furthermore, the feedback data in step (2) includes subjective evaluations of “temperature discomfort” and “light too bright” submitted by the elderly through the voice assistant, as well as heart rate variability (HRV) and blood oxygen saturation change data collected by the wearable device. The model updates the personalized demand threshold range every 72 hours based on the feedback data.

[0008] Compared with the prior art, the technical solution of this application has the following beneficial effects: 1. This invention, through in-depth analysis of individual data on the elderly, can accurately meet the unique needs of each senior citizen for their indoor environment, improving their living comfort and quality of life. It achieves automatic and intelligent control of the indoor environment, reducing the inconvenience of manual operation for the elderly, while improving the accuracy and timeliness of environmental control. 2. This invention links the indoor environment with the health status of the elderly, enabling timely detection of the impact of environmental factors on health, and providing early warnings and interventions, thus helping to prevent diseases and ensure the health and safety of the elderly. It innovatively applies AI technology to the design and control of elderly-friendly indoor environments, providing new ideas and methods for the intelligent development of indoor environments, demonstrating significant novelty and innovation. Attached Figure Description

[0009] Figure 1 This is a flowchart of the AI-based personalized map construction method for elderly-friendly indoor environments according to the present invention. Figure 2 This is a flowchart illustrating the implementation of the AI-based personalized indoor environment map construction method of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Please see Figure 1 and Figure 2 The AI-based personalized map construction method for elderly-friendly indoor environments in this embodiment is as follows: 1. Data Collection: Environmental sensor data: Deploy various environmental sensors, such as temperature and humidity sensors, air quality sensors, and light sensors, to collect various parameters of the indoor environment in real time. Elderly behavior data: Using devices such as cameras and infrared sensors, monitor the activity trajectory and behavioral habits of the elderly indoors (such as sleeping, getting up, walking, etc.). Health data: Collect physiological health data of the elderly, such as heart rate, blood pressure, and sleep quality, through wearable devices or smart home terminals. 2. AI Data Analysis: Data preprocessing: Cleaning, denoising, and normalizing the collected data to ensure accuracy and usability. Feature extraction: Using machine learning algorithms, key features are extracted from environmental data, behavioral data, and health data. For example, features such as activity frequency and activity time distribution are extracted from behavioral data. Model training: Build deep learning models, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), and train them on preprocessed data to establish a correlation model between environmental parameters, elderly behavior, and health status. 3. Personalized map construction: Needs Analysis: Based on the analysis results of the AI ​​model and combined with the subjective feedback of the elderly, we determine the individualized needs of each elderly person for the indoor environment. For example, some elderly people may be more sensitive to temperature changes and need more precise temperature control. Atlas Generation: Based on indoor spatial layout, this project visualizes individualized environmental needs, behavioral patterns, and health-related information to construct a personalized atlas of age-friendly indoor environments. The atlas includes suitable environmental parameters for different areas, space optimization suggestions based on behavioral habits, and environmental early warning information related to health conditions.

[0012] 4. Intelligent control and feedback: Intelligent control: Based on personalized data patterns, it automatically controls smart home devices, such as adjusting air conditioning temperature, adjusting light brightness and color temperature, and controlling air purifiers, to achieve intelligent and personalized control of the indoor environment. Feedback mechanism: Continuously monitor the elderly’s actual reactions to the indoor environment and changes in their health status, input the feedback data into the AI ​​model, and dynamically optimize and adjust the personalized map and control strategy.

