Detecting Discontinuous and Occluded Boundaries from Point Clouds of Building Interiors

Range scans of occupied building interiors will often generate a cumulative point cloud with disconnected regions due to varying data density and the presence of numerous partially occluded objects. Boundary detection methods based on surface normal vectors and curvature have difficulty in accurately representing occluded region boundaries because of the geometric uncertainty associated with the underlying polygonal surfaces used to determine the desired geometric parameters. An occluded boundary detection algorithm that works directly on point clouds without the need to reconstruct rough underlying surface models is presented in this paper. The algorithm uses a side-ratio constraint to identify the discontinuous boundary points which lie along successive scan lines. The basic principle is that the distance between the immediate neighboring points at the discontinuous boundary exhibits a large disparity when compared to the other points in a contiguous surface. The side ratio distance defines the spatial separation of the proceeding and succeeding data points on the local grid. The algorithm is also able to handle small density inconsistencies by continuously comparing the side-ratios of the nearest points within a preset window. Spurious point data incorrectly identified as discontinuous boundary points are removed using a density-based outlier detection technique. The effectiveness of the two-step algorithm is demonstrated on real-world data acquired using a FARO® LS 880 laser scanner.