Safety service danger signal recognition system and method

Abstract

The present invention discloses a safety clothing danger signal recognition system and method, the system includes a main trigger device worn on a person, and a sound detection device and a danger signal monitoring device installed on the safety clothing, the main trigger device includes a first microprocessor unit, a three-axis accelerometer, a three-axis gyroscope, a pressure sensor and a first Bluetooth transmission module, the three-axis accelerometer, the three-axis gyroscope, the pressure sensor and the first Bluetooth transmission module are all electrically connected to the first microprocessor unit; the sound detection device includes a second microprocessor unit, a dual microphone array, a conversion module and a second Bluetooth transmission module. The present invention records the behavior status information of the operating personnel in dangerous places according to the information collected by the safety clothing worn by the operating personnel in dangerous places, so as to realize the monitoring and protection of the operating personnel, reduce safety hazards, and improve the safety of the personnel themselves.

Description

Technical Field

[0001] The invention relates to a safety clothing danger signal recognition system and method, belonging to the technical field of data monitoring. Background Art

[0002] With the rapid development of economy, injuries have become an important public health problem threatening people's health. According to statistics, there are about 200 million injuries of various types in my country every year, and the death toll reaches more than 700,000, accounting for about 9% of the total number of deaths, and it is the fifth cause of death after nausea, cerebrovascular disease, respiratory disease and heart disease.

[0003] Injuries can be divided into accidental injuries, intentional injuries, violence and homicide according to intent, among which accidental injuries account for the largest proportion. According to estimates by the WHO, with the acceleration of urbanization and population mobility, the number of deaths caused by injuries will increase by 65% ​​worldwide, reaching 8.4 million, and the threat posed by accidental injuries will continue to rise.

[0004] Unintentional injuries caused by the inability to correctly identify danger signals pose a serious threat to human health and also bring a heavy burden to society. How to avoid accidental injuries has become a public health issue. Many times, the injuries are not directly caused by the danger itself, but by the inability to correctly identify danger signals. Therefore, the response to injuries should be based on prevention, identifying danger signals before the danger occurs. In recent years, with the continuous enrichment of multimodal information and the rapid development of deep learning technology, more and more researchers have begun to study danger signal recognition technology, striving to improve the accuracy of danger signal recognition, and reduce harm to people through prediction, reminders, alarms, remediation and other means. This type of technology has good application value and broad market prospects. Summary of the invention

[0005] The purpose of the present invention is to provide a safety clothing danger signal recognition system and method, which can record the behavioral status information of operators in dangerous situations, thereby realizing the monitoring and protection of operators, reducing safety hazards, and improving the safety of the operators themselves.

[0006] The technical solution of the present invention is as follows: a safety clothing danger signal identification system, comprising a main trigger device worn on a person and a sound detection device and a danger signal monitoring device installed on the safety clothing, the main trigger device comprising a first microprocessing unit, a three-axis accelerometer, a three-axis gyroscope, a pressure sensor and a first Bluetooth transmission module, the three-axis accelerometer, the three-axis gyroscope, the pressure sensor and the first Bluetooth transmission module are all electrically connected to the first microprocessing unit; the sound detection device comprises a second microprocessing unit, a dual microphone array, a conversion module and a second Bluetooth transmission module, the dual microphone array, the conversion module and the second Bluetooth transmission module are all electrically connected to the second microprocessing unit; the danger signal detection device comprises a receiving module, a computing module and an analyzing module, the danger signal detection device is wirelessly connected to the first Bluetooth transmission module and the second Bluetooth transmission module through the receiving module.

[0007] Furthermore, the sound detection device is installed on the shoulder of the safety suit, and the danger signal monitoring device is installed on the waist of the safety suit.

