CN106529433A - Queue pace uniformity evaluation method based on sound signals - Google Patents

Abstract

The present invention provides a brand new digital and automatic evaluation mode for evaluating the uniformity of the troop review queues by analyzing and processing the footstep sound signals of the troop review queues to obtain the uniformity of the queue paces. The technical scheme adopted by the present invention is that the queue pace uniformity evaluation method based on the sound signals comprises 1) a sound acquisition and noise reduction step; 2) a pace extraction step of extracting the pace coordinates to determine the concrete position of each pace in the sound signals; 3) a pace uniformity analysis step characterized in that an S curve of the ideal standard uniform pace sound signals is constructed, the correlation analysis is carried out on the S curve and an accumulated energy curve of the actual sound signals, the larger is a correlation coefficient, the more similar are the actual footstep signals and the standard uniform footstep sound signals, namely, the actual footstep signals are more uniform. The queue pace uniformity evaluation method based on the sound signals of the present invention is mainly applied to the analysis and processing of the footstep sound signals.

Description

Queue march in step degree evaluation methodology based on acoustical signal

Technical field

The present invention evaluates the march in step journey of queue of reviewing troops by process is analyzed to queue step acoustical signal of reviewing troops Degree, so as to corresponding judgement is made to the concordance of team posture.Formation, the standard of queue inspection and achievement is made to embody more Plus directly perceived and science.Concretely relate to the queue march in step degree evaluation methodology based on acoustical signal.

Background technology

Formation, no matter in colleges and universities' military training or in military training, is all the exercise sense of organizational discipline, steel one's will product The important way of matter, is team training mode generally accepted in China.And the neat degree of queue paces is even more queue instruction Practice the concentrated reflection of effect.At present, for queue march in step degree, still there is no a kind of scientific and effective evaluation method.Especially It is for queue is reviewed troops the evaluation of effect, substantially by the way of being evaluated by reviewer's subjective feeling.Instruct in queue There is no the guidance of science during white silk, fully rely on mechanization repetition training to produce a desired effect yet.Digitized with Today of information-based high speed development, this seems relatively backward unavoidably.For existing technical thought, can be used for carrying out queue The Digital evaluation mode of regularity detection mainly has the moving object identification based on computer vision and follows the trail of, and based on biography The action statistics of sensor network compares such two kinds of Research Thinkings.

Computer vision is obtaining the technology with processing information by the anthropomorphic vision mechanism of computer mould.It is mainly The image of human motion is recorded, and detection, extraction, tracking people a series of actions to people are analyzed and retouch from image State.Based on the solution of sensor network, it is recognizing by the acceleration signal for gathering and producing when processing human motion Different human actions, recognizes position of human body by gathering ground vibration signal.Then its diversity and dependency are calculated.

Time cost, resources costss required for the exploitation of both the above method is all of a relatively high, and difficulty is larger.So this It is bright to adopt the thinking being analyzed by queue acoustical signal.Language is mainly reflected in based on the research of acoustical signal both at home and abroad at present The field such as sound identification and process.Foreign countries defined a series of the theoretical of the Digital Signal Processing of maturations and calculated the sixties in 20 century Method, such as digital filter etc..Research in terms of for sound regularity is provided substantial amounts of reference by this.Meanwhile, also permitted both at home and abroad The research of the aspects such as many automobile models based on acoustical signal are recognized, animal species judge, mechanical disorder detection.But these grind The popular feature studied carefully is all based on frequency domain and signal is analyzed.And for the step acoustical signal of queue of reviewing troops, neat person and The footsteps that irregular person sends is almost the same on frequency domain, it is difficult to distinguish, it is therefore desirable to signal is entered from time domain angle Row analysis.

In modern medical service clinical diagnosises, there is a kind of clinical diagnosis technology based on analysis of PCG Signal.This technology profit Time-frequency distributions model with wavelet transformation etc. carries out time-frequency combination analysis to heartbeat sound.This is believed by sound for the present invention Number the neat degree of queue paces is carried out evaluating there is provided inspiring.

