A safety early warning method and system for outdoor personnel

By collecting and analyzing the elevation value, speed and sign value sequences in outdoor sports, using time-series encoder and Gaussian modeling, the problem of the safety warning results of outdoor personnel in the existing technology is solved, and the movement status of outdoor personnel is achieved accurate reflection and accuracy of safety warning results.

CN118658623BActive Publication Date: 2024-11-22ZHEJIANG NATUREHIKE SPORTING PRODUCTS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411111967.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-22
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

The prior art cannot accurately obtain the safety warning results of each outdoor person during outdoor sports, and ignores the exercise situation.

Method used

Collect the elevation value sequence, speed sequence and sign value sequence of outdoor personnel, use the time series encoder to extract the state vector, build a correlation matrix, calculate the deviation value, combine the sign value to judge the safety warning results, and determine the normal interval through Gaussian modeling to ensure that Warning accuracy.

Benefits of technology

It realizes accurate reflection of the exercise status of outdoor personnel and accurate acquisition of safety warning results, ensuring the accuracy of safety warning results for each outdoor personnel.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118658623B_ABST
    Figure CN118658623B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of personnel safety warning, and in particular to a safety warning method and system for outdoor personnel, the method comprising: collecting the elevation value sequence, speed sequence and vital sign value sequence of outdoor personnel within a preset time period; extracting features of the elevation value sequence, speed sequence and vital sign value sequence based on a time series encoder to obtain the state vector of the current collection moment and the previous adjacent collection moment; constructing the correlation matrix of the outdoor personnel at the current collection moment and the previous adjacent collection moment based on the correlation between any state vectors; calculating the deviation value of the correlation matrix between the current collection moment and the previous adjacent collection moment; obtaining the safety warning result of the outdoor personnel at the current collection moment based on the deviation value and the vital sign value at the current collection moment. Through the technical solution of the present application, the safety warning result of each outdoor personnel can be accurately obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of personnel safety warning technology, and in particular to a safety warning method and system for outdoor personnel. Background Art

[0002] Outdoor sports are a group of collective activities held in natural places, including mountaineering, rock climbing, camping, picnics, streams, exploration and other projects. Most outdoor sports are adventurous, so the safety of outdoor personnel is particularly important during outdoor sports.

[0003] At present, a patent application document with publication number CN115760518A discloses a method, device, electronic device and storage medium for monitoring personnel activities, wherein the method includes: determining a target area for outdoor activities of a target person; wherein the target area includes a plurality of target sub-areas; authenticating the target person; displaying risk warning information of each target sub-area to the target person who has passed the identity authentication; and obtaining location information of the target person who has passed the identity authentication, and transmitting the location information to a server, wherein the risk warning information is a safety factor level of each target sub-area determined based on the terrain and the risk factor of the construction task itself.

[0004] However, although the above method can obtain the safety factor level of the target person in each target sub-area, it ignores the movement of outdoor personnel in the target area and cannot accurately obtain the safety warning results for each outdoor person. Summary of the invention

[0005] In order to solve the above technical problems, the present application provides a safety warning method and system for outdoor personnel, which can accurately obtain the safety warning results of each outdoor personnel.

[0006] In a first aspect, the present application provides a safety warning method for outdoor personnel, which collects elevation value sequences, speed sequences and vital sign value sequences of outdoor personnel within a preset time period, wherein the vital sign value is heart rate or blood oxygen saturation; performs feature extraction on the elevation value sequence, the speed sequence and the vital sign value sequence based on a timing encoder to obtain a state vector of a current collection moment and a previous adjacent collection moment, wherein the state vector includes an elevation state, a speed state and a vital sign value state; constructs a correlation matrix of the outdoor personnel at the current collection moment and the previous adjacent collection moment based on the correlation between any state vectors; calculates a deviation value of the correlation matrix between the current collection moment and the previous adjacent collection moment; and obtains a safety warning result of the outdoor personnel at the current collection moment based on the deviation value and the vital sign value at the current collection moment.

