A deep intelligent perception monitoring method for conference rooms

Abstract

This invention discloses a deep intelligent perception monitoring method for conference rooms. To overcome the problems of existing monitoring technologies, which are relatively simple, have simple monitoring indicators, lack versatility, and are not highly intelligent, the invention includes the following steps: S1: constructing a deep multidimensional Taylor network perception model whose input is sensor data and whose output is monitoring indicators; S2: real-time acquisition of sensor data and data preprocessing using empirical mode decomposition; S3: monitoring the conference room's daily activities, personnel health, and major events using the deep multidimensional Taylor network perception model; S4: determining whether monitoring anomalies occur through comparison, and issuing an alarm if so. Data fusion technology and deep learning are used to comprehensively determine anomalies, effectively improving the intelligence, security, and reliability of the monitoring process.

Description

Technical Field

[0001] The present invention relates to the field of conference room monitoring, and in particular to a deep intelligent perception monitoring method for a conference room. Background Art

[0002] In recent years, with the development of technologies such as artificial intelligence and data fusion, the demand for intelligent conference room monitoring has become increasingly demanding. Researching an intelligent monitoring system for conference rooms is of both theoretical and practical significance. Existing monitoring systems are mostly based on image processing techniques. These systems first acquire and preprocess images, then perform feature extraction to analyze anomalies and monitor abnormalities. These systems employ relatively simple monitoring techniques and indicators, lacking versatility and a low level of intelligence. Furthermore, monitoring accuracy relies heavily on image processing, resulting in poor robustness when modeling complex scenarios. Therefore, developing an accurate and versatile intelligent conference room monitoring system is crucial.

[0003] For example, a "Conference Room Monitoring System, Sensor Device, and Conference Room Monitoring Method" disclosed in Chinese patent literature, with publication number CN107525592B, includes a server, a wireless gateway, and a first sensor device. The first sensor device is connected to the wireless gateway and is configured to transmit information to the wireless gateway regarding whether the conference room in which it is installed is currently occupied. The wireless gateway is connected to the server and is configured to transmit information regarding whether the conference room is currently occupied to the server. The server is connected to the wireless gateway and is configured to receive and monitor whether the conference room is currently in use based on the information transmitted by the wireless gateway. However, this solution utilizes a relatively simple monitoring technology and simple monitoring indicators, lacks versatility, and is not highly intelligent. Summary of the Invention

[0004] The present invention mainly solves the problems that existing monitoring technology is relatively single, monitoring indicators are relatively simple, lack of versatility, and low degree of intelligence; it provides a deep intelligent perception monitoring method for conference rooms, which uses data fusion technology and deep learning to make comprehensive judgments on abnormal situations, effectively improving the intelligence, safety and reliability of the monitoring process.

[0005] The above technical problems of the present invention are mainly solved by the following technical solutions:

[0006] A method for deep intelligent perception monitoring of a conference room includes the following steps:

[0007] S1: Construct a deep multi-dimensional Taylor network perception model whose input is the sensor data and output is the monitoring indicators;

[0008] S2: Collect sensor data in real time and use empirical mode decomposition method to preprocess the data;

[0009] S3: Monitor the daily life, personnel health and major events of the conference room through a deep multi-dimensional Taylor network perception model;

[0010] S4: Determine whether there is any abnormality in the monitoring by comparison, and issue an alarm if there is any abnormality in the monitoring.

[0011] This solution integrates data processing, data fusion, and artificial intelligence technologies to create a holistic design for daily conference room monitoring, personnel health monitoring, and major event monitoring. It utilizes a deep, multidimensional Taylor network perception model to effectively improve monitoring accuracy and complete overall monitoring. This invention addresses the issues of versatility and accuracy in monitoring models by developing a universal method and application for deep intelligent perception models. Leveraging artificial intelligence, data fusion, and other technologies, this solution provides a novel method and application implementation for deep intelligent perception models, resulting in a wide monitoring range, comprehensive monitoring indicators, and a high level of intelligence.

[0012] Preferably, the sensor comprises:

[0013] Millimeter-wave radar monitors people's heartbeat and breathing;

[0014] Surveillance cameras collect daily information about the conference room;

[0015] Smoke sensor, collects smoke concentration in the conference room;

[0016] Temperature sensor, collecting the temperature of the conference room;

[0017] Humidity sensor, collects humidity in the conference room;

[0018] Oxygen concentration sensor, collects oxygen concentration in the conference room.

[0019] The deep intelligent perception model serves as a daily monitoring model for conference rooms, a personnel health monitoring model, and a major event monitoring model. Through a multi-sensor data fusion strategy, it collects data from millimeter-wave radars, surveillance cameras, temperature sensors, humidity sensors, and oxygen concentration sensors as model inputs, thereby more accurately obtaining abnormal monitoring information and improving monitoring accuracy.

