C++ R-Tree Library - Do I have to compute the bouding boxes? - c++

Question

I'm trying to use the following library here (the templated version) but in the example shown in the library the user defines the bounding boxes. In my problem I have data of unknown dimensionality each time, so I don't know how to use it. Apart from this, shouldn't the R-Tree be able to calculate the bounding boxes each time there is an insertion?
This is the sample code of the library, as you can see the user defines the bounding boxes each time:
#include <stdio.h>
#include "RTree.h"
struct Rect
{
Rect() {}
Rect(int a_minX, int a_minY, int a_maxX, int a_maxY)
{
min[0] = a_minX;
min[1] = a_minY;
max[0] = a_maxX;
max[1] = a_maxY;
}
int min[2];
int max[2];
};
struct Rect rects[] =
{
Rect(0, 0, 2, 2), // xmin, ymin, xmax, ymax (for 2 dimensional RTree)
Rect(5, 5, 7, 7),
Rect(8, 5, 9, 6),
Rect(7, 1, 9, 2),
};
int nrects = sizeof(rects) / sizeof(rects[0]);
Rect search_rect(6, 4, 10, 6); // search will find above rects that this one overlaps
bool MySearchCallback(int id, void* arg)
{
printf("Hit data rect %d\n", id);
return true; // keep going
}
void main()
{
RTree<int, int, 2, float> tree;
int i, nhits;
printf("nrects = %d\n", nrects);
for(i=0; i<nrects; i++)
{
tree.Insert(rects[i].min, rects[i].max, i); // Note, all values including zero are fine in this version
}
nhits = tree.Search(search_rect.min, search_rect.max, MySearchCallback, NULL);
printf("Search resulted in %d hits\n", nhits);
// Iterator test
int itIndex = 0;
RTree<int, int, 2, float>::Iterator it;
for( tree.GetFirst(it);
!tree.IsNull(it);
tree.GetNext(it) )
{
int value = tree.GetAt(it);
int boundsMin[2] = {0,0};
int boundsMax[2] = {0,0};
it.GetBounds(boundsMin, boundsMax);
printf("it[%d] %d = (%d,%d,%d,%d)\n", itIndex++, value, boundsMin[0], boundsMin[1], boundsMax[0], boundsMax[1]);
}
// Iterator test, alternate syntax
itIndex = 0;
tree.GetFirst(it);
while( !it.IsNull() )
{
int value = *it;
++it;
printf("it[%d] %d\n", itIndex++, value);
}
getchar(); // Wait for keypress on exit so we can read console output
}
An example of what I want to save in an R-Tree is:
-------------------------------
| ID | dimension1 | dimension2|
-------------------------------
| 1 | 8 | 9 |
| 2 | 3 | 5 |
| 3 | 2 | 1 |
| 4 | 6 | 7 |
-------------------------------

Answer 1

Dimensionality
There will be some limit in your requirements to the dimensionality. This is because computers only have infinite storage so cannot store an infinite number of dimensions. Really it is a decision for you how many dimensions you wish to support. The most common numbers of course are two and three. Do you actually need to support eleven? When are you going to use it?
You can do this either by always using an R-tree with the maximum number you support, and passing zero as the other coordinates, or preferably you would create several code paths, one for each supported number of dimensions. I.e. you would have one set of routines for two-dimensional data and another for three dimensional, and so on.
Calculating the bounding box
The bounding box is the rectangle or cuboid which is aligned to the axes, and completely surrounds the object you wish to add.
So if you are inserting axis-aligned rectangles/cuboids etc, then the shape is the bounding box.
If you are inserting points, the min and max of each dimension are just the point value of that dimension.
Any other shape, you have to calculate the bounding box. E.g. if you are inserting a triangle, you need to calculate the rectangle which completely surrounds the triangle as the bounding box.
The library can't do this for you because it doesn't know what you are inserting. You might be inserting spheres stored as centre + radius, or complex triangle mesh shapes. The R-Tree can provide the spatial index but needs you to provide that little bit of information to fill in the gaps.