行人检测（一）hog

OpenCV HOG PeopleDetector：从训练到检测

转载：http://blog.youtueye.com/work/opencv-hog-peopledetector-trainning.html

opencv2.0 以上版本提供了行人检测的方法，以opencv2.2为例，该示例为opencv安装目录下的：/samples/cpp/peopledetect.cpp，测试效果如下：

我们先看看示例代码的内容：

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111

using namespace cv; using namespace std;   void help() {      printf (              "\nDemonstrate the use of the HoG descriptor using\n"              "  HOGDescriptor::hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());\n"              "Usage:\n"              "./peopledetect (<image_filename> | <image_list>.txt)\n\n" ); }   int main( int argc, char ** argv) {      Mat img;      FILE * f = 0;      char _filename[1024];        if ( argc == 1 )      {          printf ( "Usage: peopledetect (<image_filename> | <image_list>.txt)\n" );          return 0;      }      img = imread(argv[1]);        if ( img.data )      {          strcpy (_filename, argv[1]);      }      else      {          f = fopen (argv[1], "rt" );          if (!f)          {              fprintf ( stderr, "ERROR: the specified file could not be loaded\n" );              return -1;          }      }        HOGDescriptor hog;      hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());      namedWindow( "people detector" , 1);        for (;;)      {          char * filename = _filename;          if (f)          {              if (! fgets (filename, ( int ) sizeof (_filename)-2, f))                  break ;              //while(*filename && isspace(*filename))              //  ++filename;              if (filename[0] == '#' )                  continue ;              int l = strlen (filename);              while (l > 0 && isspace (filename[l-1]))                  --l;              filename[l] = '\0' ;              img = imread(filename);          }          printf ( "%s:\n" , filename);          if (!img.data)              continue ;            fflush (stdout);          vector<Rect> found, found_filtered;          double t = ( double )getTickCount();          // run the detector with default parameters. to get a higher hit-rate          // (and more false alarms, respectively), decrease the hitThreshold and          // groupThreshold (set groupThreshold to 0 to turn off the grouping completely).          hog.detectMultiScale(img, found, 0, Size(8,8), Size(32,32), 1.05, 2);          t = ( double )getTickCount() - t;          printf ( "tdetection time = %gms\n" , t*1000./cv::getTickFrequency());          size_t i, j;          for ( i = 0; i < found.size(); i++ )          {              Rect r = found[i];              for ( j = 0; j < found.size(); j++ )                  if ( j != i && (r & found[j]) == r)                      break ;              if ( j == found.size() )                  found_filtered.push_back(r);          }          for ( i = 0; i < found_filtered.size(); i++ )          {              Rect r = found_filtered[i];              // the HOG detector returns slightly larger rectangles than the real objects.              // so we slightly shrink the rectangles to get a nicer output.              r.x += cvRound(r.width*0.1);              r.width = cvRound(r.width*0.8);              r.y += cvRound(r.height*0.07);              r.height = cvRound(r.height*0.8);              rectangle(img, r.tl(), r.br(), cv::Scalar(0,255,0), 3);          }          imshow( "people detector" , img);          int c = waitKey(0) & 255;          if ( c == 'q' || c == 'Q' || !f)              break ;      }      if (f)          fclose (f);      return 0; }

你会发现代码中没有类似人脸检测时的分类器（模型）的载入过程，而是用

1	hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());

载入默认模型。找到hog的源码，在/modules/objdetect/hog.cpp，从第907行开始的代码如下：

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

vector< float > HOGDescriptor::getDefaultPeopleDetector() {      static const float detector[] = {         0.05359386f, -0.14721455f, -0.05532170f, 0.05077307f,         0.11547081f, -0.04268804f, 0.04635834f, -0.05468199f, 0.08232084f,         0.10424068f, -0.02294518f, 0.01108519f, 0.01378693f, 0.11193510f,         0.01268418f, 0.08528346f, -0.06309239f, 0.13054633f, 0.08100729f,         -0.05209739f, -0.04315529f, 0.09341384f, 0.11035026f, -0.07596218f,         -0.05517511f, -0.04465296f, 0.02947334f, 0.04555536f,         -3.55954492e-003f, 0.07818956f, 0.07730991f, 0.07890715f, 0.06222893f,         0.09001380f, -0.03574381f, 0.03414327f, 0.05677258f, -0.04773581f,         0.03746637f, -0.03521175f, 0.06955440f, -0.03849038f, 0.01052293f,         0.01736112f, 0.10867710f, 0.08748853f, 3.29739624e-003f, 0.10907028f,         0.07913758f, 0.10393070f, 0.02091867f, 0.11594022f, 0.13182420f,         0.09879354f, 0.05362710f, -0.06745391f, -7.01260753e-003f,....

