三维重建PCL：点云单侧面正射投影

终于把点云单侧面投影正射投影的代码写完了，为一个阶段，主要使用平面插值方法，且只以XOY平面作为的正射投影面。有些凑合的地方，待改进。

方法思路：使用Mesh模型，对每一个表面进行表面重建。借助OpenCV Mat类型对投影平面进行内点判断，对内点位置进行插值。

OpenCV cv::polylines 和lines 进行画图的时候都会出现问题，因此在某些时刻无法使用连通域查找的方法进行内点检测，应该重写line方法。

1.使用Mesh载入ply模型，和同步载入点云，也可以从mesh直接Copy点云。

		pcl::PolygonMesh cloudMesh;

		pcl::io::loadPolygonFileOBJ(ViewPath, cloudMesh);

		pcl::fromPCLPointCloud2(cloudMesh.cloud, *cloud);

		ViewPath = "D:/DataSet/RGB_data/teapot.pcd";

		pcl::io::savePCDFileASCII(ViewPath, *cloud);//一定要注意高和宽进行赋值

		pcl::visualization::PCLVisualizer  Viewer;//pcl::visualization::PCLVisualizer  ViewerMesh;

		Viewer.addPolygonMesh(cloudMesh);

		int FrameX = 1000;

		int FrameY = 1000;

		int FrameZ = 1000;

		int Centroid = 0;

		int num = 12;

		float gap = 3.141592653/num;

		Eigen::Vector4f  ViewPoint( 0.0, 0.0, 0.0, 1);//使用弧度

		ViewPoint[0] = gap*i;

		cv::Mat imgGray = viewEx->getCloudViewByEdge(

				cloud, cloudView, cloudMesh, ViewPath, FrameX, FrameY, FrameZ, Centroid, ViewPoint);

2. 使用平面填充方法进行投影...

//使用多边形填充的方法进行投影

//获取点云侧面投影

//输入：点云的点集、边集

cv::Mat CViewExtract::getCloudViewByEdge(

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloud,

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloudView,

	pcl::PolygonMesh  &cloudMesh,

	std::string ViewPath,

	int FrameX, int FrameY, int FrameZ,

	int Centroid,

	Eigen::Vector4f  &ViewPoint)

{

	int BbxSize = FrameX;

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloudTrans(new pcl::PointCloud<pcl::PointXYZ>);

	this->viewTrans(cloud, cloudTrans, ViewPoint);

	//对点云进行侧面投影

	std::vector<pcl::PointCloud<pcl::PointXYZ>> surfaces;

	pcl::PointCloud<pcl::PointXYZ>::Ptr  surface(new pcl::PointCloud<pcl::PointXYZ>);

	//计算平面

	genSurfaceFromVertices(cloudMesh.polygons, cloudTrans, surface);//由cloud替换cloudtrans，mesh只是一个索引

	cv::Mat imgGray = getViewer(surface, cloudTrans, cloudView);

	return imgGray;

}

3. 子函数

视点变换

float CViewExtract::viewTrans(

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloudSrc,

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloudTrans,

	Eigen::Vector4f  &AngleTrans)

{

	//1. Trans the VIew...

	float AngleA, AngleB, AngleC;//声明视角//初始化 作为原始角度

	AngleA = AngleTrans[0];//49.0/pi;

	AngleB = AngleTrans[1];//78.9/pi;

	AngleC = AngleTrans[2];//34.8/pi;

	int size = cloudSrc->points.size();

	cloudTrans->resize(0);

	cloudTrans->reserve(size);

	//位姿识别角度变换矩阵/

	Eigen::Matrix4f TransX, TransY, TransZ;

	//初始化三个矩阵！变换！

	TransX << 1, 0, 0, 0,

		0, cos(AngleA), -sin(AngleA), 0,

		0, sin(AngleA), cos(AngleA), 0,

		0, 0, 0, 1;//

	TransY << cos(AngleB), 0, sin(AngleB), 0,

		0, 1, 0, 0,

		-sin(AngleB), 0, cos(AngleB), 0,

		0, 0, 0, 1;

