如需转载请注明本博网址：http://blog.csdn.net/ding977921830/article/details/47733363。
一 训练框架
训练人脸检測分类器须要三个步骤：
(1) 准备正负样本集，分别放到两个目录里。
我使用的是麻省理工的那个人脸库。大家能够网上搜一下。
（2）把正样本集生成正样本描写叙述文件（*.vec），把负样本集生成负样本集合文件。详细怎么操作请參考我博客中的另外两篇文章，各自是http://blog.csdn.net/ding977921830/article/details/45913789和http://blog.csdn.net/ding977921830/article/details/45914137。
（3）利用........opencvsourcesappshaartraininghaartraining.cpp训练分类器。
二 建立project
我使用的是vs2012和opencv2.4.9，事实上，使用其它的版本号也区别不多大。
1 配置opencv2.4.9和vs2012，这个网上有非常多资料，我就不啰嗦了哈。
2 在vs中新建project，把opencv库中的以下文件........opencvsourcesappshaartraining加入到project中，在解决方式资源管理器中，分别加入头文件和源文件，加入好后，内容例如以下：
三 程序
上面main.cpp的内容也就是haartraining.cpp中的程序，详细内容例如以下：
//M*/
/*
* haartraining.cpp
*里面有部分參数我是稍作改动
*<a target=_blank href="http://blog.csdn.net/ding977921830/article/details/47733363">http://blog.csdn.net/ding977921830/article/details/47733363</a>
* Train cascade classifier
*/
#include <cstdio>
#include <cstring>
#include <cstdlib>
using namespace std;
#include "cvhaartraining.h"
int main( int argc, char* argv[] )
{
int i = 0;
char* nullname = (char*)"(NULL)";
char* vecname = NULL;
char* dirname = NULL;
char* bgname = NULL;
bool bg_vecfile = false;
int npos = 2000; //保证npos与nneg的比例为1:2至1：：3之间比較好
int nneg = 4000;
int nstages = 3; //为了节约时间能够把把设置为1，或2或3。当然也能够设置十几或二十几。只是，我没有耐心实验
int mem = 200;
int nsplits = 1;
float minhitrate = 0.995F;
float maxfalsealarm = 0.5F;
float weightfraction = 0.95F;
int mode = 0;
int symmetric = 1;
int equalweights = 0;
int width = 20;
int height = 20;
const char* boosttypes[] = { "DAB", "RAB", "LB", "GAB" };
int boosttype = 0; //选用DAB
const char* stumperrors[] = { "misclass", "gini", "entropy" };
int stumperror = 0; //选用misclass
int maxtreesplits = 0;
int minpos = 500;
if( argc == 1 )
{
printf( "Usage: %s
-data <dir_name>
"
" -vec <vec_file_name>
"
" -bg <background_file_name>
"
" [-bg-vecfile]
"
" [-npos <number_of_positive_samples = %d>]
"
" [-nneg <number_of_negative_samples = %d>]
"
" [-nstages <number_of_stages = %d>]
"
" [-nsplits <number_of_splits = %d>]
"
" [-mem <memory_in_MB = %d>]
"
" [-sym (default)] [-nonsym]
"
" [-minhitrate <min_hit_rate = %f>]
"
" [-maxfalsealarm <max_false_alarm_rate = %f>]
"
" [-weighttrimming <weight_trimming = %f>]
"
" [-eqw]
"
" [-mode <BASIC (default) | CORE | ALL>]
"
" [-w <sample_width = %d>]
"
" [-h <sample_height = %d>]
"
" [-bt <DAB | RAB | LB | GAB (default)>]
"
" [-err <misclass (default) | gini | entropy>]
"
" [-maxtreesplits <max_number_of_splits_in_tree_cascade = %d>]
"
" [-minpos <min_number_of_positive_samples_per_cluster = %d>]
",
argv[0], npos, nneg, nstages, nsplits, mem,
minhitrate, maxfalsealarm, weightfraction, width, height,
maxtreesplits, minpos );
return 0;
}
for( i = 1; i < argc; i++ )
{
/*if( !strcmp( argv[i], "-data" ) )
{
dirname = argv[++i];
}
else if( !strcmp( argv[i], "-vec" ) )
{
vecname = argv[++i];
}
else if( !strcmp( argv[i], "-bg" ) )
{
bgname = argv[++i];
}*/
if( !strcmp( argv[i], "-data" ) ) //前面这三个条件里面的内容我稍作改动
{
dirname = argv[i];
}
else if( !strcmp( argv[i], "-vec.vec" ) )
{
vecname = argv[i];
}
else if( !strcmp( argv[i], "-bg.txt" ) )
{
bgname = argv[i];
}
else if( !strcmp( argv[i], "-bg-vecfile" ) )
{
bg_vecfile = true;
}
