Query-Points Visibility Constraint Minimum Link Paths
in Simple Polygons

One of the problems in the field of Robotics and Computational Geometry is finding a minimum link path between two points in a simple polygon. A minimum link path between two points $s$ and $t$ is a chain of line segments (links) connecting them inside a simple polygon $P$ with $n$ vertices that has the minimum number of links. The link distance between $s$ and $t$ is defined as the number of links in a minimum link path. Finding a minimum link path between two fixed points inside a simple polygon was first studied by Suri [15]. He introduced an $O(n)$ time algorithm for this problem. Afterwards, Ghosh [8] presented an alternative algorithm, which also runs in $O(n)$ time. To solve this problem in a polygonal domain, Mitchell et al. [14] proposed an incremental algorithm that runs in time $O(n^{2}\log^{2}n)$, where $n$ is the total number of edges of the obstacles. On the other hand, a more general framework was established for minimum link paths by Suri [16] based on Shortest Path Map (SPM). By the construction of SPM from a fixed point, the simple polygon $P$ is divided into faces of equal link distance from that point in linear time. With this property, Arkin et al. [2] developed an algorithm for computing the link distance between two arbitrary query points inside $P$. The algorithm computes window partitioning (SPM) of $P$ from every vertex of $P$ and from every extension point of the visibility graph of $P$. It can be seen that endpoints of these windows divide edges of $P$ into $O(n^{2})$ atomic segments. These segments satisfy the property that the combinatorial type of $SPM(x)$ is the same for all points $x$ in the interior of each segment. Thus, the algorithm requires $O(n^{3})$ preprocessing time and answers a link distance query in $O(\log n)$ time.

In many applications, it is required for a robot to have direct visibility from a viewpoint during its motion [5]. Some examples are moving guards, resource collectors, wireless communications, etc. Minimum link paths have important applications in Robotics since turns are costly while straight line movements are inexpensive in robot motion planning. In the minimum link paths problem with point visibility constraint, the aim is to find a minimum link path between two points $s$ and $t$ such that there is at least one point on the path from which a given viewpoint $q$ is visible (a $q$-$visible$ path [18]).

The constrained version of minimum link paths problem was studied in [18] for three fixed points $s$, $t$ and $q$ inside a simple polygon. In this paper, we study the query version of the problem inside a simple polygon $P$ with $n$ vertices for two cases. First, suppose that two fixed points $s$ and $t$ are given, the goal is to find a $q$-$visible$ path between $s$ and $t$ for an arbitrary query point $q$. Second, we consider the same problem for three arbitrary query points $s$, $t$ and $q$. We propose two algorithms with $O(n)$ and $O(n^{3})$ preprocessing time for the above cases, respectively. The link distance of a $q$-$visible$ path between $s$, $t$ as well as the path are provided in time $O(\log n)$ and $O(m+\log n)$, respectively, for both cases, where $m$ is the number of links.

Similarly, a $q$-$visible$ path is defined for the Euclidean metric. In this case, the query version of the shortest Euclidean path problem for two fixed points $s$, $t$ and a query point $q$ inside a simple polygon with $n$ vertices was studied in [13]. The given algorithm preprocesses the input in $O(n^{3})$ time and provides $O(\log n)$ query time. A simpler form of the problem was studied in [12] in which the goal is to find the shortest Euclidean path from $s$ to view $q$ (without going to a destination point). The algorithm requires $O(n^{2})$ preprocessing time and answers a query in $O(\log n)$ time. Arkin et al. [3] improved the preprocessing time to $O(n)$ for simple polygons. Also, they built a data structure of size $O(n^{2}2^{\alpha(n)}\log n)$ that can answer each query in $O(n\log^{2}n)$ time for a polygonal domain with $h$ holes and $n$ vertices, where $\alpha(n)$ is the inverse Ackermann function. Recently, a new data structure of size $O(n\log h+h^{2})$ was presented for this problem that can answer each query in $O(h\log h\log n)$ time [17].

The main differences between approaches in the Euclidean metric and the link distance metric are as follows. Optimal paths that are unique under the Euclidean metric need not be unique under the link distance metric. Also, Euclidean shortest paths only turn at reflex vertices while minimum link paths can turn anywhere. Thus, minimum link paths problems are usually more difficult to solve than equivalent Euclidean shortest path problems.

The main idea of the proposed algorithms given in this paper is to consider an edge of the visibility polygon (called $e_{q}$) for a query point $q$ as a separator chord inside $P$, i.e., if $s$ and $t$ lie in different sides of such a chord, an optimal link path between them will be the answer. Otherwise, a $q$-$visible$ path should have a non-empty intersection with the side of the chord containing $q$. Therefore, the problem can be reduced to find an appropriate edge and optimal contact points between a $q$-$visible$ path and the other side of the edge. To answer the queries efficiently, we preprocess the input using map overlay [7], point location [6], ray shooting [9] and shortest path map [16] techniques.

