An intelligent detection method for falling behavior in dangerous areas of coal mines

Abstract

The present invention provides an intelligent detection system and method for fall behavior in dangerous areas of coal mines. The system: a router is installed on the inspection robot, and the inspection robot runs in the working space of the mine; the router is used to send WiFi signals to smart phones; the smart phones are configured with Nexmon firmware, which is worn on the body of the mine personnel and is used to obtain CSI data from the WiFi signal, and at the same time, it is used to obtain the acceleration data of the mine personnel; the processing terminal is located in the monitoring center, and a behavior acquisition module, a state judgment module, a preprocessing module, a behavior segmentation module and a behavior detection module are arranged inside. Method: Acquire CSI data and acceleration data; judge the motion state of the receiving end and select features; perform data preprocessing; segment the data of suspected fall behavior from the continuous behavior; perform feature extraction, and judge the fall behavior. The system and method can accurately and efficiently detect fall behavior in dangerous areas.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent detection, and in particular relates to an intelligent detection system and method for falling behavior in dangerous areas of coal mines. Background Art

[0002] In the daily production process of coal mines, there are usually great safety hazards. How to effectively ensure the safe production of coal mines is an issue that requires long-term attention. In the actual production process, the behavior of miners is directly related to some accidents in coal mines.

[0003] At present, wireless communication technology has been widely used in the field of miner behavior perception. At present, the technologies for the perception of unsafe behaviors of miners mainly include three types: wearable device-based, vision-based, and wireless signal-based. The perception of unsafe behaviors based on wearable devices mainly relies on the sensors carried by the subjects, relying on the sensors to collect relevant information in the environment, and then realize the perception of the scene and personnel status. The perception of unsafe behaviors based on vision mainly uses cameras or video cameras to collect data in the environment, and further analyzes the acquired images or videos to realize the perception of personnel behavior. The perception of unsafe behaviors based on wireless signals mainly utilizes the characteristics that wireless signals will change when propagating in the environment. By extracting features and performing relevant analysis on the collected signals, the characteristics of the signal propagation space are obtained from them to realize the perception of specific scenes or specific targets. For the current perception technology for unsafe behaviors of miners, the behavior perception method based on wearable devices has a high implementation cost. At the same time, its comfort is poor, which affects the experience of the subjects, and the miners wearing equipment unrelated to work while working may affect related operations. Limited by the influence of the complex environment of the mine, the behavior perception method based on vision is difficult to match the environmental characteristics of low illumination and high dust in the mine. Behavior perception based on wireless signals mostly relies on network card devices, and the amount of signal data collected is limited. At the same time, the network card must also be equipped with a computer for sending and receiving, which makes it difficult for the sending and receiving end to move flexibly and difficult to deploy in a mine environment. Based on the shortcomings of the existing technology, it is urgent to provide a new intelligent detection system and method for fall behavior in dangerous areas of mines. Summary of the invention

[0004] In view of some problems existing in the prior art, the present invention provides a system and method for intelligent detection of falls in dangerous areas of coal mines. The system has a simple structure and low investment cost, and can accurately and efficiently detect falls in dangerous areas, thereby facilitating timely and effective response measures, greatly improving the safety factor of mine personnel operations. The method has simple implementation steps, low implementation cost, high detection efficiency, accurate detection results, and is suitable for large-scale promotion and application.

[0005] In order to achieve the above-mentioned object, the present invention provides an intelligent detection system for fall behavior in dangerous areas of coal mines, including an inspection robot, a smart phone and a processing terminal;

[0006] The inspection robot is equipped with a router, and the number of the inspection robots is one or more, and the inspection robots are operated in the working space of the mine; the router is used as a transmitting end to send a WiFi signal to the smart phone;

[0007] The smartphone is configured with Nexmon firmware, which is worn by the mine personnel and is used to receive WiFi signals and obtain CSI data from the WiFi signals. At the same time, it is used to obtain the acceleration data of the mine personnel and to send the obtained CSI data and acceleration data to the processing terminal in a wireless manner;

[0008] The processing terminal is located in the monitoring center, and is internally provided with a behavior collection module, a state judgment module, a preprocessing module, a behavior segmentation module and a behavior detection module; wherein the behavior collection module is used to collect CSI data and acceleration data; the state judgment module is used to judge the motion state of the receiving end based on the CSI data and the acceleration data, and is used to select suitable features according to the motion state; the preprocessing module is used to perform data preprocessing on the selected features; the behavior segmentation module is used to extract signals of suspected falling behaviors from the CSI data, and is used to obtain key behaviors in continuous behaviors; the behavior detection module is used to extract features of the acquired key behaviors, and is used to judge the behavior results based on the machine learning model.

