CN103761832A - Wearable human body collision early warning and protection device and early warning and protection method thereof - Google Patents

Abstract

The invention relates to a wearable human body collision early warning and protection device. The wearable human body collision early warning and protection device comprises a belt body, wherein multiple air bags are arranged in the middle of the belt body at intervals, multi-element sensor sets, a micro-silicon accelerometer and gyroscope and an alarm unit are installed at positions, above the air bags and below the air bags, of the belt body, the output end of each multi-element sensor set and the output end of the micro-silicon accelerometer and gyroscope are connected with the input end of a main controller through a unit controller, and the output end of the main controller is connected with the input end of each air bag and the input end of the alarm unit through the unit controller. The invention further discloses an early warning and protection method of the wearable human body collision early warning and protection device. According to the wearable human body collision early warning and protection device and the early warning and protection method of the wearable human body collision early warning and protection device, acquired information is compared with an experience data set through the main controller, the number of suspicious approaching objects and the approaching distance of the suspicious approaching objects are given, an executive command is given, then an audible alarm is given out, and the air bags are opened under control; meanwhile, a preset text message is sent to a specific mobile phone through a wireless message sending unit, and the main controller makes a decision to select two-level protection execution actions.

Description

A kind of wearable human body anti-collision warning protective device and early warning protection method thereof

Technical field

The present invention relates to anti-collision warning guard technology field, especially a kind of wearable human body anti-collision warning protective device and early warning protection method thereof.

Background technology

Motion detection is the Data Detection about human body itself, and at present research mainly concentrates on, and sportsman's DATA REASONING, falls down detection etc., and Main Means is to realize application supervisory system by video or Inertial Measurement Unit.Fast and effeciently obtain body motion information, can make special population obtain necessary relief, yet, the early warning of human body collision detection and protection or a blank towards pedestrian, the walking along the street of children and special population is safely to keep the safety in production, ensure socially harmonious necessary links, avoid pedestrian impact, there is important economy and social effect.

Along with the development and application of sensor technology, the sensor such as, sound optical, electrical to modern hyperchannel from the sensor of initial single channel, the data volume of corresponding acquisition is corresponding growth also.Yet, any technology for information acquisition, there is its precision and measure limitation, and signal is inevitably subject to environmental interference, to the multiple sensors from Same Scene, because the type of its sensor is different with measuring principle, the existence of scene Self-variation and various interference, there is in varying degrees distortion or malfunctioning in the information of obtaining.

Summary of the invention

Primary and foremost purpose of the present invention is to provide a kind of and comprises that two-stage protection performs an action, volume is little, is easy to carry, and may be worn on wearable human body anti-collision warning protective device with it.

For achieving the above object, the present invention has adopted following technical scheme: a kind of wearable human body anti-collision warning protective device, comprise waistband body, a plurality of air bags are arranged at the interval, place, centre position of waistband body, on waistband body, be positioned at the upper of air bag, multielement bar group is installed in below, micro-silicon accelerometer and gyroscope, alarm unit, multielement bar group, micro-silicon accelerometer and gyrostatic output terminal with by cell controller, be connected with the input end of master controller, the output terminal of master controller by cell controller with respectively with air bag, the input end of alarm unit is connected.

Described multielement bar group is comprised of sonac, air pressure probe and infrared sensor, described cell controller by range cells, approach judging unit, unstability judging unit and performance element and form, on described waistband body, be positioned at air bag above or below wireless messages transmitting element is installed; The output terminal of sonac is connected with the input end of range cells, the output terminal of infrared sensor is connected with the input end that approaches judging unit, the output terminal of air pressure probe is connected with range cells, the input end that approaches judging unit respectively, micro-silicon accelerometer is connected with the input end of unstability judging unit with gyrostatic output terminal, and the output terminal of described performance element is connected with the input end of air bag, alarm unit, wireless messages transmitting element respectively.

Described alarm unit adopts loudspeaker.

The number of described air bag is four to six, and each air bag is positioned on same center line.

Described range cells, the output terminal that approaches judging unit, unstability judging unit are all connected with the input end of master controller, and the output terminal of master controller is connected with the input end of performance element.

A kind of early warning protection method that another object of the present invention is to provide wearable human body anti-collision warning protective device, the method comprises the step of following order:

(1) power-up initializing;

(2) information of master controller to multielement bar group, micro-silicon accelerometer and gyroscope output, acquiescence is in normal operating condition;

(3) data sequence that master controller collects sensor is extracted characteristic quantity, and judges whether to be about to occur collision situation according to support vector machines, if the determination result is YES, starts alarm unit and reports to the police, and continues Information Monitoring; Otherwise, return to step (2);

(4) after alarm unit is reported to the police, master controller judges whether to occur obvious collision status, if the determination result is YES, opens air bag, and starts wireless messages transmitting element and carry out wireless transmission; If the determination result is NO, determine whether false-alarm.

