W4算法基于背景建模的运动目标检测

W4算法对每个像素点统计三个特征值：最小灰度值、最大灰度值以及最大帧间差分值。所有像素点的特征值就构成了背景模型，比较当前帧与背景模型就能检测出运动的目标。具体步骤可分为三部分：模型初始化、目标检测以及模型更新。

模型初始化：首先遍历初始数帧，并对每个像素点计算灰度中值和标准差，然后计算每个像素点帧间灰度变化的最大值、最小值以及最大帧间差分值。计算时间过长，没有进行优化。

目标检测：

#include <iostream>
#include<opencv2/opencv.hpp>
#include <vector>
using namespace cv;
using namespace std;

//每个像素的灰度中值
bool comp(int x, int y)
{
	return x > y;
}
void sort_vector(vector <int> v)
{
	
	sort(v.begin(), v.end(), comp);
}

//每个像素的灰度值标准差
double mean(vector <int> v) {
	double ans = 0.0;
	for (int i = 0; i < v.size(); i++) {
		ans += v[i];
	}
	ans = ans / v.size();
	return ans;
}

double vector_stddev(vector <int> v) {
	double a1 = mean(v);
	double a2 = 0.0;
	for (int i = 0; i < v.size(); i++) {
		a2 = (a1 - v[i]) * (a1 - v[i]);
	}
	a2 = sqrt(a2 / v.size());
	return a2;
}

//相邻帧像素灰度差分值
int vector_diff(vector <int> v) {
	vector<int> temp;
	for (int i = 0; i < v.size()-1; i++) {
		int result = abs(v[i + 1] - v[i]);
		temp.push_back(result);
	}
	int max = *max_element(temp.begin(), temp.end()); ;
	return max;
}

//相邻帧像素灰度差分值的平均值
double mean_diff(vector <int> v) {
	vector<int> temp;
	for (int i = 0; i < v.size() - 1; i++) {
		int result = abs(v[i + 1] - v[i]);
		temp.push_back(result);
	}
	double a1 = mean(temp);
	return a1;
}



int main()
{
	VideoCapture capture;
	Mat frame,framegray,result;
	//保存目标轨迹  
	std::vector<Point> pt;
	//capture.open(0);
	frame = capture.open("D:/testimage/video/fish.mp4");	//F:/opencv/sources/amples/data/vtest.avi  D:/testimage/CXK.avi D:/testimage/video/fish.mp4
	if (!capture.isOpened())
	{
		printf("can not open camera \n");
		return -1;
	}

	int iVideoTime = capture.get(CAP_PROP_FRAME_COUNT); // capture.get(CAP_PROP_FPS);
	int Frame_start= capture.get(CAP_PROP_POS_FRAMES);
	capture.read(frame);
	
	vector <double> all_diff;
	cvtColor(frame, result, COLOR_RGB2GRAY);
	//namedWindow("input", WINDOW_AUTOSIZE);
	for (int i = 0; i < frame.rows; i++)
	{
		vector <int> v_median;
		for (int j = 0; j < frame.cols; j++)
		{
			//temp1.push_back(gray_value);
			while (capture.read(frame) && Frame_start < 20)
			{
				cvtColor(frame, framegray, COLOR_RGB2GRAY);
				int gray_value = framegray.at<uchar>(i, j);
				//cout << gray_value << endl;
				v_median.push_back(gray_value);
				Frame_start++;
				//imshow("input", frame);
				//waitKey(10);
			}

			/*判断当前像素与灰度中值差值的绝对值是否小于2倍标准差*/
			
			double temp = vector_stddev(v_median);
			double error = abs(framegray.at<uchar>(i, j) - v_median[v_median.size() / 2.0]) - 2 * temp;
			if (error < 0) {
			
				int diff = vector_diff(v_median);
				double Mean_diff = mean_diff(v_median);
				cout << "d(x)相邻帧像素差分值：" << diff << endl;
				cout << "d(x)相邻帧像素差分值的平均值：" << Mean_diff << endl;
				sort_vector(v_median);

				cout << "m(x)最大值：" << v_median[0] << endl;
				cout << "n(x)最小值：" << v_median[v_median.size() - 1] << endl;

				cout << "坐标（" << i << "，" << j << "）的灰度中值：" << v_median[v_median.size() / 2.0] << endl;
				cout << "坐标（" << i << "，" << j << "）的灰度标准差：" << temp << endl;
				cout << "error：" << error << endl;
				int gray_value = framegray.at<uchar>(i, j);

				if (abs(gray_value - v_median[0]) < 2 * Mean_diff || abs(gray_value - v_median[v_median.size() - 1]) < 2 * Mean_diff) {

					result.at<uchar>(i, j) = 0;
				}
				else
				{
					result.at<uchar>(i, j) = 255;
				}
				
			}

			
			Frame_start = 0;
			frame = capture.open("D:/testimage/video/fish.mp4");
			cvtColor(frame, framegray, COLOR_RGB2GRAY);
		}
		cout << "--------"<< i <<"---------" << endl;
		Frame_start = 0;
		frame = capture.open("D:/testimage/video/fish.mp4");
		cvtColor(frame, framegray, COLOR_RGB2GRAY);
	}
	namedWindow("result", WINDOW_AUTOSIZE);
	imshow("result", result);
	capture.release();
	system("pause");
	return 0;

}