[0009] In the present invention, a router is installed on the inspection robot, which can facilitate the use of the autonomous mobile inspection capability of the inspection robot to transmit WiFi signals to the mine personnel at different positions in the working area, and can effectively ensure that the smart phones worn by the mine personnel in different areas can all receive the WiFi signal. By configuring the Nexmon firmware in the smart phone, it is convenient to accurately and reliably obtain CSI data from the WiFi signal. By enabling the smart phone and the processing terminal to communicate wirelessly, it is convenient for the processing terminal to obtain the CSI data obtained by the smart phone and the acceleration data of the mine personnel in a timely manner, and it can be free from the restrictions of harsh environmental conditions; by providing a state judgment module in the processing terminal, it is convenient to accurately judge the motion state of the mine personnel based on the CSI data and the acceleration data, and then select suitable features according to different motion states, which is conducive to obtaining more accurate detection results. By providing a preprocessing module in the processing terminal, it is convenient to preprocess the data of the selected features, further improving the accuracy of the detection results. By setting up a behavior segmentation module in the processing terminal, it is easy to extract signals of suspected fall behaviors from the CSI data, and then obtain key behaviors from continuous behaviors, which effectively reduces the amount of calculation and further ensures the detection accuracy. By setting up a detection module in the processing terminal, it is easy to extract features of key behaviors, which is conducive to obtaining accurate detection results through machine learning models. The system has a simple structure and low investment cost. It can accurately and efficiently detect fall behaviors in dangerous areas, which is conducive to taking effective countermeasures in a timely manner, greatly improving the safety factor of mine personnel operations.

[0010] Furthermore, in order to ensure that the CSI data can be accurately and reliably acquired, the model of the smart phone is Nexus 6P.

[0011] Furthermore, in order to ensure powerful processing capability, the processing terminal is an industrial computer.

[0012] Furthermore, in order to ensure that smartphones at different positions in the working space can communicate with the processing terminal in real time, several wireless relay devices are installed at intervals in the working space of the mine, and the smartphones are connected to the processing terminal through the wireless relay devices.

[0013] The present invention also provides a method for intelligently detecting falling behavior in a dangerous area of ​​a coal mine, using an intelligent detection system for falling behavior in a dangerous area of ​​a coal mine, comprising the following steps:

[0014] Step 1: Sense the behavior of personnel in dangerous areas underground and obtain CSI data and acceleration data;

[0015] S11: Use the router installed on the inspection robot as the transmitter to send out WiFi signals; use the smartphone configured with Nexmon firmware as the receiver to receive WiFi signals and obtain CSI data from the WiFi signals; at the same time, use the built-in acceleration sensor in the smartphone to obtain the acceleration data of the mine personnel;

[0016] S12: The smartphone sends the CSI data and the acceleration data to the behavior collection module in the processing terminal by wireless communication. After receiving the CSI data and the acceleration data, the behavior collection module sends them to the state judgment module.

[0017] Step 2: Using the state judgment module to judge the motion state of the receiving end based on the acceleration data extracted in step 1, and selecting features according to the judgment result;

[0018] S21: Calculate the acceleration autocorrelation function ρ according to formula (1) A (τ,f);

[0019]

[0020] Where τ is the time lag of the autocorrelation function, f is the acceleration frequency, and δ(τ) represents the Dirac function. is the variance of A(t,f), σ 2 (f) is the measurement noise, ρ s (τ,f) is the error caused by human motion;

[0021] S22: Based on the autocorrelation function ρ of the three-axis acceleration data A (τ,f), judge the motion state of the receiving end; when lim τ→0 ρ A When (τ,f)>0, the receiving end is judged to be in motion. τ→0 ρ A When (τ,f) = 0, the receiving end is judged to be in a stationary state;

[0022] S23: using different features to detect behaviors according to the state judgment result; when the receiving end is moving, acceleration and amplitude are selected as features, and when the receiving end is stationary, amplitude and phase difference are selected as features;

[0023] Step 3: Preprocess the features selected in step 2 through the preprocessing module;

[0024] S31: performing denoising processing on the selected characteristic signal;