显然，行人检测所用到的模型数据被做为常量写在源代码中，但这些模型数据数据是如何的来的呢？如果我要用我自己的样本tainning ，然后用的到的model去detect该如何操作呢？别急，我们从opencv行人检测所用到的算法说起。

opencv行人检测所用到的算法源自Navneet Dalal和Bill Triggs在2005's CVPR上的文章Histograms of Oriented Gradients for Human Detection.这里是英文和中文的介绍。作者所在的研究机构(INRIA：French National Institute for Research in Computer Science and Control,法国国家计算机技术和控制研究所)发布了这套算法的源码：INRIA Object Detection and Localization Toolkit.

那么，opencv中的PeopleDetector是否就原封不动地使用了这套算法呢，为了求证这一问题，笔者首先比较了两者的模型数据：下载OLTbinaries.zip，找到/HOG/model_4BiSVMLight.alt文件（二进制形式的数据），用代码将其中数据保存为文本格式（这个要根据上面的源码learcode.zip），其内容如下：

1 2 3 4 5 6 7 8 9 10 11 12

0.05359386 -0.14721455 -0.05532170 0.05077307 0.11547081 -0.04268804 0.04635834 -0.05468199 0.08232084 0.10424068 -0.02294518 0.01108519 0.01378693 0.11193510 0.01268418 0.08528346 -0.06309239 0.13054632 0.08100729 -0.05209739 -0.04315529 0.09341384 0.11035026 -0.07596217 -0.05517511 -0.04465296 0.02947334 0.04555536 -0.00355954 0.07818956 0.07730991 0.07890716 0.06222893 0.09001381 -0.03574381 0.03414327 0.05677258 -0.04773581 0.03746637 -0.03521175 0.06955440 -0.03849038 0.01052293 0.01736112 0.10867710 0.08748853 0.00329740 0.10907028 0.07913758 0.10393069 0.02091867 0.11594022 0.13182420 0.09879354 0.05362710 -0.06745391 -0.00701261 0.00524702 0.03236255 0.01407916 0.02207983 0.02537322 0.04547948 0.07200756 0.03129894 -0.06274468 0.02107014 0.06035208 0.08636236 0.00453164 0.02193363 0.02309801 0.05568166 -0.02645093 0.04448695 0.02837519 0.08975695 0.04461517 0.08975355 0.07514391 0.02306982 0.10410084 0.06368385 0.05943465 0.00458421 0.05220337 0.06675851 0.08358569 0.06712102 0.06559004 -0.03930481 -0.00915937 -0.05897915 0.02816453 0.05032348 0.06780671 0.03377650 -0.00060942 -0.01795146 -0.03083684 -0.01302475 -0.02972313 0.00788707 -0.03525961 -0.00250398 0.05245084 0.11791293 -0.02167498 0.05299332 0.06640524 0.05190265 -0.00827317 0.03033127 0.05842173 -0.00401050 -0.00625106 0.05862958 -0.02465461 0.05546781 -0.08228195 ....

数值居然同HOGDescriptor::getDefaultPeopleDetector()中的是一样！那么，你就可以用上面的OLT去trainning你的样本，然后把得到的model数据替换getDefaultPeopleDetector()中的数据，你便可以去进行你要的目标检测啦！

为了验证这一想法的正确性和可行性，笔者做了些实验，在Ubuntu10.4 g++4.4.5环境中，步骤如下：

1. 下载release版的程序：OLTbinaries.zip
2. 下载样本：INRIAPerson.tar
3. 在目录OLTbinaries/下建立两个文件夹test, train. 将INRIAperson/Test/neg拷贝到test/下，INRIAperson/Train/neg拷贝到train/下;将INRIAperson/test_64x128_H96拷贝到test/下重命名为pos，将INRIAperson/train_64x128_H96拷贝到train/下重命名为pos;
4. 将test/neg , test/pos各自文件夹中的所有图片文件名分别输出到neg.list, pos.list, 并放置在test/下; 同样地操作在train/。

