Density-Invariant Registration of Multiple Scans for Aircraft Measurement

Abstract

In the aviation industry, the demand for high accuracy airplane product is growing, which makes precise production of airplane parts and accurate manufacturing increasingly important. To this end, it is crucial to be able to accurately measure the whole surface of an aircraft. Three-dimensional laser scanner is widely utilized to capture the local shapes, represented as 3D point clouds, of an object from different viewpoints. Multiview registration of point clouds is therefore a critical step to obtain the whole shape of an object. In this paper, we propose a global registration framework to simultaneously align multiple point clouds with target detection and hierarchical optimization for aircraft inspection. By placing some targets (i.e., markers) surrounding an aircraft, we first scan the aircraft by putting a laser scanner around the aircraft at various stations, resulting in a number of laser scans which contain the point clouds of aircraft parts as well as targets. By detecting the centers of targets automatically, all partial point clouds are initially aligned to the global coordinate system. Furthermore, we tackle the influence of non-uniform distribution of point cloud density on registration accuracy, which has not been extensively studied so far, due to the large size of the aircraft. Existing approaches cannot directly apply to large size point clouds registration due to the aforementioned challenge. To address this issue, we propose a density-invariant area-based method to measure the overlapped region. On this basis, a hierarchical optimization registration method is used to achieve multiview registration of aircraft point clouds, and thereby the entire geometry shape of the aircraft is accurately obtained. A variety of experiments on real raw data demonstrate the effectiveness and robustness of our proposed framework.