[0013] Example This embodiment uses a two-bedroom apartment (70㎡, including a living room, master bedroom, secondary bedroom, bathroom, and kitchen) of 75-year-old Grandma Zhang, who lives alone, as the application scenario. The specific implementation process is as follows: I. Preliminary Preparations 1.1 Equipment Selection and Deployment Environmental sensors: Sensirion SHT31 temperature and humidity sensor (measurement range -40℃-125℃, accuracy ±0.3℃ / ±2%RH) and Pantene G5 air quality sensor (detects PM2.5 / PM10, accuracy ±5μg / m³) were selected. 3 The OSRAMSFH7770 light sensor (measurement range 0-10000lx, accuracy ±10lx) is deployed in 5 sets, one set in each functional area. The sensor communicates with the local gateway via a LoRa wireless module. The sampling interval is set to 10 minutes / time. The gateway is installed in a concealed place in the living room ceiling. Behavior monitoring equipment: Hikvision DS-2CD3T46DWD-I3 smart cameras (with human contour recognition and facial blurring functions, supporting 1080P resolution) are used, installed in three locations: the corner of the living room, the doorway of the master bedroom, and the doorway of the bathroom. The lenses are pointed towards the activity area and away from privacy areas. They are paired with Hyman HM-806 passive infrared sensors (detection distance 8m, angle 110°), deployed one-to-one with the cameras to achieve dual behavior capture of "visual + infrared". After the device data is processed by the edge computing box (Huawei Atlas200DK), only structured data such as behavior trajectory, time to get up / walk / stay are output, without storing the original images. Health data collection device: Grandma Zhang wore a Huawei Watch GT3 Pro senior smartwatch (supports heart rate monitoring, accuracy ±2 beats / minute; blood pressure monitoring, accuracy ±3mmHg; sleep quality score, resolution 1 point). The watch was connected to the Huawei Smart Home Terminal (Huawei Whole House Smart Host SE) via Bluetooth 5.0 to upload health data in real time. Smart home devices: Midea KFR-35GW / BP3DN8Y-YA401(1) inverter air conditioner (supports precise temperature adjustment to 0.5℃) and Opple Zhilang series smart ceiling light (supports brightness adjustment from 0-100% and color temperature adjustment from 2700K to 6500K) installed in the living room / master bedroom; Xiaomi air purifier 4Pro (CADR value 400m) installed in the bathroom / kitchen. 3 The device includes a / h humidifier and a Little Bear JSQ-A50U1 humidifier (supports humidity setting range of 30%-80%, accuracy ±5%RH). All devices are connected to the Huawei Smart Home system and support remote command control. 1.2 Software and Model Preparation Data processing software: A Python 3.9 environment was deployed locally, using the Pandas library for data cleaning, the NumPy library for data normalization, and the Scikit-learn library for principal component analysis (PCA) feature extraction; a TensorFlow 2.8 deep learning framework was built in the cloud, and the model training environment was constructed on an Ubuntu 20.04 server configured with an Intel Xeon E5-2680v4 CPU and an NVIDIA Tesla V100 GPU to ensure model training efficiency. Pre-trained model: The LSTM model was pre-trained using publicly available elderly behavior recognition datasets (such as UTKinect-Action3D and NTURGB+D). The input layer dimension was set to 64 (including 32-dimensional environmental features, 16-dimensional behavioral features, and 16-dimensional health features), the hidden layer was set to 3 layers (with 128, 64, and 32 neurons respectively), and the output layer dimension was set to 8 (corresponding to 8 categories of environment-behavior-health association results). The accuracy of the model after pre-training reached 89%, laying the foundation for subsequent personalized training. II. Specific Implementation Steps 2.1 Data Collection Phase (7 days) Environmental data acquisition: Five sets of sensors collect data at a frequency of 10 minutes per acquisition. For example, the living room sensor collected data at 8:00 on the first day, showing a temperature of 26.2℃, humidity of 52%, PM2.5 concentration of 18μg / m³, and light intensity of 450lx; the master bedroom sensor collected data at 23:00, showing a temperature of 24.5℃, humidity of 48%, and light intensity of 15lx. All data is transmitted to the local gateway in real time and then synchronized to the cloud server via the 4G network. Behavioral data collection: The camera and infrared sensor work together. When Grandma Zhang got up from the master bedroom at 7:30 on the third day, the infrared sensor triggered the detection signal first. The camera captured the human body contour trajectory, and the edge computing box analyzed and concluded that "getting up behavior, lasting 15 seconds"; when she watched TV in the living room at 10:00, the device recorded "static dwelling behavior, lasting 120 minutes". 200-300 pieces of structured behavioral data are generated every day, without original image storage. Health data collection: The smartwatch monitors Grandma Zhang's health indicators in real time. At 2:00 AM on the 5th day, it recorded a heart rate of 68 beats / minute and a blood oxygen saturation of 96%. At 7:00 AM, it generated a sleep quality score of 78 points for the previous night (deep sleep duration of 2.1 hours and light sleep duration of 4.3 hours). The health data is updated every 5 minutes and synchronized to the smart home terminal. 