OpenCV + CPP 系列（圩一）视频分析（BSM，inRange）

文章目录

使用Opencv处理视频跟踪对象时，借助视频原先的背景与当前的获取帧进行比较，有效地区分开视频中的背景和前景。这种方法我们称为背景减除(

Background Subtraction)。

​​OpenCV—python 视频分析背景提取与前景提取​​ 理论介绍。

一、背景消除建模（BSM）与对象跟踪

实现原理：

BS算法

• 图像分割(GMM – 高斯混合模型)​​BackgroundSubtractorMOG2​​
• 机器学习(KNN –K个最近邻)​​BackgroundSubtractorKNN​​ （效果略好于GMM）

createBackgroundSubtractorMOG2(

int history=500, 　　　　

double varThreshold=16,

bool detectShadows=true

)

---

createBackgroundSubtractorKNN(

int history=500,

double dist2Threshold=400.0,

bool detectShadows=true

)

history：用于训练背景的帧数，默认为500帧，如果不手动设置learningRate，history就被用于计算当前的learningRate，此时history越大，learningRate越小，背景更新越慢。

varThreshold：方差阈值，用于判断当前像素是前景还是背景。一般默认16，如果光照变化明显，如阳光下的水面，建议设为25,36，值越大，灵敏度越低；

detectShadows：是否检测影子，设为true为检测，false为不检测，检测影子会增加程序时间复杂度，如无特殊要求，建议设为false

头文件 ​​machine_learning_all.h​​：

#pragma once
#include <opencv2/opencv.hpp>
#include <iostream>

using namespace cv;
using namespace std;


class Machine_learning{
public:
  void video_analysis(const char* mp4_path);
  void video_obj_track(const char* mp4_path);
  void video_color_analysis(const char* mp4_path)
};

主函数​​main.cpp​​

#include "machine_learning_all.h"


int main(int argc, char** argv) {
    const char* mp4_path = "D:\\Desktop\\01.mp4";

    Machine_learning ml;
    ml.video_analysis(mp4_path);
    ml.video_obj_track(mp4_path);
    ml.video_color_analysis(mp4_path);

    waitKey(0);
    destroyAllWindows();
    return 0;
}

效果展示

void Machine_learning::video_analysis(const char* mp4_path) {
  VideoCapture capture(mp4_path);
  if (!capture.isOpened()) {
    cout << "opencv video failed!" << endl;
  }

  // 存储分析后的视频
  double fps = capture.get(CAP_PROP_FPS);
  Size size = Size(capture.get(CAP_PROP_FRAME_WIDTH), capture.get(CAP_PROP_FRAME_HEIGHT));
  VideoWriter writer("D:\\Desktop\\text.mp4", CAP_PROP_FOURCC, fps, size, true);

  //初始化BGS
  Ptr<BackgroundSubtractor> pMOG2 = cv::createBackgroundSubtractorMOG2();
  Ptr<BackgroundSubtractor> pKNN = cv::createBackgroundSubtractorKNN();

  Mat frame, bsmask_MOG2, bsmask_KNN;
  Mat kernel = getStructuringElement(MORPH_RECT, Size(3, 3));
  while (capture.read(frame))
  {
    pMOG2->apply(frame, bsmask_MOG2);
    pKNN->apply(frame, bsmask_KNN,-1);

    // 消除小块噪声
    morphologyEx(bsmask_MOG2, bsmask_MOG2, MORPH_OPEN, kernel);
    morphologyEx(bsmask_KNN, bsmask_KNN, MORPH_OPEN, kernel);

    // 可视化
    imshow("bsmask_MOG2", bsmask_MOG2);
    imshow("bsmask_KNN", bsmask_KNN);
    writer.write(bsmask_MOG2);
    if (waitKey(50) == 27) {
      break;
    }
  }
  capture.release();
}

视频目标跟踪二

void Machine_learning::video_obj_track(const char* mp4_path) {
  VideoCapture capture(mp4_path);
  if (!capture.isOpened()) {
    cout << "video read failed!" << endl;
  }

  vector<vector<Point>> contours;
  Mat frame, mog_mask, hue_img, mask_img, hist_img, back_project;
  Ptr<BackgroundSubtractor> pMOG2 = cv::createBackgroundSubtractorMOG2();
  Mat Kernel = getStructuringElement(MORPH_RECT, Size(3, 3));
  while (capture.read(frame)) {
    pMOG2->apply(frame, mog_mask);
    threshold(mog_mask, mog_mask, 100, 255, THRESH_BINARY);
    morphologyEx(mog_mask, mog_mask, MORPH_OPEN, Kernel);
    findContours(mog_mask, contours, RETR_EXTERNAL, CHAIN_APPROX_SIMPLE);
    int count = 0;
    for (size_t i = 0; i < contours.size(); i++) {
      double area = contourArea(contours[i]);
      if (area < 80) {continue;}                                       // 去除块状噪点
      Rect selection = boundingRect(contours[i]);
      if (selection.width < 30 || selection.height < 30) {continue;}   // 去除条状噪点
      count++;
      rectangle(frame, selection, Scalar(0, 0, 255), 2, 8);
      cout << "numText: " << count << endl;
      putText(frame, to_string(count), Point(selection.x, selection.y), FONT_HERSHEY_PLAIN, 1, Scalar(255, 0, 0), 1, 8);
    }
    imshow("frame", frame);
    if (waitKey(50) == 27) {
      break;
    }
  }
  capture.release();
}

二、基于颜色目标跟踪

pyrMeanShiftFiltering

实现流程

• inRange过滤
• 形态学操作提取
• 轮廓查找
• 外接矩形获取
• 位置标定

void Machine_learning::video_color_analysis(const char* mp4_path) {
  VideoCapture capture(mp4_path);
  if (!capture.isOpened()) {
    cout << "opencv video failed!" << endl;
  }
  Mat frame, mean_img, bsmask_KNN;
  vector<vector<Point>> contours;
  Rect rect(0, 0, 0, 0);
  Point2f center(0, 0);
  float radius = 0.0;
  Mat kernel = getStructuringElement(MORPH_ELLIPSE, Size(5, 5));
  while (capture.read(frame))
  {
    //平滑图像 -> 选择颜色区域 -> 消除小块噪声 -> 膨胀检测区域
    pyrMeanShiftFiltering(frame, mean_img, 10.0, 10.0);
    inRange(mean_img, Scalar(0, 127, 0), Scalar(120, 255, 120), mean_img);
    morphologyEx(mean_img, mean_img, MORPH_OPEN, kernel);
    dilate(mean_img, mean_img, kernel, Point(-1, -1), 4);

    findContours(mean_img, contours, RETR_EXTERNAL, CHAIN_APPROX_SIMPLE);
    if (contours.size() > 0) {
      double maxArea = 0;
      for (int i = 0; i < contours.size(); i++){
        double area = contourArea(contours[i]);
        if (area > maxArea) {
          maxArea = area;
          rect = boundingRect(contours[i]);
          minEnclosingCircle(contours[i], center, radius);

          rectangle(frame, rect, Scalar(0, 0, 255), 2, 8);
          circle(frame, center, radius, Scalar(255, 255, 0), 2, 8);
        }
      }
    }
    imshow("mean_img", mean_img);
    imshow("frame", frame);
    if (waitKey(50) == 27) {
      break;
    }
  }
  capture.release();
}