Shulou(Shulou.com)06/01 Report--

The editor will share with you how to use matlab MFCC+GMM 's security incident acoustic detection system. I hope you will get something after reading this article. Let's discuss it together.

A brief introduction to Acoustic Detection of Safety incidents (with lunwen) 1 background of the selected topic

The issue of public security is one of the focused topics of social security and stability. In recent years, detection technology and monitoring automation are profoundly changing people's lives. Especially in the field of security, closed-circuit television CCTV (Closed Circuit Television), video stream analysis, intelligent monitoring and other new technologies have been widely used, which greatly improve the management efficiency of security monitoring. However, it is worth noting that the monitoring methods based on video streams inevitably have some congenital defects, such as blind areas of vision, vulnerable to lighting and other problems, as well as unclear semantics for event detection. The means of monitoring are not comprehensive enough. Pure video means are often not as sensitive and effective as acoustic detection and analysis methods in shooting, explosion, violent terrorist attacks, crowd panic and other semantic public security emergencies. Acoustic event detection mainly uses some acoustic processing methods to depict the acoustic characteristics of the field audio stream, and then combined with the appropriate classifier for detection and classification, so as to achieve the detection and analysis of acoustic events in the audio stream. Public safety incident detection based on acoustics has extensive application value and application prospect in many fields, such as anti-terrorism, maintaining stability, social security and so on. This paper focuses on the corresponding acoustic detection methods for two kinds of public safety emergencies: shooting and explosion.

2 current research situation

The research on gunshot focuses on the research and analysis of muzzle wave and ballistic wave. Wu Songlin et al deeply analyzed the causes and theoretical waveforms of ballistic shock waves based on the aerodynamic model of projectiles. Jiang Hao and others analyzed muzzle waves and ballistic shock waves fired by small caliber weapons, and designed a DOA model based on muzzle shock waves to estimate the trajectory of projectiles. Lu Huiyang analyzed the role of ballistic wave and muzzle wave in gunshot detection and location, and designed a set of software and hardware system of gunshot location and ranging based on triangular microphone array.

For the study of acoustic signal processing and acoustic characteristics, Zhao Li and others gave the commonly used methods of signal windowing, endpoint detection and the calculation of common acoustic features. Han Jiqing and others made a comprehensive introduction to the acoustic event detection technology and common models. Xu Dawei and others compared the endpoint detection methods based on different sound signal characteristics, and analyzed their ability to resist noise and real-time operation.

In the research of signal processing and acoustic event detection of gunshot, Jiang Xiaowei and Zhang Wen et al. [7] denoised muzzle wave by low-pass filtering and spectral subtraction, and obtained muzzle wave waveform which is highly close to the theoretical waveform in the experiment, as shown in figure 1.22, and proposed that correlation analysis can be used for gunshot detection. Zhang Kegang and others studied the method of endpoint detection of gunshot signal based on short-term energy analysis, and proposed to use duration processing to eliminate instantaneous high energy noise. Zhang Tao, Zhang Wen, Zhu Qiangqiang and others pointed out that MFCC can be used as the feature of the target segment for further classification and detection of the classifier.

For the classifier of acoustic event detection, Clavel et al discussed gunshot detection in the monitoring environment, and selected 13-dimensional features as input features of GMM model through PCA. Liu Liwei et al proposed to use 10-order median filter to deal with the energy sequence in endpoint detection, and GMM was used to classify target segments according to MFCC features. Zhu Qiangqiang analyzed the three feature selection algorithms of Logo, FFS and Adaboost, and used the feature selection algorithm to select the full set of features composed of time domain features, frequency domain features, perceptual domain features and features based on autocorrelation function, and finally input them into GMM for classification. Pimentel et al proposed a method to determine the number of clustering centers in the clustering algorithm by analyzing the WSS index in the clustering process.

With regard to the acoustic event database, in order to solve the bottlenecks and difficulties encountered in the current data-driven (data-driven) acoustic computing research, the Music Technology Research Group of Universitat Pompeu Fabra, Barcelona, where Fonseca et al is based, launched the Freesound Datasets project and established a large-scale public database Freesound based on crowdsourcing (crowdsourcing) with a wide range of audio categories. Mesaro and others of signal processing Department of Tampere University of Technology (Tampere University of Technology,TUT) initiated event detection challenge TUT Sound Events Challenge and acoustic scene detection challenge Acoustic Scene Classification Challenge, which accelerated the related research of event detection and scene analysis based on acoustics.