List of references：

Kong Fei. the Detection And Tracking [D] based on personnel's step signal of multichannel sensor. Nanjing：Institutes Of Technology Of Nanjing, 2006.

Zhang Guoliang, Wang Zhanni, Wang Tian etc. dynamic hand gesture recognition summary [J] of Appliance computer vision. Huaqiao University's journal (natural science edition), 2014, (6)：653-658.

Guo Juntao, Lin Sencai, Tian Xiaodong etc. the detection of cardiechema signals and process [J]. modern instrument, 2008,14 (2)： 36-38.

Li Weizhi. the design method [J] based on Underwater Acoustic channels. scientific and technological information：Scholarly edition, 2008, (17).

The content of the invention

To overcome the deficiencies in the prior art, it is contemplated that proposition is drawn by being analyzed to queue step acoustical signal The neat degree of queue paces, the regularity evaluation for queue of reviewing troops provide a kind of brand-new digitized, Assessment mode. The technical solution used in the present invention is that, based on the queue march in step degree evaluation methodology of acoustical signal, step is as follows：

1) collection of sound and noise reduction

First, from hardware point of view, using the pick up with directivity, and according to the propagation characteristic section of live sound The angles and position for choosing pick up are learned, the step acoustical signal of high s/n ratio as far as possible is gathered with this；Tied using time-frequency The nonlinear filtering noise reduction based on wavelet transformation for closing；

2) extraction of paces

Acoustical signal first to collecting carries out zero-crossing rate process, and zero-crossing rate ZCR (zero-crossing rate) is Refer to the ratio of the sign change of a signal, the formula of zero-crossing rate is：

Wherein, Z_nFor n-th point of zero-crossing rate, N-1 is that zero passage window is long, X_nM () is m-th point after nth point of amplitude；

Sgn [X] is sign function, for whether judging the unknown quantity X in bracket more than 0, i.e.,：

By the extraction to paces coordinate, particular location of each paces in acoustical signal is determined；

3) march in step degree analysis

By the S curve for building the neat paces acoustical signal of ideal standard so as to the cumlative energy with actual sound signal Curve carries out correlation analysiss, and correlation coefficient is bigger, illustrates that actual step signal is more close with the neat step acoustical signal of standard, I.e. actual step signal is more neat.

The construction step of S curve is to need to obtain the fundamental formular and relevant parameter of logic distribution, its distribution function F (x) Fundamental formular with probability density f (x) is：

Wherein

Above-mentioned distribution function F (x) be thus referred to as population growth curve, i.e. S curve, independent variable x represent population presence when Between sequence, s is slope of the curve at μ values, and μ values are the corresponding time point of S curve maximum slope, and μ values occur in energy sum At 1/2.

The cumlative energy for calculating actual step signal is：

Wherein E (i) corresponding energy at i-th point for signal；

According to the theoretical basiss of logic distribution, if the integral distribution curve of its corresponding preferable s curve is：

Wherein L is the cumlative energy of actual waveform with i represents i-th time point, and μ is reached for actual waveform cumlative energy Time point during L/2, s is the key of whole formula, and system is by trying to achieve the s of the point of actual waveform cumlative energy 40% to 60% It is averaged after value, with the approximate s values as preferable S curve of the meansigma methodss tried to achieve.

Using Pearson correlation coefficients (Pearson correlation coefficient) to two curve S₁And S₂Enter Row correlation test, Pearson correlation coefficients computing formula are as follows：

Wherein ρ_{X, Y}Represent the correlation coefficient of two variables of X and Y, molecule cov (X, Y) is the covariance of X and Y, denominator σ X σ Y It is the product of two variable standard deviations, E is to expect, it is clear that the standard deviation for requiring X and Y can not all be 0, it is linear when two variables When relation strengthens, correlation coefficient tends to 1 or -1.Tend to 1 during positive correlation, when negatively correlated, tend to -1, when two variable independence phases Relation number is 0, according to correlation test principle, takes the cumlative energy distribution waveform array that X and Y is respectively original sound signal Neat degree is represented with by corresponding ideal S curve array, the correlation coefficient for calculating.

