TLD（Tracking-Learning-Detection）學(xué)習(xí)與源碼理解之（六）

zouxy09@qq.com

http://blog.csdn.net/zouxy09

????? 下面是自己在看論文和這些大牛的分析過程中，對代碼進(jìn)行了一些理解，但是由于自己接觸圖像處理和機(jī)器視覺沒多久，另外由于自己編程能力比較弱，所以分析過程可能會有不少的錯誤，希望各位不吝指正。而且，因為編程很多地方不懂，所以注釋得非常亂，還海涵。

?

TLD.h

#include <opencv2/opencv.hpp> #include <tld_utils.h> #include <LKTracker.h> #include <FerNNClassifier.h> #include <fstream>//Bounding Boxes struct BoundingBox : public cv::Rect {BoundingBox(){}BoundingBox(cv::Rect r): cv::Rect(r){} //繼承的話需要初始化基類 public:float overlap; //Overlap with current Bounding Boxint sidx; //scale index };//Detection structure struct DetStruct {std::vector<int> bb;std::vector<std::vector<int> > patt;std::vector<float> conf1;std::vector<float> conf2;std::vector<std::vector<int> > isin;std::vector<cv::Mat> patch;};//Temporal structure struct TempStruct {std::vector<std::vector<int> > patt;std::vector<float> conf;};struct OComparator{ //比較兩者重合度OComparator(const std::vector<BoundingBox>& _grid):grid(_grid){}std::vector<BoundingBox> grid;bool operator()(int idx1,int idx2){return grid[idx1].overlap > grid[idx2].overlap;} };struct CComparator{ //比較兩者確信度？CComparator(const std::vector<float>& _conf):conf(_conf){}std::vector<float> conf;bool operator()(int idx1,int idx2){return conf[idx1]> conf[idx2];} };class TLD{ private:cv::PatchGenerator generator; //PatchGenerator類用來對圖像區(qū)域進(jìn)行仿射變換FerNNClassifier classifier;LKTracker tracker;//下面這些參數(shù)通過程序開始運行時讀入parameters.yml文件進(jìn)行初始化///Parametersint bbox_step;int min_win;int patch_size;//initial parameters for positive examples//從第一幀得到的目標(biāo)的bounding box中（文件讀取或者用戶框定），經(jīng)過幾何變換得//到 num_closest_init * num_warps_init 個正樣本int num_closest_init; //最近鄰窗口數(shù) 10int num_warps_init; //幾何變換數(shù)目 20int noise_init;float angle_init;float shift_init;float scale_init;從跟蹤得到的目標(biāo)的bounding box中，經(jīng)過幾何變換更新正樣本（添加到在線模型？）//update parameters for positive examplesint num_closest_update;int num_warps_update;int noise_update;float angle_update;float shift_update;float scale_update;//parameters for negative examplesfloat bad_overlap;float bad_patches;///Variables //Integral Images 積分圖像，用以計算2bitBP特征（類似于haar特征的計算） //Mat最大的優(yōu)勢跟STL很相似，都是對內(nèi)存進(jìn)行動態(tài)的管理，不需要之前用戶手動的管理內(nèi)存cv::Mat iisum;cv::Mat iisqsum;float var;//Training data//std::pair主要的作用是將兩個數(shù)據(jù)組合成一個數(shù)據(jù)，兩個數(shù)據(jù)可以是同一類型或者不同類型。//pair實質(zhì)上是一個結(jié)構(gòu)體，其主要的兩個成員變量是first和second，這兩個變量可以直接使用。//在這里用來表示樣本，first成員為 features 特征點數(shù)組，second成員為 labels 樣本類別標(biāo)簽std::vector<std::pair<std::vector<int>,int> > pX; //positive ferns <features,labels=1> 正樣本std::vector<std::pair<std::vector<int>,int> > nX; // negative ferns <features,labels=0> 負(fù)樣本cv::Mat pEx; //positive NN example std::vector<cv::Mat> nEx; //negative NN examples//Test data std::vector<std::pair<std::vector<int>,int> > nXT; //negative data to Teststd::vector<cv::Mat> nExT; //negative NN examples to Test//Last frame dataBoundingBox lastbox;bool lastvalid;float lastconf;//Current frame data//Tracker databool tracked;BoundingBox tbb;bool tvalid;float tconf;//Detector dataTempStruct tmp;DetStruct dt;std::vector<BoundingBox> dbb;std::vector<bool> dvalid; //檢測有效性？？std::vector<float> dconf; //檢測確信度？？bool detected;//Bounding Boxesstd::vector<BoundingBox> grid;std::vector<cv::Size> scales;std::vector<int> good_boxes; //indexes of bboxes with overlap > 0.6std::vector<int> bad_boxes; //indexes of bboxes with overlap < 0.2BoundingBox bbhull; // hull of good_boxes //good_boxes 的 殼，也就是窗口的邊框BoundingBox best_box; // maximum overlapping bboxpublic://ConstructorsTLD();TLD(const cv::FileNode& file);void read(const cv::FileNode& file);//Methodsvoid init(const cv::Mat& frame1,const cv::Rect &box, FILE* bb_file);void generatePositiveData(const cv::Mat& frame, int num_warps);void generateNegativeData(const cv::Mat& frame);void processFrame(const cv::Mat& img1,const cv::Mat& img2,std::vector<cv::Point2f>& points1,std::vector<cv::Point2f>& points2,BoundingBox& bbnext,bool& lastboxfound, bool tl,FILE* bb_file);void track(const cv::Mat& img1, const cv::Mat& img2,std::vector<cv::Point2f>& points1,std::vector<cv::Point2f>& points2);void detect(const cv::Mat& frame);void clusterConf(const std::vector<BoundingBox>& dbb,const std::vector<float>& dconf,std::vector<BoundingBox>& cbb,std::vector<float>& cconf);void evaluate();void learn(const cv::Mat& img);//Toolsvoid buildGrid(const cv::Mat& img, const cv::Rect& box);float bbOverlap(const BoundingBox& box1,const BoundingBox& box2);void getOverlappingBoxes(const cv::Rect& box1,int num_closest);void getBBHull();void getPattern(const cv::Mat& img, cv::Mat& pattern,cv::Scalar& mean,cv::Scalar& stdev);void bbPoints(std::vector<cv::Point2f>& points, const BoundingBox& bb);void bbPredict(const std::vector<cv::Point2f>& points1,const std::vector<cv::Point2f>& points2,const BoundingBox& bb1,BoundingBox& bb2);double getVar(const BoundingBox& box,const cv::Mat& sum,const cv::Mat& sqsum);bool bbComp(const BoundingBox& bb1,const BoundingBox& bb2);int clusterBB(const std::vector<BoundingBox>& dbb,std::vector<int>& indexes); };

