看了一天的论文，总结出手势识别的三种方法：
1、通过肤色识别
2、通过haar-like基
3、通过HandUv
说下各个优缺点
1、通过肤色识别
设定人的肤色的范围，监测在这个范围内的 肤色InRange，然后画出给定的区域。
例子：
#include "cv.h"
#include "cxcore.h"
#include "highgui.h"
#include "math.h"
#include <iostream>
#include <stdio.h>
#include <string.h>
#include <conio.h>
#include <sstream>
#include <time.h>
using namespace std;
int main()
{
int c = 0;
CvSeq* a = 0;
CvCapture* capture = cvCaptureFromCAM(0);//从对摄像头的初始化捕获
if(!cvQueryFrame(capture)) cout<<"Video capture failed, please check the camera."<<endl;
else cout<<"Video camera capture status: OK"<<endl;
CvSize sz = cvGetSize(cvQueryFrame( capture));//得到摄像头图像大小
IplImage* src = cvCreateImage( sz, 8, 3 );//3通道，每个通道8位
IplImage* hsv_image = cvCreateImage( sz, 8, 3);//
IplImage* hsv_mask = cvCreateImage( sz, 8, 1);
IplImage* hsv_mask2 = cvCreateImage( sz, 8, 1);
IplImage* hsv_edge = cvCreateImage( sz, 8, 1);
{
int c = 0;
CvSeq* a = 0;
CvCapture* capture = cvCaptureFromCAM(0);//从对摄像头的初始化捕获
if(!cvQueryFrame(capture)) cout<<"Video capture failed, please check the camera."<<endl;
else cout<<"Video camera capture status: OK"<<endl;
CvSize sz = cvGetSize(cvQueryFrame( capture));//得到摄像头图像大小
IplImage* src = cvCreateImage( sz, 8, 3 );//3通道，每个通道8位
IplImage* hsv_image = cvCreateImage( sz, 8, 3);//
IplImage* hsv_mask = cvCreateImage( sz, 8, 1);
IplImage* hsv_mask2 = cvCreateImage( sz, 8, 1);
IplImage* hsv_edge = cvCreateImage( sz, 8, 1);
CvScalar hsv_min = cvScalar(0, 30, 80, 0);//最小像素的RGB值
CvScalar hsv_max = cvScalar(20, 150, 255, 0);//最大像素的RGB值
//CvScalar hsv_min = cvScalar(30, 30, 30, 0);//最小像素的RGB值
//CvScalar hsv_max = cvScalar(200, 200, 200, 0);//最大像素的RGB值
//
CvMemStorage* storage = cvCreateMemStorage(0);//分配大小为0的内存空间
CvMemStorage* areastorage = cvCreateMemStorage(0);
CvMemStorage* minStorage = cvCreateMemStorage(0);
CvMemStorage* dftStorage = cvCreateMemStorage(0);
CvSeq* contours = NULL;
//
cvNamedWindow( "src",1);
//在屏幕上创建一个窗口，第一个参数为窗口标题，第二个参数为窗口属性，
//设置为0（默认值），或者CV_WINDOW_AUTOSIZE,设置为0，则窗口不会因图像的大小而改变
//图像只能在窗口中根据窗口的大小进行拉伸或缩放；设置为CV_WINDOW_AUTOSIZE时，窗口会根据图像
//的实际大小进行自动拉伸或缩放。
//cvNamedWindow( "hsv-msk",1);
//cvNamedWindow( "contour",1);
//////
CvScalar hsv_max = cvScalar(20, 150, 255, 0);//最大像素的RGB值
//CvScalar hsv_min = cvScalar(30, 30, 30, 0);//最小像素的RGB值
//CvScalar hsv_max = cvScalar(200, 200, 200, 0);//最大像素的RGB值
//
CvMemStorage* storage = cvCreateMemStorage(0);//分配大小为0的内存空间
CvMemStorage* areastorage = cvCreateMemStorage(0);
CvMemStorage* minStorage = cvCreateMemStorage(0);
CvMemStorage* dftStorage = cvCreateMemStorage(0);
CvSeq* contours = NULL;
//
cvNamedWindow( "src",1);
//在屏幕上创建一个窗口，第一个参数为窗口标题，第二个参数为窗口属性，
//设置为0（默认值），或者CV_WINDOW_AUTOSIZE,设置为0，则窗口不会因图像的大小而改变