[0025] S32: Fuse the CSI data and acceleration data; obtain the timestamps of the two data, with the CSI timestamp sequence T1 = {t1, t2, ..., tm} is used as the reference, and the timestamp sequence T2 of the acceleration data is T2 = {t′1, t′2, …, t′ n}, first calculate the Euclidean distance d of the two time series according to formula (2), then find the timestamp sequence T′2 of the acceleration data that makes d the minimum value, then select the corresponding acceleration data according to the timestamp sequence T′2, and realize splicing and fusion with the CSI data;

[0026]

[0027] S33: after performing unwrapping and linear fitting processing on the phases obtained by the two antennas at the receiving end of the smartphone, a phase difference is calculated;

[0028] Step 4: Use the behavior segmentation module to segment the preprocessed data into key behaviors and obtain data on suspected falling behaviors in continuous behaviors;

[0029] S41: The collected CSI data is an M×N column matrix, where M represents the length of the collected signal data and N represents the number of subcarriers; a sliding window with a length of L and a step size of d is used to traverse the CSI matrix, and the variance var of each sliding window is calculated i ,

[0030] S42: Calculate the sliding window center index i according to formula (3);

[0031]

[0032] Where n represents the sliding window number, n = 1, 2, ..., M-L+1;

[0033] S43: define a threshold value threshold by formula (4), and save the data and index higher than the threshold value threshold as key behaviors, so as to segment the key behaviors suspected of falling from the CSI matrix;

[0034]

[0035] Step 5: Use the behavior detection module to extract features of the key behaviors obtained in step 4, and send the extracted features to the machine learning model for classification. The classification results are used to determine whether there is a fall behavior in the dangerous area of ​​the coal mine.

[0036] In the present invention, a router installed on the inspection robot is used as a transmitter. Since the inspection robot has the ability to move autonomously in the working space of the mine, the sensing range is significantly expanded. By using a smartphone equipped with Nexmon firmware to collect CSI data, 256 subcarriers can be collected at a bandwidth of 80MHz, and rich channel information can be obtained. Through this information, the signal scattering, reflection and power attenuation phenomena that occur with the change of the transmission distance of the carrier can be effectively analyzed, and the changes in the physical environment that cause the channel state to change can be inferred, that is, a non-contact intelligent perception process is realized. Since CSI is extremely sensitive to environmental changes, it can capture the subtle movements of people in a static environment, which is conducive to accurately detecting whether the mine personnel wearing smartphones have fallen. In addition, since smartphones are easy to carry, there is no need for mine personnel to wear additional equipment, which not only has a low implementation cost, but also does not affect the experience of the person being detected, and does not affect the normal operation of the mine personnel. Furthermore, smartphones are equipped with built-in accelerometers, so smartphones can obtain CSI data on the one hand, and the wearer's acceleration data on the other hand, which can provide reliable data support for behavior detection and ensure that the detection method can cope with the perception scenarios of movement and stillness at the receiving end. Using the variance of the sliding window as the threshold, it is possible to accurately segment the fall-like behavior from the continuous behavior, which can not only significantly reduce the amount of calculation in the subsequent detection process, but also help to obtain more accurate detection results.

[0037] This method perceives the behavior of mine personnel based on wireless signals. It is not restricted by harsh working environments and has good applicability. At the same time, it only needs CSI data and acceleration data to accurately judge whether there is a fall behavior. It has simple implementation steps, low implementation cost, small amount of calculation, high detection efficiency, accurate detection results, and is suitable for large-scale promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of the present invention;

[0039] Figure 2 It is a topological structure diagram of the present invention.

[0040] In the picture: 1. Mine; 2. Inspection robot; 3. Smartphone. DETAILED DESCRIPTION

[0041] The present invention is further described with reference to the accompanying structural drawings below.

[0042] like Figure 2 As shown, the present invention provides an intelligent detection system for falling behavior in dangerous areas of coal mines, including an inspection robot 2, a smart phone 3 and a processing terminal;

[0043] The inspection robot 2 is equipped with a router. There are one or more inspection robots 2 running in the working space of the mine 1. The router is used as a transmitting end to send a WiFi signal to the smart phone 3.