When false-alarm is judged, if the determination result is YES, give tacit consent in normal operating condition, otherwise, judge whether to be about to occur collision situation.

Judge whether to be about to occur that the determination methods of collision situation is specially:

(1) extract each sensor characteristics of multielement bar group, comprise temporal signatures, frequency domain character and time trend feature

A, the extraction of temporal signatures

Temporal signatures is the Time-domain Statistics characteristic obtaining according to various kinds of sensors output waveform, adopts following index:

Absolute mean: $\left|\stackrel{-}{x}\right|=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|$

Root-mean-square value: $X_{\mathrm{rms}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|}$

The degree of bias: $a=\sqrt{\frac{1}{6N}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^3$

Kurtosis: $\mathrm{\β}=\sqrt{\frac{N}{24}}[\frac{1}{N}{\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^4-3]$

In formula, x _ifor the discrete-time series of sensor signal, N is sequence samples number, and μ is average, and σ is variance;

B, the extraction of frequency domain character

Various kinds of sensors output sequence has typical frequecy characteristic separately, and it is carried out to spectrum analysis, obtains amplitude-frequency sequence, extracts the range value in several particular frequencies in amplitude-frequency sequence;

C, the extraction of time trend feature

The original signal recording by sensor is definite status flag signal and the stack of random disturbance.The signal sequence that is provided with original measurement is { x _t, t=1,2 ..., N}, this sequence can be expressed as a trend term and random entry and:

x _t=d _t+ξ _t

In formula, d _tfor trend term, it is a certain definite function of time t; ξ _tfor random entry, it has reflected the random composition of signal;

When the trend of extraction feature, first need time series to carry out de-noising, remove random entry, retain trend term, and then utilize various trend analysis to extract feature;

Adopt Mann-Kendall check to detect sequence variation trend to sensor and extract, for time series { x _t, t=1,2 ..., N}, the null hypothesis H of Mann-Kendall check ₀for stochastic variable and time independence, compute statistics T

$T=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{T}_i$

In formula,

at null hypothesis H ₀under, if there is not trend in time series, when N is larger, statistic T is similar to Normal Distribution, and has two formulas below to set up:

When n>10, calculate test statistics Z

If sequence calculates Z sometime---Sequence Trend characteristic quantity;

(2) according to characteristic quantity, set up the basic model of the SVM of limit risk classification

If the characteristic set obtaining from sensor input is by two classes---be about to occur collision and normally form, if feature x[i] correspondence belongs to the 1st class, y[i]=1, if feature x[i] correspondence belongs to the 2nd class, y[i]=-1, have so training sample set x[i], y[i] }, i=1,2,3 ..., n, ask optimal classification face wx-b=0, when the situation for linearly inseparable, use core inner product K (x[i], x[j]), by kernel function, be mapped to corresponding vectorial inner product in higher dimensional space, replace x[i] x[j];

Svm classifier training step is as follows:

A, training set is chosen as: T={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _l, y _l) ∈ (R ⁿ* Y) ^l, wherein, x _i∈ R ⁿbe a collection of mark safety or dangerous sensing data characteristic quantity corresponding amount, y _i∈ Y={-1,1}, i=1 ..., l; y _i=+1 represents it is to be about to bump, y _i=-1 represents the numerical value sign of normal travel condition; Corresponding i the moment of i, the characteristic quantity that each has the sensing sequence of the fixed number of continuous renewal to obtain constantly, each has decision content y constantly _iproduce, i and i+1 are the time interval constantly; Represent element number in training set, by l training sample, supported vector machine SVM, then does decision-making by SVM and judges;

B, selects suitable punishment parameters C >0 and kernel function K (x, x'), and its Kernel Function can be selected in several typical kernel functions any one;

C, constructs and solves convex quadratic programming problem: $\underset{a}{\min }\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{y}_i{y}_j{a}_i{a}_{j}K(x_i,x_j)-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i$

Meet

0≤α i≤C must separate

D, calculates b ^*: choose be positioned at open interval (0, the α in C) ^*component

calculate

$b^{*}=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j);$

E, structure decision function F (x)=sgn (g (x)), wherein $g\left(x\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j)+b^{*};$

After training completes, bring current sensor data characteristic quantity into decision function, according to decision function result of calculation, judge that obtaining present case is dangerous or safe judgement, if decision function be on the occasion of, represent safety, if decision function is negative value, represent to be about to occur collision.