[0020] Preferably, the deep multi-dimensional Taylor network perception model is:

[0021]

[0022] in, for Metafunction Expand into The total number of product terms of the sub-approximating polynomials;

[0023] For the The weight coefficient of the product term,

[0024] For the variables in the product terms The number of times, and d is the number of layers.

[0025] First, a multidimensional Taylor network model is introduced, which is a three-layer network with a simple structure and strong approximation performance. Then, the multidimensional Taylor network is improved by adding multiple intermediate layers to obtain a deep multidimensional Taylor network model. Finally, the wake-sleep algorithm is used to learn and optimize the network parameters.

[0026] Preferably, the deep multi-dimensional Taylor network perception model is trained using a wake-sleep algorithm; the wake-sleep algorithm includes two steps: pre-training and fine-tuning.

[0027] Each layer of the multidimensional Taylor network perception model is trained independently and unsupervised to ensure that the feature vectors are mapped to different feature spaces while preserving as much feature information as possible. The final layer of the multidimensional Taylor network perception model is fully connected, receiving the output feature vectors of the previous layer as its input feature vectors for supervised training of the entity relationship classifier. Each layer of the multidimensional Taylor network perception model only ensures that the weights within that layer are optimally mapped to the feature vectors of that layer, not the entire multidimensional Taylor network perception model. Therefore, a backpropagation network also propagates error information from top to bottom to each layer of the multidimensional Taylor network perception model, fine-tuning the entire deep multidimensional Taylor network perception model.

[0028] Preferably, the empirical mode decomposition method includes:

[0029] The original sensor data x(t) is decomposed and the decomposition result is:

[0030]

[0031] Among them, IMF i (t) is the i-th order eigenmode function component;

[0032] r n (t) is the residual component;

[0033] n is the order of the IMF;

[0034] Data denoising: remove some high-frequency components from the decomposed data to obtain the processed components;

[0035] Data reconstruction is to reorganize the data after removing some high-frequency components to obtain the data processed by the empirical mode decomposition method.

[0036] Previous data processing methods, such as short-time Fourier transforms and wavelet analysis, fall into the category of global analysis. This is because they rely on the selection of basis functions, making the processing results difficult to guarantee. This solution, however, uses Empirical Mode Decomposition (EMD) to preprocess the data. EMD is a data-driven, adaptive, nonlinear, time-varying signal decomposition method that decomposes the data, removes some high-frequency components, and then reconstructs the remaining components to improve data accuracy.

[0037] Preferably, the intrinsic mode function is obtained by the following process:

[0038] 1) Calculate the mean envelope m1(t) of the original data

[0039]

[0040] Among them, e + (t) is the maximum envelope;

[0041] e - (t) is the minimum envelope;

[0042] Get new data with low frequency removed

[0043]

[0044] Repeat the above process until If the definition of the eigenmode function is satisfied, then

[0045]

[0046] 2) Calculate and obtain new data r1(t)

[0047] r1(t)=x(t)-imf1(t)

[0048] 3) Repeat the above process until the nth order intrinsic mode function component or residual component r n (t) is less than the preset value or when the residual component r n When (t) is a monotonic function or a constant, the empirical mode decomposition stops and the corresponding intrinsic mode function components are obtained.

[0049] Maximum envelope e + (t) and the minimum envelope e -(t) is obtained by fitting all the maximum and minimum points of the original data x(t) through the cubic spline function.

[0050] The beneficial effects of the present invention are:

[0051] The deep multi-dimensional Taylor network perception model uses a multi-sensor data fusion strategy to collect millimeter-wave radar, surveillance cameras, temperature sensors, humidity sensors, and oxygen concentration sensors as model inputs, and more accurately obtains daily conference room monitoring information, personnel health monitoring information, and major event monitoring information, thereby improving monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a method for deep intelligent perception monitoring of a conference room according to the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention will be further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0054] Example:

[0055] A method for monitoring a conference room using deep intelligent perception is provided in this embodiment. Figure 1 As shown, the following steps are included:

[0056] S1: Construct a deep multi-dimensional Taylor network perception model whose input is the sensor data and output is the monitoring indicators.

[0057] Drawing on deep learning and artificial intelligence theories, we design a universal deep intelligent perception model. This model boasts high modeling accuracy, a simple structure, self-learning capabilities, strong versatility, and ease of implementation. First, we introduce a multidimensional Taylor network model, a three-layer network with a simple structure and strong approximation performance. We then improve the multidimensional Taylor network by adding multiple intermediate layers to create a deep multidimensional Taylor network model. Finally, we utilize a wake-sleep algorithm to learn and optimize the network parameters.

