下面这段代码利用OpenCV实现了最为简单的基于自适应背景更新的运动目标检测算法。
即：
运动前景提取——背景减除
foreground = |frame - background| > threshold
更新背景模型——滑动平均滤波
background = background + (frame - background) * alpha = background * (1 - alpha) + frame * alpha
//运动前景检测——基于自适应背景更新
//Author:
#include <iostream>
#include <string>
#include <opencv2/opencv.hpp>
int main(int argc, char** argv)
{
std::string videoFileName = "../test.avi";
int threshold = 25 ; // 二值化阈值
float alpha = 0.01; // 更新速度 [0, 1]
cv::VideoCapture capture;
capture.open(videoFileName);
if (!capture.isOpened())
{
std::cout<<"cannot open video"<<std::endl;
return -1;
}
cv::Mat foregroundImg;
cv::Mat foregroundMat;
cv::Mat backgroundImg;
cv::Mat backgroundMat;
cv::Mat frame;
cv::Mat grayImg;
cv::Mat grayMat;
while (capture.read(frame))
{
cv::cvtColor(frame, grayImg, CV_BGR2GRAY);
grayImg.convertTo(grayMat, CV_32FC1);
if (backgroundMat.empty())
{
grayImg.copyTo(backgroundImg);
grayImg.convertTo(backgroundMat, CV_32FC1);
}
// 背景减除
cv::absdiff(grayMat, backgroundMat, foregroundMat);
// 自适应背景更新
cv::addWeighted(backgroundMat, alpha, foregroundMat, 1-alpha, 0, backgroundMat);
// 二值化，获取前景像素点
cv::threshold(foregroundMat, foregroundMat, threshold, 255, CV_THRESH_BINARY);
// 为了显示用，将CV_32FC1转换为CV_8U
cv::convertScaleAbs(foregroundMat, foregroundImg);
cv::convertScaleAbs(backgroundMat, backgroundImg);
cv::imshow("frame", frame);
cv::imshow("foreground", foregroundImg);
cv::imshow("background", backgroundImg);
if (cv::waitKey(25) > 0)
{
break;
}
}
return 0;
}