3 algorithm flow design

Part of the source code for ii = 1:24% 7 8 wav is the background sound and the background + gunshot, 21-24 is the explosive sound if 8

< ii && ii < 21 continue; end % 取信号 file_name = strcat('gun',num2str(ii)); file_name = strcat(file_name,'.wav'); fprintf('reading %s...\n',file_name); [y,fs] = audioread(file_name); sz = size(y); gun = (y(:,1))'; % 单声道 % 原信号 figure(ii); p2 = abs(fft(gun)/length(gun)); % size(gun) % size(1:length(gun)/2+1) gun_fft = p2(1:length(gun)/2+1); gun_fft(2:end-1) = 2*gun_fft(2:end-1); f = fs*(0:(length(gun)/2))/length(gun); subplot(3,2,1);plot(gun);xlabel('t / s');title('signal'); subplot(3,2,2);plot(f,gun_fft);title('spectrum');xlabel('frequency / Hz'); % 短时能量分析 N = 300; % 窗宽（张克刚） inc = 100; % 帧移（张克刚） win = hamming(N); % frameout: num x N % t: num x 1, centers of frames % energy: 1 x num [frameout,t,energy]=enframe(y,win,inc); t = t'; % 自适应短时能量阈值分割 %size(energy) threshold = min(energy)+0.2*(max(energy)-min(energy)); processed_energy = energy; for i = 1:length(energy) processed_energy(i) = 0; if energy(i) >

= threshold processed_energy (I) = 1; end% fprintf ('% d:% f >% f =% d\ nEnergy (I), threshold,processed_energy (I)); end subplot (3 energy,'b'); title ('energy'); hold on;plot (threshold*ones (size (energy)),' g'); subplot (3 size (energy) 5); plot (processed_energy) Title ('binarized energy')% duration analysis thr = 30;% continuous sampling point cnt = 0; for I = 1:length (processed_energy) if processed_energy (I) = = 1 if cnt > 0 cnt = cnt+1;% counter cumulative elseif cnt = = 0 cnt = 1 % initialization counter end if i = = length (processed_energy) & & cnt

< thr processed_energy((i-cnt):i) = 0; end elseif processed_energy(i) == 0 if cnt >

0 if cnt

< thr processed_energy((i-cnt):i) = 0; end end cnt = 0; end %fprintf('%f, %f\n',i,processed_energy(i)); end subplot(3,2,3);hold on;plot(processed_energy*max(energy),'r');hold off; subplot(3,2,5);hold on;plot(processed_energy,'r');hold off;endfunction [f,t,eng,zcr]=enframe(x,win,inc)%ENFRAME split signal up into (overlapping) frames: one per row. [F,T]=(X,WIN,INC)%% F = ENFRAME(X,LEN) splits the vector X(:) up into% frames. Each frame is of length LEN and occupies% one row of the output matrix. The last few frames of X% will be ignored if its length is not divisible by LEN.% It is an error if X is shorter than LEN.%% F = ENFRAME(X,LEN,INC) has frames beginning at increments of INC% The centre of frame I is X((I-1)*INC+(LEN+1)/2) for I=1,2,...% The number of frames is fix((length(X)-LEN+INC)/INC)%% F = ENFRAME(X,WINDOW) or ENFRAME(X,WINDOW,INC) multiplies% each frame by WINDOW(:)%% The second output argument, T, gives the time in samples at the centre% of each frame. T=i corresponds to the time of sample X(i). %nx=length(x);nwin=length(win);if (nwin == 1) len = win;else len = nwin;endif (nargin < 3) inc = len;endlen = nwin;nf = fix((nx-len+inc)/inc);f=zeros(nf,len);indf= inc*(0:(nf-1)).';inds = (1:len);f(:) = x(indf(:,ones(1,len))+inds(ones(nf,1),:));if (nwin >

1) w = win (:)'; f = f. * w (ones (nf,1),:); endt = floor ((1+len) / 2) + indf;%fprintf ('size of f\ n'); szf = size (f);% ff = f (:). * f (:); for I = 1:szf (1)% ff = f (iPert:). * f (iPert:)% ff = abs (f (iArt:));% eng (I) = sum (ff) Eng (I) = 0; zcr (I) = 0; for j = 1:szf (2) eng (I) = eng (I) + abs (f (I)); if jung1

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.