TLD.cpp

/** TLD.cpp** Created on: Jun 9, 2011* Author: alantrrs*/#include <TLD.h> #include <stdio.h> using namespace cv; using namespace std;TLD::TLD() { } TLD::TLD(const FileNode& file){read(file); }void TLD::read(const FileNode& file){///Bounding Box Parametersmin_win = (int)file["min_win"];///Genarator Parameters//initial parameters for positive examplespatch_size = (int)file["patch_size"];num_closest_init = (int)file["num_closest_init"];num_warps_init = (int)file["num_warps_init"];noise_init = (int)file["noise_init"];angle_init = (float)file["angle_init"];shift_init = (float)file["shift_init"];scale_init = (float)file["scale_init"];//update parameters for positive examplesnum_closest_update = (int)file["num_closest_update"];num_warps_update = (int)file["num_warps_update"];noise_update = (int)file["noise_update"];angle_update = (float)file["angle_update"];shift_update = (float)file["shift_update"];scale_update = (float)file["scale_update"];//parameters for negative examplesbad_overlap = (float)file["overlap"];bad_patches = (int)file["num_patches"];classifier.read(file); }//此函數(shù)完成準(zhǔn)備工作 void TLD::init(const Mat& frame1, const Rect& box, FILE* bb_file){//bb_file = fopen("bounding_boxes.txt","w");//Get Bounding Boxes//此函數(shù)根據(jù)傳入的box（目標(biāo)邊界框）在傳入的圖像frame1中構(gòu)建全部的掃描窗口，并計算重疊度buildGrid(frame1, box);printf("Created %d bounding boxes\n",(int)grid.size()); //vector的成員size()用于獲取向量元素的個數(shù)///Preparation//allocation//積分圖像，用以計算2bitBP特征（類似于haar特征的計算）//Mat的創(chuàng)建，方式有兩種：1.調(diào)用create（行，列，類型）2.Mat（行，列，類型（值））。iisum.create(frame1.rows+1, frame1.cols+1, CV_32F);iisqsum.create(frame1.rows+1, frame1.cols+1, CV_64F);//Detector data中定義：std::vector<float> dconf; 檢測確信度？？//vector 的reserve增加了vector的capacity，但是它的size沒有改變！而resize改變了vector//的capacity同時也增加了它的size！reserve是容器預(yù)留空間，但在空間內(nèi)不真正創(chuàng)建元素對象，//所以在沒有添加新的對象之前，不能引用容器內(nèi)的元素。//不管是調(diào)用resize還是reserve，二者對容器原有的元素都沒有影響。//myVec.reserve( 100 ); // 新元素還沒有構(gòu)造, 此時不能用[]訪問元素//myVec.resize( 100 ); // 用元素的默認(rèn)構(gòu)造函數(shù)構(gòu)造了100個新的元素，可以直接操作新元素dconf.reserve(100);dbb.reserve(100);bbox_step =7;//以下在Detector data中定義的容器都給其分配grid.size()大小（這個是一幅圖像中全部的掃描窗口個數(shù)）的容量//Detector data中定義TempStruct tmp; //tmp.conf.reserve(grid.size());tmp.conf = vector<float>(grid.size());tmp.patt = vector<vector<int> >(grid.size(), vector<int>(10,0));//tmp.patt.reserve(grid.size());dt.bb.reserve(grid.size());good_boxes.reserve(grid.size());bad_boxes.reserve(grid.size());//TLD中定義：cv::Mat pEx; //positive NN example 大小為15*15圖像片pEx.create(patch_size, patch_size, CV_64F);//Init Generator//TLD中定義：cv::PatchGenerator generator; //PatchGenerator類用來對圖像區(qū)域進(jìn)行仿射變換/*cv::PatchGenerator::PatchGenerator ( double _backgroundMin,double _backgroundMax,double _noiseRange,bool _randomBlur = true,double _lambdaMin = 0.6,double _lambdaMax = 1.5,double _thetaMin = -CV_PI,double _thetaMax = CV_PI,double _phiMin = -CV_PI,double _phiMax = CV_PI ) 一般的用法是先初始化一個PatchGenerator的實例，然后RNG一個隨機(jī)因子，再調(diào)用（）運算符產(chǎn)生一個變換后的正樣本。*/generator = PatchGenerator (0,0,noise_init,true,1-scale_init,1+scale_init,-angle_init*CV_PI/180,angle_init*CV_PI/180,-angle_init*CV_PI/180,angle_init*CV_PI/180);//此函數(shù)根據(jù)傳入的box（目標(biāo)邊界框），在整幀圖像中的全部窗口中尋找與該box距離最小（即最相似，//重疊度最大）的num_closest_init個窗口，然后把這些窗口 歸入good_boxes容器//同時，把重疊度小于0.2的，歸入 bad_boxes 容器//首先根據(jù)overlap的比例信息選出重復(fù)區(qū)域比例大于60%并且前num_closet_init= 10個的最接近box的RectBox，//相當(dāng)于對RectBox進(jìn)行篩選。并通過BBhull函數(shù)得到這些RectBox的最大邊界。getOverlappingBoxes(box, num_closest_init);printf("Found %d good boxes, %d bad boxes\n",(int)good_boxes.size(),(int)bad_boxes.size());printf("Best Box: %d %d %d %d\n",best_box.x, best_box.y, best_box.width, best_box.height);printf("Bounding box hull: %d %d %d %d\n", bbhull.x, bbhull.y, bbhull.width, bbhull.height);//Correct Bounding Boxlastbox=best_box;lastconf=1;lastvalid=true;//Printfprintf(bb_file,"%d,%d,%d,%d,%f\n",lastbox.x,lastbox.y,lastbox.br().x,lastbox.br().y,lastconf);//Prepare Classifier 準(zhǔn)備分類器//scales容器里是所有掃描窗口的尺度，由buildGrid()函數(shù)初始化classifier.prepare(scales);///Generate Data// Generate positive datageneratePositiveData(frame1, num_warps_init);// Set variance thresholdScalar stdev, mean;//統(tǒng)計best_box的均值和標(biāo)準(zhǔn)差例如需要提取圖像A的某個ROI（感興趣區(qū)域，由矩形框）的話，用Mat類的B=img(ROI)即可提取//frame1(best_box)就表示在frame1中提取best_box區(qū)域（目標(biāo)區(qū)域）的圖像片meanStdDev(frame1(best_box), mean, stdev);//利用積分圖像去計算每個待檢測窗口的方差//cvIntegral( const CvArr* image, CvArr* sum, CvArr* sqsum=NULL, CvArr* tilted_sum=NULL );//計算積分圖像，輸入圖像，sum積分圖像, W+1×H+1，sqsum對象素值平方的積分圖像，tilted_sum旋轉(zhuǎn)45度的積分圖像//利用積分圖像，可以計算在某象素的上－右方的或者旋轉(zhuǎn)的矩形區(qū)域中進(jìn)行求和、求均值以及標(biāo)準(zhǔn)方差的計算，//并且保證運算的復(fù)雜度為O(1)。 