	TransZ << cos(AngleC), -sin(AngleC), 0, 0,

		sin(AngleC), cos(AngleC), 0, 0,

		0, 0, 1, 0,

		0, 0, 0, 1;

	//点云模型角度变换

	Eigen::Vector4f Centroid;

	Centroid << 0, 0, 0, 0;

	//pcl::compute3DCentroid(*cloudSrc, Centroid);

	for (int idx = 0; idx < cloudSrc->points.size(); ++idx){

		Eigen::Vector4f PointSrc, PointDest;//维数一致！

		PointSrc[0] = cloudSrc->points[idx].x - Centroid[0];

		PointSrc[1] = cloudSrc->points[idx].y - Centroid[1];

		PointSrc[2] = cloudSrc->points[idx].z - Centroid[2];

		//PointSrc[3] = 1;

		PointDest = (TransX*(TransY*(TransZ*PointSrc)));//创建矩阵无效！

		//cloudSrc->points[idx].x = PointDest[0] + Centroid[0];

		//cloudSrc->points[idx].y = PointDest[1] + Centroid[1];

		//cloudSrc->points[idx].z = PointDest[2] + Centroid[2];

		//cloudSrc->points[idx].rgb = cloudSrc->points[idx].rgb;

		pcl::PointXYZ p;

		p.x = PointDest[0] + Centroid[0];

		p.y = PointDest[1] + Centroid[1];

		p.z = PointDest[2] + Centroid[2];

		//p.x *= 5; p.y *= 5; p.z *= 5;

		cloudTrans->push_back(p);

		//cloudTrans->points[idx].rgb = cloudSrc->points[idx].rgb;

	}

	return 1.0;

}

重建表面

//仍然产生整数的空隙，应该把原始点云扩充到四个整数邻域//前N个为原始点云

int CViewExtract::genSurfaceFromVertices(std::vector< ::pcl::Vertices>  &vertices,

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloud,

	pcl::PointCloud<pcl::PointXYZ>::Ptr  surfaces)

{

	int size = vertices.size();

	for ( int i = 0; i < size; ++i ){

		pcl::PointCloud<pcl::PointXYZ>::Ptr

			surface(new pcl::PointCloud<pcl::PointXYZ>);//(&surfaces[i]);//

		genSurfaceFromVertices( vertices[i], cloud, surface, i);

			for ( auto p : surface->points){

				surfaces->points.push_back(p);

			}

		surface->clear();

	}

	return size;

}

//从表面获取点云，对单个面获取点云

int CViewExtract::genSurfaceFromVertices(const pcl::Vertices &vertice,

	const pcl::PointCloud<pcl::PointXYZ>::Ptr  cloud,

	pcl::PointCloud<pcl::PointXYZ>::Ptr  surface,

	int idx)

{

	int size = 0;

	int nV = vertice.vertices.size();

	pcl::PointCloud<pcl::PointXYZ>::Ptr Votex(new pcl::PointCloud<pcl::PointXYZ>);

	for (int i = 0; i < nV; ++i){

		pcl::PointXYZ p(cloud->points[vertice.vertices[i]]);

		Votex->points.push_back(p);

	}

	int bx, by, bz;

	std::vector<std::pair<float, float> > minmax(3);

	//findMinMax( cloud, minmax );

	MathCal::findMinMax(Votex, minmax);

	bx = ceil(minmax[0].second - minmax[0].first);

	by = ceil(minmax[1].second - minmax[1].first);

	bz = ceil(minmax[2].second - minmax[2].first);

	//bx *= 10; by *= 10; bz *= 10;

	//取平面上的点//以Z轴为正射方向

	std::vector<cv::Point2f> vetexs(0);//生成顶点

	int xmin = minmax[0].first;

	int ymin = minmax[1].first;

	int zmin = minmax[2].first;

	for (int i = 0; i < vertice.vertices.size(); ++i){

		int idx = vertice.vertices[i];

		pcl::PointXYZ p = cloud->points[idx];

		cv::Point2f p2 = cv::Point2f(p.x - xmin, p.y - ymin);

		//p2.x *= 10;p2.y *= 10;

		vetexs.push_back(p2);