//图像只能在窗口中根据窗口的大小进行拉伸或缩放；设置为CV_WINDOW_AUTOSIZE时，窗口会根据图像
//的实际大小进行自动拉伸或缩放。
//cvNamedWindow( "hsv-msk",1);
//cvNamedWindow( "contour",1);
//////
IplImage * background=cvLoadImage("c:\\002.jpg");
if (background==0)
{
cout<<"open background error"<<endl;
}
else
cout<<"open background ok!"<<endl;
CvRect rect=cvRect(0,0,background->width,background->height);
cvSetImageROI(background,rect);
IplImage * src3=cvCreateImage(cvGetSize(background),background->depth,background->nChannels);
if (background==0)
{
cout<<"open background error"<<endl;
}
else
cout<<"open background ok!"<<endl;
CvRect rect=cvRect(0,0,background->width,background->height);
cvSetImageROI(background,rect);
IplImage * src3=cvCreateImage(cvGetSize(background),background->depth,background->nChannels);
while( c != 27)//27为ASCII键值（ESC），
{
IplImage* bg = cvCreateImage( sz, 8, 3);//
cvRectangle( bg, cvPoint(0,0), cvPoint(bg->width,bg->height), CV_RGB( 255, 255, 255), -1, 8, 0 );//画矩形，参数：Image，两个顶点坐标，线的颜色，线的厚度
bg->origin = 1;
for(int b = 0; b< int(bg->width/10); b++)//画网格
{
cvLine( bg, cvPoint(b*20, 0), cvPoint(b*20, bg->height), CV_RGB( 200, 200, 200), 1, 8, 0 );//画竖线
cvLine( bg, cvPoint(0, b*20), cvPoint(bg->width, b*20), CV_RGB( 200, 200, 200), 1, 8, 0 );//画横线
}
{
IplImage* bg = cvCreateImage( sz, 8, 3);//
cvRectangle( bg, cvPoint(0,0), cvPoint(bg->width,bg->height), CV_RGB( 255, 255, 255), -1, 8, 0 );//画矩形，参数：Image，两个顶点坐标，线的颜色，线的厚度
bg->origin = 1;
for(int b = 0; b< int(bg->width/10); b++)//画网格
{
cvLine( bg, cvPoint(b*20, 0), cvPoint(b*20, bg->height), CV_RGB( 200, 200, 200), 1, 8, 0 );//画竖线
cvLine( bg, cvPoint(0, b*20), cvPoint(bg->width, b*20), CV_RGB( 200, 200, 200), 1, 8, 0 );//画横线
}
src = cvQueryFrame( capture);//得到一帧图像
cvSaveImage("c:\\src.bmp",src);
cvCvtColor(src, hsv_image, CV_BGR2HSV);//色彩空间转换，HSV
cvCvtColor(src, hsv_image, CV_BGR2HSV);//色彩空间转换，HSV
cvSaveImage("c:\\hsv.bmp",hsv_image);
cvInRangeS (hsv_image, hsv_min, hsv_max, hsv_mask);//检查数组元素是否在两个数量之间，输出新图像hsv_mask，从3通道到1通道
//hsv_mask的数据要么为0，要么为1，在min和max范围内为1，得到ROI区域，找到在像素RGB范围内的数据
//cvSmooth( hsv_mask, hsv_mask, CV_MEDIAN, 27, 0, 0, 0 );
cvSaveImage("c:\\mask.bmp",hsv_mask);
cvInRangeS (hsv_image, hsv_min, hsv_max, hsv_mask);//检查数组元素是否在两个数量之间，输出新图像hsv_mask，从3通道到1通道
//hsv_mask的数据要么为0，要么为1，在min和max范围内为1，得到ROI区域，找到在像素RGB范围内的数据
//cvSmooth( hsv_mask, hsv_mask, CV_MEDIAN, 27, 0, 0, 0 );
cvSaveImage("c:\\mask.bmp",hsv_mask);
//cvSmooth( hsv_mask, hsv_mask2, CV_MEDIAN, 27, 0, 0);//图像平滑，
cvSmooth( hsv_mask, hsv_mask2, CV_MEDIAN, 3, 0, 0, 0 );
cvSmooth( hsv_mask, hsv_mask2, CV_MEDIAN, 3, 0, 0, 0 );
//CV_MEDIAN (median blur) - 对图像进行核大小为param1×param1 的中值滤波 (i.e. 邻域是方的).
cvCanny(hsv_mask2, hsv_edge, 1, 3, 5);
//cvCanny(hsv_mask, hsv_edge, 1, 3, 5);//采用 Canny 算法做边缘检测
cvCanny(hsv_mask2, hsv_edge, 1, 3, 5);
//cvCanny(hsv_mask, hsv_edge, 1, 3, 5);//采用 Canny 算法做边缘检测
cvFindContours( hsv_mask, storage, &contours, sizeof(CvContour), CV_RETR_LIST, CV_CHAIN_APPROX_SIMPLE, cvPoint(0,0) );
//对单通道图像检索轮廓，返回第一个轮廓的指针
CvSeq* contours2 = NULL;
double result = 0, result2 = 0;
while(contours)//得到最大的两个感兴趣区
{
result = fabs( cvContourArea( contours, CV_WHOLE_SEQ ) );//计算感兴趣区域的像素点数
if ( result > result2) {result2 = result; contours2 = contours;};
contours = contours->h_next;
}
if ( contours2 )//最大的感兴趣区，ROI
{
//cout << "contours2: " << contours2->total << endl;
CvRect rect = cvBoundingRect( contours2, 0 );//返回一个2d矩形的点集合
cvRectangle( bg, cvPoint(rect.x, rect.y + rect.height), cvPoint(rect.x + rect.width, rect.y), CV_RGB(200, 0, 200), 1, 8, 0 );
//在bg的矩形区域中画rect的图形
//cout << "Ratio: " << rect.width << ", " << rect.height << ", " << (float)rect.width / rect.height << endl;
int checkcxt = cvCheckContourConvexity( contours2 );//检测输入的轮廓是否是凸的
//cout << checkcxt <<endl;
CvSeq* hull = cvConvexHull2( contours2, 0, CV_CLOCKWISE, 0 );//二维凸包
CvSeq* defect = cvConvexityDefects( contours2, hull, dftStorage );//凸包中的缺陷
if( defect->total >=40 ) {cout << " Closed Palm " << endl;}
else if( defect->total >=30 && defect->total <40 ) {cout << " Open Palm " << endl;}
else{ cout << " Fist " << endl;}
cout << "defet: " << defect->total << endl;
//对单通道图像检索轮廓，返回第一个轮廓的指针
CvSeq* contours2 = NULL;
double result = 0, result2 = 0;
while(contours)//得到最大的两个感兴趣区
{
result = fabs( cvContourArea( contours, CV_WHOLE_SEQ ) );//计算感兴趣区域的像素点数
if ( result > result2) {result2 = result; contours2 = contours;};
contours = contours->h_next;
}
if ( contours2 )//最大的感兴趣区，ROI
{
//cout << "contours2: " << contours2->total << endl;
CvRect rect = cvBoundingRect( contours2, 0 );//返回一个2d矩形的点集合
cvRectangle( bg, cvPoint(rect.x, rect.y + rect.height), cvPoint(rect.x + rect.width, rect.y), CV_RGB(200, 0, 200), 1, 8, 0 );
//在bg的矩形区域中画rect的图形
//cout << "Ratio: " << rect.width << ", " << rect.height << ", " << (float)rect.width / rect.height << endl;
int checkcxt = cvCheckContourConvexity( contours2 );//检测输入的轮廓是否是凸的
//cout << checkcxt <<endl;
CvSeq* hull = cvConvexHull2( contours2, 0, CV_CLOCKWISE, 0 );//二维凸包
CvSeq* defect = cvConvexityDefects( contours2, hull, dftStorage );//凸包中的缺陷
if( defect->total >=40 ) {cout << " Closed Palm " << endl;}
else if( defect->total >=30 && defect->total <40 ) {cout << " Open Palm " << endl;}
else{ cout << " Fist " << endl;}
cout << "defet: " << defect->total << endl;
CvBox2D box = cvMinAreaRect2( contours2, minStorage );//包围所有点的轮廓的最小矩形
//cout << "box angle: " << (int)box.angle << endl;
cvCircle( bg, cvPoint(box.center.x, box.center.y), 3, CV_RGB(200, 0, 200), 2, 8, 0 );//画圆
cvEllipse( bg, cvPoint(box.center.x, box.center.y), cvSize(box.size.height/2, box.size.width/2), box.angle, 0, 360, CV_RGB(220, 0, 220), 1, 8, 0 );//椭圆
//cout << "Ratio: " << (float)box.size.width/box.size.height <<endl;
}
//cvShowImage( "hsv-msk", hsv_mask); hsv_mask->origin = 1;
//IplImage* contour = cvCreateImage( sz, 8, 3 );
//cout << "box angle: " << (int)box.angle << endl;
cvCircle( bg, cvPoint(box.center.x, box.center.y), 3, CV_RGB(200, 0, 200), 2, 8, 0 );//画圆
cvEllipse( bg, cvPoint(box.center.x, box.center.y), cvSize(box.size.height/2, box.size.width/2), box.angle, 0, 360, CV_RGB(220, 0, 220), 1, 8, 0 );//椭圆
//cout << "Ratio: " << (float)box.size.width/box.size.height <<endl;
}
//cvShowImage( "hsv-msk", hsv_mask); hsv_mask->origin = 1;
//IplImage* contour = cvCreateImage( sz, 8, 3 );
cvDrawContours( bg, contours2, CV_RGB( 0, 200, 0), CV_RGB( 0, 100, 0), 1, 1, 8, cvPoint(0,0));//绘制轮廓的图像到bg图像中
cvShowImage( "src", src);
//cvShowImage("src",src3);
//contour->origin = 1; cvShowImage( "contour", contour);
//cvReleaseImage( &contour);
//cvShowImage("src",src3);
//contour->origin = 1; cvShowImage( "contour", contour);
//cvReleaseImage( &contour);
cvNamedWindow("bg",0);
cvShowImage("bg",bg);
cvNamedWindow( "ROI",0);
cvShowImage("ROI",hsv_mask2);
//cvShowImage("ROI",hsv_mask2);
cvReleaseImage( &bg);
cvShowImage("bg",bg);
cvNamedWindow( "ROI",0);
cvShowImage("ROI",hsv_mask2);
//cvShowImage("ROI",hsv_mask2);
cvReleaseImage( &bg);
c = cvWaitKey( 100);
}
//////
cvReleaseCapture( &capture);
cvDestroyAllWindows();
}
//////
cvReleaseCapture( &capture);
cvDestroyAllWindows();
}
缺陷在于不能区分脸的颜色和手的颜色，而且识别出来的是全部在这个范围内的东西，难以满足要求。
缺陷在于不能区分脸的颜色和手的颜色，而且识别出来的是全部在这个范围内的东西，难以满足要求。
他提供的是“虚拟鼓”比较有特色，是
原理
1.原始视频->人脸识别
2.根据人脸的颜色分布，找到手
3.去除手和脸以外的区域
3.1 对手和脸进行连通区域划分
3.2 对双手进行运动分析，满足条件则击鼓
2.根据人脸的颜色分布，找到手
3.去除手和脸以外的区域
3.1 对手和脸进行连通区域划分
3.2 对双手进行运动分析，满足条件则击鼓
他的方法从部分上克服了上述缺陷，但是从他提供的视频可以看出来他穿的一个长袖的衣服，而且衣服颜色和肤色差距很大，如果是短袖的衣服效果估计不会怎么样了。