integral(frame1, iisum, iisqsum);//級聯(lián)分類器模塊一：方差檢測模塊，利用積分圖計算每個待檢測窗口的方差，方差大于var閾值（目標(biāo)patch方差的50%）的，//則認(rèn)為其含有前景目標(biāo)方差；var 為標(biāo)準(zhǔn)差的平方var = pow(stdev.val[0],2) * 0.5; //getVar(best_box,iisum,iisqsum);cout << "variance: " << var << endl;//check variance//getVar函數(shù)通過積分圖像計算輸入的best_box的方差double vr = getVar(best_box, iisum, iisqsum)*0.5;cout << "check variance: " << vr << endl;// Generate negative datagenerateNegativeData(frame1);//Split Negative Ferns into Training and Testing sets (they are already shuffled)//將負(fù)樣本放進(jìn) 訓(xùn)練和測試集int half = (int)nX.size()*0.5f;//vector::assign函數(shù)將區(qū)間[start, end)中的值賦值給當(dāng)前的vector.//將一半的負(fù)樣本集 作為 測試集nXT.assign(nX.begin()+half, nX.end()); //nXT; //negative data to Test//然后將剩下的一半作為訓(xùn)練集nX.resize(half);///Split Negative NN Examples into Training and Testing setshalf = (int)nEx.size()*0.5f;nExT.assign(nEx.begin()+half,nEx.end());nEx.resize(half);//Merge Negative Data with Positive Data and shuffle it//將負(fù)樣本和正樣本合并，然后打亂vector<pair<vector<int>,int> > ferns_data(nX.size()+pX.size());vector<int> idx = index_shuffle(0, ferns_data.size());int a=0;for (int i=0;i<pX.size();i++){ferns_data[idx[a]] = pX[i];a++;}for (int i=0;i<nX.size();i++){ferns_data[idx[a]] = nX[i];a++;}//Data already have been shuffled, just putting it in the same vectorvector<cv::Mat> nn_data(nEx.size()+1);nn_data[0] = pEx;for (int i=0;i<nEx.size();i++){nn_data[i+1]= nEx[i];}///Training //訓(xùn)練 集合分類器（森林） 和 最近鄰分類器 classifier.trainF(ferns_data, 2); //bootstrap = 2classifier.trainNN(nn_data);///Threshold Evaluation on testing sets//用樣本在上面得到的 集合分類器（森林） 和 最近鄰分類器 中分類，評價得到最好的閾值classifier.evaluateTh(nXT, nExT); }/* Generate Positive data* Inputs:* - good_boxes (bbP)* - best_box (bbP0)* - frame (im0)* Outputs:* - Positive fern features (pX)* - Positive NN examples (pEx)*/ void TLD::generatePositiveData(const Mat& frame, int num_warps){/*CvScalar定義可存放1—4個數(shù)值的數(shù)值，常用來存儲像素，其結(jié)構(gòu)體如下：typedef struct CvScalar{double val[4];}CvScalar;如果使用的圖像是1通道的，則s.val[0]中存儲數(shù)據(jù)如果使用的圖像是3通道的，則s.val[0]，s.val[1]，s.val[2]中存儲數(shù)據(jù)*/Scalar mean; //均值Scalar stdev; //標(biāo)準(zhǔn)差//此函數(shù)將frame圖像best_box區(qū)域的圖像片歸一化為均值為0的15*15大小的patch，存在pEx正樣本中g(shù)etPattern(frame(best_box), pEx, mean, stdev);//Get Fern features on warped patchesMat img;Mat warped;//void GaussianBlur(InputArray src, OutputArray dst, Size ksize, double sigmaX, double sigmaY=0, // int borderType=BORDER_DEFAULT ) ;//功能：對輸入的圖像src進(jìn)行高斯濾波后用dst輸出。//src和dst當(dāng)然分別是輸入圖像和輸出圖像。Ksize為高斯濾波器模板大小，sigmaX和sigmaY分別為高斯濾//波在橫向和豎向的濾波系數(shù)。borderType為邊緣擴(kuò)展點插值類型。//用9*9高斯核模糊輸入幀，存入img 去噪？？GaussianBlur(frame, img, Size(9,9), 1.5);//在img圖像中截取bbhull信息（bbhull是包含了位置和大小的矩形框）的圖像賦給warped//例如需要提取圖像A的某個ROI（感興趣區(qū)域，由矩形框）的話，用Mat類的B=img(ROI)即可提取warped = img(bbhull);RNG& rng = theRNG(); //生成一個隨機(jī)數(shù)Point2f pt(bbhull.x + (bbhull.width-1)*0.5f, bbhull.y+(bbhull.height-1)*0.5f); //取矩形框中心的坐標(biāo) int i(2)//nstructs樹木（由一個特征組構(gòu)建，每組特征代表圖像塊的不同視圖表示）的個數(shù)//fern[nstructs] nstructs棵樹的森林的數(shù)組？？vector<int> fern(classifier.getNumStructs());pX.clear();Mat patch;//pX為處理后的RectBox最大邊界處理后的像素信息，pEx最近鄰的RectBox的Pattern，bbP0為最近鄰的RectBox。if (pX.capacity() < num_warps * good_boxes.size())pX.reserve(num_warps * good_boxes.size()); //pX正樣本個數(shù)為 仿射變換個數(shù) * good_box的個數(shù)，故需分配至少這么大的空間int idx;for (int i=0; i< num_warps; i++){if (i>0)//PatchGenerator類用來對圖像區(qū)域進(jìn)行仿射變換，先RNG一個隨機(jī)因子，再調(diào)用（）運算符產(chǎn)生一個變換后的正樣本。