else if( !strcmp( argv[i], "-npos" ) )
{
npos = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-nneg" ) )
{
nneg = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-nstages" ) )
{
nstages = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-nsplits" ) )
{
nsplits = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-mem" ) )
{
mem = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-sym" ) )
{
symmetric = 1;
}
else if( !strcmp( argv[i], "-nonsym" ) )
{
symmetric = 0;
}
else if( !strcmp( argv[i], "-minhitrate" ) )
{
minhitrate = (float) atof( argv[++i] );
}
else if( !strcmp( argv[i], "-maxfalsealarm" ) )
{
maxfalsealarm = (float) atof( argv[++i] );
}
else if( !strcmp( argv[i], "-weighttrimming" ) )
{
weightfraction = (float) atof( argv[++i] );
}
else if( !strcmp( argv[i], "-eqw" ) )
{
equalweights = 1;
}
else if( !strcmp( argv[i], "-mode" ) )
{
char* tmp = argv[++i];
if( !strcmp( tmp, "CORE" ) )
{
mode = 1;
}
else if( !strcmp( tmp, "ALL" ) )
{
mode = 2;
}
else
{
mode = 0;
}
}
else if( !strcmp( argv[i], "-w" ) )
{
width = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-h" ) )
{
height = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-bt" ) )
{
i++;
if( !strcmp( argv[i], boosttypes[0] ) )
{
boosttype = 0;
}
else if( !strcmp( argv[i], boosttypes[1] ) )
{
boosttype = 1;
}
else if( !strcmp( argv[i], boosttypes[2] ) )
{
boosttype = 2;
}
else
{
boosttype = 3;
}
}
else if( !strcmp( argv[i], "-err" ) )
{
i++;
if( !strcmp( argv[i], stumperrors[0] ) )
{
stumperror = 0;
}
else if( !strcmp( argv[i], stumperrors[1] ) )
{
stumperror = 1;
}
else
{
stumperror = 2;
}
}
else if( !strcmp( argv[i], "-maxtreesplits" ) )
{
maxtreesplits = atoi( argv[++i] );
}
else if( !strcmp( argv[i], "-minpos" ) )
{
minpos = atoi( argv[++i] );
}
}
printf( "Data dir name: %s
", ((dirname == NULL) ? nullname : dirname ) );
printf( "Vec file name: %s
", ((vecname == NULL) ? nullname : vecname ) );
printf( "BG file name: %s, is a vecfile: %s
", ((bgname == NULL) ? nullname : bgname ), bg_vecfile ? "yes" : "no" );
printf( "Num pos: %d
", npos );
printf( "Num neg: %d
", nneg );
printf( "Num stages: %d
", nstages );
printf( "Num splits: %d (%s as weak classifier)
", nsplits,
(nsplits == 1) ? "stump" : "tree" );
printf( "Mem: %d MB
", mem );
printf( "Symmetric: %s
", (symmetric) ? "TRUE" : "FALSE" );
printf( "Min hit rate: %f
", minhitrate );
printf( "Max false alarm rate: %f
", maxfalsealarm );
printf( "Weight trimming: %f
", weightfraction );
printf( "Equal weights: %s
", (equalweights) ? "TRUE" : "FALSE" );
printf( "Mode: %s
", ( (mode == 0) ? "BASIC" : ( (mode == 1) ? "CORE" : "ALL") ) );
printf( "Width: %d
", width );
printf( "Height: %d
", height );
//printf( "Max num of precalculated features: %d
", numprecalculated );
printf( "Applied boosting algorithm: %s
", boosttypes[boosttype] );
printf( "Error (valid only for Discrete and Real AdaBoost): %s
",
stumperrors[stumperror] );
printf( "Max number of splits in tree cascade: %d
", maxtreesplits );
printf( "Min number of positive samples per cluster: %d
", minpos );
cvCreateTreeCascadeClassifier( dirname, vecname, bgname,
npos, nneg, nstages, mem,
nsplits,
minhitrate, maxfalsealarm, weightfraction,
mode, symmetric,
equalweights, width, height,
boosttype, stumperror,
maxtreesplits, minpos, bg_vecfile );
return 0;
}
我的命令行參数为："D:vs2012projects rain_opencv_main rain_cascadeDebug est.exe" "-data" "-vec.vec" "-bg.txt"
。详细设置方法是 调试----属性----配置属性----调试---命令參数
1 注意命令行參数中间要有空格的。
2 当中第一个你要改动为你自己电脑上project的绝对路径；
3 "-data" 是存放训练好的分类器，须要预先建立好一个的空目录。
4 "-vec.vec" 是我的正样本描写叙述文件。
5 "-bg.txt"是我的负样本集合文件。
四 训练结果
1 dos操作窗体
2 data目录的内容为：
我的0文件里训练了6个弱文类器。1文件里含有9个弱分类器。2目录下有17个弱分类器，每个目录就是一个级联stage。显然是越来越复杂的哈。
3 以文件0为例，里面的内容为：
6
1
2
7 1 6 10 0 -1
9 1 2 10 0 3
haar_x3
4.792333e-002 0 -1
-1.845703e+000 1.845703e+000
1
2
1 3 18 12 0 -1
1 7 18 4 0 3
haar_y3
2.389797e-001 0 -1
-1.396623e+000 1.396623e+000
1
3
2 16 6 4 0 -1
2 16 3 2 0 2
5 18 3 2 0 2
haar_x2_y2
6.900427e-003 0 -1
-9.798445e-001 9.798445e-001
1
2
10 0 10 1 0 -1
10 0 5 1 0 2
haar_x2
1.219139e-002 0 -1
-5.156118e-001 5.156118e-001
1
2
0 0 10 1 0 -1
5 0 5 1 0 2
haar_x2
1.014664e-002 0 -1
-7.365732e-001 7.365732e-001
1
2
9 14 5 3 0 -1
9 15 5 1 0 3
haar_y3
-6.578934e-003 0 -1
7.885281e-001 -7.885281e-001
-3.758514e+000
-1
-1
4 xml文件
到这里我们的训练分类器最终出来的，XML文件能够在在vs中直接调用了。xml文件的内容你看是跟上面data文件里的内容是严格一一相应的，我摘录当中部分内容（也就是0目录部分）例如以下：
<?xml version="1.0"?>
<opencv_storage>
<_-data type_id="opencv-haar-classifier">
<size>
20 20</size>
<stages>
<_>
<!-- stage 0 -->
<trees>
<_>
<!-- tree 0 -->
<_>
<!-- root node -->
<feature>
<rects>
<_>
7 1 6 10 -1.</_>
<_>
9 1 2 10 3.</_></rects>
<tilted>0</tilted></feature>
<threshold>4.7923330217599869e-002</threshold>
<left_val>-1.8457030057907104e+000</left_val>
<right_val>1.8457030057907104e+000</right_val></_></_>
<_>
<!-- tree 1 -->
<_>
<!-- root node -->
<feature>
<rects>
<_>
1 3 18 12 -1.</_>
<_>
1 7 18 4 3.</_></rects>
<tilted>0</tilted></feature>
<threshold>2.3897969722747803e-001</threshold>
<left_val>-1.3966230154037476e+000</left_val>
<right_val>1.3966230154037476e+000</right_val></_></_>
<_>
<!-- tree 2 -->
<_>
<!-- root node -->
<feature>
<rects>
<_>
2 16 6 4 -1.</_>
<_>
2 16 3 2 2.</_>
<_>
5 18 3 2 2.</_></rects>
<tilted>0</tilted></feature>
<threshold>6.9004269316792488e-003</threshold>
<left_val>-9.7984451055526733e-001</left_val>
<right_val>9.7984451055526733e-001</right_val></_></_>
<_>
<!-- tree 3 -->
<_>
<!-- root node -->
<feature>
<rects>
<_>
10 0 10 1 -1.</_>
<_>
10 0 5 1 2.</_></rects>
<tilted>0</tilted></feature>
<threshold>1.2191389687359333e-002</threshold>
<left_val>-5.1561182737350464e-001</left_val>
<right_val>5.1561182737350464e-001</right_val></_></_>
<_>
<!-- tree 4 -->
<_>
<!-- root node -->
<feature>
<rects>
<_>
0 0 10 1 -1.</_>
<_>
5 0 5 1 2.</_></rects>
<tilted>0</tilted></feature>
<threshold>1.0146640241146088e-002</threshold>
<left_val>-7.3657321929931641e-001</left_val>
<right_val>7.3657321929931641e-001</right_val></_></_>
<_>
<!-- tree 5 -->
<_>
<!-- root node -->
<feature>
<rects>
<_>
9 14 5 3 -1.</_>
<_>
9 15 5 1 3.</_></rects>
<tilted>0</tilted></feature>
<threshold>-6.5789339132606983e-003</threshold>
<left_val>7.8852808475494385e-001</left_val>
<right_val>-7.8852808475494385e-001</right_val></_></_></trees>
<stage_threshold>-3.7585139274597168e+000</stage_threshold>
<parent>-1</parent>
<next>-1</next></_>
<_>