Judge whether that the method that occurs obvious collision status is specially: according to sensor time sequence, obtain temporal signatures and frequency domain character, by temporal signatures and frequency domain character, through support vector machines, obtain decision-making again and judge, wherein the micro-silicon accelerometer of Main Basis and gyrostatic time series;

(1) extract micro-silicon accelerometer and gyrostatic time series feature, comprise temporal signatures and frequency domain character

A, the extraction of temporal signatures

Temporal signatures is the Time-domain Statistics characteristic obtaining according to various kinds of sensors output waveform, adopts following index:

Absolute mean: $\left|\stackrel{-}{X}\right|=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|$

Root-mean-square value: $X_{\mathrm{rms}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|}$

The degree of bias: $a=\sqrt{\frac{1}{6N}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^3$

In formula, x _ifor the discrete-time series of sensor signal, N is sequence samples number, and μ is average, and σ is variance;

B, the extraction of frequency domain character

Various kinds of sensors output sequence has typical frequecy characteristic separately, and it is carried out to spectrum analysis, obtains frequency domain distribution and corresponding amplitude;

(2) according to characteristic quantity, set up the basic model of the SVM of limit risk classification

If the characteristic set obtaining from sensor input is by two classes---be about to occur collision and occur that obviously collision forms, if feature x[i] correspondence belongs to the 1st class, y[i]=1, if feature x[i] correspondence belongs to the 2nd class, y[i]=-1, have so training sample set x[i], y[i] }, i=1,2,3, n, asks optimal classification face wx-b=0, when the situation for linearly inseparable, with core inner product K (x[i], x[j]), by kernel function, be mapped to corresponding vectorial inner product in higher dimensional space, replace x[i] x[j];

Svm classifier training step is as follows:

A, training set is chosen as: T={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _l, y _l) ∈ (R ⁿ* Y) ^l, wherein, x _i∈ R ⁿbe a collection of mark safety or dangerous sensing data characteristic quantity corresponding amount, y _i∈ Y={-1,1}, i=1 ..., l; y _i=+1 represents it is that obviously collision, y occur _i=-1 represents the numerical value sign that is about to bump; Corresponding i the moment of i, the characteristic quantity that each has the sensing sequence of the fixed number of continuous renewal to obtain constantly, each has decision content y constantly _iproduce, i and i+1 are the time interval constantly; L represents element number in training set, and by l training sample, supported vector machine SVM, then does decision-making by SVM and judge;

B, selects suitable punishment parameters C >0 and kernel function K (x, x'), and its Kernel Function can be selected in several typical kernel functions any one;

C, constructs and solves convex quadratic programming problem: $\underset{a}{\min }\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{y}_i{y}_j{a}_i{a}_{j}K(x_i,x_j)-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i$

Meet 0≤α i≤C must separate a*@@@aI*...aI*h@

D, calculates b ^*: choose be positioned at open interval (0, the α in C) ^*component

calculate

$b^{*}=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j);$

E, structure decision function F (x)=sgn (g (x)), wherein $g\left(x\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j)+b^{*};$

After training completes, bring current sensor data characteristic quantity into decision function, according to decision function result of calculation, judge that obtaining present case is dangerous or safe judgement, if decision function be on the occasion of, represent to be about to occur collision, if decision function is negative value, represent to occur obviously collision.

As shown from the above technical solution, the present invention obtains ranging information by sonac, air pressure probe, and infrared sensor and air pressure probe obtain and approach basis for estimation, by micro-silicon accelerometer and gyroscope, obtains unstability information; Range cells with approach judging unit and have circulation reciprocal process, ranging information and approach information reporting to master controller, by with empirical data set pair ratio, planning and the decision making algorithm of design will provide suspicious approaching thing number and approach distance in advance, and assign fill order, control airbag-releasing, and carry out audible alarm, by wireless messages transmitting element, predefined short message is sent on the mobile phone of appointment simultaneously, master controller judges status according to sensing data, and assigns corresponding protection and perform an action.The present invention adopts waistband design, can be worn on easily on human body, and volume is little, is easy to carry.

Accompanying drawing explanation

Fig. 1 is that the present invention is worn on human body, the wearing schematic diagram that air bag is not opened;

Fig. 2,3 is respectively structural representation of the present invention;

Fig. 4 is the view of airbag-releasing in Fig. 1;

Fig. 5 is collection of the present invention and carries out block diagram;

Fig. 6 is one-piece construction block diagram of the present invention;

Fig. 7 is workflow diagram of the present invention.

