这篇博客记录了利用高斯混合模型来进行目标追踪的学习过程。
有关高斯混合模型的原理请见 http://blog.csdn.net/smuevian/article/details/70169811
一、算法原理及过程
这部分内容摘自论文《基于高斯背景模型的运动目标检测与跟踪 》,为简便起见,我直接贴原文的图,方便以后查询:
算法的流程图如下:
二、Matlab程序实现:
clc,clear;
filename='E:\课题\背景建模\code\1\1.mp4';
source = VideoReader(filename);
t=cputime;
% ----------------------- frame size variables -----------------------
fr = read(source,4); % read in 1st frame as background frame
fr_bw = rgb2gray(fr); % convert background to greyscale
fr_size = size(fr); % size of the frame!
width = fr_size(2);
height = fr_size(1);
fg = zeros(height, width); %fg is the matrix of foreground
bg_bw = zeros(height, width); %bg_bw
% --------------------- MOG variables -----------------------------------
C = 1; % number of gaussian components (typically 3-5)
M = 1; % number of background components
D = 2.5; % positive deviation threshold
alpha = 0.01; % learning rate (between 0 and 1) (here is 0.01)
thresh = 0.45; % foreground threshold (0.25 or 0.75 in general)
sd_init = 6; % initial standard deviation(初始化标准方差) (for new components) var = 36 here
w = zeros(height,width,C); % initialize weights array
mean = zeros(height,width,C); % pixel means
sd = zeros(height,width,C); % pixel standard deviations(标准方差)
u_diff = zeros(height,width,C); % difference of each pixel from mean
p = alpha/(1/C); % initial p variable (used to update mean and sd)
rank = zeros(1,C); % rank of components (w/sd)
% --------------------- initialize component means and weights -----------
pixel_depth = 8; % 8-bit resolution
pixel_range = 2^pixel_depth -1; % pixel range (# of possible values)
% 对每个像素点,定义K个高斯模型来表征图像中的各个像素点的特征
for i=1:height
for j=1:width
for k=1:C
mean(i,j,k) = rand*pixel_range; % means random (0-255)
w(i,j,k) = 1/C; % weights uniformly dist
sd(i,j,k) = sd_init; % initialize to sd_init
end
end
end
%--------------------- process frames -----------------------------------
LengthFrame_source=source.Duration*source.FrameRate-4;
for n = 4:int8(LengthFrame_source)
fr = read(source,n); % read in frame
fr_bw = rgb2gray(fr); % convert frame to grayscale,转换成灰度图像
% calculate difference of pixel values from mean
for m=1:C
u_diff(:,:,m) = abs(double(fr_bw) - double(mean(:,:,m)));
end
% update gaussian components for each pixel,参数更新
for i=1:height
for j=1:width
match = 0;
for k=1:C
if (abs(u_diff(i,j,k)) <= D*sd(i,j,k)) % pixel matches component
match = 1; % variable to signal component match
% update weights, mean, sd, p
w(i,j,k) = (1-alpha)*w(i,j,k) + alpha;
p = alpha/w(i,j,k);
mean(i,j,k) = (1-p)*mean(i,j,k) + p*double(fr_bw(i,j));
sd(i,j,k) = sqrt((1-p)*(sd(i,j,k)^2) + p*((double(fr_bw(i,j)) - mean(i,j,k)))^2);
else % pixel doesn't match component
w(i,j,k) = (1-alpha)*w(i,j,k); % weight slighly decreases
end
end
w(i,j,:) = w(i,j,:)./sum(w(i,j,:)); %normalize the weight w(i,j,k),so [sum of w(i,j,k)=1],对权重进行归一化
bg_bw(i,j)=0; %bg_bw的含义是什么
for k=1:C
bg_bw(i,j) = bg_bw(i,j)+ mean(i,j,k)*w(i,j,k);
end
% if no components match, create new component,替换权重最小的高斯分布
if (match == 0)
[min_w, min_w_index] = min(w(i,j,:));
mean(i,j,min_w_index) = double(fr_bw(i,j));
sd(i,j,min_w_index) = sd_init;
end