   <pre class="brush: shell; gutter: true">amadeuzou@ubuntu:~/Projects/opencv/OLTbinaries/test$ ls ./neg &gt; neg.list
   amadeuzou@ubuntu:~/Projects/opencv/OLTbinaries/test$ ls ./pos &gt; pos.list</pre>

5. 到这里，样本数据便准备好了，那么，只要修改OLTbinaries/runall.sh相关参数然后这些此脚本，一小时左右的时间，便会在OLTbinaries/HOG/下产生一个model_4BiSVMLight.alt文件，你的模型数据便保存在这里面。到这里，你便成功trainning了一个model。

注意事项：

• runall.sh中第5行，按你的正负样本数目修改：

  1	HardOption= " --poscases 2416 --negcases 1218 "

• runall.sh中第21行，按你的样本文件夹所在(InDir)及输出文件所在(OutDir)修改：

  1 2 3 4

  OutDir=. /HOG InDir=./ OutFile=$OutDir /record CMDLINE=$OutDir /record

• trainning过程中会产生两个G左右的临时文件在OutDir（=./HOG）中，所以要确保硬盘空间足够，结束时删除，只留model_4BiSVMLight.alt。
• 整个trainning过程分4步，有4条屏幕输出信息，最可能出现的错误是样本文件路径不对，可在pos.list neg.list 中用图像文件的绝对路径。

  1 2 3 4

  First iteration complete Hard examples created Doing second learning Second iteration complete

• 如果你用的是自己的样本，注意修改其他参数（待究），如正样本的大小：

  1 2	WIDTH=64; export WIDTH HEIGHT=128; export HEIGHT

有了模型，怎么去做目标检测呢？你可以做以下的试验：

1. 使用bin在OLTbinaries/bin/中classify_rhog： classify_rhog [待检测图片] [目标位置数据结果保存的文本文件] [模型文件] -i [位置画在图像文件]

   amadeuzou@ubuntu:~/Projects/opencv/OLTbinaries/bin$ ./classify_rhog  person-1.jpg  result.txt model_4BiSVMLight.alt -i result.jpg

   结果：

2. 使用lib在OLTbinaries/lib/中：

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98

   /*   * =============================================   *   *       Filename:  lib-detector.cpp   *   *    Description:  Code to detect object   *   *       Compiler:  gcc   *   *         Author:  Amadeu zou   *         URL:  www.youtueye.com   *   * =============================================   */ #include <cv.h> #include <highgui.h>   #include <string> #include <iostream> #include <algorithm> #include <iterator>   #include <X11/Xlib.h> #include <Imlib2.h>   #include <lear/interface/windetect.h>// change this path as appropriate. using namespace std;   std::list<DetectedRegion> detector( char * imgf, WinDetectClassify windetect, LinearClassify* classifier) {      std::list<DetectedRegion> detections;      // read image      Imlib_Image image = imlib_load_image(imgf);      // if the load was successful      if (image)      { // set the image we loaded as the current context image to work on          imlib_context_set_image(image);      } else {          //std::cerr << "Unable to read image: " << argv[1] << std::endl;          return detections;      }      int width  = imlib_image_get_width(),          height = imlib_image_get_height();        typedef unsigned char uchar;      DATA32* data = imlib_image_get_data_for_reading_only();      uchar* imagedata = new uchar[3*width*height* sizeof (uchar)];      for ( int j= 0; j< height; ++j)      for ( int i= 0; i< width; ++i) {          uchar* pixel = imagedata+(i+j*width)*3;          int argb = data[i+j*width];          pixel[0] = static_cast <uchar>((argb & 0x00FF0000)>>16);          pixel[1] = static_cast <uchar>((argb & 0x0000FF00)>> 8);          pixel[2] = static_cast <uchar>((argb & 0x000000FF)    );      }      imlib_free_image();        // now get detections      windetect.test(*classifier, detections, imagedata, width, height);        delete [] imagedata;      return detections; }   int main( int argc, char ** argv) {        if (argc != 4) {          std::cout << "Error" << std::endl;          return 0;      }        char modelpath[256];      strcpy (modelpath,argv[2]);      string model_file(modelpath) ;      // initialize the person detector. All default parameters are set for person detector.      WinDetectClassify windetect; // use default person detector.      RHOGDenseParam desc;      LinearClassify* classifier = NULL; // initialize it to 64x128 person detector.      classifier = new LinearClassify(model_file, 0);      windetect.init(&desc); // initialize the descriptor computation        std::list<DetectedRegion> detections;      detections = detector(argv[1], windetect, classifier);        //draw region in image      IplImage* img = cvLoadImage(argv[1],1);      for (list<DetectedRegion>::iterator itor=detections.begin();itor!=detections.end();++itor)      {           cvRectangle(img,cvPoint(itor->x,itor->y),cvPoint(itor->x+itor->width,itor->y+itor->height),cvScalar(0,0,255),2);        }      cvSaveImage(argv[3],img);      cvReleaseImage(&img);      //print detections      std::copy(detections.begin(), detections.end(), std::ostream_iterator<DetectedRegion>(std::cout, "\n" ));        return 0; }

