peopledetect学习，来自opencv中文论坛

OpenCV2.0提供了行人检测的例子，用的是法国人Navneet　Dalal最早在CVPR2005会议上提出的方法。
最近正在学习它，下面是自己的学习体会，希望共同探讨提高。
1、VC 2008 Express下安装OpenCV2.0--可以直接使用2.1，不用使用CMake进行编译了，避免编译出错
      这是一切工作的基础，感谢版主提供的参考：http://www.opencv.org.cn/index.php/VC_2008_Express下安装OpenCV2.0
2、体会该程序
在DOS界面，进入如下路径：　C:\OpenCV2.0\samples\c　　peopledetect.exe　filename.jpg
其中filename.jpg为待检测的文件名
3、编译程序
　创建一个控制台程序，从C:\OpenCV2.0\samples\c下将peopledetect.cpp加入到工程中；按步骤1的方法进行设置。编译成功，但是在DEBUG模式下生成的EXE文件运行出错，很奇怪 。
改成RELEASE模式后再次编译，生成的EXE文件可以运行。
4程序代码简要说明
1) getDefaultPeopleDetector() 获得3780维检测算子（105 blocks with 4 histograms each and 9 bins per histogram there are 3,780 values）--（为什么是105blocks？）
2).cv::HOGDescriptor hog;　创建类的对象　一系列变量初始化 　
winSize(64,128), blockSize(16,16), blockStride(8,8),
cellSize(8,8), nbins(9), derivAperture(1), winSigma(-1),
histogramNormType(L2Hys), L2HysThreshold(0.2), gammaCorrection(true)
3). 调用函数：detectMultiScale(img, found, 0, cv::Size(8,8), cv::Size(24,16), 1.05, 2);　
　　参数分别为待检图像、返回结果列表、门槛值hitThreshold、窗口步长winStride、图像padding margin、比例系数、门槛值groupThreshold；通过修改参数发现，就所用的某图片，参数0改为0.01就检测不到，改为0.001可以；1.05改为1.1就不行，1.06可以；2改为1可以，0.8以下不行，（24,16）改成（0,0）也可以，（32,32）也行
该函数内容如下
(1) 得到层数　levels　
某图片（530，402）为例，lg(402/128)/lg1.05=23.4　则得到层数为24
　(2)　循环levels次，每次执行内容如下
HOGThreadData& tdata = threadData[getThreadNum()];
Mat smallerImg(sz, img.type(), tdata.smallerImgBuf.data);
　　　　调用以下核心函数
detect(smallerImg, tdata.locations, hitThreshold, winStride, padding);
其参数分别为，该比例下图像、返回结果列表、门槛值、步长、margin
该函数内容如下：
(a)得到补齐图像尺寸paddedImgSize
(b)创建类的对象　HOGCache cache(this, img, padding, padding, nwindows == 0, cacheStride); 在创建过程中，首先初始化 HOGCache::init,包括：计算梯度 descriptor->computeGradient、得到块的个数105、每块参数个数36　
　　　 (c)获得窗口个数nwindows，以第一层为例，其窗口数为（530＋32*2-64）/8＋1、（402＋32*2-128）/8＋1　＝67*43＝2881，其中（32,32）为winStride参数，也可用（24,16）
(d)在每个窗口执行循环，内容如下
在105个块中执行循环，每个块内容为：通过getblock函数计算HOG特征并归一化，36个数分别与算子中对应数进行相应运算；判断105个块的总和　s >= hitThreshold 则认为检测到目标　
4）主体部分感觉就是以上这些，但很多细节还需要进一步弄清。
5、原文献写的算法流程
文献NavneetDalalThesis.pdf　78页图5.5描述了The complete object detection algorithm.
前2步为初始化，上面基本提到了。后面2步如下
For each scale Si = [Ss, SsSr, . . . , Sn]
(a) Rescale the input image using bilinear interpolation
(b) Extract features (Fig. 4.12) and densely scan the scaled image with stride Ns for　object/non-object detections
(c) Push all detections with t(wi) > c to a list
Non-maximum suppression
(a) Represent each detection in 3-D position and scale space yi
(b) Using (5.9), compute the uncertainty matrices Hi for each point
(c) Compute the mean shift vector (5.7) iteratively for each point in the list until it converges to a mode
(d) The list of all of the modes gives the final fused detections
(e) For each mode compute the bounding box from the final centre point and scale