[0008] At the same time, the present invention also provides a safety clothing danger signal recognition method, comprising the following steps:

[0009] S1, obtaining trigger information, location information and sound information of personnel;

[0010] S2, design an independent trigger method according to the trigger information: use the main trigger device to collect the acceleration, angular velocity and joint pressure data of the personnel, then use the FIR filter to filter the collected data to obtain an effective measurement signal, use the gradient descent method to fuse the denoised acceleration, angular velocity and joint pressure data to obtain the target function, use this target function as the main decision function, and use this method as the main decision method;

[0011] S3, based on the sound information, a sound recognition decision method is designed using a deep learning method: first, a sound detection device is used to collect the sound information of the person, and then the sound analog signal is converted into a digital signal through a conversion module, and the sound digital signal is subjected to a wavelet threshold denoising method to reduce the noise, so as to obtain an effective sound signal, and the short-time energy, short-time average zero-crossing rate, Mel cepstral coefficient and linear prediction cepstral coefficient of the sound signal are extracted, and finally a decision tree support vector machine is used to perform data analysis on the above sound features to obtain a dangerous human voice signal target function, and this target function is used as an auxiliary decision function, and this method is used as an auxiliary decision method;

[0012] S4, design the final decision method based on the main decision method and the auxiliary decision method: use the linear programming equation to integrate the results of the main decision method and the auxiliary decision method.

[0013] The main decision-making steps mentioned above are:

[0014] Step 1: Use the main trigger device to collect the acceleration a of the person x 、a y 、a z , angular velocity ω x ,ω y ,ω z , and the magnetic strength m x 、m y 、m z ;

[0015] Step 2: Collect the acceleration a x 、a y 、a z , angular velocity w x 、w y 、w z And the magnetic intensity m, use FIR filter to filter and get effective measurement signal;

[0016] Step 3: Establish the navigation coordinate system and the carrier coordinate system, then the rotation matrix is:

[0017]

[0018] Where ψ is the yaw angle, ρ is the pitch angle, and θ is the roll angle. Note:

[0019]

[0020] The three attitude angles are:

[0021]

[0022] Attitude quaternion q = (q0 q1 q2 q3) T and the gyroscope angular velocity ω x ,ω y ,ω z The satisfied relationship is:

[0023]

[0024] The relationship between the attitude quaternion and the attitude matrix can be expressed as:

[0025]

[0026] The relationship between the three-axis accelerometer and the pitch angle and roll angle is:

[0027] [a x a y a z ] T=[g·sinθcosρ -g·sinρ -g·cosθcosρ] T

[0028]

[0029] The relationship between the three-axis magnetometer and the yaw angle is:

[0030]

[0031]

[0032] Step 4: Use the gradient descent method to fuse the denoised acceleration, angular velocity and magnetic intensity data. The iterative formula is:

[0033]

[0034]

[0035] The objective function is obtained as:

[0036]

[0037] This objective function is used as the main decision method.

[0038] The above-mentioned decision-making assistance method steps are:

[0039] Step 1: Use the sound detection device (2) to collect the sound information of the person;

[0040] Step 2: Convert the sound analog signal into a sound digital signal through a conversion module (22);

[0041] Step 3: Use the wavelet threshold denoising method to reduce the noise of the sound digital signal to obtain an effective sound signal;

[0042] Step 4: Extract the short-term energy of the sound signal Short-term average zero-crossing rate Mel cepstral coefficient MFCC and linear prediction cepstral coefficient LPCC;

[0043] Step 5: Use a decision tree support vector machine to perform data analysis on the above sound features to obtain a dangerous human voice signal recognition result δ, where 0 represents a non-dangerous signal and 1 represents a dangerous signal, and use this recognition result δ as an auxiliary decision-making method.

[0044] The above-mentioned final decision method is to use the linear programming equation to integrate the results of the main decision method and the auxiliary decision method; the linear regression equation is: y=w1f(q)+w2δ, where w1 and w2 are weights, which can be adjusted according to the situation, and are set to 10 and 1 respectively here.

[0045] Due to the adoption of the above technical scheme, the advantages of the present invention are: the present invention records the behavioral status information of workers in dangerous places based on the information collected by the safety clothing worn by the workers in dangerous places, thereby realizing the monitoring and protection of the workers, reducing safety hazards, and improving the safety of the workers themselves. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the structure of the present invention;

[0047] Figure 2 is a schematic structural diagram of a main trigger device in an embodiment of the present invention;

[0048] Figure 3 is a schematic structural diagram of a sound detection device in an embodiment of the present invention;

[0049] Figure 4 is a schematic diagram of the structure of a danger signal monitoring device in an embodiment of the present invention;

[0050] Figure 5 It is a schematic diagram of the process of the present invention;

[0051] Figure 6 It is a flow chart of the main decision-making method of the present invention;

[0052] Figure 7 It is a flowchart of the auxiliary decision-making method of the present invention;

