Kinect运用OpenNI产生点云
点云数据是由depthmap得到，利用OpenNI中depthgenerator node获取depth map，再将depth map转为real world的坐标值。
OpenNI的depth map的像素值代表的是实际空间位置到Kinect所在平面的深度值，即到该Kinect平面的距离，单位为mm；
利用DepthGenerator 中GetDepthMap或者getMetaData均可获得深度图像的像素值；GetMetaData函数是将像素值填充到DepthMetaData的对象中，GetDepthMap是直接读取像素值。
XnPoint3D是保持类型为float的x、y、z的结构，为将像素值转为点云，须利用DepthGenerator的ConvertProjectToRealWorld函数：
XnStatus ConvertProjectiveToRealWorld (XnUInt32 nCount, const XnPoint3D aProjective[], XnPoint3D aRealWorld[]) const
其中nCount表示点数数目，aProjective表示投影坐标下所有(x,y,z)的值，aRealWorld表示实际空间坐标值，该坐标系的坐标原点是：depth map的中心投影到Kinect所在平面上的点。x-y平面即为Kinect所在平面：
z轴方向代表了Kinect镜头所对方向。OpenNI获得的深度值即为实际坐标空间的z值，也就是通过ConvertProjectToRealWorld函数是不改变z值得，x，y是depth map的索引位置（x：为column y：为row），须将x，y转为实际坐标空间的x，y值；围观OpenNI的源码，ConvertProjectToRealWorld函数实现是通过单位换算得到的：
- 获得field of view（视角）值：获取horizontal和vertical的的视角值，即Kinect水平和垂直视角；
- 单位换算比例：α表示水平视角角度值 、β表示垂直视角角度值，换算关系RealWorldXtoZ=2tan(α/2) RealWorldYtoZ = 2tan(β/2)，即得到了depth map的宽度值w/高度值h与距离d的比值
- 求出实际的x，y值：NormalizedX = x / outputMode.nXRes - 0.5，即深度图像左侧在x轴负方向上，右侧在x轴正方向上；同样NormalizedY = 0.5 - y / outputMode.nYRes，上侧在y轴正方向上，下侧在y轴负方向上
最后x = NormalizedX * Z * RealWorldXtoZ y = NormalizedX * Z * RealWorldXtoZ
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
欲使用ConvertProjectiveToRealWorld函数须将projective下(x,y,z)值保存：XnPoint3D *projectivePointCloud = new XnPoint3D [pointNumber] 再将depth map的(x,y)索引对应的像素值保存到projectivePointCloud ，定义XnPoint3D *realWorldPointCloud = new XnPoint3D [pointNumber] ，利用ConvertProjectiveToRealWorld（pointNumber， projectivePointCloud ， realWorldPointCloud )完成Projective到RealWorld的数据转换。
最后就可以利用OpenGL将realWorldPointCloud的点云 进行绘制了...
部分代码如下：
#include <XnCppWrapper.h>
#include <stdio.h>
using namespace xn;
/**
* use 1.create object 2.set output mode 3.initilization 4.update and processing
* ex.. 1.DepthmapPointCloud cloud; 2.cloud.setOutputMode(getDefaultOutputMode()); 3.cloud.init(); 4. ... ;
*/
class DepthmapPointCloud
{
public:
DepthmapPointCloud(Context &contex, DepthGenerator &depthGenerator)
: m_hContex(contex), m_hDepthGenerator(depthGenerator), m_nStatus(XN_STATUS_OK),
m_pProjecttiveCloud(0), m_pRealworldCloud(0) {}
~DepthmapPointCloud()
{
stop();
}
void init();
void stop();
void setOutputMode(const XnMapOutputMode &outputMode);
void updataPointCloud(); // updata the point cloud datas
inline XnPoint3D* getPointCloudData()
{
return m_pRealworldCloud;
}
inline const XnUInt32 getPointCloudNum() const
{
return m_nOutputMode.nXRes*m_nOutputMode.nYRes;
}
inline const XnMapOutputMode& getXYRes() const
{
return m_nOutputMode;
}
inline const XnMapOutputMode getDefaultOutputMode()
{
// set the depth map output mode
XnMapOutputMode outputMode = {XN_VGA_X_RES, XN_VGA_Y_RES, 30};
return outputMode;
}
/**
* test: return projective point cloud data
*/
inline XnPoint3D* getRawPointData()
{
return m_pProjecttiveCloud;
}
/**
* test: get field of view
*/
inline void getFieldofView(XnFieldOfView &FOV)const
{
m_hDepthGenerator.GetFieldOfView(FOV);
}
private:
Context &m_hContex;
DepthGenerator &m_hDepthGenerator;
XnStatus m_nStatus;
XnMapOutputMode m_nOutputMode;
XnPoint3D *m_pProjecttiveCloud;
XnPoint3D *m_pRealworldCloud;
inline void printError()
{
if (m_nStatus != XN_STATUS_OK)
{
printf("Error: %s", xnGetStatusString(m_nStatus));
exit(-1);
}
}
DepthmapPointCloud(const DepthmapPointCloud &rhs); // don't define copy constructor
DepthmapPointCloud& operator=(const DepthmapPointCloud &rhs); // don't define assignment operator
};
void DepthmapPointCloud::setOutputMode(const XnMapOutputMode &outputMode)
{
m_nOutputMode.nFPS = outputMode.nFPS;
m_nOutputMode.nXRes= outputMode.nXRes;
m_nOutputMode.nYRes = outputMode.nYRes;
}
void DepthmapPointCloud::init()
{
m_nStatus = m_hContex.Init();
printError();
m_nStatus = m_hDepthGenerator.Create(m_hContex);
printError();
// set output mode
m_nStatus = m_hDepthGenerator.SetMapOutputMode(m_nOutputMode);
printError();
// start generating
m_nStatus = m_hContex.StartGeneratingAll();
printError();
// allocate memory to projective point cloud and real world point cloud
m_pProjecttiveCloud = new XnPoint3D[getPointCloudNum()];
m_pRealworldCloud = new XnPoint3D[getPointCloudNum()];
}
void DepthmapPointCloud::stop()
{
m_hContex.Release();
delete [] m_pProjecttiveCloud;
delete [] m_pRealworldCloud;
}
void DepthmapPointCloud::updataPointCloud()
{
try
{
// update to next frame data
m_nStatus = m_hContex.WaitOneUpdateAll(m_hDepthGenerator);
// printf("DataSize: %d\n", m_hDepthGenerator.GetDataSize()); // test
// store depthmap data to projective cloudpoint data
XnUInt32 index, shiftStep;
for (XnUInt32 row=0; row<m_nOutputMode.nYRes; ++row)
{
shiftStep = row*m_nOutputMode.nXRes;
for (XnUInt32 column=0; column<m_nOutputMode.nXRes; ++column)
{
index = shiftStep + column;
m_pProjecttiveCloud[index].X = (XnFloat)column;
m_pProjecttiveCloud[index].Y = (XnFloat)row;
m_pProjecttiveCloud[index].Z = m_hDepthGenerator.GetDepthMap()[index]*0.001f; // mm -> m
}
}
if (m_nStatus != XN_STATUS_OK) throw m_nStatus;
// convert projective pointcloud to real world pointcloud
m_hDepthGenerator.ConvertProjectiveToRealWorld(m_nOutputMode.nXRes*m_nOutputMode.nYRes, m_pProjecttiveCloud, m_pRealworldCloud);
}
catch (...)
{
printError();
stop();
}
} 最后看看OpenGL下点云显示效果：..(*^__^*) ..
现在数据还是原始数据，比较粗糙，准备继续研究怎么将数据过滤平滑，让显示更好，呵呵，本身要做的事确是怎么识别手，跟踪手势，当然这是我最终目标了，
搞成点成果，后面继续努力吧...
附件下载