In Section 2, we introduce the problem definition and notation. Section 3 gives the basic lemmas and definitions. Section 4 shows the main idea and flow of the algorithm. Section 5 describes our algorithm for single query point, and Section 6 generalizes this algorithm to triple query points. Section 7 concludes with some open problems.

Let $P$ be a simple polygon in the plane with $n$ vertices. For three points $s$, $t$ and $q$ inside $P$, the goal is to preprocess the input to answer two types of queries:

1. $1)$

   Given a query point $q$, find all $q$-$visible$ paths between fixed points $s$ and $t$ in $P$ (single query).

2. $2)$

   Given three query points $s$, $t$ and $q$, find a $q$-$visible$ path between $s$ and $t$ in $P$ (triple query).

We use the following notation throughout the paper:

1. $\bullet$

   $V(x):$ the visibility polygon of a point $x$ $\in$ $P$

2. $\bullet$

   $\pi_{L}(x,y):$ a minimum link path from a point $x$ to a point $y$ inside $P$

3. $\bullet$

   $\pi_{E}(x,y):$ the shortest Euclidean path from a point $x$ to a point $y$ inside $P$

4. $\bullet$

   $MLP(x,y,q):$ a $q$-$visible$ (minimum link) path from a point $x$ to a point $y$ inside $P$

5. $\bullet$

   $|X|:$ the link distance of a minimum link path $X$

6. $\bullet$

   $n(X):$ the number of members of a set $X$

7. $\bullet$

   $Pocket(x):$ invisible regions of $P$ from a point $x$, which are separated from $V(x)$

8. $\bullet$

   $SPM(x):$ the shortest path map (window partition) of $P$ with respect to a point or line segment $x$ [16]

More precisely, we are looking for a minimum link path between $s$ and $t$ that should have a non-empty intersection with $V(q)$ for both cases. Each region of $Pocket(q)$ has exactly one edge in common with $V(q)$, called an edge of $V(q)$. However, there is no need to compute all edges of $V(q)$. Indeed, a single edge $e_{q}$ of $V(q)$ would be sufficient to find a $q$-$visible$ path (not any such edge, see Section 4.1).

Without preprocessing the query can be answered in linear time [18], but here our goal is to achieve a logarithmic query time using a preprocessing (this would be optimal as indicated in [2]). We define two types of output for either case mentioned above: one to find out $\lvert MLP(s,t,q)\rvert$ in $O(\log n)$ time, and another to report $MLP(s,t,q)$ in $O(m+\log n)$ time, where $m=\lvert MLP(s,t,q)\rvert$.

A planar subdivision is a partition of the 2-dimensional plane into finitely many vertices, edges and faces $(V,E,F)$. One type of planar subdivision is simply connected planar subdivision (SCPS) with an additional restriction: any closed paths lying completely in one region of a SCPS can be topologically contracted to a point. Therefore, a region in a SCPS cannot contain any other regions [7]. According to Euler’s formula for SCPS, we have the following equation: $n(V)-n(E)+n(F)=2$.

As stated above, the notion of SPM introduced in [16] is central to our discussion. Indeed, $SPM(x)$ denotes the SCPS of $P$ into faces with the same link distance to a point or line segment $x$. $SPM(x)$ has an associated set of windows, which are chords of $P$ that serve as boundaries between adjacent faces. Starting and end points of a window $w$ will be denoted by $\alpha(w)$ and $\beta(w)$, respectively. We have the following two lemmas as the basic fundamental properties of SPM:

The value of each face of $SPM(x)$ is defined as the link distance from $x$ to any points of that face. These values are added to each face during the construction of $SPM(x)$ in Lemma 3.1. Let $||SPM(x)||$ denote the number of faces of $SPM(x)$. Also, let $F_{x}(i)$ be a face of $SPM(x)$ and let $||F_{x}(i)||$ denote the value of $F_{x}(i)$, where $1\leq i\leq||SPM(x)||$ (see Figure 1).

The window tree $WT(x)$ denotes the planar dual of $SPM(x)$. It has a node for each face and an arc between two nodes if their faces share an edge. $WT(x)$ is rooted at $x$ and each node of it is labeled with a window. The computation of $WT(x)$ takes $O(n)$ time [16]. According to [16], $F_{x}(i)$ is generated by its corresponding window $w_{i}\in WT(x)$ and vice versa, i.e., we define $G_{x}(w_{i})=F_{x}(i)$ and $w_{i}=G^{-1}_{x}(F_{x}(i))$ ($1\leq i\leq||SPM(x)||$). If $i$ is not specified, we define $G_{x}(w(x))=F_{x}$ and $w(x)=G^{-1}_{x}(F_{x})$, where $w(x)$ is a window of $WT(x)$ and $F_{x}$ is the corresponding face of $SPM(x)$.

