Given a 2D space inhabited by circle shapes (grey), is there an easy way to find the largest empty circular area (blue)? I guess I have to look at space partioning, maybe quadtrees, but I'm curious as to whether there might be existing solutions, especially for circular shapes?
My idea (neither thought through or implemented, and computationally expensive):
Divide the space into a grid, the finer, the more precise
At every grid point, calculate D, the smallest distance to any of the neighbouring circles C_i (= distance to center of C_i - radius of C_i)
Choose the grid point that maximizes D
Related
In my current project I need to calculate the intersection area of triangles and the unit squares in an infinite grid.
For every triangle (given by three pairs of floating point numbers) I need to know the area (in the interval (0,1]) it has in common with every square it intersects.
Right now I convert both (the triangle and the square) to polygons and use Sutherland-Hodgman polygon clipping to calculate the intersection polygon, which I then use to calculate its area.
This approach now shows to be a performance bottleneck in my application. I guess a more specialized (analytical) algorithm would be much faster. Is there a standard solution for this problem, or do you have any idea? I only need the areas, not the shape of the intersections.
Your polygon are convex. There are some algorithms for convex polygons faster than general ones. I've used O'Rourke algorithm with success (code from his book here, I believe that good description exists). Note that some values may be precomputed for your squares.
If your polygons not always intersect, then you may at first check the fact of intersection with separating axes method.
Another option to try- Liang-Barski algorithm for clipping every triangle edge by square.
Edit: You can quickly find all intersections of triangle edges with grid using algorthm of Amanatides and Woo (example in grid traversal section here)
To process this task with hi performance , i suggest some modifications of
Vatti line sweep clipping.
http://en.wikipedia.org/wiki/Vatti_clipping_algorithm
Stepping from minimal Y vertex of your Triangle make such steps:
sort vertexes by Y coordinate
step Y higher to MIN(nextVertex.Y, nextGridBottom)
Calculate points of intersection of grid with edges.
Collect current trapezoid
repeat from step2 until vertex with highest Y coordinate.
Split trapezoids by X coordinate if required.
here is example of Trapezoidalization in X direction
http://www.personal.kent.edu/~rmuhamma/Compgeometry/MyCG/PolyPart/polyPartition.htm
It illustrate main idea of line sweep algorithm. Good luck.
You are not mentioning what precision you are looking for. In case you are looking for a analytical method, disregard this answer, but if you just want to do antialiasing I suggest a scanline edge-flag algorithm by Kiia Kallio. I have used it a few times and it is quite fast and can be set up for very high precision. I have a java implementation if you are interested.
You can take advantage of the regular pattern of squares.
I'm assuming the reason this is a bottleneck is because you have to wait while your algorithm finds all squares intersecting any of the triangles and computes all the areas of intersection. So we'll compute all the areas, but in batches for each triangle in order to get the most information from the fewest calculations.
First, as explained by others, for each edge of the triangle, you can find the sequence of squares that edge passes through, as well as the points at which it crosses each vertical or horizontal edge of a square.
Do this for all three sides, keeping a list of all the squares you encounter, but keep only one copy of each square. It may be useful to store the squares in multiple lists, so that all squares on a given row are all kept in the same list.
When you've found all squares the triangle's edges pass through, if two of those squares were on the same row, any squares between those two that are not in the list are completely inside the triangle, so 100% of each of those squares is covered.
For the other squares, the calculation of area can depend on how many vertices of the triangle are in the square (0, 1, 2, or 3) and where the edges of the triangle intersect the sides of the square. You can summarize all the cases in a few pencil-and-paper drawings, and come up with calculations for each one. For example, when an edge of triangle crosses two sides of the square, with one corner of the square on the "outside" side of the edge, that corner is one angle of a small triangle "cut off" by that edge of the larger triangle; use the points of intersection on the square's sides to compute the area of the small triangle and deduct it from the area of the square. If two points instead of one are "outside", you have a trapezoid whose two base lengths are found from the points of intersection, and whose height is the width of the square; deduct its area from the square. If three points are outside, deduct the entire area of the square and then add the area of the small triangle.
One vertex of the large triangle inside the square, three corners of the square outside that angle: draw a line from the remaining corner to the triangle's vertex, so you have two small triangles, deduct the entire square and add those triangles' areas. Two corners of the square outside the angle, draw lines to the vertex to get three small triangles, etc.
I'm phrasing this so that you always assume you start with the entire area of the square and reduce the area by some amount depending on how the edge of the triangle intersects the square. That way, in the case where the edges of the triangle intersect the square more than twice--such as one edge cuts across one corner of the square and another edge cuts across a different corner, you can just deduct the area cut off by the first edge, then deduct the area cut off by the second edge.
This will be a considerable number of special cases, though you can take advantage of symmetry; for example, you don't have to write the complete calculation for "cut off a triangle in one corner" four times.
You'll write a lot more code than if you just took someone's convex-polygon library off the shelf, and you will want to test the living daylights out of it to make sure you didn't forget to code any cases, but once you get it working, it shouldn't take much more effort to make it reasonably fast.
In N (~ 500) dimensions, I wish to find out the largest sphere or rectangle such that the sphere/rectangle does not contain already existing points. The entire set of points is bounded in an axis-aligned rectangular box (lower and upper bounds on the values).
Is there any known polynomial time method/code that I can use to solve my problem?