	}

	//生成图像//使用OpenCV画出对应二维图片

	cv::Mat img = cv::Mat::zeros(by + 1, bx + 1, CV_8UC3);

	cv::Mat _lableImg;

	std::vector<std::vector<cv::Point > > foreAreas;

	//wishchin::findInliners2d( img, vetexs, _lableImg, foreAreas );

	MathCal::findInliners2dNoCon(img, vetexs, _lableImg, foreAreas);

	float zmean = 0;

	if (foreAreas.size()>0)

	{

		size = foreAreas[0].size();

		//获取平面方程//Ax + By + Cz + D

		//std::vector<float> getPlaneParam(std::vector<cv::Point2f> vetexs);

		std::vector<pcl::PointXYZ> VotexP;

		for (int i = 0; i < vetexs.size(); ++i){

			pcl::PointXYZ p(vetexs[i].x, vetexs[i].y, (cloud->points[vertice.vertices[i]].z - zmin));

			VotexP.push_back(p);

			zmean += cloud->points[vertice.vertices[i]].z;

		}

		zmean /= vetexs.size();

		std::vector<float> abcd = MathCal::getPlaneParam(VotexP);

		//从平面上取点

		surface->points.resize(0);

		float x, y, z;//Mat xy的方向与 PCL是相反的！！！

		for (int i = 0; i < size; ++i){

			x = foreAreas[0][i].x;

			y = foreAreas[0][i].y;

			//x = bx + 0 - x;

			//y = by + 0 - y;

			z = 0-(abcd[0] * x + abcd[1] * y + abcd[3]) / abcd[2];

			pcl::PointXYZ p(x,y, z);

			p.x += xmin; p.y += ymin; p.z += zmin;//移到原位

			surface->points.push_back(p);

		}

		surface->height = 1;

		surface->width = size;

	}

	return size;

}

寻找多边形的内点

//寻找多边形的内点//取整数点//只能取凸多边形

//通过判断各个边的左边右边来进行计算//通过计算在多边形的内侧外侧计算-有点慢

//不使用连通域查找//

int MathCal::findInliners2dNoCon(cv::Mat &img, std::vector<cv::Point2f> &vetexs,

	cv::Mat &_lableImg, std::vector<std::vector<cv::Point > > &foreAreas)

{

	int size = 0;

	//获取多边形边集

	std::vector<std::vector<cv::Point2f>>  edges(0);

	if (vetexs.size() > 2)

	{

		std::vector<cv::Point2f> edge(0);

		edge.push_back(cv::Point2f(vetexs[vetexs.size() - 1]));

		edge.push_back(cv::Point2f(vetexs[0]));

		edges.push_back(edge);

		for (int i = 1; i < vetexs.size(); ++i)

		{

			edge.resize(0);

			edge.push_back(cv::Point2f(vetexs[i - 1]));

			edge.push_back(cv::Point2f(vetexs[i]));

			edges.push_back(edge);

		}

	}

	//测试

	//bool isIn =isInliner(cv::Point2f(2, 538), vetexs, edges);//true

	//bool isIn = isInliner(cv::Point2f(476, 258), vetexs, edges);//false

	//bool isIn = isInliner(cv::Point2f(704, 137), vetexs, edges);

	//bool isIn = isInliner(cv::Point2f(6, 11), vetexs, edges);

	//取多边形的质心

	//从质心开始查找连通域//需要提前染色

	std::vector<cv::Point2d> inliners;

	cv::Point2d seed(-1, -1);

	bool findseed = false;

	std::vector<cv::Point > foreArea;

	for (int i = 0; i < img.rows; ++i)

	{

		unsigned char *ptrm = img.ptr<unsigned char>(i);

		for (int j = 0; j < img.cols; ++j)

		{

			int c = *ptrm;

			bool isIn = false;

			isIn = isInliner(cv::Point2f(j, i), vetexs, edges); //！！！！！出现错误！待调试！//出现了两个方向都奇异的直角三角形

			if (isIn){

				seed.x = j;

				seed.y = i;

				foreArea.push_back(seed);