[0044] The smart phone 3 is configured with Nexmon firmware, which is worn by the mine personnel and is used to receive WiFi signals and obtain CSI data from the WiFi signals. At the same time, it is used to obtain the acceleration data of the mine personnel and to send the obtained CSI data and acceleration data to the processing terminal in a wireless manner;

[0045] The processing terminal is located in the monitoring center, and is internally provided with a behavior collection module, a state judgment module, a preprocessing module, a behavior segmentation module and a behavior detection module; wherein the behavior collection module is used to collect CSI data and acceleration data; the state judgment module is used to judge the motion state of the receiving end based on the CSI data and the acceleration data, and is used to select suitable features according to the motion state; the preprocessing module is used to perform data preprocessing on the selected features; the behavior segmentation module is used to extract signals of suspected falling behaviors from the CSI data, and is used to obtain key behaviors in continuous behaviors; the behavior detection module is used to extract features of the acquired key behaviors, and is used to judge the behavior results based on the machine learning model.

[0046] In order to ensure that the CSI data can be accurately and reliably obtained, the model of the smartphone is Nexus 6P.

[0047] In order to ensure powerful processing capability, the processing terminal is an industrial computer.

[0048] In order to ensure that smartphones at different positions in the working space can communicate with the processing terminal in real time, several wireless relay devices are arranged at intervals in the working space of the mine 1, and the smartphone 3 is connected to the processing terminal via the wireless relay devices.

[0049] In the present invention, a router is installed on the inspection robot, which can facilitate the use of the autonomous mobile inspection capability of the inspection robot to transmit WiFi signals to the mine personnel at different positions in the working area, and can effectively ensure that the smart phones worn by the mine personnel in different areas can all receive the WiFi signal. By configuring the Nexmon firmware in the smart phone, it is convenient to accurately and reliably obtain CSI data from the WiFi signal. By enabling the smart phone and the processing terminal to communicate wirelessly, it is convenient for the processing terminal to obtain the CSI data obtained by the smart phone and the acceleration data of the mine personnel in a timely manner, and it can be free from the restrictions of harsh environmental conditions; by providing a state judgment module in the processing terminal, it is convenient to accurately judge the motion state of the mine personnel based on the CSI data and the acceleration data, and then select suitable features according to different motion states, which is conducive to obtaining more accurate detection results. By providing a preprocessing module in the processing terminal, it is convenient to preprocess the data of the selected features, further improving the accuracy of the detection results. By setting up a behavior segmentation module in the processing terminal, it is easy to extract signals of suspected fall behaviors from the CSI data, and then obtain key behaviors from continuous behaviors, which effectively reduces the amount of calculation and further ensures the detection accuracy. By setting up a detection module in the processing terminal, it is easy to extract features of key behaviors, which is conducive to obtaining accurate detection results through machine learning models. The system has a simple structure and low investment cost. It can accurately and efficiently detect fall behaviors in dangerous areas, which is conducive to taking effective countermeasures in a timely manner, greatly improving the safety factor of mine personnel operations.

[0050] like Figure 1 As shown, the present invention also provides a method for intelligently detecting falls in dangerous areas of coal mines, using an intelligent detection system for falls in dangerous areas of coal mines, comprising the following steps:

[0051] Step 1: Sense the behavior of personnel in dangerous areas underground and obtain CSI data and acceleration data;

[0052] S11: Using the router installed on the inspection robot 2 as a transmitter to send out a WiFi signal; using the smartphone 3 configured with Nexmon firmware as a receiver to receive the WiFi signal and obtain CSI data from the WiFi signal; at the same time, using the built-in acceleration sensor in the smartphone 3 to obtain the acceleration data of the mine personnel;

[0053] S12: The smart phone 3 sends the CSI data and the acceleration data to the behavior collection module in the processing terminal by wireless communication. After receiving the CSI data and the acceleration data, the behavior collection module sends them to the state judgment module.

[0054] Step 2: Using the state judgment module to judge the motion state of the receiving end based on the acceleration data extracted in step 1, and selecting features according to the judgment result;

[0055] S21: Calculate the acceleration autocorrelation function ρ according to formula (1) A (τ,f);

[0056]

[0057] Where τ is the time lag of the autocorrelation function, f is the acceleration frequency, and δ(τ) represents the Dirac function. is the variance of A(t,f), σ 2 (f) is the measurement noise, ρ s (τ,f) is the error caused by human motion;

[0058] S22: When the receiving end is in motion, lim τ→0 ρ s (τ,f)=1,lim τ→0 ρ A (τ,f)>0; when the receiving end is in a stationary state, lim τ→0 ρ A (τ,f)=0;

[0059] Calculate the autocorrelation function ρ for the three-axis acceleration data A (τ,f), judge the motion state of the receiving end; when lim τ→0 ρ A When (τ,f)>0, the receiving end is judged to be in motion. τ→0 ρ A When (τ,f) = 0, the receiving end is judged to be in a stationary state;