Embodiment

A kind of wearable human body anti-collision warning protective device, comprise waistband body 1, a plurality of air bags 2 are arranged at the interval, place, centre position of waistband body 1, on waistband body 1, be positioned at the upper of air bag 2, multielement bar group 3 is installed in below, micro-silicon accelerometer and gyroscope 4, alarm unit 5, multielement bar group 3, the output terminal of micro-silicon accelerometer and gyroscope 4 with by cell controller 7, be connected with the input end of master controller 8, the output terminal of master controller 8 by cell controller 7 with respectively with air bag 2, the input end of alarm unit 5 is connected, as shown in Figure 1, by multielement bar group 3 to be similar to the wearable arrangement of waistband, be placed on it human body, separated by a distance, place a kind of sensor, make the coverage of various sensors and air bag 2 can expand 360 degree to, realize omnibearing monitoring and protection.

As shown in Figure 2,3, 4, the number of described air bag 2 is four to six, and each air bag 2 is positioned on same center line.Four protective air-bags are independent Hu Bu UNICOM separately, is spacedly distributed on waistband body 1.Air bag 2 is arranged in to the top of sensor; human body waistband place; when air bag 2 is opened; can protect the trunk of human body to avoid collision; also use alarm unit and wireless messages transmitting element, when air bag 2 is opened soon, alarm unit is just started working simultaneously; meanwhile, wireless messages transmitting element sends information to predetermined mobile phone.

As Fig. 5, shown in 6, pass through sonac, air pressure probe obtains ranging information, by infrared sensor and air pressure probe, obtain and approach basis for estimation, by micro-silicon accelerometer and gyroscope 4, obtain unstability information, range cells and approach and have circulation reciprocal process between judging unit, ranging information and approach information reporting to master controller 8, by with empirical data set pair ratio, planning and the decision making algorithm of design will provide suspicious approaching thing number and approach distance in advance, and assign and carry out intention, according to current state and execution intention, to performance element, send concrete fill order, control air bag 2, alarm unit 5 and 6 actions of wireless messages transmitting element.Multielement bar group 3, together with human body intention signal, as the input of multielement bar data fusion, is differentiated state by multielement bar data fusion, then produces acoustics early warning, and airbag restraint is carried out and the protection of wireless transmission two-stage performs an action.

As shown in Figure 6, described multielement bar group 3 is comprised of sonac, air pressure probe and infrared sensor, described cell controller 7 by range cells, approach judging unit, unstability judging unit and performance element and form, on described waistband body 1, be positioned at air bag 2 above or below wireless messages transmitting element 6 is installed, the output terminal of sonac is connected with the input end of range cells, the output terminal of infrared sensor is connected with the input end that approaches judging unit, the output terminal of air pressure probe respectively with range cells, the input end that approaches judging unit is connected, the output terminal of micro-silicon accelerometer and gyroscope 4 is connected with the input end of unstability judging unit, the output terminal of described performance element respectively with air bag 2, alarm unit 5, the input end of wireless messages transmitting element 6 is connected, described range cells, approach judging unit, the output terminal of unstability judging unit is all connected with the input end of master controller 8, the output terminal of master controller 8 is connected with the input end of performance element.Described alarm unit 5 adopts loudspeakers, and audio alert " please notes and dodges " and with specific identifier sound.This protective device can be divided into three layers: ground floor is sensing and right of execution, the main sensing of being responsible for real time data, and according to condition discrimination result, carry out the execution of three large tasks; The second layer is cell controller 7; The 3rd layer is master controller 8.

As shown in Figure 7, during this device busy, its specific works flow process is as follows:

The first step, power-up initializing;

Second step, the information of 8 pairs of multielement bar groups 3 of master controller, micro-silicon accelerometer and gyroscope 4 outputs, acquiescence is in normal operating condition;

The 3rd step, the data sequence that 8 pairs of sensors of master controller collect is extracted characteristic quantity, and judges whether to be about to occur collision situation according to support vector machines, if the determination result is YES, starts alarm unit 5 and reports to the police, and continues Information Monitoring; Otherwise, return to second step;

The 4th step, after alarm unit 5 is reported to the police, master controller 8 judges whether to occur obvious collision status, if the determination result is YES, opens air bag 2, and starts wireless messages transmitting element 6 and carry out wireless transmission; If the determination result is NO, determine whether false-alarm.When false-alarm is judged, if the determination result is YES, give tacit consent in normal operating condition, otherwise, judge whether to be about to occur collision situation.

In other words, when approaching information and ranging information and meet empirical function or surpass to subscribe threshold value, system gets the hang of 2, when unstability judgement is set up, and extremely approaches after information and abnormal ranging information keep a period of time, and system gets the hang of 3.Wherein, carrying out one is acoustics alarm; Carrying out two is airbag-releasing and wireless transmission; State one is normal operation; State two is the alert status that are about to occur collision situation; State three is to have occurred obvious collision status.

