[转载]Kinect开发教程五：OpenNI获取人体骨架

http://blog.csdn.net/chenxin_130/article/details/6950480

临近毕业，小斤最近一直忙活着相关事宜，教程这边也搁浅了一阵。前几篇教程介绍了OpenNI的一些基本范例以及手势应用，但如果光用Kinect识别一些手势，总有点杀鸡用牛刀的感觉。在大部分体感应用中，获取骨架的步骤都不可缺少，这也是小斤一直想写的专题。

好了，废话不多说了，让我们进入正题吧！

在OpenNI库的enum XnSkeletonJoint中，定义了24个人体的关节，如下：

XN_SKEL_HEAD          = 1,   XN_SKEL_NECK            = 2,
XN_SKEL_TORSO         = 3,   XN_SKEL_WAIST           = 4,
XN_SKEL_LEFT_COLLAR       = 5,   XN_SKEL_LEFT_SHOULDER       = 6,
XN_SKEL_LEFT_ELBOW       = 7, XN_SKEL_LEFT_WRIST          = 8,
XN_SKEL_LEFT_HAND          = 9,   XN_SKEL_LEFT_FINGERTIP   =10,
XN_SKEL_RIGHT_COLLAR   =11,   XN_SKEL_RIGHT_SHOULDER   =12,
XN_SKEL_RIGHT_ELBOW       =13, XN_SKEL_RIGHT_WRIST          =14,
XN_SKEL_RIGHT_HAND      =15,   XN_SKEL_RIGHT_FINGERTIP   =16,
XN_SKEL_LEFT_HIP          =17,   XN_SKEL_LEFT_KNEE            =18,
XN_SKEL_LEFT_ANKLE       =19, XN_SKEL_LEFT_FOOT            =20,
XN_SKEL_RIGHT_HIP          =21,   XN_SKEL_RIGHT_KNEE          =22,
XN_SKEL_RIGHT_ANKLE       =23,   XN_SKEL_RIGHT_FOOT          =24

小斤试下来，目前可使用的有14个关节，如下图：

先上代码：

[cpp] view plain copy

#include <stdlib.h>
#include <iostream>
#include <vector>
#include <XnCppWrapper.h>
#include <XnModuleCppInterface.h>
#include "cv.h"
#include "highgui.h"
using namespace std;
using namespace cv;
//#pragma comment (lib,"cv210")
//#pragma comment (lib,"cxcore210")
//#pragma comment (lib,"highgui210")
//#pragma comment (lib,"OpenNI")
//【1】
xn::UserGenerator userGenerator;
xn::DepthGenerator depthGenerator;
xn::ImageGenerator imageGenerator;
/*
XN_SKEL_HEAD = 1, XN_SKEL_NECK = 2,
XN_SKEL_TORSO = 3, XN_SKEL_WAIST = 4,
XN_SKEL_LEFT_COLLAR = 5, XN_SKEL_LEFT_SHOULDER = 6,
XN_SKEL_LEFT_ELBOW = 7, XN_SKEL_LEFT_WRIST = 8,
XN_SKEL_LEFT_HAND = 9, XN_SKEL_LEFT_FINGERTIP =10,
XN_SKEL_RIGHT_COLLAR =11, XN_SKEL_RIGHT_SHOULDER =12,
XN_SKEL_RIGHT_ELBOW =13, XN_SKEL_RIGHT_WRIST =14,
XN_SKEL_RIGHT_HAND =15, XN_SKEL_RIGHT_FINGERTIP =16,
XN_SKEL_LEFT_HIP =17, XN_SKEL_LEFT_KNEE =18,
XN_SKEL_LEFT_ANKLE =19, XN_SKEL_LEFT_FOOT =20,
XN_SKEL_RIGHT_HIP =21, XN_SKEL_RIGHT_KNEE =22,
XN_SKEL_RIGHT_ANKLE =23, XN_SKEL_RIGHT_FOOT =24
*/
//a line will be drawn between start point and corresponding end point
int startSkelPoints[14]={1,2,6,6,12,17,6,7,12,13,17,18,21,22};
int endSkelPoints[14]={2,3,12,21,17,21,7,9,13,15,18,20,22,24};
// callback function of user generator: new user
void XN_CALLBACK_TYPE NewUser( xn::UserGenerator& generator, XnUserID user,void* pCookie )
{
cout << "New user identified: " << user << endl;
//userGenerator.GetSkeletonCap().LoadCalibrationDataFromFile( user, "UserCalibration.txt" );
generator.GetPoseDetectionCap().StartPoseDetection("Psi", user);
}
// callback function of user generator: lost user
void XN_CALLBACK_TYPE LostUser( xn::UserGenerator& generator, XnUserID user,void* pCookie )
{
cout << "User " << user << " lost" << endl;
}
// callback function of skeleton: calibration start
void XN_CALLBACK_TYPE CalibrationStart( xn::SkeletonCapability& skeleton,XnUserID user,void* pCookie )
{
cout << "Calibration start for user " << user << endl;
}
// callback function of skeleton: calibration end
void XN_CALLBACK_TYPE CalibrationEnd( xn::SkeletonCapability& skeleton,XnUserID user,XnCalibrationStatus calibrationError,void* pCookie )
{
cout << "Calibration complete for user " << user << ", ";
if( calibrationError==XN_CALIBRATION_STATUS_OK )
{
cout << "Success" << endl;
skeleton.StartTracking( user );
//userGenerator.GetSkeletonCap().SaveCalibrationDataToFile(user, "UserCalibration.txt" );
}
else
{
cout << "Failure" << endl;
//For the current version of OpenNI, only Psi pose is available
((xn::UserGenerator*)pCookie)->GetPoseDetectionCap().StartPoseDetection( "Psi", user );
}
}
// callback function of pose detection: pose start
void XN_CALLBACK_TYPE PoseDetected( xn::PoseDetectionCapability& poseDetection,const XnChar* strPose,XnUserID user,void* pCookie)
{
cout << "Pose " << strPose << " detected for user " << user << endl;
((xn::UserGenerator*)pCookie)->GetSkeletonCap().RequestCalibration( user, FALSE );
poseDetection.StopPoseDetection( user );
}
void clearImg(IplImage* inputimg)
{
CvFont font;
cvInitFont( &font, CV_FONT_VECTOR0,1, 1, 0, 3, 5);
memset(inputimg->imageData,255,640*480*3);
}
int main( int argc, char** argv )
{
char key=0;
int imgPosX=0;
int imgPosY=0;
// initial context
xn::Context context;
context.Init();
xn::ImageMetaData imageMD;
IplImage* cameraImg=cvCreateImage(cvSize(640,480),IPL_DEPTH_8U,3);
cvNamedWindow("Camera",1);
// map output mode
XnMapOutputMode mapMode;
mapMode.nXRes = 640;
mapMode.nYRes = 480;
mapMode.nFPS = 30;
// create generator
depthGenerator.Create( context );
depthGenerator.SetMapOutputMode( mapMode );
imageGenerator.Create( context );
userGenerator.Create( context );
//【2】
// Register callback functions of user generator
XnCallbackHandle userCBHandle;
userGenerator.RegisterUserCallbacks( NewUser, LostUser, NULL, userCBHandle );
//【3】
// Register callback functions of skeleton capability
xn::SkeletonCapability skeletonCap = userGenerator.GetSkeletonCap();
skeletonCap.SetSkeletonProfile( XN_SKEL_PROFILE_ALL );
XnCallbackHandle calibCBHandle;
skeletonCap.RegisterToCalibrationStart( CalibrationStart,&userGenerator, calibCBHandle );
skeletonCap.RegisterToCalibrationComplete( CalibrationEnd,&userGenerator, calibCBHandle );
//【4】
// Register callback functions of Pose Detection capability
XnCallbackHandle poseCBHandle;
userGenerator.GetPoseDetectionCap().RegisterToPoseDetected( PoseDetected,&userGenerator, poseCBHandle );
// start generate data
context.StartGeneratingAll();
while( key!=27 )
{
context.WaitAndUpdateAll();
imageGenerator.GetMetaData(imageMD);
memcpy(cameraImg->imageData,imageMD.Data(),640*480*3);
cvCvtColor(cameraImg,cameraImg,CV_RGB2BGR);
// get users
XnUInt16 userCounts = userGenerator.GetNumberOfUsers();
if( userCounts > 0 )
{
XnUserID* userID = new XnUserID[userCounts];
userGenerator.GetUsers( userID, userCounts );
for( int i = 0; i < userCounts; ++i )
{
//【5】
// if is tracking skeleton
if( skeletonCap.IsTracking( userID[i] ) )
{
XnPoint3D skelPointsIn[24],skelPointsOut[24];
XnSkeletonJointTransformation mJointTran;
for(int iter=0;iter<24;iter++)
{
//XnSkeletonJoint from 1 to 24
skeletonCap.GetSkeletonJoint( userID[i],XnSkeletonJoint(iter+1), mJointTran );
skelPointsIn[iter]=mJointTran.position.position;
}
depthGenerator.ConvertRealWorldToProjective(24,skelPointsIn,skelPointsOut);
//【6】
for(int d=0;d<14;d++)
{
CvPoint startpoint = cvPoint(skelPointsOut[startSkelPoints[d]-1].X,skelPointsOut[startSkelPoints[d]-1].Y);
CvPoint endpoint = cvPoint(skelPointsOut[endSkelPoints[d]-1].X,skelPointsOut[endSkelPoints[d]-1].Y);
cvCircle(cameraImg,startpoint,3,CV_#0000ff,12);
cvCircle(cameraImg,endpoint,3,CV_#0000ff,12);
cvLine(cameraImg,startpoint,endpoint,CV_#0000ff,4);
}
}
}
delete [] userID;
}
cvShowImage("Camera",cameraImg);
key=cvWaitKey(20);
}
// stop and shutdown
cvDestroyWindow("Camera");
cvReleaseImage(&cameraImg);
context.StopGeneratingAll();
context.Shutdown();
return 0;
}

【1】对于人体骨架的获取，小斤声明了UserGenerator这个生成器，UserGenerator具有检测新的User（以下称为人物）出现或者离开，获取画面中的人物数，人物位置信息，与上一教程介绍的GestureGenerator类似，通过注册回调函数的方式，一旦其检测到了动静（如人物出现），那么相应的回调函数就会被调用。

【2】小斤为UserGenerator注册了NewUser和LostUser两个回调函数，对应人物出现和人物消失。

【3】这里出现了一个新的Capability，SkeletonCapability。小斤为了避免混淆，常常将Capability理解为生成器的一种能力，比如SkeletonCapability就可以理解UserGenerator获取人物骨架信息的能力。

在获取人物骨架前，首先要进行标定的工作，因此SkeletonCapability需要注册两个回调函数CalibrationStart和CalibrationEnd，分别在人物标定开始与结束时调用。（在较早版本的OpenNI中，接口名可能有所变化）

【4】与【3】类似，userGenerator.GetPoseDetectionCap()获取了一个PoseDetectionCapability，这个Capability可以检测人物的特定姿势，目前来说，只支持Psi姿势，如图：

小斤并为其注册了回调函数PoseDetected，在检测到人物的Psi姿势时，会调用该函数。

将【2】【3】【4】的回调函数串联起来看，（1）人物出现会触发NewUser()，开始Pose检测；（2）检测到Pose会触发PoseDetected()，请求标定；（3）标定开始触发CalibrationStart()；（4）标定结束触发CalibrationEnd()，如果标定成功，那么调用SkeletonCapability的StartTracking()开始跟踪对应的人物。

【5】通过GetSkeletonJoint()方法，可以得到对应关节的XnSkeletonJointTransformation，这个结构体包含position和orientation，position中又包含一个position和fConfidence，分别代表关节的位置和可信度，orientation同样如此，包含关节的运动方向和可信度。这里小斤对24个关节都进行了操作，但能得到位置信息的只有14个。

这些步骤得到的position信息，是一个真实场景的3D坐标，需要通过投影转换到屏幕坐标，转换过程通过ConvertRealWorldToProjective()方法实现。

【6】为了更直观地输出显示，可以各个关节通过直线连接起来，形成一个人体的骨架。小斤定义了startSkelPoints和endSkelPoints数组，两个数组的值一一对应，代表一组起点终点的关节对，将每组起点和终点通过直线连接，比如HEAD与NECT与TORSO等。

整个程序启动后，先将身体正对摄像头（至少露出头部和上半身），控制台会显示“New user identified”，然后做出Psi姿势，在Pose Psi detected后，程序开始标定工作，此时维持Psi姿势数秒，标定成功后，骨架就会正确显示出来了。祝大家玩得愉快。

posted @ 2012-01-16 10:44 火腿骑士阅读(545) 评论(0) 编辑收藏举报

刷新页面返回顶部

[转载]Kinect开发教程五：OpenNI获取人体骨架

公告