基于精确背景补偿的动态目标检测.zip

#include <opencv2/opencv.hpp> #include <opencv2/xfeatures2d.hpp> #include <opencv2/highgui.hpp> #include "iostream" #include "func.h" using namespace cv; using namespace std; using namespace xfeatures2d; #define pi 3.1415926 //使用说明 void Func::attention() { printf("[0]表示前一帧，[1]表示当前帧\n\n"); } //读视频文件 void Func::readVideo(String &videoFile) { capture.open(videoFile); //capture.open(0); if (!capture.isOpened()) { cout << "Could not initialize capturing...\n"; system("pause"); } width = capture.get(CV_CAP_PROP_FRAME_WIDTH); height = capture.get(CV_CAP_PROP_FRAME_HEIGHT); surfDetector = SURF::create(1000); } //读frame void Func::readFrame() { capture.read(cur_frame); //读当前帧 cvtColor(cur_frame, cur_gray, COLOR_BGR2GRAY); if (prev_gray.empty()) { cur_gray.copyTo(prev_gray); //更新前一帧 cur_frame.copyTo(prev_frame); //更新前一帧 } } //SURF特征检测 void Func::surfDetect() { Pts[0].clear(); Pts[1].clear(); desc[0].release(); desc[1].release(); surfDetector->detectAndCompute(prev_gray, Mat(), Pts[0], desc[0]); surfDetector->detectAndCompute(cur_gray, Mat(), Pts[1], desc[1]); //drawKeypoints(cur_frame, Pts[1], cur_frame, Scalar(0, 0, 255)); //printf("SURF检测点数：%d、%d \n", Pts[0].size(), Pts[1].size()); } //KNN匹配 void Func::BfMatch() { knnMatches[0].clear(); matcher.knnMatch(desc[0], desc[1], knnMatches[0], 2); //正向匹配 printf("KNN匹配点对数：%d \n", knnMatches[0].size()); //BBF搜索算法 //BBFmatches.clear(); BBFpts[0].clear(); BBFpts[1].clear(); for (size_t i = 0; i < knnMatches[0].size(); i++) { const DMatch& bestMatch = knnMatches[0][i][0]; //汉明距离最小 const DMatch& betterMatch = knnMatches[0][i][1]; //汉明距离次小 float distanceRatio = bestMatch.distance / betterMatch.distance; //利用汉明距离粗略过滤匹配点对 if (distanceRatio < minRatio) { //BBFmatches.push_back(bestMatch); BBFpts[0].push_back(Pts[0][bestMatch.queryIdx].pt); BBFpts[1].push_back(Pts[1][bestMatch.trainIdx].pt); } } printf("对称约束前：BBFpts[0].size = %d \n", BBFpts[0].size()); knnMatches[1].clear(); matcher.knnMatch(desc[1], desc[0], knnMatches[1], 2); //反向匹配 BBFpts_rev[0].clear(); BBFpts_rev[1].clear(); for (size_t i = 0; i < knnMatches[1].size(); i++) { const DMatch& bestMatch_rev = knnMatches[1][i][0]; //汉明距离最小 const DMatch& betterMatch_rev = knnMatches[1][i][1]; //汉明距离次小 float distanceRatio = bestMatch_rev.distance / betterMatch_rev.distance; //利用汉明距离粗略过滤匹配点对 if (distanceRatio < minRatio) { //BBFmatches.push_back(bestMatch); BBFpts_rev[0].push_back(Pts[0][bestMatch_rev.trainIdx].pt); BBFpts_rev[1].push_back(Pts[1][bestMatch_rev.queryIdx].pt); } } //printf("对称约束前：BBFpts_rev[0].size = %d, BBFpts_rev[1].size = %d \n", BBFpts_rev[0].size(), BBFpts_rev[1].size()); /*printf("---------------------------------------------------------\n"); for (size_t i = 0; i < BBFpts[0].size(); i++) { printf("(%d, %d) (%d, %d)\n", BBFpts[0][i].x, BBFpts[0][i].y, BBFpts[1][i].x, BBFpts[1][i].y); } printf("~~~~~~~~~~~~~~\n"); for (size_t j = 0; j < BBFpts_rev[0].size(); j++) { printf("(%d, %d) (%d, %d)\n", BBFpts_rev[0][j].x, BBFpts_rev[0][j].y, BBFpts_rev[1][j].x, BBFpts_rev[1][j].y); } printf("---------------------------------------------------------\n");*/ //对称约束 pts_opti[0].clear(); pts_opti[1].clear(); for (size_t i = 0; i < BBFpts[0].size(); i++) { for (size_t j = 0; j < BBFpts_rev[0].size(); j++) { //满足对称约束，则跳过 if ((BBFpts[0][i].x == BBFpts_rev[0][j].x) && (BBFpts[0][i].y == BBFpts_rev[0][j].y)) { if ((BBFpts[1][i].x == BBFpts_rev[1][j].x) && (BBFpts[1][i].y == BBFpts_rev[1][j].y)) { //位于四角且边界向内10像素范围内的点保留 if (BBFpts[1][i].x >= 20 && BBFpts[1][i].x <= (width - 20) && BBFpts[1][i].y >= 20 && BBFpts[1][i].y <= (height - 20)) { pts_opti[0].push_back(Point2f(BBFpts[0][i].x, BBFpts[0][i].y)); pts_opti[1].push_back(Point2f(BBFpts[1][i].x, BBFpts[1][i].y)); } } } } } printf("对称约束后：pts_opti[1].size = %d \n", pts_opti[1].size()); /*for (size_t i = 0; i < pts_opti[1].size(); i++) { printf("(%d, %d) \n", pts_opti[1][i].x, pts_opti[1][i].y); } */ } /*********************** 待优化 **********************/ //自适应外点滤除 void Func::filtForePts() { //获取边角处的特征点 getCornerPts(0.2, 0.2); if (pts_corner[1].size() < 10) { getCornerPts(0.3, 0.3); } // printf("四角的点数：%d \n", pts_corner[1].size()); //自适应外点滤除 pts_BGD[0].clear(); pts_BGD[1].clear(); pts_FGD[0].clear(); pts_FGD[1].clear(); if (pts_corner[1].size() < 10) { //不做外点滤除 if (pts_corner[1].size() <= 3) { for (size_t i = 0; i < pts_opti[1].size(); i++) { pts_BGD[0].push_back(pts_opti[0][i]); pts_BGD[1].push_back(pts_opti[1][i]); } } else { for (size_t i = 0; i < pts_corner[1].size(); i++) { pts_BGD[0].push_back(pts_corner[0][i]); pts_BGD[1].push_back(pts_corner[1][i]); } } } else { //外点滤除 H = myGetAffineTransform(pts_corner[0], pts_corner[1]); //计算初始 H cout << "初始 H：" << endl; cout << H << endl; calcPtsErrs(); //计算距离残差 int errN = ptsErrs.size(); //printf("errN: %d \n", errN); //迭代min max 距离残差 float minPtsErr = 100.0, maxPtsErr = 0.0; for (size_t i = 0; i < ptsErrs.size(); i++) { float ptsErr = ptsErrs[i]; if (ptsErr > maxPtsErr) maxPtsErr = ptsErr; if (ptsErr < minPtsErr) minPtsErr = ptsErr; } //printf("minPtsErr：%.2f, maxPtsErr:%.2f \n", minPtsErr, maxPtsErr); //ptsErrs_num分配空间 ptsErrs_num.clear(); for (int i = 0; i <= L; i++) { ptsErrs_num.push_back(0); } float ptsErr_element = (maxPtsErr - minPtsErr) / ((L + 1) * 1.0); //分级因子 if (ptsErr_element != 0.0) { //统计每级（L）个数 for (size_t i = 0; i < errN; i++) { int a = int((ptsErrs[i] - minPtsErr) / ptsErr_element); //商即为级数 //printf(" %d", a); if (a <= L) { ptsErrs_num[a]++; } else { ptsErrs_num[L]++; } } //计算距离残差各等级概率 ptsErrs_prob.clear(); for (size_t i = 0; i <= L; i++) { ptsErrs_prob.push_back(ptsErrs_num[i] * 1.0 / (errN*1.0)); //printf("%d -- %.2f\n", ptsErrs_num[i], ptsErrs_prob[i]); //printf("%d ", ptsErrs_num[i]); } //printf("\n"); //从0-L之间遍历找到使sigma最大的m float maxSigma = 0.0; for (int m = 1; m < L; m++) { //计算背景（前景）点集概率和 PtsBGD_prob = 0.0; PtsFGD_prob = 0.0; for (size_t i = 0; i <= L; i++) { if (i <= m) { //背景 PtsBGD_prob += ptsErrs_prob[i]; } else if (i > m) { //前景 PtsFGD_prob += ptsErrs_prob[i]; } } //计算背景（前景）点集的均值 PtsBGD_aver = 0.0; PtsFGD_aver = 0.0; for (size_t i = 0; i <= L; i++) { if (i <= m) { PtsBGD_aver += i * ptsErrs_prob[i] / PtsBGD_prob; } else if (i > m) { PtsFGD_aver += i * ptsErrs_prob[i] / PtsFGD_prob; } } //计算类间方差 float sigma = PtsBGD_prob * pow(PtsBGD_aver, 2) + PtsFGD_prob * pow(PtsFGD_aver, 2); //更新前(背)景阈值 if (sigma > maxSigma) { maxSigma = sigma; th = m; } } //printf("th = %d \n", th); //分离前景和背景点 for (size_t i = 0; i < errN; i++) { if (ptsErrs[i] < ((th + 1)*ptsErr_element + minPtsErr)) { //背景点 pts_BGD[0].push_back(pts_opti[0][i]); pts_BGD[1].push_back(pts_opti[1][i]); } else { //前景点 pts_FGD[0].push_back(pts_opti[0][i]); pts_FGD[1].push_back(pts_opti[1][i]); } } } } // printf("背景点：%d, 前景点数：%d \n", pts_BGD[1].size(), pts_FGD[1].size()); float