[0007] In one embodiment, the timing encoder is a recurrent neural network; based on the timing encoder, feature extraction is performed on the elevation value sequence, the speed sequence and the eigenvalue sequence to obtain the state vector of the current acquisition moment and the previous adjacent acquisition moment, including: the elevation values ​​in the elevation value sequence are sequentially input into the timing encoder to obtain the timing features corresponding to each elevation value, the timing features corresponding to the elevation value at the previous adjacent acquisition moment are used as the elevation state at the previous adjacent acquisition moment, and the timing features corresponding to the elevation value at the current acquisition moment are used as the elevation state at the current acquisition moment; the instantaneous speed in the speed sequence is sequentially input into the timing encoder to obtain the speed state at the current acquisition moment and the previous adjacent acquisition moment; the eigenvalues ​​in the eigenvalue sequence are sequentially input into the timing encoder to obtain the eigenvalue state at the current acquisition moment and the previous adjacent acquisition moment.

[0008] In one embodiment, the training method of the timing encoder includes: collecting multiple historical sequences, the historical sequences include elevation value historical sequences, speed historical sequences and vital sign value historical sequences; for a historical sequence, inputting an arbitrary value and all values ​​before the arbitrary value into the timing encoder in sequence to obtain the timing feature corresponding to the arbitrary value; inputting the timing feature into a fully connected neural network to obtain a prediction result of the next adjacent value of the arbitrary value in the historical sequence; calculating the cross-entropy loss function of the prediction result and the next adjacent value, and performing back propagation based on the cross-entropy loss function to update the timing encoder and the fully connected neural network to complete one training; iteratively training the timing encoder until the cross-entropy loss function is less than the set loss or the number of iterations is greater than the set number, and stopping to obtain a trained timing encoder.

[0009] In one embodiment, the correlation matrix is ​​a square matrix with 3 rows and 3 columns, and the correlation between the eigenvalue state and the elevation state is the inner product between the eigenvalue state and the elevation state.

[0010] In one embodiment, the deviation value satisfies the relationship:

[0011] ;in, is the correlation matrix of the previous adjacent acquisition time, is the correlation matrix at the current acquisition moment, Representation calculation The Euclidean distance of is the deviation value.

[0012] In one embodiment, the safety warning result of the outdoor personnel at the current collection moment obtained based on the deviation value and the vital sign value at the current collection moment includes: in response to the deviation value being in a first normal range, and the vital sign value at the current collection moment being in a second normal range, the safety warning result of the outdoor personnel at the current collection moment is safe; in response to the deviation value not being in the first normal range, or the vital sign value at the current collection moment not being in the second normal range, the safety warning result of the outdoor personnel at the current collection moment is that there is a safety hazard, and an alarm is issued; wherein, the first normal range is the value range of the deviation value when the outdoor personnel is in a normal state, and the second normal range is the value range of the vital sign value when the outdoor personnel is in a normal state.

[0013] In one embodiment, the method for obtaining the first normal interval includes: collecting multiple historical deviation values ​​of the outdoor personnel within a historical time; performing anomaly point detection on all historical deviation values ​​to obtain the abnormal value of each historical deviation value, and marking the historical deviation value whose abnormal value is less than the abnormal threshold as the core historical deviation value, performing Gaussian modeling on all core historical deviation values ​​to obtain a first Gaussian model; and determining the first normal interval based on the 3 times standard deviation criterion and the first Gaussian model.

[0014] In one embodiment, the method for obtaining the first normal interval includes: performing Gaussian modeling on the historical deviation values ​​of other outdoor personnel to obtain a general Gaussian model, wherein the general Gaussian model includes multiple first Gaussian components; for a first Gaussian component, inputting multiple historical deviation values ​​of the outdoor personnel into the first Gaussian component respectively to obtain the response probability of each historical deviation value in the first Gaussian component; calculating the 0th order statistics, 1st order statistics and 2nd order statistics of the outdoor personnel in the first Gaussian component based on the response probability, and correcting the weight, mean and standard deviation of the first Gaussian component; after correcting all the first Gaussian components in the general Gaussian model, obtaining a second Gaussian model of the outdoor personnel; determining the first normal interval according to the 3 times standard deviation criterion and the second Gaussian model.

[0015] In one embodiment, the second Gaussian model includes a plurality of second Gaussian components, and determining the first normal interval according to the 3 times standard deviation criterion and the second Gaussian model includes: for a second Gaussian component, determining the minimum allowable probability of the second Gaussian component based on the 3 times standard deviation criterion, and the minimum allowable probability satisfies the relationship:

[0016] ;in, For the The mean of the second Gaussian components, For the The standard deviation of the second Gaussian component, Indicates The second Gaussian component, For the The invention relates to a method for determining a minimum allowable probability of a second Gaussian component; performing weighted summation of all minimum allowable probabilities based on the weight of each second Gaussian component to obtain a minimum target probability of the second Gaussian model; and taking all deviation values ​​whose output probability of the second Gaussian model is greater than or equal to the minimum target probability as the first normal area.

