PROJECT TITLE :

Building Types’ Classification Using Shape-Based Features and Linear Discriminant Functions

ABSTRACT:

In this paper, the applicability and performance of linear discriminant analysis (LDA) for building varieties’ classification are investigated. Building models at a level of detail 1 (LoD1) are derived from realty cadastral building footprints and digital surface models from stereoscopic airborne pictures. In several experiments for two cities in Germany (Berlin and Munich), we have a tendency to initial evaluate the discriminatory power of twenty six totally different shape-based options which describe the physiognomy of individual buildings in terms of 1-D (e.g., length), two-D (e.g., area), and three-D (e.g., volume) features. While one-D options show low contributions to the discrimination of the five building varieties, we tend to observe high contributions of the three-D shape index and a pair of-D measures of compactness. In an exceedingly second group of experiments, the scale of coaching samples for the classification method is investigated with the end result that a size of tenpercent of the whole range of labeled features is practicable in terms of size and accuracy. In an exceedingly third battery of experiments, the chosen options and training sample size are used for the classification of building sorts ensuing in kappa values of 0.ninety four for both cities. In the ultimate experiments, the geographical transfer between the 2 cities is investigated reaching kappa values of 0.93 and zero.91, respectively. The tests show that a easy linear classifier like LDA will handle building varieties’ classification while not abundant user interaction compared to more advanced classification methods but is limited when similar building varieties (e.g., perimeter block development and block development) are to be discriminated.