   编译：

   1	g++ `pkg-config --cflags --libs opencv` -O3 -o lib-detector lib-detector.cpp   -I.  -I /usr/include -L. -lcmdline -lcvip -lclassifier -llearutil -lboost_filesystem-gcc  -lblitz -L /usr/lib -lImlib2 -lfreetype -lz -L /usr/X11R6/lib -lX11 -lXext -ldl -lm

   结果：

   1 2 3 4

   amadeuzou@ubuntu:~ /Projects/opencv/OLTbinaries/lib $ . /lib-detector person-1.jpg   model_4BiSVMLight.alt  result.jpg     298    215    145    290    2.2674 1.10256009e-01      13      9    237    475   3.71704 1.31164089e-01     234     -7    230    460   3.59693 1.35627717e-01

3. 使用源码在learcode/app/中classify_rhog.cpp：配置及使用见learcode/README
4. 将模型数据copy到opencv的getDefaultPeopleDetector()，但要重新编译objdetect链接，没必要这么搞。从model中读取数据如下（共classifier->length() = 3780 + 1数据，1为classifier->linearbias）：

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

   int main( int argc, char ** argv) {        string model_file = "model_4BiSVMLight.alt" ;        LinearClassify* classifier = NULL;        classifier = new LinearClassify(model_file, 0);        cout<< "classifier->length() : " <<classifier->length()<<endl;      FILE * fptr = fopen ( "hog.txt" , "w" );          for ( int i = 0;i<classifier->length();i++){              fprintf (fptr, "%1.8f " ,classifier->linearwt[i]);              if (0 == (i+1) % 10) fprintf (fptr, "%s" , "\n" );          }      fprintf (fptr, "%1.8f " ,classifier->linearbias);      fclose (fptr);        return 0;   }

5. opencv直接载入model: 将peopeldetect.cpp第49行hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector())里面的vector换成你的模板数据，模板数据从model_4BiSVMLight.alt中载入，方法参照learcode/lib/windetect.cpp第1175行处LinearClassify::LinearClassify(std::string& modelfile, const int verbose)：