[0058] Sensors include:

[0059] Millimeter-wave radar monitors people's heartbeat and breathing;

[0060] Surveillance cameras collect daily information about the conference room;

[0061] Smoke sensor, collects smoke concentration in the conference room;

[0062] Temperature sensor, collects the temperature of the conference room to ensure the comfort of the personnel;

[0063] Humidity sensor, collects humidity in the conference room to ensure personnel comfort;

[0064] Oxygen concentration sensor collects oxygen concentration in the conference room to meet the comfort level of personnel and ensure the health and safety of participants.

[0065] The deep multi-dimensional Taylor network perception model is:

[0066]

[0067] in, for Metafunction Expand into The total number of product terms of the sub-approximating polynomials;

[0068] For the The weight coefficient of the product term,

[0069] For the variables in the product terms The number of times, and d is the number of layers.

[0070] The deep multi-dimensional Taylor network perception model is trained using a wake-sleep algorithm; the wake-sleep algorithm includes two steps: pre-training and fine-tuning.

[0071] Each layer of the multidimensional Taylor network perception model is trained independently and unsupervised to ensure that the feature vectors are mapped to different feature spaces while preserving as much feature information as possible. The final layer of the multidimensional Taylor network perception model is fully connected, receiving the output feature vectors of the previous layer as its input feature vectors for supervised training of the entity relationship classifier. Each layer of the multidimensional Taylor network perception model only ensures that the weights within that layer are optimally mapped to the feature vectors of that layer, not the entire multidimensional Taylor network perception model. Therefore, a backpropagation network also propagates error information from top to bottom to each layer of the multidimensional Taylor network perception model, fine-tuning the entire deep multidimensional Taylor network perception model.

[0072] The deep multi-dimensional Taylor network perception model is used to construct the daily monitoring model of the conference room, the health monitoring model of the participants, and the major event monitoring model.

[0073] S2: Collect sensor data in real time and use empirical mode decomposition method to preprocess the data.

[0074] Previous data processing methods, such as short-time Fourier transform and wavelet analysis, fall into the category of global analysis. The reason for this is that they all rely on the selection of basis functions, making it difficult to guarantee the processing results. This embodiment, however, utilizes Empirical Mode Decomposition (EMD) to preprocess the data. EMD is a data-driven, adaptive, nonlinear, time-varying signal decomposition method that decomposes the data, removes some high-frequency components, and then reconstructs the remaining components.

[0075] The empirical mode decomposition method specifically includes:

[0076] The original sensor data x(t) is decomposed and the decomposition result is:

[0077]

[0078] Among them, IMF i (t) is the i-th order eigenmode function component

[0079] r n (t) is the residual component;

[0080] n is the order of the IMF.

[0081] The intrinsic mode functions are obtained through the following process:

[0082] 1) Calculate the mean envelope m1(t) of the original data

[0083]

[0084] Among them, e + (t) is the maximum envelope;

[0085] e - (t) is the minimum envelope.

[0086] Maximum envelope e + (t) and the minimum envelope e - (t) is obtained by fitting all the maximum and minimum points of the original data x(t) through the cubic spline function.

[0087] Get new data with low frequency removed

[0088]

[0089] generally, It is not a stationary data and does not meet the two conditions of the IMF. Repeat the above process until If the definition of the eigenmode function is satisfied, then

[0090]

[0091] 2) Calculate and obtain new data r1(t)

[0092] r1(t)=x(t)-imf1(t)

[0093] 3) Repeat the above process until the nth order intrinsic mode function component or residual component r n (t) is less than the preset value or when the residual component r n When (t) is a monotonic function or a constant, the empirical mode decomposition stops and the corresponding intrinsic mode function components are obtained.

[0094] Data denoising is done by analyzing high-frequency IMF components, low-frequency IMF components and residuals. At the same time, since noise energy is mostly concentrated in high frequencies, after repeated experiments and verification, some high-frequency components are removed from the decomposed data to obtain the processed components.

[0095] Data reconstruction, the data after removing some high-frequency components imf i ′(t) is reorganized to obtain the data processed by the empirical mode decomposition method.

[0096] S3: Monitor the daily life, personnel health and major events of the conference room through the deep multi-dimensional Taylor network perception model.

[0097] As a comprehensive monitoring model, the deep multi-dimensional Taylor network perception model collects data from various sensors as input. By comprehensively utilizing multi-sensor fusion strategies, it can more accurately obtain abnormal monitoring information and improve monitoring accuracy.

[0098] As a health monitoring model, the deep multi-dimensional Taylor network perception model collects millimeter-wave radar, surveillance camera, temperature sensor, humidity sensor, and oxygen concentration sensor as the model input through a multi-sensor data fusion strategy, and obtains personnel health abnormality monitoring information more accurately, thereby improving monitoring accuracy.