[0017] In a second aspect of the present application, a safety warning system for outdoor personnel is also provided, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a safety warning method for outdoor personnel described in the first aspect of the present application is implemented.

[0018] The technical solution of this application has the following beneficial technical effects:

[0019] Through the technical solution provided by the present application, firstly, the elevation value sequence, speed sequence and vital value sequence of outdoor personnel within a preset time period are collected, and the elevation state, speed state and vital value state at the current collection moment are obtained by using a timing encoder, and the correlation matrix of the outdoor personnel at the current collection moment is constructed based on the correlation between any states. The correlation matrix at the current collection moment comprehensively reflects the correlation between the elevation state, speed state and vital value state of the outdoor personnel at the current collection moment, and can accurately reflect the movement state of the outdoor personnel. Furthermore, the deviation value of the correlation matrix between the current collection moment and the previous adjacent collection moment is obtained, and the accurate safety warning result at the current collection moment is obtained by judging whether the deviation value and the vital value at the current collection moment are in the normal range.

[0020] Furthermore, taking into account the differences in physical fitness and athletic ability of different outdoor personnel, resulting in different normal ranges of deviation values ​​for different outdoor personnel, multiple historical deviation values ​​of the outdoor personnel in historical time are used to determine the normal range of the deviation value of the outdoor personnel.

[0021] Furthermore, considering that the number of historical deviation values ​​of an outdoor person is small, it is difficult to ensure the accuracy of the normal range of deviation values ​​based only on the historical deviation values ​​of the outdoor person. First, the historical deviation values ​​of other outdoor persons are used to construct a Gaussian model common to all outdoor persons, and then the historical deviation values ​​of the outdoor persons are used to correct the background Gaussian model to obtain an accurate Gaussian model of the outdoor person, and then the 3 times standard deviation criterion is used to obtain the accurate normal range of deviation values ​​for the outdoor person, thereby ensuring the accuracy of the safety warning results for each outdoor person. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-restrictive manner, and the same or corresponding numbers represent the same or corresponding parts, wherein.

[0023] Figure 1 It is a flow chart of a safety warning method for outdoor personnel according to an embodiment of the present application.

[0024] Figure 2 It is a structural block diagram of a safety warning system for outdoor personnel according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0026] According to a first aspect of the present application, the present application provides a safety warning method for outdoor personnel, which is used to provide safety warnings to outdoor personnel during outdoor sports. Figure 1 FIG. 1 is a flow chart of a safety warning method for outdoor personnel according to an embodiment of the present application. Figure 1 As shown, the safety warning method for outdoor personnel includes steps S101 to S105, which are described in detail below.

[0027] S101, collecting an altitude value sequence, a speed sequence and a vital sign value sequence of an outdoor person within a preset time period, wherein the vital sign value is a heart rate or blood oxygen saturation.

[0028] In one embodiment, for any outdoor person, the wearable smart device worn by the outdoor person collects the elevation value, instantaneous speed and vital sign value of the outdoor person at each collection moment within a preset time period, wherein the elevation value is the altitude of the outdoor person, and the vital sign value is used to reflect the physical condition of the outdoor person, and the vital sign value is the heart rate or blood oxygen saturation.

[0029] The preset time period includes the current collection moment and a set number of collection moments before the current collection moment, and the collection moment is related to the collection frequency of the wearable smart device, wherein the wearable smart device may be a smart phone or a smart bracelet. For example, if the collection frequency of the wearable smart device is 10 times per minute, then every 6 seconds is a collection moment, and the preset time period may be the current collection moment and 5 minutes before the current collection moment, that is, the set number is 50.

[0030] All elevation values ​​are arranged in the order of collection time to obtain the elevation value sequence of outdoor personnel in a preset time period. Using the same method, the elevation value sequence, speed sequence and vital sign value sequence can be obtained.

[0031] S102, performing feature extraction on the elevation value sequence, the velocity sequence and the eigenvalue sequence based on a time series encoder to obtain a state vector of a current acquisition moment and a previous adjacent acquisition moment, wherein the state vector includes an elevation state, a velocity state and an eigenvalue state.