Judge whether to be about to occur that the determination methods of collision situation is specially:

(1) extract each sensor characteristics of multielement bar group, comprise temporal signatures, frequency domain character and time trend feature

A, the extraction of temporal signatures

Temporal signatures is the Time-domain Statistics characteristic obtaining according to various kinds of sensors output waveform, adopts following index:

Absolute mean:

Root-mean-square value: $X_{\mathrm{rms}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|}$

The degree of bias: $a=\sqrt{\frac{1}{6N}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^3$

Kurtosis: $\mathrm{\β}=\sqrt{\frac{N}{24}}[\frac{1}{N}{\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^4-3]$

In formula, x _ifor the discrete-time series of sensor signal, N is sequence samples number, and μ is average, and σ is variance;

B, the extraction of frequency domain character

Various kinds of sensors output sequence has typical frequecy characteristic separately, and it is carried out to spectrum analysis, obtains amplitude-frequency sequence, extracts the range value in several particular frequencies in amplitude-frequency sequence;

C, the extraction of time trend feature

The original signal recording by sensor is definite status flag signal and the stack of random disturbance.The signal sequence that is provided with original measurement is { x _t, t=1,2 ..., N}, this sequence can be expressed as a trend term and random entry and:

x _t=d _t+ξ _t

In formula, d _tfor trend term, it is a certain definite function of time t; ξ _tfor random entry, it has reflected the random composition of signal;

When the trend of extraction feature, first need time series to carry out de-noising, remove random entry, retain trend term, and then utilize various trend analysis to extract feature;

Adopt Mann-Kendall check to detect sequence variation trend to sensor and extract, for time series { x _t, t=1,2 ..., N}, the null hypothesis H of Mann-Kendall check ₀for stochastic variable and time independence, compute statistics T

$T=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{T}_i$

In formula, at null hypothesis H ₀under, if there is not trend in time series, when N is larger, statistic T is similar to Normal Distribution, and has two formulas below to set up:

When n>10, calculate test statistics Z

If sequence calculates Z sometime---Sequence Trend characteristic quantity;

(2) according to characteristic quantity, set up the basic model of the SVM of limit risk classification

If the characteristic set obtaining from sensor input is by two classes---be about to occur collision and normally form, if feature x[i] correspondence belongs to the 1st class, y[i]=1, if feature x[i] correspondence belongs to the 2nd class, y[i]=-1, have so training sample set x[i], y[i] }, i=1,2,3 ..., n, ask optimal classification face wx-b=0, when the situation for linearly inseparable, use core inner product K (x[i], x[j]), by kernel function, be mapped to corresponding vectorial inner product in higher dimensional space, replace x[i] x[j];

Svm classifier training step is as follows:

A, training set is chosen as: T={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _l, y _l) ∈ (R ⁿ* Y) ^l, wherein, x _i∈ R ⁿbe a collection of mark safety or dangerous sensing data characteristic quantity corresponding amount, y _i∈ Y={-1,1}, i=1 ..., l; y _i=+1 represents it is to be about to bump, y _i=-1 represents the numerical value sign of normal travel condition; Corresponding i the moment of i, the characteristic quantity that each has the sensing sequence of the fixed number of continuous renewal to obtain constantly, each has decision content yi to produce constantly, and i and i+1 are the time interval constantly; L represents element number in training set, and by l training sample, supported vector machine SVM, then does decision-making by SVM and judge;

B, selects suitable punishment parameters C >0 and kernel function K (x, x'), and its Kernel Function can be selected in several typical kernel functions any one;

C, constructs and solves convex quadratic programming problem: $\underset{a}{\min }\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{y}_i{y}_j{a}_i{a}_{j}K(x_i,x_j)-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i$

Meet

0≤α i≤C must separate

D, calculates b ^*: choose be positioned at open interval (0, the α in C) ^*component

calculate

$b^{*}=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j);$

E, structure decision function F (x)=sgn (g (x)), wherein $g\left(x\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j)+b^{*};$

After training completes, bring current sensor data characteristic quantity into decision function, according to decision function result of calculation, judge that obtaining present case is dangerous or safe judgement, if decision function be on the occasion of, represent safety, if decision function is negative value, represent to be about to occur collision.