2.2 AI Data Analysis Phase (Day 8) Data preprocessing: The Pandas library was used to process 10,080 environmental data points, 1,800 behavioral data points, and 2,016 health data points collected over 7 days. Outliers (such as temperature data of -10℃ caused by sensor failure) were removed, and noise (such as misjudged data of "staying for 0.5 seconds" in behavioral data) was filtered out. Min-Max normalization was used to map all data to the [0,1] interval to ensure data consistency. Feature extraction: The PCA algorithm was used to reduce the dimensionality of the preprocessed data. Three core features were extracted from the environmental data: "temperature fluctuation range, humidity stability, and light change trend"; three core features were extracted from the behavioral data: "number of times to get up per day, average walking speed, and percentage of static stay"; and three core features were extracted from the health data: "heart rate variability coefficient, sleep quality fluctuation, and mean blood pressure". Finally, a 9-dimensional feature vector was obtained, with a feature contribution rate of 92%, which effectively reduced data redundancy. Personalized model training: The 9-dimensional feature vectors were divided into a training set (6300 features) and a test set (2700 features) in a 7:3 ratio. The pre-trained LSTM model was then input into the training set and trained using the Adam optimizer (learning rate set to 0.001) and the mean squared error (MSE) loss function. After 50 iterations, the model's accuracy on the test set improved to 94.5%. For example, the model can accurately identify the pattern that "when the temperature in the master bedroom is below 23℃ and Grandma Zhang's sleep quality score is below 70, the correlation between the two reaches 0.87". 2.3 Personalized Atlas Construction Phase (Day 9) Needs Analysis: Combining the model output results with Grandma Zhang's interview feedback (Grandma Zhang stated that she is afraid of being cold when sleeping and needs brighter warm light when watching TV in the living room), we determined the threshold for personalized needs: Master bedroom (sleep area): Temperature 23.5℃-25℃, humidity 45%-55%, light ≤20lx. Strong light / noise interference is prohibited during sleep. When getting up (7:00-8:00), the light will automatically gradually increase to 100lx. Living room (activity area): Temperature 25℃-26.5℃, humidity 40%-50%, PM2.5≤30μg / m³, during TV viewing hours (9:00-11:00, 19:00-21:00), light color temperature set to 3000K, brightness 70%; Bathroom (safe area): Temperature ≥24℃, humidity ≤60%. When people get up to use the toilet (22:00-6:00), the lights will automatically turn on to 50lx to avoid glare. Map Generation: A 3D space model of a two-bedroom apartment was built using SketchUp 2023 software, and the layout dimensions of each functional area (living room 25㎡, master bedroom 15㎡, bathroom 6㎡, etc.) were imported. Using the Python Matplotlib library, personalized demand thresholds, behavioral hotspots (such as the living room sofa area being a static hotspot, labeled "staying for about 3 hours per day") and health warning thresholds (such as triggering a yellow warning when the temperature in the master bedroom is below 23℃) were overlaid as layers to generate a visual map. The map supports mouse click interaction; clicking on the master bedroom area displays detailed parameters. The map was also exported as a PDF and sent to Grandma Zhang's family for remote viewing. 2.4 Intelligent Control and Feedback Phase (Continued operation from day 10 onwards) Intelligent control: Personalized mapping integrates with Huawei's smart home system to achieve automatic control. At 22:30 on the 10th day, the temperature in the master bedroom dropped to 23.2℃, which was below the lower threshold. The system automatically sent a command to the air conditioner to set the temperature to 24℃. The temperature rose back to 23.8℃ 30 minutes later. At 10:00 AM on the 11th day, Grandma Zhang was watching TV in the living room. The system detected a light intensity of 320 lx and a color temperature of 5000 K, and automatically adjusted the light brightness to 70% and the color temperature to 3000 K, which met the "warm light" requirement. At 3:00 AM on the 12th day, Grandma Zhang got up to use the toilet. The bathroom's infrared sensor was triggered, and the lights automatically turned on to 50 lx. They automatically turned off 5 minutes after she finished using the toilet. Feedback and Optimization: The system continuously collects feedback data and dynamically optimizes the map and strategy. On the 14th day, Grandma Zhang reported through the voice assistant that "the air conditioner in the living room is too cold." The system recorded this subjective evaluation and, combined with the health data at the same time (heart rate fluctuation ±3 beats / minute, no abnormalities), adjusted the air conditioner fan speed in the living room from "high" to "medium" and updated the "Recommended air conditioner fan speed in the living room" parameter in the graph. On the 21st day, the model was iterated and optimized based on feedback data from the past two weeks. The suitable temperature range for the master bedroom was slightly adjusted to 24℃-25.5℃ (originally 23.5℃-25℃). Since the data showed that Grandma Zhang's sleep quality score improved by an average of 5 points when the temperature was above 24℃, the optimized map was pushed to Grandma Zhang's family's APP through the smart home terminal. After confirmation that there were no objections, it was officially put into use. III. Implementation Results Verification 3.1 Improved comfort One month after implementation, based on Grandma Zhang's daily subjective ratings (1-10 points), the average score for living comfort increased from 6.2 points before implementation to 8.8 points. Among them, "sleep comfort" increased from 5.8 points to 9.1 points, and "activity area comfort" increased from 6.5 points to 8.5 points, confirming the effectiveness of personalized environmental control. 3.2 Improvement in health indicators Data from the smartwatch showed that Grandma Zhang's deep sleep duration increased from 1.8 hours / night to 2.5 hours / night, her blood pressure fluctuation range decreased from ±8 mmHg to ±5 mmHg, and the number of abnormal heart rates (below 60 beats / minute or above 80 beats / minute) decreased from 4 times per week to 1 time per week, indicating that the regulation of the relationship between environment and health has played a positive role. 3.3 Intelligent Efficiency The system's automatic adjustment response time averages 2.3 seconds (from data anomaly to device execution command), eliminating the need for Grandma Zhang to operate it manually and solving her problems of "not knowing how to adjust the air conditioner and not being able to see the light buttons clearly." Family members only need to remotely view the chart 1-2 times per month, significantly reducing the caregiving burden and verifying the practicality and convenience of the method.

[0014] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0015] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based elderly-friendly indoor environment individualized map construction method, characterized by, The method comprises the following steps: (1) Data collection: deploying environmental sensors in key indoor areas, collecting indoor temperature, humidity, air quality and light intensity parameters through the environmental sensors; using a combination of cameras with privacy protection function and infrared sensors to monitor the activity trajectory of the elderly in the indoor, as well as sleep, get up, walk behavior data through the combination of devices; using wearable devices or smart home terminals to collect physiological health data such as heart rate, blood pressure and sleep quality of the elderly in real time; (2) AI data analysis: sequentially performing outlier rejection, noise filtering and normalization preprocessing on the multi-dimensional data collected in step (1); using principal component analysis (PCA) algorithm or wavelet transform algorithm to extract environmental parameter features, behavior time distribution features and health data fluctuation features from the preprocessed multi-dimensional data; Constructing a deep learning model based on long short-term memory network (LSTM), dividing the data set after feature extraction into training set and test set in the ratio of 7:3, and iteratively training the deep learning model through Adam optimizer to obtain a correlation model that can associate environmental parameters, behavior patterns and health status; (3) Personalized atlas construction: based on the output results of the correlation model trained in step (2), combining the feedback information obtained by interviewing the elderly and their families, determine the personalized indoor environment demand threshold of each region and time period; using 3D modeling software to load the indoor space layout basic model, marking the appropriate environmental parameter range of each region, behavior hotspot area and health risk environment warning threshold, and superimposing them on the indoor space layout basic model in the form of visual layers to generate a personalized indoor environment friendly atlas for the elderly containing space coordinates, parameter threshold and warning identifier; (4) Intelligent regulation and feedback: establish communication connection between the personalized atlas generated in step (3) and the intelligent home control system, and automatically send regulation instructions to air conditioners, lights, humidifiers and air purifiers according to the parameters in the personalized atlas; Continuously collect the feedback data of the elderly on the indoor environment and the health index change data of the elderly, input the collected feedback data and health index change data into the correlation model every 24 hours, iteratively optimize the parameters of the correlation model, and dynamically update the personalized atlas and the corresponding environmental regulation strategy according to the optimized model parameters. 2.