[0032] In one embodiment, the time series encoder is a recurrent neural network. Specifically, the time series encoder can adopt the network structure of LSTM or GRU; LSTM (Long Short Term Memory) is an existing recurrent neural network; GRU (Gate Recurrent Unit) is also an existing recurrent neural network, which is an improvement on LSTM and solves problems such as long-term memory and gradient in back propagation; LSTM and GRU can both extract features from time series data and obtain the time series features corresponding to each acquisition moment in the time series data in turn.

[0033] It should be noted that when the time series encoder adopts LSTM, each acquisition moment corresponds to a long-term state and a short-term state. The long-term state or the short-term state can be used as the time series feature corresponding to the acquisition moment, and this application does not impose any restrictions.

[0034] In one embodiment, feature extraction is performed on the elevation value sequence, the speed sequence and the eigenvalue sequence based on a timing encoder to obtain a state vector of the current acquisition moment and the previous adjacent acquisition moment, including: inputting the elevation values ​​in the elevation value sequence into the timing encoder in sequence to obtain a timing feature corresponding to each elevation value, using the timing feature corresponding to the elevation value at the previous adjacent acquisition moment as the elevation state at the previous adjacent acquisition moment, and using the timing feature corresponding to the elevation value at the current acquisition moment as the elevation state at the current acquisition moment; inputting the instantaneous speed in the speed sequence into the timing encoder in sequence to obtain the speed state at the current acquisition moment and the previous adjacent acquisition moment; inputting the eigenvalues ​​in the eigenvalue sequence into the timing encoder in sequence to obtain the eigenvalue states at the current acquisition moment and the previous adjacent acquisition moment.

[0035] The elevation state, the speed state and the eigenvalue state correspond to the state vectors, all state vectors have the same size, and the value of the size is related to the network structure of the timing encoder.

[0036] It should be noted that the time series encoder can also adopt the network structure of Transformer. Transformer abandons the traditional recurrent neural network. The entire network structure is completely composed of the self-attention mechanism, which can simultaneously obtain the time series features corresponding to each moment in the time series data.

[0037] In one embodiment, in order to ensure that the time series encoder can accurately obtain the state vector of the current acquisition moment and the previous adjacent acquisition moment, the time series encoder needs to be trained. Specifically, the training method of the time series encoder includes: collecting multiple historical sequences, the historical sequences include elevation value historical sequences, speed historical sequences and vital sign value historical sequences; for a historical sequence, inputting any numerical value and all numerical values ​​before the arbitrary numerical value into the time series encoder in sequence to obtain the time series feature corresponding to the arbitrary numerical value; inputting the time series feature into the fully connected neural network to obtain the prediction result of the next adjacent numerical value of the arbitrary numerical value in the historical sequence; calculating the cross entropy loss function of the prediction result and the next adjacent numerical value, and performing back propagation based on the cross entropy loss function to update the time series encoder and the fully connected neural network to complete one training; iteratively training the time series encoder until the cross entropy loss function is less than the set loss or the number of iterations is greater than the set number, and stopping to obtain a trained time series encoder.

[0038] Among them, the set loss value is 0.001, and the set number of times is 800.

[0039] S103, constructing a correlation matrix of the outdoor personnel at the current collection moment and the previous adjacent collection moment based on the correlation between any state vectors.

[0040] In one embodiment, for the last adjacent acquisition moment, its elevation state can reflect the movement of outdoor personnel along the elevation value plane before the last adjacent acquisition moment; its speed state can reflect the movement of outdoor personnel along the horizontal plane before the last adjacent acquisition moment; and the vital sign value state can reflect the change of outdoor personnel's physical state during the movement before the last adjacent acquisition moment. Similarly, for the current acquisition moment, its elevation state can reflect the movement of outdoor personnel along the elevation value plane before the current acquisition moment; its speed state can reflect the movement of outdoor personnel along the horizontal plane before the current acquisition moment; and the vital sign value state can reflect the change of outdoor personnel's physical state during the movement before the last adjacent acquisition moment.