generator(frame, pt, warped, bbhull.size(), rng);for (int b=0; b < good_boxes.size(); b++){idx = good_boxes[b]; //good_boxes容器保存的是 grid 的索引patch = img(grid[idx]); //把img的 grid[idx] 區(qū)域（也就是bounding box重疊度高的）這一塊圖像片提取出來//getFeatures函數(shù)得到輸入的patch的用于樹的節(jié)點，也就是特征組的特征fern（13位的二進(jìn)制代碼）classifier.getFeatures(patch, grid[idx].sidx, fern); //grid[idx].sidx 對應(yīng)的尺度索引pX.push_back(make_pair(fern, 1)); //positive ferns <features, labels=1> 正樣本}}printf("Positive examples generated: ferns:%d NN:1\n",(int)pX.size()); }//先對最接近box的RectBox區(qū)域得到其patch ,然后將像素信息轉(zhuǎn)換為Pattern， //具體的說就是歸一化RectBox對應(yīng)的patch的size（放縮至patch_size = 15*15），將2維的矩陣變成一維的向量信息， //然后將向量信息均值設(shè)為0，調(diào)整為zero mean and unit variance（ZMUV） //Output: resized Zero-Mean patch void TLD::getPattern(const Mat& img, Mat& pattern, Scalar& mean, Scalar& stdev){//將img放縮至patch_size = 15*15，存到pattern中resize(img, pattern, Size(patch_size, patch_size));//計算pattern這個矩陣的均值和標(biāo)準(zhǔn)差//Computes a mean value and a standard deviation of matrix elements.meanStdDev(pattern, mean, stdev);pattern.convertTo(pattern, CV_32F);//opencv中Mat的運算符有重載， Mat可以 + Mat; + Scalar; + int / float / double 都可以//將矩陣所有元素減去其均值，也就是把patch的均值設(shè)為零pattern = pattern - mean.val[0]; }/* Inputs:* - Image* - bad_boxes (Boxes far from the bounding box)* - variance (pEx variance)* Outputs* - Negative fern features (nX)* - Negative NN examples (nEx)*/ void TLD::generateNegativeData(const Mat& frame){//由于之前重疊度小于0.2的，都?xì)w入 bad_boxes了，所以數(shù)量挺多，下面的函數(shù)用于打亂順序，也就是為了//后面隨機(jī)選擇bad_boxesrandom_shuffle(bad_boxes.begin(), bad_boxes.end());//Random shuffle bad_boxes indexesint idx;//Get Fern Features of the boxes with big variance (calculated using integral images)int a=0;//int num = std::min((int)bad_boxes.size(),(int)bad_patches*100); //limits the size of bad_boxes to tryprintf("negative data generation started.\n");vector<int> fern(classifier.getNumStructs());nX.reserve(bad_boxes.size());Mat patch;for (int j=0;j<bad_boxes.size();j++){ //把方差較大的bad_boxes加入負(fù)樣本idx = bad_boxes[j];if (getVar(grid[idx],iisum,iisqsum)<var*0.5f)continue;patch = frame(grid[idx]);classifier.getFeatures(patch, grid[idx].sidx, fern);nX.push_back(make_pair(fern, 0)); //得到負(fù)樣本a++;}printf("Negative examples generated: ferns: %d ", a);//random_shuffle(bad_boxes.begin(),bad_boxes.begin()+bad_patches);//Randomly selects 'bad_patches' and get the patterns for NN;Scalar dum1, dum2;//bad_patches = (int)file["num_patches"]; 在參數(shù)文件中 num_patches = 100nEx=vector<Mat>(bad_patches);for (int i=0;i<bad_patches;i++){idx=bad_boxes[i];patch = frame(grid[idx]);//具體的說就是歸一化RectBox對應(yīng)的patch的size（放縮至patch_size = 15*15）//由于負(fù)樣本不需要均值和方差，所以就定義dum，將其舍棄getPattern(patch,nEx[i],dum1,dum2);}printf("NN: %d\n",(int)nEx.size()); }//該函數(shù)通過積分圖像計算輸入的box的方差 double TLD::getVar(const BoundingBox& box, const Mat& sum, const Mat& sqsum){double brs = sum.at<int>(box.y+box.height, box.x+box.width);double bls = sum.at<int>(box.y+box.height, box.x);double trs = sum.at<int>(box.y,box.x + box.width);double tls = sum.at<int>(box.y,box.x);double brsq = sqsum.at<double>(box.y+box.height,box.x+box.width);double blsq = sqsum.at<double>(box.y+box.height,box.x);double trsq = sqsum.at<double>(box.y,box.x+box.width);double tlsq = sqsum.at<double>(box.y,box.x);double mean = (brs+tls-trs-bls)/((double)box.area());double sqmean = (brsq+tlsq-trsq-blsq)/((double)box.area());//方差=E(X^2)-(EX)^2 EX表示均值return sqmean-mean*mean; }void TLD::processFrame(const cv::Mat& img1,const cv::Mat& img2,vector<Point2f>& points1,vector<Point2f>& points2,BoundingBox& bbnext, bool& lastboxfound, bool tl, FILE* bb_file){vector<BoundingBox> cbb;vector<float> cconf;int confident_detections=0;int didx; //detection index///Track 跟蹤模塊if(lastboxfound && tl){ //tl: train and learn//跟蹤track(img1, img2, points1, points2);}else{tracked = false;}///Detect 檢測模塊detect(img2);///Integration 綜合模塊//TLD只跟蹤單目標(biāo)，所以綜合模塊綜合跟蹤器跟蹤到的單個目標(biāo)和檢測器檢測到的多個目標(biāo)，然后只輸出保守相似度最大的一個目標(biāo)if (tracked){bbnext=tbb;lastconf=tconf; //表示相關(guān)相似度的閾值lastvalid=tvalid; //表示保守相似度的閾值printf("Tracked\n");if(detected){ // if Detected//通過 重疊度 對檢測器檢測到的目標(biāo)bounding box進(jìn)行聚類，每個類其重疊度小于0.5clusterConf(dbb, dconf, cbb, cconf); // cluster detectionsprintf("Found %d clusters\n",(int)cbb.size());for (int i=0;i<cbb.size();i++){//找到與跟蹤器跟蹤到的box距離比較遠(yuǎn)的類（檢測器檢測到的box），而且它的相關(guān)相似度比跟蹤器的要大if (bbOverlap(tbb, cbb[i])<0.5 && cconf[i]>tconf){ // Get index of a clusters that is far from tracker and are more confident than the trackerconfident_detections++; //記錄滿足上述條件，也就是可信度比較高的目標(biāo)box的個數(shù)didx=i; //detection index}}//如果只有一個滿足上述條件的box，那么就用這個目標(biāo)box來重新初始化跟蹤器（也就是用檢測器的結(jié)果去糾正跟蹤器）if (confident_detections==1){ //if there is ONE such a cluster, re-initialize the trackerprintf("Found a better match..reinitializing tracking\n");bbnext=cbb[didx];lastconf=cconf[didx];lastvalid=false;}else {printf("%d confident cluster was found\n", confident_detections);int cx=0,cy=0,cw=0,ch=0;int close_detections=0;for (int i=0;i<dbb.size();i++){//找到檢測器檢測到的box與跟蹤器預(yù)測到的box距離很近（重疊度大于0.7）的box，對其坐標(biāo)和大小進(jìn)行累加if(bbOverlap(tbb,dbb[i])>0.7){ // Get mean of close detectionscx += dbb[i].x;cy +=dbb[i].y;cw += dbb[i].width;ch += dbb[i].height;close_detections++; //記錄最近鄰box的個數(shù)printf("weighted detection: %d %d %d %d\n",dbb[i].x,dbb[i].y,dbb[i].width,dbb[i].height);}}if (close_detections>0){//對與跟蹤器預(yù)測到的box距離很近的box 和 跟蹤器本身預(yù)測到的box 進(jìn)行坐標(biāo)與大小的平均作為最終的//目標(biāo)bounding box，但是跟蹤器的權(quán)值較大bbnext.x = cvRound((float)(10*tbb.x+cx)/(float)(10+close_detections)); // weighted average trackers trajectory with the close detectionsbbnext.y = cvRound((float)(10*tbb.y+cy)/(float)(10+close_detections));bbnext.width = cvRound((float)(10*tbb.width+cw)/(float)(10+close_detections));bbnext.height = cvRound((float)(10*tbb.height+ch)/(float)(10+close_detections));printf("Tracker bb: %d %d %d %d\n",tbb.x,tbb.y,tbb.width,tbb.height);printf("Average bb: %d %d %d %d\n",bbnext.x,bbnext.y,bbnext.width,bbnext.height);printf("Weighting %d close detection(s) with tracker..\n",close_detections);}else{printf("%d close detections were found\n",close_detections);}}}}else{ // If NOT trackingprintf("Not tracking..\n");lastboxfound = false;lastvalid = false;//如果跟蹤器沒有跟蹤到目標(biāo)，但是檢測器檢測到了一些可能的目標(biāo)box，那么同樣對其進(jìn)行聚類，但只是簡單的//將聚類的cbb[0]作為新的跟蹤目標(biāo)box（不比較相似度了？？還是里面已經(jīng)排好序了？？），重新初始化跟蹤器if(detected){ // and detector is definedclusterConf(dbb,dconf,cbb,cconf); // cluster detectionsprintf("Found %d clusters\n",(int)cbb.size());if (cconf.size()==1){bbnext=cbb[0];lastconf=cconf[0];printf("Confident detection..reinitializing tracker\n");lastboxfound = true;}}}lastbox=bbnext;if (lastboxfound)fprintf(bb_file,"%d,%d,%d,%d,%f\n",lastbox.x,lastbox.y,lastbox.br().x,lastbox.br().y,lastconf);elsefprintf(bb_file,"NaN,NaN,NaN,NaN,NaN\n");///learn 學(xué)習(xí)模塊if (lastvalid && tl)learn(img2); }/*Inputs: * -current frame(img2), last frame(img1), last Bbox(bbox_f[0]). *Outputs: *- Confidence(tconf), Predicted bounding box(tbb), Validity(tvalid), points2 (for display purposes only) */ void TLD::track(const Mat& img1, const Mat& img2, vector<Point2f>& points1, vector<Point2f>& points2){//Generate points//網(wǎng)格均勻撒點（均勻采樣），在lastbox中共產(chǎn)生最多10*10=100個特征點，存于points1bbPoints(points1, lastbox);if (points1.size()<1){printf("BB= %d %d %d %d, Points not generated\n",lastbox.x,lastbox.y,lastbox.width,lastbox.height);tvalid=false;tracked=false;return;}vector<Point2f> points = points1;//Frame-to-frame tracking with forward-backward error cheking//trackf2f函數(shù)完成：跟蹤、計算FB error和匹配相似度sim，然后篩選出 FB_error[i] <= median(FB_error) 和 //sim_error[i] > median(sim_error) 的特征點（跟蹤結(jié)果不好的特征點），剩下的是不到50%的特征點tracked = tracker.trackf2f(img1, img2, points, points2);if (tracked){//Bounding box prediction//利用剩下的這不到一半的跟蹤點輸入來預(yù)測bounding box在當(dāng)前幀的位置和大小 tbbbbPredict(points, points2, lastbox, tbb);//跟蹤失敗檢測：如果FB error的中值大于10個像素（經(jīng)驗值），或者預(yù)測到的當(dāng)前box的位置移出圖像，則//認(rèn)為跟蹤錯誤，此時不返回bounding box；Rect::br()返回的是右下角的坐標(biāo)//getFB()返回的是FB error的中值if (tracker.getFB()>10 || tbb.x>img2.cols || tbb.y>img2.rows || tbb.br().x < 1 || tbb.br().y <1){tvalid =false; //too unstable prediction or bounding box out of imagetracked = false;printf("Too unstable predictions FB error=%f\n", tracker.getFB());return;}//Estimate Confidence and Validity//評估跟蹤確信度和有效性Mat pattern;Scalar mean, stdev;BoundingBox bb;bb.x = max(tbb.x,0);bb.y = max(tbb.y,0);bb.width = min(min(img2.cols-tbb.x,tbb.width), min(tbb.width, tbb.br().x));bb.height = min(min(img2.rows-tbb.y,tbb.height),min(tbb.height,tbb.br().y));//歸一化img2(bb)對應(yīng)的patch的size（放縮至patch_size = 15*15），存入patterngetPattern(img2(bb),pattern,mean,stdev);vector<int> isin;float dummy;//計算圖像片pattern到在線模型M的保守相似度classifier.NNConf(pattern,isin,dummy,tconf); //Conservative Similaritytvalid = lastvalid;//保守相似度大于閾值，則評估跟蹤有效if (tconf>classifier.thr_nn_valid){tvalid =true;}}elseprintf("No points tracked\n");}//網(wǎng)格均勻撒點，box共10*10=100個特征點 void TLD::bbPoints(vector<cv::Point2f>& points, const BoundingBox& bb){int max_pts=10;int margin_h=0; //采樣邊界int margin_v=0;//網(wǎng)格均勻撒點int stepx = ceil((bb.width-2*margin_h)/max_pts); //ceil返回大于或者等于指定表達(dá)式的最小整數(shù)int stepy = ceil((bb.height-2*margin_v)/max_pts);//網(wǎng)格均勻撒點，box共10*10=100個特征點for (int y=bb.y+margin_v; y<bb.y+bb.height-margin_v; y+=stepy){for (int x=bb.x+margin_h;x<bb.x+bb.width-margin_h;x+=stepx){points.push_back(Point2f(x,y));}} }//利用剩下的這不到一半的跟蹤點輸入來預(yù)測bounding box在當(dāng)前幀的位置和大小 void TLD::bbPredict(const vector<cv::Point2f>& points1,const vector<cv::Point2f>& points2,const BoundingBox& bb1,BoundingBox& bb2) {int npoints = (int)points1.size();vector<float> xoff(npoints); //位移vector<float> yoff(npoints);printf("tracked points : %d\n", npoints);for (int i=0;i<npoints;i++){ //計算每個特征點在兩幀之間的位移xoff[i]=points2[i].x - points1[i].x;yoff[i]=points2[i].y - points1[i].y;}float dx = median(xoff); //計算位移的中值float dy = median(yoff);float s;//計算bounding box尺度scale的變化：通過計算 當(dāng)前特征點相互間的距離 與 先前（上一幀）特征點相互間的距離 的//比值，以比值的中值作為尺度的變化因子if (npoints>1){vector<float> d;d.reserve(npoints*(npoints-1)/2); //等差數(shù)列求和：1+2+...