Plane extraction from inhomogeneous point clouds and plane-to-plane alignments

Abstract

Over the last few years, increasing demands for building interior surveying have brought the challenge of acquiring the geometric information of indoor environments effectively and efficiently. Though the development of robotic engineering had provided some preliminary solutions for building the virtual world for the robotics, their precision and accuracy could not satisfy the applications in Architecture, Engineering, and Construction (AEC). Various solutions have been introduced with limitations in robust feature extraction, working range, accessibility, detail levels, coverage, and reliability. Meanwhile, popular methods used for extracting features, especially planes, from point clouds for mobile mappings were not sufficient, accurate and robust enough for establishing 6 Degree-of-Freedom (DOF) point alignment workfows. A novel method for plane extraction from low-resolution inhomogeneous point clouds captured by multi-line Mobile Laser Scanners (MLS), the Enhanced Line Simplification (ELS) algorithm, was proposed and developed. The method employed raw data acquisition sequence to form point grids for analyzing curvatures and identifying feature points along raw scanlines and dedicated virtual scanlines. By clustering identifed line segments concerning scanline directions, patches were detected and merged to form planes. The method eliminated the calculation of the local estimated normals and overcame the over-segmentation problem in processing noisy data. Utilizing the plane extraction results, a plane-to-plane point cloud alignment workflow was presented for aligning point clouds captured on a mobile platform to the same frame. The workfow recovered the rotation and translation relationships between frames by identifying common planes and non-linear optimization. The implementation of the coarse-to-fine procedure and the shortest-path initialization strategy waived the use of Inertial Measurement Units (IMU) and other positioning sensors. A backpack prototype, which adopted two multi-line laser scanners as the primary sensors, was designed to test the performance of the proposed methods in multiple scenarios. The results showed that the proposed hardware system and the processing workflow could achieve acceptable accuracy and reliability in typical indoor and specific outdoor environments. The robustness of this IMU-free point cloud alignment workflow was verified as well, which could be applied in future mobile mapping and sensor fusion applications.