[0099] The deep multi-dimensional Taylor network perception model, used as a major event monitoring model, collects data from millimeter-wave radar and surveillance cameras as input. By comprehensively utilizing multi-sensor fusion strategies, it accurately derives information on abnormal human behavior, enabling timely handling of abnormal individuals and ensuring the safety of attendees.

[0100] S4: Determine whether there is any abnormality in the monitoring by comparison, and issue an alarm if there is any abnormality in the monitoring.

[0101] By judging the behavior set and comparing it with the normal behavior set, it is determined whether the monitoring is normal.

[0102] This solution leverages a deep multi-dimensional Taylor network perception model and a multi-sensor data fusion strategy, collecting data from millimeter-wave radar, surveillance cameras, temperature sensors, humidity sensors, and oxygen concentration sensors as model inputs. This allows for more accurate monitoring of daily meeting room activities, personnel health, and major events, improving monitoring accuracy. This solution offers a wide monitoring range, comprehensive monitoring indicators, and a high level of intelligence.

[0103] It should be understood that the embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope limited by the appended claims of the application.

Claims

1. A deep intelligent perception monitoring method for a conference room, characterized in that: The following steps are involved: S1: Construct a deep multidimensional Taylor network perception model with sensor data as input and monitoring indicators as output; the deep multidimensional Taylor network perception model includes multiple intermediate layers, each of which is a deep multidimensional Taylor network perception model network; each layer of the deep multidimensional Taylor network perception model network is trained independently and unsupervised; the last layer of the deep multidimensional Taylor network perception model adopts a fully connected network, receiving the output feature vector of the previous layer network as its input feature vector, and supervised training of the entity relationship classifier. Each layer of the deep multidimensional Taylor network perception model network ensures that the weights within its own layer are optimally mapped to the feature vector of that layer; the back propagation network propagates the error information from top to bottom to each layer of the deep multidimensional Taylor network perception model, and fine-tunes the entire deep multidimensional Taylor network perception model network; S2: Collect sensor data in real time and use empirical mode decomposition method to preprocess the data; S3: A deep multi-dimensional Taylor network perception model is used to monitor daily activities, personnel health, and major events in the conference room. Millimeter-wave radar, surveillance cameras, temperature sensors, humidity sensors, and oxygen concentration sensors are used as inputs for personnel health monitoring; millimeter-wave radar and surveillance cameras are used as inputs for major event monitoring. S4: Determine whether there is any abnormality in the monitoring by comparison, and issue an alarm if there is any abnormality in the monitoring.

2. The method for deep intelligent perception monitoring of a conference room according to claim 1, characterized in that: The sensor comprises: Millimeter-wave radar monitors people's heartbeat and breathing; Surveillance cameras collect daily information about the conference room and personnel behavior; Smoke sensor, collects smoke concentration in the conference room; Temperature sensor, collecting the temperature of the conference room; Humidity sensor, collects humidity in the conference room; Oxygen concentration sensor, collects oxygen concentration in the conference room.

3. The method for deep intelligent perception monitoring of a conference room according to claim 1, characterized in that: The deep multi-dimensional Taylor network perception model is: in, for Metafunction Expand into The total number of product terms of the sub-approximating polynomials; For the The weight coefficient of the product term, For the variables in the product terms The number of times, and d is the number of layers.

4. A method for deep intelligent perception monitoring of a conference room according to claim 1 or 3, characterized in that: The deep multi-dimensional Taylor network perception model is trained using a wake-sleep algorithm; the wake-sleep algorithm includes two steps: pre-training and fine-tuning.

5. The method for deep intelligent perception monitoring of a conference room according to claim 1, characterized in that: The empirical mode decomposition method includes: The original sensor data x(t) is decomposed and the decomposition result is: Among them, IMF i (t) is the i-th order eigenmode function component; r n (t) is the residual component; n is the order of the IMF; Data denoising: remove some high-frequency components from the decomposed data to obtain the processed components; Data reconstruction is to reorganize the data after removing some high-frequency components to obtain the data processed by the empirical mode decomposition method.

6. The method for deep intelligent perception monitoring of a conference room according to claim 5, characterized in that: The intrinsic mode function is obtained by the following process: 1) Calculate the mean envelope m1(t) of the original data Among them, e + (t) is the maximum envelope; e - (t) is the minimum envelope; Get new data with low frequency removed Repeat the above process until If the definition of the eigenmode function is satisfied, then 2) Calculate and obtain new data r1(t) r1(t)=x(t)-imf1(t) 3) Repeat the above process until the nth order intrinsic mode function component or residual component r n (t) is less than the preset value or when the residual component r n When (t) is a monotonic function or a constant, the empirical mode decomposition stops and the corresponding intrinsic mode function components are obtained.

Images