以下内容节选自文献NavneetDalalThesis.pdf，把重要的部分挑出来了。其中保留了原文章节号，便于查找。

4. Histogram of Oriented Gradients Based Encoding of Images
Default Detector.
As a yardstick for the purpose of comparison, throughout this section we compare results to our
default detector which has the following properties: input image in RGB colour space (without
any gamma correction); image gradient computed by applying [?1, 0, 1] filter along x- and yaxis
with no smoothing; linear gradient voting into 9 orientation bins in 0_–180_; 16×16 pixel
blocks containing 2×2 cells of 8×8 pixel; Gaussian block windowing with _ = 8 pixel; L2-Hys
(Lowe-style clipped L2 norm) block normalisation; blocks spaced with a stride of 8 pixels (hence
4-fold coverage of each cell); 64×128 detection window; and linear SVM classifier. We often
quote the performance at 10?4 false positives per window (FPPW) – the maximum false positive
rate that we consider to be useful for a real detector given that 103–104 windows are tested for
each image.
4.3.2 Gradient Computation
The simple [?1, 0, 1] masks give the best performance.
4.3.3 Spatial / Orientation Binning
Each pixel contributes a weighted vote for orientation based on the orientation of the gradient element centred on it.
The votes are accumulated into orientation bins over local spatial regions that we call cells.
To reduce aliasing, votes are interpolated trilinearly between the neighbouring bin centres in both orientation and position.
Details of the trilinear interpolation voting procedure are presented in Appendix D.
The vote is a function of the gradient magnitude at the pixel, either the magnitude itself, its square, its
square root, or a clipped form of the magnitude representing soft presence/absence of an edge at the pixel. In practice, using the magnitude itself gives the best results.
4.3.4 Block Normalisation Schemes and Descriptor Overlap
good normalisation is critical and including overlap significantly improves the performance.
Figure 4.4(d) shows that L2-Hys, L2-norm and L1-sqrt all perform equally well for the person detector.
such as cars and motorbikes, L1-sqrt gives the best results.
4.3.5 Descriptor Blocks
R-HOG.
For human detection, 3×3 cell blocks of 6×6 pixel cells perform best with 10.4% miss-rate
at 10?4 FPPW. Our standard 2×2 cell blocks of 8×8 cells are a close second.
We find 2×2 and 3×3 cell blocks work best.
4.3.6 Detector Window and Context
Our 64×128 detection window includes about 16 pixels of margin around the person on all four
sides.
4.3.7 Classifier
By default we use a soft (C=0.01) linear SVM trained with SVMLight [Joachims 1999].We modified
SVMLight to reduce memory usage for problems with large dense descriptor vectors.
---------------------------------
5. Multi-Scale Object Localisation
the detector scans the image with a detection window at all positions and scales, running the classifier in each window and fusing multiple overlapping detections to yield the final object detections.
We represent detections using kernel density estimation (KDE) in 3-D position and scale space. KDE is a data-driven process where continuous densities are evaluated by applying a smoothing kernel to observed data points. The bandwidth of the smoothing kernel defines the local neighbourhood. The detection scores are incorporated by weighting the observed detection points by their score values while computing the density estimate. Thus KDE naturally incorporates the first two criteria. The overlap criterion follows from the fact that detections at very different scales or positions are far off in 3-D position and scale space, and are thus not smoothed together. The modes (maxima) of the density estimate correspond to the positions and scales of final detections.
Let xi = [xi, yi] and s0i denote the detection position and scale, respectively, for the i-th detection.
the detections are represented in 3-D space as y = [x, y, s], where s = log(s’).
the variable bandwidth mean shift vector is defined as (5.7)

For each of the n point the mean shift based iterative procedure is guaranteed to converge to a mode2.
Detection Uncertainty Matrix Hi.
One key input to the above mode detection algorithm is the amount of uncertainty Hi to be associated with each point. We assume isosymmetric covariances, i.e. the Hi’s are diagonal matrices.
Let diag [H] represent the 3 diagonal elements of H. We use scale dependent covariance
matrices such that diag
[Hi] = [(exp(si)_x)2, (exp(si)_y)2, (_s)2] (5.9)
where _x, _y and _s are user supplied smoothing values.