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, The environmental sensors in step (1) use Bluetooth Low Energy (BLE) communication protocol or LoRa wireless communication protocol; the environmental sensors are respectively deployed in the living room, bedroom, bathroom and kitchen area of the indoor; the collection frequency of the environmental sensors is set to 5-15 minutes / time; the sampling accuracy of the environmental sensors meets the requirements of temperature measurement error ±0.5℃, humidity measurement error ±3%RH, light measurement error ±10lx and PM2.5 concentration measurement error ±5μg / m³. 3.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, The camera in step (1) has human body contour recognition function and face blur processing function; the infrared sensor adopts passive infrared detection (PIR) technology; the image data collected by the camera and the detection data collected by the infrared sensor are fused, the fused data are subjected to behavior recognition through an edge computing module, and the behavior recognition delay is not more than 100 ms; the original image data collected by the camera is not stored in a local device or a cloud server. 4.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, In step (2), the data preprocessing further includes data time sequence alignment processing, through which the environment data, behavior data and health data are unified and synchronized to a millisecond level timestamp; in the feature extraction process, an attention mechanism (Attention) is adopted to strengthen the weight proportion of the "getting up frequency" and "walking speed change" key features in the behavior data. 5.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, In step (2), the LSTM model includes an input layer, three hidden layers and an output layer; the number of neurons in the three hidden layers is set to 128, 64 and 32 respectively; the Dropout technology is adopted to suppress the overfitting phenomenon in the LSTM model training process, and the Dropout probability is set to 0.2; the number of training iterations of the LSTM model is not less than 50 rounds; the loss function of the LSTM model adopts mean square error (MSE). 6.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, In step (3), the individualized environment demand threshold specifically includes: sleeping period bedroom temperature 22-26℃, bedroom humidity 40-60%, bedroom light intensity ≤50lx; active period living room temperature 24-28℃, living room humidity 35-55%, living room light intensity 150-300lx; the individualized atlas supports touch operation and voice interaction query function; the warning signs in the individualized atlas adopt three colors of red, yellow and green, corresponding to high, medium and low health risk levels respectively. 7.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, In step (4), the transmission of the regulation and control instructions between the smart home control system and the individualized atlas adopts MQTT communication protocol; when it is monitored that the sleep quality score of the elderly decreases by ≥20% and the indoor humidity is lower than 40%, the smart home control system automatically increases the working power of the humidifier by 30-50%, and records the sleep data changes of the elderly within 2 hours after regulation and control; in the iterative optimization process of the correlation model, the incremental learning (Incremental Learning) technology is adopted, and the correlation model parameters can be updated without retraining all historical data. 8.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, It also includes a privacy protection step: the collected behavior data and health data are stored by using blockchain encryption technology; only authorized users can access the encrypted stored behavior data and health data after fingerprint recognition or face recognition verification; the environment sensor and the wearable device both have data local desensitization processing function; the transmission data between the environment sensor, the wearable device and the data receiving end adopts end-to-end encryption algorithm. 9.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein, The 3D modeling software in step (3) is SketchUp software or Blender software; the personalized atlas is supported to be exported in PDF format or PNG format after being generated; the personalized atlas can be synchronized to a mobile terminal APP of the children of the old people, and remote viewing of the personalized atlas and feedback submission of indoor environment requirements of the children of the old people are realized through the mobile terminal APP. 10.The AI-based, elderly-friendly indoor environment individualization map construction method of claim 1, wherein: The feedback data in step (2) include subjective evaluation of "temperature is not suitable" and "light is too bright" submitted by the old people through a voice assistant, and heart rate variation coefficient (HRV) and blood oxygen saturation change data collected by a wearable device, and the model updates the personalized requirement threshold range once every 72 hours according to the feedback data.

Images