[0041] When the outdoor personnel are in a stationary state, the vital sign values ​​of the outdoor personnel are basically stable, and when the outdoor personnel move along the elevation value plane and the horizontal plane, the vital sign values ​​of the outdoor personnel will also change accordingly. Therefore, the vital sign value state is affected by the elevation state and the speed state; in addition, under the premise of the same vital sign value, when the outdoor personnel move on the same elevation value plane, the instantaneous speed of the outdoor personnel is relatively large, and when the outdoor personnel cross different elevation value planes, the instantaneous speed of the outdoor personnel will decrease. Therefore, the speed state will be affected by the elevation state and the vital sign value state; in summary, there is a certain correlation between the vital sign value state, the elevation state and the speed state. Based on the correlation between the vital sign value state, the elevation state and the speed state, a correlation matrix of the outdoor personnel at the current collection time and the previous adjacent collection time is constructed.

[0042] Specifically, the correlation matrix is ​​a square matrix with 3 rows and 3 columns, and the correlation between the eigenvalue state and the elevation state is the inner product between the eigenvalue state and the elevation state.

[0043] In this way, the correlation matrix at the current collection time comprehensively reflects the correlation between the elevation state, speed state and vital sign value state of outdoor personnel at the current collection time.

[0044] S104, calculating the deviation value of the correlation matrix between the current collection moment and the previous adjacent collection moment.

[0045] In one embodiment, the deviation value satisfies the relationship:

[0046] ;in, is the correlation matrix of the previous adjacent acquisition time, is the correlation matrix at the current acquisition moment, Representation calculation The Euclidean distance of is the deviation value.

[0047] S105, obtaining a safety warning result of the outdoor person at the current collection moment based on the deviation value and the physical sign value at the current collection moment.

[0048] In one embodiment, the deviation value is the Euclidean distance of the correlation matrix between the current collection moment and the previous adjacent collection moment, which comprehensively considers the elevation state, speed state and vital sign value state of the current collection moment and the previous adjacent collection moment, and can accurately reflect the state changes of outdoor personnel during outdoor sports.

[0049] In one embodiment, the safety warning result of the outdoor personnel at the current collection moment obtained based on the deviation value and the vital sign value at the current collection moment includes: in response to the deviation value being in a first normal range, and the vital sign value at the current collection moment being in a second normal range, the safety warning result of the outdoor personnel at the current collection moment is safe; in response to the deviation value not being in the first normal range, or the vital sign value at the current collection moment not being in the second normal range, the safety warning result of the outdoor personnel at the current collection moment is that there is a safety hazard, and an alarm is issued; wherein, the first normal range is the value range of the deviation value when the outdoor personnel is in a normal state, and the second normal range is the value range of the vital sign value when the outdoor personnel is in a normal state.

[0050] The second normal interval is manually set by those skilled in the art.

[0051] In one embodiment, the method for obtaining the first normal interval includes: collecting multiple historical deviation values ​​of the outdoor personnel within a historical time; performing anomaly point detection on all historical deviation values ​​to obtain the abnormal value of each historical deviation value, and marking the historical deviation value whose abnormal value is less than the abnormal threshold as the core historical deviation value, performing Gaussian modeling on all core historical deviation values ​​to obtain a first Gaussian model; and determining the first normal interval based on the 3 times standard deviation criterion and the first Gaussian model.

[0052] Wherein, the first Gaussian model is a single Gaussian model or a Gaussian mixture model. If the first Gaussian model is a single Gaussian model, the first Gaussian model includes a first mean and the first standard deviation ; The first normal interval can be determined according to the 3 times standard deviation criterion, and the first normal interval satisfies the relationship: If the first Gaussian model is a Gaussian mixture model, the method for obtaining the first normal interval is consistent with the method of "determining the first normal interval based on the 3 times standard deviation criterion and the second Gaussian model" in the subsequent description, and will not be repeated here.

[0053] Among them, the LOF algorithm (Local Outlier Factor) can be used to obtain the outlier value of each historical deviation value. The LOF algorithm is a distance-based outlier point detection algorithm. The larger the outlier value, the greater the probability that the historical deviation value is abnormal. The value of the abnormal threshold is 1.

[0054] In this way, the first normal interval is accurately determined by utilizing the historical deviation values ​​of the outdoor personnel themselves within the historical period. The first normal interval can reflect the range of deviation values ​​of the outdoor personnel in a normal state, thereby avoiding the differences in physical fitness and athletic ability between different outdoor personnel, which may lead to inaccurate safety warning results.