[0053] Figure 8 It is a schematic diagram of the flow of the present invention performing data analysis on sound features. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0055] Embodiment of the present invention: The structural diagram of the safety clothing danger signal recognition system is as follows Figures 1 to 4As shown, it includes a main trigger device 1 worn on a person, a sound detection device 2 and a danger signal monitoring device 3 installed on safety clothing, the main trigger device 1 includes a first microprocessor unit 10, a three-axis accelerometer 11, a three-axis gyroscope 12, a pressure sensor 13 and a first Bluetooth transmission module 14, and the three-axis accelerometer 11, the three-axis gyroscope 12, the pressure sensor 13 and the first Bluetooth transmission module 14 are all electrically connected to the first microprocessor unit 10; the sound detection device 2 includes a second microprocessor unit 20, a dual microphone array 21, a conversion module 22 and a second Bluetooth transmission module 23, and the dual microphone array 21, the conversion module 22 and the second Bluetooth transmission module 23 are all electrically connected to the second microprocessor unit 20; the danger signal detection device 3 includes a receiving module 30, a calculation module 31 and an analysis module 32, and the danger signal detection device 3 is wirelessly connected to the first Bluetooth transmission module 14 and the second Bluetooth transmission module 23 through the receiving module 30.

[0056] The functions of the above components are described as follows:

[0057] A main trigger device 1, used to obtain the posture information of the personnel;

[0058] A sound detection device 2, used to obtain the sound information of a person;

[0059] The danger signal monitoring device 3 is used to integrate the posture information and sound information of the personnel, calculate the real-time status of the personnel, and use deep learning and multimodal information fusion to determine the degree of danger of the personnel.

[0060] In this embodiment, the main trigger device 1 is worn by the operator in dangerous places to obtain the operator's posture information such as acceleration, direction and pressure; the sound detection device 2 is installed on the shoulder of the safety suit to obtain and process the operator's sound information. The danger signal monitoring device 3 is installed on the waist of the safety suit to integrate the operator's posture information and sound information, calculate the operator's real-time status, and judge the operator's degree of danger based on this status.

[0061] like Figure 2 As shown, the main trigger device 1 includes:

[0062] A microprocessing unit 10, used for processing the acquired posture information;

[0063] The three-axis accelerometer 11 is used to obtain the acceleration information of the person in the three directions of X, Y and Z. If the acceleration is greater than the set value, the main trigger device 1 will be triggered to enter the excitation state and send out a trigger signal;

[0064] A three-axis gyroscope 12, used to collect angular velocity and direction information of a person;

[0065] Pressure sensor 13, used to detect the pressure state of each joint point;

[0066] The Bluetooth transmission module 14 is used to send acceleration information, angular velocity information and pressure information to the danger signal detection device 3, where these information are collectively referred to as trigger information;

[0067] The pressure status of each joint point includes: neck pressure, shoulder pressure, elbow pressure, hip pressure, knee pressure and foot pressure information.

[0068] The direction information includes: angular velocity change, rotation angle offset and equilibrium position offset.

[0069] like Figure 3 As shown, the sound detection device 2 includes:

[0070] A micro-processing unit 20, used for processing the acquired human voice information;

[0071] Dual microphone array 21, used to obtain human voice information;

[0072] The conversion module 22 is used to convert the analog signal collected by the dual microphone array 21 into a digital signal;

[0073] The Bluetooth transmission module 23 is used to send the human voice information to the danger signal detection device 3.

[0074] In a specific embodiment, the dual microphone array 21 collects the voice of a person, and then transmits the sound signal to the conversion module 22 for analog-to-digital conversion. The microprocessing unit 20 packages the received digital sound signal and sends it to the Bluetooth transmission module 23. The Bluetooth transmission module 23 transmits the sound signal to the danger signal detection device 3.

[0075] like Figure 4 As shown, the danger signal detection device 3 includes:

[0076] A receiving module 30, used for receiving the trigger signal, positioning signal and sound signal;

[0077] A calculation module 31 is used to calculate the trigger information, location information and sound information of the personnel;

[0078] The analysis module 32 is used to fuse the trigger information and sound information of the personnel, analyze and identify the status of the personnel.

