Spatial Analysis
Spatial Analysis
As already stated, architectural design is based on arranging spaces. This led us to the idea that another way to “look” at the drawing is to globally analyze the large whit e “loops”, which are candidates for representing rooms, and to propagate the analysis from these rooms to the walls. We therefore started investigating this second, more original approach.
In order to rapidly evaluate the potential of such an approach, we chose to make a first prototype working on near-perfect data, i.e. simple bitmap images drawn with an editor. Thus, we have less noise and vectorization problems than in real-world scanned drawings.
Text/graphics separation works very smoothly in this case (Fig. 18), as do vectorization and arcs detection (Fig. 19) and loop extraction.
KITCHEN | ROOM 1 | ROOM 2 | ROOM 3 | | | | | CORRIDOR | | | | | | | | | Figure 19: Arcs and segments of | Figure 17: Perfect plan. | Figure 18: Graphics of Fig. 17. | |
Doors and Windows
This step is a small variation of the method; we have only included more configurations (fig. 20). Some of these configurations contain small loops; they are recognized by taking as first candidates, among the detected loops, those whose area is smaller than a given threshold, and by matching these candidates with the models. The other configurations are searched for by looking for a single line aligned with the contours of a room loop.
Figure 20: Different ways to draw doors and windows: with small loops (top), with a single line (bottom).
From loops to rooms
All loops in the image which have not been recognized as belonging to a window or a door are considered as potential rooms. As all rooms must have at least one door, we eliminate the loops for which no associated segment has been used in a door or window recognition. They are considered to belong to other elements, unrecognized at this moment, such as tables, shower, etc.
In addition, uniformity of
As already stated, architectural design is based on arranging spaces. This led us to the idea that another way to “look” at the drawing is to globally analyze the large whit e “loops”, which are candidates for representing rooms, and to propagate the analysis from these rooms to the walls. We therefore started investigating this second, more original approach.
In order to rapidly evaluate the potential of such an approach, we chose to make a first prototype working on near-perfect data, i.e. simple bitmap images drawn with an editor. Thus, we have less noise and vectorization problems than in real-world scanned drawings.
Text/graphics separation works very smoothly in this case (Fig. 18), as do vectorization and arcs detection (Fig. 19) and loop extraction.
KITCHEN | ROOM 1 | ROOM 2 | ROOM 3 | | | | | CORRIDOR | | | | | | | | | Figure 19: Arcs and segments of | Figure 17: Perfect plan. | Figure 18: Graphics of Fig. 17. | |
Doors and Windows
This step is a small variation of the method; we have only included more configurations (fig. 20). Some of these configurations contain small loops; they are recognized by taking as first candidates, among the detected loops, those whose area is smaller than a given threshold, and by matching these candidates with the models. The other configurations are searched for by looking for a single line aligned with the contours of a room loop.
Figure 20: Different ways to draw doors and windows: with small loops (top), with a single line (bottom).
From loops to rooms
All loops in the image which have not been recognized as belonging to a window or a door are considered as potential rooms. As all rooms must have at least one door, we eliminate the loops for which no associated segment has been used in a door or window recognition. They are considered to belong to other elements, unrecognized at this moment, such as tables, shower, etc.
In addition, uniformity of