Wavelet transformation is boir4_4 small echos.

The characteristics of of the invention and beneficial effect are：

Evaluation for queue march in step degree of the invention provides the evaluation side of a kind of brand-new digitized, automatization Formula.Avoid the problem forbidden by the evaluation result caused because of anthropic factors such as individual cognition differences.Can greatly increase simultaneously Formation and review troops scene it is scientific and interesting.The method can be widely applied to the formation of army or colleges and universities and Review troops in joint performance.

Description of the drawings：

Fig. 1 original sound signals.

Signal after Fig. 2 wavelet de-noisings.

Fig. 3 zero-crossing rates.

The energy accumulation distribution curve of mono- step acoustical signals of Fig. 4.

Fig. 5 population growth s curves.

Fig. 6 logic distribution probability density curves and integral distribution curve.

The cumlative energy curve of Fig. 7 actual waveforms.

Fig. 8 approximate ideal S curves.

Fig. 9 invention flow charts.

Specific embodiment

It is obvious that step sound of the neat queue when advancing be also neat, concentrate, it is rhythmical, foot landing and The hanging time substantially can be distinguished.And the relatively low queue of regularity, the step sound in traveling process be then it is mixed and disorderly, Scattered, rhythm is relatively low, and the time of foot landing is that dispersion is irregular.Therefore by further analyses and comparison, can Substantially distinguished with the neat degree to queue step sound.

This programme will be made up of following step:The collection of sound and noise reduction, the extraction of paces and regularity point Analysis.

1. the collection of sound and noise reduction

It is to obtain clear pure step acoustical signal to the premise that queue step sound is analyzed.First, from hardware Angle is set out, and adopts with the pick up compared with high directivity, and chooses pick up according to the propagation characteristic science of live sound Angles and position, gather the step acoustical signal of high s/n ratio as far as possible with this.The step sound that Fig. 1 is arrived for actual acquisition Signal is original sound signal, then further carries out noise reduction from software respective to which.Due to queue review troops scene sound into Divide complexity, except echo signal is footsteps message extra, also there are musical sound, personnel's sound of speech and other various complexity Natural noise.It is and noise signal and step signal have on frequency domain, traditional based on impact The noise reduction mode effect of response filter is very undesirable.

The nonlinear filtering noise reduction mode based on wavelet transformation that the present invention is combined using a kind of time-frequency.Wavelet de-noising mode Obtain commonly used in the case of the time-frequency binding analysis that frequency is overlapped.According to wavelet theory, for unstable letter Number, the shake of signal can be effectively filtered using wavelet analysises.By the time subdivision to high-frequency region, the frequency of low frequency region Rate is segmented, interference component and key signaling components that system can effectively in resoluting signal, makes next step computing more accurate Property.Tested by great amount of samples, we have selected boir4_4 small echos, and this small echo is biorthogonal compactly supported wavelet, with linear Phase place, and had the application of maturation in JPEG2000 standards.Meanwhile, this wavelet shapes and time domain voice signal have necessarily Degree of fitting, can realize preferable noise reduction.Fig. 2 is signal of the actual signal after wavelet de-noising, it can be seen that scene Noise composition is substantially filtered.

2. the extraction of paces

Need to extract paces information after noise reduction process is carried out to signal, individually to analyze each step. Acoustical signal of the present invention first to collecting carries out zero-crossing rate process, and zero-crossing rate (zero-crossing rate, ZCR) is referred to The ratio of the sign change of one signal, such as signal become negative or reverse from positive number.This feature is in Speech comparison, voice Identification and music information retrieval field are used widely, and are to tapping the principal character classified by sound.It is usually used in analysis The rhythmic feature of one section of sound.The formula of zero-crossing rate is：

Wherein, Z_nFor n-th point of zero-crossing rate, N-1 is that zero passage window is long, X_nM () is m-th point after nth point of amplitude.