[0079] In the specific implementation, the personnel wear safety clothing with a main trigger device 1, a sound detection device 2 and a danger signal detection device 3. The main trigger device 1 collects posture information such as acceleration, direction and pressure, and sends the information to the danger signal detection device 3 through the Bluetooth transmission module 13. The sound detection device 2 collects sound signals, converts the signals into digital signals through the conversion module 22, and then sends the information to the danger signal detection device 3 through the Bluetooth module 23. The danger signal detection device 3 performs data analysis and status judgment on the received information.

[0080] like Figure 5 As shown, the embodiment of the present invention also provides a safety service danger signal recognition method, comprising the following steps:

[0081] S1, obtaining trigger information, location information and sound information of personnel;

[0082] S2, designing an independent triggering method according to the triggering information: using the main triggering device 1 to collect the acceleration, angular velocity and joint pressure data of the personnel, and then filtering the collected data using the FIR filter to obtain an effective measurement signal, using the gradient descent method to fuse the denoised acceleration, angular velocity and joint pressure data to obtain an objective function, using this objective function as the main decision function, and using this method as the main decision method;

[0083] S3, based on the sound information, a sound recognition decision method is designed using a deep learning method: first, the sound detection device 2 is used to collect the sound information of the person, and then the sound analog signal is converted into a digital signal through the conversion module 22, and the sound digital signal is subjected to a wavelet threshold denoising method to reduce the noise, so as to obtain an effective sound signal, and the short-time energy, short-time average zero-crossing rate, Mel cepstral coefficient and linear prediction cepstral coefficient of the sound signal are extracted, and finally a decision tree support vector machine is used to perform data analysis on the above-mentioned sound features to obtain a dangerous human voice signal target function, and this target function is used as an auxiliary decision function, and this method is used as an auxiliary decision method;

[0084] S4, design the final decision method based on the main decision method and the auxiliary decision method: use the linear programming equation to integrate the results of the main decision method and the auxiliary decision method.

[0085] like Figure 6 As shown, the main decision-making steps are:

[0086] Step 1: Use the main trigger device 1 to collect the acceleration a of the person x 、a y 、a z , angular velocity ω x ,ω y ,ω z , and the magnetic strength mx 、m y 、m z ;

[0087] Step 2: Collect the acceleration a x 、a y 、a z , angular velocity w x 、w y 、w z And the magnetic intensity m, use FIR filter to filter and get effective measurement signal;

[0088] Step 3: Establish the navigation coordinate system and the carrier coordinate system, then the rotation matrix is:

[0089]

[0090] Where ψ is the yaw angle, ρ is the pitch angle, and θ is the roll angle. Note:

[0091]

[0092] The three attitude angles are:

[0093]

[0094] Attitude quaternion q = (q0 q1 q2 q3) T and the gyroscope angular velocity ω x ,ω y ,ω z The satisfied relationship is:

[0095]

[0096] The relationship between the attitude quaternion and the attitude matrix can be expressed as:

[0097]

[0098] The relationship between the three-axis accelerometer and the pitch angle and roll angle is:

[0099] [a x a y a z ] T =[g·sinθcosρ -g·sinρ -g·cosθcosρ] T

[0100]

[0101] The relationship between the three-axis magnetometer and the yaw angle is:

[0102]

[0103]

[0104] Step 4: Use the gradient descent method to fuse the denoised acceleration, angular velocity and magnetic intensity data. The iterative formula is:

[0105]

[0106]

[0107] The objective function is obtained as:

[0108]

[0109] This objective function is used as the main decision method.

[0110] like Figure 7 As shown, the steps of the auxiliary decision-making method are:

[0111] Step 1: Use the sound detection device 2 to collect the sound information of the person;

[0112] Step 2: Convert the analog sound signal into a digital sound signal through the conversion module 22;

[0113] Step 3: Use the wavelet threshold denoising method to reduce the noise of the sound digital signal to obtain an effective sound signal;

[0114] Step 4: Extract the short-term energy of the sound signal Short-term average zero-crossing rate Mel cepstral coefficient MFCC and linear prediction cepstral coefficient LPCC;

[0115] Step 5: Use decision tree support vector machine to analyze the above sound features, see Figure 8 The analysis and processing process is: frame extraction feature - adaptive segmentation - DT-SVM anomaly detection - recognition result, and the dangerous human voice signal recognition result δ is obtained, where 0 is a non-dangerous signal and 1 is a dangerous signal, and this recognition result δ is used as an auxiliary decision-making method.