+(npoints-1)for (int i=0;i<npoints;i++){for (int j=i+1;j<npoints;j++){//計算 當(dāng)前特征點相互間的距離 與 先前（上一幀）特征點相互間的距離 的比值（位移用絕對值）d.push_back(norm(points2[i]-points2[j])/norm(points1[i]-points1[j]));}}s = median(d);}else {s = 1.0;}float s1 = 0.5*(s-1)*bb1.width;float s2 = 0.5*(s-1)*bb1.height;printf("s= %f s1= %f s2= %f \n", s, s1, s2);//得到當(dāng)前bounding box的位置與大小信息//當(dāng)前box的x坐標(biāo) = 前一幀box的x坐標(biāo) + 全部特征點位移的中值（可理解為box移動近似的位移） - 當(dāng)前box寬的一半bb2.x = round( bb1.x + dx - s1);bb2.y = round( bb1.y + dy -s2);bb2.width = round(bb1.width*s);bb2.height = round(bb1.height*s);printf("predicted bb: %d %d %d %d\n",bb2.x,bb2.y,bb2.br().x,bb2.br().y); }void TLD::detect(const cv::Mat& frame){//cleaningdbb.clear();dconf.clear();dt.bb.clear();//GetTickCount返回從操作系統(tǒng)啟動到現(xiàn)在所經(jīng)過的時間double t = (double)getTickCount();Mat img(frame.rows, frame.cols, CV_8U);integral(frame,iisum,iisqsum); //計算frame的積分圖 GaussianBlur(frame,img,Size(9,9),1.5); //高斯模糊，去噪？int numtrees = classifier.getNumStructs();float fern_th = classifier.getFernTh(); //getFernTh()返回thr_fern; 集合分類器的分類閾值vector <int> ferns(10);float conf;int a=0;Mat patch;//級聯(lián)分類器模塊一：方差檢測模塊，利用積分圖計算每個待檢測窗口的方差，方差大于var閾值（目標(biāo)patch方差的50%）的，//則認(rèn)為其含有前景目標(biāo)for (int i=0; i<grid.size(); i++){ //FIXME: BottleNeck 瓶頸if (getVar(grid[i],iisum,iisqsum) >= var){ //計算每一個掃描窗口的方差a++;//級聯(lián)分類器模塊二：集合分類器檢測模塊patch = img(grid[i]);classifier.getFeatures(patch,grid[i].sidx,ferns); //得到該patch特征（13位的二進(jìn)制代碼）conf = classifier.measure_forest(ferns); //計算該特征值對應(yīng)的后驗概率累加值tmp.conf[i]=conf; //Detector data中定義TempStruct tmp; tmp.patt[i]=ferns;//如果集合分類器的后驗概率的平均值大于閾值fern_th（由訓(xùn)練得到），就認(rèn)為含有前景目標(biāo)if (conf > numtrees*fern_th){ dt.bb.push_back(i); //將通過以上兩個檢測模塊的掃描窗口記錄在detect structure中}}elsetmp.conf[i]=0.0;}int detections = dt.bb.size();printf("%d Bounding boxes passed the variance filter\n",a);printf("%d Initial detection from Fern Classifier\n", detections);//如果通過以上兩個檢測模塊的掃描窗口數(shù)大于100個，則只取后驗概率大的前100個if (detections>100){ //CComparator(tmp.conf)指定比較方式？？？nth_element(dt.bb.begin(), dt.bb.begin()+100, dt.bb.end(), CComparator(tmp.conf));dt.bb.resize(100);detections=100;} // for (int i=0;i<detections;i++){ // drawBox(img,grid[dt.bb[i]]); // } // imshow("detections",img);if (detections==0){detected=false;return;}printf("Fern detector made %d detections ",detections);//兩次使用getTickCount()，然后再除以getTickFrequency()，計算出來的是以秒s為單位的時間（opencv 2.0 以前是ms）t=(double)getTickCount()-t; printf("in %gms\n", t*1000/getTickFrequency()); //打印以上代碼運行使用的毫秒數(shù)// Initialize detection structuredt.patt = vector<vector<int> >(detections,vector<int>(10,0)); // Corresponding codes of the Ensemble Classifierdt.conf1 = vector<float>(detections); // Relative Similarity (for final nearest neighbour classifier)dt.conf2 =vector<float>(detections); // Conservative Similarity (for integration with tracker)dt.isin = vector<vector<int> >(detections,vector<int>(3,-1)); // Detected (isin=1) or rejected (isin=0) by nearest neighbour classifierdt.patch = vector<Mat>(detections,Mat(patch_size,patch_size,CV_32F));// Corresponding patchesint idx;Scalar mean, stdev;float nn_th = classifier.getNNTh();//級聯(lián)分類器模塊三：最近鄰分類器檢測模塊for (int i=0;i<detections;i++){ // for every remaining detectionidx=dt.bb[i]; // Get the detected bounding box indexpatch = frame(grid[idx]);getPattern(patch,dt.patch[i],mean,stdev); // Get pattern within bounding box//計算圖像片pattern到在線模型M的相關(guān)相似度和保守相似度classifier.NNConf(dt.patch[i],dt.isin[i],dt.conf1[i],dt.conf2[i]); // Evaluate nearest neighbour classifierdt.patt[i]=tmp.patt[idx];//printf("Testing feature %d, conf:%f isin:(%d|%d|%d)\n",i,dt.conf1[i],dt.isin[i][0],dt.isin[i][1],dt.isin[i][2]);//相關(guān)相似度大于閾值，則認(rèn)為含有前景目標(biāo)if (dt.conf1[i]>nn_th){ // idx = dt.conf1 > tld.model.thr_nn; % get all indexes that made it through the nearest neighbourdbb.push_back(grid[idx]); // BB = dt.bb(:,idx); % bounding boxesdconf.push_back(dt.conf2[i]); // Conf = dt.conf2(:,idx); % conservative confidences}}//打印檢測到的可能存在目標(biāo)的掃描窗口數(shù)（可以通過三個級聯(lián)檢測器的）if (dbb.size()>0){printf("Found %d NN matches\n",(int)dbb.size());detected=true;}else{printf("No NN matches found.