Judge whether that the method that occurs obvious collision status is specially: according to sensor time sequence, obtain temporal signatures and frequency domain character, by temporal signatures and frequency domain character, through support vector machines, obtain decision-making again and judge, wherein the micro-silicon accelerometer of Main Basis and gyrostatic time series;

(1) extract micro-silicon accelerometer and gyrostatic time series feature, comprise temporal signatures and frequency domain character

A, the extraction of temporal signatures

Temporal signatures is the Time-domain Statistics characteristic obtaining according to various kinds of sensors output waveform, adopts following index:

Absolute mean: $\left|\stackrel{-}{x}\right|=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|$

Root-mean-square value: $X_{\mathrm{rms}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|}$

The degree of bias: $a=\sqrt{\frac{1}{6N}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^3$

In formula, x _ifor the discrete-time series of sensor signal, N is sequence samples number, and μ is average, and σ is variance;

B, the extraction of frequency domain character

Various kinds of sensors output sequence has typical frequecy characteristic separately, and it is carried out to spectrum analysis, obtains frequency domain distribution and corresponding amplitude;

(2) according to characteristic quantity, set up the basic model of the SVM of limit risk classification

If the characteristic set obtaining from sensor input is by two classes---be about to occur collision and occur that obviously collision forms, if feature x[i] correspondence belongs to the 1st class, y[i]=1, if feature x[i] correspondence belongs to the 2nd class, y[i]=-1, have so training sample set x[i], y[i] }, i=1,2,3, n, asks optimal classification face wx-b=0, when the situation for linearly inseparable, with core inner product K (x[i], x[j]), by kernel function, be mapped to corresponding vectorial inner product in higher dimensional space, replace x[i] x[j];

Svm classifier training step is as follows:

A, training set is chosen as: T={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _l, y _l) ∈ (R ⁿ* Y) ^l, wherein, x _i∈ R ⁿbe a collection of mark safety or dangerous sensing data characteristic quantity corresponding amount, y _i∈ Y={-1,1}, i=1 ..., l; y _i=+1 represents it is that obviously collision, y occur _i=-1 represents the numerical value sign that is about to bump; Corresponding i the moment of i, the characteristic quantity that each has the sensing sequence of the fixed number of continuous renewal to obtain constantly, each has decision content y constantly _iproduce, i and i+1 are the time interval constantly; L represents element number in training set, and by l training sample, supported vector machine SVM, then does decision-making by SVM and judge;

B, selects suitable punishment parameters C >0 and kernel function K (x, x'), and its Kernel Function can be selected in several typical kernel functions any one;

C, constructs and solves convex quadratic programming problem: $\underset{a}{\min }\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{y}_i{y}_j{a}_i{a}_{j}K(x_i,x_j)-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i$

Meet

0≤α i≤C must separate

D, calculates b ^*: choose be positioned at open interval (0, the α in C) ^*component

calculate

$b^{*}=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j);$

E, structure decision function F (x)=sgn (g (x)), wherein $g\left(x\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j)+b^{*};$

After training completes, bring current sensor data characteristic quantity into decision function, according to decision function result of calculation, judge that obtaining present case is dangerous or safe judgement, if decision function be on the occasion of, represent to be about to occur collision, if decision function is negative value, represent to occur obviously collision

In a word, the present invention obtains ranging information by sonac, air pressure probe, and infrared sensor and air pressure probe obtain and approach basis for estimation, by micro-silicon accelerometer and gyroscope 4, obtains unstability information; Range cells with approach judging unit and have circulation reciprocal process, ranging information and approach information reporting to master controller 8, by with empirical data set pair ratio, planning and the decision making algorithm of design will provide suspicious approaching thing number and approach distance in advance, and assign fill order, controlling air bag 2 opens, and carry out audible alarm, by wireless messages transmitting element, predefined short message is sent on the mobile phone of appointment, two kinds of protection perform an action and according to field condition, judge decision-making by master controller simultaneously.The present invention adopts waistband design, can be worn on easily on human body, and volume is little, is easy to carry.

Claims (9)

1. a wearable human body anti-collision warning protective device, it is characterized in that: comprise waistband body, a plurality of air bags are arranged at the interval, place, centre position of waistband body, the upper and lower that is positioned at air bag on waistband body is installed multielement bar group, micro-silicon accelerometer and gyroscope, alarm unit, multielement bar group, micro-silicon accelerometer and gyrostatic output terminal with by cell controller, be connected with the input end of master controller, the output terminal of master controller by cell controller be connected with the input end of air bag, alarm unit respectively.

2. wearable human body anti-collision warning protective device according to claim 1, it is characterized in that: described multielement bar group is comprised of sonac, air pressure probe and infrared sensor, described cell controller by range cells, approach judging unit, unstability judging unit and performance element and form, on described waistband body, be positioned at air bag above or below wireless messages transmitting element is installed; The output terminal of sonac is connected with the input end of range cells, the output terminal of infrared sensor is connected with the input end that approaches judging unit, the output terminal of air pressure probe is connected with range cells, the input end that approaches judging unit respectively, micro-silicon accelerometer is connected with the input end of unstability judging unit with gyrostatic output terminal, and the output terminal of described performance element is connected with the input end of air bag, alarm unit, wireless messages transmitting element respectively.

