Improved Iterative CD-Spline Approach for Building Boundary Regularization Using Airborne LiDAR Data

Building boundary regularization/modeling is an important task in urban mapping using remote sensing data for contour representation. In the previous Iterative Changeable Degree-Spline (ICDS) approach, it was noticed that contour regularization can be affected by the selection of critical points (CPs) (corner points), responsible for identifying the end points of each contour segment. In the ICDS strategy, CPs are detected using Douglas–Peucker algorithm followed by an angle-based constraint; then, a single set of CPs is used in a regularization process, not being possible to guarantee that the derived result corresponds to the best contour model. To overcome this limitation, we proposed the Improved Iterative Changeable Degree-Spline (IICDS) that consists of testing several CP configurations, resulting in multiple contour models for each building. In the proposed strategy, the optimal contour is automatically selected by evaluating the number of CPs and minimizing the distance between the initially extracted and regularized contours. In the experiments, two airborne LiDAR datasets acquired in different urban environments were utilized. The obtained results indicated the potential of the IICDS approach for simple and complex contours, presenting F_score and positional accuracy around 95% and 0.3 m, respectively.