\n");detected=false;} }//作者已經(jīng)用python腳本../datasets/evaluate_vis.py來完成算法評估功能，具體見README void TLD::evaluate(){ }void TLD::learn(const Mat& img){printf("[Learning] ");///Check consistency//檢測一致性BoundingBox bb;bb.x = max(lastbox.x,0);bb.y = max(lastbox.y,0);bb.width = min(min(img.cols-lastbox.x,lastbox.width),min(lastbox.width,lastbox.br().x));bb.height = min(min(img.rows-lastbox.y,lastbox.height),min(lastbox.height,lastbox.br().y));Scalar mean, stdev;Mat pattern;//歸一化img(bb)對應(yīng)的patch的size（放縮至patch_size = 15*15），存入patterngetPattern(img(bb), pattern, mean, stdev);vector<int> isin;float dummy, conf;//計算輸入圖像片（跟蹤器的目標(biāo)box）與在線模型之間的相關(guān)相似度confclassifier.NNConf(pattern,isin,conf,dummy);if (conf<0.5) { //如果相似度太小了，就不訓(xùn)練printf("Fast change..not training\n");lastvalid =false;return;}if (pow(stdev.val[0], 2)< var){ //如果方差太小了，也不訓(xùn)練printf("Low variance..not training\n");lastvalid=false;return;}if(isin[2]==1){ //如果被被識別為負(fù)樣本，也不訓(xùn)練printf("Patch in negative data..not traing");lastvalid=false;return;}/// Data generation 樣本產(chǎn)生for (int i=0;i<grid.size();i++){ //計算所有的掃描窗口與目標(biāo)box的重疊度grid[i].overlap = bbOverlap(lastbox, grid[i]);}//集合分類器vector<pair<vector<int>,int> > fern_examples;good_boxes.clear(); bad_boxes.clear();//此函數(shù)根據(jù)傳入的lastbox，在整幀圖像中的全部窗口中尋找與該lastbox距離最小（即最相似，//重疊度最大）的num_closest_update個窗口，然后把這些窗口 歸入good_boxes容器（只是把網(wǎng)格數(shù)組的索引存入）//同時，把重疊度小于0.2的，歸入 bad_boxes 容器getOverlappingBoxes(lastbox, num_closest_update);if (good_boxes.size()>0)generatePositiveData(img, num_warps_update); //用仿射模型產(chǎn)生正樣本（類似于第一幀的方法，但只產(chǎn)生10*10=100個）else{lastvalid = false;printf("No good boxes..Not training");return;}fern_examples.reserve(pX.size() + bad_boxes.size());fern_examples.assign(pX.begin(), pX.end());int idx;for (int i=0;i<bad_boxes.size();i++){idx=bad_boxes[i];if (tmp.conf[idx]>=1){ //加入負(fù)樣本，相似度大于1？？相似度不是出于0和1之間嗎？fern_examples.push_back(make_pair(tmp.patt[idx],0));}}//最近鄰分類器vector<Mat> nn_examples;nn_examples.reserve(dt.bb.size()+1);nn_examples.push_back(pEx);for (int i=0;i<dt.bb.size();i++){idx = dt.bb[i];if (bbOverlap(lastbox,grid[idx]) < bad_overlap)nn_examples.push_back(dt.patch[i]);}/// Classifiers update 分類器訓(xùn)練classifier.trainF(fern_examples,2);classifier.trainNN(nn_examples);classifier.show(); //把正樣本庫（在線模型）包含的所有正樣本顯示在窗口上 }//檢測器采用掃描窗口的策略 //此函數(shù)根據(jù)傳入的box（目標(biāo)邊界框）在傳入的圖像中構(gòu)建全部的掃描窗口，并計算每個窗口與box的重疊度 void TLD::buildGrid(const cv::Mat& img, const cv::Rect& box){const float SHIFT = 0.1; //掃描窗口步長為 寬高的 10%//尺度縮放系數(shù)為1.2 （0.16151*1.2=0.19381），共21種尺度變換const float SCALES[] = {0.16151,0.19381,0.23257,0.27908,0.33490,0.40188,0.48225,0.57870,0.69444,0.83333,1,1.20000,1.44000,1.72800,2.07360,2.48832,2.98598,3.58318,4.29982,5.15978,6.19174};int width, height, min_bb_side;//Rect bbox;BoundingBox bbox;Size scale;int sc=0;for (int s=0; s < 21; s++){width = round(box.width*SCALES[s]);height = round(box.height*SCALES[s]);min_bb_side = min(height,width); //bounding box最短的邊//由于圖像片（min_win 為15x15像素）是在bounding box中采樣得到的，所以box必須比min_win要大//另外，輸入的圖像肯定得比 bounding box 要大了if (min_bb_side < min_win || width > img.cols || height > img.rows)continue;scale.width = width;scale.height = height;//push_back在vector類中作用為在vector尾部加入一個數(shù)據(jù)//scales在類TLD中定義：std::vector<cv::Size> scales;scales.push_back(scale); //把該尺度的窗口存入scales容器，避免在掃描時計算，加快檢測速度for (int y=1; y<img.rows-height; y+=round(SHIFT*min_bb_side)){ //按步長移動窗口for (int x=1; x<img.cols-width; x+=round(SHIFT*min_bb_side)){bbox.x = x;bbox.y = y;bbox.width = width;bbox.height = height;//判斷傳入的bounding box（目標(biāo)邊界框）與 傳入圖像中的此時窗口的 重疊度，//以此來確定該圖像窗口是否含有目標(biāo)bbox.overlap = bbOverlap(bbox, BoundingBox(box));bbox.sidx = sc; //屬于第幾個尺度//grid在類TLD中定義：std::vector<BoundingBox> grid;//把本位置和本尺度的掃描窗口存入grid容器grid.push_back(bbox);}}sc++;} }//此函數(shù)計算兩個bounding box 的重疊度 //重疊度定義為 兩個box的交集 與 它們的并集 的比 float TLD::bbOverlap(const BoundingBox& box1, const BoundingBox& box2){//先判斷坐標(biāo)，假如它們都沒有重疊的地方，就直接返回0if (box1.x > box2.x + box2.width) { return 0.0; }if (box1.y > box2.y + box2.height) { return 0.0; }if (box1.x + box1.width < box2.x) { return 0.0; }if (box1.y + box1.height < box2.y) { return 0.0; }float colInt = min(box1.x + box1.width, box2.x + box2.width) - max(box1.x, box2.x);float rowInt = min(box1.y + box1.height, box2.y + box2.height) - max(box1.y, box2.y);float intersection = colInt * rowInt;float area1 = box1.width * box1.height;float area2 = box2.width * box2.height;return intersection / (area1 + area2 - intersection); }//此函數(shù)根據(jù)傳入的box1（目標(biāo)邊界框），在整幀圖像中的全部窗口中尋找與該box1距離最小（即最相似， //重疊度最大）的num_closest個窗口，然后把這些窗口 歸入good_boxes容器（只是把網(wǎng)格數(shù)組的索引存入） //同時，把重疊度小于0.2的，歸入 bad_boxes 容器 void TLD::getOverlappingBoxes(const cv::Rect& box1,int num_closest){float max_overlap = 0;for (int i=0;i<grid.size();i++){if (grid[i].overlap > max_overlap) { //找出重疊度最大的boxmax_overlap = grid[i].overlap;best_box = grid[i]; }if (grid[i].overlap > 0.6){ //重疊度大于0.6的，歸入 good_boxesgood_boxes.push_back(i);}else if (grid[i].overlap < bad_overlap){ //重疊度小于0.2的，歸入 bad_boxesbad_boxes.push_back(i);}}//Get the best num_closest (10) boxes and puts them in good_boxesif (good_boxes.size()>num_closest){//STL中的nth_element()方法找出一個數(shù)列中排名第n（下面為第num_closest）的那個數(shù)。