Consider the shortest path maps of $P$ with respect to the points $s$ and $t$. To find the intersection of the two maps $SPM(s)$ and $SPM(t)$, the map overlay technique is employed. One of the well-known algorithms for this purpose was introduced by Finke and Hinrichs [7]. The algorithm computes the overlay of two SCPSs in optimal time $O(n+r)$, where $n$ is the total number of edges of the two SCPSs and $r$ is the number of intersections between the edges in the worst case. Both the input and output SCPSs are represented by the quad view data structure (a trapezoidal decomposition of a SCPS) [7].

The intersection of $SPM(s)$ and $SPM(t)$ creates a new SCPS inside $P$. We call each face of this SCPS a Cell. Also, $Ce$ is defined as a set such that each member of it points to a Cell. By the construction of Cells, each Cell ($Ce(i)$) is the intersection of two faces, one face from $SPM(s)$ and another face from $SPM(t)$, i.e., $Ce(i)$= $F_{s}(j)\cap F_{t}(k)$. The value of each Cell (the value of $Ce(i)$) is defined as $||Ce(i)||=||F_{s}(j)||+||F_{t}(k)||$ ($1\leq i\leq n(Ce)$, $1\leq j\leq||SPM(s)||$ and $1\leq k\leq||SPM(t)||$). These values are added to each Cell during the overlaying, and will later be used for computing $|MLP(s,t,q)|$ (see Figure 2).

Let $W_{s}$ and $W_{t}$ be the set of windows of $SPM(s)$ and $SPM(t)$, respectively, and $W=(W_{s}\cup W_{t})$. The following lemma shows the number of Cells ($n(Ce)$):

As depicted in the example in Figure 2, $n(Ce)=32$, $n(W_{s})=n(W_{t})=14$, $n(W_{s}\cup W_{t})=28$ and $r=3$. The number of the windows of SPM inside $P$ depends on the number of the reflex vertices of $P$. Since the total internal angles of $P$ is $(n-2)*180$ degrees, $n(W_{s})$ and $n(W_{t})\leq n-3$. According to Lemma 3.3, each window of $W_{s}$ intersects at most two windows of $W_{t}$ and vice versa. Thus, each window in $W_{s}$ or $W_{t}$ contains at most two intersection points inside $P$, i.e., $r\leq 2*\min(n(W_{s}),n(W_{t}))\leq 2(n-3)$. Based on Lemma 3.4 and the above argument, we have:

$n(Ce)=n(W)+r+1\leq n(W_{s})+n(W_{t})+r+1\leq(n-3)+(n-3)+2(n-3)+1=4n-11=O(n)$.

The visibility graph of $P$, denoted by $V_{G}(P)$ is an undirected graph of the visibility relation on the vertices of $P$. $V_{G}(P)$ has a node for every vertex of $P$ and an edge for every pair of visible vertices inside $P$. Consider a visibility graph edge $e$. The extension points of $e$ refer to the intersections of the boundary of $P$ with the line containing $e$. $V_{G}(P)$ and its extension points can be computed in time proportional to the size of $V_{G}(P)$, i.e., $O(E)$, where $E$ is the number of edges in $V_{G}(P)$ [11]. Indeed, the main algorithm in [11] outputs the edges of $V_{G}(P)$ in sorted order around each vertex.

A more complicated algorithm is presented in [9] for the ray shooting problem in a simple polygon. This algorithm has $O(n)$ (instead of $O(E)$) preprocessing time and the same query time ($O(\log n)$). Note that $E=\Omega(n)=O(n^{2})$. We must use this algorithm for single query (Section 5), but Lemma 3.8 suffices for triple query (Section 6).

The following cases ($Q(x)$) may occur for a query point $q$:

1. $Q(a)$

   At least one of $s$ or $t$ is visible from $q$, i.e., $s$ or $t$ in $V(q)$. In this case, $MLP(s,t,q)=\pi_{L}(s,t)$.

2. $Q(b)$

   The points $s$ and $t$ are in two different regions of $Pocket(q)$. Again in this case, $MLP(s,t,q)=\pi_{L}(s,t)$. Since $\pi_{L}(s,t)$ crosses $V(q)$, it is a $q$-$visible$ path.

3. $Q(c)$

   Both $s$ and $t$ are in the same region in $Pocket(q)$. In this case, $e_{q}$ is the common edge between this region of $Pocket(q)$ and $V(q)$. Therefore, $MLP(s,t,q)$ should have a non-empty intersection with the side of $e_{q}$, where $q$ lies.