			}

			++ptrm;

		}

	}

	if (foreArea.size()>0){

		foreAreas.push_back(foreArea);

	}

	size = foreAreas.size();

	return size;

}

获取平面方程

//获取平面方程//Ax + By + Cz + D

std::vector<float> MathCal::getPlaneParam(const std::vector<pcl::PointXYZ> &votexs)

{

	std::vector<float> abcd;

	if (votexs.size() < 3){

		return abcd;

	}

	else

	{//取前三个点计算平面

		float x1, x2, x3, y1, y2, y3, z1, z2, z3;

		x1 = votexs[0].x; x2 = votexs[1].x; x3 = votexs[2].x;

		y1 = votexs[0].y; y2 = votexs[1].y; y3 = votexs[2].y;

		z1 = votexs[0].z; z2 = votexs[1].z; z3 = votexs[2].z;

		float A = y1*(z2 - z3) + y2*(z3 - z1) + y3*(z1 - z2);

		float B = z1*(x2 - x3) + z2*(x3 - x1) + z3*(x1 - x2);

		float C = x1*(y2 - y3) + x2*(y3 - y1) + x3*(y1 - y2);

		float D = -(x1*(y2*z3 - y3*z2) + x2*(y3*z1 - y1*z3) + x3*(y1*z2 - y2*z1));

		abcd.push_back(A); abcd.push_back(B); abcd.push_back(C);

		abcd.push_back(D);

	}

	return abcd;

}

int MathCal::findMinMax(pcl::PointCloud<pcl::PointXYZ>::Ptr cloud,

	std::vector<std::pair<float, float> > &minmax)

{

	float minX = 10000000;

	float minY = 10000000;

	float minZ = 10000000;

	float maxX = -10000000;

	float maxY = -10000000;

	float maxZ = -10000000;

	for (int i = 0; i < cloud->points.size(); ++i)

	{

		pcl::PointXYZ p(cloud->points[i]);

		if (minX >p.x) minX = p.x;

		if (minY > p.y) minY = p.y;

		if (minZ > p.z) minZ = p.z;

		if (maxX < p.x) maxX = p.x;

		if (maxY < p.y) maxY = p.y;

		if (maxZ < p.z) maxZ = p.z;

	}

	minmax.resize(0);

	minmax.push_back(std::pair<float, float>(minX, maxX));

	minmax.push_back(std::pair<float, float>(minY, maxY));

	minmax.push_back(std::pair<float, float>(minZ, maxZ));

	return 1;

}

//获取点云，直接从上一步获取

cv::Mat  CViewExtract::getViewer(const pcl::PointCloud<pcl::PointXYZ>::Ptr  surface,

	const pcl::PointCloud<pcl::PointXYZ>::Ptr  cloud,

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloudView)

{

	//获取大包围盒

	int bx, by, bz;

	std::vector<std::pair<float, float> > minmax(3);

	MathCal::findMinMax(surface, minmax);

	float xmin = minmax[0].first;

	float ymin = minmax[1].first;

	float zmin = minmax[2].first;

	bx = ceil(minmax[0].second - minmax[0].first);

	by = ceil(minmax[1].second - minmax[1].first);

	bz = ceil(minmax[2].second - minmax[2].first);

	std::vector<float> bbx;

	bbx.push_back(bx); bbx.push_back(by); bbx.push_back(bz);

	//std::vector<bool > visibies(surface->points.size() );//直接重新生成点，不取浮点数

	//生成图像//使用OpenCV画出对应灰度图片

	cv::Mat img = cv::Mat::zeros(by + 1, bx + 1, CV_32FC1);

	//for ( pcl::PointXYZ p: surface->points )

	for (int i = 0; i < surface->points.size(); ++i)

	{

		pcl::PointXYZ p = surface->points[i];

		int x = p.x - xmin;

		int y = p.y - ymin;

		float z = p.z - zmin + 1;

		//取最大Z//必须使用四邻域

		int x1 = floor(x); int x2 = ceil(x); //if (x1 < 0) x1 = 0;

		int y1 = floor(y); int y2 = ceil(y); //if (y1 < 0) y1 = 0;

		MathCal::cutValue(x1, 0, img.cols - 1);

		MathCal::cutValue(x2, 0, img.cols - 1);

		MathCal::cutValue(y1, 0, img.rows - 1);

		MathCal::cutValue(y2, 0, img.rows - 1);

		if ( img.at<float>(y1, x1) < z) img.at<float>(y1, x1) = z;

		if ( img.at<float>(y1, x2) < z) img.at<float>(y1, x2) = z;

		if ( img.at<float>(y2, x2) < z) img.at<float>(y2, x2) = z;

		if ( img.at<float>(y2, x1) < z) img.at<float>(y2, x1) = z;

	}

	//补加原始点云的四邻域//原始点云已添加，不再重复补偿，原始点云已删除

	pcl::PointCloud<pcl::PointXYZ>::Ptr  cloudFourNear(new pcl::PointCloud<pcl::PointXYZ>);

	for (int i = 0; i < cloud->points.size(); ++i)

	{

		pcl::PointXYZ p = cloud->points[i];

		float x = p.x - xmin;

		float y = p.y - ymin;

		float z = p.z - zmin + 1;

		int x1 = floor(x); int x2 = ceil(x); //if (x1<0) x1 = 0; if (x2<0) x2 = 0;

		int y1 = floor(y); int y2 = ceil(y); //if (y1<0) y1 = 0; if (y2<0) y2 = 0;

		MathCal::cutValue(x1, 0, img.cols - 1);

		MathCal::cutValue(x2, 0, img.cols - 1);

		MathCal::cutValue(y1, 0, img.rows - 1);

		MathCal::cutValue(y2, 0, img.rows - 1);

		//重复填充四邻域

		//若未被填充，则填充

		if ( 0.0001> img.at<float>(y1, x1) ) img.at<float>(y1, x1) = z;

		if (0.0001> img.at<float>(y1, x2)) img.at<float>(y1, x2) = z;

		if (0.0001> img.at<float>(y2, x2)) img.at<float>(y2, x2) = z;

		if (0.0001> img.at<float>(y2, x1)) img.at<float>(y2, x1) = z;

	}

	cloudView->resize(0);

	cv::Mat imgGray = cv::Mat::zeros(by + 1, bx + 1, CV_8UC1);

	float x, y, z;

	for (int i = 0; i < img.rows; ++i)

	{

		float *ptr = img.ptr<float>(i);

		unsigned char *ptrg = imgGray.ptr<unsigned char>(i);

		for (int j = 0; j < img.cols; ++j)

		{

			if (*ptr > 0)

			{

				x = j - xmin;

				y = i - ymin;

				z = *ptr - zmin-1;

				cloudView->points.push_back(pcl::PointXYZ(x, y, z));

				if (z < 0) z = 0;

				if (z >255) z = 255;

				*ptrg = (unsigned char)z;

			}

			++ptr;

			++ptrg;

		}

	}

	cloudView->height = 1;

	cloudView->width = cloudView->points.size();

	//cv::flip(imgGray, imgGray, 2);

	//cv::imshow("imgGray", imgGray);

	//cv::waitKey(0);

	return imgGray;

}

void MathCal::cutValue(int &inv, const int start, const int end)

{

    if (inv < start) inv = start;

    if (inv > end) inv = end;

    //return inv;

}

输出结果；

通过传入viewpoint输出不同的位姿可见面

三维重建PCL：点云单侧面正射投影的更多相关文章

PCL点云库：ICP算法
ICP(Iterative Closest Point迭代最近点)算法是一种点集对点集配准方法.在VTK.PCL.MRPT.MeshLab等C++库或软件中都有实现,可以参见维基百科中的ICP Alg ...
恒天云单节点部署指南--OpenStack H版本虚拟机单节点部署解决方案
本帖是openstack单节点在虚拟机上部署的实践.想要玩玩和学习openstack的小伙伴都看过来,尤其是那些部署openstack失败的小伙伴.本帖可以让你先领略一下openstack的魅力.本I ...
PCL中点云数据格式之间的转化
(1) 关于pcl::PCLPointCloud2::Ptr和pcl::PointCloud<pcl::PointXYZ>两中数据结构的区别 pcl::PointXYZ::PointXYZ ...
PCL点云库中的坐标系（CoordinateSystem）
博客转载自:https://blog.csdn.net/qq_33624918/article/details/80488590 引言世上本没有坐标系,用的人多了,便定义了坐标系统用来定位.地理坐标 ...
Windows下安装PCL点云库
原文链接:http://blog.csdn.net/u012337034/article/details/38270109 简介: 在Windows下安装PCL点云库的方法大概有两种: ...
Windows 8 64位系统在VS2010 32位软件上搭建 PCL点云库开发环境
Windows 8 64位系统在VS2010 32位软件上搭建 PCL点云库开发环境下载PCL For windows 软件包到这个网站下载PCL-All-In-One Installer: ...
PCL学习之：将超声数据按照PCL点云方式发布出去
前言:基于2D激光雷达的机器人,想让它跑自动导航,众所周知有2个比较明显的缺陷,1,那就是普通的激光雷达无法对玻璃或是镜面物体有反映; 2,机器人避障时只能对某一个平面的物体有反映,超过或者低于这个平 ...
PCL点云特征描述与提取（1）
3D点云特征描述与提取是点云信息处理中最基础也是最关键的一部分,点云的识别.分割,重采样,配准曲面重建等处理大部分算法,都严重依赖特征描述与提取的结果.从尺度上来分,一般分为局部特征的描述和全局特征的 ...
PCL—点云分割（最小割算法）低层次点云处理
1.点云分割的精度在之前的两个章节里介绍了基于采样一致的点云分割和基于临近搜索的点云分割算法.基于采样一致的点云分割算法显然是意识流的,它只能割出大概的点云(可能是杯子的一部分,但杯把儿肯定没分割出 ...

随机推荐

python爬虫【第1篇】
一.文件读写 1.打开文件 # 以读文件模式代开new.txt f=open(r"c:\new.txt",“r”) f=open("c:\new.txt",“r ...
IIS网站无法启动，提示另一个程序正在使用此文件
s还用netstat -ano命令,观察哪一个进程正在使用80端口,任务管理中勾选PID,看看是哪个程序,关掉即可
一张图轻松记住PHP的*类*以及private和protected的区别
上图概要的说了下php类的特性,类的方法同属性类似. 图中B类继承自A类,B是A的子类,$x和$y都是B的实例化对象. 1. 原型引用:[A:: . B:: ],仅限public stati ...
ubuntu将mysql、nginx添加到环境变量中
vim /etc/profile 添加 export PATH="$PATH:/usr/local/mysql/bin" export PATH="$PATH:/usr/ ...
[译]NUnit--Installation(三)
Installation NUnit安装程序默认安装文件路径为C:\Program Files\NUnit 2.6.2.根据用户选择安装的选项,安装文件有三个子文件夹:bin.doc.samples. ...
在MAC端查看win7
在MAC端查看win7,在finder中打开网络,输入win7地址,填入用户名和密码,就可以了
Flask的flask-sqlalchemy
flask-sqlalchemy是Flask和SQLAlchemy的管理者 - db = SQLAlchemy() - 包含配置 - 包含ORM基类 - 包含create_all - engine - ...
再读headfirst-原则与模式摘录
原则 1.找到原则中可能需要变化之处,把它们独立出来,不要和那些不需要变化的代码混在一起 2.针对接口编程,而不是针对实现编程 3.多用组合,少用继承 4.依赖倒置原则:要依赖抽象,不要依赖具体类(不 ...
nodejs实现验证码
http://www.9958.pw/post/nodejs_lesson http://www.9958.pw/post/nodejscapp
log4j日志基本配置
Log4j有三个主要的组件:Loggers(记录器),Appenders (输出源)和Layouts(布局).这里可简单理解为日志类别,日志要输出的地方和日志以何种形式输出.综合使用这三个组件可以轻松 ...

三维重建PCL：点云单侧面正射投影

三维重建PCL：点云单侧面正射投影的更多相关文章

随机推荐

热门专题