The term t(wi) provides the weight for each detection. For linear SVMs we usually use threshold = 0.
the smoothing parameters _x, _y,and _s used in the non-maximum suppression stage. These parameters can have a significant impact on performance so proper evaluation is necessary. For all of the results here, unless otherwise noted, a scale ratio of 1.05, a stride of 8 pixels, and _x = 8, _y = 16, _s = log(1.3) are used as default values.
A scale ratio of 1.01 gives the best performance, but significantly slows the overall process.
Scale smoothing of log(1.3)–log(1.6) gives good performance for most object classes.
We group these mode candidates using a proximity measure. The final location is the ode corresponding to the highest density.
----------------------------------------------------
附录　A. INRIA Static Person Data Set
The (centred and normalised) positive windows are supplied by the user, and the initial set of negatives is created once and for all by randomly sampling negative images.A preliminary classifier is thus trained using these. Second, the preliminary detector is used to exhaustively scan the negative training images for hard examples (false positives). The classifier is then re-trained using this augmented training set (user supplied positives, initial negatives and hard examples) to produce the final detector.
INRIA Static Person Data Set
As images of people are highly variable, to learn an effective classifier, the positive training examples need to be properly normalized and centered to minimize the variance among them. For this we manually annotated all upright people in the original images.
The image regions belonging to the annotations were cropped and rescaled to 64×128 pixel image windows. On average the subjects height is 96 pixels in these normalised windows to allow for an approximately16 pixel margin on each side. In practise we leave a further 16 pixel margin around each side of the image window to ensure that flow and gradients can be computed without boundary effects. The margins were added by appropriately expanding the annotations on each side before cropping the image regions.

//<------------------------以上摘自datal的博士毕业论文

关于INRIA Person Dataset的更多介绍，见以下链接
http://pascal.inrialpes.fr/data/human/
Original Images
            Folders 'Train' and 'Test' correspond, respectively, to original training and test images. Both folders have three sub folders: (a) 'pos' (positive training or test images), (b) 'neg' (negative training or test images), and (c) 'annotations' (annotation files for positive images in Pascal Challenge format).
Normalized Images
        Folders 'train_64x128_H96' and 'test_64x128_H96' correspond to normalized dataset as used in above referenced paper. Both folders have two sub folders: (a) 'pos' (normalized positive training or test images centered on the person with their left-right reflections), (b) 'neg' (containing original negative training or test images). Note images in folder 'train/pos' are of 96x160 pixels (a margin of 16 pixels around each side), and images in folder 'test/pos' are of 70x134 pixels (a margin of 3 pixels around each side). This has been done to avoid boundary conditions (thus to avoid any particular bias in the classifier). In both folders, use the centered 64x128 pixels window for original detection task.
Negative windows
        To generate negative training windows from normalized images, a fixed set of 12180 windows (10 windows per negative image) are sampled randomly from 1218 negative training photos providing the initial negative training set. For each detector and parameter combination, a preliminary detector is trained and all negative training images are searched exhaustively (over a scale-space pyramid) for false positives (`hard examples'). All examples with score greater than zero are considered hard examples. The method is then re-trained using this augmented set (initial 12180 + hard examples) to produce the final detector. The set of hard examples is subsampled if necessary, so that the descriptors of the final training set fit into 1.7 GB of RAM for SVM training.

//------------------------------------------------------______________>

原作者对 OpenCV2.0 peopledetect 进行了2次更新
https://code.ros.org/trac/opencv/changeset/2314/trunk
最近一次改为如下：
---------------------
#include "cvaux.h"
#include "highgui.h"
#include <stdio.h>
#include <string.h>
#include <ctype.h>
using namespace cv;
using namespace std;
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());
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).
int can = img.channels();
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*0.1）;
r.y += cvRound(r.height*0.07);
r.height = cvRound(r.height*0.1）;
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;
}

更新后可以批量检测图片！


将需要批量检测的图片，构造一个TXT文本，文件名为filename.txt, 其内容如下
1.jpg
2.jpg
......

然后在DOS界面输入 peopledetect filename.txt ，　即可自动检测每个图片。

//////////////////////////////////////////////////////////////////------------------------------Navneet Dalal的OLT工作流程描述

Navneet Dalal在以下网站提供了INRIA Object Detection and Localization Toolkit
http://pascal.inrialpes.fr/soft/olt/
Wilson Suryajaya Leoputra提供了它的windows版本
http://www.computing.edu.au/~12482661/hog.html
需要　Copy all the dll's (boost_1.34.1*.dll, blitz_0.9.dll, opencv*.dll) into "<ROOT_PROJECT_DIR>/debug/"
Navneet Dalal提供了linux下的可执行程序，借别人的linux系统，运行一下，先把总体流程了解了。
下面结合OLTbinaries\readme和OLTbinaries\HOG\record两个文件把其流程描述一下。
1.下载 INRIA person detection database 解压到OLTbinaries\；把其中的'train_64x128_H96' 重命名为 'train' ； 'test_64x128_H96' 重命名为 'test'.
2．在linux下运行 'runall.sh' script.
等待结果出来后，打开matlab 运行 plotdet.m 可绘制 DET曲线；
------这是一步到位法--------------------------------------------------
-------此外，它还提供了分步执行法-------------------------------------
1、由pos.lst列表提供的图片，计算正样本R－HOG特征，pos.lst列表格式如下
train/pos/crop_000010a.png
train/pos/crop_000010b.png
train/pos/crop_000011a.png
------以下表示-linux下执行语句（下同）------
./bin//dump_rhog -W 64,128 -C 8,8 -N 2,2 -B 9 -G 8,8 -S 0 --wtscale 2 --maxvalue 0.2 -- epsilon 1 --fullcirc 0 -v 3 --proc rgb_sqrt --norm l2hys -s 1 train/pos.lst 　HOG/train_pos.RHOG
2.计算负样本R－HOG特征
./bin//dump_rhog -W 64,128 -C 8,8 -N 2,2 -B 9 -G 8,8 -S 0 --wtscale 2 --maxvalue 0.2 -- epsilon 1 --fullcirc 0 -v 3 --proc rgb_sqrt --norm l2hys -s 10 train/neg.lst HOG/train_neg.RHOG
3.训练
./bin//dump4svmlearn -p HOG/train_pos.RHOG -n HOG/train_neg.RHOG HOG/train_BiSVMLight.blt -v
4.创建　model file: HOG/model_4BiSVMLight.alt
./bin//svm_learn -j 3 -B 1 -z c -v 1 -t 0 HOG/train_BiSVMLight.blt HOG/model_4BiSVMLight.alt
5.创建文件夹
mkdir -p HOG/hard
6.分类
./bin//classify_rhog train/neg.lst HOG/hard/list.txt HOG/model_4BiSVMLight.alt -d HOG/hard/hard_neg.txt -c HOG/hard/hist.txt -m 0 -t 0 --no_nonmax 1 --avsize 0 --margin 0 --scaleratio 1.2 -l N -W 64,128 -C 8,8 -N 2,2 -B 9 -G 8,8 -S 0 --wtscale 2 --maxvalue 0.2 --
epsilon 1 --fullcirc 0 -v 3 --proc rgb_sqrt --norm l2hys
--------
false +/- 分类结果会写入 HOG/hard/hard_neg.txt
7.　将hard加入到neg,再次计算RHOG特征
./bin//dump_rhog -W 64,128 -C 8,8 -N 2,2 -B 9 -G 8,8 -S 0 --wtscale 2 --maxvalue 0.2 -- epsilon 1 --fullcirc 0 -v 3 --proc rgb_sqrt --norm l2hys -s 0 HOG/hard/hard_neg.txt OG/train_hard_neg.RHOG --poscases 2416 --negcases 12180 --dumphard 1 --hardscore 0 -- memorylimit 1700
8.再次训练
./bin//dump4svmlearn -p HOG/train_pos.RHOG -n HOG/train_neg.RHOG -n HOG/train_hard_neg.RHOG HOG/train_BiSVMLight.blt -v 4
9.得到最终的模型
./bin//svm_learn -j 3 -B 1 -z c -v 1 -t 0 HOG/train_BiSVMLight.blt HOG/model_4BiSVMLight.alt
Opencv中用到的3780 个值，应该就在这个模型里面model_4BiSVMLight.alt，不过它的格式未知，无法直接读取，但是可以研究svm_learn程序是如何生成它的;此外，该模型由程序classify_rhog调用，研究它如何调用，估计是一个解析此格式的思路
10.创建文件夹
mkdir -p HOG/WindowTest_Negative
11.负样本检测结果
./bin//classify_rhog -W 64,128 -C 8,8 -N 2,2 -B 9 -G 8,8 -S 0 --wtscale 2 --maxvalue 0.2 --epsilon 1 --fullcirc 0 -v 3 --proc rgb_sqrt --norm l2hys -p 1 --no_nonmax 1 --nopyramid 0 - -scaleratio 1.2 -t 0 -m 0 --avsize 0 --margin 0 test/neg.lst HOG/WindowTest_Negative/list.txt HOG/model_4BiSVMLight.alt -c HOG/WindowTest_Negative/histogram.txt
12.创建文件夹
mkdir -p HOG/WindowTest_Positive
13.正样本检测结果
./bin//classify_rhog -W 64,128 -C 8,8 -N 2,2 -B 9 -G 8,8 -S 0 --wtscale 2 --maxvalue 0.2 --　epsilon 1 --fullcirc 0 -v 3 --proc rgb_sqrt --norm l2hys -p 1 --no_nonmax 1 --nopyramid 1 -t 0 -m 0 --avsize 0 --margin 0 test/pos.lst HOG/WindowTest_Positive/list.txt 　HOG/model_4BiSVMLight.alt -c HOG/WindowTest_Positive/histogram.txt

////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

如何制作训练样本

分析了原作者的数据集，结合网上一些资料，下面描述如何制作训练样本
1、如何从原始图片生成样本
对比INRIAPerson\INRIAPerson\Train\pos（原始图片），INRIAPerson\train_64x128_H96\pos（生成样本）可以发现，作者从原始图片裁剪出一些站立的人，要求该人不被遮挡，然后对剪裁的图片left-right reflect。以第一张图片为例crop001001，它剪裁了2个不被遮挡的人，再加上原照片，共3张，再加左右镜像，总共6张。
2、裁剪
　可利用基于opencv1.0的程序imageclipper，进行裁剪并保存,它会自动生成文件名并保存在同一路径下新生成的imageclipper文件夹下。
3.改变图片大小
　可以利用Acdsee软件，Tools/open in editor,进去后到Resize选项;　tools/rotate还可实现left-right reflect

自己编了一个程序，批量改变图片大小，代码见下一楼

4. 制作pos.lst列表
　 进入dos界面，定位到需要制作列表的图片文件夹下，输入 dir /b> pos.lst，即可生成文件列表；

/////////////////////////

#include "cv.h"
#include "highgui.h"
#include "cvaux.h"


int main(int argc,char * argv[])
{
IplImage* src ;
IplImage* dst = 0;

CvSize dst_size;

FILE* f = 0;
char _filename[1024];
int l;

f = fopen(argv[1], "rt");
if(!f)
{
fprintf( stderr, "ERROR: the specified file could not be loaded\n");
return -1;
}

for(;;)
{
char* filename = _filename;
if(f)
{
if(!fgets(filename, (int)sizeof(_filename)-2, f))
break;
if(filename[0] == '#')
continue;
l = strlen(filename);
while(l > 0 && isspace(filename[l-1]))
--l;
filename[l] = '\0';
src=cvLoadImage(filename,1);
}

dst_size.width = 96;
dst_size.height = 160;
dst=cvCreateImage(dst_size,src->depth,src->nChannels);
cvResize(src,dst,CV_INTER_LINEAR);//////////////////
char* filename2 = _filename;char* filename3 = _filename; filename3="_96x160.jpg";
strncat(filename2, filename,l-4);
strcat(filename2, filename3);

cvSaveImage(filename2, dst);

}
if(f)
fclose(f);

cvWaitKey(-1);
cvReleaseImage( &src );
cvReleaseImage( &dst );

return 0;
}