Sgn [X] is sign function, for whether judging the unknown quantity X in bracket more than 0, i.e.,：

By the calculating of zero-crossing rate, the interval in multiple paces can be separated with single paces.I.e. in compartment, mistake The value of zero rate is always zero；In the single paces stage, comparatively dense is accordingly compared in the distribution of zero-crossing rate.After being zero-crossing rate process such as Fig. 3 Sound waveform.By the extraction to paces coordinate, you can to determine particular location of each paces in acoustical signal.

3. march in step degree analysis

Paces need individually to carry out regularity analysis to each step after the completion of extracting.Such as Fig. 1, in step sound signal waveform In, where substantially can distinguish is the sound for stepping a step, where is the interval between two steps, and this is due in wave-shape amplitude Difference, that is, the difference in intensity of sound.When ignoring the factors such as sound reflection, sound superposition, and everyone footsteps During the difference of sound, the size for thinking intensity of sound that can be approximate be proportional to the number that produces this step sound number.Time domain On signal waveform in, the intensity distributions of sound can represent when stepping same step not the number of foot landing in the same time, that is, This queue member steps the neat degree of this step.Based on assumed above, the intensity distributions of step acoustical signal are carried out point Analysis.

Take out a step therein and do cumlative energy distribution curve, obtain curve as shown in Figure 4, send out through mass data analysis It is existing, the curve tendency approximate Logic profile accumulation energy curve.Namely in nature generally existing population growth curve. In natural environment, the growth curve of population is " S " type curve (also referred to as logistic curve), such as Fig. 5.Population reaches ring The laden weight or load capacity of the maximum that border can be born, referred to as environment, is represented with " K ".Population quantity when K/2 is reached, Population quantity linearly rises, and this period claims exponential phase of growth, and this point is the key point for affecting population quantity.If The thinking for being distributed S curve with logic analyzes the problem of step sound, it is found that ignore the characteristics such as sound transmission, diverging, reflection Impact, the cumlative energy of the step acoustical signal that a step is neatly concentrated is gradually increasing at the beginning, connects to foot and ground During tactile key point, acoustic energy rate of increase highest, as foot leaves ground, acoustic energy rate of increase is reduced to zero, accumulation energy Amount reaches maximum.Therefore, step acoustical signal is similar S curve in Energy distribution.

Therefore can pass through to build the S curve of the neat paces acoustical signal of ideal standard so as to actual sound signal Cumlative energy curve carries out correlation analysiss, and correlation coefficient is bigger, illustrates actual step signal and the neat footsteps message of standard Number more close, i.e., actual step signal is more neat.Therefore the degree of relevancy of the two just embodies paces to a certain extent Neat degree.

Want to build preferable S curve, need to obtain the fundamental formular and relevant parameter of logic distribution.Logic is distributed It is that the fundamental formular of the mathematical model of a comparative maturity, its distribution function F (x) and probability density f (x) is：

Wherein

Above-mentioned distribution function F (x) be thus referred to as population growth curve, i.e. S curve, independent variable x represent population presence when Between sequence, s is slope of the curve at μ values, and μ values are the corresponding time point of S curve maximum slope, and μ values occur in energy sum At 1/2.

Fig. 6 is corresponding probability density curve and integral distribution curve.

The cumlative energy for calculating actual step signal is：

Wherein E (i) corresponding energy at i-th point of time for signal.

According to the theoretical basiss of logic distribution, if the integral distribution curve of its corresponding preferable s curve is：

Wherein L is the cumlative energy of actual waveform with i represents i-th time point, and μ is reached for actual waveform cumlative energy Time point during L/2, s is the key of whole formula, and system is by trying to achieve the s of the point of actual waveform cumlative energy 40% to 60% It is averaged after value, with the approximate s values as preferable S curve of the meansigma methodss tried to achieve.String acoustical signal is carried out by the method Calculate, the cumlative energy curve and approximate ideal S curve of actual waveform can be obtained, such as Fig. 7, Fig. 8 is the figure of wherein two steps Picture.