[0055] In another embodiment, due to the small number of historical deviation values ​​of the outdoor personnel, it is difficult to ensure the accuracy of the first Gaussian model based only on the historical deviation values ​​of the outdoor personnel. However, due to the differences in physical fitness and athletic ability of different outdoor personnel, using the historical deviation values ​​of other outdoor personnel will lead to inaccurate safety warning results. Therefore, this embodiment draws on the idea of ​​the GMM-UBM algorithm. First, the historical deviation values ​​of other outdoor personnel are used to construct a Gaussian model common to all outdoor personnel, and then the historical deviation values ​​of the outdoor personnel are used to correct the background Gaussian model to obtain an accurate Gaussian model for the outdoor personnel.

[0056] The GMM-UBM algorithm is a Gaussian modeling method used in the field of voiceprint recognition. It is used to establish a Gaussian model of the voiceprint characteristics of a specific speaker. This application transfers the Gaussian modeling method originally used in the field of voiceprint recognition to the field of personnel safety warning, so as to obtain an accurate Gaussian model of the outdoor personnel under the condition of limited historical deviation values, thereby ensuring the accuracy of the outdoor personnel safety warning results.

[0057] Specifically, the method for obtaining the first normal interval includes: performing Gaussian modeling on the historical deviation values ​​of other outdoor personnel to obtain a general Gaussian model, wherein the general Gaussian model is a Gaussian mixture model including multiple first Gaussian components; for a first Gaussian component, inputting the multiple historical deviation values ​​of the outdoor personnel into the first Gaussian component respectively to obtain the response probability of each historical deviation value in the first Gaussian component; calculating the 0th order statistics, 1st order statistics and 2nd order statistics of the outdoor personnel in the first Gaussian component based on the response probability, and correcting the weight, mean and standard deviation of the first Gaussian component; after correcting all the first Gaussian components in the general Gaussian model, obtaining the second Gaussian model of the outdoor personnel; determining the first normal interval according to the 3 times standard deviation criterion and the second Gaussian model.

[0058] Wherein, the general Gaussian model satisfies the relationship:

[0059] ;in, represents any deviation value, is the number of all first Gaussian components in the general Gaussian model, For the The first Gaussian component, For the The weight of the first Gaussian component, For the The mean of the first Gaussian components, For the The standard deviation of the first Gaussian component, is the deviation value The output result of the corresponding general Gaussian model. The general Gaussian model is constructed using the historical deviation values ​​of other outdoor personnel, and can reflect the overall level of the deviation values ​​of all outdoor personnel; preferably, the number of all first Gaussian components is 2-5.

[0060] The number of historical deviation values ​​of the outdoor personnel is recorded as , with the The historical deviation value is an example. The historical deviation value is The response probability of the first Gaussian component is:

[0061] ;

[0062] in, For the Historical deviation values In the The output result under the first Gaussian component is For the Historical deviation values The corresponding output of the general Gaussian model is, For the The historical deviation value is The response probability of the first Gaussian component.

[0063] Specifically, after obtaining each historical deviation value in the After the response probability of the first Gaussian component, the outdoor personnel can be calculated The 0th order statistics, 1st order statistics and 2nd order statistics on the first Gaussian component, the 0th order statistics satisfies the relationship:

[0064] ;in, Indicates that the outdoor personnel are The 0th-order statistic on the first Gaussian component, is the number of historical deviation values ​​of the outdoor personnel, For the The historical deviation value is The response probability of the first Gaussian component. The first-order statistic satisfies the relationship:

[0065] ;in, Indicates that the outdoor personnel are The first-order statistics on the first Gaussian component, For the The second-order statistics satisfy the relationship:

[0066] ;in, Indicates that the outdoor personnel are The second-order statistics on the first Gaussian component.

[0067] In one embodiment, correcting the weight, mean and standard deviation of the first Gaussian component includes: calculating a correction coefficient of the first Gaussian component, wherein the correction coefficient satisfies the relationship:

[0068] ;in, For the outdoor personnel The 0th-order statistic on the first Gaussian component, is the preset coefficient, the value range of the preset coefficient is 8~20. In this embodiment, the preset coefficient ; Based on the correction coefficient, the weight, mean and standard deviation of the first Gaussian component are corrected, and the correction process satisfies the following relationship:

[0069] ;

[0070] ;

[0071] ;

[0072] in, is the number of historical deviation values ​​of the outdoor personnel, , and The outdoor personnel are respectively The 0th-order statistics, 1st-order statistics, and 2nd-order statistics on the first Gaussian component, , and Respectively The weights, mean and standard deviation of the first Gaussian components, is the correction factor, , and Respectively The corrected weight, mean and standard deviation of the first Gaussian component.