[0060] S23: using different features to detect behaviors according to the state judgment result; when the receiving end is moving, acceleration and amplitude are selected as features, and when the receiving end is stationary, amplitude and phase difference are selected as features;

[0061] Step 3: Preprocess the features selected in step 2 through the preprocessing module;

[0062] S31: performing denoising processing on the selected characteristic signal;

[0063] S32: Fuse the CSI data and acceleration data; obtain the timestamps of the two data, with the CSI timestamp sequence T1 = {t A ,t2,…,t m} is used as the reference, and the timestamp sequence T2 of the acceleration data is T2 = {t′1, t′2, …, t′ n}, first calculate the Euclidean distance d of the two time series according to formula (2), then find the timestamp sequence T′2 of the acceleration data that makes d the minimum value, then select the corresponding acceleration data according to the timestamp sequence T′2, and realize splicing and fusion with the CSI data;

[0064]

[0065] S33: after performing unwrapping and linear fitting processing on the phases obtained by the two antennas at the receiving end of the smartphone, a phase difference is calculated;

[0066] Step 4: Use the behavior segmentation module to segment the preprocessed data into key behaviors and obtain data on suspected falling behaviors in continuous behaviors;

[0067] S41: The collected CSI data is an M×N column matrix, where M represents the length of the collected signal data and N represents the number of subcarriers; a sliding window with a length of L and a step size of d is used to traverse the CSI matrix, and the variance var of each sliding window is calculated i ,

[0068] S42: Calculate the sliding window center index i according to formula (3);

[0069]

[0070] Where n represents the sliding window number, n = 1, 2, ..., M-L+1;

[0071] S43: define a threshold value threshold by formula (4), and save the data and index higher than the threshold value threshold as key behaviors, so as to segment the key behaviors suspected of falling from the CSI matrix;

[0072]

[0073] Step 5: Use the behavior detection module to extract features of the key behaviors obtained in step 4, and send the extracted features to the machine learning model for classification. The classification results are used to determine whether there is a fall behavior in the dangerous area of ​​the coal mine.

[0074] In the present invention, a router installed on the inspection robot is used as a transmitter. Since the inspection robot has the ability to move autonomously in the working space of the mine, the sensing range is significantly expanded. By using a smartphone equipped with Nexmon firmware to collect CSI data, 256 subcarriers can be collected at a bandwidth of 80MHz, and rich channel information can be obtained. Through this information, the signal scattering, reflection and power attenuation phenomena that occur with the change of the transmission distance of the carrier can be effectively analyzed, and the changes in the physical environment that cause the channel state to change can be inferred, that is, a non-contact intelligent perception process is realized. Since CSI is extremely sensitive to environmental changes, it can capture the subtle movements of people in a static environment, which is conducive to accurately detecting whether the mine personnel wearing smartphones have fallen. In addition, since smartphones are easy to carry, there is no need for mine personnel to wear additional equipment, which not only has a low implementation cost, but also does not affect the experience of the person being detected, and does not affect the normal operation of the mine personnel. Furthermore, smartphones are equipped with built-in accelerometers, so smartphones can obtain CSI data on the one hand, and the wearer's acceleration data on the other hand, which can provide reliable data support for behavior detection and ensure that the detection method can cope with the perception scenarios of movement and stillness at the receiving end. Using the variance of the sliding window as the threshold, it is possible to accurately segment the fall-like behavior from the continuous behavior, which can not only significantly reduce the amount of calculation in the subsequent detection process, but also help to obtain more accurate detection results.

[0075] This method perceives the behavior of mine personnel based on wireless signals. It is not restricted by harsh working environments and has good applicability. At the same time, it only needs CSI data and acceleration data to accurately judge whether there is a fall behavior. It has simple implementation steps, low implementation cost, small amount of calculation, high detection efficiency, accurate detection results, and is suitable for large-scale promotion and application.

