[M] (Un) Even Space
[Un]Even Space is an image reading machine. By using a dataset of 200 architectural plan drawings and tools like image erosion, the project generates a map showing the analysis of the complexity and dimension of the room segmentation in the original drawings.
SCI 6459 Mechatronic Optics
Instructed by Andrew Witt
image processing, data mining, plan analysis, image erosion
Architectural plans have been the primary tools for architects and engineers to understand a building’s configuration, style, and programmatic arrangement. From Borromini’s San Carlo all Quatro Fontane’s undulating wall divisions and thickened poche areas, to Kahn’s The Dominican Motherhouse’s diverse variation on the size, rotation and dimension of different programs, and even to traditional Japanese home’s interconnecting and homogenous domestic space, those plans all provide us a different reading to the spatial quality. Speculating based on the general room distribution and homogeneity, we could grasp a basic understanding on where does a building lay in the spectrum of public and private, contradiction and repetition, figure and field, relative specificity and relative generosity.
We input a selected data set of 200 architectural plan drawings from architecture-featured websites, filtering out drawings with unclear divisions, with colors, or with unclear graphic elements. Then by using a grasshopper script we extract the boundaries and remove the noise from the plan drawing. This process transform a rasterized image into a collection of computer readable vectors. We then thicken the wall geometry in order to approximately obtain the enclosed areas on the inside, and eventually they are offset and adjusted to match the original room size from the original.
Using the extracted line works that represent rooms of each drawing, we calculated the reconstructed area “floor plan” and created a histogram that illustrates the the distribution of room sizes, frequency and the standard variation value.