[0073] Understandably, The weight, mean and standard deviation in the first Gaussian component are replaced by the corrected weight, mean and standard deviation to obtain the first The second Gaussian component is a second Gaussian component; after the weights, means and standard deviations of all the first Gaussian components in the general Gaussian model are corrected in the same way, the second Gaussian model of the outdoor personnel can be obtained, and the second Gaussian model can reflect the range of deviation values ​​of the outdoor personnel under normal conditions. Different outdoor personnel have different second Gaussian models.

[0074] In one embodiment, the second Gaussian model includes a plurality of second Gaussian components, and determining the first normal interval according to the 3 times standard deviation criterion and the second Gaussian model includes: for a second Gaussian component, determining the minimum allowable probability of the second Gaussian component based on the 3 times standard deviation criterion, and the minimum allowable probability satisfies the relationship:

[0075] ;in, For the The mean of the second Gaussian components, For the The standard deviation of the second Gaussian component, Indicates The second Gaussian component, For the The invention relates to a method for determining a minimum allowable probability of a second Gaussian component; performing weighted summation of all minimum allowable probabilities based on the weight of each second Gaussian component to obtain a minimum target probability of the second Gaussian model; and taking all deviation values ​​whose output probability of the second Gaussian model is greater than or equal to the minimum target probability as the first normal area.

[0076] Wherein, the minimum target probability satisfies the relationship:

[0077] ;in, For the The weight of the second Gaussian component, For the The minimum allowed probability of the second Gaussian component, is the number of all second Gaussian components, is the minimum target probability.

[0078] In this way, an accurate second Gaussian model of an outdoor person is obtained by utilizing the limited historical deviation value of the outdoor person, and then an accurate first normal interval is determined.

[0079] The technical principle and implementation details of the safety warning method for outdoor personnel of the present application are introduced above through specific embodiments. Through the technical solution provided by the present application, the elevation value sequence, speed sequence and vital sign value sequence of outdoor personnel in a preset time period are first collected, and the elevation state, speed state and vital sign value state at the current collection moment are obtained by using a timing encoder, and the correlation matrix of outdoor personnel at the current collection moment is constructed based on the correlation between any states. The correlation matrix at the current collection moment comprehensively reflects the correlation between the elevation state, speed state and vital sign value state of outdoor personnel at the current collection moment, and can accurately reflect the motion state of outdoor personnel. Further, the deviation value of the correlation matrix between the current collection moment and the previous adjacent collection moment is obtained, and the accurate safety warning result at the current collection moment is obtained by judging whether the deviation value and the vital sign value at the current collection moment are in the normal range.

[0080] According to the second aspect of the present application, the present application also provides a safety warning system for outdoor personnel. Figure 2 is a structural block diagram of a safety warning system for outdoor personnel according to an embodiment of the present application. Figure 2 As shown, the system 50 includes a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a safety warning method for outdoor personnel according to the first aspect of the present application is implemented. The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, whose settings and functions are known in the art, so they are not repeated here.