這個函數(shù)運行后//在good_boxes[num_closest]前面num_closest個數(shù)都比他大，也就是找到最好的num_closest個box了std::nth_element(good_boxes.begin(), good_boxes.begin() + num_closest, good_boxes.end(), OComparator(grid));//重新壓縮good_boxes為num_closest大小good_boxes.resize(num_closest);}//獲取good_boxes 的 Hull殼，也就是窗口的邊框getBBHull(); }//此函數(shù)獲取good_boxes 的 Hull殼，也就是窗口（圖像）的邊框 bounding box void TLD::getBBHull(){int x1=INT_MAX, x2=0; //INT_MAX 最大的整形數(shù)int y1=INT_MAX, y2=0;int idx;for (int i=0;i<good_boxes.size();i++){idx= good_boxes[i];x1=min(grid[idx].x,x1); //防止出現(xiàn)負(fù)數(shù)？？y1=min(grid[idx].y,y1);x2=max(grid[idx].x + grid[idx].width,x2);y2=max(grid[idx].y + grid[idx].height,y2);}bbhull.x = x1;bbhull.y = y1;bbhull.width = x2-x1;bbhull.height = y2 -y1; }//如果兩個box的重疊度小于0.5，返回false，否則返回true bool bbcomp(const BoundingBox& b1,const BoundingBox& b2){TLD t;if (t.bbOverlap(b1,b2)<0.5)return false;elsereturn true; }int TLD::clusterBB(const vector<BoundingBox>& dbb,vector<int>& indexes){//FIXME: Conditional jump or move depends on uninitialised value(s)const int c = dbb.size();//1. Build proximity matrixMat D(c,c,CV_32F);float d;for (int i=0;i<c;i++){for (int j=i+1;j<c;j++){d = 1-bbOverlap(dbb[i],dbb[j]);D.at<float>(i,j) = d;D.at<float>(j,i) = d;}}//2. Initialize disjoint clusteringfloat L[c-1]; //Levelint nodes[c-1][2];int belongs[c];int m=c;for (int i=0;i<c;i++){belongs[i]=i;}for (int it=0;it<c-1;it++){//3. Find nearest neighborfloat min_d = 1;int node_a, node_b;for (int i=0;i<D.rows;i++){for (int j=i+1;j<D.cols;j++){if (D.at<float>(i,j)<min_d && belongs[i]!=belongs[j]){min_d = D.at<float>(i,j);node_a = i;node_b = j;}}}if (min_d>0.5){int max_idx =0;bool visited;for (int j=0;j<c;j++){visited = false;for(int i=0;i<2*c-1;i++){if (belongs[j]==i){indexes[j]=max_idx;visited = true;}}if (visited)max_idx++;}return max_idx;}//4. Merge clusters and assign levelL[m]=min_d;nodes[it][0] = belongs[node_a];nodes[it][1] = belongs[node_b];for (int k=0;k<c;k++){if (belongs[k]==belongs[node_a] || belongs[k]==belongs[node_b])belongs[k]=m;}m++;}return 1;}//對檢測器檢測到的目標(biāo)bounding box進(jìn)行聚類 //聚類（Cluster）分析是由若干模式（Pattern）組成的，通常，模式是一個度量（Measurement）的向量，或者是多維空間中的 //一個點。聚類分析以相似性為基礎(chǔ)，在一個聚類中的模式之間比不在同一聚類中的模式之間具有更多的相似性。 void TLD::clusterConf(const vector<BoundingBox>& dbb,const vector<float>& dconf,vector<BoundingBox>& cbb,vector<float>& cconf){int numbb =dbb.size();vector<int> T;float space_thr = 0.5;int c=1; //記錄 聚類的類個數(shù)switch (numbb){ //檢測到的含有目標(biāo)的bounding box個數(shù)case 1:cbb=vector<BoundingBox>(1,dbb[0]); //如果只檢測到一個，那么這個就是檢測器檢測到的目標(biāo)cconf=vector<float>(1,dconf[0]);return;break;case 2:T =vector<int>(2,0);//此函數(shù)計算兩個bounding box 的重疊度if (1 - bbOverlap(dbb[0],dbb[1]) > space_thr){ //如果只檢測到兩個box，但他們的重疊度小于0.5T[1]=1;c=2; //重疊度小于0.5的box，屬于不同的類}break;default: //檢測到的box數(shù)目大于2個，則篩選出重疊度大于0.5的T = vector<int>(numbb, 0);//stable_partition()重新排列元素，使得滿足指定條件的元素排在不滿足條件的元素前面。它維持著兩組元素的順序關(guān)系。//STL partition就是把一個區(qū)間中的元素按照某個條件分成兩類。返回第二類子集的起點//bbcomp()函數(shù)判斷兩個box的重疊度小于0.5，返回false，否則返回true （分界點是重疊度：0.5）//partition() 將dbb劃分為兩個子集，將滿足兩個box的重疊度小于0.5的元素移動到序列的前面，為一個子集，重疊度大于0.5的，//放在序列后面，為第二個子集，但兩個子集的大小不知道，返回第二類子集的起點c = partition(dbb, T, (*bbcomp)); //重疊度小于0.5的box，屬于不同的類，所以c是不同的類別個數(shù)//c = clusterBB(dbb,T);break;}cconf=vector<float>(c); cbb=vector<BoundingBox>(c);printf("Cluster indexes: ");BoundingBox bx;for (int i=0;i<c;i++){ //類別個數(shù)float cnf=0;int N=0,mx=0,my=0,mw=0,mh=0;for (int j=0;j<T.size();j++){ //檢測到的bounding box個數(shù)if (T[j]==i){ //將聚類為同一個類別的box的坐標(biāo)和大小進(jìn)行累加printf("%d ",i);cnf=cnf+dconf[j];mx=mx+dbb[j].x;my=my+dbb[j].y;mw=mw+dbb[j].width;mh=mh+dbb[j].height;N++;}}if (N>0){ //然后求該類的box的坐標(biāo)和大小的平均值，將平均值作為該類的box的代表cconf[i]=cnf/N;bx.x=cvRound(mx/N);bx.y=cvRound(my/N);bx.width=cvRound(mw/N);bx.height=cvRound(mh/N);cbb[i]=bx; //返回的是聚類，每一個類都有一個代表的bounding box}}printf("\n"); }

總結(jié)

以上是生活随笔為你收集整理的TLD（Tracking-Learning-Detection）学习与源码理解之（六）的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。