Claims

1. A method for intelligently detecting falls in dangerous areas of coal mines, using an intelligent detection system for falls in dangerous areas of coal mines, the intelligent detection system for falls in dangerous areas of coal mines comprising a patrol robot (2), a smart phone (3) and a processing terminal; the patrol robot (2) is equipped with a router, the number of the patrol robots (2) is one or more, and the patrol robots (2) are operated in the working space of a mine (1); the router is used as a transmitter to send a WiFi signal to the smart phone (3); the smart phone (3) is configured with Nexmon firmware, which is worn by a mine worker, is used to receive the WiFi signal, and obtain CSI data from the WiFi signal, and is also used to obtain acceleration data of the mine worker, and is used to send the obtained CSI data and acceleration data to the processing terminal in a wireless manner; the processing terminal is located in a monitoring center, and is internally provided with a behavior acquisition module, a state judgment module, a preprocessing module, a behavior segmentation module and a behavior detection module; wherein, The behavior collection module is used to collect CSI data and acceleration data; the state judgment module is used to judge the motion state of the receiving end based on the CSI data and acceleration data, and to select suitable features according to the motion state; The preprocessing module is used to perform data preprocessing on the selected features; the behavior segmentation module is used to extract the signal of suspected falling behavior from the CSI data, and to obtain the key behavior in the continuous behavior; The behavior detection module is used to extract features of the acquired key behaviors and to judge the behavior results based on the machine learning model; It is characterized by comprising the following steps: Step 1: Sense the behavior of personnel in dangerous areas underground and obtain CSI data and acceleration data; S11: using a router installed on the inspection robot (2) as a transmitter to send out a WiFi signal; using a smartphone (3) configured with Nexmon firmware as a receiver to receive the WiFi signal and obtain CSI data from the WiFi signal; at the same time, using an acceleration sensor built into the smartphone (3) to obtain acceleration data of the mine personnel; S12: The smart phone (3) sends the CSI data and the acceleration data to the behavior collection module in the processing terminal by wireless communication. After receiving the CSI data and the acceleration data, the behavior collection module sends them to the state judgment module. Step 2: Using the state judgment module to judge the motion state of the receiving end based on the acceleration data extracted in step 1, and selecting features according to the judgment result; S21: Calculate the acceleration autocorrelation function according to formula (1) ; （1）； In the formula, is the time lag of the autocorrelation function, is the acceleration frequency, represents the Dirac function, for The variance of To measure noise, It is the error caused by human body movement; S22: Autocorrelation function based on triaxial acceleration data , determine the motion state of the receiving end; when When , it is judged that the receiving end is in motion. When , it is judged that the receiving end is in a stationary state; S23: using different features to detect behaviors according to the state judgment result; when the receiving end is moving, acceleration and amplitude are selected as features, and when the receiving end is stationary, amplitude and phase difference are selected as features; Step 3: Preprocess the features selected in step 2 through the preprocessing module; S31: performing denoising processing on the selected characteristic signal; S32: Fuse the CSI data and acceleration data; obtain the timestamps of the two data, and use the CSI timestamp sequence As a benchmark, the timestamp sequence of acceleration data In the sampling, first calculate the Euclidean distance of the two time series according to formula (2): d , and then look for the envoy d Get the timestamp sequence of the acceleration data with the minimum value , then, according to the timestamp sequence Select the corresponding acceleration data and merge it with the CSI data; （2）； S33: after performing unwrapping and linear fitting processing on the phases obtained by the two antennas at the receiving end of the smartphone, a phase difference is calculated; Step 4: Use the behavior segmentation module to segment the preprocessed data into key behaviors and obtain data on suspected falling behaviors in continuous behaviors; S41: The collected CSI data is M×N A column matrix, where M Indicates the length of the collected signal data. N Indicates the number of subcarriers; the length used is L , the step length is d The sliding window traverses the CSI matrix and calculates the variance of each sliding window ; S42: Calculate the sliding window center index according to formula (3) i ; （3）； In the formula, n represents the sliding window number, n=1,2,…,M-L+1 ; S43: Define the threshold value by formula (4) threshold , and will be above the threshold threshold The data and index are saved as key behaviors to segment the key behaviors of suspected falls from the CSI matrix; （4）； Step 5: Use the behavior detection module to extract features of the key behaviors obtained in step 4, and send the extracted features to the machine learning model for classification. The classification results are used to determine whether there is a fall behavior in the dangerous area of ​​the coal mine.

2. According to claim 1, a method for intelligent detection of fall behavior in dangerous areas of coal mines is characterized in that: The model of the smartphone is Nexus 6P.

3. A method for intelligent detection of falls in dangerous areas of coal mines according to claim 1 or 2, characterized in that: The processing terminal is an industrial computer.

4. A method for intelligent detection of falls in dangerous areas of coal mines according to claim 1 or 2, characterized in that: A plurality of wireless relay devices are arranged at intervals in the working space of the mine (1), and the smart phone (3) is connected to the processing terminal via the wireless relay devices.

Images