3. wearable human body anti-collision warning protective device according to claim 1, is characterized in that: described alarm unit adopts loudspeaker.

4. wearable human body anti-collision warning protective device according to claim 1, is characterized in that: the number of described air bag is four to six, and each air bag is positioned on same center line.

5. wearable human body anti-collision warning protective device according to claim 2, it is characterized in that: described range cells, the output terminal that approaches judging unit, unstability judging unit are all connected with the input end of master controller, and the output terminal of master controller is connected with the input end of performance element.

6. according to the early warning protection method of the wearable human body anti-collision warning protective device described in any one in claim 1 to 5, the method comprises the step of following order:

(1) power-up initializing;

(2) information of master controller to multielement bar group, micro-silicon accelerometer and gyroscope output, acquiescence is in normal operating condition;

(3) data sequence that master controller collects sensor is extracted characteristic quantity, and judges whether to be about to occur collision situation according to support vector machines, if the determination result is YES, starts alarm unit and reports to the police, and continues Information Monitoring; Otherwise, return to step (2);

(4) after alarm unit is reported to the police, master controller judges whether to occur obvious collision status, if the determination result is YES, opens air bag, and starts wireless messages transmitting element and carry out wireless transmission; If the determination result is NO, determine whether false-alarm.

7. the early warning protection method of wearable human body anti-collision warning protective device according to claim 6, it is characterized in that: when false-alarm is judged, if the determination result is YES, give tacit consent in normal operating condition, otherwise, judge whether to be about to occur collision situation.

8. the early warning protection method of wearable human body anti-collision warning protective device according to claim 6, is characterized in that: judge whether to be about to occur that the determination methods of collision situation is specially:

(1) extract each sensor characteristics of multielement bar group, comprise temporal signatures, frequency domain character and time trend feature

A, the extraction of temporal signatures

Temporal signatures is the Time-domain Statistics characteristic obtaining according to various kinds of sensors output waveform, adopts following index:

Absolute mean: $\stackrel{-}{X}\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|$

Root-mean-square value: $X_{\mathrm{rms}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|}$

The degree of bias: $a=\sqrt{\frac{1}{6N}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^3$

Kurtosis: $\mathrm{\β}=\sqrt{\frac{N}{24}}[\frac{1}{N}{\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^4-3]$

In formula, x _ifor the discrete-time series of sensor signal, N is sequence samples number, and μ is average, and σ is variance;

B, the extraction of frequency domain character

Various kinds of sensors output sequence has typical frequecy characteristic separately, and it is carried out to spectrum analysis, obtains amplitude-frequency sequence, extracts the range value in several particular frequencies in amplitude-frequency sequence;

C, the extraction of time trend feature

The original signal recording by sensor is definite status flag signal and the stack of random disturbance.The signal sequence that is provided with original measurement is { x _t, t=1,2 ..., N}, this sequence can be expressed as a trend term and random entry and:

x _t=d _t+ξ _t

In formula, d _tfor trend term, it is a certain definite function of time t; ξ _tfor random entry, it has reflected the random composition of signal;

When the trend of extraction feature, first need time series to carry out de-noising, remove random entry, retain trend term, and then utilize various trend analysis to extract feature;

Adopt Mann-Kendall check to detect sequence variation trend to sensor and extract, for time series { x _t, t=1,2 ..., N}, the null hypothesis H of Mann-Kendall check ₀for stochastic variable and time independence, compute statistics T

$T=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{T}_i$

In formula,

at null hypothesis H ₀under, if there is not trend in time series, when N is larger, statistic T is similar to Normal Distribution, and has two formulas below to set up:

When n>10, calculate test statistics Z

If sequence calculates Z sometime---Sequence Trend characteristic quantity;

(2) according to characteristic quantity, set up the basic model of the SVM of limit risk classification

If the characteristic set obtaining from sensor input is by two classes---be about to occur collision and normally form, if feature x[i] correspondence belongs to the 1st class, y[i]=1, if feature x[i] correspondence belongs to the 2nd class, y[i]=-1, have so training sample set x[i], y[i] }, i=1,2,3 ..., n, ask optimal classification face wx-b=0, when the situation for linearly inseparable, use core inner product K (x[i], x[j]), by kernel function, be mapped to corresponding vectorial inner product in higher dimensional space, replace x[i] x[j];

Svm classifier training step is as follows:

A, training set is chosen as: T={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _l, y _l) ∈ (R ⁿ* Y) ^l, wherein, x _i∈ R ⁿbe a collection of mark safety or dangerous sensing data characteristic quantity corresponding amount, y _i∈ Y={-1,1}, i=1 ..., l; y _i=+1 represents it is to be about to bump, y _i=-1 represents the numerical value sign of normal travel condition; Corresponding i the moment of i, the characteristic quantity that each has the sensing sequence of the fixed number of continuous renewal to obtain constantly, each has decision content y constantly _iproduce, i and i+1 are the time interval constantly; L represents element number in training set, and by l training sample, supported vector machine SVM, then does decision-making by SVM and judge;

B, selects suitable punishment parameters C >0 and kernel function K (x, x'), and its Kernel Function can be selected in several typical kernel functions any one;

C, constructs and solves convex quadratic programming problem: $\underset{a}{\min }\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}{y}_i{y}_j{a}_i{a}_{j}K(x_i,x_j)-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i$

Meet

0≤α _i≤ C must separate

D, calculates b ^*: choose be positioned at open interval (0, the α in C) ^*component

calculate

$b^{*}=y_j-\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j);$

E, structure decision function F (x)=sgn (g (x)), wherein $g\left(x\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{a}_i^{*}{y}_{i}K(x_i,x_j)+b^{*};$

After training completes, bring current sensor data characteristic quantity into decision function, according to decision function result of calculation, judge that obtaining present case is dangerous or safe judgement, if decision function be on the occasion of, represent safety, if decision function is negative value, represent to be about to occur collision.

9. the early warning protection method of wearable human body anti-collision warning protective device according to claim 6, it is characterized in that: judge whether that the method that occurs obvious collision status is specially: according to sensor time sequence, obtain temporal signatures and frequency domain character, by temporal signatures and frequency domain character, through support vector machines, obtain decision-making again and judge, wherein the micro-silicon accelerometer of Main Basis and gyrostatic time series;

(1) extract micro-silicon accelerometer and gyrostatic time series feature, comprise temporal signatures and frequency domain character

A, the extraction of temporal signatures

Temporal signatures is the Time-domain Statistics characteristic obtaining according to various kinds of sensors output waveform, adopts following index:

Absolute mean: $\left|\stackrel{-}{x}\right|=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|$

Root-mean-square value: $X_{\mathrm{rms}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|x_i\right|}$

The degree of bias: $a=\sqrt{\frac{1}{6N}}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}\right)}^3$

In formula, x _ifor the discrete-time series of sensor signal, N is sequence samples number, and μ is average, and σ is variance;

B, the extraction of frequency domain character

Various kinds of sensors output sequence has typical frequecy characteristic separately, and it is carried out to spectrum analysis, obtains frequency domain distribution and corresponding amplitude;

(2) according to characteristic quantity, set up the basic model of the SVM of limit risk classification

If the characteristic set obtaining from sensor input is by two classes---be about to occur collision and occur that obviously collision forms, if feature x[i] correspondence belongs to the 1st class, y[i]=1, if feature x[i] correspondence belongs to the 2nd class, y[i]=-1, have so training sample set x[i], y[i] }, i=1,2,3, n, asks optimal classification face wx-b=0, when the situation for linearly inseparable, with core inner product K (x[i], x[j]), by kernel function, be mapped to corresponding vectorial inner product in higher dimensional space, replace x[i] x[j];

Svm classifier training step is as follows:

A, training set is chosen as: T={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _l, y _l) ∈ (R ⁿ* Y) ^l, wherein, x _i∈ R ⁿbe a collection of mark safety or dangerous sensing data characteristic quantity corresponding amount, y _i∈ Y={-1,1}, i=1 ..., l; y _i=+1 represents it is that obviously collision, y occur _i=-1 represents the numerical value sign that is about to bump; Corresponding i the moment of i, the characteristic quantity that each has the sensing sequence of the fixed number of continuous renewal to obtain constantly, each has decision content y constantly _iproduce, i and i+1 are the time interval constantly; L represents element number in training set, and by l training sample, supported vector machine SVM, then does decision-making by SVM and judge;

B, selects suitable punishment parameters C >0 and kernel function K (x, x'), and its Kernel Function can be selected in several typical kernel functions any one;

C, constructs and solves convex quadratic programming problem:

Meet 0≤α _i≤ C must separate

D, calculates b ^*: choose be positioned at open interval (0, the α in C) ^*component calculate

E, structure decision function F (x)=sgn (g (x)), wherein

After training completes, bring current sensor data characteristic quantity into decision function, according to decision function result of calculation, judge that obtaining present case is dangerous or safe judgement, if decision function be on the occasion of, represent to be about to occur collision, if decision function is negative value, represent to occur obviously collision.

Images