Finally, the preferable S curve corresponding to the integrated energy curve of actual waveform carries out correlation test.It is corresponding Correlation coefficient can represent the neat degree of the step acoustical signal.

Conventional Pearson correlation coefficients (Pearson correlation coefficient) carry out correlation test.Skin Ademilson correlation coefficient is also Pearson's product moment correlation of coefficient, is the statistic for reacting two variable similarity degrees.In other words Can be used to calculate two vectorial similarities.Pearson correlation coefficients computing formula is as follows：

Wherein ρ_{X, Y}Represent the correlation coefficient of two variables of X and Y, molecule cov (X, Y) is the covariance of X and Y, denominator σ X σ Y It is the product of two variable standard deviations, E is to expect.Obviously require that the standard deviation of X and Y can not all be 0.It is linear when two variables When relation strengthens, correlation coefficient tends to 1 or -1.Tend to 1 during positive correlation, when negatively correlated, tend to -1.When two variable independence phases Relation number is 0.According to correlation test principle, the cumlative energy distribution waveform array that X and Y is respectively original sound signal is taken Neat degree is represented with by corresponding ideal S curve array, the correlation coefficient for calculating.

Realize that the flow chart of the present invention is overall as shown in Figure 9.According to the thinking of technical scheme, first from hardware point of view, lead to The professional recording microphone of heart-shaped sensing is crossed, the step audio direction of queue of reviewing troops is directed at, is collected the sound of high s/n ratio as far as possible Message number.Next, process can be analyzed to the acoustical signal for collecting by softwares such as labview or matlab.Letter Number process is mainly made up of following step：The noise reduction of primary signal, the extraction of single paces, the regularity analysis of paces. According to above-mentioned theory basis, signal is carried out after respective handling, you can obtain the neat degree of step acoustical signal for collecting.

Claims (4)

1. a kind of queue march in step degree evaluation methodology based on acoustical signal, is characterized in that, step is as follows：

1) collection of sound and noise reduction

First, from hardware point of view, using the pick up with directivity, and selected according to the propagation characteristic science of live sound Angles and the position of pick up are taken, the step acoustical signal of high s/n ratio as far as possible is gathered with this；Combined using time-frequency Nonlinear filtering noise reduction based on wavelet transformation；

2) extraction of paces

Acoustical signal first to collecting carries out zero-crossing rate process, and zero-crossing rate ZCR (zero-crossing rate) refers to one The ratio of the sign change of individual signal, the formula of zero-crossing rate is：

$Z_n=\frac{1}{2}\underset{m=0}{\overset{N-1}{\Σ}}|\mathrm{sgn}\[X_n\left(m\right)\]-\mathrm{sgn}\[X_n(m-1)\]|$

Wherein, Z_nFor n-th point of zero-crossing rate, N-1 is that zero passage window is long, X_nM () is m-th point after nth point of amplitude；

Sgn [X] is sign function, for whether judging the unknown quantity X in bracket more than 0, i.e.,：

$\mathrm{sgn}\&lsqb;X\&rsqb;=\left\{\begin{array}{cc}1,& (X>0)\\ -1,& (X<0)\end{array}\right.$

By the extraction to paces coordinate, particular location of each paces in acoustical signal is determined；

3) march in step degree analysis

By the S curve for building the neat paces acoustical signal of ideal standard so as to the cumlative energy curve with actual sound signal Correlation analysiss are carried out, correlation coefficient is bigger, illustrate that actual step signal is more close with the neat step acoustical signal of standard, i.e., in fact Border step signal is more neat.