rank = w(i,j,:)./sd(i,j,:); % calculate component rank
rank_ind = [1:1:C];
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% sort rank values,高斯分布由大到小排列
for k=2:C
for m=1:(k-1)
if (rank(:,:,k) > rank(:,:,m))
% swap max values
rank_temp = rank(:,:,m);
rank(:,:,m) = rank(:,:,k);
rank(:,:,k) = rank_temp;
% swap max index values
rank_ind_temp = rank_ind(m);
rank_ind(m) = rank_ind(k);
rank_ind(k) = rank_ind_temp;
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%may be it can be cope like this:
%[rank,rank_ind]=sort(rank);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% calculate foreground,计算前景
match = 0;
k=1;
fg(i,j) = 0;
fg2(i,j) = 0;
while ((match == 0)&&(k<=M)) % M=3:number of background components
if (w(i,j,rank_ind(k)) >= thresh)
if (abs(u_diff(i,j,rank_ind(k))) <= D*sd(i,j,rank_ind(k)))
fg(i,j) = 0;
fg2(i,j)=0;
match = 1;
else
%fg(i,j) = fr_bw(i,j);
%fg(i,j) = fr(i,j);
fg(i,j) = 1;
fg2(i,j)=255;
end
end
k = k+1;
end
end
end
% se=strel('line',5,0);
% se=strel('square',3);
%
% fg2=imdilate(fg2,se);
% 对图像进行形态学处理
se=strel('rectangle',[2 2]);
fg2=bwmorph(fg2,'skel',Inf);
fg2=imfill(fg2,'holes');
fg2=bwareaopen(fg2,5);
fg2=imdilate(fg2,se);
fg2=bwareaopen(fg2,1);
fg2 = bwareaopen(fg2,500);
figure(1);imshow(fg2);
%fr是读取视频的每帧图像
%fg,skin是一个大小为[height,width,C]的矩阵,若像素点为前景,则对应的点为1
skin=fg;
fg(:,:,1)= double(fr(:,:,1)).*skin(:,:,1);
fg(:,:,2)= double(fr(:,:,2)).*skin(:,:,1);
fg(:,:,3)= double(fr(:,:,3)).*skin(:,:,1);
figure(2),imshow(fr)
figure(3),imshow(uint8(bg_bw))
figure(4),imshow(uint8(fg))
figure(4),subplot(2,2,1),imshow(fr)
subplot(2,2,2),imshow(uint8(bg_bw))
subplot(2,2,3),imshow(uint8(fg))
subplot(2,2,4),imshow(uint8(fg2))
Mov1(n) = im2frame(uint8(fg),gray); % put frames into movie
% Mov2(n) = im2frame(uint8(bg_bw),gray); % put frames into movie
Mov3(n) = im2frame(uint8(fg2),gray); % put frames into movie
end
movie2avi(Mov1,'mixture_of_gaussians_output','fps',15); % save movie as avi
movie2avi(Mov2,'mixture_of_gaussians_background','fps',15); % save movie as avi
movie2avi(Mov3,'mixture_of_gaussians_out','fps',15);
figure;imshow(uint8(bg_bw));
figure;imshow(uint8(fg)) ;
time=cputime-t;
fprintf('%f',time);
上述代码与流程图的步骤稍有不符,主要是在k三、利用opencv实现该模型
#include
#include
#include
using namespace std;
using namespace cv;
int main() {
//视频路径
string FilePath = "E:/study/book/opencv/opencv-2-cookbook-src-master/images/bike.avi";
//读取视频
VideoCapture capture(FilePath);
if (!capture.isOpened()) {
cout << "Can't open the video!" << endl;
return 1;
}
//读取帧率
double rate = capture.get(CV_CAP_PROP_FPS);
int delay = 1000 / rate;
Mat frame;
Mat foreground;
Mat mask;
namedWindow("Extracted Foreground");
Ptrbgsubstractor = createBackgroundSubtractorMOG2();
bgsubstractor->setVarThreshold(20);
bool stop(false);
while (!stop) {
if (!capture.read(frame))
{
break;
}
bgsubstractor->apply(frame, mask, 0.01);
imshow("mask", mask);
if (waitKey(delay) > 0)
stop = true;
}
capture.release();
waitKey();
return 0;
}
http://www.cnblogs.com/tornadomeet/archive/2012/06/02/2531565.html
利用opencv底层函数来实现这个模型的源代码:
http://blog.csdn.net/zouxy09/article/details/9622401,要注意二楼的回复,按照他的方法,代码可以运行成功,但效果不是很好。