The two well known algorithms: i) the largest empty rectangle within a rectangle (http://www.cs.princeton.edu/~chazelle/pubs/ComputLargestEmptyRectangle.pdf) and, ii) finding largest empty circle within location constraints (http://www.cs.dartmouth.edu/reports/TR86-130.pdf) do not work.
Although the complexity of above algorithms is N log N or N^2 log N, where N is the number of already existing points, the complexity is also a linear function of the number of vertices of the convex hull or the bounding polygon. A rectangle in 500 dimensions will have 2^500 corners which makes the above techniques infeasible.
Ideally, I am looking for a method (it does not have to be exact) that can determine the largest cirle/rectangle in polynomial time in N (number of points) and D (the dimension).
Thank you.
One possible heuristic solution is to restrict the large rectangle so that it's axis-aligned. In this case, a rectangle can be bounded by at most 2 * d points, where each point represents a bounding min or max to one or more dimensions. It can be then be determined if a point is inside that rectangle in only O(d).
A heuristic method to make use of this is to pick 2 points, and use those to form an initial bounding box, then randomly add points. If the point is inside the box, shrink or split the box. If the point is outside of the box, try to make the box bigger. A single pass should take O(d * N) if the box is shrunk instead of split.
The quality perhaps can be improved a bit by ensuring that no point is within the bounding box formed by the two points. It might be ideal to find all pairs of points such that no point is within the bounding box, as when converted to a graph, they should help with expanding the solution in a less random way. Dynamic programming likely leads to an algorithm that is better than O(d*N^3) perhaps O(d*N^2) or better.
I have a set of 2D points, unorganized, and I want to find the "contour" of this set (not the convex hull). I can't use alpha shapes because I have a speed objective (less than 10ms on an average computer).
My first approach was to compute a grid and find the outline squares (squares which have an empty square as a neighbor). So I think I downsized efficiently my numbers of points (from 22000 to 3000 roughly). But I still need to refine this new set.
My question is : how do I find the real outlines points among my green points ?
After a weekend full of reflexions, I may have found a convenient solution.
So we need a grid, we need to fill it with our points, no difficulty here.
We have to decide which squares are considered as "Contour". Our criteria is : at least one empty neighbor and at least 3 non empty neighbors.
We lack connectivity information. So we choose a "Contour" square which as 2 "Contour" neighbors or less. We then pick one of the neighbor. From that, we can start the expansion. We just circle around the current square to find the next "Contour" square, knowing the previous "Contour" squares. Our contour criteria prevent us from a dead end.
We now have vectors of connected squares, and normally if our shape doesn't have a hole, only one vector of connected squares !
Now for each square, we need to find the best point for the contour. We select the one which is farther from the barycenter of our plane. It works for most of the shapes. Another technique is to compute the barycenter of the empty neighbors of the selected square and choose the nearest point.
The red points are the contour of the green one. The technique used is the plane barycenter one.
For a set of 28000 points, this techniques take 8 ms. CGAL's Alpha shapes would take an average 125 ms for 28000 points.
PS : I hope I made myself clear, English is not my mothertongue :s
You really should use the alpha shapes. Maybe use only green points as inputs of the alpha alpha algorithm.
In an infinite 2D space there are a set of lines, each line having a start and end point, and a time of creation: Line(p0, p1, t).
I want to find the lines that should be rendered in a top-down view of this 2D space (higher values of t show up closer to the viewport, not that it should be relevant.)
The intuitive answer is "check if either point is within the viewport coordinates," but this falls down when the points are further apart than the viewport area covers.
The other idea I had was using something like geohash, this would limit precision i.e. maximum zoom level of the viewport. The idea is enumerating the hashes of the cells intersected and storing them. This way querying is a matter of asking the right question.
Are there any ideal solutions? Has this been solved before?
I think you need to check two conditions: one that the rectangle of viewport overlaps the rectangle with corners (p0,p1) and the second that some corners of viewport rectangle are on the different sides of the whole line which contains line segment (p0,p1).
The task of finding rectangle overlap can be solved very effectively for very large number of rectangles using R-trees (http://en.wikipedia.org/wiki/R-tree).
The second task can be reduced to checking signs of the cross product of (p1-p0) x (corner_coordinate-p0)
(all three quantities taken as 3-d vectors with third coordinate equal to zero, the result will be vector along the perpendicular direction). There should be corners with the opposite sign of this cross product.
Summed area table is a pre-filtering technique to avoid aliasing in texture map. I would like to know how it works and what are the advantages and disadvantages over Mipmap. Thanks
A summed area table replaces the X*Y texture, T, with an X*Y set of values, S, where entry x,y in S is the sum of all pixels of T contained in the axis-aligned rectangle from [0,0] to [x, y] inclusive.
Given that you want to compute the average of an arbitrary axis-aligned rectangle of pixels of T, you just need to sample the corner locations in S and do some trivial additions and subtractions. (Hmm. It seems that wikipedia sums (pun intended) this up quite succinctly)
The advantages of the SAT is that it will quickly give you the correct results for any arbitrary axis-aligned rectangle (rather than being limited to power-of-2 squares as in MIP mapping) but
You won't get automatic hardware support as you do with MIP maps
the SAT can be huge because, assuming you start summing from the top left texel, the bottom right needs to be able to equal the sum of all the texels. For a 1K*1K texture, you'd need 11 extra bits of precision per channel!