This project proposes a new methodology for building roof boundary regularization based on the CD-Spline concept. This concept was first explored during the PhD. The main advantage of the CD-Spline is related to modeling of complex contours, since it considers polynomial functions of different degrees. The results obtained in the PhD indicated the potentiality of CD-spline in the boundary regularization applied to LiDAR data. Despite this, some limitations were identified: the result is directly influenced by the corner point identification and by the presence of outliers in the boundary; modeling errors in occlusion regions caused by the presence of vegetation near buildings; and difficulty in identifying and representing small segments. In order to overcome these problems, a new approach is proposed, which includes the use of LiDAR data and aerial images integrated, automatic selection of corner points by means of the residual analysis, inclusion of outlier identification method and determination of a weight function for the boundary points. The experiments will be performed considering two datasets, both composed of aerial images and LiDAR data. The quality evaluation will be carried out through qualitative and quantitative analysis, being highlighted the advantages and limitations of proposed method. It is expected that the new methodology allows the mitigation of the highlighted problems and, consequently, the achievement of more accurate boundaries.
News published in Agência FAPESP Newsletter about the scholarship: