R树是一种多级平衡树,它是B树在多维空间上的扩展。在R树中存放的数据并不是原始数据,而是这些数据的最小边界矩形(MBR),空间对象的MBR被包含于R树的叶结点中。在R树空间索引中,设计一些虚拟的矩形目标,将一些空间位置相近的目标,包含在这个矩形内,这些虚拟的矩形作为空间索引,它含有所包含的空间对象的指针。虚拟矩形还可以进一步细分,即可以再套虚拟矩形形成多级空间索引。
R+树,在R树的构造中,要求虚拟矩形一般尽可能少地重叠,并且一个空间对通常仅被一个虚拟矩形所包含。但空间对象千姿百态,它们的最小矩形范围经常重叠。 R+ 改进R树的空间索引,为了平衡,它允许虚拟矩形相互重叠,并允许一个空间目标被多个虚拟矩形所包含。
在Boost.Geometry中有R树的实现,它依赖Boost.Container, Boost.Core, Boost.Move, Boost.MPL, Boost.Range, Boost.Tuple.这些库。R树的元素都是box(矩形)和整数索引值。R树的实现在Geometry中被很好封装,如果使用它,最主要的需要掌握它的查询技巧。先介绍个简单的例子,希望读者能有个大概 的映像,源码如下:
#include <boost/geometry.hpp>
#include <boost/geometry/core/point_type.hpp>
#include <boost/geometry/geometry.hpp>
#include <boost/geometry/geometries/point_xy.hpp>
#include <boost/geometry/geometries/box.hpp>
#include <boost/assign.hpp>
#include <boost/geometry/index/rtree.hpp>
#include <boost/foreach.hpp>
#include <vector>
#include <iostream>
namespace bg = boost::geometry;
namespace bgi = boost::geometry::index;
typedef bg::model::d2::point_xy<double, boost::geometry::cs::cartesian> DPoint; //双精度的点
typedef bg::model::box<DPoint> DBox; //矩形
typedef std::pair<DBox, unsigned> Value;
int main()
{
//创建R树 linear quadratic rstar三种算法
bgi::rtree<Value, bgi::quadratic<16>> rtree;//采用quadratic algorithm,节点中元素个数最多16个
//填充元素
for (unsigned i = 0; i < 10; ++i)
{
DBox b(DPoint(i + 0.0f, i + 0.0f), DPoint(i + 0.5f, i + 0.5f));
rtree.insert(std::make_pair(b, i));//r树插入外包围矩形 i为索引
}
//查询与矩形相交的矩形索引
DBox query_box(DPoint(0, 0), DPoint(5, 5));
std::vector<Value> result_s;
rtree.query(bgi::intersects(query_box), std::back_inserter(result_s));
//查找5个离点最近的索引
std::vector<Value> result_n;
rtree.query(bgi::nearest(DPoint(0, 0), 5), std::back_inserter(result_n));
//显示值
std::cout << "spatial query box:" << std::endl;
std::cout << bg::wkt<DBox>(query_box) << std::endl;
std::cout << "spatial query result:" << std::endl;
BOOST_FOREACH(Value const& v, result_s)
std::cout << bg::wkt<DBox>(v.first) << " - " << v.second << std::endl;
std::cout << "knn query point:" << std::endl;
std::cout << bg::wkt<DPoint>(DPoint(0, 0)) << std::endl;
std::cout << "knn query result:" << std::endl;
BOOST_FOREACH(Value const& v, result_n)
std::cout << bg::wkt<DBox>(v.first) << " - " << v.second << std::endl;
return 0;
}
在源代码中,有详细的注释,请读者先阅读,从中可以看出能够非常的简单的构建一颗R树,然后非常方便的往R树里添加矩形索引。另外一方面Geometry中提供的R树功能非常多,主要包括3类方式找到目标对象。
1.空间几何关系
上图是官方提供的一张图,表示查询的几何关系。查询样式如下:
rt.query(index::contains(box), std::back_inserter(result));
rt.query(index::covered_by(box), std::back_inserter(result));
rt.query(index::covers(box), std::back_inserter(result));
rt.query(index::disjont(box), std::back_inserter(result));
rt.query(index::intersects(box), std::back_inserter(result));
rt.query(index::overlaps(box), std::back_inserter(result));
rt.query(index::within(box), std::back_inserter(result));
2.最近邻查询
std::vector<Value> returned_values;
Point pt(/*...*/);
rt.query(bgi::nearest(pt, k), std::back_inserter(returned_values));//返回最近的k个空间索引
Segment seg(/*...*/);
rt.query(bgi::nearest(seg, k), std::back_inserter(returned_values));
3.函数查询
bool is_red(Value const& v) //函数
{
return v.is_red();
}
struct is_red_o //函数对象
{
template <typename Value>
bool operator()(Value const& v)
{
return v.is_red();
}
}
rt.query(index::intersects(box) && index::satisfies(is_red),
std::back_inserter(result));
rt.query(index::intersects(box) && index::satisfies(is_red_o()),
std::back_inserter(result));
//lambda表达式
#ifndef BOOST_NO_CXX11_LAMBDAS
rt.query(index::intersects(box) && index::satisfies([](Value const& v) { return v.is_red(); }),
std::back_inserter(result));
#endif
R树在几何计算中,最常用的还是空间几何关系查询,同时查询的第一个条件参数还可以采用&&运算符来组合条件,单个可采用非(!)来表示相反的条件,形式如下所示:
//Pred1 && Pred2 && Pred3 && ....
rt.query(index::intersects(box1) && !index::within(box2),
std::back_inserter(result));
rt.query(index::intersects(box1) && !index::within(box2) && index::overlaps(box3),
std::back_inserter(result));
index::query(rt, index::nearest(pt, k) && index::within(b), std::back_inserter(returned_values));
BOOST_FOREACH(Value & v, rt | index::adaptors::queried(index::nearest(pt, k) && index::covered_by(b)))
; // do something with v
最后简单介绍下一个多边形构建R树的例子,源代码如下所示:
#include <boost/geometry.hpp>
#include <boost/geometry/geometries/point.hpp>
#include <boost/geometry/geometries/box.hpp>
#include <boost/geometry/geometries/polygon.hpp>
#include <boost/geometry/index/rtree.hpp>
#include <cmath>
#include <vector>
#include <iostream>
#include <boost/foreach.hpp>
namespace bg = boost::geometry;
namespace bgi = boost::geometry::index;
typedef bg::model::point<double, 2, bg::cs::cartesian> DPoint;
typedef bg::model::box<DPoint> DBox;
typedef bg::model::polygon<DPoint, false, false> DPolygon; // ccw, open polygon
typedef std::pair<DBox, unsigned> DValue;
int main()
{
std::vector<DPolygon> polygons;
//构建多边形
for (unsigned i = 0; i < 10; ++i)
{
//创建多边形
DPolygon p;
for (float a = 0; a < 6.28316f; a += 1.04720f)
{
float x = i + int(10 * ::cos(a))*0.1f;
float y = i + int(10 * ::sin(a))*0.1f;
p.outer().push_back(DPoint(x, y));
}
//插入
polygons.push_back(p);
}
//打印多边形值
std::cout << "generated polygons:" << std::endl;
BOOST_FOREACH(DPolygon const& p, polygons)
std::cout << bg::wkt<DPolygon>(p) << std::endl;
//创建R树
bgi::rtree< DValue, bgi::rstar<16, 4> > rtree; //最大最小
//计算多边形包围矩形并插入R树
for (unsigned i = 0; i < polygons.size(); ++i)
{
//计算多边形包围矩形
DBox b = bg::return_envelope<DBox>(polygons[i]);
//插入R树
rtree.insert(std::make_pair(b, i));
}
//按矩形范围查找
DBox query_box(DPoint(0, 0), DPoint(5, 5));
std::vector<DValue> result_s;
rtree.query(bgi::intersects(query_box), std::back_inserter(result_s));
//5个最近点
std::vector<DValue> result_n;
rtree.query(bgi::nearest(DPoint(0, 0), 5), std::back_inserter(result_n));
// note: in Boost.Geometry the WKT representation of a box is polygon
// note: the values store the bounding boxes of polygons
// the polygons aren't used for querying but are printed
// display results
std::cout << "spatial query box:" << std::endl;
std::cout << bg::wkt<DBox>(query_box) << std::endl;
std::cout << "spatial query result:" << std::endl;
BOOST_FOREACH(DValue const& v, result_s)
std::cout << bg::wkt<DPolygon>(polygons[v.second]) << std::endl;
std::cout << "knn query point:" << std::endl;
std::cout << bg::wkt<DPoint>(DPoint(0, 0)) << std::endl;
std::cout << "knn query result:" << std::endl;
BOOST_FOREACH(DValue const& v, result_n)
std::cout << bg::wkt<DPolygon>(polygons[v.second]) << std::endl;
return 0;
}
4.作者寄语
合理的脚本代码可以有效的提高工作效率,减少重复劳动。