A sensor-based automatic control method for sliding doors

Abstract

The application provides a sensor-based automatic control method for sliding doors, which comprises the following steps: obtaining the frequency, stay time and number of people of relevant personnel entering and exiting a specific area in different time periods, establishing a flow model to reflect the personnel flow situation; recording the disinfection frequency of the environment inside and outside the area, the type and concentration of disinfectant used, analyzing the effective time and disinfection effect of the disinfectant under different conditions, and ensuring the continuous cleanliness of the specific area; collecting gesture data of different personnel, analyzing the accuracy, recognition speed and error rate of the gesture data, and obtaining operation state data of gesture recognition; inputting the gesture recognition data and disinfection step data into a comprehensive control system, and judging whether to open the area entrance by analyzing the completion degree of the disinfection step and the recognition accuracy of the gesture.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a sensor-based automatic control method for sliding doors. Background Technology

[0002] Managing personnel access to specific areas presents several technical contradictions and challenges. The flow of people within these areas is complex and constantly changing, especially during emergencies when entry and exit are frequent. Ensuring data integrity and accuracy without disrupting normal operations is a major challenge. While collecting data on disinfection measures and agents can analyze their effectiveness and create models, the unique environment, diverse types of disinfectants, and varying disinfection procedures among staff present significant technical challenges. Selecting appropriate data collection points and methods, and organically integrating them with personnel flow data, is also a challenge. One particularly difficult aspect is using gesture recognition sensors to collect and analyze gestures, combined with personnel disinfection procedures, to determine whether to allow access. This requires gesture recognition to adapt to the complex and changing environment and to tightly integrate gesture data with disinfection effectiveness models to build a real-time and reliable access control strategy. Furthermore, accurately predicting the likelihood and urgency of emergencies based on real-time personnel flow and disinfection effectiveness data, and promptly adjusting access control strategies in emergencies to quickly open access and allow emergency personnel to enter the area, is another technical hurdle. Summary of the Invention

[0003] This invention provides a sensor-based automatic control method for sliding doors, mainly comprising:

[0004] The frequency, duration of stay, and number of people entering and leaving a specific area at different time periods are obtained to establish a flow model to reflect the flow of people.

[0005] Record the frequency of disinfection in the area and outside the region, the types and concentrations of disinfectants used, analyze the effective time and disinfection effect of disinfectants under different conditions, and ensure the continuous cleanliness of specific areas;

[0006] Record the disinfection steps, time, and disinfectant usage before personnel enter a specific area, monitor the implementation of disinfection steps, ensure the effectiveness of disinfection operations, and obtain data on the maintenance of a clean environment within the environment;

[0007] Collect gesture data from different individuals, analyze the accuracy, recognition speed, and error rate of the gesture data, and obtain the operational status data of gesture recognition;

[0008] The gesture recognition data and disinfection step data are input into the integrated control system. The system analyzes the completion rate of the disinfection steps and the correctness of the gesture recognition to determine whether to open the area entrance.

[0009] Real-time monitoring of earthquake and fire data within the area; analysis of early warning signals for earthquake or fire emergencies; detection of abnormal changes, including rapid changes in earthquake vibration data and fire smoke concentration; prediction of emergency situations; and determination of whether it is necessary to open the gates for evacuation if the integrated control system detects earthquake or fire signals.

[0010] When the sensor-controlled sliding door automatic system detects an emergency, it responds quickly, switches to emergency mode, cancels the gesture recognition requirement, ensures that people can quickly enter or leave the area, and issues an alarm to notify relevant personnel to take emergency measures.

[0011] After the emergency ends, collect and analyze all data from the emergency response process, including personnel flow, system response speed, and the accuracy of gate control. Based on the analysis results, optimize the sensor-based automatic sliding door control system.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0013] This invention discloses a sensor-based automatic control method for sliding doors, used to monitor and manage personnel flow, disinfection implementation, and emergency response within a specific area. A flow model is established to reflect personnel movement, helping to optimize personnel distribution and flow strategies, reduce congestion, and improve the efficiency of area security management. By recording the disinfection steps, time, and disinfectant usage before personnel enter the specific area, and monitoring the execution of disinfection steps, the effectiveness of disinfection operations is ensured, providing a safe environment for personnel within the area. Combining gesture recognition data with disinfection step execution data, and analyzing this data to determine whether to open the area entrance, this integration improves the safety and efficiency of the sensor-based automatic sliding door control system. In emergency situations, this invention can respond quickly, eliminating the need for gesture recognition, ensuring personnel can quickly enter or leave the area, and issuing alarms to notify relevant personnel to take emergency measures. Overall, this invention not only improves the safety and cleanliness of a specific area but also optimizes emergency response strategies, providing a higher level of protection and convenience for personnel within the area. Attached Figure Description

[0014] Figure 1 This is a flowchart of a sensor-based automatic control method for sliding doors according to the present invention.

[0015] Figure 2 This is a schematic diagram of a sensor-based automatic control method for sliding doors according to the present invention.

[0016] Figure 3 This is another schematic diagram of a sensor-based automatic control method for sliding doors according to the present invention. Detailed Implementation

[0017] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0018] like Figure 1-3 This embodiment of a sensor-based automatic control method for sliding doors may specifically include:

[0019] Step S101: Obtain the frequency, duration of stay, and number of people entering and leaving a specific area at different time periods, and establish a flow model to reflect the flow of people.

[0020] Based on the unique identification information of the personnel, the time when personnel enter and exit a specific area is recorded; by analyzing the changes in the personnel's location coordinates, the flow direction and movement trajectory of personnel within the specific area are determined; based on the frequency of personnel in different time periods, the distribution of personnel flow frequency in different time periods is obtained; by calculating the changes in the personnel's location within the specific area, the duration of each person's stay is obtained; based on the personnel flow frequency and number of people in different time periods, a statistical report reflecting the overall personnel flow trend is generated; a flow model is established based on historical personnel flow data to obtain the predicted future personnel flow; the personnel flow prediction results are combined with the regional map to obtain a heat map showing the personnel flow trajectory and stay in different areas.

[0021] For example, unique identification information of individuals is obtained, and their timestamp data for entering and exiting specific areas is recorded. Based on changes in the location coordinates of individuals and timestamp information, their flow direction and movement trajectory within the specific area are determined. Specific areas can be medical institutions such as hospitals and clinics, food processing and packaging areas, research laboratories and biosafety laboratories, public transportation such as airports, train stations, and buses, and educational institutions such as schools and kindergartens. The frequency of individuals within different time periods is statistically analyzed, and time series analysis methods such as moving averages and exponential smoothing are used to characterize the trend of flow frequency, obtaining the distribution of personnel flow frequency in different time periods. By analyzing changes in the location of individuals within the specific area, the dwell time of each individual is calculated, and the average dwell time of all individuals within each time period is calculated, summarizing individual dwell times into the overall data. Combining the personnel flow frequency, number of people, and dwell time within different time periods, methods such as clustering and association rule mining are used to discover personnel flow patterns, generating a statistical report reflecting the overall personnel flow trend. The raw personnel flow data is cleaned, transformed, and feature-extracted to prepare for training machine learning models. This study employs recurrent neural networks such as LSTM to model time-series data, uses a Seq2Seq model for multi-step prediction, and optimizes model hyperparameters through cross-validation and grid search. A flow model is built based on historical population movement data to predict future population flows. The results of the population flow analysis are combined with regional maps to visually display the flow trajectories and heatmaps of people's stays in different areas, providing a basis for regional management decisions. The established flow model and prediction results are quantitatively and qualitatively evaluated, and the flow model is iteratively optimized based on practical application effects. When obtaining unique identification information for individuals, technologies such as RFID, QR codes, or facial recognition can be used. For example, each person can wear a uniquely coded RFID card, recording the timestamps of their entry and exit from specific areas with millisecond-level accuracy. Then, the Kalman filter algorithm is used to smooth the position coordinates of individuals. Combined with the timestamp information, the displacement vector between two adjacent time points is calculated at 1-second intervals to determine the flow direction. Simultaneously, by connecting these displacement vectors, the movement trajectory of individuals can be obtained. Next, a sliding time window method was used, with a 10-minute window and a 5-minute sliding step, to statistically analyze the frequency of people within different time periods. Within each window, exponential smoothing was applied to calculate the average frequency of people movement, with a smoothing coefficient set to 0.6 to better reflect the frequency trend. For each person, their dwell time was calculated by accumulating the time difference between two adjacent time points within a specific area. Then, the average dwell time for all people was calculated in time intervals. Based on the above data, K-means clustering was used to classify the people movement patterns into 3 to 5 categories, each representing different movement characteristics.After preprocessing the raw data, including denoising and normalization, features such as personnel ID, timestamps, location coordinates, and flow direction are extracted and input into an LSTM model for training, resulting in a flow model. Denoising can include box plot analysis, such as the IQR method, to identify outliers in the data and decide whether to remove or replace them; these outliers can be replaced with the median or mean. Normalization aims to adjust features at different scales to the same scale range, facilitating model learning and comparison; it can convert the data to a standard normal distribution, using the formula xnorm = σx - μ, where μ is the mean and σ is the standard deviation. The flow model can be configured with three hidden layers, each with 128 neurons, using the Adam optimizer with a learning rate of 0.001. After 50 epochs of training, the flow model can predict personnel flow within the next hour with an average accuracy of over 85%. Personnel flow trajectories are plotted on a regional map using arrows of different colors, with arrow thickness representing flow frequency; a heatmap is used to display the distribution of personnel dwell time, with darker colors indicating longer dwell times. By comparing the data with actual pedestrian flow data, the predictive performance of the flow model can be evaluated, and the flow model parameters can be continuously adjusted to improve the accuracy of the prediction.

[0022] Step S102: Record the disinfection frequency, type and concentration of disinfectant used in the area and outside the area, analyze the effective time and disinfection effect of disinfectant under different conditions, and ensure the continuous cleanliness of the specific area.

[0023] Based on the usage and contamination levels of specific areas, corresponding disinfection frequency plans are developed; the effectiveness of different types and concentrations of disinfectants is evaluated, and the kill rate of bacteria and viruses under different conditions is tested; the concentration and spraying time of the disinfectants are adjusted according to their chemical composition and environmental conditions; the cleanliness of specific areas is assessed to obtain predicted potential infection risks; the effective time of different disinfection schemes under different conditions is calculated by combining bacterial types, virus types, and environmental humidity factors to obtain an optimized disinfection process; based on the environmental layout and distribution of contamination hotspots in specific areas, disinfection robots are used to clean disinfection dead corners and high-risk areas to improve disinfection coverage and efficiency.

[0024] For example, environmental monitoring sensors are deployed inside and outside a specific area to transmit collected environmental parameters such as temperature and humidity to a comprehensive control system. Based on the usage, layout characteristics, and contamination level of the specific area, a scientific disinfection frequency plan is developed, and automatic reminders and recording functions are set up in the comprehensive control system to ensure that disinfection work is carried out according to plan. The effectiveness of different types and concentrations of disinfectants is evaluated, using standard methods such as quantitative suspension sterilization tests and surface sterilization tests to test aspects such as sterilization efficiency, residual toxicity, and material compatibility. A disinfectant effectiveness database is established to provide a basis for selecting the optimal disinfection scheme. Intelligent control modules are installed on disinfection equipment to automatically adjust the operating parameters of the disinfection equipment, such as spray pressure and atomization particle size, based on the chemical composition of the disinfectant and environmental conditions to optimize the disinfection effect. Based on a monitoring platform for a specific area, information such as environmental sensor data, disinfection records, and bacterial detection results is integrated. Machine learning algorithms, such as support vector machines and random forests, are used to assess the cleanliness status of the specific area in real time and predict potential infection risks. Correlation analysis is used to explore the relationship between environmental factors and disinfection effectiveness, providing data support for the evaluation model. A disinfection effectiveness evaluation model was established, comprehensively considering factors such as bacterial type, virus type, and environmental humidity. Using methods such as multiple linear regression and time series analysis, and through steps including data collection and preprocessing, feature selection and extraction, model training and validation, and model optimization and application, the effective time of different disinfection schemes under specific conditions was calculated, and the disinfection process was dynamically optimized. SHT31 temperature and humidity sensors were deployed in a specific area, collecting data every 5 minutes and transmitting the data to the integrated control system via the Zigbee protocol. The ATP biofluorescence rapid detection method was used to quantify the degree of contamination. Based on the assessment results of the contamination level in the specific area, a disinfection scheme of four times a day, each lasting 20 minutes, was set, and scheduled tasks and log records were created. Three commonly used disinfectants, such as peracetic acid and quaternary ammonium salts, were selected and samples were prepared at concentrations of 0.1%, 0.5%, and 1%. Bactericidal tests were conducted on five common bacteria, including Staphylococcus aureus and Escherichia coli, using the European standard EN13697. A logarithmic value (LRV) ≥4 was used as the pass criterion. Statistical analysis of the experimental data was performed using the Pandas library in Python to determine the optimal combination of disinfectant types and concentrations. Before determining the optimal disinfectant types and concentrations, cytotoxicity assessments of candidate disinfectants were conducted to ensure they were harmless to humans. A C8051F microcontroller and solenoid valve control module were integrated into the disinfection equipment to dynamically adjust the disinfectant atomization particle size between 20-30 μm based on ambient temperature and humidity, maintaining the spray pressure between 0.2-0.4 MPa to ensure droplet suspension time and coverage. A monitoring platform based on the Spring Boot framework was built, using Redis to cache sensor data, and a scheduled task was used daily at midnight to aggregate and clean the data.Eight feature variables, including temperature, humidity, and disinfectant concentration, were selected to train a support vector machine regression model. This model was used to score and predict the cleanliness status of the area, and 5-fold cross-validation was used to evaluate the model's performance, achieving an average R² value of over 0.85. A collaborative filtering algorithm was used to analyze the correlation between environmental factors and disinfection effectiveness, constructing a similarity matrix and generating association rules and confidence scores to provide a calibration basis for the effectiveness evaluation model. A disinfection effectiveness evaluation model was established by combining multiple linear regression and ARIMA time series models, using temperature, humidity, disinfectant type and concentration as independent variables and effective disinfection time as the dependent variable. Model training and parameter optimization were performed. The mean absolute error (MAE) of the disinfection effectiveness evaluation model on the test set was controlled within 15 minutes. Based on real-time environmental data and disinfection records, it can dynamically predict the optimal disinfection duration under different conditions, guiding the cleaning and disinfection work in the area.

[0025] Step S103: Record the disinfection steps, time and disinfectant usage before personnel enter the specific area, monitor the implementation of the disinfection steps, ensure the effectiveness of the disinfection operation, and obtain the maintenance data of the clean environment within the environment.

[0026] The entry time and sequence of each person are recorded by RFID readers installed in the disinfection channel; the disinfection operation process is standardized to guide personnel to complete the disinfection according to the steps and duration of the standardized disinfection operation process, while the disinfection process is monitored in real time to determine whether the disinfection operation is in place; air samples are collected before and after personnel enter the channel to identify the types and concentrations of bacteria in the air samples and to evaluate the effectiveness of the disinfection operation; the environment of specific areas is sampled and tested regularly to analyze the types and quantities of pathogenic microorganisms in the environment and to evaluate the cleanliness of the specific area and the long-term effectiveness of the disinfection measures.

[0027] For example, a facial recognition system and infrared thermometer are installed at the entrance of a specific area to collect the identity information and body temperature data of people entering in real time, and the data is uploaded to a central management platform. Using radio frequency identification (RFID) technology, each person is equipped with a unique electronic tag, and RFID readers are installed in the disinfection channel to automatically record the entry time and sequence of personnel. The disinfection operation process is standardized, and a visual operation interface can be designed to guide personnel to complete the disinfection according to the standardized steps and duration. Simultaneously, cameras and a target detection and attitude estimation model based on convolutional neural networks monitor the disinfection process in real time to determine whether the operation is in place. Disinfectant spraying devices and flow meters are installed in the disinfection channel. Based on physiological parameters such as height, weight, and body surface area of ​​personnel, combined with the physicochemical properties and disinfection efficacy of the disinfectant, the spraying volume and spraying time are optimized using a target detection and attitude estimation model based on convolutional neural networks, and the type and amount of disinfectant used for each disinfection is recorded. A nanoparticle detector was used to collect air samples before and after personnel entry. Gas chromatography-mass spectrometry (GC-MS) was used to identify the types and concentrations of bacteria in the air, assessing the immediate effectiveness of disinfection operations. The long-term effectiveness of disinfection measures was evaluated by comparing and analyzing air quality trends at the entrance and within the designated area. Regular environmental sampling and testing of specific areas were conducted. Microbial communities were classified and functionally annotated using 16S rRNA and metagenomics methods. Unsupervised learning algorithms, such as isolated forest and local anomaly factor analysis, were used to evaluate the cleanliness of the environment and the long-term effectiveness of disinfection measures. Data collected at each stage was cleaned and transformed using ETL tools to build a data warehouse and data mart. OLAP technology was applied for multi-dimensional analysis, and the data was aggregated to a central management platform. At the entrance, a high-definition camera and RetinaFace facial recognition algorithm were used for real-time identification of personnel, achieving an accuracy rate of over 99.5%. Simultaneously, an MLX90614 infrared thermometer was used to measure body temperature, with a temperature error of less than 0.2℃. Using RFID UHF readers and electronic tags, personnel identity verification and entry time recording are completed within 0.5 seconds, with a reading distance of up to 5 meters. The OpenPose human posture estimation model detects the coordinates of key body parts in real time to determine whether disinfection actions are standardized, achieving a detection accuracy of over 98%. Combining personnel physiological parameters and the chemical composition of the disinfectant, multi-physics coupling simulation is used to calculate the optimal spray volume and spray time. For example, for a 75kg person, using a 500mg / L hydrogen peroxide disinfectant, the spray volume is 20ml, and the spray time is 15 seconds. A TSI3330 nanoparticle detector is used to sample airborne bacterial aerosols, with a particle size detection range of 10-1000nm and a sampling flow rate of 16.7L / min. Qualitative and quantitative analysis of air samples is performed using an Agilent 7890B gas chromatograph and a 5977A mass spectrometer, with a detection limit of 0.1pg / mL.Comparative analysis revealed that the bacterial concentration in the disinfected area decreased by an average of over 90%, and the concentration remained at 100 CFU / m³ for 30 consecutive days of monitoring. 3 The following data was generated. Metagenomic sequencing of environmental samples was performed using the Illumina NovaSeq 6000 sequencing platform, achieving an average sequencing depth of 50X and a data volume of 20GB. Species classification and abundance estimation were performed on the sequencing data using Kraken2 and Bracken software. The dominant bacteria were identified as *Acinetobacter* and *Bacillus*, while the relative abundance of potential pathogens such as *Staphylococcus aureus* remained consistently below 0.01%. Anomaly detection was performed using the LocalOutlierFactor algorithm, successfully identifying two instances of abnormal bacterial increases due to insufficient disinfectant usage, with 100% accuracy. Data cleaning and feature engineering were performed using Pandas and NumPy libraries, constructing a data warehouse containing 1 million records and 50 fields. Multidimensional analysis was performed using GolangCube, with an average query response time of less than 1 second. By combining Apriori association rule mining and LightGBM decision tree model, it was found that the frequency of hand disinfection and the concentration of disinfectant are the most important factors affecting the disinfection effect. Based on this, the disinfection process was optimized, and the average compliance rate increased from 85% to over 95%.

[0028] Step S104: Collect gesture data from different people, analyze the accuracy, recognition speed and error rate of the gesture data, and obtain the operation status data of gesture recognition.

[0029] The system acquires hand movement sequence images and thermal imaging data of individuals entering a specific area and transmits them to a server in real time for processing. Key features of the hand movement sequence images are extracted and classified, and gesture trajectories are modeled to identify different gesture types. By calculating the speed and angle changes during gesture movement, it determines whether an abnormal gesture event has been triggered. If an abnormal gesture event is triggered, gesture recognition is paused, and the veracity of the event is reviewed. If the event is confirmed, it is determined whether it represents an intrusion by an unauthorized individual. For different individuals' gesture habits, a personalized gesture template library is used for template matching to achieve adaptive adjustment of gesture recognition.

[0030] For example, a 3D depth camera and infrared sensor array are installed at the entrance of the area to collect images of hand gesture sequences and thermal imaging data of people entering the area. The raw data is transmitted to an edge server for processing in real time. A gesture recognition algorithm based on skeletal point detection is adopted. Key features of the gesture images are extracted through a convolutional neural network, and then a long short-term memory network is used to model the gesture trajectory. Different gesture types are identified through the gesture recognition model. Compared with traditional methods such as hidden Markov models and dynamic time correction, this algorithm has higher accuracy and real-time performance in complex gesture recognition, achieving an accuracy of over 95% on public datasets. Using infrared sensor data, the speed and angle changes during the gesture movement are calculated. Through statistical analysis of different gesture types, a reasonable threshold range is obtained. When the speed or angle change exceeds the normal range, such as an upward waving speed exceeding 1.5 m / s, a gesture abnormality event is triggered, and the abnormal data is sent to the monitoring platform for review. If a gesture abnormality event is triggered, gesture recognition is first suspended, and the abnormality event is reviewed to determine whether it is an intrusion by an unauthorized person. Once an unusual gesture is detected, the system automatically records the time, location, relevant video footage, and facial features of the person entering (without infringing on privacy), packaging this information into an incident report. This report is instantly sent to the monitoring platform via an encrypted channel, simultaneously triggering an SMS or app push notification to security personnel, indicating "suspected unauthorized intrusion." Upon receiving the notification, monitoring platform staff immediately review the incident report, reviewing relevant video footage and combining it with other on-site monitoring footage and historical records to determine whether the gesture truly represents abnormal behavior or is a false alarm. If manual review confirms the unusual gesture as a potential security threat, security personnel immediately take action, such as blocking the entrance, on-site intervention, or notifying a higher-level security response team. The collected gesture data undergoes preprocessing, employing median filtering to remove random noise, morphological closing operations to remove holes in the gesture contour, and extracting features such as the gesture contour and keypoint coordinates to construct a gesture feature dataset containing 100,000 samples. Support Vector Machine (SVM) algorithms are used to classify gesture features. Grid search is employed to optimize the gesture recognition model's hyperparameters, such as the penalty coefficient C and the gamma value of the radial basis function kernel. Five-fold cross-validation is used to evaluate the model's performance, and the parameter combination with the highest F1 score is selected, reducing the gesture recognition error rate to below 1% and the recognition latency to within 100 milliseconds. To further improve recognition accuracy and robustness, a personalized gesture template library is integrated into the gesture recognition process, and template matching is performed using a dynamic time warp algorithm. The dynamic time warp algorithm calculates the similarity between two gesture sequences by finding the optimal nonlinear alignment. For example, for two gesture sequences of length n and m, an n*m cost matrix is ​​constructed, and dynamic programming is used to solve for the optimal matching path, achieving adaptive adjustment of gesture recognition. This can improve the recognition accuracy by more than 5% in complex environments.The system monitors key indicators such as recognition accuracy, average response time, and CPU utilization of the gesture recognition system in real time. When these indicators exceed preset thresholds, such as recognition accuracy falling below 90% or response time exceeding 500 milliseconds, an alarm and fault diagnosis process is automatically triggered to ensure the stable operation of the gesture recognition system. The sensor-based automatic control system for sliding doors communicates with the gesture recognition system. A comprehensive performance evaluation of the gesture recognition system is conducted, designing test cases encompassing different gesture types, environmental conditions, and user groups. Quantitative analysis is performed on indicators such as recognition accuracy, false alarm rate, missed detection rate, real-time response time, and resource consumption. The system is compared with other mainstream gesture recognition systems to verify its effectiveness and advancement. Simultaneously, the time and space complexity of the algorithm are analyzed, and the critical path is optimized. Techniques such as CPU and GPU co-acceleration, model quantization, and gradient pruning are employed to achieve real-time performance and scalability of gesture recognition, supporting the processing of over 1000 gesture recognition requests per second. Two RealSense D435 depth cameras and three MLX90640 infrared sensors were installed at the area entrance to acquire hand motion images and infrared thermal images at a rate of 30 frames per second, which were then transmitted to an edge server via gigabit Ethernet. The OpenPose skeletal point detection algorithm was used to extract 21 key hand points, which were then input into a three-layer convolutional neural network for feature extraction. A two-layer LSTM network was then used to perform sequence learning on the 128-dimensional gesture features, with softmax outputting the probabilities of 12 gesture categories. Combined with skin surface temperature changes measured by the infrared sensors, gesture speed and angle were calculated. An abnormal gesture was identified when the temperature in the palm area dropped by more than 2.5℃ / s, or when the angular velocity of the index fingertip exceeded 200° / s, and the corresponding image frames and temperature data were uploaded to the monitoring platform. The gesture images were then subjected to Gaussian filtering, binarization, erosion, and dilation operations to extract the gesture contour and fingertip position, and normalized to 128*128 pixels. 100,000 processed images were randomly selected and divided into training and test sets in an 8:2 ratio. The optimal parameters for the SVM model were found using the RBF kernel function and grid search: C=10, gamma=0.01. The average F1 score of the 5-fold cross-validation reached 0.98, and the single-frame gesture recognition time was less than 50ms. A template library containing commonly used gestures was established for each user. Image sequences were matched using a dynamic time warping algorithm, and the recognition threshold was adaptively adjusted, improving the recognition accuracy by 8% in complex backgrounds. The gesture recognition system performance monitoring component calculates core metrics every 30 seconds. If the recognition accuracy falls below 85% three times consecutively within one minute, or the average response time exceeds 250ms, an alarm message is sent to the on-duty personnel, and the process is automatically restarted and the model is restored.One hundred typical test scenarios can be designed to conduct a continuous 72-hour stability test on the gesture recognition system. The average recognition accuracy remains above 95%, the false alarm rate and false positive rate are controlled within 1% and 0.5% respectively, and the response time for 95% of requests is less than 1%.

[0031] Step S105: Input the gesture recognition data and disinfection step data into the integrated control system. Analyze the completion of the disinfection steps and the correctness of the gesture recognition to determine whether to open the area entrance.

[0032] The gesture recognition data and disinfection step execution data are transmitted to the integrated control system in real time. This includes separating the gesture recognition results and the disinfection step completion status into independent services and communicating with the integrated control system via an RPC interface. For the gesture recognition results, the system determines whether the current gesture recognition is correct based on features such as gesture category, confidence level, and historical recognition accuracy. For the disinfection step completion status, a finite state machine model is used to determine whether the disinfection step is complete based on parameters such as disinfectant spraying time, hand coverage area, and the number of gesture changes. The results of the gesture recognition and disinfection step determinations are then comprehensively evaluated to calculate a comprehensive score.

[0033] For example, the JSON-formatted data generated by the gesture recognition system and disinfection step monitoring module is transmitted to the integrated control system in real time via the message queue middleware Kafka, with a peak data volume of up to 1000 messages per second. In the integrated control system, Redis is used to cache hot data with a 5-second expiration time to ensure data real-time performance. Simultaneously, MongoDB is used to store all historical data, supporting subsequent data analysis and auditing. MongoDB employs a sharded cluster architecture, horizontally partitioning the data according to the time dimension and creating sparse indexes on key fields, achieving a data write and query speed of up to 5000 times per second. The integrated control system adopts a microservice architecture design, dividing functions such as gesture recognition, disinfection step monitoring, data storage, rule engine, risk assessment, and equipment control into independent service units. These units communicate synchronously via REST API and asynchronously via Kafka, achieving a service throughput of over 1000 transactions per second. For judging gesture recognition results, the rule-based expert system Drools is used. Based on indicators such as gesture category, confidence level, and historical recognition accuracy, a series of if-then rules are set, such as "if gesture category = 'No Entry' and confidence level > 0.95 and accuracy > 0.99, then gesture recognition result = 'Invalid'". This comprehensively evaluates the gesture recognition results, improving the accuracy and interpretability of the judgment. For judging the completion status of disinfection steps, a finite state machine model is used, abstracting the disinfection process into four states: "Initialization", "In Progress", "Completed", and "Abnormal". Based on parameters such as disinfectant spraying time, hand coverage area, and number of gesture changes, state transition conditions and actions are set, such as "if gesture = 'Rotation' and duration > 5, then state = 'Completed'". To ensure the fault tolerance of the state machine, reasonable threshold ranges are set for each parameter in the transition conditions. When an out-of-bounds or interruption occurs, an alarm is automatically triggered and the system transitions to the "Abnormal" state. In the comprehensive evaluation phase, the Analytic Hierarchy Process (AHP) is used to compare and score multiple indicators across various dimensions, such as gesture recognition and disinfection procedures, forming a judgment matrix to calculate indicator weights. This weighted composite score vector is then formed by combining the actual scores of each indicator. A scoring threshold, such as 0.9 points, is set; scores above this value automatically trigger an opening command, while scores below this value result in refusal to open the door and a record of the reason. By introducing the Q-Learning reinforcement learning algorithm, the judgment matrix is ​​dynamically adjusted based on user feedback, enabling continuous optimization of the decision-making model and improving the accuracy of the opening strategy. The monitoring dashboard, based on Grafana, features a multi-dimensional visualization dashboard that displays key indicators such as the number of people passing through, gesture recognition accuracy, and disinfection pass rate in real time using bar charts, pie charts, and heat maps.It supports drill-down and slicing by time, location, and personnel, and provides custom query and alarm threshold settings, such as "more than 30 people passing through per minute for 5 consecutive minutes." Once the threshold is triggered, it automatically pushes email and SMS notifications, and also displays a pop-up notification on the integrated control system homepage, facilitating timely handling by on-duty personnel. The gesture recognition module generates 50 JSON-formatted recognition results per second, and the disinfection monitoring module generates 10 execution records per disinfection step. This data is transmitted to the integrated control system in real time via a Kafka cluster, with an average latency controlled within 50 milliseconds. The Redis cluster is configured with 100 shards, each storing the gesture recognition results and disinfection records for the most recent 5 seconds, achieving an average cache hit rate of 95%. A MongoDB distributed database cluster deployment scheme is adopted, configured with 4 shards to balance the load, enhancing data processing capabilities and scalability. Each shard cluster contains one primary node responsible for handling write operations, and two secondary nodes for data redundancy and read expansion, ensuring data consistency and high service availability. In terms of index design, composite indexes were specifically created for the three key fields: person_id, timestamp, and action, optimizing query performance. Thanks to this architecture, the sensor-based sliding door automatic control system can stably support high-speed data writing requirements of inserting 5000 records per second, while ensuring query efficiency, enabling 99% of query requests to respond within 100 milliseconds, greatly improving user experience and overall system processing capabilities. The integrated control system of the sensor-based sliding door automatic control system comprises 10 microservices, using Kafka for asynchronous communication. Each microservice deploys 3 Pod instances, and uses an IstioIngress gateway for Layer 7 load balancing, completing 95% of service requests within 200 milliseconds. The gesture recognition result judgment rule base includes 100 rules, each rule containing an average of 3 conditions and 1 action, deployed using Drools containerization, with an average rule matching time of 30 milliseconds. The disinfection state machine contains 4 states and 12 transition conditions, described using the SCXML standard, with an average state transition response time of 50 milliseconds. During the AHP comprehensive evaluation process, the geometric mean of the scores from 10 experts was calculated to obtain the criterion layer index weight vector as [0.2, 0.3, 0.5] and the scheme layer weight vector as [0.6, 0.4]. The weighted comprehensive score was 0.92, which is greater than the preset threshold of 0.85, thus automatically executing the door-opening command. By adding state rewards and discount factors, after 2000 rounds of training iterations, the accuracy of the Q-Learning model on the test set increased from 75% to 95%.Grafana is a powerful and flexible dashboard system that can simultaneously connect to more than ten different types of data sources, covering mainstream technologies such as Prometheus monitoring systems, InfluxDB time-series databases, and Elasticsearch search engines. This high degree of heterogeneous data source compatibility provides us with a broad foundation for data analysis. Grafana can complete data loading in less than one second on average, ensuring users receive an instant and smooth experience during monitoring and analysis, greatly improving work efficiency and response speed. Utilizing Grafana's sub-dashboard and variable binding functions, we can achieve multi-indicator linked display and drill-down analysis, and use a webhook mechanism to connect with SMS gateways and email servers to achieve automatic batch sending of alert information, with an average delivery delay of less than one minute.

[0034] Step S106: Real-time monitoring of earthquake and fire monitoring data within the area. Analyze early warning signals for earthquake or fire emergencies, detect abnormal changes, including rapid changes in earthquake vibration data and fire smoke concentration, and predict the occurrence of emergencies. If the integrated control system detects earthquake or fire signals, determine whether it is necessary to open the doors immediately for evacuation.

[0035] Earthquake and fire monitoring data are acquired using seismic acceleration sensors and smoke detectors; the seismic and fire monitoring data are preprocessed; based on the preprocessed seismic and fire monitoring data, it is determined whether an earthquake or fire has occurred, including determining the scope of the earthquake or fire event's impact by combining the spatial distribution of seismic acceleration sensors and smoke detectors; based on the determination of the scope of the earthquake or fire event's impact, the population density in a specific area, and evacuation routes, an evacuation plan is generated, which includes the evacuation order of each area and the selection of safe exits; after generating the evacuation plan, voice broadcasts or displays in the area are triggered to guide the orderly evacuation of personnel; the evacuation plan is correlated with gesture recognition and disinfection record data for analysis to determine whether the emergency opening conditions are met, and if the conditions are met, an opening command is issued.

[0036] For example, 50 seismic acceleration sensors and 100 smart smoke detectors are deployed in the area. Through Zigbee networking, the seismic and fire monitoring data are uploaded to the integrated control system in real time at 200ms intervals. The integrated control system uses the time-series database OpenTSDB to store the seismic and fire monitoring data and uses the Kafka streaming platform to perform real-time cleaning and normalization of the data, keeping the data processing latency within 500ms. Time-series prediction algorithms such as Facebook Prophet and LSTM are used to predict the trends of key parameters such as seismic acceleration and smoke concentration within the next 30 seconds, with the mean absolute percentage error (MAPE) of the prediction model controlled within 20%. By setting seismic acceleration thresholds, such as 0.2g and smoke concentration thresholds, such as 0.5mg / m³, it is determined whether an earthquake or fire has occurred. Simultaneously, combined with the spatial distribution of the sensors, clustering algorithms such as DBSCAN are used to determine the scope of the event's impact. The system comprehensively considers seismic intensity, fire level, population density in the area, and evacuation... Based on factors such as evacuation routes, and using an ant colony optimization algorithm, the optimal evacuation plan is calculated, including the evacuation order of each area and the selection of safe exits. The decision results are then sent to the control terminal in JSON format. Upon receiving the evacuation decision, the integrated control system first triggers voice broadcasts and LED displays within the area to guide orderly evacuation. Simultaneously, it correlates the evacuation decision with data such as gesture recognition and disinfection records to determine if emergency opening conditions are met. If the conditions are met, an opening command is issued. The integrated control system continuously monitors the progress of evacuation and analyzes changes in population density within a specific area using infrared thermal imaging cameras. When the population density drops below 0.1 people / square meter, the exit gates are automatically closed, and data on evacuation and incident handling are uploaded to the city's emergency management platform to provide data support for post-event assessment. Fifty seismic acceleration sensors and 100 smart smoke detectors are deployed within the area, using a LoRaWAN star network topology. Seismic and fire monitoring data are uploaded to edge computing nodes in real-time at 5-second intervals via LoRa gateways, with a transmission distance of up to 5 kilometers. Edge nodes use the OpenTSDB time-series database to store monitoring data, employing LSM tree indexes and columnar storage to optimize write performance. Data streaming cleaning and transformation are achieved through KafkaConnect, with data processing latency controlled to within 1 second. Kalman filtering is used to remove noise and estimate the state of seismic acceleration data. Combined with the ARIMA model and outlier detection, rolling predictions of acceleration trends over the next 30 seconds are made, with prediction errors controlled to within 0.1g.The DBSCAN clustering algorithm was used to spatially aggregate magnitude data points, using magnitude and smoke concentration as distance metrics. The clustering granularity was controlled by adjusting the hypersphere radius, and the AlphaShape algorithm was used to extract the boundary polygons of the clusters. These were then overlaid with a regional map to determine the impact range of the disaster. The evacuation decision-making problem within the region was modeled as a multi-objective optimization, considering factors such as evacuation path length, exit capacity, and population density distribution. An improved ant colony algorithm was used for the solution. Pheromones concentration and weight coefficients were set for evacuation paths, and state transition rules guided ants to search for the optimal path to avoid congestion. After 1000 iterations, the Pareto optimal solution set was obtained. Nine key evaluation indicators, including event risk level, evacuation time, and evacuation pressure, were designed. A weighted summation model was constructed to comprehensively evaluate emergency evacuation and generate an evacuation decision matrix. Based on 500 sets of historical emergency drill data, the CART decision tree algorithm was used for training, and the model parameters were optimized through grid search to generate door opening decision rules. Evacuation decisions are pushed to the fire alarm control panel in JSON format via the MQTT protocol, activating emergency broadcasts and evacuation indicator lights to guide personnel to safety. Infrared thermal imaging cameras capture a 30fps video stream of personnel density, and OpenCV is used for background modeling and foreground segmentation to extract parameters such as pixel percentage and density of the activity area. A first-order exponential smoothing model dynamically predicts the evacuation curve; when the personnel density in the area falls below 0.05 people / square meter for 30 consecutive seconds, fire doors automatically close and an emergency termination signal is triggered. Evacuation process data is packaged via the CAP protocol and aggregated into the city's emergency management platform as MQTT messages to generate an evacuation effectiveness assessment report. This report is also linked to GIS maps and facial recognition records to supplement clues about the emergency response process and provide data support for post-event review. Within a 1000 square meter area, 25 triaxial seismic accelerometers and 50 photoelectric smoke detectors are deployed. The sensors have a sampling frequency of 100Hz and are connected to a star network via LoRa wireless communication modules, achieving a data transmission rate of 5Kbps, a gateway receiver sensitivity of -148dBm, and a transmission distance of up to 3 kilometers. The gateway uploads data in JSON format to the edge server via the MQTT protocol. The server uses a 4-core CPU and 16GB of memory, running a time-series database based on OpenTSDB. Each data point contains fields such as device ID, timestamp, triaxial acceleration components, and smoke concentration. The database can write 10,000 data points per second. KafkaConnect is used to clean the raw data, filtering out invalid data points and normalizing the acceleration data to between 0 and 1. The data processing throughput reaches 5000 data points per second. Linear regression of the acceleration data is performed using the least squares method, constructing a third-order Kalman filter. By dynamically adjusting the state transition matrix Q and the observation noise matrix R, it adaptively tracks acceleration changes, achieving a prediction accuracy better than 90%.The DBSCAN algorithm was used to cluster earthquake source locations, with a radius parameter ε = 50 meters and a minimum number of points MinPts = 5. The clustering results were mapped to building floor plans using geocoding. The Douglas-Peucker algorithm was then used to thin the cluster boundaries, generating polygons representing the disaster impact area. A directed graph of evacuation paths was constructed based on graph theory. Evacuation time, path distance, and exit width were used as optimization objectives. An improved ant colony algorithm was employed for multi-objective solution, with 100 ants, an initial pheromone concentration τ0 = 0.1, a pheromone decay coefficient ρ = 0.7, and a fall probability Q0 = 0.9. The Pareto optimal evacuation path set was obtained after 500 iterations. An evacuation decision-making index system was constructed using the Analytic Hierarchy Process (AHP), including 12 indicators such as earthquake intensity, fire level, evacuation time, and evacuation conflict. Ten experts were invited to score each indicator pairwise, and the judgment matrix CR = 0.058 < 0.1 was calculated. After consistency testing, the final indicator weight vector ω = [0.20, 0.32, 0.15, ...]. An open-door decision model was constructed using a CART classification decision tree, trained on 500 sets of historical data, and the Gini coefficient was used as a purity measure. Optimization was performed using 10-fold cross-validation, achieving an accuracy of 95.6% on the test set. A 30m × 30m area thermal radiation imager was used to acquire human thermal radiation images, with a resolution of 640 × 480, a frame rate of 25fps, and a temperature resolution of 0.1℃. Image preprocessing was performed using OpenCV. Gaussian mixture modeling was used to extract foreground moving targets, with a learning rate α = 0.005 and a background ratio T = 0.7. Connectivity analysis was then used to count the number of people and calculate the population density in the area. A density threshold of 0.1 people / square meter was set, and a door-closing decision was triggered if the density remained below the threshold for 30 seconds. The population density data was predicted using a single exponential smoothing algorithm with a smoothing coefficient α = 0.8 and a prediction error MAPE = 8.6%. All event data and decision commands were encapsulated according to the CAP protocol and published to the emergency command and dispatch platform at MQTT QoS1 level, with an average transmission latency of 1.5 seconds, for real-time dispatch by command personnel.

[0037] Step S107: When the automatic control system of the sliding door detects an emergency, it responds quickly, switches to emergency mode, cancels the gesture recognition requirement, ensures that personnel can quickly enter or leave the area, and issues an alarm to notify relevant personnel to take emergency measures.

[0038] When the sensor-controlled sliding door system detects an emergency, it dynamically adjusts the priority of personnel passage in evacuation routes. When the congestion of the evacuation route exceeds a preset threshold, it switches evacuation paths, guiding personnel to the next best route to maximize evacuation efficiency. Emergency plans are generated using predefined rule templates and broadcast via an RPC interface to the intelligent voice system, while the plan information is displayed in real-time on the emergency command screen. Multimodal vital sign data are collected during evacuation to identify high-risk individuals, and rescue and protection procedures are initiated for them to ensure the safety of vulnerable groups. After evacuation, an event analysis report is generated, and an emergency response report is generated after a comprehensive assessment of the event's cause, process, and impact, and sent to relevant personnel.

[0039] For example, when the sensor-controlled sliding door automatic control system detects an emergency, it triggers an emergency mode switch. In emergency mode, the gesture recognition threshold for personnel passage is automatically lowered from 99% to 75%. Through optimization strategies such as cache lookup and threshold filtering, the gesture recognition response time is shortened from 1 second to less than 200 milliseconds. The priority of personnel passage in evacuation routes is dynamically increased in emergency situations. When the congestion of the evacuation route exceeds a preset threshold, the evacuation path is automatically switched, guiding personnel to the secondary route to ensure maximum evacuation efficiency. Using the real-time stream computing engine Flink, complex event detection in emergency situations is completed within 200 milliseconds. Emergency plans are automatically generated through predefined rule templates and the intelligent voice system is called via RPC interface. Using speech synthesis technology, the speech is played at a rate of 150 words per minute. The system reports emergency measures and displays the plan information in real time on the emergency command screen. It collects multimodal vital sign data during the evacuation process using a combination of facial recognition and infrared thermal imaging. This data is then aggregated and analyzed using the InfluxDB time-series database to automatically identify high-risk individuals and initiate specialized rescue and protection procedures to ensure the safety of vulnerable groups. After evacuation, an event analysis report is generated. Machine learning algorithms comprehensively evaluate the cause, process, and impact of the event, automatically generating an emergency response report in a combined text and image format, which is sent to relevant personnel via email, SMS, and other channels. The system calculates the weights of each model using the Analytic Hierarchy Process (AHP). When the weighted fusion probability of an emergency exceeds 0.95, the sensor-based sliding door automatic control system enters emergency mode. In emergency mode, the gesture recognition system in the sensor-based sliding door automatic control system automatically simplifies the gesture recognition process, recognizing only key gestures such as "waving" and "pointing." It also employs more reliable biometric authentication methods such as facial recognition and fingerprint recognition to ensure rapid passage through evacuation routes. Utilizing an improved ant colony algorithm, an objective function is constructed based on factors such as path length, channel width, and congestion index. With minimizing evacuation time as the optimization goal, personnel are dynamically guided to the optimal evacuation path. A sensor-based automatic sliding door control system invokes an intelligent voice broadcast service, using a speech rate of 200 words per minute to broadcast evacuation guidance information in a short, repetitive manner. This is combined with multiple channels, including audible and visual alarms and electronic displays, to disseminate information. Infrared thermal imaging collects personnel's body surface temperature data, and smart bracelets collect heart rate data, which is uploaded to the time-series database InfluxDB. Threshold-based rules and machine learning algorithms are used to analyze vital sign data in real time. When body temperature exceeds 39℃ or heart rate exceeds 120 beats per minute, it is identified as abnormal, and a special rescue process is initiated for high-risk individuals such as the elderly and children.During the evacuation process, the sensor-controlled sliding door system uses facial recognition and infrared thermal imaging technology to monitor the number of people in the area in real time. When no new people are detected entering for five consecutive minutes, the evacuation is considered complete, and the system automatically switches back to normal mode. When the sensor-controlled sliding door system detects an emergency, it switches to emergency mode. Advanced single-classification technologies such as One-Class SVM are used to perform precise anomaly detection on complex gesture data. This process also incorporates facial feature point matching technology and fingerprint feature hash mapping to construct a multi-layered, cascaded biometric authentication model. This model exhibits extremely high security, with a false rejection rate (FRR) as low as 0.1% and a false acceptance rate (FAR) controlled within 0.01%. To further improve the efficiency and experience of safe evacuation, an improved ant colony algorithm (ACA) is used, with evacuation time and path comfort as dual optimization objectives, to establish an adaptive evacuation path planning model. This model, through fine-tuning key parameters such as population size (100), maximum iterations (500), pheromone importance factor α=1, expected importance factor β=5, and pheromone evaporation coefficient ρ=0.5, combined with the automatic tuning capabilities of the particle swarm optimization algorithm, significantly reduced evacuation time from 3 minutes to less than 1 minute. In health monitoring, OpenCV was used to process infrared thermal imaging video, parsing temperature information of human body regions, including average temperature and temperature distribution histograms. Connectivity analysis was used to accurately count the number of people on site, and heart rate data collected by smart bracelets was integrated. A vital sign anomaly detection model developed using a one-dimensional convolutional neural network was applied to a 60-second continuous physiological indicator sequence using a sliding window analysis, making predictions every 30 seconds. The prediction accuracy of the vital sign anomaly detection model exceeded 85%. All collected vital sign data were efficiently stored in the InfluxDB time series database, employing an hourly data sharding strategy and a 3x compression ratio to ensure efficient data management. Simultaneously, Grafana was used to achieve intuitive visualization of vital sign trends, facilitating real-time monitoring and analysis. The sensor-based automatic sliding door control system also deploys the Prometheus monitoring platform. Utilizing time series prediction models such as LSTM and combined with the dynamic baseline method, it immediately triggers an alarm mechanism when data deviates from the normal range by more than three standard deviations (3σ), effectively providing early warning of potential problems. Furthermore, by integrating the Jaeger microservice tracing tool to trace microservice call chains, it automatically delimits the root causes of anomalies such as latency, slow database queries, and cache penetration in the sensor-based automatic sliding door control system. This improves the fault response speed and resolution efficiency of the system, comprehensively ensuring its stable operation and user safety.

[0040] Step S108: After the emergency ends, collect and analyze all data from the emergency response process, including personnel flow, system response speed, and accuracy of gate control; based on the analysis results, optimize the sensor-based automatic sliding door control system.

[0041] Based on the end time of the emergency, determine the data collection time range for the emergency response process; analyze the collected personnel flow data, including modeling and predictive analysis of the response speed data of the sensor-based sliding door automatic control system; collect the analysis results of the door control accuracy data and comprehensively evaluate the overall effectiveness of the emergency response process; based on the evaluation results of the overall effectiveness of the emergency response process, optimize the sensor control parameters to improve the sensitivity and stability of the sensor-based sliding door automatic control system; and deploy the optimized sensors in the sensor-based sliding door automatic control system.

[0042] For example, the data collection time range for the emergency response process is determined based on the end time of the emergency. The data collection interval (once per minute) and data quality requirements, such as ensuring data integrity and accuracy, are also considered. If the emergency lasted 30 minutes, the data collection time range is from 10 minutes before the event to 20 minutes after the event, totaling 60 minutes. The collected personnel flow data is then analyzed. Within the 30 minutes of the emergency, 500 people passed through the gate, including 450 evacuees and 50 rescue personnel. Cluster analysis is applied to identify high-risk areas and personnel flow patterns, providing a basis for optimizing evacuation routes. Machine learning algorithms are used to model and predict the response speed data of the sensor-based sliding door automatic control system. Analysis of historical data shows that when the response time exceeds 5 seconds, the failure rate of the sensor-based sliding door automatic control system increases to over 10%. Image recognition processing is performed on the gate control accuracy data. Analysis of 1000 gate opening / closing images shows a recognition accuracy rate of 98%. The results of all data analyses are aggregated to comprehensively evaluate the overall effectiveness of the emergency response process. A weighted scoring model was established, such as: Overall Score = Average Response Time * Weight of Average Response Time + Gate Control Accuracy * Weight of Gate Control Accuracy. Different weights were assigned based on the impact of different indicators on the overall effect. With a personnel evacuation efficiency of 90%, an average response time of 3 seconds for the sensor-based sliding door automatic control system, and a gate control accuracy of 98%, the overall score was 95 out of 100. Based on the evaluation results, the sensor control parameters were optimized to improve the sensitivity and stability of the sensor-based sliding door automatic control system. After adjusting the sensor parameters, verification tests were conducted in a simulated environment to ensure that the new parameters, while improving response speed, would not cause false alarms or missed alarms. The trigger distance of the infrared sensor was adjusted from 50 cm to 30 cm, improving the response speed of the sensor-based sliding door automatic control system. The optimized sensor control parameters were deployed in the sensor-based sliding door automatic control system. The control parameters of 100 sliding doors were updated to be consistent with the optimized parameters. The operational data of the optimized sensor-based sliding door automatic control system was obtained, and an operational monitoring report was generated for subsequent optimization. Data from a week of operation showed that the optimized sensor-based automatic sliding door control system reduced the average response time by 1 second and lowered the failure rate.

[0043] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A sensor-based automatic control method for sliding doors, characterized in that, The method includes: The frequency, duration of stay, and number of people entering and leaving a specific area at different time periods are obtained to establish a flow model to reflect the flow of people. Record the frequency of disinfection in the area and outside the region, the types and concentrations of disinfectants used, analyze the effective time and disinfection effect of disinfectants under different conditions, and ensure the continuous cleanliness of specific areas; Record the disinfection steps, time, and disinfectant usage before personnel enter a specific area, monitor the implementation of disinfection steps, ensure the effectiveness of disinfection operations, and obtain data on the maintenance of a clean environment within the environment; Collect gesture data from different individuals, analyze the accuracy, recognition speed, and error rate of the gesture data, and obtain the operational status data of gesture recognition, including: acquiring hand movement sequence images and thermal imaging data of individuals entering a specific area, and transmitting the hand movement sequence images and thermal imaging data to the server for processing in real time; Key features of hand motion sequence images are extracted and classified, gesture trajectories are modeled, and different gesture types are identified. By calculating the speed and angle changes during the gesture movement, it is determined whether an abnormal gesture event has been triggered; If a gesture abnormality event is triggered, the gesture recognition is paused and the authenticity of the gesture abnormality event is verified. If true, determine whether it was an intrusion by an unauthorized person; The gesture recognition data and disinfection step data are input into the integrated control system. The completion rate of the disinfection steps and the correctness of the gesture recognition are analyzed to determine whether to open the area entrance. This includes: the gesture recognition data and disinfection step execution data are transmitted to the integrated control system in real time. This includes separating the gesture recognition results and the disinfection step completion status into independent services and communicating with the integrated control system through an RPC interface. The judgment of the gesture recognition result is based on the characteristics of gesture category, confidence level and historical recognition accuracy to determine whether the current gesture recognition is correct; To determine the completion status of the disinfection steps, a finite state machine model is used. Based on parameters such as disinfectant spraying time, hand coverage area, and number of gesture changes, the completeness of the disinfection steps is determined. The results of the gesture recognition and the judgment of the disinfection steps are comprehensively evaluated to calculate a comprehensive score; Real-time monitoring of earthquake and fire data within the area; analysis of early warning signals for earthquake or fire emergencies; detection of abnormal changes, including rapid changes in earthquake vibration data and fire smoke concentration; prediction of emergency situations; and determination of whether it is necessary to open the gates for evacuation if the integrated control system detects earthquake or fire signals. When the sensor-controlled sliding door automatic control system detects an emergency, it will respond quickly, switch to emergency mode, cancel the requirement for gesture recognition, ensure that personnel can quickly enter or leave the area, and at the same time issue an alarm to notify relevant personnel to take emergency measures, including: when the sensor-controlled sliding door automatic control system detects an emergency, it will dynamically adjust the personnel passage priority of the evacuation route in the emergency. When the congestion of the evacuation route exceeds the preset threshold, the evacuation route is switched to guide people to the second-best route to ensure maximizing evacuation efficiency. Emergency plans are generated using predefined rule templates and broadcast through an intelligent voice system via an RPC interface. At the same time, the plan information is displayed in real time on the emergency command screen. Collect multimodal vital signs data during the evacuation process, identify high-risk groups among the evacuees, initiate rescue and protection procedures for high-risk groups, and ensure the safety of vulnerable groups; After the emergency ends, collect and analyze all data from the emergency response process, including personnel flow, system response speed, and the accuracy of gate control. Based on the analysis results, optimize the sensor-based automatic sliding door control system.

2. The method according to claim 1, wherein, The process of obtaining the frequency, duration of stay, and number of relevant personnel entering and exiting a specific area at different time periods, and establishing a flow model to reflect personnel flow, includes: Based on the unique identification information of the personnel, record the time when the personnel enter and exit a specific area; By analyzing changes in the location coordinates of personnel, the direction of their movement and their trajectory within a specific area can be determined. Based on the frequency of people in different time periods, the distribution of people flow frequency in different time periods is obtained; The duration of each person's stay is obtained by calculating the changes in their location within the specific area. Based on the frequency and number of people moving within different time periods, a statistical report reflecting the overall trend of people movement is generated. A flow model is built based on historical population flow data to obtain predictions of future population flow. By combining the predicted population movement results with regional maps, a heat map showing the movement trajectory and residence of people in different areas is obtained.

3. The method according to claim 1, wherein, The recording includes the frequency of disinfection of the internal and external environments, the types and concentrations of disinfectants used, and the analysis of the effective time and disinfection effect of disinfectants under different conditions to ensure the continuous cleanliness of specific areas, including: Develop corresponding disinfection frequency plans based on the usage and contamination levels of specific areas; The effectiveness of different types and concentrations of disinfectants was evaluated, and the kill rate of bacteria and viruses under different conditions was tested. The concentration and spraying time of the disinfectant are adjusted according to its chemical composition and environmental conditions. Assess the cleanliness of a specific area to obtain a predicted potential risk of infection; By combining factors such as bacterial type, virus type, and environmental humidity, the effective time of different disinfection schemes under different conditions is calculated to obtain an optimized disinfection process; Based on the environmental layout and pollution hotspots of a specific area, disinfection robots are used to clean disinfection dead spots and high-risk areas to improve disinfection coverage and efficiency.

4. The method according to claim 1, wherein, The record document details the disinfection steps, timing, and disinfectant usage before personnel enter a specific area, monitors the execution of disinfection steps to ensure the effectiveness of disinfection operations, and obtains maintenance data on the cleanliness of the environment, including: The entry time and sequence of each person are recorded by RFID readers installed in the disinfection channel; Standardize the disinfection process and guide personnel to complete the disinfection according to the steps and duration of the standardized disinfection process. At the same time, monitor the disinfection process in real time to determine whether the disinfection operation is in place. Air samples were collected before and after personnel entered the area to identify the types and concentrations of bacteria in the air samples and to evaluate the effectiveness of the disinfection procedure. Regularly sample and test the environment in specific areas to analyze the types and quantities of pathogenic microorganisms in the environment, and evaluate the cleanliness of the specific area and the long-term effectiveness of disinfection measures.

5. The method according to claim 1, wherein, The process involves collecting gesture data from different individuals, analyzing the accuracy, recognition speed, and error rate of the gesture data, and obtaining operational status data for gesture recognition, including: To address the different gesture habits of various individuals, a personalized gesture template library is established for template matching, enabling adaptive adjustments in gesture recognition.

6. The method according to claim 1, wherein, The real-time monitoring area collects earthquake and fire monitoring data, analyzes early warning signals for earthquake or fire emergencies, detects abnormal changes, including rapid changes in earthquake vibration data and fire smoke concentration, predicts the occurrence of emergencies, and, if the integrated control system detects earthquake or fire signals, determines whether it is necessary to open the main gate for emergency evacuation, including: Earthquake and fire monitoring data are obtained through earthquake acceleration sensors and smoke detectors; The earthquake and fire monitoring data are preprocessed. Based on the preprocessed earthquake and fire monitoring data, determine whether an earthquake or fire has occurred, including combining the spatial distribution of earthquake acceleration sensors and smoke detectors to determine the scope of the impact of earthquake or fire events; Based on the assessment of the impact range of an earthquake or fire event, the population density and evacuation routes in a specific area, an evacuation plan is generated. The evacuation plan includes the evacuation order and the selection of safe exits for each area. After generating the evacuation plan, trigger voice broadcasts or display screens within the area to guide the orderly evacuation of personnel; The evacuation plan is correlated with gesture recognition and disinfection record data to determine whether the emergency opening conditions are met. If the conditions are met, an opening command is issued.

7. The method according to claim 1, wherein, When the sensor-controlled sliding door automatic system detects an emergency, it responds quickly, switches to emergency mode, disables gesture recognition, ensures personnel can quickly enter or leave the area, and simultaneously issues an alarm to notify relevant personnel to take emergency measures, including: After the evacuation is completed, an incident analysis report is generated. At the same time, an emergency response report is generated after a comprehensive assessment of the cause, process and impact of the incident, and then sent to the relevant person in charge.

8. The method according to claim 1, wherein, After the emergency ends, all data from the emergency response process is collected and analyzed, including personnel flow, system response speed, and the accuracy of door control. Based on the analysis results, the sensor-based automatic sliding door control system is optimized, including: Determine the time range for collecting emergency response process data based on the end time of the emergency situation; The collected personnel flow data is analyzed, including modeling and predictive analysis of the response speed data of the sensor-based automatic control system for sliding doors; The analysis results of gate control accuracy data are used to comprehensively evaluate the overall effectiveness of the emergency response process. Based on the evaluation results of the overall effectiveness of the emergency response process, the sensor control parameters are optimized to improve the sensitivity and stability of the sensor-based automatic control system for sliding doors. In a sensor-based automatic control system for sliding doors, optimized sensors are deployed.

Images