Analysis of visibility within urban landscapes (external and/or internal visibility) raises additional problems. With buildings and streets represented as 2D polygons, rather than 3D grids, a first level of visibility analysis can be carried out, with applications including questions such as where to place CCTV cameras or security guards in order to ensure all points are visible from the selected locations. This in turn involves an NP-hard optimisation problem (see further Section 7.1.4), that of determining the minimum number of such points to ensure complete coverage. The problem also extends in a natural manner to 3D environments. Rana (2004b) has developed efficient heuristic procedures for solving such problems based on a combination of systematic evaluation of visibility from a fine grid of initial observer positions, and then reducing this set by a ranking rule (Ranking and Overlap Elimination, or ROPE procedure). By stochastically iterating this procedure using randomly selected observer positions (S-ROPE) the final count of observer points may be reduced and thus approach optimality. An ArcView add-on (Isovist Analyst) is available to perform this operation. The procedure is illustrated in Figure 6‑24 for an area of central London (Aldwych). A sample isovist region is shown in green (street areas), based on one of the initial test observer locations shown as a red dot symbol. The blue dotted box symbols indicate the 27 locations identified by the program using the S-ROPE algorithm with around 8000 starting points as the (approximately) optimal positions for surveillance cameras, ensuring 100% coverage. Restricting the initial solution set to medial axis regions determined by simple distance transform methods (as described in Section 4.4.2.2) shows an immediate further improvement of these results (Rana, 2006). It is interesting to consider how difficult it would be for architects, surveillance specialists or traffic monitoring units to perform such analysis manually.