[0116] The above-mentioned final decision method is to use the linear programming equation to integrate the results of the main decision method and the auxiliary decision method; the linear regression equation is: y=w1f(q)+w2δ, where w1 and w2 are weights, which can be adjusted according to the situation, and are set to 10 and 1 respectively here.

[0117] In an embodiment of the present invention, based on the posture information and sound signals collected by the safety clothing worn by the test personnel, the collected data is analyzed to evaluate the safety status of the personnel, thereby performing real-time monitoring of the personnel, reducing the risk level of danger, and improving the efficiency of safety assurance.

[0118] In addition, the above is a detailed introduction to the safety clothing danger signal identification method based on multimodal information fusion and deep learning provided by the embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for identifying danger signals of safety clothing, characterized in that The following steps are involved: S1, obtaining trigger information, location information and sound information of personnel; S2, designing an independent triggering method according to the triggering information: using the main triggering device (1) to collect the acceleration, angular velocity and joint pressure data of the personnel, and then filtering the collected data using a FIR filter to obtain an effective measurement signal, using the gradient descent method to fuse the denoised acceleration, angular velocity and joint pressure data to obtain an objective function, using this objective function as the main decision function, and using this method as the main decision method; S3, based on the sound information, a sound recognition decision method is designed using a deep learning method: first, a sound detection device (2) is used to collect the sound information of the person, and then the sound analog signal is converted into a digital signal through a conversion module (22), and the sound digital signal is subjected to a wavelet threshold denoising method to reduce the noise, thereby obtaining an effective sound signal, extracting the short-time energy, short-time average zero-crossing rate, Mel cepstral coefficient and linear prediction cepstral coefficient of the sound signal, and finally using a decision tree support vector machine to perform data analysis on the above-mentioned sound features to obtain a dangerous human voice signal target function, using this target function as an auxiliary decision function, and using this method as an auxiliary decision method; S4, design the final decision method based on the main decision method and the auxiliary decision method: use the linear programming equation to integrate the results of the main decision method and the auxiliary decision method; The main decision-making method steps are: Step 1: Use the main trigger device (1) to collect the acceleration a of the person x 、a y 、a z , angular velocity ω x ,ω y ,ω z , and the magnetic strength m x 、m y 、m z ; Step 2: Collect the acceleration a x 、a y 、a z , angular velocity w x 、w y 、w z And the magnetic intensity m, use FIR filter to filter and get effective measurement signal; Step 3: Establish the navigation coordinate system and the carrier coordinate system, then the rotation matrix is: Where ψ is the yaw angle, ρ is the pitch angle, and θ is the roll angle; Denoted as: The three attitude angles are: Attitude quaternion q = (q0 q1 q2 q3) T and the gyroscope angular velocity ω x ,ω y ,ω z The satisfied relationship is: The relationship between attitude quaternion and attitude matrix can be expressed as: The relationship between the three-axis accelerometer and the pitch angle and roll angle is: [a x a y a z ] T =[g·sinθcosρ-g·sinρ-g·cosθcosρ] T The relationship between the three-axis magnetometer and the yaw angle is: Step 4: Use the gradient descent method to fuse the denoised acceleration, angular velocity and magnetic intensity data. The iterative formula is: The objective function is obtained as: This objective function is used as the main decision-making method; The steps of the auxiliary decision-making method are: Step 1: Use the sound detection device (2) to collect the sound information of the person; Step 2: Convert the sound analog signal into a sound digital signal through a conversion module (22); Step 3: Use the wavelet threshold denoising method to reduce the noise of the sound digital signal to obtain an effective sound signal; Step 4: Extract the short-term energy of the sound signal Short-term average zero-crossing rate Mel Cepstral Coefficients MFCC and Linear Prediction Cepstral Coefficients LPCC; Step 5: Use a decision tree support vector machine to perform data analysis on the above sound features to obtain a dangerous human voice signal recognition result δ, where 0 represents a non-dangerous signal and 1 represents a dangerous signal, and use this recognition result δ as an auxiliary decision-making method.

2. The safety clothing danger signal recognition method according to claim 1, characterized in that: The final decision method is to use a linear programming equation to fuse the results of the main decision method and the auxiliary decision method; the linear programming equation is: y=w1f(q)+w2δ, where w1 and w2 are weights.

Images