[0081] In the present application, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in this application may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0082] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. A safety warning method for outdoor personnel, characterized in that: The early warning method comprises: Collecting the altitude value sequence, speed sequence and vital sign value sequence of outdoor personnel within a preset time period, wherein the vital sign value is heart rate or blood oxygen saturation; Based on a time series encoder, feature extraction is performed on the elevation value sequence, the speed sequence and the eigenvalue sequence to obtain a state vector of a current acquisition moment and a previous adjacent acquisition moment, wherein the state vector includes an elevation state, a speed state and an eigenvalue state; Constructing a correlation matrix of the outdoor personnel at the current collection moment and the last adjacent collection moment based on the correlation between any state vectors; Calculate the deviation value of the correlation matrix between the current acquisition time and the previous adjacent acquisition time; Obtaining a safety warning result of the outdoor person at the current collection time based on the deviation value and the physical sign value at the current collection time; Wherein, the elevation value is the altitude where the outdoor personnel are located; The training method of the temporal encoder comprises: Collecting multiple historical sequences, the historical sequences include elevation value historical sequences, velocity historical sequences and vital sign value historical sequences; For a historical sequence, an arbitrary value and all values ​​before the arbitrary value are sequentially input into a time series encoder to obtain a time series feature corresponding to the arbitrary value; Inputting the time series feature into a fully connected neural network to obtain a prediction result of the next adjacent value of any value in the historical sequence; Calculating a cross entropy loss function between the prediction result and the next adjacent value, and performing back propagation based on the cross entropy loss function to update the temporal encoder and the fully connected neural network to complete a training; Iteratively training the time series encoder until the cross entropy loss function is less than a set loss or the number of iterations is greater than a set number, thereby obtaining a trained time series encoder; The correlation matrix is ​​a square matrix of 3 rows and 3 columns, and the correlation between the eigenvalue state and the elevation state is the inner product between the eigenvalue state and the elevation state; The correlation matrix at the current collection time comprehensively reflects the correlation between the elevation state, speed state and vital sign value state of outdoor personnel at the current collection time; The safety warning result of the outdoor person at the current collection time obtained based on the deviation value and the physical sign value at the current collection time includes: In response to the deviation value being in the first normal range, and the vital sign value at the current collection time being in the second normal range, the safety warning result of the outdoor person at the current collection time is safe; In response to the deviation value not being in the first normal range, or the vital sign value at the current collection time not being in the second normal range, the safety warning result of the outdoor personnel at the current collection time is that there is a safety hazard, and an alarm is issued; The first normal interval is the value range of the deviation value when the outdoor personnel is in a normal state, and the second normal interval is the value range of the physical sign value when the outdoor personnel is in a normal state; The method for obtaining the first normal interval includes: Performing Gaussian modeling on historical deviation values ​​of other outdoor personnel to obtain a universal Gaussian model, wherein the universal Gaussian model includes a plurality of first Gaussian components; For a first Gaussian component, a plurality of historical deviation values ​​of the outdoor personnel are respectively input into the first Gaussian component to obtain a response probability of each historical deviation value in the first Gaussian component; Calculating the 0th order statistic, the 1st order statistic and the 2nd order statistic of the outdoor personnel in the first Gaussian component based on the response probability, and modifying the weight, mean and standard deviation of the first Gaussian component; After correcting all the first Gaussian components in the general Gaussian model, a second Gaussian model of the outdoor person is obtained; Determine a first normal interval according to the 3 times standard deviation criterion and the second Gaussian model; The second Gaussian model includes a plurality of second Gaussian components, and determining the first normal interval according to the three-times standard deviation criterion and the second Gaussian model includes: For a second Gaussian component, the minimum permissible probability of the second Gaussian component is determined based on the 3-times standard deviation criterion, and the minimum permissible probability satisfies the relationship: ;in, For the The mean of the second Gaussian components, For the The standard deviation of the second Gaussian component, Indicates The second Gaussian component, For the The minimum allowed probability of the second Gaussian component; Performing a weighted summation on all minimum allowed probabilities based on the weight of each second Gaussian component to obtain a minimum target probability of the second Gaussian model; All deviation values ​​whose output probability of the second Gaussian model is greater than or equal to the minimum target probability are taken as the first normal area.

2. A safety warning method for outdoor personnel according to claim 1, characterized in that: The time series encoder is a recurrent neural network; based on the time series encoder, feature extraction is performed on the elevation value sequence, the speed sequence and the vital sign value sequence to obtain a state vector of the current acquisition time and the previous adjacent acquisition time, including: Inputting the elevation values ​​in the elevation value sequence into the timing encoder in sequence, obtaining the timing characteristics corresponding to each elevation value, taking the timing characteristics corresponding to the elevation value at the last adjacent acquisition moment as the elevation state at the last adjacent acquisition moment, and taking the timing characteristics corresponding to the elevation value at the current acquisition moment as the elevation state at the current acquisition moment; Inputting the instantaneous speeds in the speed sequence into the timing encoder in sequence to obtain the speed state at the current acquisition moment and the previous adjacent acquisition moment; The eigenvalues ​​in the eigenvalue sequence are sequentially input into the timing encoder to obtain the eigenvalue states at the current acquisition moment and the previous adjacent acquisition moment.

3. A safety warning method for outdoor personnel according to claim 1, characterized in that: The deviation value satisfies the relationship: ;in, is the correlation matrix of the previous adjacent acquisition time, is the correlation matrix at the current acquisition moment, Representation calculation The Euclidean distance of is the deviation value.

4. A safety warning system for outdoor personnel, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a safety warning method for outdoor personnel according to any one of claims 1 to 3 is implemented.