%demonstration of particle filtering
%by Paul Sundvall, KTH Signals sensors and systems 2004
clear,clc,close all

%let x be the position of the boat
%create a model for the movement of the boat
xmin=-10;
xmax=10;
xpmin=-.3;
xpmax=.3;
N=10000;
m=1;%kg
kk=1;%N/m
c=.5;%N/s
F0=25;%N
dT=.05;%s
M=400;%timesteps


%first calculate the true sequence of disturbance force
wk=(rand(M,1)-.5)*2*F0;


af=[1 -2 1].*(dT^-2)+[0 kk/m 0]+[1 0 -1]./2/dT*c/m;
bf=1/m;
xtrue=filter(bf,af,wk);
xptrue=filter([1 -1]/dT,1,xtrue);

% plot(xtrue)
% title('true x')
% xlabel('time step')
% ylabel('x (m)')
% pause

%calculate the measurement
sigma=.3;%measurent noise std deviation.
a=.2;%constant for the average slope of the bottom surface
b=0;
z=sin(xtrue)+a*xtrue+b*xtrue.*xtrue + randn(M,1)*sigma; %first part is the bottom surface, last term is the measurement noise

%initialization
xk=rand(N,1)*(xmax-xmin)+xmin;
xpk=zeros(N,1);

pik=repmat(1/N,N,1); %the propability that we are in state xk, xpk.
figure(3)
set(3,'doublebuffer','on','position',[239   291   681   343])
t=linspace(xmin,xmax,N)';
h1=line(t,sin(t)+a*t+b*t.*t,'marker','none');%t is just a temporary variable to draw the bottom surface
h3=line(xtrue(1),0,'marker','o','linestyle','none');

%plot the boat
boat.x=[-2.4 -2  -1   0   1 2 2 -2.4]';
boat.y=[   1  0 -.2 -.3 -.3 -.3 .8 1]'+5;
hboat=line(boat.x+xtrue(1),boat.y,'color','r');% line(boat.x,boat.y,'color','m');
%plot the depth measurement
hecho=line(xtrue(1)*[1 1]',[z(1) boat.y(4)]','color','r','linestyle','--');
%plot the water surface
hwater=line([xmin xmax],boat.y(2)*[1 1],'color','b','linestyle','--');

%histogram is plotted also to show p(x) (the propability density function)
h_hist=line(1,1,'color','k');
xlabel('Position x (m)')
xlim([xmin,xmax])
ylim([-3 ceil(max(boat.y))+2])

%print some text
htext(1)=text(xmin+.5,boat.y(2),'Sea surface','verticalalignment','bottom');;
htext(2)=text(boat.x(7),boat.y(2),'S/Y OptFilt','verticalalignment','bottom','horizontalalignment','right');;
htext(3)=text(xmin+.5,0,'p(x)','verticalalignment','bottom');
htext(4)=text(xmin+.5,sin(xmin+.5)+a*(xmin+.5)+b*(xmin+.5).*(xmin+.5),'Sea bottom','verticalalignment','bottom','color','b');

recordvideo=1;%to record to a video file or not
%if recordvideo
%    mov = avifile('boatdemo.avi','videoname','Particle filter example','fps',10,'quality',100);
%end

resample=.5;%to use resampling or not: 0 is no resampling, 1 is always resampling. Based on Neff/N.

Neff=zeros(M,1);

%precalculate some constants to save time
sigma_sqrt_2_pi=sigma*sqrt(2*pi);
two_sigma_square=2*sigma*sigma;

for k=1:M%loop over timesteps
    %time update step
    %predict particle i from the previous step, using random disturbance for each particle
    wk=randn(N,1)*F0;
    %predict every sample (using matrices instead of for loops to get faster execution)
    
    xk=xk+xpk*dT;
    xpk=xpk+(wk-xk*kk-xpk*(c-dT*kk))/m*dT;%ok, c-dT*kk is used instead of c to reduce the need for temporary variable on the line above
    
    %measurement update (let the measurement be z=f(x)+v where v is normal distributed with sigma std dev.  
    
    pik=pik.*exp(-(sin(xk)+a*xk+b*xk.*xk - z(k) ).^2/two_sigma_square)/sigma_sqrt_2_pi;  %ok, this is made to all particles at once.
    
    %normalize the weights
    pik=pik/sum(pik);    
    
    %resample if necessary
    if resample>0
        %only resample if a criterion is fulfilled
        %resample based on the quality of the distribution
        Neff(k)=1/sum(pik.^2);
        if Neff(k)<(resample*N) %setting resample to 1 makes this condition always true, because 1<Neff<N always!            
            Inew=rsmp(pik,N);
            xk=xk(Inew);
            xpk=xpk(Inew);
            pik=repmat(1/N,N,1); 
        end
    end
    
    
    %update the plots
    
    set(h3,'xdata',xtrue(k),'ydata',z(k))
    set(hboat,'xdata',boat.x+xtrue(k) );
    set(hecho,'xdata',xtrue(k)*[1 1]','ydata',[z(k) boat.y(4)]');
    
    
    [plotx,ploty]=histweight(xk,pik,200,[xmin xmax]);
    set(h_hist,'xdata',plotx,'ydata',ploty)
    set(htext(2),'position',[boat.x(7)+xtrue(k) boat.y(2)]);
    drawnow;
    
    %dump the frame to a video  % 此處不可正常使用
 %   if recordvideo
 %       F = getframe(gca);
 %       mov = addframe(mov,F);
 %   end
end

%if recordvideo
%    mov = close(mov);
%end
figure(4)
plot(Neff/N)
xlabel('time step')
ylabel('Particle efficiency')
title('Efficient number of particles')