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226

   /*   * =====================================================================================   *   *       Filename:  model-detector.cpp   *   *    Description:  Code to detect object   *   *       Compiler:  gcc   *   *         Author:  Amadeu zou   *         URL:  www.youtueye.com   *   * =====================================================================================   */ #include <fstream> #include <iostream> #include <vector>   #include <stdio.h> #include <string.h> #include <ctype.h>   #include "opencv2/imgproc/imgproc.hpp" #include "opencv2/objdetect/objdetect.hpp" #include "opencv2/highgui/highgui.hpp"   using namespace std; using namespace cv;   vector< float > load_lear_model( const char * model_file) {       vector< float >  detector;       FILE *modelfl;       if ((modelfl = fopen (model_file, "rb" )) == NULL)       {           cout<< "Unable to open the modelfile" <<endl;           return detector;       }         char version_buffer[10];       if (! fread (&version_buffer, sizeof ( char ),10,modelfl))       {           cout<< "Unable to read version" <<endl;           return detector;       }         if ( strcmp (version_buffer, "V6.01" ))       {          cout<< "Version of model-file does not match version of svm_classify!" <<endl;          return detector;       }       // read version number       int version = 0;       if (! fread (&version, sizeof ( int ),1,modelfl))       {           cout<< "Unable to read version number" <<endl;           return detector;       }         if (version < 200)       {           cout<< "Does not support model file compiled for light version" <<endl;           return detector;        }         long kernel_type;       fread (&(kernel_type), sizeof ( long ),1,modelfl);           { // ignore these          long poly_degree;          fread (&(poly_degree), sizeof ( long ),1,modelfl);            double rbf_gamma;          fread (&(rbf_gamma), sizeof ( double ),1,modelfl);            double  coef_lin;          fread (&(coef_lin), sizeof ( double ),1,modelfl);          double coef_const;          fread (&(coef_const), sizeof ( double ),1,modelfl);            long l;          fread (&l, sizeof ( long ),1,modelfl);          char * custom = new char [l];          fread (custom, sizeof ( char ),l,modelfl);          delete [] custom;      }        long totwords;      fread (&(totwords), sizeof ( long ),1,modelfl);        { // ignore these          long totdoc;          fread (&(totdoc), sizeof ( long ),1,modelfl);            long sv_num;          fread (&(sv_num), sizeof ( long ),1,modelfl);      }        double linearbias = 0.0;      fread (&linearbias, sizeof ( double ),1,modelfl);        if (kernel_type == 0) { /* linear kernel */          /* save linear wts also */          double * linearwt = new double [totwords+1];          int length = totwords;          fread (linearwt, sizeof ( double ),totwords+1,modelfl);            for ( int i = 0;i<totwords;i++){              float term = linearwt[i];              detector.push_back(term);          }          float term = -linearbias;          detector.push_back(term);          delete [] linearwt;        } else {          cout<< "Only supports linear SVM model files" <<endl;      }        fclose (modelfl);      return detector;   }   void help() {      printf (              "\nDemonstrate the use of the HoG descriptor using\n"              "  HOGDescriptor::hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());\n"              "Usage:\n"              "./peopledetect (<image_filename> | <image_list>.txt)\n\n" ); }   int main( int argc, char ** argv) {      Mat img;      FILE * f = 0;      char _filename[1024];        if ( argc != 3 )      {          cout<< "ERROR" <<endl;          return 0;      }      img = imread(argv[1]);        if ( img.data )      {          strcpy (_filename, argv[1]);      }      else      {          f = fopen (argv[1], "rt" );          if (!f)          {              fprintf ( stderr, "ERROR: the specified file could not be loaded\n" );              return -1;          }      }        HOGDescriptor hog;      //hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());      vector< float > detector = load_lear_model(argv[2]);      hog.setSVMDetector(detector);      namedWindow( "people detector" , 1);        for (;;)      {          char * filename = _filename;          if (f)          {              if (! fgets (filename, ( int ) sizeof (_filename)-2, f))                  break ;              //while(*filename && isspace(*filename))              //  ++filename;              if (filename[0] == '#' )                  continue ;              int l = strlen (filename);              while (l > 0 && isspace (filename[l-1]))                  --l;              filename[l] = '\0' ;              img = imread(filename);          }          printf ( "%s:\n" , filename);          if (!img.data)              continue ;            fflush (stdout);          vector<Rect> found, found_filtered;          double t = ( double )getTickCount();          // run the detector with default parameters. to get a higher hit-rate          // (and more false alarms, respectively), decrease the hitThreshold and          // groupThreshold (set groupThreshold to 0 to turn off the grouping completely).          hog.detectMultiScale(img, found, 0, Size(8,8), Size(32,32), 1.05, 2);          t = ( double )getTickCount() - t;          printf ( "tdetection time = %gms\n" , t*1000./cv::getTickFrequency());          size_t i, j;          for ( i = 0; i < found.size(); i++ )          {              Rect r = found[i];              for ( j = 0; j < found.size(); j++ )                  if ( j != i && (r & found[j]) == r)                      break ;              if ( j == found.size() )                  found_filtered.push_back(r);          }          for ( i = 0; i < found_filtered.size(); i++ )          {              Rect r = found_filtered[i];              // the HOG detector returns slightly larger rectangles than the real objects.              // so we slightly shrink the rectangles to get a nicer output.              r.x += cvRound(r.width*0.1);              r.width = cvRound(r.width*0.8);              r.y += cvRound(r.height*0.07);              r.height = cvRound(r.height*0.8);              rectangle(img, r.tl(), r.br(), cv::Scalar(0,255,0), 3);          }          imshow( "people detector" , img);          int c = waitKey(0) & 255;          if ( c == 'q' || c == 'Q' || !f)              break ;      }      if (f)          fclose (f);      return 0; }

   结果：

   1	amadeuzou@ubuntu:~ /Projects/opencv/OLTbinaries/lib $ . /model-detect  person-1.jpg model_4BiSVMLight.alt

   

到这里，opencv peopledetect从模型训练到目标检测的过程便完成了，训练过程用的是算法作者的程序，不知opencv是否已将这一过程融入。