基于粒子滤波的视频目标跟踪算法matlab源程序

clear; clc; %数据的初始化 %在HSV空间中，将三个颜色分量合成为一维特征向量时，一维向量的大小 v_count=193; N=500; %采样粒子的个数 n=52;%179;%400;%52; %视频序列中的图像帧数 first=1; %第一帧图像的名称序号 new_sita=0.20; %（new_sita）^2表示颜色信息的高斯分布方差。 vx=[0,0,0]; vy=[0,0,0]; %得出目标的移动速度 runtime=0;%求取目标速度的时候用 struct_index=0;%存储结构体的指数 %产生随机粒子的方差 sigma_x=3.5; sigma_y=3.5; %求前10帧图像与目标模板的相似度 pre_probability=zeros(1,10); %判断是否进行了重采样 resample_judge=0; %得到目标模板的初始数据 I=imread('C:\Users\lenovo\Desktop\仿真图片3\1.bmp'); imshow(I); rect = getrect(); x1 = rect(1); x2 = rect(1) + rect(3); y1 = rect(2); y2 = rect(2) + rect(4); %得到初始跟踪目标的中心坐标点 x=round((x1+x2)/2); y=round((y1+y2)/2); %得到描述目标轮廓的椭圆的长短半轴的平方 hx=((x2-x1)/3)^2; hy=((y2-y1)/3)^2; sizeimage=size(I); image_boundary_x=int16(sizeimage(2)); image_boundary_y=int16(sizeimage(1)); %将第一帧用来选择被跟踪目标的图片存入指定的文件夹中 F = getframe; mkdir('C:\Users\lenovo\Desktop\仿真程序\result'); image_source=strcat('C:\Users\lenovo\Desktop\仿真程序\result\','1.jpg'); imwrite(F.cdata,image_source); %在第一帧中手动选定的目标进行初始化操作 [H S V]=rgb_to_rank(I); [Sample_Set,Sample_probability,Estimate,target_histgram]=initialize(x,y,hx,hy,H,S,V,N,image_boundary_x,image_boundary_y,v_count,new_sita); pre_probability(1)=Estimate(1).probability; %从第二帧往后循环迭代的进行下去 for loop=2:n struct_index=struct_index+1; a=num2str(loop+first-1); b=[a,'.bmp']; b=['C:\Users\lenovo\Desktop\仿真图片3\',b]; I=imread(b); [H,S,V]=rgb_to_rank(I); %产生随机粒子 [Sample_Set,after_prop]=reproduce(Sample_Set,vx,vy,image_boundary_x,image_boundary_y,I,N,sigma_x,sigma_y,runtime); %得出被跟踪目标的在当前帧的预测位置 [Sample_probability,Estimate,vx,vy,TargetPic,Sample_histgram]=evaluate(Sample_Set,Estimate,target_histgram,new_sita,loop,after_prop,H,S,V,N,image_boundary_x,image_boundary_y,v_count,vx,vy,hx,hy,Sample_probability); %模板更新时和重采用判断时，都要用到归一化的权值Sample_probability %模板跟新 if(loop<=10)%前10帧属于特殊情况，需要额外进行处理 sum_probability=0; for p=1:loop-1 sum_probability=sum_probability+pre_probability(p); end mean_probability=sum_probability/(loop-1); else%直接求取均值 mean_probability=mean(pre_probability); end mean_probability; Estimate(loop).probability; if(Estimate(loop).probability>mean_probability) [target_histgram,pre_probability]=update_target(target_histgram,Sample_histgram,Sample_probability,pre_probability,Estimate,N,v_count,loop,resample_judge); %不进行模板更新，但是要对pre_probability进行更新操作 else if(loop>10) for k=1:9 pre_probability(k)=pre_probability(k+1); end pre_probability(10)=Estimate(loop).probability; else pre_probability(loop)=Estimate(loop).probability; end end resample_judge=0; %判断是否需要重采样 back_sum_weight=0; for judge=1:N back_sum_weight=back_sum_weight+(Sample_probability(judge))^2; end sum_weight=1/back_sum_weight; if(sum_weight<N/2) %重采样过程 usetimes=reselect(Sample_Set,Sample_probability,N); [Sample_Set,Sample_probability]=assemble(Sample_Set,usetimes,Sample_probability,N);%进行线性组合 resample_judge=1; end %得到目标运动的轨迹 if(struct_index==1) routine.x=round(Estimate(loop).x); routine.y=round(Estimate(loop).y); else routine(struct_index).x=round(Estimate(loop).x); routine(struct_index).y=round(Estimate(loop).y); end i=1; j=1; while(j<=struct_index) for new_x=routine(j).x-i:routine(j).x+i for new_y=routine(j).y:routine(j).y+i TargetPic(new_y,new_x,1)=0; TargetPic(new_y,new_x,2)=0; TargetPic(new_y,new_x,3)=255; end end j=j+1; end %画出每一帧图像中跟踪目标的预测中心点 i=1; for new_x=round(Estimate(loop).x)-i:round(Estimate(loop).x+i) for new_y=round(Estimate(loop).y)-i:round(Estimate(loop).y+i) TargetPic(new_y,new_x,1)=255; TargetPic(new_y,new_x,2)=255; TargetPic(new_y,new_x,3)=255; end end imshow(TargetPic); F = getframe; image_source=strcat('C:\Users\lenovo\Desktop\仿真程序\result\',num2str(loop),'.bmp'); imwrite(F.cdata,image_source); end %在灰度图中绘制出被跟踪目标的轨迹 im_routine=redraw_routine(image_boundary_x,image_boundary_y,routine,struct_index); figure; title('被跟踪目标的运动轨迹图'); hold on; imshow(im_routine);