2. the queue march in step degree evaluation methodology based on acoustical signal as claimed in claim 1, is characterized in that, S curve Construction step is to need to obtain the fundamental formular and relevant parameter, its distribution function F (x) and probability density f (x) of logic distribution Fundamental formular be：

$F\left(x\right)=\frac{1}{1+e^{-\frac{x-\mathrm{\&mu;}}{s}}}$

$f\left(x\right)=\frac{e^{\frac{-(x-\mathrm{\&mu;})}{s}}}{s{(1+e^{-\frac{x-\mathrm{\&mu;}}{s}})}^2}=\frac{1}{4s}{\mathrm{sech}}^2\left(\frac{x-\mathrm{\&mu;}}{2s}\right)$

Wherein

$\sec hx=\frac{2}{e^{x+}{e}^{-x}}$

Above-mentioned distribution function F (x) is thus referred to as population growth curve, i.e. S curve, and independent variable x represents the time sequence of population presence Row, s are slope of the curve at μ values, and μ values are the corresponding time point of S curve maximum slope, and μ values occur in the 1/2 of energy sum Place.

The cumlative energy for calculating actual step signal is：

$S_1=\underset{i=0}{\overset{n-1}{\&Sigma;}}E\left(i\right)$

Wherein E (i) corresponding energy at i-th point for signal；

According to the theoretical basiss of logic distribution, if the integral distribution curve of its corresponding preferable s curve is：

$S_2=\frac{L}{1+e^{-\frac{i-\mathrm{\&mu;}}{s}}}$

Wherein L is the cumlative energy of actual waveform with i represents i-th time point, when μ reaches L/2 for actual waveform cumlative energy Time point, s is the key of whole formula, and system is after trying to achieve the s values of point of actual waveform cumlative energy 40% to 60% It is averaged, with the approximate s values as preferable S curve of the meansigma methodss tried to achieve.

3. the queue march in step degree evaluation methodology based on acoustical signal as claimed in claim 1, is characterized in that, using Pierre Gloomy correlation coefficient (Pearson correlation coefficient) is to two curve S₁And S₂Carry out correlation test, Pierre Gloomy correlation coefficient computing formula is as follows：

$\begin{array}{c}{\mathrm{\&rho;}}_{X,Y}=\frac{\mathrm{cov}(X,Y)}{\mathrm{\&sigma;}X\mathrm{\&sigma;}Y}=\frac{E\left(\right(X-\mathrm{\&mu;}X)(Y-\mathrm{\&mu;}Y))}{\mathrm{\&sigma;}X\mathrm{\&sigma;}Y}\\ =\frac{E\left(XY\right)-E\left(X\right)E\left(Y\right)}{\sqrt{E\left(X^2\right)-E^2\left(X\right)}\sqrt{E\left(Y^2\right)-E^2\left(Y\right)}}=\frac{n\mathrm{\&Sigma;}\left(XY\right)-\mathrm{\&Sigma;}\left(X\right)\mathrm{\&Sigma;}\left(Y\right)}{\sqrt{n\mathrm{\&Sigma;}\left(X^2\right)-{\mathrm{\&Sigma;}}^2\left(X\right)}\sqrt{n\mathrm{\&Sigma;}\left(Y^2\right)-{\mathrm{\&Sigma;}}^2\left(Y\right)}}\end{array}$

Wherein ρ_{X, Y}The correlation coefficient of two variables of X and Y is represented, molecule cov (X, Y) is the covariance of X and Y, and denominator σ X σ Y are two The product of individual variable standard deviation, E are to expect, it is clear that require that the standard deviation of X and Y can not all be 0, when the linear relationship of two variables During enhancing, correlation coefficient tends to 1 or -1.Tend to 1 during positive correlation, when negatively correlated, tend to -1, the phase relation when two variable independence Number according to correlation test principle, takes X and Y and is respectively the cumlative energy distribution waveform array of original sound signal and right for 0 Neat degree is represented by answering preferable S curve array, the correlation coefficient for calculating.

4. the queue march in step degree evaluation methodology based on acoustical signal as claimed in claim 1, is characterized in that, wavelet transformation For boir4_4 small echos.